(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Children's Library | Biodiversity Heritage Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
Upload
See other formats

Full text of "Studies in the linguistic sciences"

410 
oT92 
7.23 
i993 
cop. 2 



NOTICE: Return or renew all Library Materialsl The Minimum Fee for 
each Lost Book Is $50.00. 

The person charging this material is responsible for 
its return to the library from which it was withdrawn 
on or before the Latest Date stamped below. 

Theft, mutilation, and underlining of books are reasons for discipli- 
nary action and may result in dismissal from the University. 
To renew call Telephone Center, 333-8400 

UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN 



ULLi 



11 nl 



i-OO'. 



L16I — O-1096 



studies in 

The Linguistic Sciences 



410 STX 

St92 

23:1 SPR 1993 COPY 2 

NIKEN ADISASMITO: Syllable structure and the nature of schwa in 

Indonesian 
Paul AGBEDOR: Verb serialization in Ewe 

Martin J. Baik and Rosa Shim: Yes, we have no bananas: 
English negative tags in cross-linguistic communication 

ELABBAS BENM amoun: The status of agreement and the agreement 

projection in Arabic 
Marvin K. L. ChinG: Examining the trustworthiness of the latest 

OED in reflecting current English 
ABDUL AZIZ DlOP: Language planning across political boundaries: 

A case study of Pulaar 
ANDREW TiLIMBE KULEMEKA: Bimoraicity in monosyllabic 

Chichewa ideophones 
Sharon R. Morrison: a re-examination of cardinal vowels and 

auditory equidistance 
Steven SCHAUFELE: The Vedic clause-initial string and universal 

grammar 

Squib 

M. LYNNE MURPHY: Discourse markers and sentential syntax 

Reviews 
Anvita Abbi (1992). Reduplication in South Asian languages: An 

areal, typological, and historical study. (Hans Henrich Hock) 
Narindar K. Aggarwal (1991). Studies on Nepali language and 

linguistics: A bibliography. (Mithilesh K. Mishra) 
John Baldwin & Peter French (1990). Forensic phonetics. (Jose 

Ignacio Hualde) 

Recent Books 

Contents of volumes 1 8 - 22 




1 
21 

43 

61 

73 

83 

107 

117 

131 

163 

169 

193 

195 
197 
201 



Department of Linguistics 
University of Illinois 



STUDIES IN THE LINGUISTIC SCIENCES 



PUBLICATION OF THE DEPARTMENT OF LINGUISTICS 

COLLEGE OF LIBERAL ARTS AND SCIENCES 

UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN 



EDITOR: Hans Henrich Hock; REVIEW EDITOR: James H. Yoon 

EDITORIAL BOARD: Elmer H. Antonsen, Eyamba G. Bokamba, Chin-Chuan 
Cheng, Jennifer S. Cole, Georgia M. Green, Hans Henrich Hock, Braj B. Kachru, 
Yamuna Kachru, Chin-W. Kim, Charles W. Kisseberth, Howard Maclay, Jerry L. 
Morgan, Rajeshwari Pandharipande, James H. Yoon, Ladislav Zgusta, and 
Alessandro Zucchi 

AIM: SLS is intended as a forum for the presentation of the latest original 
research by the faculty and especially students of the Department of Linguistics, 
University of Illinois, Urbana-Champaign. Papers by scholars not associated with 
the University of Illinois are also welcomed. 

SPECIAL ISSUES: Since its inception SLS has devoted one issue each year to 
restricted, specialized topics. A complete list of such special issues is given on the 
back cover. 

EDITORIAL ADDRESS: Editor, Studies in the Linguistic Sciences, Department 
of Linguistics, University of Illinois, 4088 Foreign Languages Building, 707 
South Mathews, Urbana, IHinois 61801; Tel.: 217-333-3563; Fax: 217-333-3466; 
E-mail: sls@kahane.cogsci.uiuc.edu 

BOOKS FOR REVIEW: Review copies of books may be sent to the Editor. 

SUBSCRIPTION: Normally, there are two issues per year. Requests for sub- 
scriptions should be addressed to SLS Subscriptions, Department of Linguistics, 
University of Illinois, 4088 Foreign Languages Building, 707 South Mathews, 
Urbana, IHinois 61801. 

UPCOMING ISSUES: Vol. 23:2: Issues in Formal Semantics, Editor: Alessandro 
Zucchi; Vol. 24:1: Papers in General Linguistics. 



Price: $8.00 (per issue) 



STUDIES IN THE LINGUISTIC SCIENCES 

Papers in General Linguistics 



EDITOR 
Hans Henrich Hock .^^i 

EDITORIAL ASSISTANT u:^-, ■ ' ■' •-' 



Sara L. Michael 



VOLUME 23, NUMBER 1 
SPRING 1993 



DEPARTMENT OF LINGUISTICS, UNIVERSITY OF ILLINOIS 
URBANA, ILLINOIS 61801 



CONTENTS 



Preface 



Niken Adisasmito: Syllable structure and the nature of schwa 

in Indonesian 1 

Paul Agbedor: Verb serialization in Ewe 2 1 

Martin J. Baik and Rosa Shim: Yes, we have no bananas: 

English negative tags in cross-linguistic communication 4 3 

Elabbas Benmamoun: The status of agreement and the agree- 
ment projection in Arabic 6 1 

Marvin K. L. Ching: Examining the trustworthiness of the latest 

OED in reflecting current English 7 3 

Abdul Aziz Diop: Language planning across political boundaries: 

A case study of Pulaar 8 3 

Andrew Tilimbe Kulemeka: Bimoraicity in monosyllabic 

Chichewa ideophones 107 

Sharon R. Morrison: A re-examination of cardinal vowels and 

auditory equidistance 117 



131 



Steven Schaufele: The Vedic clause-initial string and universal 
grammar 

Squib 

M. Lynne Murphy: Discourse markers and sentential syntax 163 

Reviews 

Anvita Abbi (1992). Reduplication in South Asian languages: 
An areal, typological, and historical study. (Hans Henrich 
Hock) 169 

Narindar K. Aggarwal (1991). Studies on Nepali language and 

linguistics: A bibliography. (Mithilesh K. Mishra) 193 

John Baldwin & Peter French (1990). Forensic phonetics. (Jose 

Ignacio Hualde) 195 

Recent Books 197 

Contents of volumes 18-22 201 



Preface 



Over the past several years the number of submissions from 
faculty and students not associated with the University of Illinois has 
greatly increased. While this increase is a welcome sign of the broad 
recognition that Studies in the Linguistic Sciences (SLS) has received 
in the field, it has also brought with it an increase in the time 
required for the editorial process. I apologize for the resulting delay 
in this issue of the journal. 

To improve communications with outside contributors, this issue 
adds an e-mail address and a fax number to the complete postal 
address of the editor. This information is provided on the inside of 
the front cover. 

Manuscripts may be submitted at any time. They should be 
DOUBLE-SPACED and adhere as closely as possible to the style sheet of 
Language, the journal of the Linguistic Society of America.. Accepted 
contributions will be returned to authors for any necessary revisions 
and for final formatting. The final version is to be formatted in ac- 
cordance with the SLS style sheet, which will be mailed along with 
the acceptance letter, and should be submitted, if possible, both in 
'hard copy' and on disk. The disk document should be in the most 
recent version of Microsoft Word for the Macintosh or IBM, on a 3.5" 
disk. 

As usual, I take great pleasure in acknowledging the help of a 
number of my colleagues who have refereed submissions for the 
present issue of Studies in the Linguistic Sciences: C.-C. Cheng, Laura 
Downing, Georgia Green, Braj B. Kachru, Yamuna Kachru, Chin-W. Kim, 
Rajoshwari Pandharipande, James Yoon, Ladislav Zgusta, and Ales- 
sandro Zucchi. 

The Department of Linguistics also is grateful for support from 
the College of Liberal Arts and Sciences toward publishing this issue, 
and technical support from the Language Learning Laboratory. Last, 
but not least, I would like to express my appreciation to Beth Creek, 
Cathy Huffman, and Eileen Sutton from the Department Office, and 
Sara Michael, my editorial assistant, for their help in preparing this 
issue. 

March 1994 Hans Henrich Hock (Editor) 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 

SYLLABLE STRUCTURE AND THE NATURE OF SCHWA IN 
INDONESIAN* 

Niken Adisasmito 

The phonological system of Indonesian allows more 
than one set of syllabification principles with regards to its 
treatment of borrowed lexical items. Schwas in borrowed 
lexical items are derived. In contrast, schwas in the native 
Indonesian lexical items are present in the underlying 
representation. 

1. Introduction 

The treatment of borrowed lexical items in a language has 
received a lot of discussion, such as by Hyman (1970), Kaye & Nykiel 
(1979), Silverman (1992), Ito & Mester (1993). In analyzing the 
phonology of the borrowed lexical items in a language, one needs to 
take into account the phonological system of the borrowing language. 
In the process of borrowing, the speaker tries to retain the form of 
the borrowed lexical items which is closest to that of the source 
language. At the same time, the (phonological) system of the 
speaker's language will 'determine' the output forms. 

In this paper, I discuss syllable structure and syllabification, as 
constraints that determine the output of borrowed lexical items in 
Indonesian. I also discuss the nature of schwa: whether it is part of 
the underlying representation or whether it is derived, by contrast- 
ing native Indonesian lexical items with bonowed ones. 

The claims of this paper are (1) that there is more than one set 
of syllabification principles in Indonesian, and (2) that schwa is 
underlying in the native Indonesian vocabulary, but inserted in the 
borrowed vocabulary. 

As mentioned in Hardjadibrata 1978, there have been various 
studies on Indonesian syllable patterns. These studies do not agree 
on what can constitute a syllable. For instance, according to 
Hardjadibrata, the 'Committee for Reforms in Indonesian Spelling' 
(CRIS) claims that the coda of a syllable can consist of a sequence of 
three consonants, while Kridalaksana lists two consonants as the 
maximum possible number for a coda. In this paper, I will discuss 
the possibility for a coda to consist of a sequence of consonants, and 
the constraints governing the occurrences of such codas. 

The data for this paper are based on my intuition as a native 
speaker of Indonesian. I consulted the dictionary of Echols & Shadily 
(1989), which has been most helpful in providing the alternations in 
the written representation of certain lexical items. These ortho- 
graphic alternations I believe reflect alternations in the pronuncia- 



2 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

tion of the borrowed lexical items. I further verified my observations 
regarding these alternations with two native speaker consultants. 

2. Theoretical framework 

In this paper, the template-based approach to syllabification 
(Ito 1986) is adopted. According to this approach, the properties of 
syllable structure fall out from the general prosodic principles. These 
principles, the settings of which are language-specific, are Maximal- 
ity, Prosodic Licensing, and Extraprosodicity. The prosodic structure 
of the syllable is further defined by syllable templates, the Sonority 
Principle, the Onset Principle, and the Coda Filter. 

These principles are sufficient to account for the data under 
discussion. 

3.0 Analysis of the data 

In the following sections, I will show the Maximal Syllable 
Template(s) of Indonesian (section 3.2) and the role of schwa in the 
syllabification of consonant clusters, based on the data in section 3.1. 
I further show that schwas are part of the underlying representation 
in native Indonesian vocabulary (section 3.3). Finally, I propose that 
there are variations in the syllabification principles among Indone- 
sian speakers; these variations, however, are not mutually contra- 
dictory, given certain language-external factors (section 3.4). 

3.1 Some relevant facts about Indonesian 

Hardjadibrata (1978) argues that there are eleven syllable 
patterns in Indonesian, as opposed to ten postulated by Kridalaksana, 
and thirteen by the CRIS. (See Hardjadibrata 1978 for further 
references.) These differences stem from disagreement on what can 
constitute a syllable, i.e., whether an onset or a coda can consist of 
more than one consonant. According to the CRIS, the maximum num- 
ber of consonants constituting a coda is three. There are, however, 
only two lexical items (see (1)) with such a coda, found in the 
Indonesian vocabulary. Ignoring these highly exceptional forms, I 
show in this section that Hardjadibrata's claim is correct. 

(1) VCCC: arts (from Dutch) 'physician' 
CVCCC: kOrps (from Dutch) 'corps' 

The syllable patterns in Indonesian are exhibited in (2) - (11) 
below, involving both native and foreign-origin words. 

In (2), it is shown that a syllable may consist of a single vowel, 
and such a syllable may be word initial or word final. 

(2) V pattern 

a-nak 'child' 

i-kat 'to tie' 

o-Iah 'to analyze' 

ba-u 'smell' 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 3 

A syllable may be onset-less, or coda-less, or may consist of a 
simple onset and a simple coda, as shown in (3) - (5). These syllable 
patterns can occur at word-initial, word-medial or word-final posi- 
tion. The data in (2) - (5) are representative of the native Indonesian 
syllable patterns (Grijns 1977). 

(3) CV pattern 

ku-rus 'skinny' 





bi-na-s a 


'to die' 




S9 -k a -rar) 


'now' 




la-m a 


'long (of time)' 


(4) 


VC pattern 






in-dah 


'beautiful (for scenery)' 




am-bun 


'dew' 




ar-ti 


'meaning' 




ra-i h 


'to reach' 




ha-u s 


'thirsty' 




ma-in 


'to play' 


(5) 


CVC pattern 






pas-ti 


'certain, exact' 




t a f)- k a s 


'skilled, fast' 




pal-su 


'fake' 




p u - s ig 


'headache' 




sio-g a h 


'to stop by' 



The syllable structure of borrowed lexical items is distinct from 
that of native ones in that it allows complex onsets and/or complex 
codas, which may consist of a sequence of two to three consonants 
(particularly the onsets). A two-consonant onset usually consists of 
an obstruent (voiced or voiceless) and a liquid (e.g. [tragis]), or a 
sibilant [s] and a voiceless obstruent (e.g. [spasi], shown in (6) - (9). 
However, as the alternative forms in (6) - (9) illustrate, there is an 
alternation in the pronunciation: either the consonants are main- 
tained as a cluster, or a schwa is inserted between them.' 

(6)CCV pattern 

sas-tra, sas-ta-ra (Sanskrit) 'literature' 

kre-dit, ka-re-dit (Dutch) 'credit' 

spe-syal, sa-pe-syal (Dutch) 'special' 

gra-tis, ga-ra-tis (Dutch) 'free of charge' 
sta-syUn, sa-ta-syUn, sa-ta-si-yUn (Dutch) 'station' 

spa-si, sa-pa-si (Dutch) 'space' 

(7)CCVC pattern 

do-brak, do-ba-rak 'to break in' 

blun-tas, ba-lun-tas 'a kind of plant' 

prak-tis, pa-rak-tis (Dutch) 'practical' 

stOp, sa-tOp (Dutch) 'stop' 

in-spek-si, in-sa-pek-si (Dutch) 'inspection' 



4 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

A two-consonant coda may consist of a voiceless obstruent and a 
sibilant [s], or a liquid and a nasal consonant, or a nasal consonant 
and a sibilant [s]. 

(8) (C)CVCC pattern 

kOm-pIeks, kOm-pl£k, kOm-pa-lek (Dutch) 'complex (housing)' 

In-deks, In-dek (Dutch) 'index' 

helm, he-lgm (Dutch) 'helmet' 

film, n-lam (Dutch) 'film' 

in-tern, in-te-ran (Dutch) 'internal' 

mo-dErn, mo-de-ran (Dutch) 'modern' 
trans-mi-gra-si, tran-sa-mi-gra-si (Dutch) 'transmigration' 

(9) VCC pattern 

Ons, On 'ounce' 

eks, ek (?) 'ex-' 

The alternation of /helm/ and /film/ differs from that of 
/kompleks/ and /indeks/. In the former case, either the consonants 
are maintained as a cluster at the coda position, or a schwa is insert- 
ed between them. In contrast, the consonant clusters at coda position 
in /kompleks/ and /indeks/ may be maintained or simplified by 
deleting the second member of the cluster. 

A three-consonant onset consists of a sibilant [s], a voiceless 
obstruent and a liquid [r], shown in (10) and (11). This kind of onset 
can occur at word-initial or word-medial position. (The lexical items 
with this kind of onset are borrowed from Dutch.) 

(10) CCCV pattern 

stra-tegi, sa-tra-tegi, sa-ta-ra-te-gi 'strategy' 

spre, S3-pre, sa-pa-re 'linen' 

(11) CCCVC pattern 

skrlp-si, sa-krip-si, sa-ka-rlp-si 'thesis' 

Strom, sa-trom, sa-ta-rom 'electrical current' 

in-struk-tur, in-sa-truk-tur, 'instructor' 
in-sa-ta-ruk-tur 

In summary, Indonesian syllable patterns seem to allow up to 
three consonants in a sequence within a syllable. These consonant 
clusters show alternations in their pronunciation, i.e. with schwas at 
word-initial and word-medial position. At word final position, the 
maximum number of consonants, in general, is two. 

3.2 Basic analysis 

Based on the syllable patterns and the alternations in the 
pronunciation of the lexical items, as presented in 3.1, the structure 
of the syllable in Indonesian is maximally CCCVCC, and minimally V. 
The latter case indicates that Indonesian syllables require at least a 
vowel to be well-formed, and that they adopt the 'Relative' Onset 
Principle. 2 When there is no prevocalic consonant available to be 
incorporated as an onset, the vowel itself is sufficient to form a 



Adisasmito: Syllabic structure and the nature of schwa in Indonesian 5 

syllable, with or without the presence of a coda. When there is a 
consonant preceding the vowel, however, the Onset Principle is over- 
ridden by the 'Universal Core Syllable Condition' (Ito 1986), shown in 
(12). 

(12) 'Universal Core Syllable Condition' 
* a a 

r\ I 

V C V 

This condition guarantees that an intervocalic consonant becomes the 
onset of the following vowel and not the coda of the preceding vowel. 

The complex onsets shown in (13), follow the Sonority Principle: 
The sonority of onsets increases towards the nucleus of the syllable, 
and the sonority of codas decreases from the nucleus (Selkirk 1982; 
Steriade 1982). This principle, however, is violated in (14) in that the 
sonority of the onset decreases towards the nucleus. In the language 
from which the lexical items in (14) originate, Dutch, complex onsets 
of this kind are licensed by assigning extraprosodic status to [s] at 
the edges of the syllable (van der Hulst 1984). 



literature' 
credit, loan' 
to break in' 
a kind of plant' 

scale' 

inspection' 

iron (from Dutch: strijken 'to iron') 

instructor' 



(13) sas-tra 
kre-dit 
do-brak 
blun-tas 

(14) ska-la 
in-spek - s i 
stri-ka 
in-struk-tur 

The fact that complex onsets are preserved in Indonesian shows that 
this onset structure is acceptable, at least for some speakers. 

The Maximal Syllable Template in Indonesian then, can be 
characterized as in (15). The condition governing the combinatorial 
possibility of the consonants is shown in the diagram in (16). Onset 
sequences of consononants violating the Sonority Principle (as in 
(14)) are allowed in that the [s] of such structures is extraprosodic; if 
the sequence of consonants respects the Sonority Principle, then the 
first segment is an obstruent (voiced or voiceless) and the second a 
liquid, [1] or [r]. 

(15) Maximal Syllable 1: OOCVOC 

(16) Onset Principle 
Ex a 



C 

I 
cont 
son 



C 



V 



[+ son] 





Ex o a 




1 r\ i\ 

C C V C V 

1 1 1 1 1 




1 1 1 1 1 
s k a 1 a 


CT 




C V 





6 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

The Maximal Syllable Template (15), together with the Onset 
Principle (16), accounts for (13) and (14), as shown in (17). 

(17) a a 
C V C C C V 

I I I I I I 

s a s t r a 
a Ex a 

V C C C V c 

I I I I I I I I 

i n s p E k s i 

The forms in (13) and (14), however, are accepted only by some 
speakers; for others, the only accepted forms are the alternations 
shown earlier (repeated in (18)). 

(18) S3-ka-la 
in-S9-pek- si 
S9-tri- ka 
in-sa-truk-tur 

For yet other speakers, the complex onsets shown in (13) and (18) 
are not tolerated either, even though they respect the Sonority 
Principle. Only the structures in (19) are well-formed for these 
speakers. 

(19) S9-ta-ri-ka 
in-S9-t9-ruk-tur 
ga-ra-tis 
pa-rak-tis 

This phenomenon suggests that the Maximal Syllable Template 
previously suggested, viz. CCCVCC,^ is not observed by all speakers. 
The Maximal Syllable Template for the forms in (18) is given in (20). 

(20) Maximal Syllable 2: OCVC 

Comparing the forms in (14) and those in (18), the extraprosodic 
segments of (14): (as in [skala]), are syllabified in (18) (e.g. [sakala]) 
in that they become the onset of a syllable followed by an inserted 
vowel, schwa. The syllabification of some of the forms in (18) is 
shown in (21). This insertion phenomenon indicates that in this 
variety of Indonesian extraprosodic elements are not allowed.** The 
Onset Principle (shown in (22)), governing the combinatorial possi- 
bility of the consonants thus, determines that if the onset of a 
syllable includes a sequence of two consonants, then the first 
segment is an obstruent (voiced or voiceless) and the second a liquid, 
[1] or [r]. 



(21) 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 

a a a 



a a a 

r^- K f\ 

C V C V C V 

I I I i I I 

s 3 k a 1 a 



C V C C V C V 

I i I I I I I 

s a t r i k a 



1^ r^- -"1^ r\ 

vccvcvccv 

i I I I I I I I I 

I n s 9 p £ k s i 



(22) Onset Principle 
a 



obstruent 
+/- voice 



C V (C) 

I 
[liquid] 



The principles of Prosodic Licensing (Ito 1986) require that all 
phonological units are licensed, either by syllabification or by 
extraprosody. If a segment at a sylllable edge is not extraprosodic 
and it is not syllabified, it will be deleted by Stray Erasure. A vowel 
slot is inserted to syllabify the [s], which otherwise would become a 
stray segment and would have to be deleted. The melodic material 
for the vowel is schwa, as in. Yoruba (Abaglo & Archangeli 1989) and 
Yawelmani (Archangeli 1984). 

The data in (19), [garatis], [paraktis], etc., show that in this 
variety of Indonesian the maximal syllable is CVC (23), implying that 
no tautosyllabic consonant cluster is allowed (24). 5 

(23) Syllable Maximal Template: CVC 

(24) Onset Principle 

*a[CCV 

To license the unsyllabified segments, a schwa is inserted, as shown 
in (25). The syllable template shown in (23) is consistent with Grijns's 
(1977) claim that the native Indonesian syllable pattern consists 
maximally of a simple onset, a nucleus, and a simple coda. 



(25) 



C V C V C V c 

I I I I I I I 

p 9 r a t is 



Tn Tn n n 

cvcvcv cv 

I I I I I I II 



s a t 



The fact that there are various syllable templates existing simul- 
taneously in the language suggests that different speakers have 



8 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

different parameter settings for the onset condition. In the following 
paragraph, I show that the Coda Condition also seems to have differ- 
ent parameter settings for different speakers. 

Complex codas in borrowed lexical items are maintained by 
some speakers, but not by others (cf. (8); the relevant information is 
repeated in (26)). 



(26) kOm-pIeks 


vs. 


kOm-plek 


helm 


vs. 


he-la m 


mo-der n 


vs. 


mo-de-ran 



Kager (1990) claims that in Dutch, a schwa-like vowel, with 
much shorter duration (relative to underlying schwa), 6 is inserted 
when consonant clusters are at syllable final position. The pronun- 
ciation of /helm/, as he indicates, is [hel^m], in which [^] signifies the 
inserted short schwa-like vowel. The presence of the schwa-like 
vowel may have been interpreted as a full schwa in the borrowing 
process of these lexical items — /helm/, /modern/ — into Indonesian. 
However, there is a clear difference, at least to some speakers, be- 
tween Dutch [hel^m] and Indonesian [helam]. This difference indicates 
that complex codas are not tolerated in the speech of some speakers. 
The alternation in the pronunciation of /kompleks/, i.e. [kOmpleks] 
and [kOmplek] also shows this intolerance of complex codas by some 
speakers. 

The Coda Condition which allows complex codas follows that of 
the source language (exhibited in (27)), i.e., the [s] is licensed as 
extraprosodic, and a nasal consonant following a liquid consonant is 
incorporated into the syllable forming a complex coda — since the 
resulting consonant cluster still observes the Sonority Principle (van 
der Hulst 1984). 

(27) Coda Condition 1:7 

a) extraprosodic [s] b) liquid-nasal cluster 

a Ex a 

(C) (C) V C s (C) (C) V C C 

[liquid] [nasal] 
The effect of the Coda Condition in (27) is exemplified in (28). 

(28) a o Ex o 

cvcccvcc cvcc 

I i I I I I I I I I I I 

kOmpleks film 

When no complex codas are tolerated, the Coda Condition licens- 
es only a single post-vocalic consonant to be a coda (29). In 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 9 

accordance with the principles of Prosodic Licensing, the unsyllab- 
ified consonant is either syllabified forming a new syllable, or the 
consonant remains and is deleted by Stray Erasure. When new 
syllable formation occurs, a schwa is inserted before the nasal and, 
observing the 'Universal Core Syllable Condition' in (12), the 
immediately preceding consonant ([1] in /film/, which now becomes 
intervocalic) is incorporated into the new syllable as its onset. Stray 
Erasure deletes the extraprosodic [s]. The syllabification is shown in 
(30). 

(29) Coda Condition 2: 
* VCC ]a 

(30) Syllabification of (28) according to Coda Condition 2:8 

cvcc CVCVC 

I I I I I I I I I 

helm ~^ he Igm 

CO o a 

CVCCCVCC CVCCCVC 

I I I I I I I 4 I I I I I I I 

kOmpleks-^ kOmplek 

The conditions set forth in (27) and (29) suggest that Indonesian 
allows different settings for its Coda Condition in the face of bor- 
rowed lexical items, i.e., one setting for complex codas, and another 
for simple codas, in addition to allowing different settings for its 
Onset Principle. 

In this section, 1 have discussed the maximal syllable templates, 
as well as the language-specific parameter settings for the Onset 
Principles and the Coda Conditions of Indonesian. I also have shown 
that some schwas are derived in Indonesian, as the result of the 
application of the syllabification principles to borrowed lexical items. 

3.3 The nature of schwa in Indonesian 

There have been two opposing claims about the nature of schwa 
in Indonesian: One is that schwa is derived, due to its predictable 
distribution (Cohn 1989; Grijns 1976); the other is that it is present 
in the underlying representation (Lapoliwa 1981). In this section I 
present evidence that schwa in the native vocabulary is in fact 
present underlyingly. 

The syllabification of borrowed lexical items discussed in section 
3.2 supports the argument that the distribution of schwa is 
predictable: Whenever a consonant cluster is not tolerated, a schwa is 
inserted to simplify the cluster. This argument seems to also be 
supported by the native Indonesian vocabulary, as shown by the 



10 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

example in (31). In these data, the Maximal Syllable Template is 
CCWC, and the consonant clusters at the onset position, respecting the 
Sonority Principle, consist of an obstruent (voiced or voiceless) and a 
liquid ([1] or [r]). An alternative pronunciation is shown in (32), 
where schwa appears to be inserted in order to break up the conso- 
nant cluster, and therefore, appears to be predictable (as argued by 
Cohn and Grijns). This alternative pronunciation, which is OPTIONAL, is 
usually determined by language-external factors, namely the 
formality of context and the rate of speech. For further discussion on 
language-external factors, see section 3.4. 

(31) brar)-kat 'to go' (32) ba-raq-kat 

cla-ka 'bad luck' ca-la-ka 

blun-tas 'kind of plant' ba-lun-tas 

krOn-cOo 'type of music genre' ka-rOn-cOo 

The predictable occurrence of schwa just shown, however, does 
not account for the presence of schwas in (33). If the presence of 
schwas in (33) were predictable, the forms in (34) should be the 
underlying representations of those in (33). 

(33) u-pa-ti 'tax' (34) *up-ti 

ga-ma-lan 'Javanese musical *gam-lan 

instrument' 

The application of the syllabification principles set forth in 
section 3.2, however, does not result in the surface forms, as demon- 
strated in (35). The main reason is that in these data all segments are 
exhaustively syllabified. Thus, there is no motivation for the schwa 
to be inserted here, since these forms are well-formed by the syllab- 
ic template proposed. One could conclude that the schwas in (33) 
must be underlying, and therefore that NOT all schwas are derived in 
the language. 

(35) o c a a 

vccv cvccvc 

I I I I I I I I I I 

*upti *gamlan 

However, one could try to salvage the argument that all schwas 
are derived by proposing a filter blocking the structures in (35) for 
other, non-templatic reasons. One may argue that Indonesian has an 
adjacency restriction on consonants that two adjacent stops are ruled 
out (hence *upti), and that a (labial) nasal cannot be adjacent to a 
liquid [1] (hence *gamlan), even when they are not tautosyllabic. 

But these filters are not supported by other data, as shown in 
(36). The data in (36) include forms where adjacent stop consonants 
and adjacent nasal/liquid sequences are well-formed. 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 1 1 

(36) sak-ti 'invincible' jum-lah 'total of addition' 
sap-tu 'Saturday' kim-lo type of dish' 
buk-ti 'evidence' im-lah 'dictation' 

Since the adjacency restrictions which were proposed to salvage 
the schwa-as-derived argument are themselves not valid, it does not 
seem possible to salvage this line of reasoning. Since schwa is there- 
fore not derived in (33), it must be underlying. 

Forms shown in (37) provide further evidence that some schwas 
MUST be underlying. If all the schwas in (37) were derived, then the 
forms in the right column without a schwa should be the underlying 
representations. 

(37) ca-ma-ti <- cmti 'whip' 
sa-ka-jap <— skjap 'in an instant' 
ka-ma-lUt <- kmlUt 'chaos' 

Assuming that these are the correct underlying forms, applying 
the syllabification principles argued for in 3.2 would result in incor- 
rect surface forms, as demonstrated in (38). Given that the language 
tolerates codas, there is no reason why, for example, *camti should 
require further syllabification as camati. 



(38) 




<5 <5 G a 

CVCCV cvccvc 

I I I I I I I I I I I 

*camti *sa kda r 

Since the correct distribution of schwa cannot be achieved based 
on the underlying forms proposed in (37) where schwa is not present 
underlyingly, these forms cannot be correct. Therefore it must be the 
case that these schwas are present underlyingly, as shown in (39). 

(39) 

<5 O a (3 <3 C5 

K N /T\ IN N /N 

cvcvcvc cvcvcvc 

I I I I I I I I I I I I I I 

sakadar kamalut 

The data in (33) and (37) have shown that schwa cannot be 
derived in at least these forms. There are three possibilities for the 
distribution of schwa in the native vocabulary: (1) all schwas are 
derived; (2) some schwas are underlying, while others are derived; 
(3) all schwas are underlying. Since the data above have indicated 
that (1) is not true, the choice is between (2) and (3). On theoretical 
grounds, it is undesirable that some schwas should be underlying 
and others derived, since there is no phonotactic evidence that they 
are distinct. 



a 

N 

C V 

1 1 


a 

N N 

C V C V 
1 1 1 1 


I 1 

C a 


I'll 
m a t i 



12 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

I propose, therefore, that ALL schwas in the native Indonesian 
vocabulary are part of the underlying representation. Under this 
view, the forms in (32): baraijkat, calaka, etc., are the underlying 
representations of the forms in (31): braijkat, claka, etc., and not vice 
versa, as proposed for the schwa-as-derived approach above. 

This, in turn, indicates that a schwa may optionally be deleted, 
provided that the resulting consonant cluster respects the Sonority 
Principle. This is the case with the forms in (31), repeated in (40), 
but is not the case in (41), where deletion of schwa results in 
consonant clusters which violate the Sonority Principle. Therefore the 
application of schwa deletion is ruled out. 

(40) baraokat -^ brar)kat 
baluntas -^ bluntas 

(41) jambatan — > *jmbatan 'bridge' 
candawan — > *cndawan 'mushroom' 

In conclusion, schwas in the native Indonesian vocabulary are 
present underlyingly, and are optionally deleted, ^ if and only if the 
resulting forms respect the Sonority Principle. 

3.4 Variations of the UR in Indonesian 

In this section, I propose that Indonesian allows more than one 
set of syllabification principles to exist simultaneously. 1 also discuss 
another dimension of the pronunciation alternations of borrowed 
lexical items, language-external factors and their interaction with the 
different sets of syllabification principles. 

As discussed in section 3.2, it seems that Indonesian allows 
different maximal syllable templates. Onset Principles, and Coda 
Conditions with regards to borrowed lexical items. Based on this 
discussion, I divide Indonesian into different varieties, which I call 
Variety A, Variety B, and Variety C. 

Variety A represents that form of Indonesian which allows 
complex onsets and complex codas in its Maximal Syllable Template: 
CCCVCC (shown earlier in (15)). The combinatorial condition govern- 
ing the occurrences of the consonant clusters is determined by the 
Onset Principle shown in (16) and the Coda Condition 1 shown in 
(27). The Onset Principle determines that a sequence of consonants is 
allowed to occur at the onset position, in that if the first segment is 
an [s] and if the following segment is less sonorous than [s], then [s] is 
extraprosodic; and if the sequence of consonants respects the 
Sonority Principle, then it is a sequence of an obstruent and a liquid. 
The Coda Condition determines that (1) if [s] is the second member of 
a consonant cluster at coda position, it is extraprosodic, and that (2) a 
nasal consonant following a liquid consonant at coda position is 
tautosyllabic. The syllabification principles of VARIETY A are shown 
in (42). 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 



13 



(42) Syllabification principles of VARIETY A 

- Maximal Syllable Template: CCCVCC 

- Onset Principle: 

Ex a 

I 



c c 

1 1 


C V 

1 


1 1 
^ - cont 


1 
[+ son] 


- son 




Coda Conditio 
extraprosod 
a 


i: 

ic [s] 
Ex 



liquid-nasal cluster 
a 



(C) (C) V C s 



(C) (C) V C 



[liquid] [nasal] 

Variety B allows complex onsets in its Maximal Syllable Tem- 
plate: CCVC (shown earlier in (18)). The consonant sequence allowed 
to occupy the onset position is an obstruent and a liquid (shown in 
(22)). The governing coda condition is Coda Condition 2, shown in 
(29), in which no consonant clusters are allowed. The syllabification 
principles characterizing VARIETY B are shown in (43). 

(43) Syllabification principles of VARIETY B 

- Maximal Syllable Template: CCVC 

- Onset Principle: 

o 



C 



C V (C) 



1 [liquid] 



[obstruent 
[+/- voice 

- Coda Condition: * VCC]a 

Variety C has the 'simplest' Maximal Syllable Template relative 
to the other two varieties: CVC. It has Coda Condition 2 governing its 
syllable structure. The syllabification principles of VARIETY C are 
shown in (44). 

(44) Syllabification principles of VARIETY C 

- Maximal Syllable Template: CVC 

- Onset Principle: * a [CCV 

- Coda Condition: * VCC]o 



14 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

These three varieties are by no means mutually exclusive. Some 
speakers may choose to use one variety for one kind of situation and 
another for a different situation. What this means is that depending 
on the situation, a speaker may apply different syllabification 
principles with regards to borrowed lexical items. 

In the following paragraphs, I discuss the kinds of situations 
which determine the contexts where speakers apply different syllab- 
ification principles. 

Grijns (1977) observes that in the Indonesian dialect of Jakar- 
ta'O there are variations in the pronunciation of borrowed lexical 
items, particularly those involving consonant clusters. The factors 
determining these variations seem to be the rate of speech, the 
register used (formal or informal), and to what extent the speaker is 
exposed to the source language. This last factor is also discussed by 
Onn (1976) for Johore Malay, where the constraint on the consonant 
clusters st-, sk-, sm- is violated when the speakers are familiar with 
the source language, or when the speakers '... aspire to [a] more 
"sophisticated" type of speech ..." (58). 

In my own speech, with regards to the lexical items previously 
discussed, the less formal the context of the conversation is, the more 
likely it is that the schwa will be heard. This is also supported by my 
consultants in that, in our conversation, the schwas in some of the 
lexical items shown earlier were consistently present, e.g. [fllam], 
[S9t0p], [s9trom],ll [modergn], [sakripsi], [satrika], etc. 

The rate of speech also proves to be significant in the occurrence 
of schwas in Indonesian, as I observed from my own speech, as well 
as that of my consultants. The faster the speech, the more likely the 
schwa is to be dropped. This phenomenon is also observed by Grijns, 
based on his survey of the lenong show which involves reading a 
dramatic text in very fast and informal speech and in which, speak- 
ing loudly seems to be necessary (Grijns 1977). Speaking loudly 
seems to have the same effect as speaking slowly in deliberate 
speech, as noted by Grijns, in that the occurrence of consonant 
clusters is low, i.e., the occurrence of schwa is high. 

It seems to be the case for Johore Malay, for the Indonesian 
dialect of Jakarta, and for Indonesian in general, that the more the 
speakers are familiar with borrowed lexical items, the more likely 
the consonant clusters are to be retained. 

Based on Grijns' observation of the three factors, the diagram 
shown in (45) shows the LIKELINESS of the presence of schwa in the 
pronunciation of borrowed lexical items. 

The likeliness of the presence of schwa in this case is to be 
understood in conjunction with the syllabification principles relevant 
to each variety. For instance, speakers of VARIETY A — assuming that 
the source language substantially influences them, in that they are 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 
(45) Diagram of [a] occurrence in borrowed lexical items 



15 



most exposure to 
source language 


least exposure to 
source language 


fast speech 


slow speech 


formal speech 


informal speech 



schwa is 
not present 



schwa is 
present 



able to use the most complex syllable template — may use the alter- 
native form closest to that of the source language during the formal 
speech. Thus, no schwa is present (e.g. [struktur]). In contrast, in the 
same formal context, a speaker of VARIETY B or VARIETY C will use 
the form with a schwa inserted, because the Onset Principle in these 
two varieties does not permit a segment to be extraprosodic. The 
result is the surface forms [S9truktur] and [sgtgruktur], respectively. 

In contrast to the conditions for the occurrence of schwa in 
BORROWED lexical items, the schwa in the NATIVE vocabulary is more 
likely to be present in slow speech, and when the context is formal 
(one obviously has maximum exposure to one's native language), as 
indicated in diagram (46). Thus, for the forms shown earlier in (31) 
and (32), [baraokat] is more likely to be used on a formal occasion, as 
well as in slow speech, and [bragkat] is used under informal and/or 
fast-speech conditions. 

(46/ Diagram of the [a] occurrence in native lexical items 



fast speech 



slow speech 



informal speech 



formal speech 



schwa is 
not present 



schwa is 
present 



4.0 Conclusion 

I have argued that the phonological system of Indonesian allows 
more than one set of syllabification principles to exist simultaneously 
in the borrowing process of lexical items from another language. The 
different syllabification principles have different maximal syllable 
templates, and different settings of the Onset Principle and the Coda 
Condition. I have also pointed out that the varieties characterized by 
the three sets of syllabification principles do not necessarily identify 
distinct speech communities, but may be used by a single speaker 
under different conditions. Schwas are derived in borrowed lexical 
items, as a means to incorporate these words into Indonesian (i.e. in 



16 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

the syllabification of consonant clusters). They are, however, present 
in the underlying representation in native Indonesian vocabulary. 

Further investigation is certainly needed to determine which 
other phonological and morphological constraints borrowed lexical 
items are subjected to in Indonesian. For instance, the complex affix- 
ation system in Indonesian may provide evidence for whether or not 
borrowed lexical items behave the way native ones do. 

NOTES 

* I am grateful to Jennifer Cole for her supervision during the 
initial writing process of this paper. Many thanks also go to Abby 
Cohn and Laura Downing for their suggestions and comments, some 
of which are incorporated in this paper, as well as to other professors 
and colleagues in the department of Linguistics who discussed the 
ideas with me. I would also like to thank Johannes Kabu and Herry 
Sutanto for their help in verifying the data. Of course, all errors are 
solely my responsibility. 

1 Donors of borrowed lexical items in Indonesian are Dutch, 
English, Arabic/Persian, Chinese, and Portuguese. It is coincidental 
that the majority of the data in this paper are of Dutch origin. For 
discussion on borrowed lexical items in Indonesian, see Spitzbardt 
1970, Lowenberg 1983, Verhaar 1984, de Vries 1988, etc. 

2 In contrast, a language may choose to require the 'Strict Onset 
Principle', in that a well-formed syllable must have an onset (Ito 
1989). 

3 At this particular point, the Coda Condition in Indonesian is not 
yet discussed. 

4 It seems to be a strain for some speakers to pronounce word- 
medial consonant clusters, as in instruksi, transmigrasi, etc. Some- 
times the [s] is completely deleted, resulting in intruksi, tranmigrasi, 
etc. A lot of speakers are quite familiar with these two words due to 
the socio-political context. The [s] deletion in this case may indicate a 
tendency in which [s] in the middle of consonant clusters: -nstr- , 
-nsm-, is not syllabifiable. 

5 Even though schwa insertion is obligatory for the speakers 
who accepts only the forms in (19), it is never obligatorily absent for 
other speakers. Its absence, however, may be preferable in formal 
situations. See section 3.4. for discussion on language-external 
factors. 

6 The occurrence of full schwas in the final syllable of Dutch 
words is shown in the following (Kager 1990): 

kolOna 'column' ritma 'rhythm' 

Ortnar 'file' kataloxas 'catalogs' 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 1 7 

1 It may be argued that the Maximal Syllable Template in (18) 
allows complex codas: CCVCC, governed by this Coda Condition. While 
this may be true in Indonesian, further research needs to be done. 
For this paper, I assume that Coda Condition 1 governs the Maximal 
Syllable Template in (15): CCCVCC, and not in (20): CCVC. 

8 The syllabification of /film/ and /kompleks/ seems to be 
problematic. Taking into account the directionality of, say, CCVC- 
syllable template mapping, if the syllabification is R^L, the resulting 
surface form is (i), in which case the surface form of /kompleks/ is 
ill-formed. 

(i) o a <5 (5 

N N K /N 

film -> f I 1 8 m 

O O (5 O <3 (5 

/N /K N /1\ /K /r\ 

kOmpleks -^ *kOmplekas 

If the syllabification is L^ R, as in (ii), the surface forms of both 
/film/ and /kompleks/ are incorrect. 

(ii) GO <5 (5 

film ^ *f I 1 m a 

<5 G <5 COO 

kompleks -> *kompleksa 

A possibility to account for this phenomenon is to assume that 
[s] in /kompleks/ is not present in the underlying representation of 
some speakers, as suggested by Cohn (personal communication). This 
suggests that the underlying representation may vary among differ- 
ent speakers, as observed for Hindi by Ohala (1974). 

9 Cohn (personal communication) has suggested that while this is 
true historically, there has been reanalysis synchronically: many of 
the optional deletion cases have been reanalyzed as insertion cases, 
although she points out that this needs to be tested through psycho- 
linguistic experiments. 

'0 This particular dialect of Indonesian is influenced by Dutch, 
Chinese, Javanese, Sundanese, and various regional languages of In- 
donesia. This is partly due to Jakarta being the administrative center. 

' ' All consultants felt that it would be more natural to 
pronounce [satOp] or [satrom] in general, even though they could pro- 
nounce those two words as [stOp] and [strom], when asked. This may 
show that monosyllabic words are avoided, when possible, in Indo- 
nesian. Grijns notes that, according to M. Zain's Kamus moderen 



1 8 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

Bahasa Indonesia, 93% of Indonesian words are bi- and trisyllabic, 
and only 7% are monosyllabic or longer than three syllables. 

REFERENCES 

Abaglo, p., & D. Archangeli. 1989. Language particular underspecifi- 
cation: Gengbe /e/ and Yoruba I'll. Linguistic Inquiry 20.457- 
480. 

Archangeli, Diana. 1984. Underspecification in Yawelmani phonology 
and morphology. MIT Ph.D. dissertation in Linguistics. 

COHN, Abigail. 1989. Stress in Indonesian and bracketing paradoxes. 
Natural Language and Linguistic Theory 7.167-216. 

DE Vries, J. W. 1988. Dutch loanwords in Indonesian. International 
Journal of Society and Languages 73.121-136. 

Echols, John, & Hassan Shadily. 1989. Kamus Indonesia Inggris. 
Jakarta: PT Gramedia. 

Grijns, C. D. 1977. Jakartan speech and Takdir Alisjahbana's plea for 
the simple Indonesian word form. Studies in Austronesian 
linguistics, ed. by S. A. Wurm, 1-34. Ann Arbor: University of 
Michigan. 

Hardjadibrata, R. 1978. Consonant clusters in Indonesian. Second 
International Conference on Austronesian Linguistics: Proceed- 
ings, ed. by S. A. Wurm, 165-180. Canberra: The Australian 
National University. 

Hyman, Larry. 1970. The role of borrowing in the justification of 
phonological grammars. Studies in African Linguistics 1:1.1-48. 

ITO, Junko. 1986. Syllable theory in prosodic phonology. University of 
Massachusetts Ph.D. dissertation in Linguistics. 

. 1989. A prosodic theory of epenthesis. Natural Language and 

Linguistic Theory 7.217-259. 

, & Armin Mester. 1993. Japanese phonology constraint domains 

and structure preservation. To appear in: A handbook of phono- 
logical theory, ed. by John Goldsmith. Blackwell: Handbooks in 
Linguistics Series. 

KaGER, Rene. 1990. Dutch schwa in moraic phonology. Papers from 
the 26th Regional Meeting of the Chicago Linguistic Society 2: 
The Parasession on the Syllable in Phonetics and Phonology, 
241-255. 

Kaye, Jonathan, & Barbara Nykiel. 1979. Loan words and abstract 
phonotactic constraints. The Canadian Journal of Linguistics 
24:2.71-93. 

Lapoliwa, Hans. 1981. A generative approach to the phonology of 
Bahasa Indonesia. (Pacific Linguistics: Series D, No. 34.) Canberra. 

LOWENBERG, Peter H. 1983. Lexical modernization in Bahasa Indonesia: 
functional allocation and variation in borrowing. Studies in the 
Linguistic Sciences 13:2.73-86. 

Oh ALA, Manjari. 1974. The abstractness controversy: experimental 
input from Hindi. Language 50.225-235. 



Adisasmito: Syllable structure and the nature of schwa in Indonesian 1 9 

Onn, Farid Mohamed. 1976. Aspects of Malay phonology and 
morphology. University of Illinois at Urbana-Champaign Ph.D. 
dissertation in Linguistics. 

Selkirk, E. 1982. The syllable. The structure of phonological repre- 
sentations, ed. by H. van der Hulst & N. Smith, 337-383. 
Dordrecht: Foris Publications. 

Silverman, Daniel. 1992. Multiple scansions in loanword phonology: 
evidence from Cantonese. Phonology 9.289-328. 

Spitzbardt, H. 1970. Sanskrit loan words in the Bahasa Indonesia. 
Beitrage zur Linguistik und Informationsverarbeitung 9.62-79. 

Steriade, Donca. 1982. Greek prosodies and the nature of syllabifi- 
cation. MIT Ph.D. dissertation in Linguistics. 

VAN DER HULST, Harry. 1984. Syllable structure and stress in Dutch. 
Cinnaminson, N.J.: Foris Publications. 

Verhaar, John W. M. (ed.). 1984. Towards a description of contempo- 
rary Indonesian: preliminary studies, part II, in NUSA: Linguistic 
Studies of Indonesian and Other Languages in Indonesian 19. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



VERB SERIALIZATION IN EWE 

Paul Agbedor 
Department of Linguistics, University of Victoria 

This paper examines serial verbal constructions in Ewe, 
a Kwa language of West Africa, in the light of the recent 
treatment of such phenomena in Yoruba by Baker (1989). 
Some of the structural characteristics of serial verbal 
constructions (SVCs) are examined. The treatment of SVCs 
in Yoruba is discussed. Some problems with the framework 
suggested by Baker are pointed out. The main area of 
concern in this paper is the object sharing phenomenon, 
which Baker assumes to be obligatory for 'true' SVCs. In 
this paper, I argue that what Baker calls 'covert coordina- 
tion' constructions are, in fact, 'true' SVCs in Ewe. As a 
result, the object sharing phenomenon is not always obliga- 
tory. The result is that the syntactic framework proposed 
for Yoruba would not work for Ewe. An alternative proposal 
is made to address the discrepancy. The idea is to have a 
framework that can account for both the object-sharing or 
'argument-sharing' and non-argument-sharing types of 
SVCs. 

1.0 Introduction 

Serial verbal constructions (SVCs) are a phenomenon commonly 
found in the Kwa languages of West Africa and Caribbean Creoles. 
They are also reported in Chinese (Li & Thompson 1973) and in 
Burmese (Matisoff 1973).^ In this paper, I will examine SVCs in Ewe 
(a Kwa language of Ghana) in the light of the principles of 
Government and Binding Theory. The paper will specifically examine 
the syntactic framework proposed by Baker (1989), and the 
problems with that framework will be outlined. I will argue, for 
instance, that the object-sharing phenomenon, which Baker (1989) 
suggests is obligatory for 'true' SVCs, is not found in all cases (at least 
for Ewe). I will also argue that certain SVC structures in Ewe pose a 
problem for Baker's model, with respect to the Projection Principle, 
and I will suggest an alternative framework for SVCs. The paper is 
organized as follows: In the next section, I discuss the major charac- 
teristics of SVCs that distinguish them from other structures in Ewe. 
In section 2.0, I discuss Baker's proposal in some detail, noting the 
problems with his framework in section 4.0. I show that Baker's 
proposal does not account for all the SVC types in Ewe, and in section 
5.0, I make an alternative proposal. 



22 



Studies in the Linguistic Sciences 23:1 (Spring 1993) 



2.0 Characteristics of SVCs 

One of the early linguists who hinted at the notion of SVCs was 
Westermann, in his 1930 Ewe grammar: 

A peculiarity of Ewe is that we often find a row of 
verbs one after the other. The chief features of this are that 
all the verbs stand next to each other without being con- 
nected, that all have the same tense or mood, and that in the 
event of their having a common subject and object, these 
stand with the first, the others remaining bare (1930:126). 

Baker (1989) describes SVCs as a construction in which a sequence of 
verbs appears in what seems to be a single clause (513). According to 
him, there is usually one tense/aspect specification for the whole 
chain of verbs. The verbs in a SVC are also believed to have a single 
structural subject and they share logical arguments. The following 
are some examples of SVCs. 

(1) Yoruba: Aje sunkun lo ile 

Aje weep go home 

'Aje wept on his way home.' (Awoyale 1989) 

(2) Haitian: Emil pran liv la bay Mari 

E take book DET give M. 

'Emil gave the book to Mary.' (Dechaine 1988) 

(3) Sranan: Kofi naki Amba kill 

Kofi hit Amba kill 

'Kofi struck Amba dead.' (Baker 1989) 

(4) Akan: Kofi too bayire dii 

Kofi bought yam ate 

'Kofi bought yam and ate.' (Campbell 1991) 

(5) Ewe: Kofi d,a nu (lu 

Kofi cook thing eat 
'Kofi cooked and ate.' 

One powerful test that has been developed over the years to 
distinguish SVCs from coordinate and purposive constructions is WH- 
extraction. If the NP argument of a verb in a SVC can be extracted by 
WH-movement, it follows that the structure cannot be a coordination 
or an embedded purpose or result clause (see Ross 1967). This 
follows from the Coordinate Structure Constraint, which restrains 
extraction from either conjunct of a coordinate construction. Consider 
the following examples in Ewe. 

(6) (a) Kofi eta nu ctu 

Kofi cook thing eat 
'Kofi cooked and ate.' 
(b) Nuka Kofi eta clu? 

Thing-which K cook eat 
'What did Kofi cook and eat?' 



Agbedor: Verb serializatiDii in Ewe 23 

(7) (a) Kofi sle agbale na Ania 

Kofi buy book give Ama 
'Kofi bought a book for Ama.' 
(b) Nuka Kofi sle na Ama? 

Thing-which K buy give Ama 
'What did Kofi buy for Ama?' 

In (6) and (7), the (b) examples are Wh-extractions of the NP 

arguments. In (7), there is an extra argument because of the 

presence of the 3-place predicate verb na (give). All the three NP 

arguments can undergo WH-extraction, showing that they are not 
coordinate structures. 

The verbs in a SVC form a complex VP with a single-event 
interpretation. One test that proves this for Ewe is negation. In Ewe, 
the negative marker is a discontinuous element, me ... o, and the 
negated constituent lies between the two elements. The first element 
is considered to be the head, and is always attached to the verb. If 
the verbs in a SVC were to have multiple-event interpretation, we 
would expect the negative marker to be attached to each verb that 
represents a separate event. This is found not to be the case for Ewe 
SVCs, showing that they represent single events. Consider the 
ungrammaticality of the (c) and (d) examples in (8). 

(8) (a) Mesle agbale na Ama 

Isg. buy book give Ama 
'I bought a book for Ama' 

(b) Nye mesle agbale na Ama o 
Isg. NEG-buy book give Ama NEC. 
'I did not buy a book for Ama' 

(c) *Nye mesle agbale me-na Ama o 
Isg. NEG-buy book NEG-give Ama NEG. 

(d) *Mesle agbale me-na Ama o 
Isg. buy book NEG-give Ama NEG. 

3. Syntactic representation of SVCs 

One of the most challenging phenomena in SVC analysis, 
according to Baker (1989), is the notion of object sharing. Stewart 
(1963) in analyzing Twi SVCs suggested that SVCs formed out of a 
sequence of two transitive verbs show an object deletion under iden- 
tity (i.e. one of the objects is deleted (normally that of V2) because it 
is identical to the object of VI)). So for example, the Ewe example in 
(9) below will be derived from (10) by the traditional generative 
transformational rule of Equi-NP Deletion. 

(9) Kofi so Ama wu 
Kofi beat Ama kill 

'Kofi beat Ama to death.' 

(10) Kofi so Ama wu Ama 
Kofi beat NP kill NP 



24 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

With the demise of transformations as they prevailed in early 
generative transformational grammar, one option that readily comes 
to mind is to posit a D-structure in which the two verbs in the SVC 
take an object NP either to their right or left, depending on the type 
of language. 

(11) [ VP [vi beat [ v2 kill [np Ama ]]]]. 

In the above structure, only the V2 directly theta-marks the NP. 
The VI does not case-mark the NP because the adjacency condition is 
violated. One alternative is to move the NP to a position between the 
two verbs. In this case the VI directly theta-marks the NP. But how 
can we account for the theta-marking and case-marking properties 
of the V2, which is a transitive verb, and selects an NP object to 
which it may assign case? 

Baker (1989) suggests a framework for analyzing SVCs in 
Yoruba. He proposes that the NP that comes between the two verbs 
in the SVC is a shared object, in that it occupies a position which is 
theta-marked by both verbs (or their projections). Thus under 
Baker's analysis, (11) above will be assigned the structure in (12). 

(12) s 

/ \ 

NP VP 
I I 

Kofi V 
I 

V 

/ I \ 

V NP VI 

^1 I I 

SO Ama wu 

Baker assumes that SVCs are double-headed — that the serial verbs 
jointly constitute a complex predicate. In the above structure, 
therefore, the VP is double-headed and the NP it contains is 
governed by both verbs. From the structure in (12) above, the theta- 
marking of the NP within the VP by the VI is straightforward, but 
the notion of V2 also theta-marking the same NP might seem doubt- 
ful. To account for this. Baker invokes the standard conditions on 
theta role assignment from Chomsky (1986), which are stated as 
follows. 

(13) a may theta-mark (3 iff: 

(a) a and p are structural sisters. 

(b) a projection of a is a structural sister of (3 

Condition (a) allows for theta-marking of the NP by VI while 
condition (b) allows for theta-marking of the NP by V2, whose pro- 
jection is a structural sister to the NP. Under Baker's analysis, theta- 
marking of the external argument is achieved by invoking Williams's 
(1984) notion that the external (argument) role of the verb 



Agbedor: Verb serialization in Ewe 25 

percolates to its maximal projection. Since VP in the structure in (12) 
is the maximal projection of both VI and V2, the external theta role 
of both verbs percolate up to it, where they are assigned to the 
subject by condition (13b) (Baker 1989:520). So the lexical theta role 
assignment properties of both verbs are satisfied and the Projection 
Principle is obeyed. In the rest of this section I examine how Baker's 
framework can handle the various types of SVCs in Ewe. 

The example in (12) above involves two transitive verbs with a 
shared NP. Each of the verbs is a 2-place predicate. There are exam- 
ples in which one of the verbs is triadic (i.e. a 3-place predicate). This 
type can fit into Baker's framework. We only have to expand the last 
V into a V and NP as in (14). 

(14) 



/ 


\ 






NP 

1 


VP 
1 






1 
Kofi 


1 

v 

1 

V 






/ 


1 


\ 




V 


NP 


v 




/ 


1 


1 


\ 


tso agbale 


V2 

1 


NP 
1 






1 

na 


1 
Ama 



In (14), the shared NP is agbale 'book'; and the V2, a triadic na 'give', 
takes an additional argument Ama. VI theta-marks the NP agbale by 
(13a) and V2 theta-marks the same NP by (13b). The V2 theta- 
marks its additional argument by (13a). 

There is a type of SVC which differs from the one in (14) in that 
the V2 takes a PP complement instead of the NP in (14). This type is 
also successfully represented in the framework under discussion, as 
shown in (15). 

(15) 



s 






/ \ 






NP VP 
1 1 






1 1 

:ofi v 






/I 


\ 




/ 1 


\ 




VI NP 

1 1 


V 
1 


\ 


tso agbal 


e V2 

1 


PP 
/ \ 




da 


cte kplo dzi 



26 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Now consider the sentence in (16) below: 

(16) Adela da tu wu xevi 
Hunter shoot gun kill bird 
'The hunter shot and killed a bird" 

In the SVCs represented so far, the two verbs have a common NP 
object which is theta-marked by both verbs. The verbs assign the 
same theta role to the NP. But in (16), the NP 'gun' between the two 
verbs receives two different theta roles from the two verbs THEME, 
from VI and INSTRUMENT from V2, under the standard conditions of 
theta-role assignment outlined in (13). Despite this assignment of 
different theta roles to the NP object, the sentence fits well into the 
syntactic frame we are considering. It is possible for the NP to 
receive two different theta roles, provided it is the same structural 
position that is involved (cf. Baker 1989:521). 

One characteristic of SVCs that comes out clearly at this stage is 
that they share at least one argument, and that this argument is not 
always the grammatical object of both verbs. 

A class of verbs in Ewe and Yoruba (Baker 1989) which raises 
some questions about their role in SVCs, is the so-called 
'bimorphemic' verbs which are made up of a bound verb and a noun 
complement referred to as bound verb complement (BVC). For 
example: 

(17) eta nu (19) dzi ha 
cook thing sing song 
'cook' 'sing' 

(18) no tsi (20) kpa ha 
drink water compose song 
'drink' 'compose' 

The controversy is whether the BVCs are syntactic objects of the 
verbs or whether they compound lexically with the verb root to form 
true intransitives. 1 suggest that the BVCs are syntactic objects to the 
verbs concerned in (17) - (20) above. Consider the following 
sentences. 

(21) E-da nu ctu 
3sg-cook thing eat 
He cooked and ate' 

(22) E-kpa ha dzi 
3sg. -compose song sing 

'He composed a song and sang' 

(23) E-ku tsi no 
3sg. -fetch water drink 

'He fetched water and drank' 

In the above examples (all grammatical and acceptable), the V2s 
have no object after them. Like most of the SVCs we have examined 
already, these examples have NPs which are theta-marked by both 



Agbedor: Verb serialization in Ewe 27 

verbs. If the verbs in (17) - (20) are true intransitives (as Baker 
assumes for Yoruba), then they should not share an argument, and 
we expect forms like (24) and (25). 

(24) *E-(ta nu: ctu nu. 
He-cook thing eat thing 

(25) *E-ku tsi no tsi 
He-fetched water drink water 

But these forms are not grammatical in Ewe. These ungrammatical 
sentences in (24) and (25) are quite different from the sentence in 
(16), where we have two verbs and two NP arguments. In (16), the 
two verbs share an argument, 'gun', but while this argument is the 
grammatical object of the first verb, it is not the object of the second. 
The second verb has its own object, xevi. In the ungrammatical (24) 
and (25), the two verbs share the same object, and since the first 
occurrence of the object is theta-marked by both verbs, there is no 
need for an overt NP object for the second verb. Moreover, other 
nouns can be substituted for the BVCs. Consider (26) and (27). 

(26) E-cta te du 
3sg.-cook yam eat 

'He cooked yam and ate' 

(27) E-no aha mu 

3sg. -drink wine be drunk 

'He got drunk by drinking wine' 

In the above examples, the nouns act as complements to the so- 
called bound verbs in grammatical SVCs. If other NPs can be 
complements of the bound verbs in question, then the BVCs in (17) - 
(20) are syntactic objects of the bound verbs. I, therefore suggest 
that the verb roots in (17) - (20) are real transitive verbs that 
subcategorize for NP objects, just like any other transitive verb in the 
language. It seems the syntactic properties of the types of verbs 
found in (17) - (20) in Ewe are similar to the corresponding examples 
in Yoruba, in which momi and jeun are both bimorphemic verbs 
made up of two morphemes as follows: 

(28) (a) mu + omi — > momi 

drink water 'drink' 

(b) je + oun > jeun 

eat something 'eat' 

The Yoruba sentence in (29) is ungrammatical, just like the Ewe (25): 

(29) *Mo bu omi mumi 

I pour water drink water 
'I poured water and drank' 

There are, however, some verb-noun pairs which can be said to 
form true intransitives in Ewe. These are shown in (30) - (33). 



28 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

(30) ku dzi (32) ve dame 

kill heart hurt stomach 

'annoy' 'annoy' 

(31)su du (33)tsi meg be 

run race remain behind 

'run' 'be late' 

This class of verb-complement pairs consists of more or less idiomat- 
ic expressions that behave like single verbs and true intransitives, in 
that no NP can occur in place of the BVCs in these examples. As 
intransitives they cannot take direct objects. They can, however, take 
other verbs in SVCs. 

(34) Devi-a su du dzo 
child-the run race go 
'The child ran away.' 

(35) Nufiala tsi megbe va suku 
Teacher remain behind come school 
'The teacher was late to school.' 

In these examples, the complements (BVCs) of the first verbs (VI) 
are not shared by the second verbs (V2), because the VI in each case 
is made up of the 'bound' verb root and the complement (BVC) to 
become an intransitive verb. 

Another class of verbs in some African languages (including Ewe 
and Yoruba (cf. Awoyale 1987)) is the class of morphologically 
complex transitive verbs which have been treated as some kind of 
serial verbs (Bamgbose 1982) or as a distinct class of their own and 
referred to as 'splitting verbs' (Awolobuyi 1969; quoted in Awoyale 
1988:21). Ewe has such examples as: 

(36) 



(38) 



The term 'splitting' is applied here to refer to the fact that the two 
verbs forming the complex can be 'split' by an intervening NP object. 
Those who hold this view regard the verb pairs as single lexical 
items. Others like Bamgbose (1982)2 regard them as relatively 
'frozen serial collocations'. I hold the latter view that these verb pairs 
form serial verb strings. They behave just like other serial verbs. If 
they are, then it follows that the two verbal elements forming the 
pair are syntactic lexical heads forming a complex VP. To show that 
they are lexical heads in a complex VP, we subject these pairs in 
SVCs to the tense/aspect test. Since TENSE or ASPECT is a functional 
element which is always attached to the verb in Ewe, 1 assume that 



xo 


se 


(37) 


cte fia 


receive 


hear 




remove show 


'believe' 






'introduce' 


bia se 




(39) 


d,3 kpo 


ask hear 






taste see 


'inquire' 






'taste' 



Agbedor: Verb serialization in Ewe 29 

this can serve as an indicator of the headedness of the verbs in the 
SVC. 

(40) (a) Kofi eta nu d-U 
K cook thing eat 
'Kofi cooked and ate.' 
(b) Kofi a-d,a nu a-d,u 

K FUT-cook thing FUT-eat 
'Kofi will cook and eat.' 

(41) (a) Kofi xo nya la se 

K receive word the hear 
'Kofi believed the message.' 
(b) Kofi a-xo nya la a-se 

K FUT-receive word the FUT-hear 
Kofi will believe the message.' 

As the (b) examples in (40) - (41) show, each of the verbs receives a 
FUTURE marker, proving that it forms a double-headed predicate. 
Example (40) involves the normal transitive verb while (41) involves 
the 'splitting' verbs under discussion. These 'splitting' verbs, though 
they are not single syntactic units, form single semantic units. They 
are a kind of fixed collocations, because the two verbs forming the 
pair in each case have a fixed semantic interpretation. 

4.0 The problem 

The examples of Ewe SVCs examined so far seem to be 
adequately accounted for by Baker's model. All involve transitive 
verbs Moreover, note that the VI takes only one argument. Now let 
us consider cases in which the VI takes an extra argument. 

(42) Kofi cte awua le ka dzi da cte xo me 

Kofi / remove / shirt / on / rope / top / put / LOC / room / in 

Kofi removed the shirt from the line and put it in the room.' 

(43) Kofi fo agbaleawo le xoa me da d,e gota 

Kofi / collect / book-thc-PI. / in / room / put / I.OC / outside 
'Kofi collected the books from the room and put them 
outside.' 

In the two examples in (42) and (43), the VI in each case assigns an 
additional theta role to the PP. Applying Baker's model to these 
examples, we would expect that the V2 would theta-mark the PP 
argument of VI. 

In the representation in (44), a projection of V2 is a sister to 
both the NP and the PPl. Note that Baker (1989) claims that the 
sharing of the NP by the two verbs is obligatory (527). That is, the 
two verbs should theta-mark the NP between them. Since the PPl in 
(44) above is an argument of VI and a structural sister to the 
projection of V2, we would expect that V2 should theta-mark the PP 
too; but it does not. This constitutes a violation of the Projection 



30 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

(44) s 



/ 


\ 




NP 
1 


VP 
1 




1 
Kofi 


1 

v 






/ 1 1 


\ 




/ /I 


\ 


/ 


' / 1 


\ 


/ 


/ 1 


\ 


VI NP PPl 


v 



I I I \ I \ 

III \ V2 PP2 
III \ I I \ 

cte awua le ka dzi da c^e xo me 

Principle and the standard conditions on theta-role assignment 
adopted by Baker. At the same time, the two sentences in (42) and 
(43) are not coordinate structures because they pass the Wh- 
extraction test. The following examples show that it is possible to 
extract an NP from any of the conjuncts. 

(42') Nuka Kofi cte le ka dzi da cte xo me? 

Thing-what / Kofi / remove / from / line / pul / in / room 
'What did Kofi remove from the line and put in the room?' 

Now let us consider examples of SVCs involving a transitive verb 
and an intransitive or two intransitives. 

(45) Kofi no tsi ku 
K drink water die 

'Kofi died by drinking water.' 

(46) Kofi tutu d,evia dze anyi 
K push child-the fall down 
'Kofi pushed the child and fell down.' 

(47) Xevia dzo dzo 
Bird fly go 

'The bird flew away.' 

In all the examples in (45) - (47), the V2 is intransitive and object- 
sharing does not apply. To account for these examples, Baker 
suggests that where the two verbs theta-mark the NP between them, 
the structure in (48a) is projected. The structure in (48b) which 
Baker terms 'covert coordination' is projected where only the first 
verb theta-marks the NP. 

According to Baker, the Projection Principle forces the V2 to theta- 
mark the NPl in (48a) above, but this requirement would not hold 
for (48b). He suggests that the V2 would not be able to theta-mark 
the NPl in the configuration in (48b) because the NP is not sister to 
V2 or any of its projections. He views the so-called 'overt coordina- 
tions' as a sequence of distinct events, whereas the true SVC is 



Agbedor: Verb serialization in Ewe 31 

(48) (a) s (b) s 

/ \ / \ 

NP VP NP VP 

I / \ 

V V V 

/ I \ / \ / \ 

VI NPl V VI NPl V2 NP2 

/ \ 
V2 NP2 

perceived as a single event. Tliis suggests that 'covert coordination' 
types are not true SVCs. 

This position, however, is not acceptable (at least for Ewe). The 
Ewe examples in (45) - (47) above all pass the WH-extraction and 
negation tests for SVCs: 

(49) (a) Kofi no tsi ku 

K drink water die 

'Kofi died by drinking water.' 

(b) Nuka Kofi no ku 
What K drink die? 
'What did Kofi drink and die?' 

(c) Kofi me no tsi ku o 

K NBG drink water die NEG 
'Kofi did not die by drinking water.' 

(50) (a) Kofi so d,evia si 

K beat child-the flee 
'Kofi beat the child and fled.' 

(b) Ameka Kofi so si? 
Person-which K beat flee 
'Who did Kofi beat and flee?' 

(c) Kofi me so (levia si o 

K NEG beat child-the flee NBG 
'Kofi did not beat the child and flee.' 

Since the sentences in (45) - (47) also pass the Wh-extraction test, it 
follows that they are not coordinate structures, but true SVCs. 

One other problem has to do with subject-sharing by the verbs 
in the SVC. As noted earlier, SVCs have the characteristic of having 
the same subject. While this is true for Ewe, Baker (1989) reports 
something different for Yoruba. The following example is taken from 
Baker (1989:529). 

(51) Olu ti omo na a subu 
Olu push child the fall 

'Olu pushed the child down.' 

This example is accounted for by Baker's model as far as object- 
sharing is concerned (i.e. the V2 theta-marks the NP between it and 
the VI). But instead of the two verbs sharing the subject 'Olu', the 
object of the first verb becomes the subject of V2. The verb 'fall' 



32 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

takes 'child' as NP within its second V-bar projection as its only 
argument (it is unaccusative). 

(52) s 

/ \ 

NP VP 
I 

V 
/ I \ 
V NP V 
I I I 

I I V 

I I I 

push child fall 
(AG,TH) (TH) (Baker, 1989:530) 

Baker suggests that this structure has two consequences for the 
intransitive V2: 

(a) V2 must be lexically capable of assigning an internal theta 
role; 

(b) its theta role must be assigned to the object of VI rather 
than to the subject of VI. 

From this example. Baker predicts that only an unaccusative type of 
intransitive verb can follow a transitive or unaccusative in a true 
SVC (i.e. V2 must always be unaccusative if it is intransitive). This 
prediction is proved wrong by example (50a) above, where the V2 
'flee' is unergative and has only an external argument. 

In the Ewe example in (46) which is similar to the Yoruba 
example in (51), the V2 does not theta-mark the NP object of VI, and 
the sentence has only one interpretation (i.e. Kofi pushed the child 
and Kofi fell down, not the boy, as in the Yoruba example). In the 
Ewe example, therefore, the subject-sharing phenomenon is 
preserved. For an Ewe equivalent to the Yoruba example in (51) 
(with the interpretation given by Baker), a pronominal third person 
singular coindexed with the object NP will have to precede the V2, as 
in the example below. ^ 

(53) Kofi tutu ctevia wo dze anyi 
K push child-the 3sg. fall down 
'Kofi pushed the child down.' 

So we see examples in which no object-sharing takes place but the 
sentences are triie SVCs. I, therefore, propose that all SVCs, whether 
object-sharing or non-object-sharing, should be regarded as true 
SVCs. This claim rejects the distinction drawn between true SVCs and 
'covert coordination'. 

One other problematic type of SVC for Baker is the one involving 
an unergative and an unaccusative. An example is (47), repeated 
below. 



Agbedor: Verb serialization in Ewe 33 

(54) Xevia dzo dzo 
bird-the fly go 
'The bird flew away.' 

As Baker observed in a note, this particular type of SVC raises a 
problem for the theta-criterion. 'Bird' receives theta-roles in two 
different positions: one as the subject of 'fly' and one as the object of 
'go'. It has been observed that the same structural position can 
receive more than one theta-role. But the case under examination 
involves two different structural positions, external argument 
position of 'fly' and internal argument position of 'go'. In a structure 
involving only an unaccusative, the base-generated object which is in 
a non-case assigning position has to move to subject position to 
receive NOMINATIVE case in order to satisfy the case-filter. 

(55) [ bird [v got]] 

But when the verb 'go' is combined with the unergative 'fly', a 
problem arises: The subject position is filled by the external 
argument of the unergative. How then do we account for the theta- 
role assignment by the unaccusative verb? 

5.0 An alternative analysis 

In the preceding section, I have tried to unearth various 
problems posed by some Ewe SVCs for the framework of Baker 1989. 
In this section, I will make an alternative proposal in an attempt to 
address the problems raised in the preceding section. 

The main issue so far is the argument sharing of the verbs in the 
SVC and the projection of this phenomenon in the tree. The standard 
conditions on theta-marking in (13) were used to account for this 
phenomenon. Now, it has been shown that certain SVCs in Ewe do not 
display object or argument sharing (see examples (45) - (47)). There 
are two possible ways to account for these non-argument-sharing 
SVCs: one is to revise the conditions on theta-marking in (13); the 
other is to modify the syntactic projection of SVCs. Whatever option 
is chosen affects the other in a way. To modify the projection would 
lead us back to the structure in (48b) in which the V2 is not able to 
theta-mark the NPl. This would account for the non-argument- 
sharing SVCs, but may not account for the argument-sharing ones. 
Moreover, the structure in (48b) looks more like a coordinate 
structure, since there are two VI heads at the same level. I therefore 
would resort to the first option (i.e., revise the conditions on theta- 
marking) and modify the projection proposed by Baker a little. 

I adopt a rather strict version of the conditions for theta- 
marking by doing away with condition (b) in (13). The revised 
condition is stated as follows: 

(13') a may theta-mark (i iff a is a structural sister of p. 

The above condition would prevent V2 from theta-marking NPl in 
the tree below. However, while these conditions can account for non- 



34 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

argument-sharing examples like (45) - (47), it would be difficult to 
account for SVCs like (6a) or (9), which are of the argument sharing 
type. To account for these types as well, I propose that the NP2 
projection in (56) below is either empty or overt, depending on the 
type of SVC. NPl and PP are made optional in (56) by putting them 
in parenthesis. This is to make it possible to use the same framework 
for verbs that do not select an NP or PP. 

(56) 



s 




/ \ 




NP VP 
1 




1 

v 




/ / 1 


\ 


/ / 1 


\ 


/ / 1 


\ 


(NPl) (PP) V 




/ \ 




V2 NP2/PP 



In argument-sharing types, NP2 would be null and coindexed with 
the NPl object of VI. We noted earlier that in SVCs where the VI 
takes an additional (PP) argument, the V2 is not able to theta-mark 
this argument thus violating the Projection Principle under Baker's 
proposal. This suggests that the object (argument) sharing phe- 
nomenon should be projected in a different way than Baker 
suggested. Applying a typical example like (57a), we shall have the 
structure in (57b). 

(57) (a) Kofi so d,evia wu 

K. beat child the kill 

'Kofi beat the child to death.' 

(b) s 

/ \ 

NP VP 

/ / I \ 

/ / I \ 

Kofi VI NPli V 
•11 / \ 

I I V2 NP2 

^1 I I I 

so ctevia wu ei 

In the above tree, VI theta-marks only NPl by the revised 
conditions on theta-marking. V2, being a transitive verb, selects an 
object NP, which is identical to that of VI. NP2 is therefore rendered 
null and coindexed with NPl. Note that in this particular SVC, NP2 
cannot be overt. It can only be overt in coordinate structures, in 
which there must be a pronominal subject for V2 and a coordinating 
conjunction as in: 



Agbedor: Verb serialization in Ewe 35 

(57a') Kofi so ctevia eye wo wu-i 

Kofi beat child-the and 3sg. kill-3sg(0BJ) 
'Kofi beat the child and he killed him.' 

My proposal, I believe, would avoid the three major problems of 
Baker's proposal, namely: 

(i) the inability of V2 to theta-mark a PP argument of VI even 
though the argument position satisfies the condition for theta- 
marking proposed by Baker; 

(ii) non-sharing of NP between VI and V2 in some SVCs; 

(iii) the problem of the 'fly go' type. 

In my proposal, the V2 would not have to theta-mark the object of 
VI, since V2 would have its own object position projected. This 
solves the second problem automatically; that is, the verbs do not 
have to share NP objects. Concerning the 'fly go' type, my proposal 
does not involve movement of the internal argument of V2 to subject 
position as is normally proposed for unaccusatives, since that posi- 
tion would already be filled by the external argument of VI. 

(54') s 

/ \ 

NP VP 

I / \ 

Xeviai V V 

I / \ 
I V NP 
I I I 

dzo dzo ei 

So the internal argument position of V2 is projected as an empty 
category co-referential with the subject of the sentence. In (57a), the 
object of V2 is understood. It is, therefore natural to assume that 
there is an empty category in the object position of V2. The question 
then arises as to what type of empty category the null object is. It is 
assumed that there are four types of empty categories, on the basis 
of the two binary features [± anaphor] and [± pronominal] (cf. 
Chomsky 1982). 

(58) (a) PRO: [+ anaphor, -t- pronominal] 

(b) pro: [- anaphor, + pronominal] 

(c) WH-trace: [- anaphor, - pronominal] 

(d) NP-trace: [+ anaphor, - pronominal] 

PRO is ruled out as a probable candidate, since it is said to be 
ungoverned at S-structure. The object position in (57b) is governed 
by the verb, since that position must be case-marked. 

NP-trace is also not a possible candidate because NP-movement 
involves movement from a theta-position to a theta-bar position. The 
null object in (57b) is in a theta position, and its antecedent is also in 
a theta position. Therefore, it cannot be NP-trace. However, the null 
object under discussion seems to share a property with NP-traces. An 



36 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

NP-trace is subject to Principle A of the Binding Theory which says 
that an anaphor must be bound in its governing category. The 
governing category for the null object in (62b) is the entire clause, 
and the null object is coindexed with an NP within its governing 
category. So in a way, the null object has something in common with 
an NP-trace, i.e., they both occur in A-positions. But they differ in 
their case-marking properties."* 

WH-trace is also ruled out because it must be bound by an 
antecedent in an A-bar position. In the examples under discussion, 
there is no A-bar binder for the trace. This leaves us with pro. 
Supposing we assume at this stage that the null object is pro. This 
raises two questions: 

(a) What are the conditions that formally license the prol 

(b) How is the content of pro determined or recovered? 

To answer these questions, let us look at the proposal by Rizzi (1986) 
that pro is subject to two requirements, formulated in what is 
termed the 'pro-drop parameter'. 

(59) The pro-dvop parameter: 

(a) pro is governed by X^ 

(b) Let X be the licensing head of an occurrence of pro; 
then pro has the grammatical specification of the 
features on X coindexed with it. 

Condition (a) can be satisfied in (57b); the object position is licensed 
by the verb wu 'kill'. The problem is with condition (b). In Italian, a 
typical pro-drop language, the content of pro in subject position is 
recoverable from the rich morphology of the verb (i.e. from strong 
agreement features). But for pro in object position, Rizzi (1986) 
suggests a distinction between English and Italian in the way null 
objects are licensed. He claims that an occurrence of pro in a verb- 
governed position is allowed in Italian but not in English. He argues 
that in Italian, the understood object is syntactically 'active' in that it 
can act as a controller, as a binder, and as a subject of predication for 
adjunct and small clauses, whereas the null object in English appears 
to be syntactically 'inert' in the same environment (502). Compare 
the following sentences from English and Italian. 

(60) (a) This leads people [PRO to conclude what follows] 
(b) *This leads [PRO to conclude what follows] 

(61) (a) Questo conduce la gente alia seguente conclusione 
(b) Questo conduce — alia seguente conclusione 

(Rizzi, 1986:503) 

In (60b), we find that we cannot delete the object controller, whereas 
in (61b) we can. This suggests that in object-control structures in 
English, the object NP controller must be overtly represented. Ewe 
follows English in this respect. Consider the following. 



Agbedor: Verb serialization in Ewe 37 

(62) (a) Esia nana amewo susuna be nuwuwua cto 

This make-HAB people think-HAB that end-the arrive 
This makes people think that the end is near.' 
(b) *Esia nana — susuna be nuwuwua d,o 

In (62b) the object cannot be deleted. Rizzi (1986) also points out 
that in Italian, argument small clauses selected by causative verbs 
can take null subjects having the same interpretive and formal 
properties as the null objects. 

(63) (a) Questa musica rende [ — allegri] 

This music renders — happy [+pl] 
(b) Certe medicine rendono [ — piu intelligenti/calmi] 

Certain medicines render — more intelligent/calm [+pl] 

The English glosses in (63) are ungrammatical. The missing null 
object in the Italian examples must be present in the English glosses 
to be grammatical in English. Ewe behaves just like English in this 
respect too. 

(64) (a) Atike ac^e-wo woa [ame drozii] 

medicine INDE.-pl make person weak 
'Some drugs make people weak.' 
(b) *Atike at^ewo waa — drozii 
'Some drugs make weak.' 

(64b) is ungrammatical because of the missing small clause subject. 

Evidence adduced so far points to the fact that the null object 
being proposed for Ewe SVC may not be pro. In fact the discussion so 
far suggests that there is no structural NP position. But as has been 
pointed out, the failure of Baker's model to satisfy the Projection 
Principle suggest that the object sharing phenomenon should be 
projected in a different way. One plausible way is to project an 
empty NP object for V2, an NP which will be co-indexed with the NP 
object of VI. Moreover, pro as a pronominal must be free in its 
governing category (i.e. it is subject to Principle B of the Binding 
Theory). The null object being proposed here is quite different, in 
that it is co-indexed with an NP within its governing category. It will 
also be shown that this null object is bound by the NP object of VI. 
Also, there is no evidence that Ewe allows object pro independently 
of SVCs. Therefore, the null object in question cannot be pro. 

Raposo (1986) proposed for European Portuguese that the empty 
category in object position is a variable. 5 According to Principle C of 
the binding theory (Chomsky 1981), a variable, like other referring 
expressions, cannot be coreferential with a c-commanding nominal 
occurring in an argument position. This is because variables, like 
other referring expressions cannot be A-bound. Pronominals, on the 
other hand, are not subject to Principle C and can, therefore, be 
coreferential with a c-commanding argument (as long as these 
arguments do not occur in the governing category (GC) of the 



38 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

pronominal). This rules out the null object being a variable and 
brings us back to a point mentioned earlier. 6 

We noted earlier that the null object being proposed shares a 
characteristic with NP-traces, but that it cannot be NP-trace because 
it is in a case-marked and a theta position. The shared characteristic 
is that the null object is bound in its governing category (i.e., it has as 
its antecedent an argument in its GC). This position satisfies the 
condition for an anaphor. I, therefore, propose that the null object in 
the Ewe SVC structure in (57b) is an 'empty anaphor' (cf. Saxon 
1989, 1990; Chung 1989). We will have a base-generated NP 'empty 
anaphor' as the object of V2, and this would be coindexed with the 
NP object of VI. This gives us the structure in (65). 

(65) S 

/ \ 

NP VP 

/ I \ 
V NPli V 
/ \ 
V2 NP2 
I 

ei 
[+ana] 

Since this empty category is an anaphor, it must obey Principle A of 
the binding theory. We shall now explore the conditions on this bind- 
ing principle and see how far the structure in (65) fits into it. 

(66) Binding Theory:. 

Principle A 

An anaphor must be bound in its governing category. 

(67) A-binding 

a binds P iff 

(i) a is in an A-position; 

(ii) a c-commands (3; 

(iii) a and p are co-indexed. 

(66) involves two notions, 'binding' and 'governing category'. (67) 
outlines the conditions for binding. The first is that a must be in an 
A-position. In (65), the NP binder is in an A-position. The second 
condition is that a must c-command p. Here, 1 will adopt the revised 
version of c-command which is also known as m-command, and 
which is stated in Sells 1985:39 as follows: 

(68) C-command (revised definition) 

a c-commands p iff every maximal projection dominating a 
also dominates p 

Under this interpretation of c-command, the governing category of 
the empty anaphor being proposed will be the entire clause. This 
empty anaphor will be bound by the NP which is within the GC, thus 
satisfying Principle A in (66). 



Agbedor: Verb serialization in Ewe 3 9 

Now let us consider the other problematic type of SVC (the one 
involving unergative and unaccusative verbs). I propose the same 
base-generated 'empty anaphor' as above for the argument of V2. 

(69) S 

/ \ 

NPi VP 

I / \ 

Xeviai V V 
I / \ 
VI V2 NP 

I I I 

dzo dz6 ei 

[ +ana] 

As noted earlier, the only argument of the V2 is internal, and if this 
verb occurs alone in a structure, then the D-structure argument 
moves to subject position to receive NOMINATIVE case. But here it is 
combined with an unergative, whose only argument is external. 
Therefore, when these two verbs combine, the subject position is 
already filled by the external argument of the VI and, therefore, the 
object of V2 cannot move there to receive case. So it must be empty, 
and since it is coreferential with the subject, it must be an anaphor. 
Its GC is the entire clause. The subject 'bird' m-commands the empty 
anaphor and they are co-indexed. 

One question that needs to be addressed is whether the empty 
anaphor, ana, needs case. I assume that ana, being the object of a 
transitive verb, must be case-marked. The position is governed and 
theta marked and, therefore, nothing prevents the empty anaphor 
from receiving case. Moreover, WH-traces and pro are case-marked. 
So we can say that the empty anaphor is case-marked. The case on 
the empty anaphor will make the theta position visible and allow the 
predicate to assign its theta role.^ 

6.0 Conclusion 

I have been discussing Ewe serial verbal constructions within a 
framework suggested by Baker (1989). I noted that the Ewe data 
pose certain problems for Baker's framework, especially with respect 
to the Projection Principle and the notion of object-sharing. Baker's 
distinction between a true and non-true SVCs has been rejected (at 
least for Ewe). An alternative proposal has been made for handling 
SVCs in Ewe. It was found necessary to modify the conditions on 
theta-assignment and the syntactic projection of SVCs to make room 
for both object-sharing and non-object-sharing types. The notion of 
'empty anaphor' was introduced as the null object of V2 in the SVC. 
Object-sharing is achieved by coindexation of the empty anaphor 
with the object of VI. This proposal has to be tested against other 
SVC languages to determine its universality. Moreover, the notion of 



40 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

empty anaphor or 'Little ana' (Saxon 1989, 1990) is quite new and 
needs to be tested by further research. 

Despite the extensive work on SVCs in the past twenty years, 
there are still more questions than answers. The issue of what 
constitutes a true SVC is not clear. Also pertinent is the question of 
what constitutes a main verb in SVCs. Should there even be a main 
verb and a subordinate one? What principles determine the order of 
verbs in SVCs? These and other questions need to be addressed in 
future research. 

NOTES 

1 Reported in Awoyale 1988. 

2 Cited in Awoyale 1988. 

3 There seems to be a pragmatic issue here. The pronominal 
third person that distinguishes the two sentences in (51) and (53) is 
always coreferential with the immediately preceding NP. It is the 
subject of the embedded clause and, being a pronominal, it should be 
free in its governing category. The antecedent of this pronominal is 
the object of the matrix verb, which is outside the GC of the pronom- 
inal. The V2 dze anyi seems to be a fixed collocation forming an 
intransitive verb. The issue as to whether it is an unaccusative is not 
clear, and I do not intend going into that here. 

'^ This issue will be taken up again later in this section. 

5 Reported in Cole 1987:597. 

6 It may be possible to say that the null object is a variable left 
by an empty operator. But assuming that an operator moves only 
into [Spec, CP], the issue will be where to locate the operator in the 
tree in (57b) for example. Carstens (1988) is reported to have taken 
on that issue (see Baker 1989). 

^ Case assignment for the empty anaphor in (69) is, however, 
problematic. It may be possible to propose that the empty anaphor 
in (69) is case-marked, with the case realized on its antecedent. This 
needs to be further explored. 

REFERENCES 

Awoyale, Y. 1985. On the semantic interpretation of serial verb 
constructions. West African languages in education: Papers from 
the 12th West African Linguistics Conference, ed. by K. Williams, 
144-174. Port Harcourt. 

. 1987. Perspectives on verb serialization. Niger Congo syntax 

and semantics, ed. by V. Manfredi, 1.3-35. Boston: Boston 
University Press. 

Baker, M. 1989. Object sharing and projection in serial verb 
constructions. Linguistic Inquiry 20.513-553. 



Agbedor: Verb serialization in Ewe 4 1 

BAMGBOSE, a. 1982. Issues in the analysis of serial verb constructions. 

Journal of West African Languages 12.2-21. 
Campbell, R. 1991. Argument-sharing serial verbs in Akan. 

University of Pennsylvania, MS. 
Carstens, V. 1988. Yoruba serial verbs. Paper presented at the 2nd 

Niger-Congo Syntax and Semantics Workshop, MIT, Cambridge, 

MA, April 1988. 
Chomsky, N. 1982. Some concepts and consequences of the theory of 

government and binding. Cambridge, MA: MIT Press. 
. 1986. Knowledge of language: Its nature, origin, and use. New 

York: Praeger. 
Chung, S. 1989. On the notion 'null anaphor' in Chamorro. The null 

subject parameter, ed. by O. Jaeggli & K. Safir, 143-184. 

Dordrecht/Boston/London: Kluwer Academic Publishers. 
Cole, p. 1987. Null objects in universal grammar. Linguistic Inquiry 

18.597-612. 
DecHAINE, R. 1988. Towards a typology of serial constructions in 

Haitian. Niger Congo syntax and semantics, ed. by V. Manfredi, 

1.49-64. Boston: Boston University Press. 
Ekundayo, S.A., & F.N. Akinnaso. 1983. Yoruba serial verb string 

commutability constraints. Lingua 60.115-133. 
Haegeman, L. 1991. Introduction to government and binding theory. 

Oxford: Basil Blackwell. 
Laniran, Y., & O. Sonalya. 1988. The lexical nature of Yoruba 

serialization. Niger Congo syntax and semantics, ed. by V. 

Manfredi, 1.37-48. Boston: Boston University Press. 
Larson, R. K. 1991. Some issues in verb serialization. Serial verbs: 

grammatical, comparative, and cognitive approaches, ed. by C. 

Lefebvre. Amsterdam/Philadelphia: Benjamins. 
Lefebvre, C. 1991. Take serial verb constructions in Fon. Serial verbs: 

grammatical, comparative, and cognitive approaches, ed. by C. 

Lefebvre. Amsterdam/Philadelphia: Benjamins. 
Ll, C. N., & S. Thompson. 1973. Serial verb constructions in Mandarin 

Chinese: subordination or coordination? Chicago: Chicago 

Linguistic Society. 
MATLSOFE, J. 1973. The grammar of Lahu. Berkeley: University of 

California Press. 
MUYSKEN, P. 1988. Parameters for serial verbs. Niger Congo syntax 

and semantics, ed. by V. Manfredi, 1.65-75. Boston: Boston 

University Press. 
Oyelaran, O. 1982. On the scope of the serial verb construction in 

Yoruba. Studies in African Linguistics 13:2.109-146. 
RIZZI, L. 1986. Null objects in Italian and the theory of pro. Linguistic 

Inquiry 17:3.501-557. 
ROSS, J. R. 1967. Constraints on variables in syntax. MIT Ph.D. 

dissertation in Linguistics. 
Saxon, L. 1989. Agreement in Dogrib: Inflection or cliticization? 

Theoretical perspectives on Native American languages, ed. by D. 



42 SMfesiB*eIiiiewsttcSaeHxs23tUS|!ni*I993) 

B. Gens & K. ifidielsMMu 149-162. Aibaay. NY: Sate Uiii^^rsit> of 

New Ycrk Press. 
. 1990. Reflexive i£Teeiiieat binding. Proceediags of the 16th 

Ajuiaal Meetug of the Beikeley Linfoisdc Societ>\ Special 

SessKM oa A — e j m ^mi Luigustics. Berkeley: Berkeley Linguistic 

Socie: 
SCH.\CETf ^ F 1~~- A MMHtnBsfonauioMal accoorit of serial vertvs. 

Sr.iies u Afrkaa Luigvistks, Sapplemeat 5.2S3-270. 
5r?rA. XL 1987. The syalax of seri^ ii«fbs. A m st e idMn: Beajamins. 
5r.lS. P. 19S5. Lectares m co^nratiTe syntactic theory. Stanfcxii: 

Ccrrer for the Sta^ of Laagaage aad laformatioa 
STr^ ^ xT. J. 1963. Some lesirictioas oa obfects ia Twi. Joniaal of West 

LMiSaages 1.145-149. 
We5 ; }«, D- 193a A stady of Ae Ewe li-£_^£c _:: - Oxford 

L^Bi««rsity Press. 
WlLi:*^f> E 1vS4 Grumnatica] relatio-f _-:i_:>tic Inquiry 15.639- 



Studies in the Linguistics Sciences 
Volume 23, Number 1, Spring 1993 



YES, WE HAVE NO BANANAS: 

ENGLISH NEGATIVE TAGS IN CROSS-LINGUISTIC 

COMMUNICATION* 

Martin J. Baikand Rosa J. Shim 

A study was conducted with Korean-English bilinguals 
in both Korean and English. Two groups of bilinguals, 
proficient and non-proficient in English, were asked to 
answer a series of negative tag questions. Results indicate 
the following: First, the proficient group was generally 
more accurate than the non-proficient group in answer- 
ing English questions. Second, Korean-English bilinguals 
were not aware of the constraints in English positive 
disagreement strategies; thus they do not feel that posi- 
tive disagreement is any more difficult than negative 
agreement. Third, learning the English system does not 
have any effect on the native language system; thus they 
are able to code-switch freely without difficulty. 

1. Introduction 

As human beings we depend largely on language to commu- 
nicate with other people in society. In the ever-increasingly complex 
world that we are living in, cross-linguistic communication is a 
common feature. On the campus of the University of Illinois, for 
example, there is a large body of foreign students who come from 
diverse linguistic backgrounds. The question that arises in such a 
situation is whether or not we truly understand each other by what 
we mean rather than by what we say. 

In the profession of language teaching and learning, awareness 
of cross-linguistic interaction has brought about various types of 
studies that show contrasts between the learner's language and the 
target language. Contrastive Analysis (Dulay & Burt, 1972, 1974, 
1975; Dulay et al. 1982), Error Analysis (Corder, 1971; Richards, 
1971), Interlanguage Hypothesis (Selinker, 1972). and Markedness 
Hypothesis (Eckman. 1977) can be considered major developments in 
the recent past. Unfortunately, most of the efforts in these fields 
have come from the English Language Teaching (ELT) profession, and 
there is an unmistakable bias in these studies that treat the systems 
in English as the norm and the systems in other languages as either 
deficient, or as a departure from the norm. 

In this study we try to break away from such a bias. Our 
concern is in describing the use of one simple but major system of 
communication, that of expressing agreement or disagreement to 
negative opposite-polarity tag questions in English, and a comparable 
construction in Korean. 



44 Studies in the Linguistic Sciences 23:1 

It has been noted by Pope (1976) that in English, Yes/No 
answers correspond with the positivity or negativity of the propo- 
sition that follows the answers while there are other languages in the 
world in which Yes/No answers signal agreement/ disagreement 
with the proposition in the question. The former system is referred 
to as the positive-negative system and the latter is called the 
agreement-disagreement system. Korean is one of the languages that 
can been described as having an agreement-disagreement system. 

The two systems do not differ in answering simple questions 
with positive propositions. In most cases, comparable structures of 
questions exist in both English and Korean and the same answers are 
given in both languages. Thus for question (1) and question (2), the 
answers are not different in the two languages. 

(1) Are you a student? 

Yes. (I am a student) / No. (I am not a student) 

(2) hakseng- ip- nika? 
student BE QM 

Ye. (hakseng- ipnita) / Aniyo (haksengi- anipnita) 
Yes (student BE) / No (student not BE) 

However, problems arise when negative questions are asked in 
the two different systems. In the following section of this paper, we 
will give a brief description of the two systems of answering nega- 
tive tags and some of the pragmatic conditions that are related to the 
use of negative opposite-polarity tags in English and a comparable 
construction in Korean. Then, we will discuss some of the practical 
problems that arise from the differences in the two systems. Follow- 
ing this discussion we will outline the present study and present the 
results. Finally, we will discuss the results in relation to previously 
mentioned problems and suggest further areas of research. 

2. Negative tag questions: Answering systems and pragmatic 
concerns 

When Yes/No answers are given to negative opposite-polarity 
tags in English, the prescriptive English grammar that we impose on 
non-native speakers of English mandates that answers must match 
the positivity or negativity of the proposition that follows the an- 
swers. For example, when question (3) is asked, the answers can be 
Yes, I did in disagreement or No in agreement with the questioner's 
assumption that the answerer did not go to the library. Since the 
positive proposition / did is in disagreement with the negative 
assumption in the question. Pope (1976) calls this POSITIVE 
Disagreement. Similarly, negative answers are labelled NEGATIVE 
Agreement since the negative proposition in the answer is in agree- 
ment with the assumption of the question. 

(3) You didn't go to the library, did you? 
Yes, I did. / No. (I didn't.) 



Baik «& Shim: Yes, we have no bananas 45 

In addition to Pope's simple description of answers to negative 
tags in English, Houck (1991) discusses the pragmatic implications in 
asking such questions. In the case of falling-intonation opposite- 
polarity tag questions, she claims that the questioner has to assign a 
strong probability to the truth of the proposition and at the same 
time assume that the answerer would also assign a strong probability 
to the truth of the proposition. 

The implication of Houck's contention is that the necessary 
pragmatic context for falling-intonation opposite-polarity tag ques- 
tions are met only when the question is asked in confirmation of the 
truth of the proposition in the question. If this is in fact true, then 
the expected answer to a negative falling-intonation opposite- 
polarity tag question is always No. 

However, it does not necessarily follow that we only hear the 
answer No to negative tag questions in real life. Regardless of the 
expectations, we sometimes do need to answer such questions with a 
disagreeing Yes, and when we do, it is not sufficient to simple say 
Yes. Rather, we normally repeat the tag in the answer in order to 
achieve successful communication. The point noted by Pope which 
explains the need to give fuller answers in positive disagreement is 
our tendency to use the word Yes in agreement. This is what we do 
in answers to positive questions. However, this tendency is not 
fulfilled when answering negative questions since the answer Yes 
here is used to disagree rather than agree. This explains the need to 
say something more. 

Whether the reasons lie in our tendency to use Yes in agreement 
rather than in disagreement, or in the fact that pragmatic conditions 
are not fulfilled in positive disagreement. Pope agrees with Bellugi's 
(1967) observations that the use of the word Yes as a positive 
disagreement strategy is the last category of negation to be acquired 
by children. 

In Korean there is no parallel construction to the English tags. 
However, it is possible to ask negative questions which have similar 
pragmatic constraints as English negative tags. In this sense, there is 
a comparable construction in Korean that matches question (3): 

(4) tosekwan -e an -ka ss -ci -yo? 

library to not go PAST QM HON 

(You) didn't go to the library, did you? 

In question (4) the question marker (QM) -ci has the function of 
asking confirmation of the truth of the proposition in the question. 
The morpheme ss marks the past tense of the verb ka 'go', and -yo is 
an honorific morpheme. Thus the pragmatic constraints on question 
(4) are similar to that of question (3). 

It is interesting to note, however, that the expected answer to 
this question is exactly the opposite of English. Although we have 
already seen that Ye means Yes and Aniyo means No, the answer Ye 



46 Studies in the Linguistic Sciences 23: 1 

is given to confirm the assumptions of the negative question. At the 
same time, the answer Ye signals that the negative proposition that is 
given in the question is in fact true. When the negative answer Aniyo 
is given to this question, it is used to contradict the assumptions of 
the question. In addition to disagreeing with the assumptions, Aniyo, 
in this case, signals that the negative proposition of the question is 
false. In other words, this signals positive disagreement since the 
true proposition that naturally follows the answer is a positive one. 

A further interesting observation is that, unlike in English, this 
does not cause problems in giving simple answers in Korean. The 
reasons is that the answers Ye/Aniyo signal clear agreement/ dis- 
agreement with the proposition in the question, and these answers 
do not have a direct relationship with the proposition of the ex- 
tended answer. Thus misunderstanding is unlikely to occur, and 
Koreans do not feel the need to supply extended statements to 
positive disagreement answers. 

3. Problems in cross-linguistic communication 

The problems are predictable when we have two systems of 
interaction that are identical on the one hand and antithetical on the 
other. Once Korean-English bilinguals establish the similarity in the 
English system and the Korean system of answering positive 
questions, they will invariably transfer that knowledge when 
answering negative questions. The consequences of this transfer are 
that there is a possibility for a mismatch between one's meaning and 
one's utterance when a Korean answers negative questions in English. 

In the real world, this type of transfer can cause serious prob- 
lems for the non-native speaker if native speakers are not aware of 
differences in answering systems in different languages. Gumperz 
(1982) reports of a Philippino doctor who was charged with perjury 
when in fact he was simply misunderstood by the FBI agents who 
were not aware that the doctor was using a different system of 
linguistic interaction. The case was eventually dismissed when the 
defense attorneys called in a linguist to closely examine the doctor's 
speech. The doctor was shown to be using English in a way that was 
not familiar to monolingual English speakers, and one of the major 
areas of conflict was in the system of answering negative questions. 
The doctor's native language, Tagalog, is a language that has the 
agreement-disagreement system, and occasionally he made mistakes 
when giving answers to negative questions, leading to apparent 
contradictions in his testimony. The following is an example of his 
speech at a Navy hearing (Gumperz 176): 

(5) Q: It's the testimony by Lieutenant Commander Gilbert that 
you did not attend the briefing. 
Yes. 

You did attend it? 
No. 



(6) 


A: 


(7) 


Q: 


(8) 


A: 



Baik & Shim: Yes, we have no bananas 47 

As the above example clearly shows, the Philippine doctor is misun- 
derstood by the questioner when he gives answer (6). The questioner 
takes this answer to be positive disagreement even though the 
answer is not elaborated on with an extended prepositional state- 
ment. The point is that the doctor was being consistent in his use of 
the agreement-disagreement strategy, and this was interpreted by 
monolingual native speakers of English to be a contradiction in his 
testimony. 

4. Aims of the study 

The study presented in this paper was conducted for two main 
purposes: First it was to investigate the uses (or misuses) of Yes/No 
answers to English negative tag questions by two groups of Korean- 
English bilinguals, those that are proficient and those that are non- 
proficient in English. We have established that there is a source of 
miscommunication which is a direct result of antithetical rules that 
are applied to answering negative questions in English and Korean. 
Cowan (1983) gives the generalization that it is precisely in such 
circumstances that there is the greatest possibility of errors. Never- 
theless, we do not believe that this is a permanent condition. Human 
cognitive abilities enable us to encompass many conflicting systems 
of thought. It should not be impossible to attain a level of proficiency 
in English that allows proficient non-native English speakers to 
answer negative tags to near-native accuracy. Thus it is expected 
that Korean-English bilinguals that have achieved the level of pro- 
ficiency in English that is deemed appropriate for graduate level 
study will be more accurate in answering negative tag questions in 
English than the non-proficient, intermediate level English learners. 

The second question asked in this study stems from the differ- 
ent answering strategies in the two languages when giving answers 
of positive disagreement. How many Korean-English bilinguals are 
aware that they not only need to give a completely different answer 
in English but also that they need to elaborate on their answers of 
positive disagreement? Will they feel more constrained in giving a 
simple Yes/No answer when the answer is in positive disagreement? 
In other words, will they make more mistakes in positive disagree- 
ment situations than in negative agreement situations? 

The third question that we wanted to look at in this study was 
whether the learning and using of a different system of answering 
negative questions had any effect on the bilingual's first language. In 
other words, as a Korean-English bilingual becomes more proficient 
in the English answering system, would this have a negative effect on 
the Korean answering system? A related questions is whether 
Korean-English bilinguals in general will have difficulty maintaining 
the answering system of their native language if for some reason 
they need to code-switch to Korean from an English speaking situa- 
tion. In these circumstances, will they become less accurate in giving 
answers to Korean negative questions? 



48 Studies in the Linguistic Sciences 23:1 

5. Methodology of the study 

In order to answer the above research questions, two groups of 
Korean-English bilinguals, non-proficient, intermediate-level English 
learners that were enrolled in the Intensive English Institute at the 
University of Illinois, and proficient English speakers that were 
graduate students at the University of Illinois, were asked 16 nega- 
tive tag questions in English and Korean. There were 17 subjects in 
each group. The mean length of stay of the students in the US was six 
months, and three years and ten months, respectively. 

Both English and Korean questions were masked as asking 
opinions on four commercial products: jeans, diamonds, cars, and 
cola. Advertisements were taken from magazines on Levis, diamonds, 
Toyota, and Coca Cola. Similar advertisements were chosen for 
Korean questions. The Korean questionnaire was a direct translation 
of the English questionnaire. 

In order not to give the subjects the impression that they were 
asked only negative questions, an equal number of positive questions 
were asked. Four possibilities were given for answers to English 
questions: Yes, No, Don't know, and Don't understand the question. 
For Korean questions, the possibility that the subjects would not 
understand the question was remote, hence the fourth option was 
not given. 

We had to be sure that the answers to the negative questions 
were either correct or incorrect depending on the simple Yes/No 
answers that the subjects gave. Thus all of the negative questions 
were very simple in that there was no ambiguity in the truth or 
falsity of the proposition in these questions. Discussions with the 
subjects after the test confirmed this fact. The positive questions, 
however, asked for the subjects' genuine opinion so as to disguise the 
purposes of the test to some extent. The questionnaire is given in full 
detail in Appendix A. 

Both the English questions and the Korean questions were audio- 
taped with approximately equal limit of time (7 seconds) to answer 
for each question. All of the English negative questions were asked 
with falling-intonation tags. Negative questions in Korean were asked 
with normal rising intonation which implies the same pragmatic 
conditions as the English falling-intonation opposite-polarity tag 
questions. 

When all the subjects arrived at the testing site, they were 
asked to read the cover sheet for the English questionnaire. Then 
they were given photocopies of the four advertisements which they 
could examine at leisure. They were allowed to discuss the product 
or ask questions about the product before they started answering 
questions on them. The subjects were told that they will be asked to 
answer questions about the four products. When everyone was 
ready, the tape containing the questions was played back. Subjects 



Baik & Shim: Yes, we have no bananas 



49 



were not allowed to speak during the test. Upon completion of the 
English questionnaire, the same process was followed for the Korean 
questionnaire. 

At the end of the testing sessions, completed questionnaires 
were collected, and the subjects were informally interviewed regard- 
ing the difficulty of the questions asked, whether or not they guessed 
what the purposes of the tests were, if they felt that they had to say 
something more than a simple Yes or No, and various other questions 
concerning the products in the questionnaires. A full list of questions 
asked at the informal interview is given in Appendix B. 

6. Results 

The accuracy rate was calculated in percentages of number of 
correct answers over number of answers given. Since the subjects 
had the option of saying 'Don't know' when they in fact did not know 
the answer to a question, the items for which a subject gave 'Don't 
know' as the answer was not included in the analysis. Results from 
the proficient speakers are given in Table 1. and results from the 
non-proficient speakers are given in Table 2. PD refers to POSITIVE 
DISAGREEMENT and NA refers to NEGATIVE AGREEMENT in each table. 







English 




Stay 


Overall 


PD 


NA 


nth) 










100.00% 


100.00% 


100.00% 




100.00% 


100.00% 


100.00% 




100.00% 


100.00% 


100.00% 




86.66% 


75.00% 


100.00% 




93.75% 


88.89% 


100.00% 




80.00% 


62.50% 


100.00% 




93.75% 


88. 89%^ 


100.00% 




87.50%' 


100.00% 


71.43% 




93.33% 


87.50% 


100.00%- 




100.00% 


100.00% 


100.00% 




86.66% 


100.00% 


66.67% 




87.50% 


100.00% 


71.43% 




93.33% 


87.50% 


100.00% 




93.75% 


88.89% 


100.00% 




92.86% 


100.00% 


85.71% 



n = 15 
Length - 

(Year:Mo 
10:03 
1:04 
8:07 
3:04 
2:06 
4:07 
4:03 
5:05 
6:03 
1:02 
2:00 
1:04 
4:09 
5:04 
1:03 

Table 1. Accuracy rate of proficient Korean-English 

bilinguals 

Out of 17 possible subjects for each group, data from two 
subjects in the proficient group were excluded from the analysis. 
Accuracy rates for the two subjects excluded from the proficient 
group are given in Table 3. 





Korean 




Overall 


PD 


N A 


93.33% 


100.00% 


85.71% 


100.00% 


100.00% 


100.00% 


93.33% 


88.89% 


85.71% 


100.00% 


100.00%. 


100.00% 


93.75% 


88.89% 


100.00% 


81.25% 


100.00% 


57.14% 


86.67% 


100.00% 


71.43% 


100.00% 


100.00% 


100.00% 


100.00% 


100.00% 


100.00% 


75.00% 


66.67% 


85.71% 


100.00% 


100.00% 


100.00% 


87.50% 


77.78% 


100.00% 


83.33%. 


66.67% 


100.00% 


93.33% 


87.50% 


100.00% 


100.00% 


100.00% 


100.00% 



50 



Studies in the Linguistic Sciences 23:1 



n = 16 




English 








Korean 




Length - Stay 


Overall 


PD 


NA 




Overall 


PD 


NA 


(Year:Month) 
















0:03 


92.86% 


87.50% 


100.00% 




93.75% 


100.00% 


85.71% 


0:03 


100.00% 


100.00% 


100.00% 




93.33% 


100.00% 


85.71% 


0:04 


18.75% 


11.11% 


28.57% 




100.00% 


100.00% 


100.00% 


0:05 


12.50% 


11.11% 


14.29% 




87.50% 


88.89% 


85.71% 


0:10 


0.00% 


0.00% 


0.00% 




100.00% 


100.00% 


100.00% 


0:10 


37.50% 


22.22% 


57.14% 




68.75% 


77.78% 


57.14% 


0:07 


12.50% 


0.00% 


28.57% 




86.67% 


87.50% 


85.71% 


0:11 


93.75% 


100.00% 


85.71% 




100.00% 


100.00% 


100.00% 


0:09 


33.33% 


37.50% 


28.57% 




86.67% 


75.00% 


100.00% 


0:03 


12.50% 


0.00% 


28.57% 




100.00% 


100.00% 


100.00% 


0:03 


7.14% 


0.00% 


16.67% 




86.67% 


87.50% 


85.71% 


0:11 


86.67% 


87.50% 


85.71% 




92.86% 


85.71% 


100.00% 


0:05 


6.67% 


0.00% 


16.67% 




93.75% 


88.89% 


100.00% 


0:04 


86.67% 


88.89% 


83.33% 




93.75% 


100.00% 


85.71% 


0:04 


100.00% 


100.00% 


100.007r 




93.75% 


88.89% 


100.00% 


0:05 


7.69% 


0.00% 


14.29% 




81.25% 


66.67% 


100.00% 



Table 2. Accuracy rate of non-proficient Korean-English 

bilinguals 





English 








Korean 




Length - Stay 


Overall PD 


NA 




Overall 


PD 


NA 


( y e a r : m n t h ) 














4:01 1-1 


93.75% 88.89% 


100.00% 




25.00% 


33.33% 


14.29% 


1:05 1-2 


6.25% 0.00% 


14.29% 




100.00% 


100.00% 


100.00% 



Table 3. Idiosyncratic responses from the proficient group 



As Table 3 illustrates, the first subject (henceforth I-l) had stayed in 
America for four years and one month. His responses showed that he 
had almost perfect accuracy rates for English questions while his 
accuracy rates for Korean questions were well below chance level. On 
the other hand, the second subject (henceforth 1-2) who had lived in 
America for a year and five months had very low accuracy rates for 
English questions while there were absolutely no errors in his an- 
swers for all the Korean questions. In order to avoid the risk of 
having these data distort the general picture of the results, we will 
discuss them separately. 

As for the non-proficient group, responses of one subject were 
taken out of the analysis. The reason was that he admitted to giving 
the answer 'Don't know' whenever he was not sure which answer 
was correct. He understood the questions and knew the answers, but 
he answered 'Don't know' whenever he did not have enough time to 
figure out which one of them conveyed his meaning. He was very 
conscious of the differences in the two answering systems and was 



Baik & Shim: Yes, we have no bananas 



51 



afraid of making mistakes. Thus we felt that his accuracy rates were 
not representative of his unmonitored use of Yes/No answers. 

Thus analysis of data was carried out with data from 15 subjects 
in the proficient group and data from 16 subjects in the non- 
proficient group. The means and standard deviations of all the 
different columns in Tables 1 and 2 are given in Table 4. 



Proficient 


Mean 




s.d. 


Non- 


Mean 


Proficicni 


s.d. 





English 




Overall 


PD 


NA 


92.61% 


91.94% 


93.02% 


6.01% 


11.15% 


12.58% 




44.28% 


40.36% 


49.26% 


40.46% 


44.19% 


36.807o 





Korean 




Overall 


PD 


NA 


92.50% 


91.76% 


92.38% 


8.09% 


12.25% 


13.08% 




91.17% 


90.43% 


91.96% 


8.25% 


10.49% 


11.63% 



Table 4. Results of mean and standard deviation 

As the above table illustrates, the major differences were in the 
accuracy rates for answers given to English questions by the 
proficient group (in bold) and the non-proficient group (in bold and 
underlined). Tests of variance between the two groups revealed 
differences in the variance for the two groups that were statistically 
significant (p < .001) in all three categories (Overall, PD, NA). Further- 
more, t-tests' revealed that the accuracy rate of the proficient group 
was significantly higher than that of the non-proficient group in all 
three categories (overall: p < .001; PD: p < .001; NA p < .001). 

Difference of accuracy rates within the non-proficient group for 
English (bold, underlined) and Korean (underlined) questions were 
also statistically significant (p < .001 for all three categories). Tests of 
variance also revealed that the variance for the two sample data sets 
were significantly different (p < .001 for all three categories). 

Within-group t-tests for differences between PD category and 
NA category did not reveal any statistically significant differences for 
any group or language. Further, differences between accuracy rates 
for English and Korean questions within the proficient group were 
not statistically significant, and no significant difference was found 
for the accuracy rate for Korean questions by the two groups. 

One other analysis that was carried out was the analysis of 
correlation. Since we had access to the length of stay in the US for 
each subject, correlational analyses were done so as to find out if 
there was any relationship between length of stay in the US and the 
subjects' accuracy rate for English questions. No significant correla- 
tion was found. Further correlational analyses were done to see if the 
subjects' length of stay had an effect on the accuracy rate for Korean 
questions, and no significant correlation was found. 



52 



Studies in the Linguistic Sciences 23:1 



7. Discussions and implications 

Our first research question was whether proficient bilinguals 
would be better in giving accurate Yes/No answers to negative tag 
questions in English than the non-proficient bilinguals. Our hypoth- 
esis was that they would be, and the results of the study support this 
hypothesis. In addition to the fact that proficient bilinguals attain a 
very high level of accuracy, the results also show that there is no 
relationship between length of stay in the US and the subjects' 
accuracy in answering negative tag questions. Although the subjects' 
length of stay in US ranged from a year and three months to ten 
years and three months, the subjects' level of accuracy did not reflect 
this fact. This was also true of the non-proficient speakers: There was 
no correlation between length of stay and accuracy in answering 
negative tag questions. 

The second question asked in this study was whether Korean- 
English bilinguals understood the constraint in giving answers of 
positive disagreement in English. Results of the study and results 
from interviews with the subjects indicate that Korean-English bilin- 
guals are not aware of such a constraint. Neither of the groups 
showed any difference in their accuracy to positive disagreement 
(PD) answers and negative agreement (NA) answers. When asked if 
any of them felt that they needed to elaborate on their answers, the 
responses were that sometimes they would, but that they neither felt 
nor knew that it was necessary to do so. 

The third aim of the study was to find out if learning and using 
the English system of answering negative questions would have any 
effect in the Korean answering system. Although we could not find 
perfect accuracy rates in either of the groups, the results did show 
that there is no difference in the performance of proficient and non- 
proficient groups in answering Korean questions. Therefore, what- 
ever the reasons may be for the subjects in this study to have a less 
than perfect accuracy rate in giving answers in their native language, 
it does not seem to be due to the level of mastery of a different 
system. In other words, gaining proficiency in answering English 
negative questions does not seem to contribute toward a decline in 
their level of accuracy for Korean questions. 

This brings us to the discussion of the two idiosyncratic respons- 
es in the proficient group. They are repeated here for convenience 
sake. 



Length - Stay 

(year:month) 


English 
Overall PD NA 




Korean 
Overall PD NA 


4:01 I-l 
1:05 1-2 


93.75% 88.89% 100.00% 
6.25% 0.00% 14.29% 




25.00% 33.33% 14.29% 
100.00% 100.00% 100,00% 



For I-l, it is apparent that there has been counter-transference 
from the English system to the Korean system. Although his respons- 



Baik & Shim: Yes, we have no bananas 53 

es do not fit into the general picture obtained from this study, we 
cannot rule out the possibility that using the English positive- 
negative system may sometimes result in mental restructuring of a 
Korean-English bilingual's answering system that could cause com- 
municative problems in the bilingual's native language. In the case of 
1-2, the results are completely the opposite. He showed 100% 
accuracy for the Korean questions while he was able to answer only 
one question correctly in English. Keeping in mind that his general 
level of proficiency in English placed him in the proficiency group, it 
is rather strange that he did not even match the performance of the 
non-proficiency group. An interesting observation from the cases is 
that both subjects seem to have more difficulties in the PD category 
than in the NA category. While this could be mere coincidence, it 
could also be a reflection of the acquisition sequence of the two 
categories. 

A tentative interpretation that can be obtained from the above 
idiosyncracies is that there are some individuals who fail to maintain 
two opposite systems of interactions at the same time. In one case, 
the consequence was in the restructuring of th6 whole system to fit 
the system of the second language. In the other case it resulted in 
maintaining the native language system at the expense of dis- 
regarding the system of the second language to the extent that the 
acquisition rate for the system of the second language was consid- 
erably slowed down. It is possible that 1-2 is still in the process of 
acquiring the English system. It may be taking him longer to acquire 
the different system, and once he does acquire it, he could follow the 
pattern shown by I-l and lose his proficiency in the Korean system. 

According to the results of this study, we can say that most 
proficient bilinguals would simply say 'Yes' for positive disagreement 
in the majority of instances and be correct to a certain degree. As far 
as prescriptive grammar is concerned, this is the correct answer. 
However, prescriptive grammar does not mention the fact that one 
needs to elaborate when the answer is a positive disagreement. 
When non-native speakers fail to give any elaboration, monolingual 
native speakers may feel that the response of the non-native 
speaker was rude and curt. Although propositional meaning may be 
conveyed by the simple 'Yes' for positive disagreement, it unfortu- 
nately conveys an additional message that is certainly not part of the 
answerer's intent. 

A second point to be noted is the fact that non-proficient 
bilinguals would say 'Yes' to negative tag questions, not as positive 
disagreement, but as negative agreement more than 50 percent of 
the time. The fact that they do answer the questions correctly some 
of the times adds to the difficulty of interpreting their answers. The 
consequences in cross-linguistic communication are predictable. Un- 
less the native-speaker is aware of the communicative situation (that 
there is a non-native speaker involved who may use an opposite 



54 Studies in the Linguistic Sciences 23:1 

answering system for negative questions), the native-speaker will 
make judgments about the non-native speaker similar to that of the 
the FBI agents' accusations of perjury against the Philippino doctor. 

The issue that we would like to raise at this point concerns the 
accuracy rate of both groups in Korean questions. We have estab- 
lished that level of proficiency could not be a factor in their less than 
perfect performance. One relatively uninteresting reason behind this 
result may be that the Korean test was administered right after the 
English test: It could simply be a testing effect. 

If the above factor does not account for the inaccuracies 
observed, a second straightforward cause could be the fact that all 
the subjects in this study were bilinguals living in a second language 
environment, regardless of their level of proficiency. The simple fact 
that they are exposed to a different system could have resulted in an 
error range in their native system. 

A third explanation, which we advocate is that the level of 
accuracy obtained in this study is simply a reflection of real language 
use by monolingual Koreans. In other words, we believe that even 
monolingual native-speakers of a language can sometimes give an 
answer that is different from what is prescribed in the grammar. It 
is not a reflection of their competence but a reflection of what is said 
in everyday communication. 

In a language such as Hindi, Bhatia (1974) reports the co- 
existence of three different systems of answering negative questions. 
The questions is: How do people understand what they mean? It is 
apparent from Bhatia's account of the phenomenon that the answers 
are rarely misunderstood because of various pragmatic presup- 
positions, implications, and expectations (PIEs) governing the use of 
negative questions and their answers in Hindi. This does not mean 
that Hindi speakers do not have a preferred norm. Bhatia reports 
that the positive-negative system is the predominant strategy used 
by Hindi speakers. Nevertheless, there are instances when Hindi 
speakers would use the agreement-disagreement system and suc- 
cessfully convey what they mean. 

Likewise, Americans do sometimes give answers in agreement- 
disagreement while communicating with other Americans, and they 
usually figure out the correct meaning from the context of situation. 
Koreans, too, may give answers following the positive-negative 
strategy while communicating with other Korean speakers. Although 
further studies need to be done to confirm our belief, observations of 
language use in real-life contexts lead us to believe that this is what 
is reflected in our data. 



Baik & Shim: Yes, we have no bananas 55 

8. Summary and conclusions 

In summary, the results of this study have indicated the 
following: 

1. The level of accuracy in answering negative questions by non- 
native speakers of English corresponds to their general level of 
proficiency. 

2. Non-native speakers of English who are accustomed to the 
agreement-disagreement system in their native language 
simply apply the opposite rule in answering negative questions 
in order to be accurate, but they are generally unaware of the 
pragmatic constraint on positive disagreement answers. 

3. The attainment of proficiency in a second language with a 
different system of answering negative questions does not 
adversely affect the bilingual's performance in his/her native 
language. 

Based on these results we have argued that proficiency in the 
answering system of the second language in terms of following the 
prescriptive rules does not necessarily guarantee successful commu- 
nication. Further we have pointed out that non-proficient bilinguals 
will have predictable difficulties in getting their meanings across 
when the native-speaker interactant is not aware of the cross- 
linguistic variables. Third, we postulated that competence in one lan- 
guage system may not mean 100 percent accuracy in performance. 

These discussions lead us to ponder upon the arguments by 
Kachru (1985) and Nelson (1985) that the difficulty in understanding 
institutionalized non-native varieties of English may be the result of 
attitudinal baggage rather than true unintelligibility. This may also 
be true of language learner varieties, especially in the case of an- 
swering negative questions where the source of the problem may be 
in intolerance stemming from prescriptivism. 

9. Implications for further research 

The discussion from the present study is limited to the perfor- 
mance of bilinguals that are aware of two systems of answering 
negative questions that are completely opposite to each other. We 
have found that the bilinguals do not perform at 100 percent accu- 
racy rate in their native language. We have determined that this 
could not be caused by gaining proficiency in the second language. As 
we have already discussed, it is important to find out if native 
speakers of a language actually follow the system of the specific 
language a hundred percent of the time. It may be possible that the 
90+% accuracy obtained in this study is representative of the native 
monolingual speakers' performance. However, this is only a 
speculation at this point. Further studies are needed to confirm this 
observation. 

One of the limitations of this study lies in the scope of the 
negative questions asked. We concentrated only on opposite-polarity. 



56 Studies in the Linguistic Sciences 23: 1 

falling-intonation negative questions. The reason for limiting the 
scope of this study was that the pragmatic constraints for these 
questions have already been well defined. We feel that similar 
research needs to be carried out for simple negative questions with- 
out tags. The pragmatic conditions differ, and change in intonation 
plays an important role in conveying the presuppositions, impli- 
cations, and expectations (PIEs) of the questioner. Undoubtedly these 
different PIEs could certainly have an effect on the strategies of 
answering if a speaker uses the agreement-disagreement system. 

The challenge in second language acquisition is not only in 
acquiring the syntactic structure of the language, but also in acquir- 
ing communicative competence in the language. We believe that the 
knowledge of implicit PIEs in negative questions is a significant part 
of such communicative competence. 

NOTES 

This paper was presented at the 7th Annual International 
Conference on Pragmatics and Language Learning, University of 
Illinois at Urbana-Champaign, April 1-3, 1993. 

1 Since tests of variance led us to reject the null hypothesis that 
the group variances were the same, the t-tests conducted were also 
tests comparing two means from samples of different variances. 

REFERENCES 

BelluGI, U. 1967. The acquisition of the system of negation in 
children's speech. Harvard University Ph.D. dissertation. 

Bhatia, T. K. 1974. The coexisting answering systems and the role of 
presuppositions, implications, and expectations in Hindi simplex 
Yes/No questions. Papers from the Tenth Regional Meeting 
Chicago Linguistic Society (CLS), 47-61. 

CORDER, S. P. 1971. Idiosyncratic dialects and error analysis. 
International Review of Applied Linguistics 9.147-159. 

Cow AN, R. 1983. Toward a psychological theory of interference in 
second language learning. Second language learning: 
Contrastive analysis, error analysis and related aspects, ed. by 
B. W. Robinett and J. Schachter, 109-119. 

DULAY, H. C, & M. Burt. 1972. Goofing: an indication of children's 
second language learning strategies. Language Learning 
(Urbana, IL) 22.235-252. 

. 1974. Errors and strategies in child second language acqui- 
sition. TESOL Quarterly 8.129-136. 

. 1975. Creative construction in second language learning. Papers 

in second language acquisition: Proceedings of the Sixth Annual 
Conference on Applied Linguistics, ed. by H. D. Brown, 65-80. 
Ann Arbor, Michigan: University of Michigan Press. 



Baik & Shim: Yes, we have no bananas 57 

DULAY, H. C, M. Burt, & S. Krashen. 1982. Language two. New York: 

Oxford University Press. 
ECKMAN, F. R. 1977. Markedness and the CA hypothesis. Language 

Learning (Urbana, IL) 27.315-330. 
GUMPERZ, J. J. 1982. Fact and inference in courtroom testimony. 

Language and social identity, ed. by J. J. Gumperz, 163-194. 

Cambridge: Cambridge University Press. 
HOUCK, N. 1991. Tag questions: a necessary pragmatic context. 

Pragmatics and Language Learning 2.29-38. 
KaCHRU, Y. 1985. Discourse analysis, non-native Englishes, and second 

language acquisition research. World Englishes 4:2.223-232. 
Nelson, C. L. 1985. My language, your culture: whose communicative 

competence? World Englishes 4:2.243-250. 
POPE, E. N. 1976. Questions and answers in English. The Hague: 

Mouton. 
Richards, J. C. 1971. A non-contrastive approach to error analysis. 

English Language Teaching 25.204-219. 
Selinker, L. 1972. Interlanguage. International Review of Applied 

Linguistics 10.210-231. 

APPENDICES 

Appendix A: English questionnaire 

Thank you very much for participating in this little survey. This is a 
survey to find out the level of information that is available to the 
public on four widely known commercial products: the Levis jeans, 
the Coca-Cola drink, diamonds, and Toyota cars. 

Please fill in the information on this page. In the following pages, you 
will find four pages of advertisement on the above four products. 
You will be allowed to study them at your leisure. Then you will be 
asked to answer several questions on each of the products. 

There are two types of questions. The first type of question asks for 
a Yes, No, Don't know ,or Don't understand question answer. There 
will be a time limit on each of the questions (i.e., you will be given 
approximately seven seconds to answer each question). 

Example question: 

Please circle one of the answers. 

Have you heard of the Levis jeans? 

Yes No Don't know Don't understand question 

In order to ensure your anonymity in the answers that you will give 
in this survey, we ask that you do not identify yourself on any part 
of the survey itself. Instead, we encourage you to leave your name 
and address with us on a separate information sheet so that we can 
inform you about the purposes of the survey after we have collected 



5 8 Studies in the Linguistic Sciences 23: 1 

all the data. Since the data will be collected from many different 
people on many different occasions, we are sorry that we cannot give 
out that information at this time. 

1. Length of Stay in the United States: years months 

2. Familiarity with products in survey: 

Very familiar Slightly familiar Unfamiliar 

Levis Jeans 3 2 1 

Toyota Cars 3 2 1 

Diamonds 3 2 1 

Coca-Cola 3 2 1 

Questions on Levis Jeans 

1. Is the Levis Jeans a popular brand in Korea? 

Yes No Don't know Don't understand question 

2. Jeans are not popular among young people, are they? 

3. Do you think jeans are durable? 

4. People do not usually wear jeans on formal occasions, do they? 

5. Do women like to wear jeans as much as men? 

6. Elderly people in Korea do not usually wear jeans, do they? 

7. Do you think Levis jeans are expensive? 

8. Office workers in Korea do not wear jeans to work, do they? 

Questions on Toyota cars 

1. Toyota is not made in Korea, is it? 

2. Do you like Toyota cars? 

3. Corolla is not a model of Toyota, is it? 

4. Are Toyota cars generally more expensive than BMW cars? 

5. Toyota cars are generally not cheaper than Hyundai cars, are they? 

6. Have you seen this advertisement before? 

7. Toyota is not a Japanese brand, is it? 

8. Do you think the new aero-dynamic model from Toyota is 
attractive? 

Questions on Diamonds 

1. Is the diamond the most expensive jewel? 

2. The pearl is not as expensive as the diamond, is it? 

3. The diamond is supposed to be forever, isn't it? 

4. Most people do not use the diamond as wedding gifts, do they? 

5. Will the diamond break if it falls on rocks? 

6. Women generally do not like diamonds, do they? 

7. Is the size of the diamond an important factor to the price? 

8. The clarity of the diamond is not important when choosing a 
diamond, is it? 

Questions on Coca-Cola 

1. Coca-Cola is not made from fruit juices, is it? 

2. Is Pepsi-Cola the major competition of Coca-Cola? 



Baik & Shim: Yes, we have no bananas 59 

3. There is no Diet form of Coca-Cola, is there? 

4. Do you know that there is an RC brand cola? 

5. Coca-Cola is not as famous as RC-Cola, is it? 

6. Do you think it is not healthy to drink too much Coca-Cola? 

7. Coca-Cola company does not make caffein-free products, do they? 

8. Did you know that a can of Coca-Cola contains twelve teaspoons of 
sugar? 

Appendix B: Interview questions 

0. Do you have any general comments about the test? Any 
questions? 

1 . Were the questions difficult to answer? 

2. Was the source of difficulty in the question itself or in the lack of 
information regarding the products? 

3. Did any of you answer 'Don't know' when in fact you knew the 
answer but did not know how to answer the question? 

4. We will give you the answers that we feel are correct. Please tell 
us if you disagree. (Go through the questions and discuss their 
answers.) 

5. Did any of you guess at what the purposes of this test could be? 

6. Did you feel more constrained in giving answers of positive 
disagreement, either in English or Korean? (Explain the meaning 
of positive disagreement.) 

7. What is the strategy that you use in answering English negative 
questions? 

8. Do you give prepositional statements in addition to Yes/No 
answers? 

9. Arc you aware that you are required to give an extended answer 
when the your answer is 'Yes' in positive disagreement? 

10. When you answer Korean negative questions, do you ever find 
yourself using the English system, rather than the Korean 
system? 

11. Have you ever had the experience of miscommunication either 
in English or Korean because you were using a different system of 
answering than that of the questioner? 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



THE STATUS OF AGREEMENT 
AND THE AGREEMENT PROJECTION IN ARABIC 

Elabbas Benmamoun 

This paper deals with two main issues that revolve 
around the problem of subject agreement. The first issue is 
whether there is one single pattern of subject agreement. 
The second issue is whether there are any motivations for 
the claim that agreement heads an independent syntactic 
projection. As far as the first issue is concerned, I show that 
Arabic evidence supports the claim that there are two 
independent agreement relations involving the subject and 
the verb, namely person and number. With regard to the 
second issue, I argue that there is no evidence that agree- 
ment is involved in nominative Case assignment in Arabic. 
This weakens the claim for an independent agreement 
projection, since Case has been the main syntactic reason 
for the postulation of such a projection. 

1. Introduction 

Pollock (1989) suggests an agreement projection located right 
above VP and below TP (and NegP). This raises two questions that 
have been left relatively open in most of the recent treatments of 
functivinal categories in general and of subject agreement in 
particular (Ouhalla 1988, Mahajan 1989, Chomsky 1991, Belletti 
1991, among many others). 

The first question concerns the internal structure of the agree- 
ment morpheme (its feature composition). The underlying assump- 
tion is that the node Agr contains Phi features of person, number, 
and gender which are realized by an affix under an agreement 
relation with the subject. I will present data that show that there are 
two agreement patterns in Arabic, one in number and gender and 
the other in person and gender. A verb, for example, may show both 
patterns or just one of them. 

The second question has to do with whether there are any 
conceptual and empirical reasons for postulating an independent 
agreement projection beyond the need to have a place holder to 
derive the right word order. Again the facts from Arabic do not 
support a positive answer to this question. The main focus in this 
paper will be on Arabic and particularly on Standard Arabic and 
Moroccan Arabic. 



62 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

2. Agreement features 

Consider the following sentences from Standard Arabic: 

(1) a. daxal-a t-tullaab-u 

enter-3SM the-students-M-Nom 
'The students entered.' 
b. daxal-at t-taalibaat-u 

entered-3SF the-students-F-Nom 
'The students entered.' 

(2) a. t-tullab-u daxal-uu 

the-students-F-Nom entered-3PM 
'The students entered.' 
b. t-taalibaat-u daxal-na 

the-students-F-Nom entered-3PF 
'The students entered.' 

(3) a. kaan-a t-tulaab-u ya-drus-uun 

was-3SM the-students-M-Nom Imp-3M-study-PM 
'The students were studying.' 
b. kaan-at t-taalibaat-u ya-drus-na 

was-3SF the-students-F-Nom Imp-3F-study-PF 
'The students were studying.' 

As is well known, when the verb precedes the subject, agreement is 
in gender (and person) only, as shown in (1). When the subject 
precedes the main verb as in (2), agreement is in person and 
number, in addition to gender. Moreover, in (3) the subject agrees 
with the auxiliary that precedes it in person and gender only.' 

The main question is how to account for the subject-verb 
agreement feature alternation. If we confine our attention to (1) and 
(2) we may account for the facts by positing two agreement affixes. 
One affix contains gender and person and is realized under the VS 
order as in (1). Another agreement affix contains number, person, 
and gender and is realized under SV order as in {1)} 

Another way to implement the same basic idea would be to 
posit one abstract agreement affix and allow it to be either partially 
specified (gender and person) or fully specified (number, gender and 
person) depending on the structural relation that obtains between 
the verb and the subject. According to this account the Phi features 
are encoded on one affix, with full or partial specification contingent 
on the relative order of subject and verb. 

(4) a. V S O ---> V-hAGR(P.G) 
b. S V O ---> V+AGR(N.P.G) 

However, this account cannot carry over to the embedded verb 
in (3), where the imperfective form of the lexical verb carries two 
agreement affixes. The prefix carries person and gender, and the 
suffix carries number and gender. Moreover, the suffix shows up 
only when the subject precedes the verb (3). Thus, (3) shows that we 



Benmamoun: The status oiagiceincnt and the agreement projection in Arabic 



63 



are not dealing with one agreement affix but rather with two affixes 
that happen to overlap in the feature gender. In other words, 
number and person agreement do not necessarily belong to the same 
morpheme, but rather constitute two independent morphemes. A 
plausible representation could be as shown in (5). 

(5) 



Agr 




Agr 
~P~ 


G 




G 



Further support for this conclusion comes from copular con- 
structions. In Standard Arabic (and in fact in all the dialects that I 
know of, including Moroccan Arabic), the element corresponding to 
the verb be in the present tense is realized as a pronominal. I follow 
Eid (1991) and refer to this element as a pronoun copula. The use of 
this term is for purely descriptive purposes. For detailed analyses of 
this element in Arabic the reader is referred to the important work 
of Eid (1991), who, as far as I know, was the first to point out the 
interesting agreement patterns in (6). The examples are from 
Moroccan Arabic (disregarding phonological details). 

(6) a. Omar huwa mul d-dar 

Omar 3MS owner the-house 
'Omar is the owner of the house.' 

b. Salmaa hiya mulat d-dar 
Salmaa 3FS owner-F the-house 
'Salmaa is the owner of the house.' 

c. Omar wa Sami hum mmalin d-dar 
Omar and Sami 3MP owners the-house 
'Omar and Sami are the owners of the house.' 

Confining our attention to the agreement features on the 
pronominal copula, notice that in (6) the copula apparently agrees 
with the subject in person, gender, and number. However, it is easy 
to show that agreement in person in the above examples is not with 
the subject but is a default feature on the copula. In (7), where the 
subject is a first or second person pronoun, the copula agrees with 
the subject only in gender and number; the person marking remains 
the same as in (6), viz. third person. There is no agreement in person 
as shown by the ungrammaticality of (8): 

(7) a. 7anaa huwa mul d-dar 

I-IS 3MS owner the-house 
'I am the owner of the house.' 
b. Vanti hiya mulat d-dar 

Y0U-2FS 3FS owner-F the-house 
'You are the owner of the house.' 

(8) a. *7anaa 7anaa mul d-dar 

I-IS IS owner the-house 

'I am the owner of the house.' 



64 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

b. *7anti 7anti inulat d-dar 

You-2FS 2FS owner-F the-house 
'You are the owner of the house.' 

The same agreement facts obtain with participial predicates, 
which agree with the subject in number and gender only: 

(9) a. kaan-a 7al-7atfaal-u naa7im-iin 

was-3SM the-children-Nom sleep-PM-Acc 
'The children are sleeping.' 

b. kaan-a 7al-tifl-u naa7iin-an 

was-3SM the-child-SM-Nom sleep-SM-Acc 
'The child is sleeping.' 

To sum up, the facts from the imperfective form of the verb, 
copular constructions, and participles strongly support the conclusion 
that there are two independent subject-verb agreement relations in 
Arabic, encoded by different features on different affixes; one 
number and gender affix which shows up under the SV order only, 
and another person and gender affix which shows up under both VS 
and SV orders.-^- "* 

The recognition of two agreement patterns in Arabic puts the 
debate over the agreement relation in general and the agreement 
projection in particular in a different perspective. Under an analysis 
where the derivation of inflectional morphology is derived in the 
syntax by the process of head movement that adjoins the lexical 
head to the inflection, one may need to posit more than one 
agreement projection. This may not be a negative move after all as 
long as we can conceptually and empirically ground the postulation 
of an agreement projection. The question that arises then is whether 
we can conceptually ground the agreement projection. This question 
has been addressed by latridou (1991) in the context of verb 
movement and adverb placement,- and by Rouveret (1991) in the 
context of his analysis of agreement in Welsh. In the next section I 
address this question from a different perspective, namely its role in 
Case assignment. 

3. Agreement as a projection 

The second important issue that arises in the context of any 
discussion of agreement is the problem of projection. This issue has 
not been subjected to much debate, and it seems that this has to do, 
at least partially, with the still unclear role and status of agreement. 
Whether there is an agreement projection or not raises many ques- 
tions concerning the necessary conditions for projecting a functional 
category. If we compare agreement on one hand and tense and 
affixal negation on the other, we realize that there is good evidence 
that tense and negation need to be syntactically accessible as 
independent projections rather than as features on a lexical head 
(see Benmamoun 1992 for discussion). As far as tense and negation 



Benmamoun: The status of agreement and the agreement projection in Arabic 65 

are concerned the evidence, both syntactic and semantic, for projec- 
tion seems to be strong. 

Consider negation for example. It is semantically relevant be- 
cause it contributes to the truth value of a proposition (10): 

(10) a. Omar qra le-ktaab 

Omar read the-book. 
b. Omar ma-qra-s le-ktaab 

Omar Neg-read-Neg-'' the-book 
'Omar did not read the book.' 

It is also syntactically active. For example, it licenses negative 
polarity items (11) and interacts with quantifiers to give a narrow 
scope reading for the quantifier in (12): 

(11) a. ma-zaa Hatta wahed 

Neg-came any one 
'No one came.' 
b. ma-qra Hatta ktaab 
Neg-read any book 
'He did not read any book.' 

(12) a. ma-qra-s bazaaf d-le-ktuba 

Neg-read-Neg. many of-the-books 
'He did not read many books.' 

All these facts suggest that the negative is syntactically visible 
and accessible. As to whether it heads an independent projection, the 
morphological evidence of Standard Arabic shows that the negative 
carries tense (Benmamoun 1992): 

(13) lam yaktub 
Neg-Past write-3SM 
'He did not write' 

Sentences such as (13) can be accounted for by considering the 
negative as head of a projection between TP and VP. As such it 
blocks verb movement to Tense (due to minimality by acting as a 
potential antecedent in the sense of Aoun & Li 1989 and Rizzi 1990, 
see Benmamoun 1992 for details). Consequently, the tense inflection 
is hosted by the negative head (14). Notice the fact that negation 
hosts temporal information supports its status as head of a 
projection. 

Turning to agreement, it seems to behave differently from the 
other heads. Its semantic function, if any, is not entirely clear. 
Syntactically, one possible motivation for an agreement projection 
can be found in Chomsky 1993, where subject- verb agreement, for 
example, is taken as an expression of Case assignment (also Borer 
1986). However, given the theory that tense and agreement each 
head their own projection, the problem arises about how to derive 
the generalization that tense seems to play a role, given the fact that 



66 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

(14) 




nominative Case usually (but not exclusively) obtains in tensed 
clauses. According to Chomsky (1993) this can be derived by raising 
Tense to Agr (Agr-S). This insures that the Case relation is given its 
full expression under Spec-head relation mediated via the Agr head. 
This amounts to saying that nominative Case requires that the 
subject be in Spec-head agreement relation with both tense and 
agreement. The requirement obtains by combining agreement and 
tense. *^ 

However, there are three contexts where this does not seem to 
be the case. First, the requirement is obviously not fulfilled in 
languages where the combination of tense and agreement fails. This 
is the case of negative sentences in Arabic where the negative carries 
tense and the verb carries agreement as in (13). Second, in copular 
constructions, the subject gets nominative Case even in contexts 
where there is no agreement that would mediate Case-assignment: 

(15) t-taalib-u fii 1-bayt-i 
the-student-Nom in the-house-Gen 
The student is at home.' 

We could obviously posit an agreement projection headed by an 
empty head. However, such a move makes one wonder why such an 
abstract head does not induce processes and repair strategies akin to 
Do-support in English when there is no eligible host, as is clearly the 
case in copular constructions in the present tense in Arabic. 

Third, there are cases where the NP is assigned Case by one 
head while it agrees with another head. In the context of Q-float 
(Benmamoun 1993) the NP agrees with Q, though it is assigned Case 
by the external head (Joseph Aoun p.c): 

(16) raVaytu l-7awlaad-a kulla-hum 
(I) saw the-children-Acc all-Agr 

'I saw all the children.' 

Thus, nominative Case assignment can take place in the absence of an 
agreement feature on the Case-assigning head.^ 

Nevertheless, there is a solid generalization that an intimate 
relation exists between tense and agreement in languages like 



Benmamoun: The status of agreement and the agreement projection in Arabic 6 7 

English, where absence of tense entails absence of agreement. This is 
also reflected in contexts of sentential negation and questions, where 
Do-support affects tense and agreement. 

1 would like to suggest that the agreement feature may be part 
and parcel of the morphological template of heads such as tense, 
verbs, negation, questions.^ Thus, in English, tense is specified for 
agreement while lexical categories such as verbs and adjectives are 
not. On the other hand, in Arabic, tense is not specified for agreement 
but only lexical categories and some functional categories such as the 
negative laysa.^ This implies that in contexts where verb movement 
does not take place we correctly predict the verb to carry agreement 
in Arabic but not in English. Conversely, in such contexts, we cor- 
rectly expect the categories that support tense to carry agreement in 
English but not in Arabic. This is indeed the case with tensed nega- 
tives in Standard Arabic where the tensed negative does not carry 
agreement (Benmamoun 1992). 

Consequently, the agreement relation that obtains in English 
between the subject and the verb is indeed a reflection of nomina- 
tive Case assignment but only because the agreement feature is part 
of the lexical specification of tense in that language. ^^ 

If this conclusion is correct, together with the arguments that 
Case is not mediated by an agreement morpheme (head of an 
agreement projection), we can conclude that agreement does not 
seem to correspond to a syntactic projection that would be treated on 
a par with other projections such as tense and negation.' ' 

NOTES 

' For detailed studies of agreement in Standard Arabic see 
Ayoub 1981, Fassi Fehri 1988, and Mohammad 1990, and the 
references cited there. 

2 It should be mentioned that not all analyses assume that there 
is person agreement with the postverbal subject. Thus, for example, 
in (i) it is assumed that the suffix -at carries gender only. This 
assumption can be problematic given (ii), where the same suffix is 
used in the context of a null pronominal. 

(i) ?aad-at Nadia 
return Nadia 

(ii) ?aad-at 
returned 
'She returned.' 

Under the theory of identification of null thematic pronominal 
elements, the person feature has to be present for the null 
pronominal to be identified. The crucial role of person agreement is 
shown by the fact that the subject cannot be dropped in the context 
of a predicate that does not contain the person feature as shown by 



68 



Studies in the Linguistic Sciences 23:1 (Spring 1993) 



(iii) (see Kenstowicz 1989; see also Borer 1984, 1986 and Shlonsky 
1989 for similar facts in Hebrew). This entails that the agreement 
suffix in (ii) contains the person feature to identify pro: 

(iii) *(Omar) naaVim 
Omar sleeping 
'Omar is sleeping.' 

If the suffix -at contains the person feature in (ii) as required by the 
identification condition on null pronominals (in Arabic at least), and 
since the same affix is used in (i), then it is safe to conclude that the 
latter carries person agreement as well. See also Ayoub 1981 where 
it is assumed that under VSO the verb has both person and gender 
agreement. 

3 The suffix cannot be considered a resumptive pronoun related 
to a topicalized or left-dislocated NP. The main reason against this 
view is that the suffix lacks the main pronominal feature, which is 
person. Moreover, in (3) there is no representation in which the NP 
subject is left-dislocated or topicalized. Also, the suffix does not 
constitute together with the prefix a discontinuous morpheme. If we 
assume that it is a discontinuous morpheme, we still have to explain 
why the second part of this morpheme does not show up under VS. 
This is apart from the problems of representation and derivation that 
arise if a discontinuous morpheme is posited. 

'* Shlonsky (1989) independently argues for separating agree- 
ment features. He proposes that the features person, number, and 
gender each head an independent projection. 

5 I gloss -s as Neg. though I remain agnostic about its syntactic 
status. See Benmamoun 1992. 

6 This essentially goes back to the LGB-type analysis where 
nominative Case is assigned by INFL which contains both tense and 
agreement. 

^ Another variant of the same facts involves multiple agreement 
in the same clause. In Arabic both the auxiliary and the verb agree 
with the subject as illustrated in (i) from Moroccan Arabic: 

(i) a. lewlaad kaanu taylaVbu 

the children be-Past-P play-P 
'The children were playing.' 
b. kaanu lewlaad taylaVbu 

be-Past-P the children play-P 
'The children were playing.' 

For an analysis of subject agreement in Moroccan Arabic and its 
comparison with Standard Arabic see Benmamoun 1992. See also 
Aoun, Benmamoun, & Sportiche forthcoming. 

^ This is not a new idea. It is essentially the content of the tra- 
ditional analysis of the agreement relation as concord between two 
lexical elements (adjective and noun for example). The current 



Bcnmamoun: The status of agreement and the agreement projection in Arabic 6 9 

implementation of this idea consists of restricting the agreement 
relation to take place under only one of the three grammatical rela- 
tions within the X' schema, namely Spec-head (Chomsky 1993). 

9 A more radical proposal is made by Rouveret (1991:7) who 
argues that agreement morphology can only attach to a functional 
head: 'La morphologie d'accord ne pent etre affixe qu'a une tete 
fonctionelle.' 

10 In her GLOW 1991 abstract, Ritter accounts for the fact that in 
Hebrew, in the present tense the verb agrees with the subject in 
number and gender only, while in the past and future it agrees in 
person, number, and gender, by suggesting to specify agreement as 
[gender, number] in the present (her Num) and as [person, number, 
and gender] in the past and future (her D). 

' ' Notice that even if subject agreement heads its own projec- 
tion, the facts from Standard Arabic suggest that it should be located 
below negation; otherwise verb movement to agreement will be 
blocked for the same reason that verb movement to tense in the 
context of the negative laa is blocked in Standard Arabic (see 
Benmamoun 1992). This is another problem that arises whenever 
agreement is projected as the head of its own syntactic projection. 
One then is forced to locate it in different positions relative to the 
other functional projections depending on its morpho-phonological 
distribution in the language in question (in a kind of mirror principle 
fashion). This may well be the case, but what is needed are the 
theoretical principles that underly these 'parametric' choices. The 
mirror principle of Baker (1988), that is implicit or explicit in most 
analyses of inflectional and functional categories, cannot be used in a 
principled way as long as the syntactic role of agreement as a 
projection is not well defined. The mirror principle deals primarily 
with the isomorphism between the relative order of morphemes in a 
phonological string and their (relatively) well defined syntactic scope 
(e.g, passives, causatives, reciprocals ...). It is far from clear that the 
same applies to, say, the relative ordering of tense and agreement, or 
agreement and negation. 

REFERENCES 

AOUN, J., & A. Li. 1989. Scope and constituency. Linguistic Inquiry 

20.141-172. 
, E. Benmamoun, & D. Sportiche. Forthcoming. Agreement and 

conjunction in some Arabic dialects. Linguistic Inquiry 25. 
AYOUB, G. 1981. Structure de la phrase en Arabe Standard. Universite 

de Paris VII Ph.D. dissertation in Linguistics. 
BahLOUL, M., & W. Harbert. 1993. Agreement asymmetries in Arabic. 

Proceedings of the Eleventh West Coast Conference on Formal 

Linguistics, ed. by J. Mead, 15-31. Stanford, CA: Center for the 

Study of Language and Information. 



70 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Baker, M. 1988 incorporation: A theory of grammatical function 

changing. Chicago: University of Chicago Press. 
BELLETTI, a. 1990. Generalized verb movement. Torino: Rosenberg & 

Sellier. 
Benmamoun, E. 1992. Functional and inflectional categories: Problems 

of projection, representation, and derivation. University of 

Southern California Ph.D. dissertation in Linguistics. 
. 1993. Null pronominals in the context of NP and QP. 

Proceedings of the Eleventh West Coast Conference on Formal 

Linguistics, ed. by J. Mead, 32-43. Stanford, CA: Center for the 

Study of Language and Information. 
Borer, H. 1984. Parametric syntax. Dordrecht: Foris. 

. 1986. I-Subjects. Linguistic Inquiry 17.375-416. 

Chomsky, N. 1991. Some notes on the economy of derivation and 

representation. Principles and parameters in comparative 

grammar, ed. by R. Freidin, 417-454. Cambridge, MA: MIT Press. 
. 1993. A minimalist program for linguistic theory. The view 

from Building 20: Essays in linguistics in honor of Sylvain 

Bromberger, ed. by K. Hale & S. J. Keyser. Cambridge, MA: MIT 

Press. 
ElD, M. 1991 Verbless sentences in Arabic and Hebrew. Perspectives 

on Arabic linguistics III, ed. by B. Comrie & M. Eid, 31-62. 

Amsterdam: John Benjamins. 
. 1992. Pronouns, questions, and agreement. Perspectives on 

Arabic linguistics IV, ed. by M. Eid, J. McCarthy, & E. Broselow. 

Amsterdam: John Benjamins. 
Fassi Fehri, a. 1988. Agreement in Arabic, binding and coreference. 

Agreement in natural languages: Approaches, theories, 

descriptions, ed. by M. Barlow & C. Ferguson, 107-158. Stanford, 

CA: Center for the Study of Language and Information. 
. 1989. Generalized IP structure, case, inflection, and VS word 

order. MIT Working Papers in Linguistics, 10.75-111. 
lATRlDOU, S. 1990. About Agr(P). Linguistic Inquiry 21.551-577. 
Jaeggli, O., & K. Safir. 1989. The null subject parameter and 

parametric theory. The null subject parameter, ed. by O. Jaeggli 

& K. Safir, 1-44. Dordrecht: Kluwer. 
KenstowiCZ, M. 1989. The null subject parameter in modern Arabic 

dialects. The null subject parameter, ed. by O. Jaeggli & K. Safir, 

263-275. Dordrecht: Kluwer. 
Mahajan, a. 1989. Agreement and agreement phrases. MIT Working 

Papers in Linguistics 10.217-252. 
Mohammad, M. 1990. The sentence structure of Arabic. University of 

Southern California Ph.D. dissertation in Linguistics. 
OUHALLA, J. 1988. The syntax of head movement: A study of Berber. 

University College, London, Ph.D. dissertation in Linguistics. 
POLLOCK, J-Y. 1989. Verb movement, UG, and the structure of IP. 

Linguistic Inquiry 20.365-424. 
RiTTER, E. 1991. Evidence for number as a nominal head. Generative 

Linguistics of the Old World (GLOW). Abstract. 



Benmamoun: The status of agreement and the agreement projection in Arabic 7 1 

RlZZI, L. 1990. Relativized minimality. Cambridge, MA: MIT Press. 
ROUVERET, A. 1991. La nature des prepositions conjuguees. Universite 

de Paris VIII, MS. 
SHLONSKY, U. 1989. The hierarchical representation of subject-verb 

agreement. University of Haifa, MS. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



EXAMINING THE TRUSTWORTHINESS OF THE LATEST OED 
IN REFLECTING CURRENT ENGLISH 

Marvin K. L. Ching 
Department of English, Memphis State University 

Although the second edition of the Oxford English 
Dictionary (OED) is still the paragon of lexicographical 
achievement, it suffers from seven serious flaws which 
must be rectified in future editions if it is to continue its 
supremacy. These deficiencies, though named by other 
writers with brief examples, are illustrated with a multi- 
plicity of examples in this study to allow readers to 
understand the magnitude of the task of revision and to 
decide for themselves the pragmatic possibility or impossi- 
bility of achieving the OED's, goals of accuracy and compre- 
hensiveness in future editions. It is easy to accede to the 
need for revision according to principles, but perhaps the 
size of the data nullifies the possibility of accomplishment, 
unless an innovative way is found to collect, organize, and 
revise the data. 

1.0 Introduction 

Through an investigation of contemporary American dictionar- 
ies. The Oxford Dictionary of New Words (1991), as well as words and 
expressions I have encountered in print and in other media and from 
actual speech, this study examines the second edition, the 1989 
version, of the venerable OED for its dependability in reflecting cur- 
rent English. The OED, though still venerable, is lacking in: 1) the 
inclusiveness of terms cited, 2) the updating of the newest meanings 
of old terms, 3) the updating of quotations, 4) the clarity of defini- 
tions, 5) social sensitivity to changes in world view when defining 
terms, 6) usage labels or usage notes, and 7) a systematic method of 
listing terms and defining them for reader convenience. 

Although both Harvey (1991:84-86) and Algeo (1990:131-150) 
have cited a number of these same points, readers cannot fully grasp 
the magnitude — and thus the probability or improbability of 
accomplishment — of the task for future editions of the OED without 
more specific, detailed examples. Moreover, although it is easy to 
accede to the areas for revision Harvey and Algeo have named, with 
their brief citation of specific examples of shortcomings, readers need 
to see a multitude of data to question the present method of employ- 
ing paid and volunteer workers to accomplish the mammoth task of 
maintaining the OED, as Algeo has eloquently dubbed the OED, as the 



7 4 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

'emperor' and 'the crowning achievement" of lexicography (131). But 
the explosion of lexical terms together with the explosion of publica- 
tions in English are like the modern world's warp-speed changes 
predicted by Toffler's 1971 Future shock. Thus, after examining the 
strengths and liabilities of the new second edition of the OED, this 
study proposes an additional method for acquiring data to meet the 
demands of the times, if the OED is to remain the comprehensive and 
exemplary dictionary it has been. 

Many (though not exclusively all) of the specific terms examined 
are those used by the general populace, which this writer has come 
across. They cannot be dismissed merely as 'soft terms,' because the 
OED has shifted not only from the literary world to terms in science, 
business, and medicine, but also to North American popular terms 
and slang (e.g., brain-dead, nose job, acid rain, crack, asset stripping, 
barf, and drunk tank) in its second edition (Gray 1989:95, 98). But in 
examining the specific terms used, readers can judge for themselves 
whether the OED should limit its aims in comprehensiveness, leaving 
certain terms for specialized dictionaries, or whether the missing 
gaps should be filled. 

Details and examples of problems in scientific, business, and 
medical terminology will be left to future investigators. 

A cross-comparison of American dictionaries reveals the 
adequacy or inadequacy of the OED in the seven areas named at the 
beginning. The comparison uses three general standard American 
desk dictionaries and the unabridged Random House II (2nd ed., 
1987). They, together with their abbreviated names for this study, 
are as follows, and will be referred to as 'the four dictionaries': 

Webster's Ninth New Collegiate Dictionary (1991) or W^'''; 

The American Heritage Dictionary (2nd college ed., 1991) or 

AHD; 
Webster's New World Dictionary (3rd college ed., 1988) or 

WNWD; 
The unabridged Random House Dictionary of the English 

Language (2nd ed., 1987) or RHII. 

At times, reference shall also be made to these other works: 

The Oxford Dictionary of New Words (1991) or OD of NW; 
Webster's Third New International Dictionary (1971) or Will; 
Dictionary of American Slang ( 1975 ) or D of AS; 
New Dictionary of American Slang (1986) or ND of AS; and 
Webster's Dictionary of English Usage (1989) or WD of EU. 

The OD of NW is also used because it is a wonderful source for 
2,000 high-profile current words and phrases, only a fourth of which 
have been included in the new words and senses added to the OED. 
These dictionaries were not consulted in Algeo's study (1990:143). 



Ching: Examining the trustworthiness of the latest OED 7 5 

2.0 Strengths 

No doubt the OED is still a treasure. Its quotations indicate the 
context for the first recorded appearance of a term and its growth 
and spread to other contexts. For example, we find that 1952 is the 
year for the first appearance of Ms. when the National Office of Man- 
agement Association in Philadelphia in The simplified letter directed 
that the term be used for all women addressees. After the following 
year with another quotation showing its use by the same business 
association, eight quotations in the 1970s appear, indicating its use in 
the Women's Movement. Another social trend can be observed: 
Apparently, with the growth of crime, neighborhood was com- 
pounded with watch, producing neighborhood watch, which appeared 
in print in 1972. 

The OED is also important in dating the first use of a term in 
metaphorical contexts: e.g., headhunter 'a person recruiting a top 
executive officer' (1961); and litmus test 'a touchstone or decisive 
test' (with a cultural litmus test in 1957). The OED can thus trace new 
uses of old terms, correlating them with new trends in society. 

Moreover, the quotations also include interesting conjectures on 
the etymology of a word, when it is uncertain. For example, although 
the OED states that the etymology of jazz is unknown, the numerous 
quotations surmising its origin include New Orleans creole and the 
West Coast of Africa, among other guesses. 

The OED is also up-to-date in citing words reflecting new phe- 
nomena, sometimes the only dictionary with these words, which are 
not entered in the four general dictionaries cited at the beginning. 
For example, the OED alone cites Scientology, head voice (a musical 
term for the register between chest voice and falsetto). Teflon-coated 
(the others just cite Teflon), and Ramho with its derivative forms like 
Ramboesque, Ramhoism, and Ramho-like. Moreover, unlike the three 
general standard desk dictionaries, the OED, with RHII, cites the slang 
meaning of Oreo, coined by African-Americans to indicate Blacks 
with White loyalties. The OED (like the four dictionaries) also includes 
the slang word jawboning. And the OED gives the dialectal meaning 
of viewing (a Southeastern Pennsylvania term for looking at a de- 
ceased body on display before the funeral according to Carver 
1987:265), defined only by WNWD of the four dictionaries but cited 
by RHII as an exainple for the general sense of viewing. 

The OED cites such computer terms as lap-top (omitted by AHD 
and WNWD); window (omitted by W^''' and AHD); and hack, hacking, 
and hacker. Among items from popular culture, the OED cites junk 
food (omitted by W^'''); break dancing (omitted by AHD); more com- 
pounds beginning with the word women's than RHII (which, in turn, 
contains more women's compounds than the general desk dictionar- 
ies) — namely the OED citations of women's liberation, women's 
movement , women's rights, and women's studies; televangelist (omit- 



76 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

ted by AHD and WNWD); passive smoking and passive smoker 
(omitted by the three desk dictionaries, but passive smoking cited in 
RHIiy, and access as a transitive verb (in line with the American 
dictionaries); and treats massaging the data or hooks (omitted by 
AHD and WNWD). 

Despite these achievements, however, the second edition of the 
OED lacks a number of terms used in current life, probably because 
of a heavy reliance, as the OED acknowledges in its prefatory mate- 
rials (xxii), upon two Barnhart Dictionaries of New English, the Barn- 
hart Quarterly Companion, and the record of new vocabulary words 
in book form published by Merriam-Webster. Although entirely new 
information dealing with an additional 5,000 words, combinations, 
and senses were included and integrated in the second edition of the 
OED chiefly for the first third of the alphabet where the Supplement 
was 20 or more years old (xii), many revisions are needed for a 
future third edition. 

3.0 Problems 

3.1 Terms not included 

Some of the words or phrases not presently included are now 
basic to American life and can safely be predicted to endure, because 
of present trends of life. They often appear in some of the other dic- 
tionaries cited at the beginning of this paper, though the OED omits 
these items. In the following discussion whenever other dictionaries 
include these common terms that the OED neglects, the dictionary or 
dictionaries are listed in parentheses. The OED has not included living 
will {RHII,AHD, WNWD, OD of NW)\ caregiver (RHII, though a sharper 
definition is needed); inclusive language {OD of NW under inclusive); 
and stress management, though the OED cites these related health 
phrases: stress disease, stress-free, and stress test (the last item in 
RHII); and has also omitted sexual harassment (RHII), though citing 
these compounds having to do with sex: sexual athlete, sexual revo- 
lution, and sex discrimination. 

Also missing are blended family (RHII); Moral Majority (RHII, 
WNWD, OD of NW); New Age (WNWD, OD of NW); holiday blues/ 
Christmas blues/yuletide blues; singles-bar (AHD , WNWD , RHII); and 
junk bond (W^'/', WNWD, RHII, OD of NW). Organ donor is also omit- 
ted (its meaning in the OED under donor alone). 

Words for biological classification of mother are also absent from 
the OED: namely, birth mother (but the OED lists birth parent); donor 
mother (but one of the definitions for donor in the OED pertains only 
to artificial insemination from the semen donated by a male); and 
biological mother/father/or parent (RHII under biological parent). 

Words also missing are rap, a noun or verb pertaining to music 
(RHII as a verb and labeled slang and also under rap music; WNWD 
as a sense of the verb rap; OD of NW as a noun); tabloid television; 



Ching: Examining the trustwoithiness of the latest OED 77 

gay bashing; and chaos theory, a term recently applied to literary 
criticism {OD of NW only in its mathematical and physics sense; OED 
including only the word chaology with no definition but with two 
quotations, the last being 1775). 

Unlike the terms above, the following words or phrases are 
rather new to the American public: African American (the OED, like 
the four dictionaries, lists Afro-American); the African-American use 
of dis or diss, an abbreviated verb clipped from a longer verb, per- 
haps from disrespect, dismiss, discount, or distance {WNWD, though it 
is unclear whether the African-American use is meant, because it is 
not labeled as an Americanism nor as slang; OD of NW)\ and friendly 
fire, pertaining to destruction from one's own military forces by 
mistake — a term widely used by the military in the Gulf War (RHII; 
OD of NW, which indicates that it was a euphemism used since the 
Vietnam War). 

3.2 Updated definitions needed 

Another area needing rectification is the updating of newer 
senses of old words. At times, the newer metaphorical meanings of 
old words and phrases have not been cited. The OED, for example, 
does not list the sense of lightning rod referring to a person bearing 
the criticism and punishment for others (RHII); and the meaning of 
greening as in C. Reich's The Greening of America (1970) (WNWD). 

Certain senses of words in specific contexts are also lacking; e.g., 
linkage in terms of political connection between two different issues 
so that progress in one area must occur for progress on another issue 
to take place {RHII, AH D, OD of NW); myth as an explanation of the 
world, its phenomena, and the place of human beings in it, as used 
by Lakoff & Johnson (1980:185-194) in what they call the myths of 
objectivism, subjectivism, and the experientialist synthesis; myth also 
used by more liberal religious groups to refer not only to other 
people's but also to their own beliefs of the world, their destiny, and 
the purpose of life (AHD) — not understood as a fictitious, legendary, 
or prehistoric explanation of the creation of the world; and owner- 
ship, not meaning legal possession of something, but possession in 
the sense of understanding something and making it a part of 
oneself, such as in the ownership of an idea or a poem. 

Other old words for which new senses are not given include 
mosey, 'to move along leisurely or aimlessly' with the OED only 
listing the original meaning of 'to hurry along' (but both meanings 
appear in all four dictionaries); the slang meaning of cowboy as a 
verb, meaning 'to murder recklessly and openly' (ND of AS); futon, an 
updated meaning of a Japanese mattress with a slatted wooden base 
which can be converted into a sofa for day use in the US (OD of NW); 
street people (RHII, WNWD), to mean homeless people who live on 
the streets (not as the OED'% older meaning of people who live on the 
streets, especially to protest against the values of society); and home 



78 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

boy, without the new extended definition of someone who is as 
intimate and dear to one as a person from one's own hometown 
(\y9r/i^ with only the older literal meaning; OD of NW), though pre- 
senting home boy, home girl, and home people as slang expressions 
in a 1967 American Speech quotation about its use at Southern 
African-American colleges. Moreover, shortened forms, such as 
homey and homes, and the synonym homeslice are only listed by O D 
of NW. 

3.3 Updated quotations needed 

In informing readers of how old words are used in new ways, 
the OED's updating of quotations would help. 

For many words and senses of words, the latest quotation cited 
is a nineteenth-century example. For example, the OED gives four 
quotations for humankind (cited in the four dictionaries), the first 
recorded in 1645; the last in 1860. Without quotations in the 
twentieth century, a reader would not know that this word has cur- 
rently replaced mankind in some circles, especially feminist or lib- 
eral religious groups. Similarly, the reader would not know, without 
current quotations, that working class, the British euphemism for 
lower class, is now used in American reports and in dialect studies 
(the four dictionaries). And for downsize or downsizing, the OED's 
numerous quotations are largely about automobiles, with one quota- 
tion about a remark being downsized. Quotations need to be cited on 
downsizing many other matters — style of living — clothes, shoes, 
homes — and portfolio stocks, etc. (an updating is also needed in the 
four dictionaries). 

3.4 Sharpened definitions needed 

Besides updating the newest meanings of old terms and the 
updating of quotations for new uses of old words, the OED should 
sharpen its definitions of some of its words for clarity. 

Although the OED defines carpet-bagger adequately in its deno- 
tative sense as a person interfering in politics with 'no permanent or 
genuine connexion' it does not match RHII's. description of an 'oppor- 
tunistic or exploitive outsider' or WNWD's mention of the contempt 
and resentment felt by the local populace. Another example is the 
OED'?, definition of sisterhood as a relationship of women with a 
common aim or common characteristics and calling, a term used in 
feminist descriptions, with quotations which have derogatory or 
ominous implications of the word: 'sisterhood's demonology' and '... a 
portent of what the sisterhood is now brewing up.' The definition 
should include positive meanings, such as a feeling of oneness and 
closeness in mutual understanding and support among each other, 
followed by illustrative quotations of the positive meaning (only 
WNWD gives this positive view). 



Ching: Examining the trustworthiness of the latest OED 79 

The OED also needs to refine these definitions: recycling to 
include the reuse of discarded inaterials, such as bottles, cans, and 
paper (the four dictionaries, OD of NW) — not only the recycling of 
organic wastes or the recycling of industrial processes; and demoral- 
ize to mean not only the lowering of morale especially because of 
armed force but also because of any trying adversity (the four 
dictionaries). 

Another problem with the OED's definitions is that the defini- 
tions concerning learned or technical matters would not be under- 
stood by the novice, but only by persons already familiar with the 
field. For example, although deconstruction and deconstruct are listed 
in the OED, the definition as using the method of Jacques Derrida in 
critical analysis of language and literature is arcane to most people, 
unlike the excellent definitions by RHII and WNWD. And definitions 
for new grammatical terminology are entirely inaccessible for the 
uninitiated: e.g., deep structure and surface structure (also incompre- 
hensible for the novice in the four dictionaries); case grammar (RHIf 
but also opaque for the novice); and homorganic {RHII). 

3.5 Social sensitivity needed 

Another problem with the definition of terms is that a reader 
socially sensitive to changes in world view would be displeased with 
language once accepted, but now considered offensive. 

The OED's second edition has tried to be more socially sensitive 
but falls short, for example, in its definition of canoe. Its first edition 
states that it is a 'rude craft' used by 'uncivilized nations' or 'unciv- 
ilized people' and 'savages'. In contrast, the second editions describes 
a canoe as a 'roughly-made craft' used by 'primitive societies'. But 
who is to say that these crafts are 'roughly made'? Canoes that 
traveled all the way from Samoa to Hawaii in rough waters, for 
example, cannot be just 'roughly made'. Nor is 'primitive societies' 
acceptable. Substitutes now acceptable for primitive are pre-literate, 
pre-industrial, or nontechnological. 

3.6 Usage problems 

Not only is the OED's. usage outdated as shown above, but some 
of its usage LABHLS are also outdated, and there is an absence of 
usage notes or labels where they should appear. 

Some of the usage labels are incorrect. For example, the OED 
indicates that blurb is slang, whereas the four dictionaries have no 
usage label for this term. The OED also labels misspeak as obsolete, 
although I have heard it used by my department chair and a young 
engineering professor. In comparison, VV^''' and/4//Z) do not cite 
misspeak at all, but /?////, WNWD, and Will cite this word with no 
status label indicating obsolescence. Another word labeled by the 
OED as obsolete is disinvite, although columnist George Will in one of 
his columns (1991 :A8) spoke of Linda Chavez being disinvited or of 



80 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

certain groups disinviting her because of political correctness, using 
the passive voice of the verb twice and its present participle once. In 
comparison, only RHII of the four dictionaries lists this term, and it 
bears no status label; Will lists this term, but marks it as obsolete. 

Moreover, usage labels or usage notes do not appear where they 
should. For example, there are no usage notes for mankind and 
brotherhood to indicate that they are old fashioned and offensive to 
some people if they refer generically to both males and females. In 
comparison, W^'f^, AH D , and WNWD also have no usage notes for 
mankind. But RHII does say, 'See -man,' which states, with extensive 
examples, that compounds with -man have declined in recent years, 
though it does not cite mankind specifically. One would have to look 
at WD of EU, which cites alternatives to the word if it is offensive to 
the audience: alternatives such as humankind , human beings, 
humans, and people. Similarly, for brotherhood, cited in the OED in 
the phrases brotherhood of man and universal brotherhood, no usage 
note appears. In comparison, there is also no usage note or label for 
brotherhood in the four dictionaries nor in WD of EU. 

3.7 A systematic method needed in listing and defining 
terms for reader convenience 

Another problem when looking up words is the lack of a sys- 
tematic method in listing and in defining terms. 

At times, related terms are not listed. For example, the OED lists 
AIDS as an entry, and in its definition, cites HIV as the cause, but 
there is no separate entry for HIV. In comparison, the only one of the 
four dictionaries which lacks a separate entry for HIV is A H D . 
Similarly, the OED lists pro-life and pro-lifer as one entry with the 
request to look under pro-, and in a 1979 quotation for pro-life men- 
tions pro-choicer, but there is no separate entry containing both pro- 
choice and pro-choicer, unlike W^'/', RHII, and WNWD, which list pro- 
choicer under pro-choice. AH D V\st& pro-choice with no mention of 
pro-choicer. 

Another problem in listing terms is that the older form of a 
word may be used rather than its more familiar, current form: e.g., 
the OED does not cite the clipped form blush, meaning 'rouge', as an 
entry at all, but lists only its older, fuller form, blusher (this is also 
true of W^t'',AHD, and WNWD; RHII lists blush as a noun, but asks 
the reader to look under blusher); and the OED does not cite bag lady 
(cited by the four dictionaries), today's familiar clipped form, but 
lists the compound as shopping-bag lady under shopping as an at- 
tributive. 

Another difficulty is that some phrases are not entered as 
compounds in bold print, as the OED often does before citing them in 
the illustrative quotations. Thus, the reader must plough through the 
illustrative quotations to find primal scream (RHII) and related com- 
pounds under the psychological subject heading for primal: primal 



Ching: Examining the Uiistworlhiness of llie latest OED 8 1 

screamers and primal therapy (RHII, WNWD). (W^'^' and WNWD list, 
in addition, primal scream therapy). Nor is the popular sense of 
primal scream given, for example, a 'loud, primitive scream venting 
one's pent-up frustrations for catharsis', as used on college campuses 
on an agreed-upon hour, such as midnight, during examinations; only 
the technical psychotherapeutic definition is implied in the quota- 
tions on primal therapy (like W^'l'^RHII, WNWD). 

Moreover, highlighter, meaning 'a marking pen highlighting 
words or passages', is confusingly listed immediately under the 
second sense of the verb highlight, meaning 'to tint or bleach 
portions of the hair', with "hence highlighter', rather than under the 
verb's first sense, 'to bring into prominence ... or draw attention to'. 

Also for improved clarity a parallel method of defining related 
terms is needed: e.g., deep structure and surface structure. For 
surface structure, the term is defined immediately when it is listed 
under the attributive combinations of surface; then one may look at 
its use in a quotation cited from Chomsky's Current issues in linguis- 
tic theory. For deep structure, on the other hand, no definition imme- 
diately follows its listing as a compound of deep. Instead, there is an 
admonition to 'see quot. 1965, a quotation from Chomsky's Aspects of 
a theory of syntax.' 

One thorny matter which also needs resolution is the way 
compounds are listed, either under the first word or as a separate 
entry, a situation which confuses the reader about where to look first 
for the compound. For exainple, camp meeting is an entirely separate 
entry, not listed under camp. On the other hand, human rights is list- 
ed under the word human in italicized bold print in the section on 
compounds with human, and surrogate mother is found only in quo- 
tations for surrogate without surrogate mother listed in bold letters 
and with no separate section listing compounds for surrogate before 
the quotations. Some logical rationale needs to be devised for the 
citing of compounds. 

4.0 Conclusion 

Fortunately, the OED editors in the prefatory materials acknowl- 
edge that full revision and updating are long-term goals and that the 
great accomplishment thus far has been a merger of the first edition 
with the supplements of the last two decades using computer 
technology (xi). Moreover, the editors acknowledge many of the 
problems cited in this study (Iv-lvi). Besides, the critical examination 
in this study is not meant to be a disparagement of the OED but only 
a lover's quarrel with this venerable treasure of information, to 
which we have been fondly attached and from which we have grown 
to expect high standards nearing perfection, if not perfection itself. 
Yet the task for revision is mammoth, as we have seen, and more so, 
because the OED now seeks to include not only the common literary 
and colloquial words, as in the first edition, but also more of the 



82 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

'main technical vocabulary and a large measure of dialectal usage 
and slang' (vii). In addition, the OED seeks not only to expand the 
coverage outside of the United Kingdom, but also especially North 
American English, which the editors say, 'is the greatest source of 
linguistic change' (Ivi), along with the English of other parts of the 
world so that 'the Oxford English Dictionary may continue to be an 
accurate and comprehensive register of the whole vocabulary of 
English' (Ivi). A tall order! 

If this venerable dictionary is to retain its scholarly reputation, 
it must use other procedures to glean the data necessary to make 
emendations for the seven areas named above. 

REFERENCES 

Algeo, John. 1990. The emperor's new clothes: The second edition of 

the society's dictionary. Transactions of the Philological Society 

88:2.131-150. 
The American Heritage Dictionary. 1991. 2nd college ed. Boston: 

Houghton. 
Carver, Craig. 1987. American regional dialects: A word geography. 

Ann Arbor: University of Michigan Press. 
Chapman, Robert. L. 1986. New dictionary of American slang. New 

York: Harper & Row. 
Gray, Paul. 1989. A scholarly Everest gets bigger. Time, 17 March, 

pp. 95, 98. 
Harvey, A. D. 1991. Road to rot: The new 'Oxford English dictionary'. 

Contemporary Review, August, pp. 84-86. 
Lakoff, George, & Mark Johnson. 1980. Metaphors we live by. 

Chicago: University of Chicago Press. 
The Random House Dictionary of the English Language. 1987. 2nd ed. 

unabr. New York: Random House. 
TOFFLER, Alvin. 1971. Future shock. New York: Bantam Books. 
TULLOCH, Sara. 1991. The Oxford dictionary of new words: A popular 

guide to words in the news. New York: Oxford University Press. 
Webster's Dictionary of English Usage. 1989. Springfield, MA: 

M err i am- Webster. 
Webster's Ninth New Collegiate dictionary. 1991. Springfield, MA: 

Merriam-Webster. 
Webster's New World Dictionary. 1988. 3rd college ed. New York: 

Simon and Schuster. 
Webster's THIRD New international Dictionary of the English 

Language. 1971. Unabr. Springfield, MA: G. & C. Merriam. 
Wentworth, Harold, & Stuart Berg Flexner, comp. and ed. 1975. 

Dictionary of American slang. 2nd supp. ed. New York: Crowell. 
Will, George. 1991. 'PC? Never heard of it. Commercial Appeal 

[Memphis, TN] 22 October, p. A8. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 

LANGUAGE PLANNING ACROSS POLITICAL BOUNDARIES: 
A CASE STUDY OF PULAAR* 

Abdul Aziz Diop 

Much of the previous literature on language planning 
and language policy (LP/LP) focused attention primarily on 
the discussion of the associated issues within national 
boundaries. In this study I discuss LP/LP at the supra- 
national level. I examine factors that can impede language 
planning (LP) across political boundaries in Africa, focusing 
on the case of Pulaar (Fula). I make suggestions towards the 
development of standard Pulaar and consider some of the 
problems underlying such an enterprise — problems such as 
those caused by language loyalty, (language) nationalism, 
power conflict, and language maintenance. I suggest ways 
of dealing with (some of) these problems. At the end of the 
paper I indicate directions for future research in the area of 
LP/LP for Pulaar. 

1. Introduction 

1.1. Preliminaries 

Language planning, whether carried out consciously or not, is a 
secular enterprise. Yet it remains one of the areas where human 
beings have probably been the least successful. Whenever an 
institution or a body of institutions chooses a language (or variety 
thereof) among others to serve as the medium of communication (in 
all of its aspects) between speakers of different languages or dialects, 
we have a case of language planning (Whiteley 1972). However, as 
Tollefson (1991) mentions, there has been 'a failure to relate 
language planning and policy to broader sociopolitical changes, and 
ideologies and political values passing for theoretical frameworks 
prevail'. 

Language planning has become, through the ages, more of the 
domain of politics than of linguistics. Haugen (1966a:26) stated that 
'language planning, like politics of which it is a part, is the art of the 
possible'. This statement is certainly true of LP but unlike the 
politician's goal the language planner's goal is substantially the same 
as that which the people have unconsciously accepted as their own. 
Thus, LP, unlike politics, is not the art of preventing people from 
meddling in their own affairs. 

Much of the previous literature on LP/LP (Fishman & Gupta 
1968, Ferguson 1966, Jernudd & Rubinl971, Haugen 1966a, Whitely 
1966, etc.) focused on LP at the national level, i.e. within one country 
or state; and that was pretty much the general paradigm. Of the few 



84 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

studies about LP/LP at the regional level one can cite the recent 
study by Berns (1992) which focuses on the use and status of English 
within the European Council (EC) member countries. Berns points to 
the fact that '... the special status of English as a language of wider 
communication among Europeans is recognized in its use, with French 
as the language of Council publications and documents'. This raises, 
however, a number of concerns on the part of Europeans who speak 
languages other than English. I discuss those concerns as they relate 
to Fula in section 4.2 below. 

The lack of studies on LP/LP across national boundaries or at 
the regional level seems to suggest that LP/LP must be confined to 
national boundaries. If this is correct, the question must be raised 
why this is the case. What factors, for example, impede LP at the 
regional level? Can such difficulties be overcome and what benefit 
would accrue to the nations concerned? To answer these and other 
questions we must, first of all, understand the problem and 
opportunities presented by the language under consideration here, 
Pulaar. 

1.2. The problem 

Pulaar is a West Atlantic language (Greenberg 1955) spoken in 
approximately seventeen to twenty countries in Africa. Though the 
figures may vary, the most recent ones available indicate that there 
are 10-15 million native speakers (Crystal 1987:438) speaking many 
different varieties of the language, some of which are almost 
mutually unintelligible. While certain (neighboring) countries, for 
example Senegal, Guinea, Mali, and Mauritania, have had some 
common politico-cultural ties, such ties do not necessarily exist 
between any two (or more) given countries where Fula is spoken 
(e.g. between Mali and Ethiopia). The popularity of Fula offers 
numerous opportunities as well as challenges. One such opportunity 
(or challenge depending on how one looks at it) is the possibility of 
serving as a lingua franca or language for wider communication at a 
regional level much like Kiswahili in East Africa or English within EC 
countries in Europe. A discussion of this problem in terms of LP is 
called for both for practical and theoretical purposes. There are 
classic problems that LP/LP have always faced. Four of those will be 
mentioned in this paper because they surely would constitute a 
major problem for Pulaar planners. Three of them are sociopolitical: 
(language) nationalism, language loyalty, and (international) power 
conflict; and the fourth one is linguistic and has to do with problems 
of maintaining the variety that is chosen as standard and making 
sure that it would not undergo further atomization. Before discussing 
these issues let me first outline the objectives of this paper. 

1.3. Objectives 

The main objective of this paper is to argue for the necessity of 
a cross-national language planning and language policy for Pulaar in 



Diop: Language planning across political boundaries: A case study of Puular 85 

Africa, a language policy that consists first and foremost in standard- 
izing the language at the regional level. In fact the grounds for a 
similar proposal were established as far back as 1965 following the 
creation of 'la Societe Linguistique d'Afrique Occidentale', one of the 
objectives of which was 'to study the Peul language and culture'. 
Another initiative with respect to Pulaar was taken in 1986 following 
meetings in Yaounde (Cameroon) and Niamey (Niger). One of the 
goals outlined in those meetings was 'la promotion des langues a 
vocation regionale (languages for wider communication): Peul, 
Manding' (CONFEMEN 1986). 

A secondary goal of this paper is to examine the feasibility and 
desirability of undertaking LP/LP at a supranational level. The 
discussion of these objectives will lead to the exploration of other 
issues of both practical and theoretical importance. 

The paper will to some extent draw on the theoretical model of 
language standardization of Albanian discussed by Byron (1976). The 
similarity between Albanian and the case under study is that in both 
cases we are dealing with selection (of a standard) among alternates 
of the same language. However, while in the Albanian case we are 
basically faced with two dialects, Tosk and Geg, in the case under 
study we may be dealing with many more dialects. 

The Albanian case was presented in Byron 1976 from two 
different angles: On the one hand opposing views are presented as to 
the criteria for selection of the Tosk or the Geg dialect; on the other 
hand, subsequent views center around the issue of what the base 
component of standard Albanian is; i.e. whether it is one dialect, or 
the other, or the combination of both. 

The paper will also try to bring into the discussion the 
experiences of other African countries, including that of some of the 
so-called Francophone countries in relation to 'Francophonie'. Before 
doing so, let us, first, take a look at the various contributions made to 
LP/LP. 

2. Review of literature 

Even though there is a dearth of literature on LP/LP for Pulaar, 
the issues of LP/LP have been investigated quite thoroughly by a 
number of linguists around the world. 

2.1. General literature on LP/LP 

The term language planning refers to the organized pursuit of 
solutions to language problems, typically at the national level 
(Jernudd & Gupta cited by Fishman 1972). 

'Language planning is an area of linguistics which allows us to 
consider language as one more object of human manipulation — not 
only by language specialists but also by lay persons who may change 
its basic structure through their attitudes and myths about the 
language, and their subjective reactions to language' (Rubin 1973). 



86 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Language planning has also been defined as 'the conscious, predictive 
approach to changes in language and language use' (Jernudd & 
Rubinl971). Language planning could be the result of a formal or 
informal process. The latter consists of a de facto assignment of 
certain functions, positions, or statuses to certain languages in a 
multi-lingual community. The first is the result of a deliberate and 
conscious effort on the part of decision-making bodies in a particular 
country — an effort geared towards promoting a certain language or 
certain languages (and, thereby, demoting others?) to a certain status 
in order to fulfill certain specific functions. 

It has been suggested in the sociolinguistic literature on LP/LP 
that formal language planning encompasses two major steps which 
can be further analyzed into substeps. The first step is referred to as 
status planning. As initially suggested this step is concerned with 
decision-making generally by a government body and often at the 
national level; it is the political phase of language planning. It 
involves three different but related subphases: 

• the perceived need to elevate one selected language as the 
national and/or official language of communication; 

• an attitudinal survey in order to identify attitudes of 
individuals vis-a-vis the language(s) selected; 

• a collection of demographic data on the languages. 

The second step, called corpus-planning, involves the actual 
planning/articulation of the language. This is the linguistic phase of 
language planning. Four different but not mutually exclusive 
substeps can be identified within this step: 

• identification and selection of a particular language or dialect 
(notice that this stage is very similar in essence to the third 
subphase of the political phase); 

• codification, a substep involving the adoption of an alphabet. 
Thus it is often referred to as graphicization. Codification also 
consists of selecting lexical items wherever there are some that 
are competing, writing of grammars, compilation of dictionaries, 
etc. 

• implementation: or the adoption of what has been proposed; 

• elaboration of the language functions and development of new 
registers. 

While the subphases of the political stage of language planning 
can be followed with 'relative' ease, the four substeps of the corpus 
planning (i.e. identification, codification, implementation, and 
elaboration) are highly theoretical and harder to apply to language 
planning in real life. Proper application of language planning would 
certainly require the right kind of information about the 
sociolinguistic habits of the target population and about the social 
basis for language policy in order to 'project productive directions of 
change' (Rubin 1973). 



Diop: Language planning across political boundaries: A case study of Puular 8 7 

Another issue that a number of prominent sociolinguists deal 
with in the literature on LP/LP is the question of the criteria for 
evaluation (e.g. Ray 1963, Haugen 1966b, Neustupny 1970). Ray 
(1963), for example, proposes three criteria: efficiency, rationality, 
and commonalty. The criterion which is directly relevant to the case 
under study in this paper is that of efficiency as it refers to the 
relative cost, in time and other resources, of learning and maintain- 
ing one form (for that matter, perhaps one dialect) as opposed to 
another. However, this is not to say that the other criteria are not 
important at all. They certainly are, but perhaps to a different de- 
gree. The second criterion, rationality ('adequacy' in Haugen's terms 
(1966b)) is certainly an important one in that it has to do with the 
capacity of a language to function in a wide variety of styles, genres, 
levels of discourse, etc. Neustupny (1968), on the other hand 
postulates four national needs which, in his opinion, will determine 
the ultimate fate of the lingua franca. These are: 

• raising the standard of living 

• extension of literacy 

• development of national consciousness 

• cooperation with neighboring nations. 

These needs could in fact be regarded as criteria of evaluation to 
some extent — in that language planners would have to make some 
predictions and anticipations as to what dialectal variety would be 
capable of achieving such ideals with the 'minimum' amount of 
sacrifice. 

Another set of criteria often discussed in the literature has to do 
with the issue of whether the standard should reflect the speech of 
an elite, i.e. a minority, or that of the majority. In the case under 
study it is not quite sure whether this is a relevant issue at all, given 
the multitude of dialects and the widely varying experiences from 
one country to another. 

Although the three linguists mentioned above use almost the 
same terminology to discuss the notion of criteria of evaluation, they 
all disagree with respect to what each criterion encompasses. What 
these disagreements reveal is the fact that the relative role of lin- 
guistic vs. non-linguistic factors in language planning is not entirely 
clear. What, for example, are the "patterns' and 'features' which 
Neustupny assures us constitute the field in which evaluation 
functions? Are phonological problems as tangential to language plan- 
ning as some linguists (including Ray) contend? Does usage precede 
evaluation, or vice versa? These as well as other issues are still 
unaddressed. 

Given the outline that I just presented of the general literature 
on LP/LP and on criteria for evaluation, let us now briefly discuss 
LP/LP as it relates to the African continent. 



88 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

2.2. LP/LP in African societies 

There is a large body of literature on LP/LP in Africa. Among 
some of the most well-known I can cite Whitely 1966, 1971, Fishman 
1972, Bokamba 1984a, 1984b, 1976, Bokamba & Tlou 1977; 
Labouret 1952, Sow 1977, Bamgbose 1991, Gnalibouly 1988, Ka 
1983, etc. to name a few. To this one can add numerous conferences 
by the UNESCO and the ACCT. Spencer (1971) gives a historical 
perspective of what 'led' to LP/LP in Africa — tracing contacts with 
Europeans that date back as early as 1445. He gives us an analysis of 
French/Portuguese policies in contrast with British/Belgian policies 
with respect to attitudes towards the so-called 'vernacular' 
languages. The former was that of indifference and restrictive 
assimilation, while the latter displayed a paternalistic flavor. 
According to Bokamba (1984b:2) French was imposed because of: 

• politico-cultural strategies; 

• African intellectuals' wish to 'be educated' the way people are 
educated in France; 

• various local linguistic factors. 

The French colonial language policy reached its highest stage 
when the metropolitan ordinance of Villers-Cotteret, which was 
issued in 1539 by King Francois I, was extended to the colonies in 
order to ensure that French must be the only language used in 
schools by students. 

Despite apparently different philosophies of imperialism, these 
cases share the similarity that LP is the result of a colonial policy 
whose very foundation rested on a theory of 'separate development' 
for the different races in contact with Africa. Sow (1977) takes us 
back some three to four decades to trace the first initiatives taken 
for the creation of the West African Linguistic Survey in 1956, 
followed by the creation of 'la Societe Linguistique d'Afrique 
Occidentale' (SLAO). Under the auspices of these organizations a 
number of initiatives were taken with respect to LP/LP at different 
meetings. One of the objectives of the organization was to study the 
Fula culture and language. The 'SLAO' came up with recommenda- 
tions following a meeting in Accra on April 12, 1965. These 
recommendations concerned the implementation of an alphabet and 
an orthography for Pulaar. Recommendations were also made with 
respect to various aspects of implementation of the language, but 
Sow does not make it clear which dialect was selected and elevated 
to the status of standard. That issue is as unclear as the question of 
what the procedures were for choosing one variety. For the sound 
transcription Sow goes on to mention that '.... il en est resulte 
I'adoption, apres vote, d'un systeme qui pent se definir comme suit... 
(Sow 1977:101). Thus decisions made with respect to that particular 
issue followed a vote rather than empirical/scientific research. One 
cannot help but question the validity of such an enterprise. 



Diop: Language planning across political boundaries: A case study of Puular 89 

Following a 1979 meeting by the UNESCO in Bamako (Mali), a 
recommendation was made to the effect of promoting the languages 
spoken across national boundaries as a means of inter-African com- 
munication and their use for administrative and economic purposes. 
That meeting was notorious for its lack of specifics and, as a conse- 
quence, there was no follow-up. The OAU Language Plan of Action for 
Africa was also one such attempt at planning in Africa. The plan 
proposes five specific goals, the first of which was to promote 'the 
use of certain viable African languages at national, regional, and 
international levels as official languages in place of non-indigenous 
official languages currently being used, and the adoption of such 
languages as working languages by national, regional, and continental 
institutions [....]' (Bamgbo.se 1991:127). This plan, in its internal 
anatomy (cf. Bamgbose 1991:125), has departed from the traditional 
ill-defined, pitifully vague pronouncements; but so far, it has failed 
to change attitudes of governments and policy makers in favor of 
realizing the projected goals. 

Other recent initiatives outlined by UNESCO include its 1985 
periodical on 'La definition d'une strategic relative a la promotion 
des langues Africaines'. Those initiatives were taken after a meeting 
of experts on African languages in Port-au-Prince, Haiti in 1975. 
Following that meeting two conferences were held in Yaounde 
(Cameroon) and Niamey (Niger). During those conferences a number 
of goals were outlined one of which was to promote languages that 
have 'une vocation regionale', Pulaar and Mandingue. 

Another study that dealt with the issues of LP/LP in Africa is 
published in collaboration with L'Agence de Cooperation Culturelle et 
Technique. Proposals that appear in that review are made following 
conferences of Secretaries of Education in various French West 
African States. For work in the area of LP/LP directly affecting 
Pulaar one can cite the Pulaar-Russian-French dictionary by Zoubko 
(1980); the ACCT's specialized lexicon that deals with teaching the 
natural sciences, grammar and linguistics, history and geography, 
politics, administration, and justice; the recent publication of an 
English-Pulaar dictionary by a Japanese linguist (Eguchi 1986); and 
finally, a UNESCO-sponsored dictionary of elementary Pulaar-French- 
English. Future projects of the Pulaarophones from Paris include 
helping researchers from France, England, Japan, and various coun- 
tries in Africa gather the data necessary for making an interdialectal 
dictionary, dialectal dictionaries, and bilingual interdialectal dic- 
tionaries. Unfortunately, as pointed out by Gnalibouly (1988), the 
initiative is not progressing satisfactorily. 

It is evident from the preceding discussion that the issue of 
LP/LP in Africa has changed very little. As a matter of fact, as 
mentioned by Bamgbo.se (1991), several African governments appear 
to employ avoidance or vagueness techniques, as can be illustrated 
by the fact that very few countries have definitive statements of 
language policy. But avoidance is in itself policy. An example of a 



90 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

vague policy cited by Bamgbose (1991) is Kenya's decision to adopt 
Swahili as its national language. The immediate motivation was 
political. The vagueness is reflected in the implementation steps 
recommended, the details of which were not given. In addition, the 
country's official language in which records are kept and admin- 
istration conducted is still English. To this one can add cases (e.g. 
Nigeria) where a policy is declared without the possibility to imple- 
ment it due to improper circumstances for implementation, built-in 
escape clauses that give excuses for not implementing the policy in 
question, or unspecified procedures. To all of these problems related 
to language planning in Africa one can also add problems caused by 
fluctuation in language policy due to such factors as changes in 
government or party policies, advice from various foreign organiza- 
tions, etc. Ghana is a case in point for such fluctuations (cf. Bamgbose 
1991 for further discussion). 

In the next section I will try to give a sketch of some of the most 
crucial and specific problems underlying the development of a 
language policy for Pulaar, based on what has already been achieved. 

3. Language planning in West Africa: the case of Pulaar 

At this juncture we should address at least two questions. First, 
what problems would a language policy proposed for Fula encounter? 
Second, how would one deal with these problems? Assuming that the 
problems can be addressed successfully, what benefits would the 
population derive from the language policy in question? To get a 
better picture of the situation at hand I give a profile of Pulaar. 

There are many names used to refer to this language: Fula, 
Fulani, Fulfulde, Pulaar, Haalpular, Toucouleur Fulacunda, Fellata, etc. 
In this paper I shall refer to it as Pulaar or Fula. Pulaar is a West 
Atlantic branch of the Niger-Congo family, which is in turn member 
of the larger family of the Congo-Kordofanian (cf. Greenberg 1955). It 
is spoken in at least the following countries in West and Central 
Africa: Benin, Burkina Faso, Cameroon, the Central African Republic, 
Chad, Gambia, Guinea, Guinea-Bissau, Mali, Mauritania, Niger, Nigeria, 
Senegal, Sierra Leone, Togo, and to a lesser extent in Sudan and 
Ethiopia. 

Because of differences in geographical distance/proximity it 
should come as no surprise that certain dialects of Fula are more 
mutually intelligible than others. Some of them are almost heading 
towards mutual unintelligibility. Not only do we notice considerable 
dialectal variations among its different regional dialects, but also 
within each country where it is spoken there is some kind of dialec- 
tal 'atomisation' that is worthy of notice in some cases (e.g. the 
difference between Pulaar as spoken in Northern Senegal and that 
which is spoken in the Casamance (southern) region of the same 
country). 



Diop: Language planning across political boundaries: A case study of Puular 9 1 

While it is clear that HaalPulaar'en (native speakers of Pulaar), 
who were for the most part nomad cattle herders, have migrated 
over centuries all over the African continent, settling wherever they 
could take advantage of rain and green pasture for their cattle, it is 
not yet quite certain where they exactly departed from. However, Ka 
(1983) quoting Hama (1968) who himself quoted Delafosse (1912) 
seems to support the thesis that Haalpulaar'en originated from 
somewhere between Mauritania, Senegal, and Mali. The issue is moot 
for the purposes of this paper and I do not intend to pursue it 
further. 

A number of countries where Pulaar is spoken have been or are 
currently involved in various attempts to implement concrete 
guidelines for LP/LP (Niger, Senegal, Mali, Mauritania, Guinea, etc.). 
The 'PROJET MAPE' is a materialization of such attempts. This project 
is primarily intended for LP/LP of a number of languages in that 
particular region of Africa. In the case of Pulaar a number of coun- 
tries are using the language (still at the experimental stage) both for 
purposes of education (school curricula) and in the media (radio and 
TV); cf. the countries just cited in addition to Egypt for radio. 
However, there does not seem to be any literature that shows where 
the language fits into the overall framework of LP/LP in other 
countries, such as Sierra Leone, Sudan, etc. 

The language is 'divided' into two major geolinguistic groups: the 
Eastern Dialects and the Western Dialects. There are some straight- 
forward structural differences between the two. 

One straightforward structural difference between these 
regional dialects is the fact that although all the dialects of Pulaar 
exhibit stem-initial as well as stem-final consonant alternations, 
some dialects show more alternations than others, especially in 
word-initial position (perhaps alternation is avoided for the purposes 
of 'simplification'). To illustrate, consider the following example. In 
the Westernmost dialects (e.g. Mauritania, or part of Senegal), the 
verbal stem yah- 'go' exhibits the following alternation (initially) in 
certain contexts such as plural (lb) or subject pronoun post-position 
(Ic) below: 

(1) a. mi yah-i-ii Kahaydi 
Isg. go-i-asp 
'I went to Kahaydi.' 

b. be njah-»-ii Kahaydi 
3sg. go-i-asp 

'They went to Kahaydi.' 

c. njah mi ko Kahaydi 
go Isg. foe. 

'It's Kahaydi 1 went to.' 
Or: 'I went to Kahaydi.' 



92 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

In some of the dialects this alternation is avoided (most of the 
time) by preserving the same stem shape yah in both contexts (after 
singular and plural object pronouns). 

Another difference that is worth mentioning and that is noted 
even within the same country (e.g. Guinea) is the distinction between 
degrees of formality. In Fuuta Jalon the plural for respect is used any 
time one is talking to an elderly person; failure to do so is considered 
rude. In 'Haute Guinee', on the other hand, such is not the case. While 
degree of formality may not be justification for dividing dialects into 
major groups there is yet another more crucial difference between 
the two major dialect groups, viz. in the form of the infinitive. In the 
Eastern dialects the infinitive of a verb is formed by adding gol to 
the verbal root (e.g. yah-gol 'to go'), whereas in the Western dialect 
this is achieved by adding the suffix de (e.g. yah-de 'to go'). 

There are other dialectal differences that are worth considering 
but for the purposes of this paper I will not dwell on them. Besides, 
they are problematic both for native speakers themselves and for 
lexicographers working on Pulaar (cf. Gnalibouly 1988:10 for discus- 
sion). 

As we shall see later on, the issues I have briefly outlined here 
are going to have to be taken into consideration in the choice of a 
standard dialect. At this juncture I would like to get into issues 
pertaining, directly or indirectly, to supranational policies for the 
development of standard Pulaar. 

3.1. The development of a supranational language policy for 
Fuiani 

The need in the African continent for urgent language planning 
strategies is certainly worthy of note. Though this paper is not an 
account of the history of the African continent I cannot, however, 
avoid dealing with aspects of the legacy of France to West Africa. 
French colonial language policies have had the detrimental effect of 
undermining the importance of promoting 'indigenous' African 
languages to serve national functions (or international ones, for that 
matter). There was very little (if any) positive action taken to en- 
courage the so-called 'vernacular' literatures or to standardize and 
extend 'major' languages; i.e., those that had a considerable number 
of native speakers and dialects or which had the potential of being 
languages of wider communication. One of the damages of these poli- 
cies that Africans are still paying a heavy price for is the soaring 
illiteracy rate in (West) Africa. It was assumed that only those who 
read and/or wrote and/or spoke French were educated. Thus, no 
attention was paid to the idea of educating people in their native 
languages because, after all, writing/reading in Fula or any other 
African language was not something that was regarded as crucial for 
literacy. Emphasis may have been put on French to the detriment of 
native languages for various reasons but the most important ones are 
mentioned in Bokamba 1984b:2. 



Diop: Language planning across political boundaries: A case study of Puular 93 

When we are involved in the process of LP one of the questions 
we should ask ourselves is 'why plan at all?' In order to answer that 
question we should answer another question, which is that of who is 
it that we are trying to plan for. Thus, it looks as though the answers 
that we seek lie within the projected long-term goals that we set to 
achieve. We may need to plan languages in Africa for various but 
closely intertwined reasons — some of which I would like to mention 
here. First, where most European countries may have not more than 
three or four languages to contend with, African countries may well 
have more than a hundred languages within the state or country, 
sometimes belonging to several different language families. In such a 
situation for people to communicate with one another, recourse must 
be had to some kind of lingua franca. Second, the pre-independence 
experience of African countries, in addition to political/artificial 
boundaries of the post-independence era, prevented any language 
(or dialects thereof) from assuming a pre-eminent national or supra- 
national position. Third and last, but not least, political and economic 
atomization, in addition to climatological conditions (in the case of 
nomad Fulanis) accentuated territorially based dialect distinctions. 

After this brief outline, I would like to continue this section with 
the strictest ideological and political neutrality and discuss the issue 
as an outsider. With that in mind here are what, I think, should be 
six of the various needs for which I feel a standard Pulaar is nec- 
essary: 

First, for purposes of creating communication between, 
most importantly, native speakers of the language (promotion of 
intracultural communication) wherever they might be; and then, 
between native and non-native speakers. As Cooper (1989) 
suggests, an understanding of language planning demands an 
understanding of the social changes which promote it. More and 
more nomad Pulaars are forced (by climatological conditions) to 
settle, become farmers, or go to modern schools, thereby inter- 
acting and intermarrying with their sedentary 'brethren' who for 
the most part speak a different variety of Pulaar than theirs. 
This social change, while capable of bringing about the proper 
setting for a de facto standard at a local level, cannot do so at a 
regional level. 

Second, extension of literacy; mass literacy, that is. The 
more educated the people are the more likely it would be for 
them to take an active role in the development process of their 
nation. Thus literacy leads towards greater productivity. 

Third, 'international unity'. Linguistic diversity is as much a 
source of cultural strength as a potential source of disunity and 
politico-economic weakness. Thus, dialects of one and the same 
language, at least if they threaten to become languages, are 
potentially disruptive forces for a people that is in search of 
unity because the dialects appeal to local loyalties. If language 



94 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

planning within a particular country is a major step toward 
unification of the people of that country, language planning 
across political boundaries in Africa ought to have the effect (be 
it immediate or not) of helping toward greater homogeneity and 
unity of African nations. (One might claim that this is a naive 
statement but it is only so when we look at language planning 
and its desired result as a short term relief operation — which it 
is not.) 

Religion (e.g. translation of the Koran in Pulaar since a great 
number of Haalpulaar'en are Muslims). 

'Deethnicization'. The dialectal atomization of Fula would 
ipso facto imply a certain degree of cultural divergence as well, 
which itself may mean that certain Pulaar linguistic groups may 
evolve into totally separate ethnic groups. Language standard- 
ization would, in theory, contribute toward helping native 
speakers of the different dialects of Pulaar to, in the long run, 
unite within one speech community. 

Language maintenance. This is directly correlated to (and 
could be a result of) the need outlined above under 
'deethnicization'. 

We need to plan and standardize and enrich our languages 
because not doing so would have the necessary corollary that they 
would, in Fishman's (1972) terms 'forever remain in the intellectual 
and pragmatic shadows of others that had been fortunate enough to 
undergo slow but sure enrichment and standardization, both 
consciously and unconsciously, in prior generations'. In the following 
section I suggest ways of starting such a standardization process. 

4. Suggestions towards the development of standard Pulaar 

For the choice of a standard dialect of Pulaar, while we cannot 
yet establish which dialect to choose, we can bear in mind two things. 
First, for purposes of learnability the choice of a standard dialect 
should consider dialects that show the least amount of consonant al- 
ternations possible since those alternations are a great problem even 
for native speakers. Second, for the sake of authenticity the choice of 
vocabulary (codification phase) should be based on the question of 
'which words are the most authentic ones to the language?' as op- 
posed to those that mean the same thing and yet are borrowed from 
Arabic or French. I am not, however, by any means trying to exclude 
the possibility of borrowing from other languages (of course, inter- 
dialectal borrowing would be a first priority). Jernudd (1989:53) 
mentioned, in the case of French standardization, that one general 
goal was 'the maintenance and extension of a standard spoken and 
written French language purified of unacceptable English language 
borrowings and local idiosyncrasies'. While I do agree with getting 
rid of as many local idiosyncrasies as is feasible, it is not quite clear 
what is meant by 'unacceptable' in the above quote. It is certainly 



Diop: Language planning across political boundaries: A case study of Puular 95 

not my suggestion to proceed with a puristic attitude intended to rid 
Pulaar of borrowings. They are, after all, unavoidable. Furthermore, 
when it comes to matters of (language) purism, as noted by Jernudd, 
there is no such thing as absolute purism; but purism has to be 
viewed vis-a-vis the challenging language. It is not quite sure 
whether there is a particular language that can be said to constitute a 
challenging language to Pulaar in the same way English is to French. 
The nature of foreign borrowings depends on such factors as the 
language that Pulaar comes in contact with (Wolof, Hausa, French, 
Arabic, etc.). 

The following is an enumeration of some of the suggestion that I 
would like to make with respect to the standardization and imple- 
mentation of Pulaar. We need: 

• to (first and foremost) gain greater insight into the role of 
Pulaar in any context in the continent of Africa, using an 
analytical framework referred to as a sociolinguistic profile (cf. 
Ferguson 1966 for an outline). This insight will not only doc- 
ument attitudes toward the different dialects of Pulaar, but it 
would also help, as pointed out by Berns (1992), make important 
decisions that have to do with curriculum development, material 
design, and the setting of goals and expectations; 

• to conduct an attitude survey before we set to decide on a 
standard variety, followed by descriptive studies and pilot 
projects; 

• to combine, after the first step is taken, some of the various 
dialects on a rational and scientific basis, and construct a unified 
grammar from the compounded dialects without necessarily 
oversimplifying (and therefore trivializing) that grammar. After 
this step is taken let us then be concerned with the following: 

• to work with people who are not necessarily professional 
linguists but who should participate in the regularization of the 
language through their own specialties, e.g. law, economics, 
agriculture, writing, etc. After all, they, also, are language 
'cultivators'; 

• to create primary and secondary school facilities as well as 
incentives for going to school. Initially these would be the only 
places where the standard would be taught; 

• to create terminology commissions (e.g. a supranational termi- 
nology bank) the purpose of which will be to make lists of 
needed Pulaar terms in various areas of specialization. For 
example, as mentioned in Gnalibouly 1988 some concepts, 
though commonly used nowadays such as: feelings, color taste, 
nature, animal, insect, gas, liquid, or solid, etc. have no satis- 
factory equivalents in Pulaar. The Pulaar lexicon needs to be 
expanded to include these and other terms in order for the 
language to be taken to higher altitudes in the communications 



96 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

race. The terminology commissions can use a number of 
techniques for creating new terms, some of which are outlined in 
Mezei's paper in the case of Somali (1989): using Pulaar roots or 
derivations, reviving archaic words from classical poetry and 
songs, semantic expansion, compounding or reduplication, suffix- 
ation, etc. The terms would be used in government-supported 
schools and institutions, in the first place; 

• to organize frequent meetings between Fulanists and, say, 
Kiswahilists in order to exchange ideas and to learn from each 
other's experience; 

• to materialize initiatives such as that taken by the Pulaar- 
phones of Paris which consists of trying to come up with a 
General Dictionary of the Pulaar World. Such a piece of work (I 
do recognize, it would take tremendous effort to realize) is vital 
for the Pulaarophone world, in the absence, for the time being, of 
a standard dialect; 

• to create committees (in each country involved) in charge of 
supervising the creation (and co-ordination with various other 
countries) of new technical terms and promoting linguistic inter- 
action. When things go wrong those committees will determine 
what went wrong and who is to be held responsible for that; 

• to draw some useful conclusions from the Guinean experience. 
Guinea (cf. Fall 1981 for more discussion) is one of the few 
countries in (West) Africa to develop and implement a national 
language policy in its early independence years whereby the 
country was linguistically divided into geographical areas that 
reflected linguistic distribution. Pulaar was taught in 'Moyenne 
Guinee'. Therefore, Guinea may have a lot to offer in terms of 
pedagogy and teaching experience 

Let me mention, however, that all our efforts for planning a 
standard dialect would be in vain if we do not take a major pre- 
liminary step which is outlined in the lines to follow. I suggest that 
all governments involved take one major step without which nothing 
can be achieved: promoting language institutes. These language 
institutes should hire professionally-trained linguists whose job 
would be to work in ways that would help write consistent, coherent, 
and well motivated grammars of each separate dialect of Fula. In this 
way we are going to gain more insights into the structural descrip- 
tion of each dialect and therefore, be in a better position for 
evaluating the dialects with a view to determining which of them to 
choose. Furthermore, these descriptions would certainly help decide 
what adequate theoretical framework ought to be used for standard- 
ization of Pulaar. There have been some efforts in some countries 
(Mauritania, Senegal) to promote the existence of language institutes; 
but, for the time being, they employ many unskilled 'linguists' whose 
job (though not trivial) is reduced to basic descriptive analyses of the 
language without any theoretical framework. 



Diop: Language planning across political boundaries: A case study of Puular 97 

Such language institutes would have to work closely together in 
order to yield the best results. What would then be the implications 
of such an apparently ambitious enterprise? I address this and other 
questions in the next section. 

4.1. Implications 

The significance of language planning to the development of 
Africa is beyond any shadow of doubt. One of the contributions 
would be the positive effect that it would have, in the long run, on 
literacy and education, especially in 'Francophone' West Africa. Ac- 
cording to Bokamba (1984b), the illiteracy rate in that part of Africa 
is one of the highest in the continent. French is still a psychological 
barrier to the implementation of African languages in Africa. The 
French colonial language policy is still spreading its effects on the 
continent in term of the psychosociological behavior of its inhab- 
itants. A lot of people in the continent still feel that being educated in 
French is more prestigious and is more of a priority than instruction 
in native languages. 

Bokamba (1984b) quoting Fishman (1971) (in an attempt to 
argue against the latter's typology) outlined three factors deter- 
mining the type of language policy a developing country may adopt. 
One of those has some implications for the present study; it is the 
type c decisions. Type c decisions are those that intervene in a 
situation where there are conflicting or competing great traditions, 
particularly in the absence of a superordinate threat. Language plan- 
ning across political boundaries of the type proposed in this paper 
has the advantage of helping us avoid certain conflicts (though we all 
know it would certainly have some backlash effects). I believe that it 
is possible, though costly, to achieve unity and integration through 
lessening of linguistic rivalries by using authentic African languages. 

One major implication that we should hope to be a corollary of 
LP/LP for Pulaar is that along with the experience of Swahili and 
other 'successful' experiences in Africa, the Pulaar experience might 
contribute toward the creation of major pan-African (or pan-West- 
African) languages of wider communication whose purpose, among 
others, is to create major communication networks in the continent 
and prepare African languages to face the challenging task of ex- 
pressing the Arts and Sciences — with all the advantages that might 
encompass. 

The objectives, ideals, and projected results and implications 
outlined here cannot be reached easily. Some of the problems that 
Pulaar planners may be faced with are outlined in what follows. 

4.2. Issues/problems underlying the development of a 
language policy for Pulaar 

There are various problems underlying the development of a 
language policy for Pulaar or any other language of wider commu- 



98 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

nication (LWC). One of the most important of these (which is also 
thoroughly discussed in Berns (1992)) is that of 'outsider' attitude. 
The fact that Pulaar is a LWC will certainly raise concern on the part 
of speakers of lesser widespread/known languages who might view 
any attempt to develop it for standardized use at the regional level 
as the beginning of the 'death' of their own languages. Berns dis- 
cusses similar attitudes towards English within the EC countries. 
Various EC countries have expressed their concern over the spread of 
the English language (whether it is the British or American variety) 
in their community in general and in their respective countries in 
particular. For instance, she mentions the fact that Mr. Jack Lang, 
France's Minister of Culture, has been quite vocal about English hege- 
mony and '... lack of European identity in the face of the linguistic 
dominance of English.' One way the Community tries to deal with 
such attitudes is by exploring various strategies to teach English 
without giving (non-native) learners the 'impression' that they are 
being taught a language associated with international communication. 
Rather, they are being taught a language associated with British 
Culture just in the same way they could be taught Spanish (associ- 
ated with the culture of Spain), or French (representing the culture of 
France), i.e. 'another variety of European culture and society' (Berns 
1992:12). What this shows is that people's attitudes are heavily 
influenced by the status given to the language they are being taught, 
especially if such language is not their own. Speakers of languages 
other than Fula may not care too much to be taught a foreign 
(African) language that has the bold ambition of becoming 'their 
language of international communication'. Likewise, speakers of dif- 
ferent dialects of Fula may resent a particular dialect being imposed 
on them as their standard. In the remainder of this section I explore 
these ideas in more details. 

4.2.1. (Language) nationalism 

(Language) nationalism, to some extent, can be similar to 
language loyalty. However, while the latter might essentially refer to 
attitude toward, for example, one language, the former is of a greater 
dimension and involves the attitude of an entire nation towards a 
certain issue. (Of course, the obvious question one is tempted to ask 
is 'what is a nation?' The paper does not intend/pretend to answer 
this question.) In the case of India, for example, LP involved a choice 
between totally different languages, thus, totally different identities. 
In the case under study the choice is that of a variety among 
competing varieties of the same language. The issue at stake here is 
twofold. 

On the one hand we have an incompatibility between reliance on 
a foreign language (as expressed in borrowing) and maintenance of 
(the nation's) self-esteem. The more borrowings we find in the lan- 
guage (from a foreign language), it seems, the more native speakers 
of that language are going to feel somewhat 'dependent' on the 
'donor' language. The issue here, however, is that of whether we can 



Diop: Language planning across political boundaries: A case study of Puular 99 

really do much about borrowings. Sir Samuel Johnson illustrated our 
helplessness vis-a-vis borrowings better than anyone else. He 
declared that 'the project for an English Academy failed in the face of 
what he termed 'the Spirit of English Liberty ...', and that '... sounds 
are too volatile and subtle for (legal) restraints; to enchain syllables 
and to lash the wind are equally undertaking of pride' (Johnson, cited 
in Fishman 1971). 

On the other hand, the tension between the requirements of 
modernization and those of authentification creates a sort of dialectic: 
The more stress we exert on real authenticity, therefore, the greater 
the risk of regionalism. The more stress on modernization, the 
greater the risk of loss of maintenance of self-esteem. Achieving a 
balance between the two is certainly something worthy of recog- 
nition. 

4.2.2. Language loyalty 

History suggests that human nature is essentially refractory in 
language matters. Language planning in general involves choosing 
one variety as opposed to another or other varieties as a norm. To 
the extent that this is true it gives the people who speak the chosen 
variety a prestigious position because it raises their status to that of 
norm-bearers and, therefore, may ultimately, though not necessarily 
intentionally, give them power. If we agree that language is an 
expression of personality and a sign of identity and pride, then LP 
'should seek a balance between uniformity and diversity of code" 
(Haugen 1966b:59). Of course that balance is easier talked about than 
achieved. Thus, it is often the case that those involved in the process 
of LP adopt conciliatory attitudes towards the other less 'fortunate' 
varieties. 

When we talk about loyalty in the context under consideration, 
we should distinguish between two kinds of loyalties (or perhaps 
attitudes): 

• loyalty towards one's dialectal variety 

• loyalty towards the colonial language. 

The first kind has, to some extent, been discussed in the 
previous paragraph and basically involves attachment to the variety 
that we grow up speaking and that we are so emotionally and dearly 
attached to. Thus, choosing another variety would ipso facto under- 
mine the salience of the dialect that we speak. However, typological 
data of several developed standards (e.g. Albanian) demonstrate that 
after a certain point in time we are no longer able to claim with 
certainty and precision that the standard comes exactly from a 
particular variety; and that is when speakers of the language identify 
themselves with the standard irrespective of their previous dialectal 
background. Of course, this is not something that will be seen for 
quite some time after the implementation of a norm. But at least it is 
something to look forward to. After all, LP is not an emergency relief 



100 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

operation. It is my position that dialectal preferences should be 
suppressed in the interest of important supranational needs. The 
standard language should be a unifying cultural force; and linguistic 
unity should imply unity of the nation — whatever that 'magic' word 
might mean. 

The second kind of loyalty, on the other hand, is manifested (for 
instrumental purposes?) toward a foreign language or languages. 
Thus, apart from some African supporters of the widespread use of 
national languages in education and administration during the colo- 
nial era in British West Africa, there was a tendency among educated 
Africans to see in their use the danger that progress for the African 
peoples and their integration into the modern world would thereby 
be impaired. This fear still persists. Haugen (1966a) also mentions 
the case of speakers of Garani who would rather use Spanish for 
public affairs. Taken from a purely instrumental point of view those 
attitudes can be somewhat justified. To this array of obstacles one 
should also add problems that draw their roots from egotistic battles 
of political leaders, the subject of the next section. 

4.2.3. Power conflict/elite closure 

'Everything is suffocated if one's own way is not sought and if 
another nation is blindly taken as a model' (Fishman: 1972:8). This 
quote captures the essence of how decision-makers in the various 
countries involved in the LP under discussion in this paper might 
feel vis-a-vis a decision that affects their internal policies in one way 
or the other, and yet is not totally theirs. Their fear is that of loss of 
power, prerogatives, and control; control in terms of manipulating 
decisions. Thus, countries that regard themselves as 'leaders' would 
not want to see themselves lose the power of decision and have 
decisions made for them from without as opposed to from within. 
Furthermore, the differences in the situation in each country and in 
the philosophy of the government, in part, explain the differences in 
policy. This is why a policy that works in one country may fail 
hopelessly in another. To this, one has to add further complications 
brought about by revolving-door governments which proceed, 
generally, with reshuffling and ignoring decisions made by previous 
governments; and this complicates language planning at the 
supranational level. (However, in the case under study I am 
assuming language planning under 'ideal circumstances'.) 

Another fear (on the part of decision-makers) comes from a 
perceived 'threat' that such a 'nationalist(ic)' enterprise might be a 
major step toward 'linking' of the masses to successively higher 
levels of social and political authority. 

In any case the point is that the reasons behind that power con- 
flict could stem from external or internal pressure, or a combination 
of both. An accurate investigation into the nature of this problem is 
above and beyond the scope of this paper. 



Diop: Language planning across political boundaries: A case study of Puular 101 

For the present time one of the biggest problems facing linguists 
who are interested in the project outlined here would be to gain the 
cooperation of different nations involved; and one of the challenges 
that they face is to make scholarly publications about LP/LP compre- 
hensible to policy makers so as to assume their constructive partic- 
ipation. This could perhaps be achieved by having follow-up sessions 
(after conferences) with policy makers and staff members of various 
ministries of education. Political leaders and decision-makers in 
Africa sometimes drag their feet when faced with issues of the kind 
presented in this paper, where Africans are given the option of 
working towards promoting their languages to fulfill the functions 
that were primarily assigned to the language of the colonials. Such 
proposals are often qualified as 'results of leftist movements' in the 
political jargon. Therefore, LP faces more resistance/reluctance on 
the part of politicians than on the part of the people whom it will 
directly affect. However, once we succeed in getting the decision 
makers involved, the other problem would be a linguistic one; and it 
would involve a very lengthy process of selection. As I pointed out 
earlier, in the case of Pulaar, we are dealing with a language that has 
very many dialects. 

4.2.4. Language maintenance 

Another problem that is linguistic in nature has to do with 
maintenance of the variety that is chosen to be standard. Languages 
are dynamic and have to be so lest they die. Language as an organ- 
ism (if I may use the biological metaphor) is subject to the same 
evolutionary development as other organisms. Therefore, the variety 
of Fula that would be chosen as a standard would certainly undergo 
some influences. In the paragraphs to follow I shall briefly outline 
the problem (if it is a problem at all) and try to suggest modest 
solutions to this particular issue. 

Once a variety is chosen and implemented, the next issue is to 
keep further atomization of the standard from exceeding the bound- 
aries of the 'acceptable and reasonable'. There is no consensus on this 
issue. Berns (1992) discusses two views on this (with respect to 
English, at least). One view (Kachru 1985) takes into account multiple 
and variable standards while the other (Quirk 1985, 1990) proposes 
to maintain a single internationally intelligible (native) standard. 
Given the fact that Fula is spoken in a variety of countries, covering a 
considerable amount of geographic territory, influence over any of its 
dialects could come from within as well as from without. The former 
will be the case where other dialects of the language continue to 
influence the standard because, after all, in their own homes, native 
speakers are more than likely to use their own respective dialects 
because they (may) feel (more) comfortable using them. The latter 
refers to the case where other neighboring languages exert some 
influence over the new (and therefore vulnerable) variety. In any 
case, it seems to me that it is more reasonable at this stage to use one 



102 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

standard of Pulaar for all, instead of allowing multiple and variable 
standards, because Pulaar is a long way from enjoying the inter- 
national and 'universal' use that English has enjoyed for centuries. 

The third issue concerns attitudes of parents towards their 
children if they speak to them in the standard medium that they will 
acquire at school. 

No matter what the attitudes of Pulaar speakers are vis a vis 
standard Pulaar, varieties will emerge. French (and its academy) 
illustrates this point. As I suggested earlier, a term used in a partic- 
ular dialect may enter the Terminology Bank and be used legit- 
imately just in case it has already been widely used, and provided it 
does not affect the 'standard grammar', or pose a serious threat to an 
existing term. 

By and large these are some of the (many) problems with 
standardization. In order to reduce the impact of these problems, we 
need to ask: 'Who is it that we are planning for, and for what 
reasons?' The reasons could stem from short-term as well as long- 
term goals. Some of these goals might be: 

• opening up levels of communications, 

• facilitating the publication of texts, 

• gaining a sense of national unity, 

• mainstreaming of the educational system. 

These goals could be more easily reached (and the problems 
outlined earlier reduced) through implementation of reference 
materials, textbooks, de facto language academies, perceived social 
benefits attached to the use of the standard variety, etc. These will 
not only be tools for furthering and reinforcing mother-tongue edu- 
cation but they also assure maintenance of the standard. Of course, I 
am aware of the fact that it is much easier to convince governments 
to spend time and effort to come up with textbooks and other 
teaching materials than to find ways of attaching benefits to the use 
of a particular language. That is political in nature and is left to 
political decision makers. 

5. Conclusion 

5.1. Directions for future research 

Some modest but significant steps have been taken toward 
LP/LP for the African continent in general and for Pulaar in par- 
ticular. However, in order to take those steps further we need to 
convince decision-makers to rely more and more on the collaboration 
of linguists, without whom LP would be at jeopardy because it would 
be carried out empirically (at best) rather than scientifically. 
Pulaarophones have a lot to learn from other Africanists and lin- 
guists from other parts of the world. Future research should first of 
all take the form of thorough attitude surveys of samples that would 
genuinely represent the major dialects of the language. This is going 
to be a lengthy but necessary process. After that, Pulaarophones 



Diop: Language planning across political boundaries: A case study of Puular 103 

need to outline clear goals for standardizing Pulaar and design teams 
(governmental as well as non-governmental) accordingly. 

The project would have to involve linguists from all subfields: 
phonologists, sociolinguists, semanticists, etc. Of course, they alone 
cannot achieve anything without help from the economists who are 
qualified to evaluate the cost-effectiveness of such a project; and this 
takes us back to the first thing to worry about: funding. Without 
funds this project will never take off, but will remain dormant on 
paper like many other projects in the third world. 

While it is true that some African languages may die over the 
years (of natural death?) it is not likely that this would happen to 
Pulaar, given the number of people who speak it as a native language 
and the number of countries where it is spoken. 

African heads of states and politicians in general have always 
used the word 'unity' as their 'motto'. But just as freedom is not free, 
unity is not free either. There is a price to pay for both. If unity is a 
goal that is so dear to us we should be ready to take realistic views, 
one of which has been mentioned by Bokamba (1984b:27): 'very few 
linguists and anthropologists encourage the death of languages, but 
the multiplicity of African languages makes any realist welcome such 
development for national integration purposes.' 

Should our attempts to unify this huge continent not be doomed 
to failure we have to have faith in the fact that there is a need to 
group major geographical areas in Africa under 'major' linguistic 
entities (phase 3 of the political phase of language planning outlined 
earlier). Once that step is taken, the next equally important step 
would consist of assigning (or associating) important functions to 
those languages in the society. This may not be something we can 
achieve in the immediate future, but if achieved, it would most 
certainly help Africans realize that their languages are capable of 
carrying out the functions fulfilled by French (or English, Spanish, 
Portuguese, etc.). Who knows, it might even be the beginning of the 
relegation to lower status of French and other colonial languages in 
the African continent. 

Nevertheless, in order to realize most (if not all) of our goals, we 
need to be aware that it takes efforts from the political scientist, the 
linguist, and the economist, as well as cooperative participation on 
the part of the masses to arrive at meaningful results in LP. 

Last but not least, efforts to standardize Pulaar should start first 
with standardizing the name we use to refer to its native speakers: 
Haalpulaar'en, instead of nonsensical terms such as 'toucouleur', 
'Fulani', 'Fellata', etc., most of which not only are created by colo- 
nialists but seem to suggest actually separate ethnic groups. It does 
not help matters when we see in the literature on the origin of 
HaalPulaar'en many studies that distinguish the 'toucouleurs' from 
the 'Fulanis' or 'Peulh' (cf. for example Chavane 1985) as if they were 



104 Studies in the Linguistic Sciences 23:1 (Spring 1993) ^ 

two separate ethnic groups that happened to speak the same lan- 
guage or borrow aspects of each other's culture. 

If African decision-makers consciously or unconsciously agree 
with and maintain the often implied claim that language planning in 
the form of one language for all is Utopian for the continent or its 
major subregions, they are either accepting, in one way or another, 
an already existing de facto linguistic imperialism or else they are 
making the mistake of thinking that LP/LP is a short-term relief | 
operation rather than what it is in reality: a long-term solution for 
long-standing dilemmas. 

NOTE 

*The original version of this paper was presented at the 
20th ACAL conference at the University of Illinois at Urbana- 
Champaign. It has been revised several times since then. I benefited 
from invaluable comments from Eyamba Bokamba and Braj Kachru. 
To them I would like to express my sincere appreciations. All errors 
contained in this paper are entirely my own. 

REFERENCES 

Ammon et. al. 1987. Sociolinguistics: An international handbook of the 

science of language and society. Berlin: Mouton de Gruyter. | 

Bamgbose, a. 1991. Language and nation. The language question in 

sub-saharan Africa. Edinburgh: Edinburgh University Press. 
Berns, Margie 1992. Sociolinguistics and the teaching of English in 

Europe beyond the 1990s. World Englishes 11:1.3-14. 
BOKAMBA, E. G. 1976. Authenticity and the choice of a national i 

language: the case of Zaire. Presence Africaine 99/100.104-143. j 
. 1984a. Language and literacy in West Africa. Annual review of 

applied linguistics, vol 4: Literacy issues, ed. by R. B. Kaplan, 40- 

74. Rowley, MA: Newbury House. 
. 1984b. French colonial language policies in Africa and its 

legacies. Studies in the Linguistic Sciences 14:2.1-36. 
, & J. S. Tlou. 1977. The consequences of the language policies of 

African states vis-a-vis education. Language and linguistic 

problems in Africa, ed. by P. A. Kotey & H. Der-Houssikian, 35-53. 

Columbia, SC: Hornbeam Press. 
Breton, R. 1991. The handicaps of language planning in Africa. 

Language planning: Focusschrift in honor of Joshua A. Fishman, 

ed. by David F. Marshall, 153-174. Amsterdam: Benjamins. 
Byron, J. L. 1976. Selection among alternates in language 

standardization: the case of Albanian. The Hague: Mouton. 
Chavane, Bruno A. 1985. Villages de I'ancien Tekrour. Paris: Karthala. 
CONFEMEN. 1986. Promotion et integration des langues nationales 

dans les systemes educatifs. Bilan et inventaire. Paris: Editions 

Champion. 



Diop: Language planning across political boundaries: A case study of Puular 105 

Cooper, R. L. 1989. Language planning and social change. Cambridge: 

Cambridge University Press. 
Crystal, D. 1987. The Cambridge Encyclopedia of Language. 

Cambridge: Cambridge University Press. 
DUMONT, P. 1983. Le fran^ais et les langues africaines au Senegal. 

Paris: Agence de Cooperation Culturelle et Technique. 
EguCHL P- 1986. An English-Fulfulde dictionary. Tokyo: Institute for 

the Study of Languages and Cultures of Asia and Africa. 
Fall, Aram 1981. La politique de promotion des langues nationales 

dans cinq etats de I'Afrique occidentale: Gamble, Guinee, Mali, 

Mauritanie, Senegal. UNESCO: La definition d'une strategic 

relative a la promotion des langues Africaines. Documents de la 

reunion d'experts qui a eu lieu a Conakry, Guinee, du 21-25 

septembre 1981, pp. 192-218. Paris: UNESCO. 
Ferguson, C. 1966. National sociolinguistic profile formulas. 

Sociolinguistics: Proceedings of the UCLA Sociolinguistics 

Conference, 1964, ed. by W. Bright, 309-315. The Hague: Mouton. 
FiSHMAN, J. A. 1971. Sociolinguistics: a brief introduction. Rowley, MA: 

Newbury House. 
. 1972. Language and nationalism. Two integrative essays. 

Rowley, MA: Newbury House. 

. 1974. Advances in language planning. The Hague: Mouton. 

, & Gupta. 1968. Readings in the sociology of language. The 

Hague: Mouton. 
GNALIBOULY, B. 1988. Problemes de lexicologie et de lexicographic: le 

cas du Fulfulde. Seminaire regional de I'UNESCO sur les 

problemes de lexicologie en Afrique. Niamey, Niger, MS. 
GreENBERG, J. 1955. Studies in African linguistic classification. New 

Haven, CT: Yale University Press. 
HauGEN, E. 1966a. Language conflict and language planning. 

Cambridge, MA: Havard University Press. 
. 1966b. Linguistics and language planning. Sociolinguistics, ed. 

by W. Bright, 50-71. The Hague: Mouton. 
JERNUDD, B., & J. Rubin. 1971. Can language be planned? Honolulu: 

University Press of Hawii. 
, et al. 1989. The politics of language purism. New York: Mouton 

de Gruyter. 
Ka, F. 1983. Le Pulaar en Mauritanie. Etude dialectologique Institut 

des langues nationales. Nouakchott: Institut des Langues 

Nationales. 
KacHRU, B. 1985. Standards, codification, and sociolinguistic realism: 

the English language in the outer circle. English in the world: 

Teaching and learning the language and literatures, ed. by R. 

Quirk & H. Widdowson, 11-30. Cambridge: Cambridge University 

Press. 
Labouret. 1952. La langue des Peuls ou Foulbe. Memoires de 

ITnstitut Frangais d' Afrique Noire 16. Dakar: Institute Fran9ais 

d'Afrique Noire. 



106 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

MKZEI, R. 1989. Somali language and literacy. Language Problems and 

Language Planning 13:3.211-223. 
NEUSTUPNY, J. V. 1970. Basic types of treatment of language problems. 

Linguistic Communications (Monash University) 1.77-100. 
OAU Bureau of Languages. 1973-80. Reconsideration of African 

language policies. Kampala: Organization of African Unity. 
Quirk, R. 1985. The English language in a global context. English in 

the world: Teaching and learning the language and literatures, 

ed. by R. Quirk & H. Widdowson, 1-6. Cambridge: Cambridge 

University Press. 
. 1990. Language varieties and standard language. English Today 

6.3-10. 
Ray, p. S. 1963. Language standardization. The Hague: Mouton 
Rubin, J. 1973. Language planning: current issues and research. 

Washington, DC: Georgetown University Press. 
, & Jernudd Bjorn (eds.) 1971. Can language be planned? 

Honolulu, HI: East-West Center Press. 
Sow, A. I 1977. Langues et politiques des langues en Afrique noire. 

Paris: Nubia. 
Spencer, J. (ed.). 1963. Language in Africa. London: Cambridge 

University Press. 
. 1971. Colonial language policies and their legacies. Current 

Trends in Linguistics 7.537-574. 
TOLLEFSON, J. W. 1991. Planning language, planning inequality: 

Language policy in the community. London: Longman. 
TURCOTTE, D. 1981. La politique linguistique en Afrique francophone. 

Quebec: Les Presses de I'Universite de LAVAL. 
WHITELY, W. H. 1966. Social anthropology, meaning, and linguistics. 

Man (n.s.) 1.139-157. 

1971. Language policies of independent African states. 

Linguistics in Sub-Saharan Africa, ed. by T Sebeok. Current 

Trends in Linguistics 7.548-558. 
. 1972. To plan is to choose. Bloomington, IN: Indiana University 

Press. 
Yanga, T. 1978. Language planning and onomastics in Zaire. Studies 

in African Linguistics 9:2 233-244. 
ZOUBKO, G. V. 1980. Dictionnaire Peul-Russe-Fran^ais. Moscou: Editions 

Langues Russes. 



Studies in the Linguistics Sciences 
Volume 23, Number 1, Spring 1993 

BIMORAICITY IN MONOSYLLABIC CHICHEWA IDEOPHONES* 

Andrew Tilimbe Kulemeka 
University of Louisville 

Kanerva (1990) has successfully shown that the 
Chichewa phonological word is disyllabic in underived 
major-category words such as nouns, verbs, and adjectives. 
Moreover, in such underived forms a mora is equal to a 
syllable. It follows that a phonological word is minimally 
bimoraic. 

In Chichewa ideophones a phonological word is also 
disyllabic. However, underived monosyllabic ideophones 
are also bimoraic. Thus, in underived monosyllabic ideo- 
phones the mora and syllable counts are not identical. This 
paper presents crucial evidence from cliticization, penulti- 
mate lengthening, and facts of pronunciation which reveal 
the asymmetry between the mora and syllable counts in 
monosyllabic ideophones. The paper also demonstrates how 
monosyllabic ideophones become disyllabic. Further, I raise 
an important question about the interaction between 
regular and ideophone phonology. 

1. Introduction 

Ideophones or expressive words have been identified in many 
languages of the world. For instance, expressive forms exist in 
Austronesian languages (Brandstetter 1916), including Javanese 
(Uhlenbeck 1971) and Malay (Collins 1974); in Korean (Lee 1992, Lee 
1993); in Dravidian languages, such as Kota, Toda, and Kannada 
(Emeneau 1969); in Indo-Aryan (Dimock 1957); in Thai (Haas 1946); 
in Japanese (Alfonso 1966); in Chadic (Newman 1968); and in Niger- 
Congo (Samarin 1965), particularly in Bantu (Torrend 1891, Samarin 
1971, Kulemeka 1993). 

The bulk of the literature on ideophones stresses the point that 
ideophones have a different phonological system from nouns, verbs, 
adjectives, and so forth (Childs 1988, 1989; Doke 1954; Fivaz 1963; 
Mphande 1989). While this may be correct to a certain extent, it 
needs to be substantiated. With respect to Southern Bantu, so far I 
have not come across any study which systematically investigates 
the nature and degree of phonological differences and similarities 
between ideophones on the one hand, and other lexical elements on 
the other. Most authors on Southern Bantu make statements about 
the aberrant phonology of ideophones without supporting those 
claims with empirical investigation. This study fills a gap in our 
understanding of the ideophone by carrying out an empirical 
investigation of the structure of monosyllabic ideophones. 



108 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Most writers on ideophones in Southern Bantu languages claim 
that ideophones tend to have long vowels (Fortune 1962; Zondo 
1982; Mphande 1989), even though long vowels are often not 
phonemic in those languages. This is true of Chichewa as well. But 
careful observation of the distribution of long vowels in Chichewa 
ideophones clearly reveals the following: (i) All monosyllabic ideo- 
phones have underlying long vowels (la); (ii) disyllabic ideophones 
may have underlying long or short vowels (lb); (iii) longer, 
'polysyllabic' ideophones do NOT have underlying long vowels (2). (An 
accent indicates high tone; low tone is not marked.) 

(1) a. Monosyllabic 

Ih'v.l 'intense dark" 

/dii:/ 'not saying anything' 

/gwa:/ 'hard and strong' 

b. Disyllabic 

/ce:te/ 'absolute silence' 

/"do:to:/ 'something worthwhile' 

/kwi:ci:/ 'stop suddenly' 

/p^oco/ 'fall heavily' 

/pwimp^^wi/ 'of a large object sitting' 

(2) Polysyllabic 

/k'^olop^'et^e/ 'numerous, abundant' 

/gobede/ 'hard items knocking against each other' 

/balama"t^u/ 'suddenly appear" 

The asymmetrical distribution of long vowels in (1) and (2) raises the 
question of the possible connection between vowel length and the 
minimal phonological unit in the ideophone. It seems that mono- 
syllabic ideophones minimally have to be bimoraic (see e.g. Hayes 
1981 and Hyman 1985 for issues relating to syllable weight). 

The fact that the smallest possible ideophonic form has the 
structure [(C)|i(i] and forms a single syllable is interesting when it is 
related to the fact that in the rest of the lexicon one mora equals one 
syllable in underived words (Kanerva 1990). The question that needs 
to be addressed is whether the minimal bimoraic ideophone is mono- 
or disyllabic. In order to answer such a question we need to briefly 
examine Kanerva's 1990 analysis and evidence. 

Kanerva demonstrates that underlyingly all Chichewa syllables 
are monomoraic. Bimoraic syllables are derived through penulti- 
mate-syllable vowel lengthening.! One piece of evidence for 
Kanerva's analysis involves the assignment of high tone on the 
penultimate syllable. I summarize his arguments on high tone 
assignment immediately because they are relevant to my analysis. 

Kanerva 1990 proposes that Chichewa has a tonal rule which 
assigns a high tone on the penultimate syllable of the verb. One of 
the triggers for the rule is the negative imperative marker -sa-. 
Kanerva indicates that -sa- 'removes any other tone the verb might 



Kulemeka: Bimoraicity in monosyllabic Chichewa ideophones 



109 



have and assigns a penultimate High tone' (1989:33). So, for example, 
the high tones of the subject and the object marker are assigned to 
the penultimate syllable as shown in (3). (Here, as elsewhere, 'OM' in 
the gloss refers to an object marker.) 



(3) 



p'^ika 
ti-p^'ika 
ti-ci-p^ika 
ti-sa-ci-p^ike 



'cook' 

'we-cook' 

'we-OM-cook' 

'we-Neglmp-OM-cook' 

The pattern of penultimate high tone assignment becomes critical 
when a word has an underlying sequence of vowels, and it needs to 
be determined if those vowels form one heavy bimoraic syllable, 
CVV, or two light syllables, CV.V. The elements /boola/ 'pierce' and 
/sauka/ 'suffer' are used to illustrate the point. Neither word has an 
underlying high tone (cf. Kanerva 1989, example 9). 



(4) a. boola 

ti-boola 
ti-ci- boola 
ti-sa-ci-bo61e 
b. sauka 
ti-sauka 
ti-mu-sauka 
ti-sa-mu-saiike 



pierce 
we-pierce' 
we-OM-pierce' 
we-Neg-OM -pierce' 
suffer' 
we-suffer 
we-OM-suffer' 
we-Neg-OM -suffer' 



As shown by example (4), the rule which assigns high tone to the 
penultimate syllable reads the sequence of two vowels in /boola/ 
and /sauka/ as falling into DIFFERENT syllables, schematically 
represented as follows. 



(5) 



H 



b. 



H 



^ 

s a 



/I 

ka 



bo 



/I 



The penultimate syllable high tone rule is highly productive in 
Chichewa tonology. Its appearance on the penultimate syllables in 
the examples in (4) provides one strong piece of evidence for 
treating the penultimate vowels as separate syllables. 

Further, Kanerva uses the penultimate tone assignment rule to 
also argue that the mora count is exactly identical to the syllable 
count in underived words in Chichewa. Moreover, he shows that the 
minimal phonological word for major category words in Chichewa is 
disyllabic. The last statement entails that the minimal underived 
phonological word has two moras. 

The question, then, arises whether the minimal ideophone also is 
disyllabic and bimoraic. If so, does this mean that the minimal 
ideophonic unit is equivalent to a single phonological word? 



110 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

If the smallest ideophone is equal to a disyllabic word, then it 
follows that the term 'monosyllabic ideophones' is inappropriate to 
refer to these forms. In fact it would mean that, with respect to the 
smallest word possible, ideophones do not violate the rules of regular 
phonology. 

In the following I show that the minimal ideophonic unit is not 
disyllabic and thus violates the rules of regular phonology. In fact, 
because monosyllabic ideophones (though bimoraic) do not count as 
disyllabic, they amount to less than a phonological word in the 
language, even though they are full morphological words. 

Three pieces of evidence support this position: the failure of 
clitics to attach to monosyllabic ideophones; the position of 
penultimate vowel lenghthening in ideophones which have become 
disyllabic; and differentiations of pronunciation between monosyl- 
labic and disyllabic ideophones. 

2. Clitic attachment 

Nespor & Vogel (1986), Zee & Inkelas (1990), and Kanerva 
(1990) all agree that clitics attach to full words and no less. 
Moreover, through a very detailed analysis of different types of 
bound morphemes, Inkelas (1989) shows that clitics are sensitive to 
prosodic features, rather than morphological ones. Hence clitics 
attach to FULL PHONOLOGICAL WORDS. In Other words, we can test 
whether a form counts as a phonological word in a language by 
attaching a clitic to it. 

/nso/ 'also' and the emphasizer /di7 are high-toned clitics in 
Chichewa. Like clitics in other languages 'they may never occur alone; 
that is they may not be the only element of an utterance' (Nespor & 
Vogel 1986:149; see also Inkelas & Zee 1990). Chichewa clitics are 
further distinguished from other forms of affixes, such as verbal 
extensions, because they alone can attach to all major category words 
which are minimally disyllabic, such as verbs, nouns, adverbs, 
possessive pronouns, and adjectives. In addition, clitics such as 
/-nso/ or /-di7 modify the basic meaning of their hosts and 
sometimes the entire sentence along somewhat predictable lines. 

Kanerva (1990) shows that Chichewa has a rule which assigns a 
high tone to the syllable preceding the clitic. Thus, for example, low 
tone elements as in (6) acquire high tone through the clitic high tone 
assignment rule as follows. 

In all the cases in (6) we see that a low-toned form acquires a 
high tone on the syllable preceding the clitic through the clitic high 
tone assignment rule which assigns a high tone to the left syllable of 
an element to which the clitic is attached. 

Underived monosyllabic ideophones fail to take clitics, as shown 
in (7a). In contrast, di- and polysyllabic ideophones do accept clitics, 
cf. (7b). 



Kulemeka: Bimoraicity in monosyllabic Chichewa ideophones 
(6) a. mu-nthu mu-nthu-nso 



person 



H 



'also a person' 



mu-nthu-nso 
H 



111 



(7) 



galu 



'dog' 



galu-nso 



H 

'also a dog' 



galu-nso 
H 



Monosvl 


able 






bu: 


'light a fire' 


*bu:-nso 




ci: 


'throw' 


*ci:-nso 




mbwe: 


'arrive' 


*mbwe:-nso 




my a: 


'smooth' 


*mya:-di 




bi: 


'intense dark' 


*bi:-di 




bi: 


'convincing' 


*bi:-di 




Di- & polvsvllabic 






ndo:to: 


'something 


ndo:t6:-nso 


'also something 




worthwhile' 




worthwhile' 


cololo 


'pouring liquid' 


cololo-nso 


'again pour liquid' 


psi:ti: 


'completely 


pi:ti:-di 


'indeed 




finished' 




completely 
finished' 


delu 


'pass quickly' 


delu-di 


'indeed pass 
quickly' 


gankha 


'to be and look 


gankha-nso 


'to also be and 




thick headed 
and ignorant' 




look thick headed 
and ignorant' 



As shown in (6) and (7), the high tone which comes with the clitic 
gets assigned to the syllable preceding the clitic in a di- or 
polysyllabic form. Such a syllable will naturally be preceded by one 
or more syllables. In other words, the form to which the clitic 
attaches is minimally disyllabic, a phonological word. Clitics fail to 
attach to monosyllabic ideophones. It is only when the structure of 
monosyllabic ideophones is changed to become disyllabic that clitics 
attach and assign a high tone on the clitic's preceding vowel (see 
further below). 

The low-toned ideophones in (7a) further demonstrate that the 
high tone of the clitic fails to appear. For the high tone of the clitic to 
show up, again a second syllable is required. ^ 

If the mora count were identical to the syllable count in un- 
derived monosyllabic ideophones, we would expect that clitics would 
attach to the forms in (7a) because they are bimoraic and thus would 



112 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

be disyllabic. The high tone of the clitic should surface on the second 
mora, which would be the second syllable. Since the high tone of the 
clitic fails to appear on the second mora in low-toned monosyllabic- 
bimoraic ideophones, we have to conclude that the mora count in 
monosyllabic ideophones is not identical to the syllable count. There- 
fore the strict identity between mora count and syllable count 
holding for other category words, argued for by Kanerva (1990), does 
not apply to monosyllabic ideophones. 

However, it is still interesting that the minimal ideophone is 
bimoraic. In a way, therefore, the minimal ideophone partially 
resembles the smallest phonological word, which also has two moras, 
but in separate syllables. 

A further interesting aspect of monosyllabic ideophones is that 
they do not have an upper limit in terms of the length of their 
vowels. Thus, depending on expressive needs, the vowel length of the 
minimal ideophonic unit is completely controlled by the speakers. 
For example, speakers might say Ih'v.l or [bi:::] or [bi::::] or [bi::::::] or 
[bi::::::::::].3 In contrast, one cannot expand vowels in verbs, nouns, 
adverbs, or adjectives indefinitely. Thus, excessive lengthening in 
words like /gula/ 'buy', /mwana/ 'child', and /tali/ 'tall', such as 
*[gu::la], *[mwana::], or *[ta:::li:::], is ungrammatical. 

Although the length of vowels in the minimal ideophone is 
unlimited, length by itself does not appear to compensate for phono- 
logical wordhood in Chichewa. That is, even when lengthened, 
monosyllabic ideophones do not accept clitics. 

Thus we are forced to conclude that even though the smallest 
underived ideophone is bimoraic, i.e. contains a long vowel, the two 
moras comprise one syllable. This is in marked contrast to the 
behavior of 'ordinary' bimoraic words. Now, the fact that mono- 
syllabic bimoraic ideophones have long vowels is uncontroversial and 
has been acknowledged by almost all researchers on the subject. 
What is significant, however, is the asymmetry between 'ordinary' 
words and ideophones as regards the relation of mora count to 
syllable structure. 

Note further that in order for clitics to attach to it, the mono- 
syllabic ideophone has to become disyllabic. To do this, the second 
mora to the right delinks to form a separate syllable. The evidence 
for such a process comes from the interaction of monosyllabic ideo- 
phones with high toned clitics like /-nso/ 'also' and /-di7 'indeed', 
facts of pronunciation, and the occurrence of the phrase-final rule of 
penultimate lengthening. 

Let us first examine the effects of high-toned clitics such as 
/nso/ and /di7. The structural description for cliticization requires a 
preceding form that is minimally disyllabic [aa]. Monosyllabic 
ideophones, however, do not meet this requirement. As shown in 
(7a), ordinarily these clittics therefore cannot attach to monosyllabic 
ideophones and thus fail to pass on their high tone. 



Kulemeka: Bimoraicily in monosyllabic Chichewa ideophones 



113 



However, if the second mora of a monosyllabic ideophone 
delinks and forms a separate, second syllable, then the clitics can 
attach and pass their high tone to the newly created preceding 
syllable. Note specifically that it is this second syllable alone which 
bears the high tone of the clitic. Compare the derivations in (8). (The 
derivations are illustrated with the low-tone monosyllabic ideo- 
phones /mbwe:/ 'arrive' and /bu:/ 'light a fire', plus the high-tone 
clitic /nso/ 'also'.) 

(8) a. Underlying representation 

a ^ 





b u 



Delinking and resyllabification 
H 



mbw 




e + nso 




c. Clitic high tone assignment 
H 



H 



A 

A W 

mbw e e -t- nso 



A 



o a a 



b u u + nso 



d. Surface forms 

[mbwe. e. nso] [bu.ii.nso] 

Further support for the claim that the pre-clitic vowel in (8d) is 
a separate syllable comes from pronunciation. There is an obvious 
break between it and the preceding vowel, as in [mbwe. e. nso] and 
[bu.ii.nso]. This inter-vowel break is absent in the corresponding 
unmodified monosyllabic ideophones /mbwe:/ and /bu:/ (not 
*[mbwe.e], *[bu.u]). Note that a similar break is evident in 'ordinary' 
words. For instance, the forms /boola/ and /sauka/ of (4) above are 
pronounced [bo.o.la] and [sa.u.ka], with an obvious break and tran- 
sition between the neighboring vowels. This shows that, even when 



114 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

the two vowels are identical, they are not a single long vowel but 
form separate syllables. 

A final piece of evidence in support of the analysis involves the 
common phrasal rule of penultimate vowel lengthening. Significantly, 
in structures like those of (8d), such as [mbwe.e.nso] and [bu.ii.nso], 
the process only lengthens the penultimate vowel, as in 
[mbwe.ec.nso] and [bu.u:.nso]. If the long vowels of these forms had 
not undergone resyllabification, we would expect penultimate 
lengthening to have affected the entire long vowel, yielding 
*[mbwe:(:).nso] and *[bu:(:).nso]. The fact that we get [mbwe.e:.nso] 
and [bu.u:.nso] therefore provides strong evidence for treating the 
sequences of two vowels as comprising separate syllables. 

Four pieces of evidence, then, provide strong empirical support 
for arguing that in order to accept clitics, monosyllabic ideophones 
undergo a process of delinking and resyllabification: (i) rejection of 
clitics by unmodified monosyllabic ideophones; (ii) placement of the 
clitic high tone on the SECOND syllable of modified ideophones, (iii) a 
break and transition between the first and second syllables of such 
ideophones; and (iv) the fact that penultimate vowel lengthening 
affects only the second syllable of such structures. 

3. Conclusion 

This paper has shown that Kanerva's claim (1990) that there is a 
one-to-one correspondence between mora and syllable count in 
underived words of Chichewa cannot be fully maintained. While the 
claim holds true for underived 'ordinary' words, ideophones show a 
remarkably different phonological pattern: Although bimoraic, they 
are monosyllabic, not disyllabic as expected. As such they constitute 
sub-minimal phonological words and therefore cannot host clitics, 
unless modified by delinking and resyllabification. 

These differences in prosodic phonology between ideophones 
and 'ordinary' words raise important questions which may require a 
reevaluation of the relationship between ideophone and regular 
phonology. For instance, to what extent are other conclusions about 
regular phonology applicable to ideophone rules? Is ideophone 
phonology distinct or a part of regular phonology? The extent of such 
a reevaluation can only be gauged after further and more extensive 
investigation of ideophone behavior. 

NOTES 

* A version of this paper was presented at the 24 Annual 
Conference on African Linguistics at Ohio State University, Columbus. 
I am grateful to Jonni Kanerva, Charles Kisseberth, Lupenga Mpande, 
Robert Botne, Francis Moto, and Hans Henrich Hock for their very 
useful comments which have helped me to clarify my points and 
provide additional supporting evidence. I also thank an anonymous 



Kulemeka: Bimoraicity in monosyllabic Chichewa ideophones 115 

reviewer for SLS for making pertinent observations. Of course all 
errors in the paper are entirely my fault. 

1 Kanerva (1990) does not include ideophones in his discussion. 

2 Several native speakers of Chichewa (Dr. Francis Moto, Dr. 
Lupenga Mphande, and Mr. Simango), who attended the 24th Annual 
Conference of African Linguistics at Ohio State University, have 
provided further support for the view that monosyllabic ideophones 
cannot take clitics. When asked to produce monosyllabic ideophones 
with clitics, they invariably made the ideophone disyllabic as shown 
later in this section. 

3 The symbol /::/ represents a vowel whose length is double the 
usual long length, /:::/ represents thrice the usual long length, etc. 

REFERENCES 

Alfonso, A. 1966. Japanese language patterns, volume I. Tokyo: 

Sophia University Press. 
BranDSTETTER, Renward. 1916. An introduction to Indonesian 

Linguistics. London: The Royal Asiatic Society. 
Childs, G. Tucker. 1988. The phonology of Kisi ideophones. Journal of 

African languages and linguistics 10.165-190. 
. 1989. African ideophones. Sound symbolism, ed. by Leanne 

Hinton, John Nichols, & John Ohala. Forthcoming. Cambridge: 

Cambridge University Press. 
Collins, James T. 1974. Expressives in Kedah Malay. Pacific 

linguistics: South East Asian linguistic studies 4.379-406. 
DIMOCK, E. C. 1957. Symbolic forms in Bengali. Bulletin of the Deccan 

College Research Institute 18.22-29. 
DOKE, Clement M. 1954. The southern Bantu languages. London: 

Oxford University Press. 
EMENEAU, M. B. 1969. Onomatopoeics in the Indian linguistic area. 

Language 45.274-299. 
FiVAZ, Derek. 1963. Some aspects of the ideophone in Zulu. (Hartford 

Studies in Linguistics, 4) Hartford: Hartford Seminary Founda- 
tion. 
Fortune, George. 1962. Ideophones in Shona. London: Oxford 

University Press. 
Haas, Mary. 1946. Techniques of intensifying in Thai. Word 2.127- 

130. 
Hayes, Bruce. 1981. A metrical theory of stress rules. MIT Ph.D. 

dissertation in Linguistics. (Revised version distributed by the 

Indiana University Linguistics Club.) 
HYMAN, Larry. 1985. A theory of phonological weight. Dordrecht: 

Foris Publications. 
Inkelas, Sharon. 1989. Prosodic constituency in the lexicon. Stanford 

University Ph.D. dissertation in Linguistics. 
Kanerva, Jonni. 1989. Focus and phrasing in Chichewa phonology. 

Stanford University Ph.D. dissertation in Linguistics. 



116 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

. 1990. Focus and phrasing in Chichewa phonology. New York: 

Garland. 
Lee, Jin-Seong. 1992. Phonology and sound symbolism of Korean 

ideophones. Indiana University Ph.D. dissertation in Linguistics. 
Lee, Yongsung. 1993. Topics in the vowel phonology of Korean. 

Indiana University Ph.D. dissertation in Linguistics. 
KULEMEKA, Andrew T. 1993. The status of the ideophone in Chichewa. 

Indiana University Ph.D. dissertation in Linguistics. 
McCarthy, John, & Alan Prince. 1986. Prosodic morphology. Univer- 
sity of Massachusetts at Amherst, and Brandeis University, MS. 
MPHANDE, Lupenga. 1989. A phonological analysis of the ideophone in 

Chitumbuka. University of Texas at Austin Ph.D. dissertation in 

Linguistics. 
NESPOR, Marina, & Irene Vogel. 1986. Prosodic phonology. Dordrecht: 

Foris Publications. 
Newman, Paul. 1968. Ideophones from a syntactic point of view. 

Journal of West African languages 2.107-17. 
Samarin, William J. 1965. Perspective on African ideophones. African 

Studies 24.117-121. 
. 1971. Survey of Bantu ideophones. African language studies 

12.130-168. 
TORREND, J. 1891. A comparative grammar of the South African Bantu 

languages. London: Kegan Paul, Trench, Trubner and Co. 
Uhlenbeck, E. M. 1971. Peripheral verb categories with emotive- 
expressive or onomatopoeic value in modern Javanese. Travaux 

linguistiques de Prague 4.145-156. 
Zec, Draga, & Sharon Inkelas. 1990. Prosodically constrained syntax. 

The phonology-syntax connection, ed. by Sharon Inkelas and 

Draga Zec, 365-378. Chicago: University of Chicago Press. 
ZONDO, J. 1982. Some aspects of the ideophone in Ndebele. Zambezia 

10.111-126. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 

A RE-EXAMINATION OF CARDINAL VOWELS 
AND AUDITORY EQUIDISTANCE* 

Sharon R. Morrison 

The present study examines the hypothesis of auditory 
equidistance of Jones's primary cardinal vowels. Four sets 
of previously recorded cardinal vowels spoken by Daniel 
Jones were digitally analyzed. The obtained formant peaks 
were used to compute Euclidean distance measures be- 
tween adjacent vowels in a F1/F2 vowel space. ANOVAs of 
the linear inter-vowel distances (in Hz) and distances in 
several non-linear scales (Bark, semitone, and auditory- 
perceptual) were interpreted as indicating that the percep- 
tual inter-vowel distances of the cardinal vowels were 
significantly different. Separate analyses of the front and 
back vowel series did not produce consistent results. 



1. Introduction 

The cardinal vowel system was developed by Daniel Jones in the 
early 1900's, and the first published description of cardinal vowels 
appeared in the 1917 edition of An Outline of English Phonetics (see 
Jones 1956). Cardinal vowels were intended as a descriptive system 
for vowels based on articulatory positions of the tongue. Jones taught 
that all the cardinal vowels were articulatorily equidistant, that the 
difference between any two adjacent vowels involved an equal 
displacement of the tongue. For example, the difference in tongue 
position between the first and second cardinal vowels (CVl and CV2) 
should be equal to the difference between CV2 and CV3 (see Figure 
1). The top and bottom distances between front and back vowel 
series (CVl and CV8; CV4 and CV5 respectively) were not specified 
as equidistant in Jones's system. 

8 



2 V 

4 



Figure 1. The cardinal vowel figure or trapezium 



118 Studies in the Linguistic Sciences 23:1 (Spring, 1993) 

The cardinal vowels were used almost exclusively in the British 
school of phonetics. Training in the use of cardinal vowels was a long 
and laborious procedure. The vowels were to be taught in an oral 
tradition (by Jones or his students). To use the cardinal vowels as a 
descriptive technique one learned to produce each vowel, paying 
particular attention to the position of the tongue. To describe a 
natural vowel the user would learn to produce that vowel to the 
satisfaction of the native speaker (informant) and then propri- 
oceptively compare the articulation of that vowel to the closest 
cardinal vowel. After determining the articulatory relationship 
between the natural vowel and the cardinal vowel, the natural vowel 
was placed in the appropriate location on the cardinal vowel figure 
(trapezium) along with a notation of lip rounding (Abercrombie 
1967, 1985). At some point the cardinal vowel figure came to be 
viewed as an auditory space, not an articulatory space (Ladefoged 
1960a, 1967, 1982; O'Connor 1973). 

One reason for considering the cardinal vowels as reference 
points in an auditory space is that x-rays of the cardinal vowels did 
not show articulatory equidistance of tongue position (S. Jones 1929; 
no relation to Daniel Jones). Stephen Jones did not publish a scale or 
measurements of tongue position with his x-rays, and the 
reproductions were poor. Still, the differences in tongue positions did 
not appear equidistant, especially for the back vowels (CV 4 - CVS). 
Ladefoged (1960a, 1967, see also 1982) measured differences in 
tongue position in S. Jones's (1929) x-rays and found no indication of 
articulatory equidistance. 

Ladefoged (1960b) investigated the usefulness of cardinal 
vowels as a classification tool. He compared the classification of Gaelic 
vowels by observers trained in the British school of phonetics (who 
were extensively trained in the proprioceptive use of cardinal 
vowels) and those not formally trained in the use of cardinal vowels. 
Ladefoged concluded that the 15 British-trained observers seemed 
more consistent, as a group, about the placement of the Gaelic vowels 
on the trapezium than were the three other subjects not trained 
extensively in the use of cardinal vowels. Ladefoged's conclusions 
were interesting, but they were not based on any inferential 
statistical analyses. 

Laver (1965) studied the variability, over time, of the placement 
of synthetic vowels on the cardinal vowel diagram. All of his subjects 
were trained extensively in the use of cardinal vowels. Vowel 
placement was found to vary from trial to trial, though the average 
locations for each vowel showed a high agreement within the group. 
Laver also found that, for each subject, there was no significant 
change in overall variability over time.' 

The usefulness of cardinal vowels could be furthered by 
resolution of the issue of equidistance. Obviously, cardinal vowels are 
not articulatorily equidistant (S. Jones 1929; Ladefoged 1960a; 1967, 
1982), but are they equidistant in an auditory frame work? 



Moirison: A re-examination of cardinal vowels and auditory equidistance 119 

Ladefoged (1982) indicated that, if cardinal vowels are plotted on a 
formant chart, there is no way to get the distances between the front 
vowels to be the same as the distances between the back vowels. His 
claim is probably based on the spectrographic analysis of cardinal 
vowels described by Ladefoged, (1960a; 1967). Ladefoged 1960a is a 
monograph based on his Ph.D. thesis, and a slightly revised version of 
this monograph appears as the second chapter of Ladefoged 1967. 
The main difference between the two versions is that the 1960a 
form has slightly more detail about methods used, larger figures and 
also includes a table of unsealed cardinal vowel formants. 

Ladefoged 1960a, 1967 examined the nature of vowel quality on 
a number of levels. The relevant section (or chapter) is the third one, 
'The acoustic analysis of cardinal vowels'. In this section Ladefoged 
used cardinal vov/els to study the meaning of phonetic quality. 
Eleven experienced phoneticians, all trained extensively in the use of 
cardinal vowels by Daniel Jones himself, produced multiple sets of 
cardinal vowels in various pitches. Each subject was rehearsed in the 
eight primary cardinal vowels by Jones and then produced at least 
five sets of cardinal vowels. Each vowel was produced in isolation. 
Each vowel set was criticized and discussed by Jones and the 
informant. Jones was the final judge of vowel quality and selected 
three (sometimes only two) sets of vowels for each informant (a total 
of 31 sets out of 92 sets) as 'good complete sets not in the extreme 
pitch ranges' (Ladefoged 1967:78). 

The goal of Ladefoged's acoustic analysis was to determine the 
frequency of each of the first three formants for all 248 vowels. In 
the course of analysis at least two and often more spectrograms were 
made of each vowel. All acoustic analyses were done on a modified 
Kay Sonograph. In addition to the normal frequency scale (1"=2000 
Hz) an expanded (1"=1000 Hz) and ultra-expanded (1"=500) scale 
were added. Acoustic analysis was based on wide band spectrograms 
and sections (a narrow band spectrum cross section at a given instant 
in time). Ladefoged (1960a, 1967) reported difficulty finding Fl for 
the high front and back vowels [i] and [u] (CVl and CVS) as well as 
differentiating Fl and F2 when they were close together as found in 
tokens of [a] and [o] (CVS and CV6). The second and third formants 
were also hard to find for some vowels. Ladefoged also reported 
difficulties with spurious formants in unexpected locations and 
discussed the poorly defined nature of formants. 

Ladefoged presented the results of his analyses of 31 sets of 
cardinal vowels spoken by 1 1 trained phoneticians. Many of the 
vowels could not be adequately described in terms of formant 
frequencies, especially the back vowel series. For 16 of 31 cases 
Ladefoged was unable to distinguish the first two formants for [a] 
(CVS), in 12 out of 31 times he could not determine either Fl or F2 
for [o] and [o] (CV6 and CV7), and in 26 out of 31 he could not 
completely specify formants for [u] (CVS). Formants were presented 
as hertz and mel (Stevens & Volkmann 1940) values in Ladefoged 



120 Studies in the Linguistic Sciences 23:1 (Spring, 1993) 

1960a and only mel values in Ladefoged 1967. Ladefoged 1960a; 
1967 first plotted all sets of vowels in a traditional F1/F2 inverted 
plot and found that only CV 4 and CVS ([a] and [u]) did not exhibit 
overlapping formant regions. CVl and CV2 and CV6 and CV7 ex- 
hibited the most overlap. Ladefoged then tried to use various 
combinations of the first three formants to 'normalize' formants for 
the back vowels and thereby differentiate them. None of the at- 
tempts appeared successful. Ladefoged discussed the notion that the 
phonetic quality of vowels apparently does not depend on absolute 
acoustic value of the formants, but perhaps on the relationship 
between the formants of all the vowels produced by a given speaker. 
Following this viewpoint, he plotted each sets of cardinal vowels 
spoken by each phonetician separately. In the individual plots the 
relationship between the cardinal vowels was clearer than in the 
aggregate plot. In the individual plots more vowels were differ- 
entiable. 

There are several limitations to Ladefoged 1960a, 1967. First, 
Ladefoged was unable to determine the formants of many back 
vowels, so his set of formants was not complete. Second, Ladefoged 
did not attempt any quantitative measures of the distances between 
cardinal vowels. Another problem with Ladefoged's study may have 
been how he scaled the cardinal vowel formants. The perception of 
pitch does not have a linear relationship to frequency measured in 
hertz (Stevens & Volkmann 1940) and researchers commonly 
transform formant values in an attempt to plot them in some 
perceptual space. One of the more commonly used scales is the mel 
scale of subjective pitch where a 1000-Hz tone, 40 dB above 
threshold, has a pitch of 1000 mels (Stevens & Volkmann 1940). The 
mel scale was derived from subjective pitch evaluations of pure 
tones. Because speech is a complex signal and does not have the same 
(or even similar) pitch characteristics as pure tones, the trans- 
formation of formant frequencies from hertz to mels may be a 
questionable practice. 

Other frequency scales, derived from critical bands and 
frequency difference limens, do represent the non-linear nature of 
the human auditory system (Munson & Gardner 1950). The Bark 
scale (Zwicker 1961), Miller's (1989) Auditory-Perceptual space, and 
the semitone (semit) scale are all scales which represent frequency 
in a non-linear fashion. These scales were used in the present study 
to perceptually scale cardinal vowel formants. 

The aim of the present study is to rigorously and quantitatively 
resolve the issue of auditory equidistance using linear predictive 
analysis of cardinal vowel formants. Digital analysis is not without its 
problems, such as varying results depending on the analysis 
parameters, but the results presented include at least one set of 
formant values for each cardinal vowel. Calculated measures of the 
distance between vowels were statistically analyzed for both linear 
and transformed formant values. 



Morrison: A re -examination of cardinal vowels and auditory equidistance 121 

2. Recordings of the cardinal vowels 

Each of four, independently acquired, tape-recorded sets of the 
primary cardinal vowels was digitized and analyzed using linear 
predictive coding. All four sets of vowels had been spoken by Daniel 
Jones. Three of the sets were recorded by Peter Ladefoged in 1956 as 
part of his dissertation (PL sets 3, 6, and 7) (see Ladefoged 1960a, 
1967 for recording methodology). The fourth set is from a cassette 
copy of a recording of the cardinal vowels made by Daniel Jones. 2 

3. Digital analysis 

The cassettes were played on a stereo cassette deck (Sony TC- 
WR950) and bandpass filtered between 100 and 7000 Hz using a 
'Brickwair filter (Wavetek Rockland model 751 A). The amplitudes of 
the PL3 and PL6 sets were amplified 10 dB to equalize the 
amplitudes for all the vowel sets. The cardinal vowels were then 
digitized (Data Translation DT2823 A/D-D/A board) at 20,000 
samples per second with a resolution of 16 bits. 

Interactive Laboratory System (PC-ILS) software (Signal 
Technology Inc. 1987a) was used for all speech analysis. The 
temporal centerpoint of each vowel was found, and the frame on 
each side was analyzed using the ANA command. Points 100 msec on 
both sides of the midpoint were also analyzed as a consistency check 
of analysis results. Variations in formants for all vowels were 
generally less than five percent of the centerpoint formant 
frequency. The RSO (root solving) command was then used to find 
the formant peak values (Signal Technology Inc. 1987b). 

The results of the digital analysis varied depending on the 
analysis parameters used in PC-ILS. Most vowels were best analyzed 
with an 18th order filter, preemphasis of 98% (to compensate for the 
frequency roll off found in speech signals), a Hamming window, 
analysis frame size of 200-250 points (samples), and the number of 
peaks set to eight (8). The above parameters produced reasonable 
formant values for most vowels.-^ The high back vowel CV8 was 
analyzed with a higher order filter in an attempt to differentiate 
between the first two formants. 

Because the digital analysis sometimes produced unexpected 
results, the following post hoc rules were used to make the analysis 
results more consistent. These rules have no real theoretical basis, 
but are similar in spirit to the methods used in the art of 
spectrograph reading. 

1. If the calculated bandwidth was equal to or greater than 
the value of the resonance peak, or the amplitude was 
very low, then that peak was discarded. 

2. Formants with very wide bandwidths were kept only if 
the other analysis frame for that vowel token had a peak 
close in value with a narrower bandwidth. 



122 



Studies in the Linguistic Sciences 23:1 (Spring. 1993) 



3. A few foimant peaks were listed when the value occurred 
in only one analysis frame if peaks with similar values 
were found in other samples from the same vowel. 

4. When an unexpected formant value was found in an 
analysis frame, it was dropped if a similar value did not 
also appear in at least half of the other frames for that 
token. 

4. Formant values 

The obtained formant values for the primary cardinal vowels 
are listed in Table 1. PL indicates that the vowels are from the tape 
of cardinal vowels recorded by Ladefoged (using his numbering 
system). LL avg. are the average formants extracted from the 
duplicate vowel tokens on the tape of Daniel Jones supplied by 
Lisker. 



Formant freq.(Hz): 

Set CVl CV2 CV3 CV4 CV5 CV6 CV7 CVS 



Fl 



F2 



F3 



LL avg. 
PL set 3 
PL set 6 
PL set 7 

LL avg. 
PL set 3 
PL set 6 
PL set 7 

LL avg. 
PL set 3 
PL set 6 
PL set 7 

Table 1. 



233 
295 
296 
304 

2541 
2108 
2492 
2086 

3566 
3681 
3 808 
3357 



371 
379 
427 
3 46 

2142 
2166 

2324 
2225 

2454 
3779 
4109 
35 13 



567 
476 
587 
615 

1837 
1866 
1980 
2025 

2220 
3989 
4126 
3383 



889 
966 
726 
983 

1575 
1696 
1640 
2002 

1880 
3827 
2107 
3390 



840 
616 
666 
634 

954 
813 
93 6 

888 

2396 
3450 
2310 
3304 



501 
476 
464 
461 

893 
785 
709 
704 

2389 
3660 
2703 
3322 



342 
340 
402 

327 



216 

177 
7 

170 



668 410 

629 362 

986 ? 

5 9 2 3 5 5 

2256 2396 
2924 2399 
2982 ? 
3371 ? 



Formant frequencies for the primary Cardinal 
Vowels (in Hertz) 



The first formants exhibit the expected inverse relationship with 
vowel height (Ladefoged 1982). The high vowels, CVl and CVS, have 
low Fl's and the low vowels, CV4 and CV5, have high first formants. 
The inverse relationship between vowel height and first formant 
frequency is displayed in Figure 2. Figure 2 is a traditional, inverse 
plot of the first two formants of the type first presented by Joos 
(1948). This linear frequency plot has the advantage of allowing a 
more direct visual comparison of vowel position in the theoretical 
vowel trapezium and of obtained vowel formants (Hockett 1955). For 
example, in the traditional vowel space plot, the high, front vowel 
CVl appears in the upper left corner as it does in the trapezium 
(Figure 1). The other vowels also appear in their correct positions 
relative to each others' position in the trapezium. 



Morrison: A re-examination of cardinal vowels and auditory equidistance 



123 



Y 



N 

<D 

C 

o 

•♦-' 

c 

(13 

E 

i_ 

o 

LL 



500 



1000 



1500 




I I I 



3000 



2500 



O — — O LL Average Formants 

^ ^ PL3 Fornianis 

D Q PL6 Formants 

^ ^ PL7 Formaiils 

-" — I — " — I— ' — I— J — I — I — I — I — I — I I I 1 I I I I I I 

2000 1500 1000 500 



Formant Two (Hz) 



Figure 2. Values of the first two formants of four sets 
of Jones's cardinal vowels plotted in the traditional inverse 

F2/F1 style 

The high front vowels, CVl and CV2, show the most F2 variation. 
Possibly this is related to the greater F1/F2 separation found in the 
high front vowels. Also, F2 for CVl is quite close to the F2 for CV2 in 
the PL sets. The obtained second formants for CVl are lower than 
expected for PL sets 3 and 7 (Ladefoged 1960a, 1967). The obtained 
F2 for CVS is not quite in its expected location (around 815 Hz, Lisker 
1989). These discrepancies may be due to degradation of the record- 
ing between the time of recording and the time of the experiment. 
The second formants show the expected lowering of frequency as 
Jones moves around the vowel quadrangle (see Figure 2). 

Third formant values do not appear very consistent. Extraction 
of higher vocal tract resonances is more difficult than for lower 
resonances due to the intensity roll-off found in the speech spec- 
trum. The preemphasis of the signals prior to analysis attempted to 
counter some of this high frequency loss, but may not have been 
effective in this case. Some other causes of variation of the higher 
formants might be the age or quality of the recordings or the fact 
that the signals were low-pass filtered by the microphones used for 
recording the original vowel sets. 

5. Inter-vowel distances 

In the F1/F2 plot of the vowel sets (Figure 2) the spatial 
distances between the cardinal vowels do not appear equidistant. 
But, because the issue of equidistance is difficult to resolve solely on 
the basis of visual inspection, a quantitative method was used to test 
for equidistance. Euclidean distance measures were computed for 



124 Studies in the Linguistic Sciences 23:1 (Spring, 1993) 

adjacent vowel pairs based on the values of the first two formants 
for all four vowels sets (see Table 2). (Note that Jones did not specify 
the intervals between CV4-CV5 and CV1-CV8 as equidistant with the 
remaining pairs). The final inter-vowel interval, between CV7 and 
CVS for PL set 6, could not be computed because no formants were 
extracted for CVS of PL set 6. In order to maintain the maximum 
number of degrees of freedom, the final interval distance for PL set 6 
was set at the mean interval size of the other three vowel sets. A 
repeated measures analysis of variance of the distance measure- 
ments showed a significant difference in the distances between the 
different pairs of cardinal vowels ( F(5,15) = 4.17, p<0.02). The 
conclusion is that cardinal vowels may not be equidistant in a linear 
frequency (Hz) space. 

Inter-vowel interval 

Scale Set CVl-2 CV2-3 CV3-4 CV5-6 CV6-7 CV7-8 

Hertz LL avg. 422.19 362.55 415.12 344.45 275.50 2S7.12 

PL set 3 102. OS 315.29 518.65 142.77 206.96 312. S2 

PL set 6 213.04 379.39 367.32 303.86 283.85 (294.74)^ 

PL set 7 145.21 335.20 368.72 276.96 174.64 284.12 



Bark 


LL avg. 


1.68 


1.93 


2.60 


2.72 


2.20 


2.50 




PL set 


30.80 


1.30 


3.82 


1.20 


1.73 


2.30 




PL set 


61.28 


1.72 


1.67 


2.36 


2.04 


(5.20)a 




PL set 


70.57 


2.43 


2.69 


1.99 


1.52 


2.57 


Semits 


LL avg. 


8.58 


7.81 


8.23 


9.02 


8.30 


11.61 




PL set 3 


4.36 


4.71 


12.36 


4.51 


6.98 


14.81 




PL set 6 


6.46 


6.17 


4.92 


7.90 


6.22 


(13.60)a 




PL set 7 


2.50 


10.09 


8.12 


6.83 


6.66 


14.38 


Audito 


ry LL avg 


. 0.81 


0.72 


0.75 


0.69 


0.46 


0.72 




PL set 3 


0.56 


0.49 


1.07 


0.35 


0.36 


0.75 




PL set 6 


0.58 


0.60 


0.66 


0.57 


0.55 


(I.08)b 




PL set 7 


0.15 


0.88 


0.67 


0.41 


0.43 


(0.69)c 



^ Average interval distance of the three other sets. 

" Computed from formant averages of the other sets, Fl = 188 Hz, 

F2=376 Hz. F3=2398 Hz. 
'' Computed using formant average of the other sets for F3=2398 Hz. 

Table 2. Inter-vowel distances for the primary cardinal 
vowels using linear and transformed scales 

Ladefoged (1960a, 1967) reported that many phoneticians 
considered the intervals between the first four cardinal vowels to be 
greater than the intervals between the last four vowels and pre- 
sented formant charts of 31 sets of cardinal vowels to support this 
idea. To examine this idea quantitatively, the inter-vowels distances 
for the front and back vowels were analyzed separately. The ANOVA 
of the first three intervals (front vowel series) did not produce 



Morrison: A re-examination of cardinal vowels and auditory equidistance 125 

significant results, (F(2,6) = 4.31, p<0.07). An analysis of the last 
three intervals, (back vowel series) also failed to produce statistically 
significant results, (F(2,6) = 1.18, p<0.38). These results support the 
idea that the front and back cardinal vowels may be independently 
equidistant, that is, with different distances within each series. 

Because frequency and perceived pitch are not linearly related, 
various attempts were made to perceptually scale the cardinal vowel 
formants to see if a scaling could be found where all the cardinal 
vowels were equidistant. The first scale used was the critical band or 
Bark scale (Zwicker 1961). Zwicker & Terhardt (1980) provided a 
mathematical approximation to the empirically derived Bark scale: 

(1) Bark = 13 arctan (0.76 f) + 3.5 arctan (f/7.5), 
where f is the frequency in kilohertz. 

Formant frequencies for all four vowel sets were converted to 
Barks using Equation (1). As was done for the linear inter-vowel 
distance computations, the final interval distance for PL set 6 was set 
at the mean interval size of the other three vowel sets. An analysis of 
variance revealed a significant difference in the size of the interval 
between adjacent cardinal vowels (F(5, 15) = 3.16, p<0.04). The front 
vowel intervals alone were marginally different from each other 
(F(2,6) = 4.77, p<0.06), although the back vowel intervals did not 
significantly differ from each other (F(2,6) = 3.12, p<0.12). 

Another attempt at perceptually scaling cardinal vowels to find 
equidistance used the semitone (semits) scaling. Semits are units in 
the Equal Temperment musical scale and were commonly used by 
Speech Scientists for traditional inverse formant plots (Hockett 
1955). Semits are defined in Equation (2): 

(2) semits = 121og2 (f / 16.35) 

where f is frequency measured in hertz and 16.35 is the reference 
frequency corresponding to CO in the Usually Equally Tempered Scale 
(where an octave is divided into 12 equal intervals) based on A4 = 
440 Hz (American National Standards Institute 1960). Formants for 
all four vowel sets were converted into semits using Equation 2, and 
Euclidean distances were computed for each adjacent vowel pair. The 
final interval distance for PL set 6 was set at the mean distance of 
the other three vowel sets. The results of a repeated measures anal- 
ysis of variance revealed that the inter-vowel distances using a 
semitone scaling were significantly different from each other, 
(F(5,15) = 6.18, p<0.005). In contrast to the Bark scaling, the front 
vowel series distances were not significantly different, (F(2,6) = 0.93, 
p<0.50), while the back vowel series was different, (F(2,6) = 18.66, 
p<0.005). 

Miller (1987, 1989) proposed that speech sounds are mapped 
onto phonetically relevant target zones in an auditory-perceptual 
space (APS) using log frequency ratios of FO, Fl, F2 and F3. Scaling in 
APS is an attempt to deal with vowel normalization of inter- and 



126 Studies in the Linguistic Sciences 23:1 (Spring, 1993) 

intra-speaker variations in an auditory-perceptual space. Speech is 
first processed at a sensory level. Miller's (1987, 1989) scalings (see 
Equations (3)-(6) below) were based on the speech sounds of 
American English, but the APS framework has been successfully 
extended to German and Greek vowels (Jongman, Fourakis, & Sereno 
1989). Since the APS framework appears to be generally applicable 
to vowel perception, it was used to examine the equidistance of 
cardinal vowels. Scaling was done with the following formulas. 

(3) SR = 168 (GM FO / 168)1/3 

(4) X = log (SF3 / SF2) 

(5) y = log (SFl / SR) 

(6) z = log (SF2 / SFl) 

SR is the sensory reference computed using the geometric mean of 
the fundamental frequency (FO in HZ) of the utterance (here the 
geometric mean of FO for all the vowel sets was used, GMFOall = 
151). SF1-SF3 are the frequency locations of the first three signif- 
icant spectral prominences of the short term spectral envelope of the 
vowel waveform (usually equivalent to the first three formants). 
Miller's theory posits several levels of processing with almost 
identical scaling equations (the only difference is that the sensory 
reference and formants are transformed into perceptual reference 
and formants). The sensory level equations were arbitrarily chosen 
for use in this study. 

Auditory-perceptual scaling of the four cardinal vowel set 
formants was done (using Equations (3)-(6)), and Euclidean inter- 
vowel distances were computed in APS (see Table 2). For compu- 
tational purposes the first three formants of CV8 in PL set 6 were set 
at the average of the first three formants of the other three vowel 
sets, Fl = 188 Hz, F2=376 Hz, and F3=2398 Hz. The third formant of 
CVS in PL set 7 was set to 2398 Hz. A repeated measures analysis of 
variance indicated that there was a significant difference in the size 
of the interval between the cardinal vowels (F(5,15) = 3.00, p<0.05). 
As found for the semit scaling, the front vowel interval did not 
significantly vary (F(2,6) = 1.60, p<0.30), but the back vowel 
intervals were significantly different (F(2,6) = 12.37, p<0.01) 

6. Discussion 

The cardinal vowels analyzed in this study were not equidistant 
in linear or non-linear (Bark, semit, or APS) vowel space. The 
separate analyses of the front and back vowel series did not produce 
consistent results. Both series might be equidistant using linear or 
Bark scaling. (However the front series could be considered 
marginally different using either scale. The respective p-values of 
0.07 and 0.06 could be considered weak support for the idea that the 
front vowels series is not equidistant.) The other log-based scales, 
semitone and APS, indicated that the front vowel series might be 
equidistant and the back vowel series not equidistant. While it is 



Morrison: A re-examination of cardinal vowels and auditory equidistance 127 

possible to conclusively prove that cardinal vowels are not equidis- 
tant using standard analysis of variance, negative results are more 
difficult to interpret. Negative results can only be interpreted as 
possible support for the equidistance of cardinal vowels. 

Neary (1989) and Miller (1989) both suggest that log-based 
scaling is more appropriate for perceptually normalizing vowels, so 
the Bark, semit, and auditory-perceptual scaling results probably 
better represent reality. The distance between cardinal vowels was 
not equidistant for any of these log-based scales, but conclusions 
based on the partitioned vowel series are equivocal. 

The formant center frequencies obtained in this study may not 
be correct values due to a number of factors such as poor recordings, 
problems with digital analysis, or incorrect scaling from the physical 
to perceptual domain. If new, better recordings of cardinal vowels 
could be obtained, the analyses described here could be repeated and 
maybe the results would indicate that cardinal vowels are equidis- 
tant. The best way resolve the issue of auditory equidistance would 
be to do a perceptual or psychophysical study of the perceived 
distance between cardinal vowels using trained subjects. 

Cardinal vowels can still be useful even if they are not percep- 
tually equidistant. Cardinal vowels remain valuable as common 
reference points for the description and classification of vowels, 
especially when it is not feasible to do acoustic analyses. In addition, 
they are appropriate as training tools for students of phonetics 
(Abercrombie 1985). Cardinal vowels are also suitable when a 
standard vowel set is needed for a study of some aspect of vowels. 
Cardinal vowels have been used as the stimuli in a study of rounding 
(Lisker 1989) and for testing a two-formant model of vowel 
perception (Bladon & Fant 1978). 

NOTES 

* This research was submitted in partial fulfillment of the 
requirements for a Ph.D. in the Department of Speech and Hearing 
Science, University of Illinois-Champaign. Preparation of this article 
was supported in part by Department of Health and Human Services, 
Public Health Services Grant DC 00174-08 to Robert C. Bilger. I would 
like to thank my advisor, Robert C. Bilger, for his support and 
understanding, C.-W. Kim for his helpful comments on this project 
and manuscript, and Cynthia Johnson for comments on the manu- 
script. In addition, I thank Peter Ladefoged and Leigh Lisker for 
supplying me with the recordings of cardinal vowels used in this 
study. Address correspondence to Sharon R. Morrison, Department of 
Speech and Hearing Science, 901 South Sixth Street, Champaign, IL 
61820. 



128 Studies in the Linguistic Sciences 23:1 (Spring, 1993) 

^ Note, however, that the test used to analyze variability, 
Bartlett's test, is not a robust test (Neter, Wasserman, & Kutner 
1985), so Laver's conclusion may be suspect. 

2 This recording was supplied by Leigh Lisker who reported 
(Lisker 1989) only that the set was copied onto tape from a 
commercially available record. (Possibly Linguaphone ENG 254/255, 
1955). 

3 Reasonable in this case means close to the formants found in 
other analyses of cardinal vowels (Ladefoged 1960a, 1967; and 
Lisker 1989) or for similar sounding English vowels (Peterson & 
Barney 1952). 

REFERENCES 

AbercrOMBIE, David. 1967. Elements of general phonetics. Chicago: 

Aldine Publishing. 
. 1985. Daniel Jones's teaching. Phonetic linguistics, ed. by V. 

Fromkin, 15-24. Orlando: Academic Press. 
American National Standards Institute. 1960. Acoustical Termi- 
nology, S 1.1. 
BLADON, R. a. W., & Gunnar Fant. 1978. A two formant model and the 

cardinal vowels. Speech Transmission Laboratory Quarterly 

Progress and Status Report 1978:1.1-8. 
HOCKETT, Charles F. 1955. A manual of phonology. Baltimore, MD: 

Waverly Press. 
Jones, Daniel. 1956. An outline of English phonetics (8th ed.). New 

York: E. P. Dutton. 
Jones, Stephen. 1929. Radiography and pronunciation. British Journal 

of Radiology 2.149-150. 
JONGMAN, Allard, Marios Fourakis, & Joan A. Sereno. 1989. The 

acoustic vowel space of Modern Greek and German. Language 

and Speech 32:3.221-248. 
JOOS, M. 1948. Acoustic Phonetics. Baltimore, MD: Waverly Press. 
Ladefoged, Peter. 1960a. The nature of vowel quality. Revista do 

Laboratorio de Fonetica Experimental da Faculdade de Letras da 

Universidade de Coimbra V. 
. 1960b. The value of phonetic statements. Language 36.387- 

396. 
. 1967. Three areas of experimental phonetics. London: Oxford 

University Press. 
. 1982. A course in phonetics (2nd ed.). New York: Harcourt 

Brace Jovanovich. 
Layer, J. D. M. H. 1965. Variability in vowel perception. Language 

and Speech 8.95-121. 
Lisker, Leigh. 1989. On the interpretation of vowel 'quality': The 

dimension of rounding. Journal of the International Phonetic 

Association 19:1.24-30. 



Morrison: A re-examination of cardinal vowels and auditory equidistance 129 

Miller, James D. 1987. The auditory-perceptual theory of phonetic 

recognition: A synopsis. Central Institute for the Deaf, MS. 
. 1989. Auditory-perceptual interpretation of the vowel. Journal 

of the Acoustical Society of America 85:5.2114-2134. 
MUSON, W. A.. & M. B. Gardner. 1950. Loudness patterns: a new 

approach. Journal of the Acoustical Society of America 22:2.177- 

190. 
Neary, Terrance M. 1989. Static, dynamic, and relational properties 

in vowel perception. Journal of the Acoustical Society of America 

85:5.2088-2113. 
Neter, John, William Wasserman, & Michael H. Kutner. 1985. Applied 

linear statistical models (2nd ed.). Homewood, IL: Richard Irwin. 
O'Connor, Joseph D. 1973. Phonetics. London: Pelican. 
Peterson, Gordon E., & Harold L. Barney 1952. Control methods used 

in a study of the identification of vowels. Journal of the 

Acoustical Society of America 50.678-684. 
SIGNAL Technology Inc. 1987a. PC-ILS v. 6 [Computer Program]. 

Goleta, CA: Signal Technology Inc. 
. 1987b. ILS command reference guide v. 6. Goleta, CA: Signal 

Technology Inc. 
Stevens, Stanley S., & John Volkmann. 1940. The relation of pitch to 

frequency: A revised scale. American Journal of Psychology 

53:3.329-353. 
ZWICKER, Eberhard. 1961. Subdivision of the audible frequency range 

in critical bands (Frequenzgruppen). Journal of the Acoustical 

Society of America 33:2.248. 
, & E. Terhardt. 1980. Analytical expressions for critical-band 

rate and critical bandwidth as a function of frequency. Journal of 

the Acoustical Society of America 68:5.1523-1525. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



THE VEDIC CLAUSE-INITIAL STRING 
AND UNIVERSAL GRAMMAR* 

Steven Schaufele 

Examining a variety of clause-peripheral phenomena in 
Vedic Sanskrit, this paper argues that they are not amen- 
able to an analysis in terms of projections of a functional 
head COMP, but rather that such a head, for which evidence 
exists in other languages, is best regarded as a syntactic 
coalescence of a variety of syntactic and pragmatic func- 
tions which UG allows some languages to keep syntactically 
distinct. 

1. COMP-functions 

In various portions of the literature of generative grammar, 
especially those associated with the Principles-&-Parameters 
Approach (PPA) and its historical predecessors in the Standard- 
Theoretical School, the label COMP has been associated with linguistic 
elements performing a variety of functions, hereafter referred to as 
'COMP-Functions'. In PPA, these functions are actually associated 
with a variety of structural positions associated in different ways 
with the node-label COMP and its various projections, as will be 
discussed further in §3. In this paper I argue that in Vedic Sanskrit 
the syntactic realizations of the COMP-Functions must have been 
structurally different from those assumed in PPA literature for what 
some (e.g. Whorf, cited by Bruce Nevin, LINGUIST 3-7981) have 
called 'Standard Average European' languages to the point that the 
status of the node-label COMP as a universal functional head ought at 
the very least to be reconsidered. 2 As an alternative, I suggest that 
UG may allow a range of possibilities of syntactic realizations for 
these functions. 

For the purposes of discussion, I identify the COMP-Functions 
listed in (1). 

(1) a. Syntactic Linking 

b. Pronominal-Fronting 

c. Topicalization 

d. Left-Dislocation 

By 'Syntactic Linking', I mean the linking between a verb that 
subcategorizes for a clausal complement and that complement. Such 
linking is performed by the bold words in (2). 

(2) a. I wonder if Hilary will get here before 9:00. 

b. She told Terry that I was willing to sell my copy of 
'Stairway to Heaven'. 



132 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

c. I can't see whether Phil took the blue bell or the red 
hammer. 

'Pronominal-Fronting' refers to the movement to the beginning 
of a clause of various pronominals as in (3). In modern English such 
movement affects only interrogative and relative pronominals, most 
of which begin orthographically with the digraph 'wh'; hence the 
phenomenon is typically known in generative literature as 'wh- 
fronting', even in discussions of languages in which interrogative and 
relative pronominals take radically different forms. As will become 
clear in the next section, however, cross-linguistically what I am call- 
ing 'pronominal-fronting' can have much wider application than it 
does in English, therefore justifying the broader label. 

(3) a. Whom did Sam give the money to? 

b. Where did Morgan find the keys? 

c. Hilary showed it to the farmer whom you had seen in the 
drugstore yesterday. 

'Topicalization' refers to the fronting of non-pronominals to 
clause-initial position as in (4). Generally, there is some special prag- 
matic function associated with such fronting, though it may be 
difficult to define.^ 

(4) a. Peasi I like a;. 

b. [Mary Harper]i I have yet to see a, play Rosalind. 

c. [Mary Harper]i I have yet to see a more enchanting actress 
than aj. 

'Left-Dislocation' differs from Topicalization chiefly in that it 
involves a pronominal 'copy' of the fronted constituent, as in (5).'* 

(5) a. [That man],, I don't like him,. 

b. [That man]i, I don't like hisj dog. 

c. [RalphJi, I wish hcj would take his opinions elsewhere. 

In generative literature, as noted above, the notion of a func- 
tional head labelled COMP has been invoked in describing all of these 
phenomena, though not all at the same time. Early discussion within 
generative literature of such a notion (e.g., Rosenbaum 1967, Bresnan 
1972) was concerned primarily with Syntactic Linking. Chomsky 
1977 focussed on Wh-Fronting, with some discussion of Topical- 
ization and Left-Dislocation. Subsequent work has tended to focus on 
Wh-Fronting and Topicalization, while acknowledging Syntactic 
Linking as relevant. 

In this paper I show that in Vedic Sanskrit most of these func- 
tions were syntactically distinct, and that therefore the use of the 
notion of a structural node labelled COMP and/or its various projec- 
tions in describing that language may be misleading and therefore 
inappropriate.-'' 



Schaufele: The Vedic clause-initial string and universal grammar 133 

2. Vedic COMP-realizations 

In this section I discuss how the different COMP-Functions iden- 
tified above are realized in Vedic Sanskrit. The lexical and syntactic 
strategies discussed here will be referred to as 'COMP-Realizations'. 

Sanskrit, like many other South Asian languages (cf. Hock 1989), 
avoids syntactic subordination, and in fact as noted in Schaufele 
1990b:64-66 and references cited there, it has no subordination of 
finite clauses and, to the extent it has verbs subcategorizing for 
clausal complements, they are verbs used in reporting quoted dis- 
course and the various metaphorical extensions thereof, as discussed 
in Hock 1982b; cf. (6). In such constructions, the syntactic link 
between the quoted material (represented in (6) by the material be- 
tween quotes in the Sanskrit text) and the 'matrix' (represented in 
(6) by everything else) is effected by the word iti. 

(6) a. yah indraya 'sunavama' iti aha 

REL-N / Indra-D / press- Ip.impv. / Q / say-3s.pres. 
'Who says to Indra, "Let us press"' (RV 5. 37. Id) 

b. tam ahuh 'suprajah' iti 

DEM-Ams. / say-3p.pres. / rich in progeny-N / Q 
'They call him "rich in progeny".' (RV 9.114.1c) 

c. yat va pravrddha satpate 'na marai' 

REL I ox I great-V / sovereign-V / not /die-ls.subj. 

iti manyase 

Q / think-2s.pres. 

'Or when, great sovereign, you think "I shall not die'" 

(RV 8.93.5ab) 

d. 'bhijmih' iti tvam abhi-pra-manvate janah 'nirrtih' iti 
earth-Ns. / Q / 2s. pro. -A / think-3p.pres. / peopie-Np. / Nirrti-N / Q 
tva aham pari-veda sarvatah 

2s.cl.-A / Is. pro. -N / know-Is. pres. / completely 

'People think of you as "earth", (but) I know you completely 

as "Nirrti".' (AV 6.84. led) 

e. svayam u hi eva etat veda 'idam 

himself-N / PTCL / PTCL / PTCL / DEM-A / know-3s.pres. / DEM-N 

atah karma kartavyam iti 

now / deed-N / to be done-N / Q 

'For he himself knows that this deed is now to be 

done.' (SB 1.2.5.21) 

Most if not all Indo-European languages routinely front certain 
pronominals into clause-initial position, and Vedic was no exception. 
Vedic pronominal-fronting differs from English wh-fronting in three 
obvious ways (cf. Schaufele (1988) for further discussion). 

Whereas in English only 'wh-elements', i.e., interrogatives and 
relatives, are fronted, (cf. (7)), in Vedic demonstratives were affected 
in exactly the same way; cf. (8). 



134 Studies in the Linguistic Sciences 23: 1 (Spring 1 993) 

(7) a. [Which book]i did Lynn buy ai? 

b. The book that; Lynn bought aj has been missing for 3 days. 

c. ?*Did Lynn buy which book? 

d. Lynn bought that book. 

e. ?[That book]i Lynn bought ai. 

f. *[The bookJi Lynn bought aj that has been missing for 3 
days. 

(8) a. te§ui eva enamj etat ai pari-asTnesu 

DEM-Lp. / PTCL / 3f.s.cl-A / DEM / a / siuing around-Lp. 
anagnam aj karoti 
non-naked-A / a / make-3s.pres. 

'He makes her (to be) non-naked among those sitting 
around." (SB L3.3.8) 
b. etami u eva esahj etasmaik aj visnuh 

DEM-Af. / PTCL / PTCL / DEM-Nm. / DEM-Dm. / a / Vi§nu-N 

yajiiah ak ai vikrantim vi-kramati 

sacrifice-Nm. / a / a / victorious siep-Af. / stride-3s.pres. 

'This sacrifice, Visnu, strides this victorious step for him.' 

(SB LL2.13) 

Whereas in English only one wh-element can be fronted (cf. (9)), 
in Vedic an indefinite number can be fronted; cf. (8b). In this respect, 
Vedic resembles other languages which have been discussed by e.g. 
Rudin (1988) and McDaniel (1989); cf. (10). 

(9) a. Where did who give what? 
b. * Where who what did give? 

(10) a. Kg kas mislinol so o Demiri cuminja? (Romani) 

who / whom / think-3s. / that / Demiri-N / kissed 
'Who thinks that Demir kissed whom?' 

b. Kto etc kogda skazal? (Russian) 
who / what / when / said 

'Who said what to whom?' 

c. Koj kakvo na kogo e dal? (Bulgarian) 
who / what / to / whom / has / given 

'Who gave what to whom?' 

d. Kto koitiu jak^ by napisal ksiazkq? (Polish) 
who / to whom / what kind / would / write / book 
'Who would write what kind of book to whom?' 

e. Cine cui ce ziceai ca i a promis? (Rumanian) 

who / whom-0 / what / said-2s. / that / him-0 / has / promised 
'Who did you say promised what to whom?' 

However, whereas in English any clause including one wh- 
element is unacceptable unless that wh-element is either fronted or 
highly emphasized (cf. (11)), the fronting of pronominals is not 
obligatory in Vedic; cf. (12). It is even possible to have in the same 
clause some pronominals fronted into clause-initial position while 
others remain in their DS positions; cf. (13). 



Schiiufele: The Vedic clause-initial string and universal grammar 135 

(11) a. *This man here is who? 

b. Who is this man here? 

c. This man here is who? 

(12) a. aheh yataram kam apasya indra 

serpent-G / avenger-A / INT-A / see-2s.impf. / Indra-V 
'Which avenger of the serpent did you see, O Indra? = 
Whom did you see, O Indra, as avenger of the serpent?' 
(RV 1.32.14a) 

b. hari indrasya ni-cikaya kab svit 

horses-Ad. / Indra-G / perceive-3s.perf. / INT-N / PTCL 

'Who perceived the two steeds of Indra?' (RV 10.1 14. 9d) 

c. te devah etat yajuh apasyan 

dDEM-N / gods-N / dDEM-A / Yajus-A / see-3p.inipf. 
'The gods saw this Yajus.' (TS 5.2.3.1) 

d. evam eva etat aksaram etabhih devatabhih 
thus / PTCL / dDEM-N / syllable-N / dDEM-I / deities-I 
sam-srjyate 

unite-3s.pres.pass. 

'Thus this syllable is united with these deities.' (JUB 1.1.3.8) 

(13) etam eva asmin etat santatam 

DEM-Am. / PTCL / 3s.cl.-L / DEM-Adv. / extended-Am. 

avyavachinnam dadhati etat anuvacanam 

not cut off-Am. / grant-3s.pres. / dDEM-Nn. / recitation-Nn. 

'So this recitation grants him (the state of being) extended 

(in lifespan), not cut off (by death).' (SB 1.3.5.13) 

Whereas in English what is fronted is an entire NP or PP con- 
taining a wh-element (cf. (14)), in Vedic only the pronominal word 
itself is fronted, leaving behind the NP for which it serves as deter- 
miner; cf. (15). 

(14) a. Whomi did you give ai the books? 

b. [Which books]i did you give your sister 8i? 

c. [How many papers on Celtic verb-fronting]i did you receive 

ai? 

d. [From how many of your students]; did you get papers a,? 

(15) a. etati ha vai devah [a\ vratain]Np 

dDEM-As. / PTCL / PTCL / gods-Np. / a / vow-As. 

caranti 

undertake-3p.pres. 

'The gods undertake this vow.' (SB 1.1.1.5) 

b. yabhyahj eva tanij [aj devatabhyahj^p [aj havTrhsi]Np 
REL-Dp. / PTCL / dDEM-Np. / a / deities-Dp. / a / oblations-Np. 
grhltani bhavanti 

take-pass. pari. -Np. / be-3p.prcs. 

'For which deities those oblations are taken' (AB 7.2.3) 

c. kati, ayanij adya [aj udgataj^p [asmi'n yajfie]ts[p 

INT-Ap. / DEM-Ns. / today / a / Udgatr-Ns. / DEM-Ls. / sacrifice-Ls. 



136 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

[ai stotriyah]Np stosyati 

a / stotriya-Ap. / sing-3s.fut. 

'How many stotriyas will the Udgatr sing in this sacrifice 

today?' (SB 14.4.6.12) 

An additional difference between English and Vedic, well known 
to Indo-Europeanists, is that clitic pronouns in Vedic also routinely 
get fronted into the clause-initial string in conformity with 
Wackernagel's Law (of. (16)), ^ while in English (cf. (17)) they cliticize 
obligatorily to the elements that govern them for purposes of 6- and 
Case-assignment. 

(16) a. aj vam; vayah asvasah 

Pfx. / 2d.cl.-A / winged-Np. / horses-Np. 
vahisthah [abhi prayah nasatya 

finest draught animals-Np. / to-Adp(A) / offering-A / Nasatyas-V 
[aj-vahantu = ajlylvp 
a / bring3p.impv. / a 

'May the winged horses, the very finest draught animals, 
bring you to the offering, O Nasatyas.' (RV 6.63.7ab) 

b. pra vahj sardhaya ghfsvaye tvesadyumnaya susmine 
Pfx. / 2p.cl.-G / host-D / vigorous-D / fierce-D / wild-D 
[devattam aj brahmaJfMp gayata 

inspired-A / a / hymn-A / sing-2p.impv. 

'Sing your inspired hymn to the vigorous, fierce, wild 

host.' (RV 1.37.4) 

c. vrtram hi asmaij etat jaghniise 

Vrtra-A / PTCL / 3s.cl.-D / DEM-An. / slay-perf.part.-D 

apyayanam [akurvan = ajjy 

envigorating draught-An. / make-3p.impf. / a 

'For him, (the one) having slain Vrtra, they prepared this 

envigorating draught.' (SB 1.6.4.12) 

d. te enamj etat vratam upayantam 

DEM-Np. / 3s.cl.-A / DEM-As. / vow-A / undertaking-A 

[viduh = ajJv 

know-3p.pres. / a 

'They know him (to be) the one undertaking this vow.' 

(SB 1.1.1.7) 

(17) a. Tomorrow mornin' Claude's gonna sell='eni to 

y'r=neighbors. 

b. *Tomorrow mornin' 'em Claude's gonna sell to 
y'r=neighbors. 

c. *Tomorrow 'em mornin' Claude's gonna sell to 
y'r=neighbors. 

d. *Tomorrow mornin' y'r Claude's gonna sell='em to 
neighbors. 

e. *Tomorrow y'r mornin' Claude's gonna sell='em to 
neighbors. 

f. *Tomorrow mornin' y'r 'em Claude's gonna sell to 
neighbors. 



Schaufele: The Vedic clause-initial string and universal grammar 1 37 

g. *Tomorrow y'r 'em mornin' Claude's gonna sell to 
neighbors. 

Topicalization is attested in Vedic as in probably all other human 
languages.^ But whereas in English, topicalization fronts whole NPs 
but not single words out of NPs (cf. (18)), in Vedic while the fronting 
of NPs is possible (cf. (19)) the fronting of single words out of 
phrases is preferred;^ cf. (20) and see Hock 1982a and Schaufele 
1985, 1986, 1991, and 1993a for further discussion. '^ 

(18) a. [The neighbors' dog]i I haven't heard from ai yet. 

b. *Dogi I haven't heard from the neighbors' aj yet. 

c. *Neighbors'i I haven't heard from the ai dog yet. 

(19) a. [aiigirasah suvargam lokam yatah]i 

A ngirases-Ap. / heavenly-As. / world-As. / go-pres.pari.-Ap. 
purodasah kurmah bhutva ai anu pra-asarpat 
cake-N / tortoise-N / become-ger. / a / af(er-Adp(A) / crawl-3s.impf. 
'Becoming a tortoise, the cake crawled after the Angirases 
(who were) going to heaven.' (TS 5.2.8.4) 
b. [rajfia somenaji tat vayam ai asmasu dharayamasi 

king-I / Sonia-I / DEM-A / Ip.pro.-N / a / Ip.pro-L / guard- Ip.pres. 
'We guard it for ourselves by means of King Soma.' (KS 29.2) 

(20) a. dvisantam, ha asya tat [[a, bhratrvyam]N P 

hateful-A / PTCL / 3s.cl.-G / DEM-N / a / foe-A 

abhi-ati-ricyate]vp 

remain for-3s.pres. 

'That remains for his hateful foe.' (SB 1.9.1.18) 

b. osadhlnami vai sah [[ai mulani]NP upa-amlocat]vp 
planls-G / PTCL / DEM-N / a / roots-A / hide among-3s.impf. 
'He hid among the roots of plants.' (SB 1.2.5.10) 

c. ai ha vai asmin svah ca nistyah ca 

Pfx. / PTCL / PTCL / 3s.cl.-L / own-Np. / & / su-angers-Np / & 

ai-sariisate 

a / place trust-3p.pres.mid. 

'In him both his own (relatives) and strangers place their 

trust.' (SB 1.6.4.17) 

d. manahi ha vai devah [[manusyasya ajJNP a-jananti]vp 
mind-A / PTCL / PTCL / gods-N / man-G / a / know-3p.pres. 
'The gods know the mind of man' (SB 1.1.1.7) 

e. brahmai ca vai idam agre 

Brahman-N / PTCL / PTCL / here / beginning-L 

[ai subrahma ca]NP astam 

a / Subrahman-N/PTCL / be-2d.impL 

'In the beginning, there were here Brahman and 

Subrahman.' (SB 1.1.1) 

f. rsibhyahi ca eva enam etat [aj devebhyah ca]NP 
Rsis-Dp. / PTCL / PTCL / 3s.cl.-A / DEM / a / gods-D / PTCL 



138 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

ni-vedayati 

introduce-3s.pres. 

'He introduces him to the Rsis and to the gods.' (SB 1.4.2.3) 

And finally, Left-Dislocation is attested in Vedic; cf. (21) and see 
Oertel 1926 for a detailed discussion. 

(21) a. [datvah ca ha sautemanasah mitravit ca 

Datva-N / & / PTCL / Sautemanasa-N / Mitravil-N / & 
dariistradyumnahlNP [tau ha pratidarsasya 
Dam§tradyumna-N / DEM-Nd. / PTCL / Pratidars'a-G 
vaibhavatasya svaiknasya rajfiah brahmacarinau asatuh]s 
Vaibhavata-G / Svaikna-G / king-G / pupil-Nd. / be-3d.perf. 
'Datva Sautemanasa and Mitravit Darhstradyumna — these 
were King Pratidarsa Vaibhavata Svaikna's pupils.' (JB 2.274) 

b. [atha etat sarTram]NP [tasmin na rasah asti]s 

NEXUS / DEM-Nn. / body-Nn. / DEM-Ln. / not / sap-N / be-3s.pres. 
'Now this body — there is no sap in it.' (SB 4.4.5.1) 

c. [yajnaya yajamanaya atmane]NP [tebhyah eva asisam 
sacrifice-Ds. / sacrificer-Ds. / self-Ds. / DEM-Dp. / PTCL / blessing-A 
a-saste]s 

invoke-3s.pres. 

'For the sacrifice, for the sacrificer, for himself — for these 

he invokes a blessing.' (TB 2.6.9.3) 

3. The syntax of COMP-realizations 

3.1. Standard PPA assumptions 

In PPA literature (cf. e.g. Chomsky 1986:3) a constituent COMP is 
typically assumed. Furthermore, this constituent is defined as a 
Bar-0 head of a complete projectionj_ CP, with its own complement 
and specifier. In line with standard X-theory (cf. Chomsky, ibid.), 
both complement and specifier are assumed to be themselves max- 
imal projections. COMP° itself is the site of Syntactic-Linkers, while 
its complement is the clause (IP) 'introduced' thereby. Cf. (22). 

(22) a. [ip I [vp wonder [cp[C0MP if] [iP Hilary will get here before 

9:00]]]]? 

b. [ip She [vp told Teny [cp[cOMP that] [ip I was willing to sell 
my copy of 'Stairway to Heaven']]]]. 

c. [ip I can't [vp see [cp[C0MP whether] [ip Phil took the blue bell 
or the red hanmier]]]]. 

Spec of COMP is assumed to be the landing-site of wh-movement 
as in (23). The fact that a Spec position is a maximal projection would 
account for the fact that in English and other Standard Average 
European languages pronominal-fronting affects whole phrases as in 

(24). 

(23) a. [cp[Spec Whomi] [cOMP didj] [ip Sam aj [yp give the money to 

ai]]]? 
b. [cp[Spec Wherei] [cOMP didj] [ip Morgan aj [vp find the keys 

9i]]]? 



Schaufele: The Vedic clause-initial string and universal grammar 139 

(24) a. [cp[Spec[Which booksJi] [coMP didj] [ip you aj [vp give your 
sister a,]]]? 

b. [CP[Spec[How many papers on Celtic verb-fronting]i] [coMP 
didj] [ip you aj [vp receive ai]]]? 

c. [cp[Spec[Frorri how many of your studentsJi] [coMP didj] [ip you 
aj [vp get papers ai]]]? 

The landing-site of Topicalization and Left-Dislocation may also 
be Spec of COMP.io Or it may be an adjunction-site to IP or CP. Note 
that in PPA (cf. Chomsky 1986:88) adjunction is typically assumed to 
be of maximal projections to maximal projections. Adjunction of X°- 
heads to heads is also possible; cf. Baker (1988). But the consensus is 
that an X^^^x c^n only accept another X"^^" in adjunction, and like- 
wise an X° another X°. This is consistent with the Structure- 
Preservation Constraint (cf. Emonds 1976). Thus, if Topicalization 
and/or Left-Dislocation are examples of adjunction then they involve 
maximal projections. The same conclusion, of course, follows from the 
definition of Topicalization and/or Left-Dislocation as movement to 
Spec of COMP. 

An alternative hypothesis, proposed in Chomsky 1977 and 
mentioned here for the sake of completeness and because it will be 
at least tangentially relevant to later discussion, would define an 
extra TOPIC position as aunt to (i.e., sister to the mother of) COMP. Cf. 
(25).!' On the face of it, this hypothesis would appear to define 
TOPIC as the Bar-0 head of a separate phrasal projection, with its 
own specifier and complement, which latter would presumably be 
CP. 

(25) S=TP? 



TOPIC S = CP 



COMP S = IP 



3.2. The Vedic situation 



It will be noticed that the theory outlined in the previous section 
posits a relatively large number of maximal projections and/or 
landing-sites for maximal projections and a relatively small number 
of Bar-0 nodes. COMP itself is a Bar-0 node and a second Bar-0 node 
may exist in TOPIC, but each of these has its own Specifier position, 
presumably X^^^x ^nd the possibility of adjunction creates more 
landing sites that, according to the assumption already mentioned, 
are restricted to maximal projections. 

On the other hand, the Vedic COMP-Realizations discussed in §2 
are mostly Bar-0 in nature. While phrasal topicalization is attested 
and was therefore presumably possible, the corpus as a whole and 



140 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

especially the prose portions thereof show a strong if not over- 
whelming preference for X°-fronting. And while clitic pronouns are 
syntactically NPs in their own right, the demonstratives, inter- 
rogatives, and relatives whose fronting is Sanskrit's equivalent of 
'wh-fronting' are merely NP-determiners, not phrases themselves (cf. 
Schaufele 1988 for further discussion). And, as mentioned earlier (n. 
7) and discussed in Schaufele 1991, 1993a, to assert that in spite of 
appearances all these fronted words are really bare-headed phrases 
is to drain the theory of falsifiability and therefore of interest. 
Ideally, what would be desired is a number of Bar-0 landing sites 
(which might themselves serve as adjunction sites) without accom- 
panying specifier and complement positions. Cf. further discussion in 
§5. 

The question of the location of these landing-sites vis-a-vis IP 
raises further complications. It will be noted that iti, which is 
apparently the prime candidate for the realization of the function of 
Syntactic Linking, is generally in Vedic a clause-final element. '^ jt 
therefore differs from English COMP in this respect. Generative 
linguists studying Japanese, which like Sanskrit is an obviously head- 
final language, have claimed that COMP in that language is post- 
clausal rather than pre-clausal as it is in English. However, as may be 
obvious from the examples above and will be discussed below, all the 
other 'COMP-Realizations' in Vedic Sanskrit are clause-initial phe- 
nomena. It has been suggested (Davison, Yoon p. c.) that iti might 
indeed represent a clause-final COMP, with a pre-clausal Spec posi- 
tion serving the other 'COMP-Realizations'; cf. the structure outlined 
in (26). This would be consistent with common assumptions about 
the internal structure of head-final, left-branching constituents and 
in fact with the structure assumed in Schaufele 1990b for NPs and 
other projections of lexical categories in Vedic, except that the 
analysis presented there seeks to avoid any intermediate level of 
structure between the Bar-0 head and the maximal projection 
thereof. I discuss some of the possible inadequacies of the analysis in 
(26) later, but I will assume it for the immediate discussion.' 3 

(26) 




Q = COMP? 



Schiiufele: The Vedic clause-initial string and universal grammar 141 

The other COMP-Realizations listed in §2 need to be classified as 
CLAUSE-INITIAL vs. PRE-CLAUSAL. Clause-initial elements are treated 
by the syntax as part of the clause but happen to surface at its left 
periphery; pre-clausal elements are treated by the syntax as outside 
the clause altogether. 

The principal diagnostic in Vedic for the clause-boundary, in 
terms of which clause-initiality and pre-clausality should of course 
be defined, is the location of 'sentential particles'. These elements, 
labelled 'PTCL' in all the examples in this paper, are words with 
pragmatic rather than semantic content. i"* Syntactically they rou- 
tinely follow immediately the first phonological word (cf. Schaufele 
1986, 1990b for discussion of various complications in the definition 
thereof) in whatever larger syntactic constituent defines their 
'pragmatic scope', most typically the clause. '^ Ergo, the (phonological) 
word immediately before any set of sentential particles is, by 
definition, the first word in the clause; anything before it must 
therefore be outside the clause. 

On the basis of this criterion, the Vedic realization of Left- 
Dislocation must be viewed as pre-clausal. In (21a), repeated below 
as (27), the two occurrences of the sentential particle ha must belong 
to distinct 'clauses'; the word immediately preceding the second ha, 
the resumptive pronoun tau, must therefore be the first word of its 
clause and all the material to its left must be extra-clausal. • "5 

(27) [datvah ca ha sautemanasah mitravit ca 
Datva-N / & / PTCL / Sautemanasa -N / Mitravit-N / & 
darhstradyumnah]NP [tau ha pratidarsasya 
Dam§tradyumna-N / DEM-Nd. / PTCL / Pratidars'a-G 
vaibhavatasya svaiknasya rajnah brahmacarinau asatuh]s 
Vaibhavata-G / Svaikna-G / king-G / pupil-Nd. / be-3d.perf. 
'Datva Sautemanasa and Mitravit Damstradyumna — these 
were King Pratidars'a Vaibhavata Svaikna's pupils.' (JB 2.274) 

Topicalization and pronominal-fronting in Vedic, on the other 
hand, are clearly associated with a clause-initial landing site.^'^ As 
can be seen from (20), the topicalized word is routinely followed 
immediately by the sentential-particle string. As for pronominal 
fronting, the examples repeated below as (28) show that the particle 
string is likely to follow immediately a fronted pronominal (28a); if 
there is more than one fronted pronominal, then the particle string 
will follow the first one (28b). 

(28) a. etatj ha vai devah [a, vratam]Np 

dDEM-As. / PTCL / PTCL / gods-Np. / a / vow-As. 

caranti 

undertake-3p.pres. 

'The gods undertake this vow.' (SB 1.1.1.5) 



142 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

b. etami u eva esahj etasmaik aj visnuh 

DEM-Af. / PTCL / PTCL / DEM-Nm. / DEM-Dm. / a / Vi^nu-N 

yajiiah a^ aj vikrantim vi-kramati 

sacrifice-Nm. / a / a / victorious step-Af. / slride-3s.pres. 

'This sacrifice, Visnu, strides this victorious step for him.' 

(SB 1.1.2.13) 

4. Discourse Linking 

A brief digression is in order to discuss an extra function which 
has not been among the foci of interest in the study of Standard 
Average European languages with regard to COMP, but which Vedic 
syntax invites into this examination. This is 'Discourse Linking', by 
which I mean the linking together (often by means of a special class 
of adverbs) of sentences which are syntactically and semantically 
complete in themselves but which are pragmatically linked in a 
discourse context such that said linking is a necessary aspect of their 
relevance and cooperativeness. Cf. the bold words in (29). 

(29) Greg had been wanting for weeks to buy a new pair of 
shoes. But Irene had told him that the local shoe- 
store would be having a sale next month. So Greg 
chose to wait a while longer. However, unbeknownst 
to both Irene and Greg, circumstances w^re com- 
pelling the store owner to change his plans and to 
delay the anticipated sale. Meanwhile, on the other 
side of town a new shoe store was opening, with a 
major sale of new stock. Fortunately, Greg's friend 
Hilary, who occasionally had business to do in that 
area, found out about this and told Greg. So Greg was 
able to get new shoes in spite of the delay at the local 
shoe store. 

Syntactically and semantically, every sentence in (29) could 
stand by itself without the bold words. 

Discourse Linking in Vedic is done by means of such words as 
dtha, tdtha, and tcitah. Cf. (30). These words (which in this role I shall 
label 'nexus', following Minard 1936 and Klein 1987) can be related 
to pronominal stems. Some nexus words are more obviously derived 
from pronominal stems. As Delbriick (1888:§140) notes, the nomina- 
tive masculine and neuter singular demonstratives sdh and tat can 
serve as nexus (though he doesn't use that label); cf. (31). The 
ablative singular tdsmdt can be similarly used; cf. (32). By means of 
such words, extensive discussions can be held together; cf. (33). 

(30) a. sah ha uvaca videghah mathavah ... atha 

DEM-N / PTCL / speak-3s.perf. / Videgha / Mathava-N / ... / NEXUS 

ha uvaca gotamah rahuganah ... 

PTCL / speak-3s.perf. / Gotama / Rahugana-N 

'Videgha Mathava said ... Then Gotama Rahugana 

said ...' (SB 1.4.1.17-18) 



Schiiufele: The Vedic clause-initial string and universal grammar 143 

b. tatha u eva esah etena vajrena ajyena 

NEXUS / PTCL / DEM-N / DEM-I / thunderbolt-I / butter-I 

rtun samvatsaram pra-jayati 

seasons-A / year-A / win-3s.pres. 

'Likewise with that thunderbolt (which is) the butter he 

wins the seasons, the year.' (SB 1.5.3.4) 

c. tatah dvabhyam brahmanah yajfie caranti 
NEXUS / two-I / Brahmin-Np. / sacrifice-L / use-3p.pres. 
'Therefore, Brahmins use two in the sacrifice.' (SB 1.2.4.2) 

(31) a. sah yatra ha evam rtvijah samvidanah yajhena 

NEXUS / REL / PTCL / thus / priests-Np. / agreeing-Np. / sacrifice-I 
caranti 

perform-3p.pres. 

'And when the priests in complete agreement thus perform 
the sacrifice ... ' (SB 1.5.2.15) 
b. tat svena eva enam etat payasa devah 

NEXUS / own-I / PTCL / 3s.cl.-A / thus / juice-I / gods-N 

svi-akurvata 

acquire-3p.inipf.mid. 

'So the gods acquired it by means of its own juice.' 

(SB 1.5.3.5) 

(32) tasmat gardabhah dpi analese ati 

NEXUS / donkey-N / even / poor grazing-L / beyond-Adp(A) 

anyan pas'un medyati 

other-A / animals-A / get fat-3s.pres. 

'Therefore the donkey, even in conditions of poor grazing, 

gets fat beyond other animals.' (TS 5.1.5.5) 

(33) sah [yah kapalani upa-dadhati], [sah 

NEXUS / REL-N / potsherds-A / put down-3s.pres. / DEM-N / 

upavesam a-datte] ['dhrstih asi'] iti. sah 

Upavesa-A / pick up-3s.pres. / bold-N / be-2s.pres. / Q / NEXUS 

[yat enena agnim dhrsti iva apa-carati], [tena 

REL / 3s.ci.-I / fire-A / boldly / PTCL / attack-3s.pres. / DEM-I 

atha dhrstih]. [yat enena yajfie upa-a-labhate], [upa 

NEXUS / bold-N / REL / 3s.cl.-I / sacrifice-L / touch-3s.pres. / Pfx. 

eva vai etena etat vevesti], tasmat [upavesah 

PTCL / DEM-I / DEM-A / atlend-3s.pres. / NEXUS / Upavesa-N 

nama]. 

by name 

'Now he who puts down the potsherds picks up the Upavesa 

(shovel) (saying) 'You are bold'. Now because with it he 

attacks the fire boldly (dhrsti), as it were, that's why (it is 

called) 'Dhrsti'. And because with it he touches (the coals) in 

the sacrifice, he attends (upa-vis) it (the fire) with it (the 

shovel), that's why it (the shovel) is called 'Upavesa'. 

(SB 1.2.1.3) 



14 4 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

To the extent that constituents like the bold words in (29) have 
been discussed at all in the Standard-Theoretical literature, it tends 
to be assumed that they are merely clause-introductory adverbials. 
As noted above, note 4, the quotative marker iii which has been 
identified as Vedic's only realization of the Syntactic-Linking 
function also serves a similar function on occasion, as in (34). 

(34) iti braviti vaktari raranah vasoh vasutva 

Q / say-3s.pres. / speaker-N / generous-N / good-G / goodness-I 

karavah anehah 

singers-N / guiltless-Np. 

'(Thus) says the generous speaker, "Through the goodness 

of the good, the singers are guiltless.".' (RV 10.61.12bc) 

However, such collocations do not excuse us from including iti in 
a discussion of Vedic COMP-Realizations. For one thing, as noted 
above they are very rare compared with those exemplified in (6) in 
which iti is used as a link between quoted material and the matrix. 
Furthermore, there is another usage of iti which is in my opinion 
more essentially syntactic. As noted by Hock (1982b:55-57), at least 
by the time of the Atharva-Veda, i.e., mid-late Vedic period, /'// was 
used as a means of assimilating onomatopoetic elements and other 
non-Sanskrit lexemes into the syntax of a Vedic clause without 
having to bother with fitting them out with Sanskritic morphological 
trappings as was the common strategy in the RV (cf. Hock 1982b:56). 
Cf. (35). 

(35) a. prthivyam te nipecanam bahih te astu 

ground-L / 2s.cL-G / outpouring-N / outside / 2s.cl.-G / be-3s.impv. 
'bar Iti 
splash / Q 

'May your outpouring be on the ground, outside of you, 
"splash".' (AV 1.3.1-9 refrain) 

b. ajena krnvantah sitam vrsena uksantu 

goat-I / making-Np. / cool-A / rain-I / sprinkle-?p.impv. 

'bal' I'ti 

splash / Q 

'Making (you) cool with the goat, let them sprinkle (you) 

with rain, "splash".' (AV 18.2.22) 

c. 'bhiik' I'ti abhi-gatah 'sal' iti apa-krantah 'phal' 
bounce / Q / come-Ns.part. / whist / Q / go-Ns.part. / bang 
Iti abhi-sthitah 

Q / tread-Ns.part. 

'"Bounce" it has come; "whist" it has gone; "bang" it has 

trodden.' (AV 20.135.1) 

In these clauses iti is performing a purely grammatical function, 
in a sense 'naturalizing' elements that are foreign to the grammar. 
Therefore, there can be no doubt that, at least in cases like these, its 
presence and position must receive a grammatical account as a 
functional category. 



Schaufele: The Vedic clause-initial string and universal grammar 145 

Returning to the question of Discourse Linkers, it is not enough 
to characterize thenn as clause-initial adverbials. In English, different 
Discourse Linkers are treated differently by the grammar. Witness 
the freedom of ordering exhibited by the discourse-linking pronom- 
inal therefore that is not enjoyed by the conjunction hut in (36). 
Presumably, this difference relates to the syntactic difference be- 
tween conjunctions, which typically and cross-linguistically are 
severally restricted in position, and other 'adverbials'. Even vocative 
adjuncts cannot precede clause-conjunctions in English; cf. (37). 
Clearly, returning to the distinction introduced earlier, clause- 
conjunctions in English must be pre-clausal. 

(36) a. i. Therefore, there's no way of being certain. 

ii. But there's no way of being certain. 

b. i. There's therefore no way of being certain, 
ii. *There's but no way of being certain. 

c. i. There's no way therefore of being certain, 
ii. *There's no way but of being certain. 

d. i. There's no way of being certain therefore, 
ii. *There's no way of being certain but. 

(37) a. But there's no way of ever being certain, one way or 

another. Commander. 

b. But there's no way of ever being certain, Commander, one 
way or another. 

c. But there's no way. Commander, of ever being certain, one 
way or another. 

d. But, Commander, there's no way of ever being certain, one 
way or another. 

e. ^Commander, but there's no way of ever being certain, one 
way or another. 

In Vedic, nexus words always occur at the left periphery of their 
associated clauses. But are they clause-initial or pre-clausal? In (38) 
we see examples of nexus words followed by other words which are 
in turn followed by sentential particles. So evidently in cases like 
these it is not the nexus word, the discourse-linking element, that is 
the first word in the clause but the word immediately following it. 

(38) a. atha kim u yah devesu anasnatsu 

NEXUS / INT / PTCL / PTCL / REL-N / gods-L / cating-L 
purvah as'niyat 
before-Adp(L) / eal-3s.opl. 

'But what if he should eat before the gods eat?' (SB 1.1.1.8) 
b. atho manasa vai prajapatih yajfiam 

NEXUS / mind-I / PTCL / Prajapati-N / sacrifice-A 

atanuta 

perform-3s.impf. 

'Then Prajapati performed the sacrifice with his mind.' 

(TS 1.6.8.4) 



146 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

c. sah yatha iva ha tat agneh bhavati 

NEXUS / REL-Adv. / PTCL / PTCL / DEM-N / fire-G / be-3s.pres. 

'In that case, that (feud) is (= would be) (on the part) of the 

fire.' (SB 1.1.1.21) 

However, it should be noted that there is some variation with 
regard to discourse linkers. It is possible for 'nexus' words to be fol- 
lowed immediately by sentential particles. Cf. (39). ^^ This variation is 
doubtless due to the fact that these elements, like their equivalents 
in other Indo-European languages (such as 'therefore' in (36)), are 
etymologically pronouns. In some cases they may refer directly to 
(possibly empty) constituents in the clauses they introduce, in which 
case they may be treated syntactically as real members of those 
clauses, while in others they may be serving solely as discourse 
linkers, which if they refer to anything at all are referring to the 
previous discourse context in general. Similar variation obtains in 
English; cf. (40). In (40a), parallel to constructions like (41), the word 
'then' is a pronominal referring to a specific time that has (presum- 
ably) been identified in the preceding discourse and which is 
functioning in this sentence as an adverbial but is overtly represent- 
ed solely by this pronominal; being a pronominal, in this sentence 
'then' is immediately followed by the verb, which has been fronted 
ahead of the subject in the manner typical of questions. (40b), on the 
other hand, is not asking 'will you listen to me at a particular time' 
but 'you are going to listen to me, aren't you? That's what I deduce 
from your previous remarks'. In which case the word 'then' if it is 
referential at all refers to 'the previous remarks', i.e. the entire 
discourse context. In this case, the verb is not fronted ahead of the 
subject, and 'then' is acting as a mere discourse linker. 

(39) a. uta sma asya panayanti janah jutim 

NEXUS / PTCL / 3s.cl.-G / praise-3p.pres. / people-N / zeal-A 
'And the people praise his zeal.' (RV 4.38.9ab) 
b. atha ha somah uvaca 

NEXUS / PTCL /Soma -N / say-3s.perf. 
'Then Soma said' (SB 1.6.3.21) 

(40) a. Then will you listen to me? 
b. Then you will listen to me? 

(41) a. When will you listen to me? 

b. Will you listen to me this time? 

5. On the structural organization of Vedic COMP-realizations 

To summarize, Vedic COMP-Realizations divide into three 
syntactic groups in terms of constituent order. The quotative marker 
iti, the only realization of syntactic linking, is a clause-final element. 
The 'landing-site' of left-dislocation is clearly pre-clausal; the 'nexus' 
words that serve as discourse linkers are variously pre-clausal or 
clause-initial, presumably due to their etymological status as 



Schaufele: The Vedic clause-initial string and universal grammar 147 

pronominals. The landing-sites of topicalization and pronominal fron- 
ting, on the other hand, are clearly clause-initial. 

It is in relation to this last fact that Hale's (1987) attempt to 
shoehorn Vedic Sanskrit into the general assumptions of PPA are 
most clearly at fault. Hale first fails to recognize that in Vedic 'wh- 
fronting' is not restricted to relative and interrogative pronominals 
as it is in Standard Average European languages, but that demon- 
stratives based on the stem ta- are subject to the same effects. He 
further assumes without question that the landing-site of 'wh- 
fronting' in Vedic as in Standard Average European languages is 
COMP.'^ He then defines the landing-site of topicalization as a 
distinct structural position TOPIC which he assumes is to the left of 
COMP, consistently with the above-mentioned position advocated in 
Chomsky 1977. This assumption is also consistent with the fact that, 
in the Vedic corpus, whenever a clause shows both pronominal 
fronting and topicalization, the topicalized word USUALLY precedes 
any and all fronted pronominals. This is, however, not always true. 
Especially if the topicalized word is a verb of speech, as discussed in 
Hock 1982a, it is not unheard of for it to follow a fronted pronominal, 
even though it precedes all non-fronted elements and is clearly 
included in the clause-initial string. Cf. (42). 

(42) etati ha vai uvacaj sahkah kausyah a[ 

DEM-A / PTCL / PTCL / say-3s.perf. / Sanka-N / Kau§ya-N / a 

putram aj 

son-A / a 

'On this (subject), Sahka Kausya said to his son ...' (KS 22.6) 

Hale argues that his analysis is supported by the facts of clitic- 
pronominal ordering. He claims that clitic pronouns, when they are 
fronted into the clause-initial string, cliticize specifically to COMP, 
and that his analysis, wherein wh-fronting goes to COMP while 
topicalization lands in a TOPIC position to the left of COMP, is 
supported by the fact that in the RV clitic pronouns never come 
between the topicalized word and a fronted 'wh-word', i.e. strings 
like (43b) are unattested. 

(43) a. brahma kah vah saparyati 

priest-N / INT-N / 2p.cl.-A / honor-3s.pres. 
'Which priest honors you?' (RV 8.7.20c) 
b. *brahma vah kah saparyati 

priest-N / 2p.cl.-A / INT-N / honor-3s.prcs. 

As far as I know, this claim is true of the 'Family Books' of the 
RV {mandaldh 2-7), but it is not true of either the younger portions 
of the RV (cf. (44)) nor of Vedic Prose (cf. (45); presumably. Hale 
overlooked examples such as these because they involve demonstra- 
tives rather than relative or interrogative pronominals. 20) In (44a) a 
(topicalized?) subject is followed by a relative pronominal which in 
turn is followed, not preceded as Hale's theory would predict, by a 
sentential particle. In (44b) a pronominal clitic comes between a 



148 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

clause-initial (topicalized?) subject and a relative pronominal in third 
position. In (45a-b), etdt is not associated with any nominal and must 
therefore be read as a sentential adverbial; it is therefore indu- 
bitably to be included in the clause-initial string, in which it is 
clearly preceded by the clitics asyai and enam respectively. In (45c), 
it is associated with the noun dpydyanam 'draught' from which it has 
been fronted into the clause-initial string, in which it is preceded by 
the clitic asmai. Such examples are enough to falsify Hale's analysis. 

(44) a. visnuh yat ha avat vr'sanam madacyiitam 

Vi§nu-N / REL / PTCL / assist-3s.impf. / bull-A / intoxicated-A 
'As Visnu assisted the intoxicated bull' (RV 1.85.7c) 
b. pusa nah yatha vedasam asat vrdhe 

Pu§an-N / Ip.cl.-D / REL / goods-G / be-3p.subj. / increase-D 

raksita payiih adabdhah svastaye 

protector-N / guardian-N / indeceivable-N / well-being-D 

'So that Pusan, the indeceivable protector and guardian 

for (our) well-being, may be for us increase of goods' 

(RV 1.89.5cd) 

(45) a. kruram iva vai asyai etat karoti 

cruel-A / PTCL / PTCL / 3s.f.cl.-D / DEM / do-3s.pres. 

'He does a cruel (thing), so to speak, to her.' (TS 5.1.5.1) 

b. tat svena eva enam etat payasa devah svi 

NEXUS / own-I / PTCL / 3s.cl.-A / DEM / juice-I / gods-N / own 

akurvata 

make -3p. imp f. mid. 

'So the gods made own = gained possession of it by means 

of its own juice.' (SB 1.5.3.5) 

c. vrtram hi asmai etatj jaghnuse s\ 

Vrtra-A / PTCL / 3s.cl.-D / DEM-An. / slay-pcrf.part.-D / 8 

apyayanam akurvan 

envigorating draught-An. / make-3p.impf. 

'For him, (the one) having slain Vrtra, they prepared this 

envigorating draught.' (SB 1.6.4.12) 

As Hock (1989) has made clear, the fact that clauses like (43b) 
and (45) are not attested in the RV is due not to a structural distinc- 
tion between the landing-sites of topicalization and 'wh-fronting' and 
a constraint on the landing-site of clitic-fronting but on a slight dis- 
tinction between Mantra Vedic and Vedic Prose in the details of the 
ordering of constituents within the clause-initial string. Whereas in 
Vedic Prose, constituents in the clause-initial string stack up accord- 
ing to the five-place template described in Hock (1982a) and 
outlined in (46a), with all accented fronted words except one going 
into the final position of the clause-initial string, in Mantra Vedic a 
variant template operates (46b), in which accented fronted pronom- 
inals that do not go into first position go into the third position which 
in Vedic Prose is reserved for accented particles. For details and 
further discussion cf. Schaufele 1990b:171-179, 1993b, and Hock 
1992:60-71. 



Schiiufele: The Vedic clausc-iniiial siring and universal grammar 1 49 

(46) a. W/Pro PTCL PTCL Pro Pro (Vedic Prose) 
b. W/Pro PTCL Pro/PTCL Pro (Mantra Vedic) 

Hale says nothing at all about Left-Dislocation, and although he 
notes that a 'clause-connector' (=nexus) will precede a fronted 
pronominal. Hale fails to acknowledge that it will also precede a 
topicalized word as in (47). In Schiiufele 1986 I argued from this cir- 
cumstance that, contrary to Hale's and Chomsky's assumption, TOPIC 
in Vedic must follow, not precede COMP, identifying COMP as the 
location for nexus. As I have been saying throughout this paper, this 
is an oversimplification. 

(47) a. atho manasa vai prajapatih yajnam 

NEXUS / mind-I / PTCL / Prajapali-N / sacrificc-A 
atanuta 

perform-3s.impf. 

'Then Prajapati performed the sacrifice with his mind.' 
(TS 1.6.8.4) 
b. tat svena eva enam etat payasa devah 

NEXUS / own-I / PTCL / 3s.cl.-A / thus / juicc-I / gods-N 

svf-akurvata 

acquirc-3p.impf.mid. 

'So the gods acquired it by means of its own juice.' 

(SB 1.5.3.5) 

As noted earlier, standard PPA, especially with the assumption 
that adjunction sites are only for maximal projections plus the 
'binary-branching' assumption according to which any non-terminal 
node has at most two daughters, seems to provide an excessive 
number of landing-sites for maximal-projections and/or an insuffi- 
cient number of Bar-0 landing-sites. What is needed for an adequate 
description of Vedic is: 

i. A Bar-0 landing-site to the left of the 'clause proper' 

(to be defined somehow; cf. below) for topicalized 

words and fronted pronominals; 
ii. A landing-site to the left of the 'clause-proper' for 

topicalized PHRASES as in (19); 
iii. A DISTINCT 'landing-site' to the left of the 'clause 

proper' for Left-Dislocation (cf. notes 8, 15); 
iv. A (presumably Bar-0) site for the generation of nexus 

words, which may be either clause-initial or pre- 

clausal; 
v. A site to the right of the clause proper for the 

generation of the quotative marker /'//. This site is 

likewise presumably Bar-0. 

It must be further noted that none of these elements can plau- 
sibly be said to be subcategorized for by anything or, in other words, 
to be part of any other constituent's thematic structure. It may be 
plausible to suppose that nexus words and the quotative marker are 
base-generated (although a case can be made for the insertion of iii. 



150 



Studies in the Linguistic Sciences 23:1 (Spring 1993) 



rather in the manner of a case-marker, by the grammar in cases like 
(35)), but in the absence of such words there may be no reason a 
priori to expect the base-generation of nodes corresponding to them, 
and in any case the landing-sites of topicalization and Left- 
Dislocation are clearly not mandated by 9-Theory.2i Yet Bar-0 nodes 
are manifestly necessary in order to describe these phenomena. So I 
will for the moment assume that such nodes are base-generated by 
the grammar of Vedic, probably as functional heads (cf. Gelderen 
1993 and references cited there), at least as an option. 2 2 

There are at least two options for the organization of these 
various nodes vis-a-vis each other and IP, represented in (48). (48a) 
represents a relatively flat structure, in which the various clause- 
peripheral sites are all sisters to each other and to IP. In this 
diagram, W is the landing-site for Left-Dislocation while X is the 
landing-site for phrasal topicalization and TOPIC for Bar-0 topicaliza- 
tion. (The relative ordering of W and NEXUS is unclear and I don't 
know of anything in the data that would clarify it.) E, as in some PPA 
literature, represents a supra-clausal 'expression'. In (48b), the nodes 
labelled 'NEXUS' and 'TOPIC have associated 'Specifier' nodes, repre- 
senting the landing-sites for Left-Dislocation and phrasal topicaliza- 
tion respectively. The status of the nodes Y and Z is problematical 
and will be discussed below. 



(48) a. 




W NEXUS 



X 



TOPIC 



IP 



Q(=COMP?) 



iti 




Spec NEXUS 




Spec TOPIC 

IP 



Q(=COMP?) 



iti 



Either hypothesis needs to account for pronominal fronting as 
well as topicalization. In Schiiufele 1990b I suggested that one single 
non-pronominal word could be fronted into the TOPIC° position, but 
that any number of pronominals could be adjoined to this position. It 
may be objected that, if I am going to posit Bar-0-adjunction as a 



Schaufele: The Vedic clause-initial string and universal grammar 151 

feature of Vedic grammar, why can't I assume such adjunction to, 
e.g., IP to account for topicalization instead of positing extra func- 
tional heads for which there is little or no evidence in Standard 
Average European languages? This suggestion runs into two prob- 
lems, however. One is that, as noted earlier, adjunction of X° con- 
stituents to X^iax nodes seems to violate the Structure-Preservation 
Constraint. 23 The other is that adjunction is supposed to be capable of 
unlimited iteration; if one constituent can be adjoined, why not two 
or more? But in Vedic unlimited fronting is characteristic peculiarly 
of pronominals, which is why I have suggested that specifically 
pronominal-fronting is adjunction to the node that serves as landing- 
site for lexical topicalization. This hypothesis, that (Bar-0) topical- 
ization involves movement to a Bar-0 node while pronominal- 
fronting involves adjunction to the same node, accounts for: 

i. The fact that, while an unlimited number of pronom- 
inals can be fronted, at most one non-pronominal can 
be; 

ii. The fact that both the topicalized word and the 
fronted pronominals are Bar-0 constituents; 

iii. The pragmatic parallel between topicalization and 
pronominal-fronting. 

Of the two hypotheses diagrammed in (48), (48b) has the advan- 
tage of providing a structural distinction between 'clause-initial' 
topicalization and pronominal-fronting on the one hand and 'pre- 
clausal' Left-Dislocation and nexus on the other. The facts of Vedic 
syntax require us to distinguish between the clause-initial string, the 
'clause proper' i.e., everything within the clause that is not included 
in the clause-initial string, and the aggregate of the two, which we 
might call the 'maximal clause'. In the diagram in (48b), IP is to be 
identified as the 'clause proper'; all of the 'clause-initial string' 
discussed in Hock 1982a is ultimately contained under TOPIC or its 
associated Specifier position, and the 'maximal clause' is to be 
identified with Y. The placement of sentential particles, in terms of 
which the clause-initial/pre-clausal distinction was defined, is to be 
defined in terms not of IP, the 'clause proper', but of Y, the 'maximal 
clause'. Of course, if there has been no topicalization or pronominal- 
fronting as in (49) then the TOPIC node will be empty and Y will 
reduce to Z or IP. Meanwhile, Left-Dislocation would place con- 
stituents to the left of Y, which would account for such constituents' 
being treated as outside Y (the 'maximal clause') for purposes of 
particle-placement. Nexus words likewise would be generated 
outside of Y, as required, though (being pronominal) they could also 
appear under TOPIC, like fronted pronominals, hence serving as hosts 
for particle-placement as in (39). (48a) does not account for this 
distinction between topicalization and Left-Dislocation, between 
clause-initial and pre-clausal sites, which is why I have been inclined 
to favour (48b). 



152 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

(49) [y[S[NP janakah ha vafdehah] [VP bahudaksinena 
Janaka-N / PTCL / Vaidcha-N / lavish-I 

yajfiena ije]]] 

sacrifice-I / worship-3s.perf. 

'Janaka Vaideha worshipped with = performed a lavish 

sacrifice' (SB 14.6.6.1) 

The weakness of the analysis in (48b) is that it suggests that iti, 
TOPIC, and NEXUS are 'heads' of Z, Y, and E respectively. This sugges- 
tion has in my opinion little if any justification. In English, COMP 
words like for in (50) have the power to assign Case and otherwise 
license the generation of structures and constituents; in other words, 
they behave like lexical heads. To the extent that iti 'licenses' the 
generation of constituents within larger constituents, it too behaves 
like a head, in which case the node labelled 'Z' in (48b) may actually 
correspond to CP or some such standard phrasal constituent in PPA. 
But note that the 'complement' of iti isn't necessarily a clause; it may 
be, as in (6d) or (35), a single word. This fact casts in doubt the 
precise status of Z. 

(50) We were waiting for him to bring the tapes. 

As for Y and E, there is no recognizable sense in which the nodes 
labelled TOPIC and NEXUS function as their respective heads. In 
other words, saying TOPIC is the 'head' of Y or that Y is the maximal 
projection of TOPIC is to say nothing at all; and likewise for the 
relationship between NEXUS and E. Furthermore, it is not clear from 
the data what the relationship is between TOPIC to the left of IP and 
iti to its right. (48b) implies an asymmetrical aunt-niece relationship 
(cf. also (25) and accompanying discussion), but they might as well 
be sisters, as in (48a). Standard PPA would strongly prefer the 
analysis in (48b) because it conforms better to the assumptions 
underlying the 'Universal Base Hypothesis', outlined in Travis 1984 
and Speas 1990, according to which all lexical or Bar-0 categories 
share the same projection structure. But standard PPA would 
necessarily read into this analysis implications about headship, etc. 
that are unwarranted. For the reason given above, 1 prefer the 
analysis in (48b) over that in (48a), but in doing so I warn 
stringently against the implications it would seem to bear within the 
context of X-Theory; an adaptation of the analysis in (48b) such as 
that in (51) might actually be preferable, as far as the data are 
concerned. This is why I have used the non-committal labels 'E', 'Y', 
and 'Z', and have striven to avoid the suggestion that NEXUS, TOPIC, 
or even iti is a 'head' in any grammatically meaningful sense. 

6. Conclusion 

In this paper I have argued that the functions which the 
Standard-Theoretical school has associated with the structural node- 
label COMP and its projections were in Vedic Sanskrit associated with 
a variety of structural positions which cannot plausibly be viewed as 



Schiiufele: The Vedic clause-initial string and universal grammar 153 

(51) E 




Spec NEXUS 




Spec TOPIC IP Q(=COMP?) 

Z^^ z^ I 

iti 

belonging to the projection of any one node. This is especially true 
given the necessity of two Bar-0 nodes at opposite ends of the clause, 
the location of the quotative-marker iti and the landing-site of 
topicalization and pronominal-fronting, each of which serves some of 
the COMP-Functions, and the implausibility of regarding_ some of the 
Bar-0 nodes discussed in this paper as 'heads' in the X-theoretical 
sense. 

For the reasons reported in this paper, I am inclined to reject the 
notion that UG defines a single Bar-0 node and its set of projections 
as the syntactic realization of the 'COMP-Functions' listed in (1), as is 
typical of the Standard Average European languages. Instead, I 
suggest that there is a (presumably universal) set of 'COMP- 
Functions', and that UG allows for a range of possible syntactic struc- 
tures realizing them. It seems to me likely that all of these options 
involve clause-peripheral positions, though even this hypothesis may 
prove too narrow. This is an empirical question. 

Of course, since the syntactic and pragmatic functions discussed 
in this paper are called 'COMP-Functions' because of their theoretical 
association with a specific node-label 'COMP', if we reject that node- 
label with its implications as a linguistic universal we should prob- 
ably find another name for the collective functions it tends to serve. 
However, if broad-based and thorough comparative and typological 
studies indicate that there is a strong cross-linguistic tendency for 
these functions to be 'gathered together' into the purview of a single 
Bar-0 node and its projections, then we would be justified not only in 
calling that node 'COMP' but in using that label for the set of syntactic 
and pragmatic functions associated with it as well, and languages like 
Sanskrit would serve syntactic theory primarily as exotic reminders 
that this association is not, as standard PPA assumes, inherent in UG. 



NOTES 

* An earlier version of this paper was presented at the 
Thirteenth South Asian Languages Roundtable, 26 May 1991, Urbana, 



154 



Studies in the Linguistic Sciences 23:1 (Spring 1993) 



IL. I would like to acknowledge the helpful comments of Rakesh 
Bhatt, Hans Henrich Hock, Gillian Ramchand, Nalini Rau-Murthy, and 
James Yoon, and to absolve them of any responsibility for this paper. 
In the examples, all inter-word sandhi has been undone, and accent 
is indicated if it is given in the primary source. 

The examples in this paper make use of the following abbre- 
viations: 

Grammatical abbreviations 

clitic pronoun 
dual 

feminine gender 
future tense 
gerundive 
imperfect tense 
imperative mood 
masculine gender 
middle voice 
neuter gender 
optative mood 
plural 
participle 
passive voice 
perfect tense 
present tense 
personal pronoun 
singular 
subjunctive mood 

TEXTUAL ABBREVIATIONS 

AB: Aitareya Brahman a 

AV: Atharva-Veda 

JB: Jaiminiya Brahmana 

JUB: Jaiminiya Upanisad-Brahmana 

KS: Kathaka Samhitii 

RV: Rg-Veda 

SB: Sadvimsati Brahmana 

SB: Satapatha Brahmana 

TB: Taittirlya Brahmana 

TS: Taittirlya Samhita 

^ Here, as elsewhere, references to LINGUIST list contributions 
are cited in this format. 

2 One reviewer challenged the claim that COMP is generally 
regarded in the PPA community as a linguistic universal. While 
works independent of mine (e.g. Bhatt & Yoon 1991) have challenged 
this assuinption, in practice the assumption is strong enough to be 
introduced immediately into any discussion of clause-peripheral 
phenomena in any language; cf., for the issues relevant to this paper. 



A 


accusative case 


cl. 


Ab. 


ablative case 


d. 


Adp. 


adposition 


f. 


Adp(X) 


adposition assigning 


fut. 




case X 


ger. 


Adv. 


adverb 


impf. 


D 


dative case 


impv. 


DEM 


demonstrative 


m. 


G 


genitive case 


mid. 


I 


instrumental case 


n. 


INT 


interrogative 


opt. 


L 


locative case 


P- 


N 


nominative case 


part. 





oblique case 


pass. 


Pfx. 


verbal prefix 


perf. 


PTCL 


particle 


pres. 


Q 


quotative marker 


pro. 


RFJ. 


relative pronominal 


s. 


V 


vocative case 


subj. 



Schiiufele: The Vedic clause-initial string and universal grammar 155 

especially Hale (1987, 1993). I am here recommending that, in any 
language outside Standard Average European, the existence of COMP 
must first be justified before it is used in any further syntactic 
discussion. 

3 That the fronting of non-pronominals, in this paper for con- 
venience labelled 'topicalization', can have a variety of communica- 
tive functions in addition to topicalization properly so called (for 
which (4b-c) may serve as good examples) can be seen briefly from 
the following examples. It may involve (contrastive) focus as in (i) = 
(4a); there are, of course, alternative strategies for expressing such 
focus; cf. (ii-iii). Such strategies include not only contrastive accent 
(in speech) or left-dislocation, represented in (ii-iii), but, at least in 
verb-final languages, focus-movement; for discussion cf. Schaufele 
1990a and references cited there. 'Topicalization' may also serve a 
stage-setting function as in (iv-v). It can also serve a 'backgrounding' 
function, deemphasizing a constituent or at least denying it an 
emphasis it might get in a more 'normal' position. Cf. (vi) vs. (vii). 

(i) Peas I like. 

(ii) I like peas. 

(iii) As for peas, I like them. 

(iv) In the morning, Terry left the package at the corner. 

(v) In the jungle, the mighty jungle, the lion sleeps tonight. 

(vi) With the claw (of the hammer), Brent pried out several 

nails, (most felicitous if the nails are the topic of 

subsequent narrative/discussion) 
(vii) Brent pried out several nails with the claw of the hammer. 

(most felicitous if the subsequent topic is either the 

hammer or the claw) 

Cf. Schaufele 1991, 1993a for a discussion of the variety of 
pragmatic consequences 'topicalization' seems to have had in Vedic. 

Of course, pronominal-fronting also entails certain pragmatic 
consequences, often parallel to those of 'topicalization'. As will be 
further discussed in this paper, this is one reason why I judge it 
desirable to define both kinds of fronting as having the same, or at 
least related, landing-site(s). 

'* 'Left-Dislocation' is known in more traditional grammars of 
Indo-European languages by the Greek name 'prolepsis'. 

-'' Cf. also Bhatt & Yoon 1991 for a more general discussion of the 
necessity of distributing COMP-Functions amongst distinct structural 
positions in a variety of languages, and Gelderen 1993 for the 
inappropriateness of certain functional categories, allowed and 
defined by UG, in the description of certain languages. 

6 In (16), 1 have represented the 'traces' of clitic pronouns that 
represent verbal arguments as being 'cliticized' to the verbs under 
the lexical node V, since if these pronouns had not been fronted they 
would be in these positions, forming phonological words with the 



156 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

verbs. Cf. Schiiufele 1993b and references cited there for further 
discussion. 

^ Vedic topicalization differs from pronominal-fronting most 
obviously in that while an unlimited number of pronouns can be 
fronted in Vedic (cf. above, (8b)), at most one non-pronominal can be 
topicalized. The ramifications of this distinction will be discussed 
below. 

8 As noted in Schaufele 1991, 1993a, the general Vedic 
preference for X°-fronting over the fronting of multi-word phrases is 
reflected to a great extent in text frequency; the overwhelming 
majority of fronted constituents in the entire corpus are single 
words. But it is also reflected in genre differences: In the metrical 
texts of the Mantra Vedic corpora, although the vast majority of 
fronted constituents are single words, the fronting of whole phrases 
is not uncommon. However, in the pedantic style of the Vedic Prose 
texts the (unambiguous) fronting of whole phrases is of vanishingly 
small frequency. 

9 As noted in Schaufele 1991, 1993a, Webelhuth & den Besten 
(1987) have suggested that some apparently Bar-0 topicalized con- 
stituents in, e.g., Germanic are actually X^^^ constituents. In those 
papers I noted that, given that in Vedic any (non-clitic) word can be 
topicalized (including verbal prefixes and members of conjuncts, cf. 
(20c, f)), the hypothesis proposed by Webelhuth & den Besten as a 
'safety valve' for the theoretical assumption that only maximal pro- 
jections can be topicalized would require that in Vedic every word be 
treated as a maximal projection. Given the evidence for hierarchical 
structure adduced in Schaufele 1990b, this hypothesis seems not 
only ad hoc but absurd. Cf. Schaufele 1991, 1993a for further 
discussion. 

10 Even if, as one reviewer pointed out, Left-Dislocation is 
assumed to be a base-generated structure (a very plausible assump- 
tion, even 'pre-/extratheoretically'), the material to the left of the 
clause must have a structural location. For convenience, I will in this 
paper refer to this location as a 'landing-site.' 

11 The assumption that TOPIC, if it exists, is to the left of COMP 
has not gone unchallenged in PPA. Baltin (1982:17-22) argues on the 
basis of sentences like the one below that TOPIC must be to the RIGHT 
of COMP. Huang (1982:152-53) has also argued that in Chinese TOPIC 
must be to the right of COMP. Cf. Lasnik & Saito (1992:cap. 4) for 
further discussion. 

It's obvious [[thatJcOMP [Maryi he can't stand aiJslS 

'2 Hock (1982b:45-47) notes that occasions in which iti precedes 
the quoted material, while they can be found in the RV, are 
'definitely ... in the minority compared to those in which iti follows' 
the quoted material. He tallies five occurrences of pre-quote iti in the 
RV, as opposed to 32 occurrences of post-quote /'//; however, the only 



Schaufele: The Vedic clause-initial string and universal grammar 157 

example he gives in his text is one in which the order is iti — matrix 
verb — matrix subject — quote; cf. below. Thus, it is not obvious that iti 
in this example bears any syntactic relation to the quoted material; 
as Hock himself notes, iti in this example could at least equally 
plausibly be glossed by the pronominal 'thus'. 

Iti bravTti vaktari raranah vasoh vasutva 

Q / say-3s.pres. / speaker-N / generous-N / good-G / goodness-I 

karavah / anehah 

singers-N / guiltless-Np. 

'(Thus) says the generous speaker, "Through the goodness of 

the good, the singers are guiltless.".' (RV 10.61.12bc) 

13 One reviewer suggested the following alternatives to the 
structure in (26): 

(i) Y [XP Topic [X PTCL IP]] 

in which Y is the landing-site for Left-Dislocation, and 
extra 'topic' positions can be generated by adjunction 
to IP. 
(ii) [TopP Spec [T Topic [CP Spec [C IP [C iti]]]]] 
Suggestion (i) misses the point that any 'topic' nodes in Vedic ought 
preferentially to be Bar-0, not Spec nodes or adjunction to other X^^ax 
nodes. Suggestion (ii) avoids this problem by defining a Bar-0 Topic 
node as a functional head. However, assuming that this node is the 
landing-site for topicalization and/or pronominal fronting, as sug- 
gested later in this paper, while its Spec position is the landing-site 
for Left-Dislocation, it fails to account for the fact that the Topic node 
must be clause-internal, as discussed immediately below, while its 
Spec node is extra-clausal. 

14 Cf. Klein 1978, Hock 1982a, and Schaufele 1985, 1986, 1988, 
and 1990b, and, for more general discussion of such elements, 
Kendall & Yoon 1986, Schourup 1983, Zwicky 1985, and Underhill 
1988. 

15 It should be noted that Vedic also has a small number of 
particles such as evd and iva that are phrase-bound rather than 
clause-bound; cf. especially the discussion in Schaufele 1985, 1990b 
on this subject. These particles are also glossed 'PTCL' in the ex- 
amples in this paper, but the distinction between them and the more 
important clausal particles should in any given case be either clear or 
irrelevant. 

16 The word immediately preceding the first ha, ca '&', itself 
belongs to the particle class and therefore the two together form a 
particle string ca ha identifying the preceding word datvah as the 
first word in the 'clause'. 

1"^ As is clear from this discussion, topicalization and Left- 
Dislocation are distinct in Vedic: whereas Left-Dislocation not only 
involves resumptive pronouns (cf. Oertel 1926) but is clearly pre- 
clausal, witness the evidence of distinct sentential-particle strings. 



158 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

topicalization is clearly clause-initial by the contrary evidence. 
Pragmatically, the two phenomena are probably in complementary 
distribution; I have not come across any sentences in the Vedic 
corpus involving both. Be it noted that, while topicalization is often 
used to highlight or emphasize a constituent, Vedic grammar had 
other means of accomplishing this end, such as use of the emphatic 
phrasal particle evd or focussing, as discussed in Schaufele 1990a. 

18 One reviewer asked if it is possible for particles ever to be 
clause-initial in Vedic. Generally, it is not; while not all Vedic 
discourse particles are clitics, they all behave like clitics in requiring 
preceding hosts; cf. Schaufele 1993b for further discussion. Some 
particles, e.g., nii and evd, have in the RV 'long-vowel' doublets (i.e., 
nu and evd) which act as full adverbial clause-connectors; cf. 
Schaufele 1986 for further discussion. But these are not relevant to 
this paper. 

19 Actually, of course, as noted earlier, modern PPA defines the 
landing-site of wh-movement as 'SPEC of COMP', rather than COMP°. 
This distinction is, however, irrelevant for Hale's discussion. 

20 Overwhelmingly in Vedic Prose relative and interrogative 
pronominals go into clause-initial position, so one rarely has occasion 
to find such a pronominal preceded by anything at all in a clause in 
Vedic Prose. It is not impossible, however; witness the following 
example: 

indrah ha yatra vrtraya vajram pra-jahara 

Indra-N / PTCL / REL / Vnra-D / thunderboll-A / throw-3s.pcrf. 

"When Indra had thrown (his) thunderbolt at Vitra' (SB 1.2.4.1) 

21 As noted in Schaufele 1991, 1993a, topicalization in Vedic, 
whatever its peculiariues, is an orthodox example of Move a in that 
it is movement to an A-position. 

22 I have no reason to suppose that such nodes have been 
generated in sentences in which they are not occupied 

23 Note that the Bar-0 adjunction posited in Schaufele 1990a to 
account for focus-movement in Vedic, as well as the 'incorporation' 
processes discussed in Baker 1988, all make use of Bar-0 hosts. 

REFERENCES 

Baker, Mark C. 1988. Incorporation: a theory of grammatical function 

changing. Chicago: University of Chicago Press. 
BalTIN, Mark R. 1982. A landing site theory of movement rules. 

Linguistic Inquiry 13.1-38. 
Bhatt, Rakesh, & James Yoon. 1991. On the composition of COMP and 

parameters of V2. Proceedings of the Tenth West Coast 

Conference on Formal Linguistics, ed. by Dawn Dates. 
BRESNAN, Joan. 1972. Theory of complementation in English syntax. 

MIT Ph. D. dissertation in Linguistics. 



Schaufele: The Vedic clause-initial string and universal grammar 159 

Chomsky, Noam. 1977. On w^-movement. Formal syntax, ed. by Peter 

W. Culicover, Thomas Wasow, & Adrian Akmajian, 71-132. New 

York: Academic Press. 

. 1986. Barriers. Cambridge, MA: MIT Press. 

DELBRUCK, Berthold. 1888. Altindische Syntax. (Syntaktische 

Forschungen, 5.) Halle: Buchhandlung des Waisenhauses. 
Emonds, Joseph E. 1976. A transformational approach to English 

syntax: root, structure-preserving, and local transformations. 

New York: Academic Press. 
Gelderen, Elly van. 1993. The rise of functional categories. 

Amsterdam: Benjamins. 
Hale, Mark Robert. 1987. Studies in the comparative syntax of the 

oldest Indo-Iranian languages. Harvard Ph. D. dissertation in 

Linguistics. 
. 1993. Deriving Wackernagel's Law: prosodic and syntactic 

factors determining clitic placement in the language of the 

Rigveda. Paper read at the 2nd-Position Clitic Workshop, The 

Ohio State University, Columbus, Ohio, 11 July 1993. To appear 

in the Proceedings. 
HOCK, Hans Henrich. 1982a. Clitic verbs in PIE or discourse-based 

verb fronting? Sanskrit sd hovdca gdrgyah and congeners in 

Avestan and Homeric Greek. Studies in the Linguistic Sciences 

12:2.1-38. 
. 1982b. The Sanskrit quotative: a historical and comparative 

study. Studies in the Linguistic Sciences 12:2.39-85. 
. 1989. Conjoined we stand: theoretical implications of Sanskrit 

relative structures. Studies in the Linguistic Sciences 19:1.93- 

126. 
(ed.). 1991. Studies in Sanskrit syntax. Delhi: Motilal 

Banarsidass. 
. 1992. What's a nice word like you doing in a place like this? 

Syntax vs. Phonological Form. Studies in the Linguistic Sciences 

22:1.39-87. 
Huang, Cheng-Teh James. 1982. Logical relations in Chinese and the 

theory of grammar. MIT Ph. D. dissertation in Linguistics. 
Kendall, Sue Ann, & James H. Yoon. 1986. Sentence particles as 

evidence for morphosyntactic interactions with pragmatics. 

Studies in the Linguistic Sciences 16:1.53-77. 
Klein, Jared S. 1978. The particle u in the Rigveda: a synchronic and 

diachronic study. Gottingen: Vandenhoeck & Ruprecht. 
1987. Syntactic and discourse correlates of verb-initial 

sentences in the Rigveda. Paper read at the Second Sanskrit 

Syntax Symposium, Ninth South Asian Languages Analysis 

Roundtable, Syracuse, NY, 7 June 1987. Hock 1991.123-143. 
Lasnik, Howard, & Mamoru Saito. 1992. Move a. (Current Studies in 

Linguistics, 22.) Cambridge, MA: MIT Press. 
McDaniel, Dana. 1989. Partial and multiple wh-movement. Natural 

Language and Linguistic Theory 7.565-604. 



160 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

MiNARD, Armand. 1936. La subordination dans la prose vedique. 
(Etudes sur le Satapatha-Brahmana, 1. Annales de I'Universite 
de Lyon, 2:3.) Paris: Belles Lettres. 

OERTEL, Hanns. 1926. The syntax of cases in the narrative and 
descriptive prose of the Brahmanas. 1. The disjunct use of cases. 
(Indogermanische Bibliothek. 1. Abt. Sammlung indoger- 
manischer Lehr- und Handbiicher. 1. Reihe. Grammatiken, 18.) 
Heidelberg: Carl Winter's Universitatsbuchhandlung. 

ROSENBAUM, Peter S. 1967. The grammar of English predicate 
complement constructions. Cambridge, MA: MIT Press. 

RUDIN, Catherine. 1988. On multiple question and multiple wh 
fronting. Natural Language and Linguistic Theory 6.445-501. 

SCHAUFELE, Steven. 1985. Fronting/topicalization and sentential 
particles in Vedic prose. Department of Linguistics, University 
of Illinois at Urbana-Champaign, MS. 

. 1986. Single-word topicalization in Vedic Prose: a challenge to 

Government & Binding? Paper read at the First Sanskrit Syntax 
Symposium, Eighth South Asian Languages Analysis Round- 
table, University of Illinois, Urbana, IL, 30 May 1986. Hock 
1991.153-175. 

. 1988. Where's my NP? Non-transformational analyses of Vedic 

pronominal fronting. Studies in the Linguistic Sciences 
18:2.129-162. 

1990a. A 'focus' position for subjects within the Vedic VP. 

Paper read at the Twelfth Annual South Asian Languages 
Analysis Roundtable, University of California, Berkeley, CA, 3 
June 1990. 

1990b. Free word-order syntax: the challenge from Vedic 

Sanskrit to contemporary formal syntactic theory. University of 
Illinois at Urbana-Champaign Ph.D. dissertation in Linguistics. 
Available from University Microfilms. 

. 1991. Lexical topicalization in Vedic Sanskrit and the Head 

Movement Constraint. Paper read at the 44th Kentucky Foreign 
Languages Conference, Lexington, KY, 26 April 1991. 

. 1993a. X°-fronting in Vedic Sanskrit and Binding Theory. Paper 

read at the 15th Annual South Asian Languages Analysis 
Roundtable, Iowa City, lA, 22 May 1993. 

. 1993b. Now that we're all here, where do we sit? Phonological 

ordering in the Vedic clause-initial string. Paper read at the 
2nd-Position Clitic Workshop, The Ohio State University, 
Columbus, Ohio, 11 July 1993. To appear in the Proceedings. 

SCHOURUP, Laurence C. 1983. Common discourse particles in English 
conversation. Ohio State University Working Papers in 
Linguistics, 28. Columbus: Ohio State University. 

SPEAS, Margaret J. 1990. Phrase structure in natural language. Studies 
in natural language and linguistic theory 21. Dordrecht: Kluwer. 

TRAVIS, Lise DeMena. 1984. Parameters and effects of word order 
variation. MIT Ph. D. dissertation in Linguistics. 



Schaufelc: The Vcdic claiisc-inilial string and universal grammar 1 6 1 

UNDKRHII.L. Robert. 1988. Like i.s, like, focus. American Speech 63. 
234-246. 

WhbKI.HUTH, Gert, & Hans den Besten. 1987. Adjunction and remnant 
topicalization in the Germanic SOV languages. Paper presented 
at the GLOW Conference, 30 March-2 April 1987, Venice. 

ZwiCKY, Arnold M. 1985. Clitics and particles. Language 61.283-305. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 

SQUIB 
Discourse markers and sentential syntax* 

M. Lynne Murphy 

The notion Discourse Marker or Discourse Particle (henceforth, 
DM) has, of late, claimed the attention of many linguists, who have 
tried to define the meanings or uses of these lexical items. Relatively 
little has been said, however, about the syntactic status of such 
pieces of language as oh, like, y'know, and well, though several 
syntacticians (e.g. Ross (1973), Emonds (1976), McCawley (1982), 
Espinal (1991)) have focused their concern on the Discourse Marker's 
wordier cousin, the parenthetical remark. Though Deborah James 
noted many of the syntactic peculiarities of DMs in her 1972 and 
1973 CLS papers, her data and provocative arguments for syntactic 
treatment of DMs have not received much attention in the current 
flurry of literature on DMs and parentheticals. The aim of this 
writing is to examine and refute James's argument that DMs are 
syntactically sensitive to the sentences they inhabit. Instead, I 
maintain that semantic well-formedness, rather than syntactic well- 
formedness, determines where a DM can occur. Thus, DMs are best 
treated as independent of the syntactic structure of the sentence. 

Discourse Markers are those lexical items which are used by the 
speaker to comment upon the discourse plan and goals. This defini- 
tion, from Andrea Dunn's 1990 dissertation on Swahili DMs, covers a 
large assortment of lexical items in English, which do not otherwise 
fall into traditional parts of speech, such as oh, ah, uh, certain uses of 
well, say, y'know, like, and non-conjunctive uses of so, and, hut, 
among others. Just a few examples of DMs are given in (I)-(IO). 

(1) Oh, 1 wish I were an Oscar Mayer wiener. 

(2) Ah, 1 see you've found the wine cellar. 

(3) I, like, couldn't believe that Elvis was dead. 

(4) Well, 1 always meant to buy flood insurance. 

(5) If we had, say, a jelly roll, we'd be having a lot more fun. 

(6) Now, there comes a time in every linguist's life when she 
must read Chomsky. 

(7) I never meant to electrocute my brother's goldfish, 
y'know. 

(8) But you said 1 could have a cookie! 

(9) And how am I supposed to pay for this? 

(10) So you think I'm made of money? 

DMs have a number of characteristics in common which are 
symptomatic of their discourse-comment purpose. First, they do not 
generally contribute to the truth-conditional semantics of their host 
sentences. This can be seen in comparing the DM y'know with the 



164 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

subject-verb sequence you know. In (11), where know is the matrix 
verb and you its subject, the sentence is true if and only if my 
mother's maiden name is among the things that you know. On the 
other hand, sentence (12) with DM use of y'know (12) is truth 
conditionally equivalent to (13), which has no you know at all. And 
in fact, (12) is felicitously used even, or especially, if (11) is false. 

(11) You know (that) my mother's maiden name is Bangs. 

(12) Y'know, my mother's maiden name is Bangs. 

(13) My mother's maiden name is Bangs. 

Another characteristic of DMs is their freedom in syntactic 
ordering. Any syntactic position may be preceded or followed by 
some Discourse Marker. The DM like can be used at any point in the 
sentence, to the frustration of parents of teenagers everywhere. 
Though it is unlikely that sentence (14a) would be uttered with all of 
the possible likes indicated, with the right intonational patterns, it 
can support any one, or even several, of the likes, as in (14b) or 
(14c), for example. 

(14) a. (Like) it's (like) been (like) forever (like) since (like) 

I've (like) been (like) to (like) the (like) mall (like). 

b. Like it's been forever since I've like been to the mall. 

c. It's been like forever since like I've been to like the 
mall. 

But not every DM is so flexible in its placement. For instance, 
while now and ah are both acceptable at the beginning of sentences, 
as shown in (15), now, but not ah, is acceptable at the end of a 
sentence, as in (16). 

(15) a. Now, the Jolly Green Giant doesn't like lima beans, 
b. Ah, the Jolly Green Giant doesn't like lima beans! 

(16) a. The Jolly Green Giant doesn't like lima beans, now. 
b. #The Jolly Green Giant doesn't like lima beans, ah. 

These sorts of facts about the distribution of DMs in host 
sentences inspired James to try to account for DM distribution within 
the schemata of sentential syntax. In her 1972 CLS paper, James 
makes two major claims. First, that DMs have meaning, in that they 
can be distinguished from one another by what they communicate. 
And second, that there are syntactic constraints on the distribution of 
DMs. James starts to catalogue some of these syntactic constraints, 
such as: DMs cannot break up idiomatic phrases, as in (17), or follow 
preposed manner adverbials, as in (18). 

(17) *John kicked, oh, the bucket. 

(18) *Wisely, why, Jan left early. 

In her 1973 CLS paper, James further develops her 1972 claims 
and asserts that DMs refer to parts of the sentence, and the parts of 
the sentence to which they refer are necessarily syntactic con- 



Murphy: Squib on discourse markers and sentential syntax 165 

stituents. James's use of the term "refer" here is not the usual 
semantic notion of reference. Instead, it is what I have termed here 
as "comment". So, to rephrase James's claim, DMs comment upon 
parts of the sentence, and the parts upon which they comment must 
be syntactic constituents. This can be characterized as a kind of 
scopal relation — a DM has scope over the constituent it comments 
upon, although DM scope should not be equated with other types of 
scopal relations such as quantifier scope. James's claim for the 
syntactic sensitivity of DM scope is motivated by sentences such as 
(19) and (20), in which the scopes of the DMs are just those complete 
constituents that occur immediately to the right of the DM, as 
indicated by the square brackets. 

(19) The woman who said she liked, [oM-scope oh, [vp to read]] 
sang a song. 

(20) That Kim is thinking of moving to [oM-scope ah, [np San 
Francisco]] is considered likely. 

Since the scope of oh in (19) cannot be to read sang a song, 
which is not a constituent, James concludes that the objects over 
which DMs have scope must be well-formed syntactic units. 
However, we can put James's claim to rest with one simple counter- 
example, stated in (21), where the DM simultaneously comments on 
more than one syntactic unit in the sentence. 

(21) Jackie should drive, [oM-scope oh, [pp from Nova Scotia] [pp 
to Arizona] ], just for the experience at the wheel. 

In (21), from Nova Scotia to Arizona, is not a syntactic 
constituent, rather it is a series of two prepositional phrases, each of 
which is a daughter of the larger VP. However, this string is most 
likely what the speaker intends the oh to comment upon, if we read 
the oh as indicating that the content of the string from Nova Scotia to 
Arizona was made up at the spur of the moment as exemplary start 
and finish points for Jackie's trip. Thus, James's claim that syntactic 
constituency is relevant to the scopal properties of DMs is falsified, 
although we still need to explain why the scopes of DMs are so 
frequently linked to syntactic constituency, as in (19) and (20). If 
syntactic constituency is irrelevant, then why is it that non- 
constituents like to read sang a song are not granted DM scope? 

The answer to this question lies in the fact that well-formed 
syntactic units reflect well-formed semantic units. That is, syntactic 
units are meaningful, and only units of meaning are commented 
upon by Discourse Markers. This jibes with Dunn's definition of DMs 
as commenting on the speaker's goals or plan in the discourse. It is 
only by communicating with the audience via meaningful units of 
language that the speaker can successfully complete her discourse 
plan, or parts thereof. Thus, DMs comment upon meaningful units of 
the discourse in order to clarify how the statement of those units 
contributes to the speaker's goals. Although it is not a syntactic 
constituent, from Nova Scotia to Arizona is a meaningful unit in the 



166 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

discourse, referring to a path by designating its two extremes. In 
commenting upon this string, oh indicates that the specifics of the 
path were extemporaneously produced, and therefore it may be the 
case that those specifics were not terribly relevant to the success- 
fulness of the discourse plan. In other words, the oh indicates that 
the path from Nova Scotia to Arizona is a good one for exemplifying 
the speaker's point, but that some other path might have worked 
equally as well. 

Having established that DMs are not syntactically sensitive in 
their scopal properties, we are left with no motivation for attributing 
syntactic properties to DMs. Recall that DMs defy distributional 
generalizations, making them an unlikely syntactic category. While 
nouns, for instance, might be syntactically interchangeable in noun 
phrases, DMs are not necessarily interchangeable in any particular 
sentence. Furthermore, DMs play no syntactic roles in the sentence, 
since they are neither subcategorized nor specified for. Finally, there 
is little evidence for the position of DMs in the hierarchical structure 
of the sentence. For instance, in (21), what would be the mother of 
ohl The first PP, in which case oh from Nova Scotia would be a 
constituent? The VP {drive, oh, from Nova Scotia to Arizona)! The 
sentence? Each possibility presents problems for syntactic theory, 
since none of these proposed constituents passes the traditional tests 
for constituency: they don't form prosodic units, nor are they moved, 
copied, or deleted as constituents. 

The argument that DMs are not syntactic constituents of their 
host sentences can be extended to parentheticals, which can serve 
the same purpose as lexical DMs, making meta-discourse comment, 
as in (22). 

(22) Jackie should drive — [dm -scope what would be a good 
route — [pp from Nova Scotia] [pp to Arizona] ], just for the 
experience at the wheel. 

Recognition that neither DMs nor parentheticals are properly within 
the sentential syntax puts to rest the arguments among Ross (1973), 
Emonds (1976), and McCawley (1982) as to what the constituency 
status of parentheticals is. Accounts in three-dimensional syntax 
(Espinal 1991) or autolexical syntax (as formulated by Sadock 1991) 
may be more promising, as such accounts allow for multiple syntactic 
dimensions in which DM/parenthetical and sentence structures could 
exist independently. It is unclear, however, how the separate 
syntactic structures for the host sentence and the DM/parenthetical 
would interact semantically such that the theory can represent the 
DM's scope over semantic units in the sentence. 

To conclude, I have shown here, contra James 1973, that there is 
no motivation for treating DMs and their host sentences as composite 
syntactic structures. Instead, it seems that DMs interact with their 
host sentences with reference to semantic rather than syntactic units. 
While multidimensional theories of language structure hold promise 



Murphy: Squib on discourse markers and sentential syntax 1 67 

for the syntactic representation of these sentences, the semantic 
interaction of DMs and their hosts deserves further investigation. 

NOTES 

* I am grateful to many members of the University of Illinois 
Department of Linguistics for comments and discussion concerning 
this work, especially Jennifer Cole, Georgia Green, and Jerry Morgan. 

REFERENCES 

Dunn, Andrea Susan. 1990. The pragmatics of selected discourse 

markers in Swahili. University of Illinois at Urbana-Champaign 

Ph.D. dissertation in Linguistics. 
Emonds, Joseph E. 1976. A transformational approach to English 

syntax: Root, structure-preserving, and local transformations. 

New York: Academic Press. 
Espinal, M. Teresa. 1991. The representation of disjunct constituents. 

Language 67:4.726-762. 
James, Deborah. 1972. Some aspects of the syntax and semantics of 

interjections. CLS 8.162-172. 
1973. Another look at, say, some grammatical constraints, on, oh, 

interjections and hesitations. CLS 9.242-251. 
McCawley, James D. 1982. Parentheticals and discontinuous 

constituent structure. MS. University of Chicago (printed in 

Linguistic Inquiry 13). 
ROSS, John R. 1973. Slifting. The formal analysis of natural languages, 

ed. by M. Gross, M. Halle, & M. Schiitzenberger. The Hague: 

Mouton. 
SadoCK, Jerrold M. 1991. Autolexical syntax. Chicago: University of 

Chicago Press. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



REVIEW 

Anvita Abbi. Reduplication in South Asian languages: An are- 
al, typological, and historical study. New Delhi: Allied Publish- 
ers, 1992; pp. xxii, 193. 

Hans Henrich Hock 

South Asia is one of the paradigm cases of linguistic converg- 
ence, involving the three major linguistic families Indo-Aryan, Dravi- 
dian, and Munda, as well as some of the Iranian and Tibeto-Burman 
languages, Khasi (Austro-Asiatic like Munda, but belonging to a dif- 
ferent subfamily), the language isolate Burushaski, and languages 
such as Nahali whose genetic affiliation is controversial. The features 
most prominently mentioned as defining the convergence area are (i) 
a contrast between dental and retroflex consonants; (ii) SOV order; 
(iii) the widespread use of absolutives and other non-finite devices 
instead of, or beside, clausal subordination, and (iv) the use of quota- 
tive constructions marked by a post-citation quotative particle. It is 
generally acknowledged that Indo-Aryan, Dravidian, and Munda 
form the core of the convergence area.' See especially the synchronic 
work of Masica (1988) and Ramanujan & Masica (1969). The his- 
torical developments giving rise to convergence are still a matter of 
controversy; cf. e.g. Emeneau 1954, 1956, 1962, 1980, Kuiper 1967a, 
b, 1991, and Kaufman & Thomason 1988 vs. Hock 1975, 1982, 1984 
and the yet different interpretation in Tikkanen 1987. 

The volume under review attempts to demonstrate that another 
important feature characterizes the South Asian convergence area, 
viz. REDUPLICATION. 

Abbi (A.) uses the term 'reduplication' in a fairly broad sense to 
refer to all of the highlighted structures in (1) - (4), although some- 
times she distinguishes structures of the type (1) and (2) from those 
of the type (3) and (4) as 'morphological' vs. 'lexical'. 2 (The examples 
in (1) and (2) are from Sanskrit, (3) from Hindi, and (4) from Khasi.) 
Her major emphasis is on the type (3); for it is this type, especially 
structures like (3cd) with iterated non-finite verbs, which she claims 
characterizes the core of the South Asian convergence area. 

(1) Reduplicated structures from Skt. tar- 'cross over' 

a. Reduplicated perfect: ta-tar-a 

b. Reduplicated desiderative: ti-tlr-sa-ti 

c. Reduplicated intensive: tar-tarl-ti 

(2) a. adhikadhika- 'very much, very many' 

adhika-adhika- (from adhika- 'much, many') 
b. ekaika- 'each one (in turn)' 
eka-eka- (from eka- 'one') 



170 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

(3) a. ravaiikTbar] bar! arhkherh thTrh 

Ravana's/big/big/eyes/be-past 
'Ravana had big, big (= very big) eyes.' 

b. unko ek ek phal de do 
they-dat/one/one/ fruit/give/give 
'Give them one fruit each.' 

c. vah pan becte becte bola 

[verbal adverb] 
he/betel /selling/selling/speak -past 
'He spoke while selling betel.' 

d. vah baithe baithe thak gaya 

[verbal adverb] 
he/sit- perfective/sit- perfective/tired/go 
'He got tired from sitting (for a long time).' 

(4) ban lyait si lyait ogala thait 

[ interfixed reduplication'] 
inf in itive-marker/walk/'infix'-particle/walk/l -past/tired 
'I am tired of walking continuously.' 

A. bases her claims on an impressive amount of data, much of 
them coming from her own fieldwork. Within modern South Asia she 
examines the thirty-three languages and dialects listed in Table I. 
For comparison and contrast she adds data from earlier stages of 
South Asian languages and from related languages outside the sub- 
continent (cf. Table H), with some additions from 'Hokkien and 
Chinese structures', some of 'the languages of Indonesia, Thailand, 
Malaysia, the Philippines, and Singapore' (8), as well as ancient Iran- 
ian (9). The total number of languages is 'about 48 to 50' (8). 



Indo-Aryan: 


1. 


Assamese 


8. 


Kashmiri 




2. 


Bengali 


9. 


Marathi 




3. 


Dakhini Urdu 


10. 


Maithili 




4. 


Dogri 


11. 


Oriya 




5. 


Gujarati 


12. 


Panjabi 




6. 


Hindi-Urdu 


13. 


Sindhi 




7. 


Konkani 


14. 


Sadari 


Dravidian: 


1. 


Kannada 


4. 


Malayalam 




2. 


Kodagu 


5. 


Tamil 




3. 


Kurukh 


6. 


Telugu 


Tibeto-Burman: 


1. 


Gangte 


4. 


Paite 




2. 


Kabui 


5. 


Thado 




3. 


Meitei 






Austro-Asiatic: 










a. Munda: 


1. 


Birjia 


5. 


Mundari 




2. 


Juang 


6. 


Santhali 




3. 


Kharia 


7. 


Sora 




4. 


Kuruku 







b. non-Munda: 8. Khasi 

Table I: Modern South Asian languages 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 171 



Indo-European: 

a. South Asian: 

b. Outside: 


1. 

2. 
4. 

5. 


Sanskrit 
Prakrit 
Greek 
Hittite 


3. 

6. 

7. 


Pali 

Pashto 
Persian 


Dravidian: 


1. 


Old Tamil 






Austro-Asiatic: 


1. 

2. 


Malay 
Iban 


3. 
4. 


Thai 
Kadazan 



Table II: Ancient and 'outside' languages 

One of A.'s major claims is that type (3) (and to some extent, 
(4)) reduplication of 'verbal adverbs' (VA) has aspectual functions. 
These are: (i) simultaneity of the actions of the VA and the main 
verb (MV) of the clause as in (3c); (ii) 'non-precipitation' where the 
MV action stops the incipient VA action, as in (5a) from Hindi; (iii) 
'continuation-duration' as in (3d) and (4); (iv) 'iteration' as in (5b) 
from Hindi; and (v) 'sequentiality', where the action of the VA is 'long 
and often continuous', while the MV action is 'short and abrupt' (53); 
cf. (5c) from Konkani. A. claims that the 'iterative' and 'continuative- 
durative' functions are found in all of the South Asian languages and 
therefore must be considered a feature defining the South Asian con- 
vergence area. 

(5) a. baris hote hote rah gal 

rain/happening/happening/stop/ went 

'It was going/about to rain, but it stopped.' 

b. vah gana sun sun kar thak gaya 

(s)he /song /listen/listen /a bsolutive-marker/ti red /went 
'He got tired of listening to songs.' 

c. uleita uleita taja doleat duka aili 
speaking/speaking/his- from/eyes/tears/came 

'He had been speaking (for a long time) and suddenly he 
began weeping.' 

Chapter IV deals with reduplicated modifiers (adjectives, quan- 
tifiers, and adverbs). A. determines four major functions and discus- 
ses the degree to which they are present in different modifier sub- 
categories. The four functions are (i) 'emphasis' as in (3a) above, (ii) 
'attenuation' as in (6a) from Hindi, (iii) 'exclusiveness' as in (6b) from 
Hindi, and (iv) the well-known feature of 'distributiveness', cf. (3b) 
above. She finds that 'emphasis' and 'distributiveness' are most 
widely attested in South Asia; 'exclusiveness' is rare in Dravidian, 
more common in Indo-Aryan and Tibeto-Burman, and most common 
in Munda; while 'attenuation' is most common in Indo-Aryan, rarer 
in Munda and Dravidian, and least common in Tibeto-Burman. 

(6) a. yah phal khatta khatta hai 

this /fruit/sour/ sour /is 
'This fruit is sourish.' 



172 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

b. aurterii aurtem mela dekhne gaim 
ladies/ladies/fair/see/went 
'Only ladies went to see the fair.' 

Chapter V examines in greater detail the geographical distribu- 
tion of the functions distinguished in Chapters 111 and IV, as they oc- 
cur with different subcategories of reduplicated structures. 

Her findings suggest a core 'Reduplication Area' which includes 
the Munda languages and the Indo-Aryan languages Hindi-Urdu, 
Panjabi, Dogri, [Sadari], and Gujarati. The other Indo-Aryan langu- 
ages, Marathi, Maithili, Oriya, Assamese, Konkani, Sindhi, Bengali, and 
Kashmiri, corresponding to Grierson's (1921) 'outer circle' of Indo- 
Aryan, agree with Tibeto-Burman and most of Dravidian in exhibit- 
ing reduplicated structures to a lesser degree. Finally, within Dravidi- 
an, Tamil and especially Malayalam are most 'resistant' to using re- 
duplication. A. concludes that 'It appears that the structures origin- 
ated somewhere in central India, perhaps in Munda languages, and 
then spread outside the Munda speech region' (96).' 

Chapter VI is concerned with the function of reduplication in 
Austro-Asiatic, comparing members of the Munda branch of the fam- 
ily with non-Munda Khasi, with side glances at non-South Asian 
members of the family, as well as other languages of Southeast Asia, 
Austronesian, and Chinese. 

A. finds that 'morphological' reduplication (of the type (1) and 
(2) above) plays a much more prominent role in Austro-Asiatic than 
in (modern) Indo-Aryan, Dravidian, and Tibeto-Burman. Within the 
Austro-Asiatic languages, however, the Munda branch stands out by 
making extensive use of 'lexically' reduplicated verbal-adverb struc- 
tures of the type (3cd) which, A. claims, are absent even in Khasi 
(138). 

This finding, if correct, would have interesting consequences for 
the historical interpretation of the geographical distribution of South 
Asian reduplicated structures. However, A.'s own data include the 
Khasi structure in (4) above (A.'s example (19), p. 130) which sug- 
gests that the distinction between Munda and Khasi is not as great as 
A. claims. 

Chapter VII, entitled 'Reduplication in classical languages: Indo- 
European and Dravidian', attempts to use historical evidence to come 
to a better understanding of the origins of South Asian reduplication. 

A. acknowledges Dressler's (1968) evidence for structures with 
'lexical' reduplication in early Indo-European languages, as well as 
Macdonell's (1916:281-282) more extensive listing of Vedic forms of 
this type. Now, Macdonell gives only two Vedic structures with iter- 
ated verb, both involving imperatives, cf. (7),^ but none with non- 
finite verbs (comparable to (3cd) above). This fact, combined with 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 173 

'[t]he total absence of any discussion on reduplicated verbal adverbs 
... by Panini or any other grammarian', leads A 'to believe that, 
perhaps, up to the period of 600 - 500 B.C. duplication of verbal 
adverbs ... had not emerged in Vedic or Classical Sanskrit' (150).''^ 

(7) a. piba-piba 'drink, drink' (RV 2.11.11) 

b. yajasva-yajasva 'sacrifice, sacrifice' (SB 12.3.4.1 (2x)) 

Chapter VIII presents A.'s major conclusions. Although these 
are certainly significant and thought-provoking, it is not quite clear 
how some of them follow from her preceding discussion. 

Apparently because of the widespread use of reduplication (in 
the larger sense) in Austro-Asiatic, A. feels that the reduplicated 
structures which she has examined are 'the gift of Munda on the In- 
dian soil' (162). On the other hand, she claims that 'the development 
and widespread use of verbal adverbs ... in reduplicated forms 
representing various aspectual elements seem to be the internal 
innovation of Indo- Aryan languages' (162), even though they are not 
yet found in Vedic and the Classical Sanskrit of Panini (161). 

A. considers Dravidian origin of these constructions unlikely for 
several reasons: (i) Brahui lacks such structures; (ii) the earliest in- 
digenous Dravidian (Tamil) grammar, Tolkappiam, 'has no reference 
to the structures under consideration'; (iii) the range of functions is 
more limited in Dravidian than in Indo-Aryan and Munda; (iv) 
Dravidian offers fewer subtypes of reduplicated constructions (163). 

Further claims include the suggestion that the widespread use 
of morphological reduplication in Munda and Proto-Indo-European 
may be attributable to early convergence 'and contact between the 
speakers of the two families of which we have no knowledge' (160).^ 

Finally, A. reiterates her earlier finding that the 'inner-circle' 
Indo-Aryan languages, Hindi-Urdu, Panjabi, Dogri, and Gujarati, to- 
gether with Munda, form the core 'reduplicated area'. The 'outer- 
circle' Indo-Aryan languages, Konkani, Kashmiri, Maithili, Marathi, 
Oriya, Bengali, and Assamese, together with much of Dravidian con- 
stitute an area with attenuated reduplication. The languages with the 
least amount of reduplication are the most peripheral. These are 
Dravidian Malayalam in the extreme southwest, and Austro-Asiatic 
(but non-Munda) Khasi as well as Tibeto-Burman Thado in the ex- 
treme northeast (166). 

As noted by Colin P. Masica in the Foreword, A.'s monograph is 
a 'significant contribution' to the study of the South Asian converg- 
ence area, addressing an aspect 'hitherto little explored', based on 
extensive fieldwork, and 'rich in original data ... It ... will repay 
careful study' (v). At the same time, as A. herself realizes (e.g. p. 9), a 
pioneering study of this type cannot yield satisfactory answers for all 
questions. 



174 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

In the following I suggest some areas for improvement in a re- 
vised edition, when such an edition will be published.^ 

First, A.'s use of the term 'reduplication' to refer to all of the 
structures in (1) - (4) is unfortunate, since the structures differ con- 
siderably, both in form and semantics. Using a single term can blur 
these distinctions and/or confuse the reader; and A.'s distinction 
between 'morphological' and 'lexical' reduplication only partly allevi- 
ates the problem. At the same time, in spite of their differences, the 
structures may exhibit various degrees of overlap, especially in func- 
tion. A fuller discussion of these similarities and differences and a 
more differentiated terminology would certainly be helpful in estab- 
lishing a clearer focus for the discussion. 

Let us begin by examining the formal similarities and differ- 
ences between these structures:^ 

The type (1) involves MORPHOLOGICAL DERIVATION similar to affix- 
ation. This is the type for which the term 'reduplication' traditionally 
is most commonly used. Like other types of derivational morphology 
it may be more or less productive; but it tends to exhibit much of the 
idiosyncrasy characteristic of derivational morphology. For instance, 
while in Sanskrit perfects, the vowel of the reduplication syllable is 
sensitive to the underlying root shape (as in u-vac-a from /uac-/ 
'speak', yu-yoj-a from /yuj-/ 'yoke'), in other forms it is sensitive to 
the surface root shape (as in the corresponding desideratives vi-vak- 
sa-ti 'desires to speak' vs. yu-yuk-sa-ti 'desires to yoke'). 

The type (2) also is morphological, but it belongs to the category 
of COMPOUNDS, formally similar to structures like Skt. divanisam '(by) 
day (and) night'. Like other compounds these structures constitute 
single lexical units, with inflectional endings added at the end of the 
compound, not to each of the component parts. Moreover, excepting 
lexicalized forms, compounds generally exhibit fewer idiosyncrasies 
than structures of the type (1). The term 'iterative compound' may 
be useful to distinguish type (2) from type (1) structures. 

Structures of the type (3) look like ITERATIONS of fully inflected 
words. (For instance, both instances of haithe in example (3d) contain 
the adverbial affix -e.) As such, they behave more like syntactic col- 
locations than like morphological compounds. The traditional term 
for this type in Sanskritist and Indo-Europeanist literature is 'amre- 
dita'; a less exotic term would be 'lexical iteration'. 

The type (4), finally, looks very similar to SYNTACTICALLY con- 
joined (or quasi-conjoined) structures of the English type (8) and the 
Vedic Sanskrit type (9). These two types differ only in terms of the 
method of signaling coordination. In the very widespread type (8), 
an overt coordinating conjunction is employed, while in (9) coordina- 
tion is signaled by the use of a case form (in the present case, the 
instrumental in 'comitative' function). ^ It is convenient to refer to 
this type as 'iterative conjuncts'. 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 175 

(8) a. better and better 

b. through and through 

c. Turning and turning in the widening gyre I The falcon can- 
not hear the falconer (Yeats, The second coming, 1) 

(9) yudh-a yudham lipa ... esi (RV 1.53.7) 
battle-instru mental/battle- accusative/to/you-go 
'You go to battle after battle/every battle.' 

From the formal perspective, then, the four different structures 
can be ranked in the order (1) - (2) - (3) - (4), with (1) most proto- 
typically morphological and (4) closest to being fully syntactic. 

However, the distinction between these types is not absolute. 
For instance, in languages without inflectional affixes, the types (2) 
and (3) are difficult to distinguish. In fact, even the Sanskrit type (3) 
exhibits behavior close or even identical to that of compounds.^ Thus, 
in accented texts, only the first of the iterated items bears accent; cf. 
(10a). 10 Moreover, clause-second particles follow the entire structure 
rather than the first element; cf. (10b) vs. (10c).' ^ 

(10) a. sa vai saihm fjya sarhmrjya 

he/particle/wipe-absolutive/wipe-absolutive 

pratapya pratapya pra yacchati (SB 1.3.1.8) 

heat-absolutive/heat-absolutive/<gives 

'He gives (the ladle to the adhvaryu) after each wiping and 

heating.' 

b. vyatiharam vyatihararh h(i) 
transpose-absolutive/tr a nsp.-absolutive/' for' (P2-p article)/ 
uttaravedirh vyagharayanti (SB 9.2.1.7) 
uttaravedi/over-sprinkle 

'For they over-sprinkle the uttaravedi, transposing again 
and again.' 

c. *vyatiharam hi vyatiharam ... 

On the other hand, where there is no evidence (as in (10)) that 
structures of the type (3) are compounds, it may be difficult to reject 
an alternative analysis of these as being 'asyndetic' (i.e. conjunction- 
less) counterparts of the syntactically (quasi-)conjoined type (4), (8), 
or (9). 

The formal overlaps between the types (2) and (3), as well as 
(3) and (4), may be significant, suggesting a possible diachronic path 
of grammaticalization, from the most 'syntactic' type (4), via the 
quasi-compound type (3), to the full-compound type {2)A^ 

Functionally, too, the types (2) - (4) exhibit a great degree of 
affinity: All three of them most commonly signal either intensitivity 
(as in (2a), (3a), and (8b)) or distributiveness and iterativity (as in 
(2b), (3b), (8a), and (9)). By generally exhibiting this very restricted, 
and highly iconic, range of functions, these types differ markedly 
from the type (1), whose range of functions is much broader and 



176 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

generally much less iconic. (The examples in (1) illustrate only a 
small subset of these functions.) 

For these reasons it is legitimate to distinguish types (2) - (4) as 

a group from type (1), by referring to them as exhibiting 'iteration', 

while type (1) exhibits the much more clearly morphological process 
of 'reduplication'. 

Functionally, even this distinction is not absolute: As is well 
known, some subtypes of morphological reduplication exhibit the 
same iconic functions of intensity and distributeness/iterativity as 
the types (2) - (4); cf. e.g. the Sanskrit 'intensive' type (Ic). Struc- 
tures of this type, therefore, may be relevant when discussing con- 
structions of the type (2) - (4). Other forms of reduplication, how- 
ever, would generally be of questionable value. 

The differences between these four different types of structure 
have significant implications for establishing genetic relationship or 
contact. 

In principle, for instance, similarities in the purely morphologic- 
al type (1) would be most probative, especially if they are highly 
idiosyncratic (rather than iconic), since such similarities are least 
likely to result from chance. In this regard, then, A.'s claimed paral- 
lelism of Austro-Asiatic and early Indo-European as regards mor- 
phological reduplication might be significant. However, this parallel- 
ism holds only at a very general level; the details of reduplication 
differ to such a degree that a special prehistoric relationship is quite 
unlikely. 

On the other side of the spectrum, the iterative conjunct type 
(4) and (8/9) has the least probative value in comparative work. It is 
highly iconic (rather than idiosyncratic), draws on the crosslinguistic- 
ally very productive syntactic process of coordination, and as a con- 
sequence is found in language after language after language. Evid- 
ence of this type, therefore, is best ignored in trying to establish a 
special (contact or genetic) relationship, unless the nature of the 
evidence happens to be highly idiosyncratic. 

A.'s best evidence lies in the area of types (2) and (3), i.e. in the 
area intermediate between the most morphological and most syn- 
tactic types. Structures of this type are more likely to exhibit idio- 
syncrasies than type (4) structures. For instance, while all varieties 
of English have the 'iterated conjunct' type (8), 'lexical iteration' is 
relatively restricted, especially in American English, where even the 
British type It's very very difficult tends to be avoided. Even a pri- 
ori, then, similarities in structures of the type (2) and (3) are more 
likely to be indicative of a special, contact or genetic, relationship. 

In addition, however, A. has been able to isolate a number of 
specific South Asian idiosyncrasies. One of these is found in the 'at- 
tenuated' and 'exclusive' functions of the types (6a) and (6b) which 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 177 

differ from the normal, iconic, intensive and distributive/iterative 
functions of these structures. 

A second idiosyncrasy consists in the uneven, but patterned, 
geographical distribution of the structures, which strongly suggests 
spread from a core area. 

A third, and perhaps most significant, idiosyncrasy lies in the 
existence of the iterated verbal-adverb type (3cd) and its special 
semantic functions, a phenomenon which does not seem to be found 
in related languages outside of India. 

So far, so good. But several questions arise. One of these is the 
typological one of how idiosyncratic these structures really are. An- 
other set of interrelated issues concerns the reliability of A.'s data: 
Are they sufficient to establish that the reduplicated structures orig- 
inated in Munda, that the 'reduplicated verbal-adverb' type is an In- 
do-Aryan innovation, and that this type developed in post-Paiiinian 
times? And can we be certain that it was absent in early Dravidian? 

Now, while 'attenuated' and 'exclusive' functions are not a usual 
crosslinguistic feature of iterated structures, there are parallels in 
other languages, as well as historical (near-)antecedents in Sanskrit. 
Both functions can be documented in English. Attenuation is ob- 
served in the expression so-so (which has parallels in many other 
European languages), even though this type is quite marginal. Some- 
thing close to exclusivity is found in the much more productive, quite 
colloquial, English pattern (11). But perhaps this pattern is of fairly 
recent origin. 

(11) It's not a mickey-mouse book; it's a book book. 

More significant is the fact that similar semantic specializations 
are found relatively early in Sanskrit. Delbriick (1888:55) notes the 
use of iterated updri 'over' and aclhds 'below' in exclusive function in 
structures like (12a); and in Delbriick 1900:151 he adds that this fact 
was already known to Panini (cf. his rule 8.1.7). Upon closer exam- 
ination, Panini's rules yield another, similar type of iterated adposi- 
tion with exclusive function (8.1.5), as well as an adjectival iterative 
compound type with attenuated function, cf. (12b). Attenuation is no 
doubt also found in the type (12c) mentioned in Delbriick 1888:55, 
572, but without discussion of its function. (The attenuated function 
is adumbrated by Harisvamin's alpdrthe 'in the meaning of little' in 
his commentary on the passage.) An exclusive function of iteration 
may even be found as early as the Rig-Veda; cf. (12d). 

(12) a. atha daksinarh bhriivam upary-upari lalatam lipa sprsati 

(SB 3.2.L29) 

then/right/brow/above-above/forehead/to/touch 
'Then he touches the forehead JUST above the right brow.' 
b. prakare gunavacanasya (8.1.12) 

'[Iteration] of an adjective in the meaning "of sorts".' 
Example: patu-patu- 'rather, somewhat hot/spicy' 



178 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

c. tasmin yavan va yavan va rasah sam asravat 
(SB 4.4.3.4) 

in- it/how-much/or/how -much/or/juice/together/flowed 
'Into it, just so much juice flowed.' 

d. tvaih-tvam aharyatha(h) ... (RV 10.96.5) 
you-you/we re -desired 

'Only you were desired ... ' (Geldner: 'immer nur du ... ') 

The fact that the attenuated and exclusive functions of iteration 
thus go back to the earliest stages of Sanskrit might be interpreted as 
indicating that these functions originated in Indo-Aryan. 

This conclusion would be compatible with A.'s observation that 
the geographic core area of attenuation consists of the Indo-Aryan 
languages. But it seems to be in conflict with A.'s claim that the core 
area of exclusivity is Munda. Unfortunately, no Munda texts (or texts 
from any other South Asian language group other than Indo-Aryan) 
can match the antiquity of Vedic Sanskrit. As a consequence it is not 
possible to judge whether at a comparable prehistoric time, Munda 
did or did not have iterated structures with exclusive (or attenuated) 
functions. We are thus unable to determine whether A.'s evidence for 
Munda being the core area of modern South Asian exclusivity should 
be interpreted as indicating that Munda was the originator of these 
feature, or whether for some reason the feature may have become 
more productive in modern Munda. 

The situation is similar as regards A.'s claim that iteration in 
general is of Munda origin. True, A. finds that iterated structures are 
most common in Munda and the neighboring Indo-Aryan languages. 
And it is certainly tempting to assume that the greater use of iter- 
ated structures in the latter, 'inner-circle' Indo-Aryan languages, as 
compared to the 'outer-circle' ones, is due to contact with Munda. 
However, here again we find that iteration is a feature found even in 
the earliest stratum of Indo-Aryan, Vedic Sanskrit, a period for 
which we simply have no counterparts in Munda or any of the other 
South Asian languages. Moreover, it is also found, to varying degrees, 
in other Indo-European languages, as well as in non-Indo-European 
languages outside South Asia. 

Before giving selected examples and references, however, it is 
important to note that for many of these languages it is much more 
difficult to find relevant evidence than for Sanskrit. Several factors 
seem to be responsible for this fact: In many languages iteration is a 
feature of the colloquial language (cf. e.g. Hofmann 1951:58-61). As a 
consequence, it may not be found freely in the formal texts that have 
been handed down (cf. Bartholomae 1907:166). Moreover, while the 
tradition of indigenous Sanskrit grammar offered a formal treatment 
of iteration under the notion amredita (Panini 8.1.1-13) and thus in- 
vited discussion of this phenomenon in western accounts, the ten- 
dency of traditional western grammars of such languages as Latin 
and Greek has been to treat iteration informally, under the heading 
of STYLISTICS (cf. e.g. Hofmann 1965:§45, Menges 1953:551, Schwyzer 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 179 

1949:613, 699-100). For these reasons, one has to be extremely care- 
ful in interpreting the information provided by traditional western 
accounts of languages other than Sanskrit (and occasionally even of 
Sanskrit). In some cases it is possible to supplement that information 
from individual articles or from primary textual research. But where 
such subsidiary information is not available, the absence of any dis- 
cussion of iteration, or the presence of just a few cursory notes, 
should not be taken to indicate that a given language had little or no 
iteration. 

Studies of the Rig-Veda, the oldest stage of Sanskrit, have es- 
tablished that lexical iteration is most common for nouns, pronouns, 
and preposition-adverbs; it is less common for adjectives and nu- 
merals; and exceedingly rare for verbs, for which there is just one 
certain example with accent only on the first item (cf. (7a) above, as 
well as note 10). See for instance Collitz 1882, Delbruck 1888:51-55, 
1900:141-153, Wackernagel 1905/Wackernagel & Debrunner 1957: 
142-148, Wackernagel 1929-30:395-396 (for numerals). 

Moreover, the Rig-Veda offers several examples of iteration 
with accent on BOTH elements (see note 10), a fact which might indic- 
ate that the normal type of lexical iteration with accent only on the 
first element is an innovation. Perhaps significantly, all of these ex- 
amples are exclamations, and the iteration merely signals emphasis, 
not the 'aspectual' functions of distributivity or iterativity. 

Interestingly, the examples of finite-verb iteration that I have 
found in the post-Rig-Veda Vedic language preponderantly are im- 
peratives (i.e. exclamatory in nature), and with one possible excep- 
tion likewise signal emphasis rather than aspectual functions (cf. (7b) 
above with note 3). Notice in this regard that iteration of impera- 
tives, vocatives, and other 'exclamatories' for emphasis is an ex- 
tremely common phenomenon, even in languages with much less 
robust evidence for lexical iteration than Sanskrit; cf. e.g. Hofmann 
1951:58-59, 1965:§45. 

These facts might support the view that verbal iteration with 
aspectual functions is in fact an innovation in Sanskrit/Indo-Aryan. 

Note however that traditional accounts of Rig-Vedic Sanskrit do 
not accord special treatment to one type of verbal iteration which 
does seem to be used in aspectual functions, viz. NON-finite verb iter- 
ation; cf. the examples in (13). To the extent that such structures are 
noted, they are classified as adjectives. And in fact, their use IS rath- 
er adjectival. 

(13) a. satah-satah pratimanarh parobhur (RV 3.31.8a) 
of being-being/equality/surpassing 
'Equal of every existing one/every being, surpassing ..." 
b. sute-sute vavrdhe ... (RV 3.36.1c) 
at pressed-pressed/grows 
'He grows at every pressed (soma) ..." 



180 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

c. jato-jato jayate vajy asya (RV 7. 90. 2d) 
born -born/is -born/ vie tor/his 

'Every born (son) of his is born a victor.' 

d. panyam-panyaih ft sotara a dhavata (RV 8.2.25ab) 
to-be-praised = to-be-praised/particle/pressers/up/stir 
'Stir up, O pressors, the one to be praised again and again.' 

At the same time, however, the iterated lexical items are VERB- 
AL adjectives, and they exhibit the aspectual functions characteristic 
of the modern South Asian type (3c/d). 

The first unambiguous examples of non-finite verb iteration 
with aspectual functions come from one of the earliest Vedic-Prose 
texts; cf. (14). Note also the somewhat later examples in (10) above. 

(14) taih samstambham-samstambham asuran ajayat 

(MS L4.14) 

they- instrumental/suppress in g-suppr./Asuras/defeated 
'Suppressing the Asuras again and again by means of these 
(victory sacrifices) he defeated (them).' 
(Similarly ibid, and MS 3.8.1, 3.10.5) 

Perhaps significantly, these unambiguously aspectual verbal 
iteration structures involve absolutives, i.e. verbal adverbs, rather 
than verbal adjectives. If the difference between the Rig-Veda and 
later Vedic Prose reflects a difference in chronology, rather than 
genre, 13 this fact may indicate that the iterated verbal-adverb con- 
struction is a post-Rig- Vedic, Vedic-Prose innovation. While this 
would support A.'s claim that the construction is an Indo-Aryan in- 
novation, it dates the innovation at a considerably earlier period.''* 

Now, as is well known, the origin and antiquity of the Sanskrit 
absolutives is a matter of controversy. Many scholars consider them 
an Indo-Aryan innovation, resulting from Dravidian influence; cf. e.g. 
Emeneau 1954, 1956, 1962, 1980, Kuiper 1967a, b, 1991, and Kauf- 
man & Thomason 1988. Following earlier suggestions I have argued 
that they may be inherited and that, at any rate, the arguments for 
Dravidian influence have not been established beyond a reasonable 
doubt (Hock 1975, 1982, 1984). While accepting the latter claim, Tik- 
kanen (1987) postulates a different, northwestern substratum as the 
source for the Sanskrit absolutive. 

Depending on one's position regarding this controversy, the fact 
that aspectual verbal iteration is first attested in Sanskrit for absolu- 
tives may suggest different conclusions. Accepting my arguments 
would make it relatively easy to follow A. in considering structures 
of the type (3cd), (10), and (14) an Indo-Aryan innovation, perhaps 
resulting from extension of the verbal-adjective pattern in (13) to 
the adverbial absolutives. Accepting the Dravidian-substratum hypo- 
thesis might be taken to support Dravidian origin for this type. This 
view would run into the difficulty that A. finds structures of this sort 
to be less common and the range of functions more limited in Dra- 
vidian than in the Indo-Aryan and Munda core area. But perhaps it 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 181 

is only the wider range of functions which is an innovation of the 
core area. Finally, accepting Tikkanen's hypothesis would open up 
yet further opportunities for speculation. 

Unfortunately, here again our ability to decide between these 
different possible scenarios is limited by the fact that none of the 
other languages is attested at anything approaching the time depth 
of Vedic Sanskrit. 

On the other hand, comparative Indo-European evidence sug- 
gests that iterated verbal-adverb structures are indeed an innova- 
tion of early Indo-Aryan. Even Avestan, the most closely related 
early Iranian language has much more restricted evidence of itera- 
tion, with examples limited to nouns, pronouns, and numerals. '5 All 
of these, to be sure, have the distributive/iterative functions which 
A. considers a feature of the entire South Asian area. But none of 
them involve verbs (whether finite or non-finite). At the same time, 
as Bartholomae (1907:166-167) notes, the dearth of examples may 
be attributable to the nature of the Avestan texts. More examples of 
iteration are found in later Iranian. See also Heston 1980.'^ 

The majority of the non-Iranian Indo-European languages offer 
even less evidence for iteration with distributive/iterative function, 
at least in their earliest stages. In early Greek, for instances, the cited 
evidence is limited to Mycenaean we-te-i-we-te-i 'year after year', 
Cypriot a-ma-ti-a-ma-ti 'day after day' (Dressier 1968), the Homeric 
iterated verbal prefix propro- 'forward and forward' (Collitz 1882: 
298, Dressier 1968, Dunkel 1981a), and a few other less obviously 
distributive forms (for which see e.g. Dressier 1968). To these forms 
Dunkel (1981a,b) adds a few more early examples of iterated verbal 
prefixes. Excepting intensive iteration, especially of imperatives, voc- 
atives, and other 'exclamatories', early Latin distributive/iterative 
examples are essentially limited to iterated interrogative pronouns in 
indefinite-pronoun function, such as quisquis 'whoever'. Dunkel 
(I981ab) adds a few possible examples with iterated preposition/ad- 
verbs. Iterated numerals in distributive function are attested quite 
late, both in Greek and in Latin. 

Interestingly, iteration seems to be much more widely attested 
in Hittite. Dressier (1968) lists several examples of iterated locative 
nominals in temporal function, such as KASKAL-.u KASKAL-.f/ 'every 
time'. To these Dunkel adds several examples of iterated preverbs or 
adverbs in distributive/iterative function (1981a, with references). 
In addition, Hittite has iterated interrogatives used as indefinites, 
and at least one example of an iterated numeral; cf. (I5a). Dressier 
(1968) claims that it is not certain that the iterated structures func- 
tion like compounds as they generally do in Sanskrit. But examples 
like (15b) suggest that they do, since sentence-initial particles (in- 
dicated by underlining) follow the entire structure, rather than the 
first element. 



182 Studies in liie Linguistic Sciences 23:1 (Spring 1993) 

(15) a. nu 1-as 1-as ... sesuwanzi liebe karastari 
now/one/one/sleep/not/fail 
(KUB III 4,5,6,17,18,19 III 5-6) 
'Now let them not fail, one by one, to sleep ..." 
b. Ml-ti Ml-ti-ma. ... (KUB III 4,5,6,17,18,19 III 12-13) 
night/night/but 
'But night after night ...' 

While most Indo-Europeanists agree that the comparative evid- 
ence of the early Indo-European languages warrants reconstructing 
some kind of iterated structure for the proto-language, they disagree 
as to what structures should be reconstructed. Delbriick (1900:152- 
153) considered it likely that Proto-Indo-European had iterated pro- 
nouns, numerals, and preposition-adverbs. Dressier (1968) claims 
that iterated locative nominals can be reconstructed, but no other 
case forms of iterated nominals. Most recently Dunkel (1981a) argues 
that only iterated prepositions can be reconstructed, since only in 
this category do we find exact equations (of the type Skt. prd pra : 
Gk. propro). Dunkel's position may be overly restrictive: In quasi- 
syntactic structures of this type it is unrealistic to expect exact 
equations, just as we should not expect exact equations in sentences. 
What we can reconstruct in such cases, and even for many pro- 
ductive morphological processes, are structural PATTERNS (cf. Hock 
1985 and 1986/1991:576-577). Moreover, Dunkel's preverbs, whose 
behavior in early Indo-European suggests that they started out as 
adverbs, are quite similar to Dressler's locative nominals, whose 
function likewise is adverbial. Hittite evidence of the type (15a), 
combined with the Indo-Iranian evidence, further suggests that 
iterated numerals may have been possible in early Indo-European. 
And the agreement of Indo-Iranian, Latin, and Hittite in using iter- 
ated interrogative pronouns as indefinites might perhaps be taken to 
suggest reconstruction of this type, as well. But note that, if a pattern 
of iteration was inherited from the proto-language, it is perfectly 
possible that it was independently extended to any one of the struc- 
tures just mentioned. That is, although we can be quite certain that 
Proto-Indo-European had a pattern of lexical iteration, it is much 
more difficult to know which lexical categories were permitted to un- 
dergo iteration. 

Significantly, however, there is no comparative evidence for 
iterated VERBAL structures, either finite or non-finite, with distribut- 
ive/iterative function.'^ The comparative Indo-European perspect- 
ive, then, seems to support A.'s claim that iterated verbal adverbs 
are an Indo-Aryan innovation 

Non-Indo-European languages outside South Asia, however, do 
provide evidence for iterated verbal structures and thus present a 
challenge to the view that the constructions are a uniquely South 
Asian phenomenon. 

To judge by the sample folk stories from non-literary Uralic 
languages in Collinder 1965, finite-verb iteration in iterative-contin- 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 183 

uative function is not uncommon in Uralic. Compare e.g. Mordvin 
sokan sokan da sizin koda 'I plow and plow/I keep plowing, and how 
tired I got'.'^ For one language, Votyak, Collinder further cites the 
struture in (16), with iterated 'modal gerundium', a construction re- 
markably similar to the South Asian type (3cd). Unfortunately, Col- 
linder does not indicate how widespread this structure is in Uralic. If 
it is quite isolated, it might either be a spontaneous internal devel- 
opment or reflect Turkic influence. 

(16) berdysa berdysa gurtaz koskem 
crying /crying/home/went-narrative 

'She went home crying and crying/crying bitterly.' 

Most important is the evidence of Turkish. In addition to iter- 
ated adverbs and adjectives in intensive/emphatic function (Lewis 
1953:56, 59), it offers a large variety of iterated structures in dis- 
tributive/iterative function, including numerals (Lewis 1953:69), ad- 
jectives modifying plural nouns (Lewis 1967:236), nouns (ibid.), 
finite verbs (ibid. 235), or non-finite gerunds in iterated verbal-ad- 
verb structures of the type (17) for which see Lewis 1967:175-176. 

(17) insan belki dogiile dogiile uslanir 

person /perhaps/being-beaten/being-bea ten/ becomes- 

well-behaved 

'Perhaps one becomes well-behaved with being constantly 

beaten.' 

Structures of the latter type are strikingly similar to the South 
Asian iterated verbal-adverb structures in (3cd). Moreover, like their 
Sanskrit counterparts, they bear accent only on their first element 
(Lewis 1953:56, 1967:176), while at the same time fully inflecting 
both of the iterated elements 

These similarities by themselves would be remarkable enough 
to provide an interesting challenge to the view that lexical iteration, 
especially of the verbal-adverb type, is a uniquely South Asian fea- 
ture. 

However, these are not the only structural similarities between 
Turkish (and Turkic in general) and the languages of South Asia. Ex- 
cepting retroflexion, all the features commonly considered character- 
istic of the South Asian convergence area recur in Turkish: basic SOV 
order; the widespread use of absolutives and other non-finite devices 
instead of, or beside, clausal subordination; and the use of quotative 
constructions marked by a post-citation quotative particle. Moreover, 
Turkish shares with Indo-Aryan and Dravidian the use of 'relative- 
correlative' constructions which, as in Dravidian, are more marginal 
than relative verbal participle structures. Compare Hock 1988, 1989, 
1992, based in part on Steever's pioneering study of finiteness in 
Dravidian (1988). '^ 

In addition to posing a significant challenge to the view that all 
of these features uniquely define the South Asian convergence area. 



184 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

these similarities raise questions which, to my knowledge, have not 
been satisfactorily addressed so far, not just in A.'s monograph, but 
in any of the studies devoted to South Asia convergence: How do we 
explain these remakable similarities? Can they be due to chance? Or 
must they reflect some kind of bilingual contact? If we are really 
dealing with chance similarities, could some or all of the South Asian 
similarities likewise be due to chance? If there was bilingual contact, 
where in time and space should it be located? And how do we 
account for the fact that the Turkic languages are more similar in 
their syntax to the Dravidian languages than to Indo-Aryan which, in 
the form of Sanskrit, had a much better chance of coming into contact 
with Turkic in Buddhist Central Asia? Or should we assume that all of 
the features reflect inheritance from an often posited, but so far not 
well established, common ancestor of Dravidian and Turkic (or Altaic, 
the larger family of which the Turkic languages are commonly con- 
sidered to be a member)? 

These are difficult questions indeed. And it would be too much 
to expect A. to find satisfactory answers in a revised edition. Never- 
theless, even if no answer is possible at this time, the evidence of 
Turkic needs to be considered more fully in assessing to what extent 
iteration, especially the verbal-adverb type, is a feature uniquely 
defining the South Asian convergence area. 

A revised version of A.'s monograph might further benefit from 
taking another, closer look at what can be known about the earlier 
history of Dravidian. First, the range of functions in which iterated 
verbal-adverb structures are used in Dravidian does not seem to 
differ significantly from their range in Indo-Aryan and Munda. From 
A.'s discussion it appears that only one function, that of 'non- 
precipitation' in (5a) above, is absent in several Dravidian languages. 
Secondly, even if the oldest Dravidian grammar, Tolkappiam, does 
not explicitly mention iterated verbal-adverb constructions, this does 
not necessarily mean that such structures did not exist. It is very 
well possible that their existence is assumed or subsumed by the 
general rules that permit lexical iteration. Only an examination of Old 
Tamil texts, and of the oldest texts in the other Dravidian languages 
with a long literary tradition, can yield some degree of certainty in 
this regard (unless there are genre-considerations that would 
preclude the use of such structures). 

At the same time, Tamil, in many respects the most con- 
servative Dravidian language, offers certain forms which might sug- 
gest that very different structures, with morphological (quasi-) 
reduplication, may have originally been used instead of some of the 
modern iterated structures — or perhaps even all? Beythan (1943: 
174) states that intensive adjectives are formed by a process of 
quasi-reduplication, with infixation of -nnaN- or some other element 
between the reduplication syllable and the stem; cf. (18a). And cer- 
tain distributive numerals likewise seem to be formed by some kind 
of reduplicative prefixation; cf. (18b) from Beythan (1943:143). 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 185 

(18) a. karutta 'black' : ka-nnaii-karutta 'very black' 
b. onru 'one' : o-vv-onru 'one each' 

beside iterated onru-v-onru 
irantu 'two' : i-vv-irantu 'two each' 
munru 'three' : mu-m-munru 'three each' 

Should structures of this type turn out to be genuine archaisms, 
rather than relatively recent reductions of originally iterated con- 
structions, this might support A.'s view that the principle of iteration 
spread from an Indo-Aryan/Munda core area to Dravidian. 

Finally, I would like to express the hope that a revised edition 
will receive greater editorial care by the publishers. There are all too 
many passages whose interpretation is unclear or which seem to be 
self-contradictory. (See for instance notes 4 and 5 above.) These pas- 
sages do injustice to the great amount of research that A. has done 
and to her extremely interesting conclusions. 

As the length of this review may suggest, A.'s monograph 
makes significant and challenging contributions to the study of South 
Asian convergence. It should be considered 'must' reading for 
scholars concerned with this topic, no matter whether they agree 
with all of her findings or not. 

NOTES 

• But note that Burushaski, in the extreme north of the area, also 
exhibits the major diagnostic features defining the South Asian con- 
vergence area. At the same time, the quotative construction is absent 
in many of the modern Indo-Aryan languages, as well as in the 
northwestern Dravidian language Brahui. (It was present in the ear- 
liest attested Indo-Aryan language, Sanskrit.) Indo-Aryan Assamese, 
in the extreme northeast, lacks the retroflex-dental contrast. (Again, 
the contrast was present in Sanskrit.) And so on. That is, as in many 
convergence areas, the isoglosses correlated with the presence of 
particular features do not form neat bundles that boldly define the 
area but rather, like dialectal isoglosses, more irregular patterns, 
especially in the transition areas between the core and the peri- 
pheral languages outside the area. Cf. e.g. Masica 1976 for South Asia 
and the discussion in Hock 1988a. 

2 In her discussion in Chapter 2 she further includes 'echo for- 
mations' of the type Hindi pani-vani 'water and the like' (from pani 
'water'), Tamil param-giram 'fruit and the like' (from param 'fruit') 
and compounds of the type Hindi dhan-daulat 'wealth' {dhan 'wealth' 
-I- daulat 'wealth, money') or Burmese thwa la 'travel' {thwa 'come' + 
la 'go'). She admits that the latter type is 'not duplicated on the pho- 
nological level'. Moreover, both types of structures do not figure pro- 
minently in her further discussion. 



186 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

The type dhan daulat, by her own admission, is not limited to 
South Asia but is also found in Burmese, Thai, and Malay. In fact, it 
recurs in Tocharian (Toch. A nom-klyu 'name-fame' = 'fame'). Old 
Irish (gaisced 'weapons' = gde 'spear' + sciath 'shield'; Meid 1968), 
Lithuanian (cf. Senn 1966:351), and many other Indo-European lan- 
guages (references in Meid 1968), as well as in Uralic (Collinder 
1965:49, 63) and Turkish (Lewis 1988:§43, 45). This type therefore 
is not a unique property of South Asia, except perhaps in terms of 
frequency of occurrence. 

'Echo words' likewise have outside parallels. Compare e.g. Turk. 
siki fiki 'intimate' (from siki 'close'; Lewis 1988:236); Engl, hurly- 
burly, pell-mell, etc.; Lat. (Plautus) at bat (pejorative, from at 'but'), 
eia beta (from eia 'up and at it'); and especially the Yiddish-based 
English type syntax-shmyntax. Most of these echo formations are not 
very productive and do not exhibit recurrent morphological charac- 
teristics, in stark contrast to the South Asian languages. Moreover, 
the only well-known PRODUCTIVE type, that of Yiddish, only has nega- 
tive connotations. But as Steever (1988) observed, the echo forma- 
tions of Tamil likewise have negative connotations. Moreover, if we 
expand our horizon beyond the (western) European languages, we do 
find at least one highly productive parallel outside South Asia, viz. 
the colloquial Turkish type Burada kutu mutu yok 'there is no box 
OR THE LIKE here' (Lewis 1953:112); and this type does not seem to 
have negative connotations. The question whether productive echo 
formation is an exclusive feature of the South Asian convergence 
area deserves further study. 

3 Even for finite verbs this list is incomplete. The Satapatha- 
Brahmana additionally offers ydja-yaja 'sacrifice, sacrifice' (1.5.3.8 
(2x), 2.5.2.30,41), veda-veda 'I know, I know' (14.6.7.5); the Aitare- 
ya-Brahmana, aireyathdm-aireyathdm 'did you divide, did you di- 
vide?' (6.15.12); the Jaiminiya-Brahmana, 7V/}'«r/-7V/}Y/r/ 'he wins again 
and again' or 'he wins, he wins (indeed)' (2.293) and daddni-daddni 
'let me give, let me give' (2.213). A complete investigation of finite- 
verb iteration in Vedic Sanskrit still is very much a desideratum. 

^ A.'s discussion on this point, unfortunately, is not very clear. 
She acknowledges one example of an iterated absolutive in the Jai- 
minlya-Brahmana which I had brought to her attention, then states 
that this 'is an absolutive construction which even Panini does not 
exclude from his treatise', and then continues, 'It is the reduplicated 
absolutives that he does not discuss at all' (151). Perhaps the latter 
sentence should read 'It is the reduplicated VERBAL ADVERBS that he 
does not discuss at all', assuming that A. makes a distinction between 
reduplicated absolutives and other 'reduplicated verbal adverb' 
structures. 



Hock: ReviewofAbbi (1992), Reduplication in South Asian languages 187 

Such a distinction might make sense for modern Indo-Aryan 
languages like Hindi; cf. e.g. A.'s example (22) on p. 45 vs. (37) on p. 
51. However, to judge by the Tamil, Telugu, and Kannada data, the 
corresponding Dravidian structures only show past (or perfective) 
ABSOLUTIVES; and similarly, the Dravidian structures corresponding to 
modern Indo-Aryan adverbial present-participle constructions like 
(3c) above have present (or imperfective) absolutives. Moreover, 
while Sanskrit does have locative (or genitive) absolute constructions 
involving quasi-adverbial case forms of participles, the absolutive is 
the main verbal adverb of Sanskrit. And the functions of Sanskrit 
iterated-absolutive structures are very much the same as A.'s mod- 
ern verbal-adverb constructions, no matter what specific adverbial 
form of the verb they might employ. In the subsequent discussion, 
therefore, I treat Vedic Sanskrit iterated absolutive structures (and 
other structures with iterated non-finite verbs) as relevant for eval- 
uating A.'s conclusions. (Note incidentally that Panini does make an 
explicit reference to iterated absolutive structures, but in the section 
on absolutives (3.4.22), not in the amredita section.) 

5 As in a number of other places, A.'s discussion of this matter is 
somewhat confusing: She first claims a special affinity between Old 
Iranian and Munda in respect to morphological reduplication, adding 
that 'At his juncture ... one cannot be very sure whether such a pa- 
rallel can be established for Vedic Sanskrit' (§8.3.1). But in the next 
section (§8.3.2) she points to 'structural parallels between Proto- 
Munda reduplicated verbs ... and Proto IE reduplicated verbs', and 
cites (Vedic) Sanskrit examples to illustrate the Indo-European pat- 
tern. 

6 For obvious reasons these comments are confined to areas 
with which I am familiar. It is to be hoped that A.'s book will receive 
similar comments from experts in other relevant areas. 

1 The following discussion does not make any claims to origin- 
ality but simply summarizes characteristics that are generally well 
known and have been widely reported and discussed in the liter- 
ature. 

8 For this type of structure see Hoffmann 1960. 

^ Panini does not appear to have considered this type a com- 
pound: He does not account for it in the compound section (2.1.1- 
2.38), but in a completely different part of his grammar (8.1.1-8, 57, 
2.95, 103, 3.12, 49, plus 3.4.22). On the other hand, the padapatha of 
the Rig-Veda analyzes structures of this type as compounds. 

10 One early exception is Rig- Vedic stuhi stuhi _ id 'praise, 
praise indeed' (RV 8.1.30). Here the emphasizer id might be respon- 
sible for the accentuation of the second element. But this explanation 
does not apply to the other certain examples with accent on both 



188 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

elements of the iterated structure, vdsat vdsat ... ndmo ndm(o) 'hail, 
hail ... honor, honor' (RV 10.115.9) and dranydny dranydny 'O wild 
woman, O wild woman' (RV 10.146.1). Wackernagel & Debrunner 
(1957:144), listing only the last example, wonder whether the single- 
accent pattern may not be a relatively recent innovation for these 
structures. Interestingly, all of these exceptions are exclamatory, 
either imperatives (8.1.30), or sacrificial exclamations (10.115.9), or 
vocatives (10.146.1). 

1 1 Elsewhere, the particles can follow the first word, rather than 
just the first constituent; cf. e.g. Hock 1982, Schaufele 1990. 

12 Wackernagel & Debrunner (1957:146-147) map out several 
ways in which Sanskrit structures of the type (3), with lexical itera- 
tion, can become iterative compounds of the type (2). Similarly, 
Panini (8.1.9-15) lists certain iterative compounds as special types of 
amredita, i.e. of lexical iteration. As Hoffmann (1960) points out, the 
quasi-conjoined type (9) above likewise has furnished iterative com- 
pounds in Vedic, a pattern which continues into Middle Indo-Aryan. 

13 Cf. Hock In Press on this difficult issue. 
i"* See also note 4 above. 

15 No single publication known to me gives all the examples. 
The following is a composite listing from Delbriick 1900:144, 148, 
Bartholomae 1907:166, Wackernagel 1929-30:395, Mayrhofer 1992: 
34-35 with references. (Duchesne-Guillemin's listing (1936:43) is dis- 
appointingly incomplete.) For complete citations see Bartholomae 
1904:svv. The examples in (i) are nominal iterations; those in (ii) 
iterations of interrogative pronouns used as indefinites (with or 
without the indefinitizing particle cit)\ (iii) an iterated demonstrative 
pronoun used as indefinite (for parallels in Sanskrit and other Indo- 
European languages see Brugmann 1904:130-131); (iv) what appears 
to be an iterative COMPOUND of a numeral, rather than lexical itera- 
tion; and (v) an iterated noun which has been reinterpreted and uni- 
verbated as an adverb, apparently in Proto-lndo-Iranian. 

(i) nmane nmane vTsi vTsi 'in home after home, clan after clan' 

narim naram 'man after man' 

man5 mano 'attack after attack' 
(ii) kafjhe kaghe apayzaire 'in each gully' 

kam kam cit aipi nmane 'with everyone in his house' 

(Bartholomae 1904:425-426 lists three similar examples) 
(iii) aem aem 'anyone' 

(iv) baevara baevaranam 'myriads after myriads' 
(v) nana 'separately', Skt. nana (id.) < *na na 'man after man' 

16 Modern Ossetic has a fair number of iterative compounds; cf. 
Abaev 1964:116. Some of these, such as raxcec-raxcec 'pulling' are 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 189 

listed as 'verbal'; and to judge by the gloss 'while hobbling', one 
example, k'uilix-k'uilix, looks like an iterated verbal-adverb struc- 
ture. However, no context is given, and no indication as to how fre- 
quently such structures are used. Independent evidence for Turkic 
influence on Ossetic (Thordarson 1989:457) makes it at least possible 
that the Ossetic structures result from contact with Turkic. 

17 Modern Lithuanian (Senn 1966:450, 480, 488) and Russian 
(Hirt 1937:50) use finite-verb iteration in iterative or continuative 
function. Perhaps this is an innovation due to convergence with 
Uralic. 

'8 Similar structures are found in Vogul (CoUinder 1965:336- 
338 passim) and Kamassian (ibid. 508). Some of the other Uralic lan- 
guages use similar structures with (borrowed) coordinating conjunc- 
tions. 

1^ To these similarities must be added the evidence of echo for- 
mations etc. discussed in note 2 above. 

TEXTUAL ABBREVIATIONS 

MS = MaitrayanT Samhita RV = Rig-Veda 

SB = Satapatha-Brahmana 

REFERENCES 

ABAEV, v. 1. 1964. A grammatical sketch of Ossetic, transl. by S. P. 

Hill. The Hague: Mouton. 
BARTHOLOMAE, Christian. 1904. Altiranisches Worterbuch. StraBburg: 

Triibner. 
1907. Review of Wackernagel 1905. Anzeiger fiir Indogermani- 

sche Sprach- und Altertumskunde 20.162-172. 
Beythan, Hermann. 1943. Praktische Grammatik der Tamilsprache. 

Leipzig: Harrassowitz. 
Brugmann, Karl. 1904. Die Demonstrativpronomina der indogermani- 

schen Sprachen. (Abhandlungen der Konigl. Siichsischen Gesell- 

schaft der Wissenschaften, philologisch-historische Klasse, 22:6.) 

Leipzig: Teubner. 
1907. Die distributiven und kollektiven Numeralia der indoger- 

manischen Sprachen. (Abhandlungen der Konigl. Sachsischen 

Gesellschaft der Wissenschaften, philologisch-historische Klasse, 

25:5.) Leipzig: Teubner. 
COLLINDER, Bjorn. 1965. An introduction to the Uralic languages. Ber- 
keley & Los Angeles: University of California Press. 
COLLITZ, Hermann. 1882. Ueber eine besondere Art vedischer Com- 

posita. Verhandlungen des Fiinften Internationalen Orientali- 

sten-Congresses, 2:2.287-298. 



190 Studies in the Linguistic Sciences 23:l(Springl993) 

DelbruCK, Bertold. 1888. Altindische Syntax. (Syntaktische Forschun- 
gen, 5.) Halle: Waisenhaus. (Repr. 1968, Darmstadt: Wissen- 
schaftliche Buchgesellschaft.) 

1900. Vergleichende Syntax der indogermanischen Sprachen, 3. 

(= Vol. 5 of K. Brugmann & B. Delbriick's GrundriB der verglei- 
chenden Grammatik der indogermanischen Sprachen.) StraB- 
burg: Triibner. 

Dressler, Wolfgang. 1968. Ved. dive-dive und die idg. Iterativkom- 
posita. Studien zur Sprachwissenschaft und Kulturkunde: Ge- 
denkschrift fiir Wilhelm Brandenstein ..., ed. by M. Mayrhofer et 
al., 39-47. (Innsbrucker Beitriige zur Kulturwissenschaft, 14.) 
Innsbruck. 

DUCHESNE-GUILLEMIN, Jacques. 1936. Les composes de I'avesta. Liege 
& Paris: Faculte de Philosophie et Lettres/Librairie E. Droz. 

DUNKEL, George. 1981a. Amredita and iteration of preverbs in Vedic 
and Hittite. Zeitschrift fiir vergleichende Sprachforschung 95. 
214-226. 

1981b. Further traces of preverbal amredita in Greek and Latin. 

Zeitschrift fiir vergleichende Sprachforschung 95.226-231. 

Emeneau, Murray B. 1954. Linguistic prehistory of India. Proceedings 
of the American Philosophical Society 98.282-292. 

1956. India as a linguistic area. Language 32.3-16. 

.1962. Bilingualism and structural borrowing. Proceedings of the 

American Philosophical Society 106.430-442. 

. 1980. Language and linguistic area: Essays selected by A. S. Dil. 

Stanford, CA: University Press. 

Grierson, J. H. 1921. Linguistic survey of India, vol. 1:1. Calcutta. 
Repr. 1968, Delhi: Motilal Banarsidass. 

Heston, Wilma L. 1980. Some areal features: Indian or Indo-lranian? 
International Journal of Dravidian Linguistics 10.180-187. 

HIRT, Hermann. 1937. Indogermanische Grammatik, 7. Heidelberg: 
Winter. 

HOCK, Hans Henrich. 1975. Substratum influence on (Rig-Vedic) Sans- 
krit? Studies in the Linguistic Sciences 5:2.76-125. 

1982. Clitic verbs in PIE or discourse-based verb fronting? 

Sanskrit sd hovdca gdrgyah and congeners in Avestan and Ho- 
meric Greek. Studies in the Linguistic Sciences 12:2.1-38. 

. 1984. (Pre-)Rig-Vedic convergence of Indo-Aryan with Dravi- 
dian? Another look at the evidence. Studies in the Linguistic 
Sciences 14:1.89-107. 

1985. Yes, Virginia, syntactic reconstruction is possible. Studies 

in the Linguistic Sciences 15:1.49-60. 

. 1986/1991. Principles of historical linguistics. 2nd rev. edition, 

1991. Berlin: Mouton de Gruyter. 

. 1988a. Historical implications of a dialectological approach to 

convergence. Historical dialectology, ed. by J. Fisiak, 283-328. 
Berlin: Mouton de Gruyter. 

1988b. Review article: Finiteness in Dravidian. (Steever 1988.) 

Studies in the Linguistic Sciences 18:2.211-31. 



Hock: Review of Abbi (1992), Reduplication in South Asian languages 191 

1989. Conjoined we stand: Theoretical implications of Sanskrit 

relative clauses. Studies in the Linguistic Sciences 19:1.93-126. 

. 1992. Reconstruction and syntactic typology: A plea for a differ- 
ent approach. Explanation in historical linguistics, ed. by G. W. 
Davis & G. K. Iverson, 105-121. Amsterdam & Philadelphia: 
Benjamins. 

. In Press. Chronology or genre? Problems in Vedic syntax. Pro- 
ceedings of the International Vedic Workshop, Harvard Uni- 
versity, June 1989, ed. by M. Witzel. 

HOFFMANN, Karl. 1960. Der vedische Typus menamenam. Zeitschrift 
fiir vergleichende Sprachforschung 76.242-248. (Repr. 1975, in 
Karl Hoffmann, Aufsatze zur Indoiranistik, ed. by J. Narten, 
Wiesbaden: Reichert.) 

HOFMANN, J. B. 1951. Lateinische Umgangssprache. 3rd ed. Heidelberg: 
Winter. 

. 1965. Lateinische Syntax and Stylistik, rev. by Z. Szantyr. Miin- 

chen: Beck. 

KRAUSE, Wolfgang, & Werner Thomas. 1960. Tocharisches Elementar- 
buch, 1. Heidelberg: Winter. 

Kuiper, F. B. J. 1967a. The genesis of a linguistic area. Indo-Iranian 
Journal 10.81-102. (Repr. in International Journal of Dravidian 
Linguistics 3.135-153 (1974).) 

. 1967b. The Sanskrit nom. sing. v/7. Indo-lranian Journal 10.103- 

125. 

. 1991. Aryans in the Rigveda. (Leiden Studies in Indo-European.) 

Amsterdam/Atlanta, GA: Rodopi. 

Lewis, G. L. 1953. (Teach yourself) Turkish. Kent: Hodder and Stough- 
ton. 

. 1988. Turkish grammar. Revised edition. Oxford/New York: Ox- 
ford University Press. 

MACDONELL, A. 1916. Vedic grammar for students. London: Oxford 
University Press. (Repr. 1955, Bombay, Calcutta, Madras: Oxford 
University Press.) 

MasiCA, Colin P. 1976. Defining a linguistic area: South Asia. Chicago: 
University of Chicago Press. 

Mayrhofer, Manfred. 1992. Etymologisches Worterbuch des Altindo- 
arischen, 2:11. Heidelberg: Winter. 

MEID, Wolfgang. 1968. Zum Dvandva-Kompositum im Irischen. Studi- 
en zur Sprachwissenschaft und Kulturkunde: Gedenkschrift fiir 
Wilhelm Brandenstein ..., ed. by M. Mayrhofer et al., 107-108. 
(Innsbrucker Beitriige zur Kulturwissenschaft, 14.) Innsbruck. 

Menges, Hermann. 1953. Repetitorium der lateinischen Syntax und 
Stilistik, rev. by A. Thierfelder. Repr. 1973, Darmstadt: Wissen- 
schaftliche Buchgesellschaft. 

RAMANUJAN, A. K., & Colin Masica. 1969. Toward a phonological typol- 
ogy of the Indian linguistic area. Current trends in linguistics, 
ed. by T. Sebeok, 5.543-577. The Hague: Mouton. 



192 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

SCHAUFELE, Steven. 1990. Free word order syntax: The challenge from 
Vedic Sanskrit to contemporary formal syntactic theory. Ur- 
bana: University of Illinois Ph.D. dissertation in Linguistics. 

SCHWYZER, Eduard. 1939. Griechische Grammatik, 1. Miinchen: Beck. 

1949. Griechische Grammatik, 2, completed by A. Debrunner. 

Miinchen: Beck. 

SENN, Alfred. 1966. Handbuch der litauischen Sprache, 1. Heidelberg: 
Winter. 

Steever, Sanford B. 1988. The serial verb formation in the Dravidian 
languages. (MLBD Series in Linguistics, 4.) Delhi: Motilal Banarsi- 
dass. 

THOMASON, Sarah Grey, & Terrence Kaufman. 1988. Language contact, 
creolization, and genetic linguistics. Berkeley & Los Angeles: 
University of California Press. 

Thordarson, Fridrik. 1989. Ossetic. Compendium linguarum iranica- 
rum, ed. by R. Schmitt, 456-479. Wiesbaden: Reichert. 

Tikkanen, Bertil. 1987. The Sanskrit gerund: A synchronic, diachron- 
ic, and typological analysis. (Studia Orientalia, 62.) Helsinki: Fin- 
nish Oriental Society. 

Wackernagel, Jacob. 1905. Altindische Grammatik, 2:1. Gottingen: 
Vandenhoeck & Ruprecht. 

1929-30. Altindische Grammatik, 3. Gottingen: Vandenhoeck & 

Ruprecht. 

, & Albert Debrunner. 1957. Altindische Grammatik, 2:1. [Second 

edition of Wackernagel 1905.] Gottingen: Vandenhoeck & Rup- 
recht. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 



REVIEW 

Narindar K. Aggarwal. Studies on Nepali language and linguis- 
tics: A bibliography. (Subject Bibliography Series, 15.) 

Gurgaon (India): Indian Documentation Service, 1991; pp. xiv, 93. 

Mithilesh K. Mishra 

Studies on Nepali language and linguistics (SNLL) is undoubtedly 
the most comprehensive and useful collection of resource materials 
(both primary and secondary) on Nepali or Gurkhali and Newari that 
has been published to this date. The author has rendered the 
researchers working in various areas related to Nepali and Newari an 
extremely valuable service by compiling at one place various source 
materials on these languages that remain scattered over the three 
continents, especially in the U.S.A., South Asia (Nepal and India), and 
the U.K. SNLL has 726 unannotated entries which include books, 
doctoral dissertations, master theses, and published and unpublished 
papers on various aspects of Nepali and Newari languages and 
linguistics (written/published in English, Hindi, Nepali and Newari). 
The entries are classified under eleven headings: (1) Bibliographies, 
dictionaries and glossaries (pp. 1-8), (2) Studies on Nepali and Newari 
(pp. 9-13), (3) Nepali and Newari grammars (Traditional) (pp. 14-22), 
(4) Descriptions of Nepali (syntax/semantics and morphology) (pp. 
23-29), (5) Descriptions of Newari (syntax/semantics and morphol- 
ogy) (PP- 30-35), (6) Phonetics and phonology (pp. 36-41), (7) Lex- 
icography and lexicology (pp. 42-47), (8) Applied Nepali linguistics: 
stylistics, socio-linguistics (pp. 48-51), (9) Pedagogical material (pp. 
52-58), (10) Historical, comparative and dialectological studies (pp. 
59-76), and (11) Nepali and language planning (pp. 77-79). Also 
included is an author index that provides cross references to the 
entries. Aggarwal's categorization of various entries is very system- 
atic, exhaustive, and useful. Especially for teachers and students of 
Nepali and Newari languages, this monograph brings to light a rich 
source of pedagogical material both in English and in Nepali and 
Newari. 

The bibliography is highly commendable for a variety of rea- 
sons. First of all, by publishing this monograph, the author has recti- 
fied an overly outdated practice of putting together widely divergent 
pieces of bibliographic information under one title such as 'A Bibli- 
ography of Nepal' (Scarecrow Press, 1973), or 'Bibliography of Nepal' 
(Royal Nepal Akademy, 1975). In the present age of information the 
need for a comprehensive and yet very focussed and user-oriented 
body of information cannot be overemphasized. SNLL meets these 
criteria of an excellent bibliography in exemplary fashion. Secondly, 
this bibliography, by consolidating information on scholarly work on 
Nepali and Newari languages and linguistics, has established an 



194 Studies in the Linguistic Sciences 22:1 (Spring 1993) 

academic bridge between the traditional scholars of Nepal and India 
and the western scholars in general. Third, the author of SNLL, by 
mapping out the entire field of scholarship on Nepali and Newari 
languages and linguistics, not only reminds us of the existence of this 
potentially fruitful but utterly neglected area of research on both 
sides of the Atlantic, but also challenges us to explore the entire field 
anew. 

Because of its comprehensive coverage of different types of re- 
source materials, SNLL is definitely going to be a part of the 'tool-kit' 
of every brand of linguist working on Nepali and Newari languages in 
particular and of scholars of Nepali and Newari in general. 



Studies in the Linguistic Sciences 
Volume 23, Number 1, Spring 1993 

REVIEW 

John Baldwin & Peter French. Forensic phonetics. London & New 
York: Pinter, 1990; pp. viii, 141. $47.50. 

Jose Ignacio Hualde 

Even though this book has two authors, it is not a coauthored 
book in the sense that the two authors claim joint responsibility for 
the text. Rather, we are told in the Foreword that chapter 3 was 
written by French, and all other chapters by Baldwin. The reader 
who skips the Foreword will experience some puzzlement since the 
book is not only written in first person singular, but the two authors 
disagree as to the preferable way to practice forensic phonetics. 

The book is intended for the interested layman. It contains six 
chapters. In chapter 1, 'Introduction', and chapter 4, 'Aspects of 
forensic phonetics', we are given information on the work of forensic 
phoneticians, and, in particular, forensic phoneticians working within 
the English judiciary system. 

Forensic phoneticians are those professionals whose job it is to 
serve as experts in legal cases in the identification of speakers by 
analyzing and comparing samples of their speech. Typically, a foren- 
sic phonetician is consulted when the police are in the possession of 
an incriminating recording and either have a suspect whose speech 
they want to have compared to that on tape, or want to obtain 
information on the features of the taped speech that would allow 
them to narrow the range of possible suspects. A forensic phonetician 
may also be consulted by the defense in order to disprove that a 
recorded voice is that of the defendant. 

Chapter 2, 'Phonetics', provides a very brief and rather super- 
ficial overview of this field, but intercalating a number of anecdotes 
which make it quite readable. Chapter 3, 'Acoustic phonetics', deals 
with those aspects of speaker identification where instrumental 
analysis offers advantages over a purely auditory examination in the 
opinion of the author (French). The chapter includes narrations of a 
number of actual cases where acoustic analysis was crucial in the 
identification of the speaker. Precisely the main difference of opinion 
between the two authors is that, whereas Baldwin has faith in 
identification by auditory means, French favors an acoustic/auditory 
methodology which combines identification by ear with instrumental 
analysis. 

Chapter 5, 'Cases', is the most entertaining one of the book. In it, 
Baldwin relates some of the most interesting cases in which he was 
consulted as an expert. As a linguist, the case which I found of 
greatest interest is one whose complicating factor was that it 
involved accent-switching on the part of the suspect. Baldwin distin- 



196 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

guishes accent-switching from the behavior of someone who puts on 
an accent different from his or her own in order to deceive. Accent- 
switchers are in command of two accents, both of them equally real 
and natural to them, and can voluntarily switch from one accent to 
the other. These are, thus, bilingual speakers whose two linguistic 
codes happen to differ solely or mostly on phonetic characteristics. 
The subject in the case was a native of Northern Ireland who had full 
command of two varieties of English spoken in that region, which are 
referred to as Scots Irish and Irish Irish. Baldwin offers the following 
comparison of the two varieties: 

Some salient features of [Scots Irish], which distinguish 
it from Irish Irish, are the widespread use of the glottal 
stop, even to reinforce voiced consonants as in 'Radley' as 
opposed to its non-occurrence; the use of a back rounded 
diphthong in words like 'cold', 'go', etc., as opposed to the 
Irish Irish front rounded diphthong; the correct pronun- 
ciation of dental fricatives, as opposed to their realization as 
dental stops, and so on. [p. 105] 

Apparently the subject had total control over these features and 
could switch from one accent to the other depending on who the 
interlocutor was. 

In Chapter 6, 'The future', Baldwin discusses two topics. One is 
the feasibility of computer-generated 'voice-prints' in a not too 
distant future. He believes the technology will be available. The 
second topic is some reforms that he would like to see implemented 
in the English judiciary system. Baldwin has a rather negative 
opinion of the adversarial system, as it is practiced in England, which 
opinion he voices in several places throughout the book. His strongest 
complaint is that many people involved in such a legal system are 
purely interested in winning, and not in determining the truth. 

Forensic phoneticians who, like other experts, work for one of 
the two parties in the adversarial system (the defense or the 
prosecution), are not always able to present all the evidence that 
they have laboriously obtained and which could be crucial for finding 
the truth in the case. For instance, if they are working for the 
defense and their results are against the interest of the defendant, 
they will be suppressed. Baldwin firmly believes that forensic 
phoneticians should be impartial experts and proposes that experts 
should work for the court, not for one of the two parties. 

The book is very readable and offers interesting insights in the 
work of phoneticians working in the judiciary system. Although, its 
interest for academic linguists is only tangential, it could be a very 
interesting book for students considering careers involving phonetics. 



Studies in the Linguistics Sciences 

Volume 23, Number 1, Spring 1993 

RECENT BOOKS 

Studies in the Linguistic Sciences gratefully accepts review copies of 
recent publications and tries to find reviewers for them. In this en- 
deavor, however, it does not always succeed. Volumes for which no 
reviewers have been found so far are publicized in this section, with 
brief indications of contents or interest. Prices are indicated where 
known. (Unless otherwise indicated, the brief descriptions below are 
by the editor.) 

Carol Lord. Historical change in serial verb constructions. 

(Typological Studies in Language, 26.) Amsterdam/Philadelphia: John 
Benjamins Publishing Company, 1993, pp. x, 273. 

The term 'serial verb' has been used in a variety of different, often con- 
flicting meanings. Lord (L) concentrates on a type of 'constructions in which 
one of the verbs is defective in some respect, such as phonological assimila- 
tion, failure to take the usual verb inflections or negation affixes, or showing 
unexpected syntactic properties, for example with respect to movement.' (3) 
She claims that verbs of this type exhibit strong crosslinguistic similarities 
both in their original meanings 'and in the functions they come to mark'. (3) 
L posits a 'directional' change through 'bleaching' or 'desemanticization', such 
as verb -> preposition -> case-marking affix and similar developments leading 
to the development of auxiliaries, adverbs, and conjunction-like elements 
(complementizers, quotative markers, adverbial subordinators, etc.). The major 
focus of the book is on West African languages. The wealth of data from these 
languages, much of which is based on L's own research, is supplemented by 
evidence of other languages, especially Caribbean Creoles and Mandarin 
Chinese. 

Chapter 1 presents a general discussion on 'Serial verbs'. Chapters 2 - 8 
are devoted to specific developments: Locative verbs -> prepositions (2); verbs 
-> recipient/benefactive marking (3); comitative verbs -^ prepositions and 
conjunctions (4); verbs -^ object markers (5); developments of special prepo- 
sitions (such as 'except' or 'more than') (6); developments of complementizers 
and subordinating conjunctions (7); and verbs -» adverbs and auxiliaries. 
Chapter 9 ('Pragmatics, typology, and teleology') addresses the implications of 
L's findings, as well as 'The "why" question'. 

Concerning the latter question, L opts for the view that grammat- 
icalization is attributable to two factors: 'the process of phonological erosion of 
grammatical forms, and the human imperative to be expressive resulting in 
the utilization of metaphor.' (249) What is missing in this account is the ele- 
ment of 'bleaching' or 'desemanticization' which L appeals to in the preface. 
One suspects that it is this semantic process of bleaching which leads to accen- 
tual 'downgrading' (including cliticizalion) and that this downgrading, in 
turn, is responsible for the fact that these verbs undergo much more radical 
'phonological erosion' than non-grammaticalized verbs. 

Morteza Mahmoudian. Modern theories of language: The em- 
pirical challenge. (Sound and Meaning: The Roman Jakobson Series 
in Linguistics and Poetics.) Durham & London: Duke University Press, 
1993, pp.xvii, 231. 

The goal of this book is to critically examine twentieth-century theories 
of language and to argue for the need to confront them with empirical evi- 



198 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

dence. Mahmoudian (M) claims that differences in terminology mask princi- 
ples shared by most linguists. These are the distinction between unit and 
system; concerns with taxonomy, rules (or other types of generalization), and 
explanatory adequacy; the issue of linguistic universals; the distinction 
between 'signifie' and 'signifiant'. 

The scope of M's coverage can perhaps most effectively be gauged by 
examining his references and citations: The four most recent references are 
from 1989 and 1988. All of these and 17 out of some 24 other citations since 1980 
can be broadly classified as being semiotic and/or as adhering to Martinet's 
functionalism. Although Chomsky is referred to most often (on nine occasions 
vs. seven each for Martinet and Hjemslev), his most recent publication re- 
ferred to by M is the 1977 French version of his presentation of the 'Extended 
Standard Theory'. The most recent publications by US linguists are a 1978 book 
by Ebeling in the Taxonomic framework and Greenberg's 1978 series on 
'Universals of human language'. 

P. A. Messelaar. La confection du dictionnaire general bi- 
lingue. Leuven: Peelers, 1990, pp. 109. 

This book is intended as an aide-memoire for the director of a lexico- 
graphical project ad as a manual for the beginning bilingual lexicographer. 
The introductory section discusses background issues, including the general 
difficulties associated with translation. The second section outlines the history 
of bilingual dictionary making. Sections 3 through 17 address specific issues 
confronting the compiler of a bilingual dictionary: The general organization 
of the dictionary (3); the required 'note to the reader' (4); the selection of en- 
tries (5); the nature of the lexicographical entry (6); the structure of the entry 
(7); the phonetic transcription (8); the typography to be employed (9); gram- 
matical information (10); information on usage (11); semantic information and 
additional information helpful in translation (12); synonymy (13); semantic 
information required to assure translational equivalency (14); issues connect- 
ed with idiomatic usage and restrictions on collocability (15); the extent to 
which non-lexical (primarily syntactic) information needs to be included (16); 
stylistic issues (17). Except for M's own work, three monolingual dictionaries, 
and one bilingual one, the bibliographical references predate 1980. 

Kenneth L. Pike. Talk, thought, and thing: the emic road to- 
ward conscious knowledge. Dallas: Summer Institute of Linguis- 
tics, 1993, pp. xii, 85. 

'This book is written for a small number of people unknown to me, who 
are disillusioned in a changing world' (vii). Pike (P) suggests that a source for 
hope may lie in 'looking into the nature of language', since language serves as 
the basic means of communication of all people (viii). Although addressing 
these issues as a linguist (within the Taxonomic framework), P emphasizes the 
central role of the 'person' and professes to 'have been helped ... by several 
philosophers', including Quine, Mavrodes, and Reeder (ix). He concludes that 
'A person grows in knowledge through the intersection of networks of pat- 
terns of patterns of ... phonological, grammatical, and referential hierarchies' 
and that 'A person needs his language to help him know himself in relation to 
his physical, social, aesthetic, and philosophical environment' (78). 

V. Prakasham ad S. V. Parasher, eds. Linguistics at large: Papers 
in general and applied linguistics. Hyderabad: Booklinks 
Corporation, 1991, pp. xvii, 191. 

Linguistics at large is a collection of papers in honor of the sixtieth 
birthday of Shivendra K. Verma, Director of the Central Institute of English 



Recent Books 199 

and Foreign Languages (CIEFI,), Hyderabad (India). In addition to the editors' 
preface and a list of Verma's publications, the volume contains sixteen articles, 
half of which are by members of the CIEFL. 

Eight papers address issues directly or indirectly connected with the 
CIEFL's primary focus on language pedagogy, especially as regards English. 
Especially interesting and provocative among these is K. Annamalai's 'When 
an adult learns his mother tongue', an article examining the special difficul- 
ties encountered by adults who have 'lost' — or even never learned — the 
language of their community in trying to learn that language as adults, as part 
of a kind of ethnic revival. 

The remaining papers address issues of the nature of 'Modern 
[generative] grammar' (K. A. Jayaseelan); 'Contemporary syntactic theories 
and traditional Indian syntax' (B. N. Patnaik); 'Length as a formative prosody 
in Telugu' (V. Prakasam); 'Schwa fronting in Hindi' as a variable-rule pheno- 
menon (P. K. Pandey); 'Addressee markers on verbs' in colloquial Tamil as a 
problem for theories of agreement (R. Amrilavalli); 'Bhartrhari [the eminent 
indigenous philosopher of language] on lexical meaning' (K. Kapoor); 'The 
sociolinguistics of code-switching' in Hindi and English (A. Kumar); and 
'Modernization of Sanskrit', mainly a study of lexical borrowing throughout 
the history of Sanskrit (H. S. Anantanarayana). Some of the papers have 
appeared elsewhere. 

The volume is well produced, except for a fair number of self-correcting 
misprints and the more serious fact that diacritics and special symbols fre- 
quently are missing or misprinted. 

F. de Saussure. Troisieme cours de lingiiistique generale 
(1910-1911) d'apres les cahiers d'Emile Constantin — Saus- 
sure's Third course of lectures on general linguistics (1910- 
1911): From the notebooks of Eniile Constantin. French text 
edited by Eisuke Komatsu; English translation by Roy Harris. Oxford, 
New York, Seoul, Tokyo: Pergamon Press, 1993, pp. xxiii, 173 (-H143). 

The notes on de Saussure's third — and last — cour.se in general linguistics 
(TCLG) which his student Emile Constantin look in 1910-19II were not avail- 
able to the editors of the famous set of de Saussure's lecture notes, Cours de 
lingiiistique generale (1916). Their existence became known only in 19.S8. (As 
noted in the foreword of TCLG, the editors of the Cours claimed to have drawn 
on de Saussure's third course, but the notes that they consulted were much less 
complete than Constantin's.) 

The importance of Constantin's notes was realized almost immediately 
after becoming available: They seemed to offer a much more reliable and di- 
rect access to de Saussure's mature thinking on general linguistics than the 
recast and transformation of his lecture notes at the hands of the editors of the 
Cours, Bally and Sechehay. Especially noteworthy are a classification of speech 
sounds in terms of six degrees of buccal opening covering both consonants 
and vowels (reminiscent of classifications by the indigenous phoneticians of 
ancient India as well as recent work by Clements) and what in the Foreword of 
TCLG is referred to as de Saussure's 'most important theoretical pronounce- 
ments about la langue'. Among the latter are such statements as 'The language 
[i.e. la langue. as distinct from langage] is located only in the brain' (69a) and 
'Strictly speaking there are no signs but differences between signs' (14 1 a), as 
well as an account of synchronic laws as distinct from diachronic ones and 
remarks on the often indirect relation between the two types of laws (115a- 
II8a). 



200 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Because of their significance, Constantin's notes were drawn on exten- 
sively by Rudolf Engler in his 1968 critical edition of the Cours; but the 
structure of that edition remained that of Bally and Sechehay's. TCLG presents 
the notes in their original structure and sequence and thus permits the reader 
to judge them in their own right. A Foreword, a note on 'The Constantin Note- 
books', and a 'Translator's Preface' are followed by a bilingual presentation of 
the text, with an edited version of the French original on the left (pp. 1-143) 
and the F.nglish translation of that version on the right (pp. la-143a). The 
book concludes with a 'Selective index of French terminology' (145-173). 



CONTENTS OF STUDIES IN THE LINGUISTICS SCIENCES, 
VOLUMES 18 - 22 

Volume 18:1 (Spring 1988) 

Papers in General Linguistics 
(Edited by Michael J. Kenstowicz) 

Carreira, Maria. The representation of diphthongs in Spanish (1-24) 

Downing, Laura J. Tonology of noun-modifier phrases in Jita (25-60) 

Irshied, Omar, & Peter Whelan. Exploring the dictionary: On teaching foreign learners of 

Arabic to use the Arabic-English dictionary (61-75) 
Kenstowicz, Michael J., Emmanuel Nikiema, & Meterwa Ourso. Tonal polarity in two 

Gur languages (77-103) 
Moshi, Lioba. A functional typology of ni in Kivunjo (Chaga) (105-134) 
Patterson, Trudi A. Some morphological and phonological interactions in Lakhota (135- 

149) 
Wong-opasi, Uthaiwan. On deriving specifiers in Spanish: Morpho-phono-syntax inter- 
actions (151-177) 

Volume 18:2 (Fall 1988) 

Papers in General Linguistics 
(Edited by Hans Henrich Hock) 

Abu-Salim, Issam M., & Hassan R. Abd-el-Jawad. Syllable patterns in Levantine Arabic 
(1-22) 

Cervin, Richard. On the notion of 'second position' in Greek (23-39) 

Gerdemann, Dale, & Erhard W. Hinrichs. UNICORN: A unification parser for attribute- 
value grammars (41-86) 

Kamwangamalu, Nkonko Mudipanu. 'C-command' and the phonology-syntax interface in 
Ciluba (87-109) 

Ourso, Meterwa A. Root control, underspecification, and ATR harmony (1 1 1-127) 

Schiiufele, Steven. Where's my NP? Non-u-ansformational analy.ses of Vedic pronominal 
fronting (129-162) 

Tsiang, Sarah. The discourse function of the absolutive in the Pancatantra (163-181) 

Zhou, Xinping. On the head movement constraint (183-210) 

Hock, Hans Henrich. Finiteness in Dravidian (Review article): Sanford B. Steever 
(1988), The serial verb formation in the Dravidian languages (21 1-233) 

Volume 19:1 (Spring 1989) 

Papers in General Linguistics 
(Edited by Hans Henrich Hock) 

Arora, Harbir, & K. V. Subbarao. Convergence and syntactic reanalysis: The case of so 

inDakkhini(l-18) 
Bundrick, Camille. An inference-based account of restrictive relative which and that (19- 

31) 
Cassimjee, Farida, & Charles W. Kisseberth. Shingazidja nominal accent (33-61) 
Chung, Raung-fu. On the representation of Kejia diphthongs (63-80) 
Gerdemann, Dale. Restriction as a means of optimizing unification parsing (81-92) 
Hock, Hans Henrich. Conjoined we stand: Theoretical implications of Sanskrit relative 

structures (93-126) 
Kachru, Braj B. World Englishes and applied linguistics (127-152) 



202 Studies in the Linguistic Sciences 23: 1 (Spring 1993) 

Kachru, Yamuna. Corpus planning for modernization: Sanskritization and Englishization 

of Hindi (153-164) 
Bhatt, Rakesh Mohan. Good mixes and odd mixes: Implications for the bilingual's 

grammar. (Squib) (165-168) 
Aitchison, Jean. Review of Rama Kant Agnihou-i (1987), Crisis of identity: Sikhs in 

England {l69-n I) 
Pandharipande, Rajeshwari. Review of Tej K. Bhatia (1987),i4 history of the Hindi 

grammatical tradition (173-179) 
Markee, Numa. Review of Dick Chamberlain & Robert Baumgardner (eds.) (1988), ESP 

in the classroom: Practice and evaluation (181-1 85) 
Zgusta, Ladislav. Review of Sumitra Katre (1987), AstadhyayT of Pdnini (1^1-193) 

Volume 19:2 (Fall 1989) 

The contribution of African linguistics to linguistic theory. Vol. 1 

(Edited by Eyamba G. Bokamba, Associate Editors: Rick Treece 

and Dorothy E. Evans) 

Introduction(vii-xi) 

Clements, G(eorge) N. African linguistics and its contributions to linguistic theory (3-39) 

Biloa, Edmond. Tuki gaps: Null resumptive pronouns or variables? (43-54) 

Childs, G. Tucker. Where do ideophones come from? (55-76) 

Ottenheimer, Harriet, & Heather Primrose. Current research on ShiNzwani ideophones 

(77-87) 
Downing, Laura J. Tone in Jita questions (91-113) 
Hym an, Larry M. Accent in Bantu: An appraisal (1 15-134) 
Timmons, Claude, & Christian Dunn. La selection morphophonologique des classes en 

kpokolo (135-151) 
Ali, Mohammed. Trends in Oromo lexicon and lexicography (155-168) 
BoUie, Robert. Reconstruction of a grammaticalized auxiliary in Bantu (169-186) 
Clamons, Cynthia Robb. Modification of the gender system in the Wollegan dialect of 

Oromo (187-195) 
Appendices to the Proceedings 

A. History of the Annual Conference on African Linguistics (199-201) 

B. Research and publications in African linguistics by students, alumni, and faculty of 

the University of Illinois (203-225) 

Volume 20:1 (Spring 1990) 

The contribution of African linguistics to linguistic theory. Vol. 2 
(Edited by Eyamba G. Bokamba, Associate Editors: Amy C. Cheatham, 
Dorothy E. Evans, and Rick Treece) 

Preface (v-vi) 

Inu-oduction (vii-ix) 

Bokamba, Eyamba G. African languages and sociolingui.stic theories (3-34) 

Bresnan, Joan. African languages and syntactic theories (35-48) 

Goldsmith, John. Phonological theory and African language phonology (49-62) 

Mufwene, Salikoko. African languages, African linguistics, and linguistic theory: A 

commentary on the plenai7 session papers (63-67) 
Broselow, Ellen, & Alice Niyondagara. Feature geomeU7 of Kimndi palatalization (7 1-88) 



Contents of volumes 1 8 - 22 203 

Clements, G(eorgc) N., & Remi Sonaiya. Underlying feature representation in Yoruba 

(89-103) 
Ka, Omar. Reduplication and prosodic constituents in Wolof ( 105- 121) 
Noske. Manuela. Vowel Harmony in Turkana (123-134) 

Ourso, Meterwa A., & Charles H. Ulrich. Sonorant-strenglhening in Lama (135-147) 
Kisseberth, Charles, & Sheila Mmusi. The tonology of the object prefix in Setswana 

(151-161) 
Mutaka, Ngessimo. The tone bearing unit in Kinande (163-172) 
Kapanga, Andre Mwamba. Language variation and language attitudes: A case study 

from Shaba Swahili (175-188) 
Wade-Lewis, Margaret. The contribution of Lorenzo Dow Turner to African linguistics 

(189-204) 

Volume 20:2 (Fall 1990) 

Linguistics for the Nineties: Papers from a lecture series in 

celebration of the Department's twenty-fifth anniversary 

(Editor: Hans Henrich Hock, Editorial Assistant: Lynne Murphy) 

Antonsen, Elmer H. Introduction (ix-xiv) 
Kahane, Henry. The esiabli.shment of Linguistics at Illinois ( 1 -2) 
Langacker, Ronald W. Cognitive Grammar: The symbolic alternative (3-30) 
Sadock, Jerrold M. A trimodular account of Yiddish syntax (3 1 -50) 
Newmeyer, Frederick J. Some i.ssues in language origins and evolution (51-68) 
Odden, David. Phonology and its interaction with syntax and morphology (69-108) 
Menn, Li.se. Aphasic language under discourse pressure: Functional syntax vs. psycho- 
linguistic function (109-122) 
Lowenberg, Peter H. Standards and norms for World Englishes: I.ssues and attitudes 

(123-137) 
Hermon, Gabriella. Syntactic theory and language acquisition: A ca.se against parameters 

(139-163) 
Sridhar, S. N. What are applied lingui.stics? (165-176) 
Index to Studies in the Linguistic Sciences, Volumes 1-19 

A. Author index (177-195) 

B. Title index (197-214) 

Volume 20:3 (Sprinj^ 1990) 
Meeting handbook. Thirteenth South Asian Languages Analysis 
Roundtable (Abstracts] 
(Edited by Hans Henrich Hock. Editorial Assistant: Lynne Murphy) 

Ahmed, Mariam. Convent English: Structun; and altitudes (17) 

Anderson, Lloyd B. Script Manager .software ft)r Indic .scripts on the Macinto.sh ( 18) 

Bagchi, Tisla. Conditionals and emphasi/crs in Bangla: Some pragmatic effects of their 

interaction (19-20) 
Bains. Gurpril. Focus movement in Hindi-Urdu (21) 
Bhatt. Rakcsh M. An essay on Kashmirit siivss (22-23) 
Boolchandani. Pushpa. On binding retlexives in Sindhi (24-25) 
Bubenik, Vit, & C. Paranjpe. Some observations on the development of West- Indo- Aryan 

pronominal .systems from Apabhrarh.sa (26) 
Butt, Miriam J., & Tracy Holloway King. Semantic ca.se in Urdu (27-28) 



204 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Chakraborty, Jayshree. Perfectivity and the resultalive state in Hindi (29-30) 

Chandrasekhar, S., & S. N. Sridhar. Case markers and prepositions in Kannada (31) 

Cole, Jennifer. Alliteration in Sindhi poetry: Evidence for phonological structure (32-33) 

Davison, Alice. Finiteness and case in Hindi-Urdu complements (34-35) 

Deshpande, Madhav. Sociolinguistic parameters of Panini's Sanskrit (36) 

Genetti, Carol. On the loss of gender distinctions in Nepali (37-38) 

Gnanam, M. Religious cum linguistic problems in modern India (39) 

Hamp, Eric P. The sources of a passive (40-41) 

Herring, Susan. From aspect to tense in Old Tamil: Evidence from narrative discourse 

(42-43) 
Hock, Hans Henrich. Syntax or Phonological Form? Reconsidering some allegedly 

syntactic phenomena of Vedic Sanskrit (44-45) 
Hook, Peter E., & Omkar Nath Koul. Kashmiri causals: Evidence for a transformational 

approach (46-47) 
Jamison, Stephanie W. Demonstratives with non-third persons in Vedic Sanskrit (48) 
Jayasuriya, Wilfrid. The web of the spider: Language and power in Sri Lanka (49) 
Joseph, Brian. Sibilant confusion in early Indie: Sanskrit pradiir (50-51) 
Kamwangamalu, Nkonko M. Advancement in some Asian and African languages (52) 

. Multilingualism and social identity: The case of Singapore (53) 

Kapoor, Kapil. Analogy as argument in Adi Saiikara (54) 

Kissock, Madelyn J. Reflexive pronouns in Vedic (55) 

Loud, John A. Issues in translating the Puranas (56) 

Mahajan, Anoop. Against wh-movement in Hindi (57-58) 

Mahajan, Gyanam. Sanskrit reduplication: A templatic approach (59-60) 

Marlow, Patrick E. Meet me in the Bazaar: A historical perspective on the origin of a North 

Indian koine (61) 
Mehrotra, Raja Ram. Sociolinguistics of verbal abuse in Hindi (62) 
Menon, A. G. Tamil verb formation (63-64) 
Moag, Rodney F. The associative case in Malayalam: Making sense of a catch-all category 

(65-66) 
Mohanty, Gopabandhu. Compound verbs in Oriya (67-68) 
Nadahalli, Jayashree. Pronouns in Kannada: Sociolinguistic implications (69) 
Nadkami, Mangesh V. On liberating English to be a world language: An Indian perspec- 
tive (70-71) 
Nihalani, Paroo. Articulatory and acoustic properties of apical and laminal stop 

consonants: A cross-language study (72) 
Pandit, P. N. A socio-cognitive approach to designing a self-instructional multi-media 

course in English communicative skills (73) 
Paolillo, John C. Functional articulation: Analyzing diglossic variation (74-76) 
Pelletier, Rosanne. Telugu negatives and non-capabilitives: Morphological structure and 

syntactic structure (77-79) 
Rai, Alok. Sammelani Hindi and Malviya Hindi: Language and politics in India between 

1875 and 1930(80) 
Ramchand, Gillian. The category of nominals in Bangla (81-83) 
Rau, Nalini. Coordination and word order (84) 
Sadanand, Kamlesh. The pure vowels of Punjabi (85) 
Sadanand, Suchilra. Malayalam syllabication (86) 
Satyanath, T. S. On change and variation of (1) in Kannada (87) 



Contents of volumes 1 8 - 22 205 

Scharf, Peter M. Assessing Sahara's arguments for the conclusion that a generic term de- 
notes just a class property (88) 
Schaufele, Steven. The Vedic clause-initial string and universal grammar (89) 
Sharma, Krishna K. Semio-linguistic aspect of dhvani siddhanta (90-91) 
Sharma, Rama Natha. Naming and expressing objects in Panini (92) 
Singh, Atamjit. The aesthetics of play in Punjabi folkloric tradition (93) 
Singh, Mona. A situation-type analysis of compound verbs (94-95) 
Sreedhar, M. V. Drastic modernization of the curricula of the teacher training courses (96- 

97) 
Sridhar, S. N., & Mark Aronoff. A lexicalist analysis of participle compounds in Kannada 

(98-99) 

, & Indira Ayyar. Aspects of the syntax of spoken Indian English (100) 

Srivastav, Veneeta. Pair-list answers in Hindi indirect questions (101-103) 
Subbarao, K. V., & Harbir Arora. Convergence and syntactic change: The case of the 

negative participles in Dakkhini (104-105) 
, & Lalitha M. The INFL nodes in non-finite clauses in Dravidian and Tibeto-Burman 

languages (106) 
Tickoo, Asha. New dimensions of word order freedom in verb-final languages (107) 
Tsiang, Sarah. Clausal vs. non-clausal subordination in Sanskrit narratives (108-109) 
Vijayakrishnan, K. G. The mental dictionary: Its role in linguistic theroy (110-111) 
Winters, Clyde A. The Harappan script: The most ancient form of Dravidian (112) 
Yatabe, Shuichi. Verbal compounds in Malayalam (113-114) 
Zakharyin, Boris A. Ergativity in the Indo-European languages of South Asia: Diachronic 

and synchronic processes (115-116) 
, & L. V. Khokhlova. The development of ergativity in Indo-European languages of 

Western India in the fifteenth through twentieth centuries (117-118) 
Zide, Norman. A sketchy history of cliticization and verb stem noun incorporation in 

Munda(119) 
Pandharipande, Rajeshwari. A grammar of politeness in Marathi (125) 
Mishra, Mithilesh K. Towards an ethnography of politeness in Maithili (126) 
Bhatia, Tej K. Directives in Panjabi and Lahanda (127) 

Verma, Manindra K. Linguistic conventions of politeness in Bhojpuri and Magahi (128) 
D'Souza, Jean. Recreating South Asian speech acLs in English: A study in linguistic trans- 
fer (131) 
Kachru, Yamuna. Speech act in the mother tongue and the other tongue (132) 
Nelson, Cecil L. On creating speech acts: The creativity of Indian English writers (133- 

134) 
Valentine, Tamara. Language and female identity in India (135) 
Ahmed, Mariam. A house divided: Confiict and rivali7 in two varieties of a language 

(139) 
Bhatia, Tej K. Transplanted languages and ethnic identity (140) 

Sridhar, Kamal K. Language minorities: Issues of identity in a global p)erspective (141) 
Bhatt, Rakesh M. Identity, conflict, and convergence: South Asia as a sociolinguistic area 

(142-143) 
Pandharipande, Rajeshwari. The question of defining the language of religion (147) 
Hock, Hans Henrich. Vasar, s'rau.sar, and other ritual particles: Their origin and use in 

Vedic ritualistic literature (148-149) 
Mishra, Mithilesh K. The role of deixis in defining ordinary vs. religious language (150) 
Anushivarani, Ali. Rabindranath Tagore's Nobel Prize: What does it mean? (153) 



206 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Wu, Yongan. Chinese responses to Tagore: Pin Hsin's poetry (154) 

Tikku, Girdhari. Aldous Huxley's The island {\55) 

Harada, Hiroko. Coleridge and Basho: The legacy of Indian monism (156) 

Volume 21:1 (Spring 1991) 

Papers in General Linguistics 
(Edited by Hans Henrich Hock, Editorial Assistant: Amy C. Cheatham) 

Branstine, Zoann. Stop/spirant alternations in Spanish: On the representation of contrast 

(1-22) 
Darzi, Ali. Compensatory lengthening in modem colloquial Tehran! Farsi (23-37) 
Hancin, Barbara J. On the phonology-moiphology interaction in Brazilian Portuguese 

vowel harmony (39-54) 
Kang, Seok Keun. Morale phonology and /l/-irregular predicates in Korean (55-66) 
Kang, Yongsoon. Coronal: Transparent or opaque? (67-79) 
Lee, Han-Gyu. Plural marker copying in Korean (81-105) 
Markee, Numa. Toward an integrated approach to language planning (107-123) 
Mtenje, Al. On Autosegmental feature-spreading in phonology: Evidence from Chiyao 

(125-145) 
Pandey, Pramod Kumar. Schwa fronting in Hindi (147-159) 
Hock, Hans Henrich. Review of R. N. Aralikatti (1989), Spoken Sanskrit in India: A 

study of sentence patterns ( 1 6 1 - 1 65) 
Yoon, James H. Review of Mark R. Baltin & Anthony S. Kroch, eds. (1989), Alternative 

conceptions of phrase structure (167-178) 
Conefrey, Theresa. Review of Deborah Tannen (1990), You just don't understand 

(179-181) 
Recent Books (183-188) 

Volume 21:2 (Fall 1991) 

Illinois studies in Korean linguistics, 2 

(Edited by Chin-W. Kim, Jerry L. Morgan, and James H-S. Yoon, 

Editorial Assistant: Amy C. Cheatham) 

Preface (v) 

List of dissertations on Korean between 1987-92 at the University of Illinois at 

Urbana-Champaign (vi) 
Ahn, Sang-Cheol. Vowel deletion and epenthesis: The vowel /' ( 1 - 1 8) 
Cho, Euiyon. Notes on some tests for subjecthood in Korean (19-29) 
Cho, Seikyung. The acquisition of English reflexives by Korean ESL learners (31-67) 
Choi, Yeon Hee. Discourse reference in written Korean folk tales (69-87) 
Kang, Seok Keun. Morale representation of ambi.syllabicity: Evidence from Korean 

(89-100) 
Kang, Yongsoon. The Locality Condition of tonal systems: With special reference to 

North Kyungsang dialect in Korean (101-112) 
Kim, Chin W., & Hyoung- Youb Kim. The character of Korean glides (113-1 25) 
Kim, Hyoung-Youb. Prosodic phonology of Korean (127-142) 
Lee, Han-gyu. The pragmatics of the pragmatic morpheme com 'a little' in Korean 

(143-166) 
McClanahan, Virginia K. TTie pragmatics of negation in Korean (167-178) 
Yoon, James Hye Suk. Inflectional structures in Korean and headedness (179-192) 



i 



Contents of volumes 18-22 207 

Volume 22:1 (Spring 1992) 

Papers in General Linguistics 
(Edited by Hans Henrich Hock, Editorial Assistant: Amy C. Cheatham) 

Preface (v) 

Endangered languages: An appeal for publications (vii-viii) 

Alho, Irja H. Distinguishing kind and set in Finnish (1-16) 

Bhatt, Rakesh Mohan. Language identity, conflict, and convergence in South Asia (17-37) 

Hock, Hans Henrich. What's a nice word like you doing in a place like this? 

Syntax vs. phonological form (39-87) 
Kraska, Iwona. From verb to clitic and nominal suffix: The Somali -e, -o nouns (89-106) 
Kuo, Pinmin. On the use and function of Chinese keshi: An explanation based on 

the notion 'inference system' (107-122) 
Mmusi, Sheila Onkaelse. OCP violations in Setsvvana: Evidence for redefining the OCP? 

(123-142) 
Prieto, Pilar. Truncation processes in Spanish (143-158) 
Hualde, Jos6 Ignacio, & Gorka Elordietta. On the lexical/post-lexical distinction: Vowel 

assimilation in Lekeitio Basque (Squib) (159-164) 
Iwasaki, Yasufumi. Review of Nanette Twine ( 199 1), Language and the modern state: 

The reform of written Japanese (165-168) 
Murphy, M. Lynne. Review of Marina Yaguello (1991), Lunatic lovers of language: 

Imaginary languages and their inventors (169-172) 
Book notices 

Goldlap, Christel (1991). Lokale Relationen im Yukatekischen: Eine 
onomasiologische Studie (Hans Henrich Hock) (173-174) 

Kachru, Branj B. (1992). The other tongue: English across cultures (Hans Henrich 
Hock) (174-176) 
Recent Books (177-182) 
Contents of volumes 17-21 (183-189) 

Volume 22:2 (Fall 1992) 

Twenty-five years of linguistic research and teaching at the 

University of Illinois at Urbana-Champaign 

(Edited by Braj B. Kachru, with the assistance of 

Amy C. Cheatham and Frances Vavrus) 

Antonsen, Elmer H. Foreword (vii-xii) 

Preface (xiii-xiv) 

Acknowledgements (xv) 

Part I: Perspectives on linguistics in the Midwest and at Illinois 

Kahane, Henry, & Braj B. Kachru. Introduction (3-5) 

Kachru, Braj B. Linguistics in the Midwestern region: Beginnings to 1973 (7-35) 

Kahane, Henry. History of the Department of Linguistics at the University of 
Illinois at Urbana-Champaign (37-39) 

Kahane, Henry. The European emigree (40-41) 

Osgood, Charles E. The tale of an eager then lonely then contented dinosaur 
(42-58) 

Lees, Robert B. How to find the right tree to bark up (59-64) 

Kachru, Braj B. Three linguistic reincarnations of a Kashmiri Pandit (65-74) 



208 Studies in the Linguistic Sciences 23:1 (Spring 1993) 

Kisseberth, Charles W. A sense of perspective (75-80) 
Part II: Memorial tributes to a builder: Henry R. Kahane 

Introduction (85) 

Antonsen, Elmer H. (87-89) 

Gamer, Roberta Kahane (89-9 1 ) I 

Kahane, Charles (91-92) I 

Weir, Morton W. (92-94) a 

Faulkner, Larry R. (94-95) | 

Zgusta, Ladislav (95-97) 

Kachru, Braj B. (97-l(X)) 
Part HI: Graduate student research 1964-1992 

Introduction (103) 

Ph.D. dissertation abstracts (105-237) 

Master's thesis abstracts (239-257) 

Research in progress up to August 1992 (259-261) 

Author index (263-265) 

Language index (267-271) 

Regional index (273-276) 

Area of concentration (277-279) 

Index of advisors (281-283) 



i 



All checks and money orders must be in U.S. Dollars, drawn on a U.S. Bank, 
and made payable to University of Illinois . Sales Tax must be included as follows (ex- 
cept tax exempt organizations): IL 71/4%, IN 5%, MI 4%, MN 6%, OH 5%, VVI 5%. Please 
include $2.00 postage per book. 

While quantities last, Volumes 7:2 through 17:2 are being offered at half-price. 

Papers in General Linguistics $2.50 $5.00 

Papers in General Linguistics $2.50 $5.00 

Papers in General Linguistics $2.50 $5.00 

Papers in General Linguistics $2.50 $5.00 

Relational Grammar and Semantics $2.50 $5.00 

(Editor: Jerry L. Morgan) 

Papers in General Linguistics $2.50 $5.00 

Studies in Arabic Linguistics $2.50 $5.00 

(Editor: Michael J. Kenstowicz) 

Papers in General Linguistics $2.50 $5.00 

Dimensions of South Asian Linguistics $2.50 $5.00 

(Editor: Yamuna Kachru) 

Papers in General Linguistics $2.50 $5.00 

Papers on Diachronic Syntax: $2.50 $5.00 

Six Case Studies 

(Editor: Hans Henrich Hock) 

Papers in General Linguistics $2.50 $5.00 

Studies in Language Variation: $2.50 $5.00 

Nonwestern Case Studies 
(Editor: Braj B. Kachru) 

Papers in General Linguistics $2.50 $5.00 

Language in African Culture and Society $2.50 $5.00 

(Editor: Eyamba G. Bokamba) 

Papers in General Linguistics $2.50 $5.00 

Linguistic Studies in Memory of $2.50 $5.00 

Theodore M. Lightner 
(Editor: Michael J. Kenstowicz) 

Papers in General Linguistics $3.00 $6.00 

Illinois Studies in Korean Linguistics $3.00 $6.00 

(Editor: Chin-W. Kim) 

Papers from the 1986 South Asian $3.00 $^^ 

Languages Analysis Roundtable 
(Editor: Hans Henrich Hock) 

Papers in General Linguistics $3.00 $6.00 



Vol. 


7:2 


Fall 1977 


Vol. 


8:1 


Spring 1978 


Vol. 


8:2 


Fall 1978 


Vol. 


9:1 


Spring 1979 


Vol. 


9:2 


Fall 1979 


Vol. 


10:1 


Spring 1980 


Vol. 


10:2 


Fall 1980 


Vol. 


11:1 


Spring 1981 


Vol. 


11:2 


Fall 1981 


Vol. 


12:1 


Spring 1982 


Vol. 


12:2 


Fall 1982 


Vol. 


13:1 


Spring 1983 


Vol. 


13:2 


Fall 1983 


Vol. 


14:1 


Spring 1984 


Vol. 


14:2 


Fall 1984 


Vol. 


15:1 


Spring 1985 


Vol. 


15:2 


Fall 1985 


Vol. 


16:1 


Spring 1986 


Vol. 


16:2 


Fall 1986 


Vol. 


17:1 


Spring 1987 


Vol 


17:2 


Fall 1987 



STUDIES IN THE LINGUISTIC SCIENCES 

The following issues are available 
(For contents of volumes 18-22, see pp. 201-208) 

Vol. 18:1 Spring 1988 Papers in General Linguistics $6.00 

Vol. 18:2 Fall 1988 Papers in General Linguistics $6.00 

Vol. 19:1 Spring 1989 Papers in General Linguistics $7.50 

Vol. 19:2 Fall 1989 The Contribution of African Linguistics $7.50 

to Linguistic Theory, Vol. 1 
(Editor: Eyamba G. Bokamba) 

Vol. 20:1 Spring 1990 The Contribution of African Linguistics $7.50 

to Linguistic Theory, Vol. 2 
(Editor: Eyamba G. Bokamba) 

Vol. 20:2 Fall 1990 Linguistics for the Nineties: $7.50 

Papers from a Lecture Series in Celebration 
of the Department's Twenty-Fifth Anniversary 
(Editor: Hans Henrich Hock) 

Vol. 20:3 Spring 1991 Meeting Handbook: Thirteenth South Asian $5.00 

Languages Analysis Roundtable, 25 - 27 May 
1991, University of Illinois at Urbana-Champaign 
(Editor: Hans Henrich Hock) 

Vol. 21:1 Spring 1991 Papers in General Linguistics $7.50 

Vol. 21:2 Falll991 Illinois Studies in Korean Linguistics, 2 $7.50 

(Editor: Chin-W. Kim) 

Vol. 22:1 Spring 1992 Papers in General Linguistics $7.50 

Vol. 22:2 Fall 1992 Twenty-five Years of Linguistic Research $10.00 

and Teaching at the University of Illinois 
at Urbana-Champaign 
(Editor: Braj B. Kachru) 



FOR EARLIER ISSUES AND ORDER INFORMATION SEE INSIDE COVER 



Orders should be sent to: 

SLS Subscriptions, Department of Linguistics 

University of Illinois 

4088 Foreign Languages Building 

707 S. Mathews 

Urbana, Illinois 61801 



The Linguistic Sciences 



VC4.lMli2XNUMMR2 



DePArrMBNT or UNot Ttsnrs 
UMVBarTY or ■ j jnob at mmANAOUMPAiow 




STUDIES IN THE LINGUISTIC SCIENCES 



publication of the department of linguistics in the 
College of liberal arts and Sciences of the 
University of Illinois at urbana-Champaign 



General editor: Elmer H. Antonsen 

Review editor: James H. Yoon 

EDITORIAL ASSISTANTS: SARA MICHAEL, MARK HONEGGER 

Editorial Board: Eyamba G. Bokamba, Chin-chuan Cheng, Jennifer S. Cole, 
Georgia M. Green, Hans Henrich Hock, Braj B. Kachni, Yamuna Kachru, Chin- 
W. Kim, Charles W. Kisseberth, Peter Lasersohn, Howard Maclay, Jerry L. 
Morgan, Rajeshwari Pandharipande, James H. Yoon, and Ladislav Zgusta. 

AIM: SLS is intended as a forum for the presentation of the latest original 
research by the faculty and students of the Department of Linguistics, University 
of Illinois, Urbana-Champaign. Scholars outside the Department and from other 
institutions are also cordially invited to submit original linguistic research for 
consideration. In all cases, articles submitted for publication will be reviewed by 
a panel of at least two experts in the appropriate field to determine suitability for 
publication. Copyright remains with the individual authors. Authors will receive 
10 offprints of their individual contributions. 

SPECIAL ISSUES: Since its inception SLS has devoted one issue each year to 
restricted, specialized topics. A complete list of such special issues is given on the 
back cover. 

BOOKS FOR REVIEW: Review copies of books may be sent to: 
Editor, Studies in the Linguistic Sciences, 
Department of Linguistics, 4088 For. Lang. Bldg., 
University of Illinois 
707 S. Mathews, 
Urbana,IL 61801, USA 

SUBSCRIPTION: Normally, there are two issues per year. Requests for sub- ^ 
scriptions should be addressed to SLS Subscriptions, Department of Linguistics, 
University of Illinois, 4088 Foreign Languages Building, 707 South Mathews, 
Urbana, Illinois 61801. 



Price: $8.00 per single issue. 



STUDIES IN THE LINGUISTIC SCIENCES 
Papers in General Linguistics 



EDITOR 
Elmer H. Antonsen 

EDITORIAL ASSISTANTS 

Sara Michael 
Mark Honegger 



VOLUME 23, NUMBER 2 
FALL 1993 



DEPARTMENT OF LINGUISTICS. UNIVERSITY OF ILLINOIS 
URBANA, ILLINOIS 61801 



EDITOR'S NOTE 



Although this issue bears the volume and issue designation 23:2 Fall 1993, 
filling a gap in the journal's twice-a-year scheduled appearance, because of un- 
avoidable circumstances it could not be published until October 1996. We have 
therefore indicated the actual publication date in square brackets where appro- 
priate. 



CORRECTION 



The Editors wish to correct an unfortunate inaccuracy in the following, pre- 
viously published article: 

Paul Agdebor: Verb seriaUzation in Ewe, Studies in the Linguistic Sciences 
23:1.21-42, Spring 1993. 

The designation of the voiceless bilabial fricative, <f >, was apparently misinter- 
preted as the IPA symbol for the voiceless alveopalatal fricative {\\, which was 
then converted to the phonological equivalent III producing an inaccurate phono- 
logical rendition of examples 7, 8, 9, 10, 12, 31, 34, 50, and 57. In all of these 
examples. III should be read as /$/. We greatly regret this error and apologize to the 
author. 

In the future, authors will be given the opportunity to proof-read their 
contributions before they go to press. 



CONTENTS 

Editor's Note and Correction iv 

SAE-YOUN CHO: Auxiliary verb constructions in Korean 1 

CaSSANDRE CRESWELL: Criticizing witii a question 25 

ABDUL Aziz DIOP: Vowel deletion in Pulaar: Rime and nuclear mergers and the issue 

of the syntax-phonology interface 33 

Hans HENRICH Hock: Subversion or convergence? The issue of pre-Vedic retro- 
flexion reexamined 73 

JONI Kay Hurley: Request formation in Ecuadorian Quichua 1 17 

NKONKO M. KAMWANGAMALU: Advancement in some Asian and African languages 137 

Emmanuel Kweku OSAM: Animacy distinctions in Akan grammar 153 

Uthaiwan Wong-OPASI: The interplay between tone, stress, and syllabification 

in Thai 165 

Mary A. Wu: Adjectival amd determinate measure phrases and NP interpretations in 

Mandarin Chinese 193 



c 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

AUXILIARY VERB CONSTRUCTIONS IN KOREAN* 

Sae-Youn Cho 

Korean auxiliary verb constructions have led to much controversy 
concerning how they can be analyzed. In analyzing this construction 
there have been at least two approaches: one is a syntactic approach, 
including the bi-clausal analysis and the VP analysis. The other is a 
lexical approach, including the compound verb analyses. Recently 
many papers on this auxiliary construction have taken either the bi- 
clausal analysis or the compound verb analyses without specifying 
why other alternatives cannot be good candidates. This paper presents 
the VP hypothesis to account for auxiliary constructions and argue 
that this analysis provides a simpler explanation of various phenom- 
ena related to this construction than the bi-clausal or the compound 
verb analyses. 

1. Introduction 

This paper provides an analysis of Korean auxiliary verb constructions' un- 
der the Head-Driven Phrase Structure Grammar (hereafter, HPSG) framework. 2 
The data in ( 1 ) show various types of auxiliary constructions in Korean where the 
first one of the bold strings in the data is either a verb (e.g. ilk-e in (la)) or an ad- 
jective (e.g. yeypp-e in (lb)), and the second one is an auxiliary verb (e.g. po-ass- 
ta in(la)).3 

(1) a. Mary-ka chayk-ul ilk-e po-ass-ta.'* 

-N book-A read-Comp try-P-Dec 
'Mary tried to read a book.' 

b. Mary-ka yeyppu-e ci-ass-ta. 

-N pretty-Comp become-P-Dec 
'Mary became pretty.' 

c. Mary-ka yeyppu-e poi-ass-ta. 

-N pretty-Comp seem-P-Dec 
'Mary seemed to be pretty.' 

d. Mary-ka chayk-ul ilk-na po-ta. 

-N book-A read-Comp seem-Dec 
'Mary seems to read a book.' 

e. Mary-ka chayk-ul ilk-eya ha-n-ta. 

-N book-A read-Comp must-Pres-Dec 
'Mary must read a book.' 

f. Mary-ka chayk-ul iik-ko iss-ta. 

-N book-A read-Comp be-Pres-Dec 
'Mary is reading a book.' 



2 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

There has been much controversy concerning the constructions in (1), cen- 
tering on two problems. One is the question of which phrasal categories, such as 
VP or S, each auxiliary verb subcategorizes for. The other is how to handle the 
morphological requirements for the subcategorized elements by each auxiliary 
verb (AUX). For example, po-ass-ta in (la) always requires a preceding verb with 
the suffix -e. If the preceding verb has a different suffix such as -eya in (le), the 
sentence is ungrammatical, as in *Mary-ka chayk-ul ilk-eya po-ass-ta. Recently 
many papers on Korean linguistic phenomena in HPSG, including Yoo 1993, as- 
sume that the two italicized strings in ( 1 ) are a compound verb, where the AUX is 
only a part of the compound verb, without specifying any reason why AUX does 
not constitute an independent category. 

This paper will show that if AUX is assumed to be an independent category 
which subcategorizes for a VP and a NP (=Subject), the AUX constructions in (1) 
can be sufficiently explained in the HPSG framework. In addition, this analysis 
can deal with the morphological requirements for the subcategorized elements, 
such as the restrictions on the occurrence of tense suffix and the suffix form 
(Comp). 

The arguments of this paper are organized in three main sections. In section 
2, three competing hypotheses, the compound verb hypothesis, the bi-clausal hy- 
pothesis and the VP hypothesis, are presented. The primary claim of the VP hy- 
pothesis, that the AUXs in (1) are an independent category, will be motivated by 
arguments regarding the scope interpretation in coordination structures in section 
2.1.1, the distributional properties of kuray+vtvh constructions in section 2.1.2, 
and verbal fronting (V^) in section 2.1.3.5 

Section 2.2 argues for AUX as an independent category by presenting argu- 
ments showing that AUX subcategorizes for a VP rather than a S in various envi- 
ronments. Section 2.2.1 demonstrates that negative polarity item (NPI) require- 
ments do not provide evidence that AUX would subcategorize for a S. Section 
2.2.2 also demonstrates that the reflexive caki-ka + Verb in the AUX construc- 
tions does not necessarily constitute a S, thus posing no problems for the AUX 
subcategorization proposal. Rather, the arguments can be used as evidence that 
AUX subcategorizes for a VP. 

Section 3 identifies two required verbal suffixes, Comp and tense, for the 
subcategorized VP, which will be integrated into a formalized account of AUX 
subcategorization in section 4. Consequently, if an AUX subcategorizes for a VP 
and a NP, a unified and intuitive explanation for the AUX constructions is 
possible. 

2. Constituency tests 

There have been at least three analyses of the AUX constructions in (1): bi- 
clausal analysis by many early transformationalists, VP analysis by Park 1990 
and compound verb analyses by Cho 1988, Sells 1991 and Yoo 1993. ^ For each 
analysis described below, there is a representation of (la), which demonstrates that 
analysis's structural claims. 



Cho: Auxiliary verb constructions in Korean 3 

The bi-clausal analysis regards AUX as a category subcategorizing for a S 
and a NP (Subject) where the S has a trace or PRO depending on the AUXs; if the 
AUX is an equi-verb, the gap is a PRO but if it is a raising verb, then it is a trace. 
The analysis treats the verb ilk-e as the verb of the embedded clause [0 chayk-ul 
ilk-e ], whereas the AUX po-ass-ta is treated as the verb of the main clause, as 
shown in (2). 



(la) Mary-ka chayk-ul ilk-e 

-N book- A read-Comp 
'Mary tried to read a book.' 

(2) Bi-clausal analysis 

S 



po-ass-ta. 

try-P-Dec 




Mary-ka PRO chayk-ul ilk-e po-ass-ta 

In this analysis the correct surface structure can be derived in terms of Equi-NP 
Deletion, since the AUX po-ass-ta is an equi-verb."^ On the other hand, if the AUX 
in a sentence is a raising verb like iss-ta the surface structure can be derived by 
Subject-to-Subject Raising. This analysis is called bi-clausal because (la) has two 
sentences where one is a main clause and the other is an embedded one, as shown 
in (2). 

The VP analysis regards the AUX as a category subcategorizing for a VP and 
a NP. This analysis treats the VP chayk-ul ilk-e as a complement of the AUX po- 
ass-ta, so that the constructions like (la) have no embedded sentence as shown in 
(3) on the next page. 

Unlike the bi-clausal analysis this does not postulate on empty category in the 
constructions. The difference between equi and raising AUXs can be distin- 
guished by the semantic CONTENT of AUXs. 

Finally, there are two different Compound analyses: Cho 1988 and Sells 
1991. Sells 1991 claims that the AUX and the preceding verb syntactically form 
a compound verb so that the compound verb subcategorizes for two NPs to ac- 
count for the example in (la). In this analysis ilk-e and po-ass-ta are each members 



4 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(3) VP analysis 




V(=AUX) 



NP V 

I I 

Mary-ka chayk-ul ilk-e po-ass-ta 

of a lexical category but syntactically form a compound verb ilk-e po-ass-ta as 
shown in (4). 

(4) Complex verb analysis by Sells 1991 

S 




Mary-ka chayk-ul ilk-e po-ass-ta 

Similarly to Sells 1991, Cho's 1988 analysis also regards two verbs as a 
compound verb, where the compound verb subcategorizes for two NPs to explain 
the example in (la). However, Cho claims that the suffixed element ilk-e is a 
gerundive nominal and that this gerundive nominal and the verb po-ass-ta mor- 
phologically form a compound verb ilk-e-po-ass-ta as in (5). 

(5) Compound verb analysis by Cho 1988 




c 



N-1 
I 
Mary-ka chayk-iol ilk-e 



po-ass-ta 



Both compound verb analyses differently predict the possibility of the oc- 
currence of adverbs between the AUX and the preceding verb since they have dif- 



Cho: Auxiliary verb constructions in Korean 



ferent compound formations. Under Sells' 1991 analysis any adverb modifying 
the AUX can occur between the two because each part of the compound verb is a 
bar-level category. But Cho's 1988 analysis predicts that no adverb modifying 
the AUX can occur between the two because the compound verb is morphologi- 
cally formed so that nothing can be placed in front of the AUX. 

Despite their differences in category assignment and compound formation, 
they make similar claims about constituency of the AUX constructions. 
Therefore, they will be grouped together as the Compound Verb analysis. 

The following sections will argue that the VP analysis is more plausible than 
the compound verb or bi-clausal analysis in accounting for the AUX 
constructions in (1). 

2.1. Evidence against compound verb analysis 

2.1.1. Coordination and scope problems 

The VP analysis provides a simpler analysis for the ambiguity of sentences 
with VP coordination than does the compound verb analysis. VP coordinations 
with AUX are possible in Korean as shown in (6). A sentence with VP coordina- 
tion like (6) can have two different interpretations, (7) and (8). To have the correct 
readings the AUX in (7) must have scope over the VP pap-ul mek-e while the 
AUX in (8) must have scope over the whole conjoined VP chayk-ul ilk-ko pap-ul 
mek-e. 



(6) John-i chayk-ul ilk-ko 



(7) 



pap-ul mek-e 



-N book-A read-and rice-A eat-Comp 
'John read a book and tried to have a meal.' OR 
'John tried to read a book and have a meal.' 

John-i [[chayk-ul ilk-ko]vp [pap-ul mek-e 

-N book(s)-A read-and rice-A eat-Comp 
'John read a book and tried to have a meal.' 



po-ass-ta. 
try-P-Dec 



po-ass-ta]vp]vp 
try-P-Dec 




John-i chayk-ul ilk-ko pap-ul mek-e po-ass-ta 

(8) John-i [[chayk-ul ilk-ko|v|. | pap-ul- mek-e]vp 
-N book-A read-and rice-A eat-Comp 

'John tried to read a book and have a meal.' 



[po-ass-ta] vlvp 
try-P-Dec 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 




John-i chayk-ul ilk-ko pap-ul mek-e po-ass-ta 

If po-ass-ta in (6) is an AUX which subcategorizes for a VP, the structures and 
their interpretations Uke (7) and (8) can both be derived from (6). If the object of 
the AUX po-ass-ta in (6) is the VP pap-ul mek-e, then its interpretation must be 
John read a book and tried to have a meal, Uke (7), whereas if the AUX takes the 
whole conjoined VP chak-ul ilk-ko pap-ul mek-e as its object its interpretation 
must be John tried to read a book and have a meal, as in (8). 

However, the compound verb analysis cannot predict that the sentence in (6) 
can have two interpretations, since it provides a representation as shown in (9). 
Under the compound verb analysis po-ass-ta is a part of the verb mek-e po-ass-ta as 
shown in (9), not an independent constituent. 

(9) 




V 

I I 

John-i chayk-\il ilk-ko pap-ul mek-e po-ass-ta 

Thus, the structure and meaning in (8) cannot be derived from (6). If the com- 
pound verb analysis is taken as correct, an additional explanation for why (7) is 
possible while (8) is not must be provided.^ Such an explanation will not be 
needed under the VP analysis to deal with VP coordination with AUX. 

2.1.2. Kuray substitution 

The VP analysis predicts the possibility of substituting a VP for the word 
kuray+sujfix in Korean while the compound verb analysis does not. An interroga- 



Cho: Auxiliary verb constructions in Korean 7 

tive like (10a) may be answered with a sentence like (10b), where a VP like pap- ul 
mek-ess-ni may be replaced with the word ku ray + suffix. Thus the word kuray- 
ess-e in (10b) as an answer for the interrogative sentence (10a) can be used instead 
of the VP pap-ul mek-ess-e, which is like do so constructions in English. 

(10) a. Mary-ka [pap-ul mek-ess-nijyp? 

-N rice-A eat-P-Q 
'Did Mary have a meal?' 
b. ung, Mary-ka kuray-ess-e. 
yes, -N do so-P-Dec 

'Yes, Mary did so.' 

Similarly, VPs with AUXs can also be replaced by the word kuray+suffjx as 
shown in (1 Ib-c), while AUXs alone cannot be replaced by it, as in (1 Id). 

(11) a. Mary-ka [[pap-ul mek-e]vp cwu-kojyp iss-ni? 

-N rice-A eat-Comp give-a-favor-of be-Q 
'Is Mary giving a favor of having a meal?' 

b. ung, Mary-ka kule-ko 
yes, -N do-so 
'Yes, Mary does so.' 

c. ung, Mary-ka kuray 
yes, -N do so 
'Yes, Mary gives a favor ol 

d. *ung, Mary-ka pap-ul 
yes, -N rice-A 

'Yes, Mary does so of having a meal.' 

As answers for the interrogative sentence containing a raising AUX iss- in (11a), 
the sentence (1 lb) has the word kule-ko replacing the VP pap-ul mek-e cwu-ko in 
(11a) and the sentence (lie) has the word kuray replacing the VP pap-ul mek-e in 
(11a). On the other hand, the AUX cwu-ko in (11a) cannot be replaced by the 
word kule in (lid) since it is not a VP. Thus the distributional behavior of kuray 
can be accounted for if the VP analysis is chosen. 

In addition, the fact that sentences (12b-c) are acceptable answers for (12a) 
but (12d) is not also shows that the word kuray + sujfix can replace only VPs, not 
AUXs alone. 

(12) a. Mary-ka [[pap-ul mek-e]vp po-ass-ni]? 

-N rice-A eat-Comp try-P-Q 
'Did Mary try to have a meal?' 

b. ung, Mary-ka kuray po-ass-e. 
yes, -N do-so try-P-Dec 
'Yes, Mary did try to do so.' 

c. ung, Mary-ka kuray-ass-e. 
yes, -N do so-P-Dec 
'Yes, Mary did so.' 



iss-e.9 




be-Dec 




cwu-ko 


iss-e. 


give-a-favor-of 


be-Dec 


doing so.' 




mek-e kule- 


ko iss-e. 


eat do-so 


be-Dec 



8 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

d. *ung, Mary-ka pap-ul kuray po-ass-e. 
yes, -N rice-A do-so try-P-Dec 

'Yes, Mary did so of having a meal.' 

As answers for the interrogative sentence containing an equi AUX po-ass-ta in 
(12a), the sentence (12b) has the word kuray replacing the VP pap-ul mek-e in 
(12a) and the sentence (12c) has the word kuray-ass-e replacing the W pap-ul 
mek-e po- in (12a). But the AUX mek-e in (12a) cannot be replaced by the word 
kuray in (12d) because it is not a VP. Again, the distributional behavior of kuray 
can be sufficiently explained if the VP analysis is taken as the correct analysis. 

However, if the compound verb analysis is chosen, an explanation must be 
provided for how a part of a verb can be replaced by the word kuray as in (1 Ib-c) 
and (12b-c) and why the verb cwu-ko or mek-e cannot be replaced by kule or 
fcwray in (lid) and (1 2d). 

Furthermore, there is another compound verb, cap-e mek-ess-ni as in (13a), 
whose constituents cannot be replaced by the word kuray as seen in (13b-c). 

(13) a. John-i thokki-lul ^j[c^Jp-t mek-ess-ni?]. 

-N rabbit-A catch-(Comp) eat-P-Q 
'Did John catch and eat a rabbit?' 

b. *ung, John-i kuray mek-ess-e. 
yes, -N do so eat-P-Dec 
'Yes, John did so and ate.' 

c. *ung, John-i cap-e kule-ess-e. 
yes, -N catch doso-P-Dec 
'Yes, John caught and did so.' 

Under the VP analysis, (13b) and (13c) are not possible answers to the interroga- 
tive sentence (13a) because cap-e mek-ess-ni in (13a) is not a phrase (VP) but a 
single word (verb). Thus a part of the verb, like cap-e or mek-ess-e, cannot be re- 
placed by the word kuray as in (13b) or (13c). Again, the compound verb analysis 
requires additional restrictions in order to account for why a part of a compound 
verb in (13a) cannot be replaced by the word kuray as in (13b) and (13c). 

Therefore, the distributional restrictions of the word kuray under the VP 
analysis follow from the generalization that VPs can be replaced by the word 
kuray. 

2.1.3. V^ fronting 

The VP analysis also provides a simpler analysis of verb phrase fronting than 
the compound verb analysis. Verbal phrases like a VP or S can be fronted when 
the gap is filled with the verb ha- as in (14), whereas the lexical category V , mek- 
ki-nun, cannot be fronted as shown in (15). 

(14) a. Mary-ka pap-ul mek-nun-ta. 

-N rice-A eat-Pres-Dec 
'Mary has a meal' 



Cho: Auxiliary verb constructions in Korean 9 

b. [pap-ul mek-ki-nun]vp Mary-ka han-ta."o 
rice- A eat-NM-T -N do-Dec 

'It is Mary who has a meal.' 

c. [Mary-ka pap-ul mek-ki-nun]s han-ta. 

-N rice-A eat-NM-T do-Dec 

'Mary has a meal.' 

(15) *[mek-ki-nun]v Mary-ka pap-ul han-ta. 

eat-NM-T -N rice-A do-Dec 

'Mary has a meal.' 

Since the VP pap-ul mek-nun-ta in (14a) is fronted, (14b) is acceptable. In the 
same way, (14c) is also acceptable because the S Mary-ka pap-ul mek-nun-ta in 
(14a) is fronted. However, (15) is unacceptable since the lexical category (V'') 
mek-nun-ta in (14a) cannot be fronted. 

Either a VP with an AUX as in (16c) or a VP without an AUX as in (16b) 
can be fronted. The possibility of VP fronting without AUX as in (16b) or with 
AUX in (16c) follows from a generalization that all V" categories can be fronted in 
Korean under the VP hypothesis. 

(16) a. Mary-ka pap-ul mek-eya han-ta. 

-N rice-A eat-VForm must-Dec 
'Mary must have a meal.' 

b. [pap-ul mek-ki-nun] Mary-ka ha-eya han-ta. 

-A eat-NM-T -N must-Comp do-Dec 

'It is Mary who must have a meal.' 

c. [pap-ul mek-eya ha-ki-nun] Mary-ka han-ta. 

-A eat-Comp must-NM-T -N do-Dec 

'It is Mary who must have a meal.' 

But under the compound verb analysis an additional restriction to explain 
why a compound verb like mek-eya han-ta in (16b) can be divided into two parts 
to be fronted is necessary. 

Cho 1988 argues that the compound verb analysis is preferable because of the 
fact that scrambling of the VP subcategorized for by AUX is impossible and ad- 
verbs immediately preceding AUX are impossible. The HPSG analysis of this 
problem is in section 4.1. 

2.2. Evidence against bi-clausal analysis 

Once the compound verb hypothesis is found incorrect, it must be deter- 
mined if the AUX subcategorizes for a S or a VP. Specifically, the problem here is 
whether the phrase chayk-ul ilk-e in (17= la) is a S or a VP. 

(17= la) Mary-ka [X chayk-ul ilk-e] po-ass-ta. 

-N book-A read-Comp try-P-Dec 

'Mary tried to read a book.' 

To show that the phrasal category that each AUX subcategorizes for is a S, 
the bi-clausal analysis provides two arguments. The first is based on negative po- 



1 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

larity items. The second concerns the occurrences of caki-ka in the X position of 
(17). However, the VP analysis is more plausible than the bi-clausal analysis be- 
cause these two arguments actually support the VP analysis rather than the bi- 
clausal analysis. 

2.2.1. Negative polarity items (NPIs) 

The VP analysis provides a simpler analysis for the distributional behaviour of (i 
NPIs than the bi-clausal analysis. Negative polarity items such as amwukes-to must 
occur with a negative element (Neg) like anh- within the same clause, as shown in 
(18) and (19). The sentences in (18a) and (19a) are acceptable because the NPI 
amwukesto and the Neg anh-ass-ta co-occur in the same clause. But (18b) and (19b) 
are unacceptable because (18b) has no Neg in the sentence and (19b) has no Neg in 
the embedded clause with the NPI. 

(18) a. 



[Mary-ka amwukesto 


mek-ci anh-ass-ta]. 


-N anything 


eat Neg-P-Dec 


'Mary ate nothing.' 




*[Mary-ka amwukesto 


mek-ass-ta]. 


-N anything 


eat-P-Dec 


'Mary ate nothing.' 




[Mary-ka amwukesto 


mek-ci anh-ass-ta-ko] 


-N anything 


eat Neg-P-Dec-Comp 



(19) a. 

John-i (Sue-lul) seltukha-ess-ta. 
-N -A persuade-P-Dec 

'John persuaded Sue that Mary ate nothing.' 
b. * [Mary-ka amwukesto mek-ass-ta-ko] 
-N anything eat-P-Dec-Comp 

John-i (Sue-lul) seltukha-ci anh-ass-ta. 

-N tell-P-Comp Neg-P-Dec 

'John persuaded Sue that Mary ate nothing.' 

Examples (18) and (19) show that the clause-mate constraint, which specifies that 
a NPI must occur with a Neg in the same clause, is needed to explain the 
constructions. 

The AUX constructions with a NPI as in (20) are also possible. Under the 
VP analysis (20a) and (20b) are predicted as grammatical because the phrases 
amwukesto mek-ci anh-a in (20a) and amwukesto mek-e in (20b) are VPs where 
the NPI in each VP observes the clause-mate constraint. 

(20) a. Mary-ka amwukesto mek-ci anh-a po-ass-ta. 

-N anything eat Neg try-P-Dec 

'Mary tried not to eat anything.' 
b. Mary-ka amwukesto mek-e po-ci anh-ass-ta. 

-N anything eat try Neg-P-Dec 

'Mary didn't try to eat anything.' 



Cho: Auxiliary verb constructions in Korean 1 1 

Thus, under the VP analysis, the AUX constructions with a NPI follow from the 
generalization that a NPI must occur with a Neg in the same clause. 

On the other hand, the bi-clausal analysis predicts that the italicized words 
amwukesto mek-ci anh-a in (20a) are a S because the AUX po-ass-ta subcategorizes 
for a S. If the strings are a S and the NPI observes the clause-mate constraint, the 
grammaticality of (20a) can be accounted for. Still, the analysis wrongly predicts 
that (20b) is ungrammatical, because the NPI in the S amwukesto mek-e violates 
the clause-mate constraint. There are two possible solutions to this problem. One 
is that the Tensed S Condition (TSC) is invoked to explain the behaviour of NPIs, 
instead of the clause-mate constraint. Because the embedded S amwukesto mek-e in 
(20b) has no Tense and TSC restricts the co-occurrence of the NPI and the Neg to 
a sentence with Tense, TSC can correctly predict that (20b) is grammatical. The 
other solution is to treat sentences like (20b) as scrambled constructions so that 
they still seem to observe the clause-mate constraint on NPIs. If the NPI 
amwukesto in (20b) is adjoined to the embedded S by Scrambling, the NPI can 
occur with the Neg in the same clause. This solution makes the correct prediction 
for the grammaticality of (20b). However, these alternatives are not preferable. 

If the bi-clausal analysis takes Tensed S Condition as the proper constraint to 
deal with the AUX constructions with a NPI, instead of the clause-mate con- 
straint, the differences in acceptability between (18b & 19b) and (20b) can be ac- 
counted for as follows. Even though all three sentences with the NPI amwukesto 
do not have a Neg in the same clause, (20b) is possible because the S amwukesto 
mek-e has no Tense, thereby not violating TSC, whereas (18b) and (19b) are im- 
possible because each clause with a NPI has a Tense and thus violates TSC. 

However, examples like (20b) show that although there is a tensed clause 
with a NPI but no Neg within the same clause, the sentence can be grammatical. 
The bi-clausal analysis in conjunction with TSC predicts that (21a) is acceptable 
because the NPI in the embedded S subcategorized for by the AUX ha- in (21a) 
does not violate TSC. But it wrongly predicts that (21b) is unacceptable because 
the NPI in the embedded S violates TSC. 

(21) a. Mary-ka [amwukesto mek-ci anh-ass-eya] ha-ess-ta. 

-N anything eat-Comp Neg-P-Comp must-P-Dec 

'Mary must not have eaten anything.' 
b. Mary-ka [amwukesto mek-ess-eya] ha-ci anh-ass-ta. 

-N anything eat-P-Comp must-Comp Neg-P-Dec 

'Mary didn't have to eat something.' 

Therefore, this solution is not a good candidate to account for these constructions. 

With the second solution of the bi-clausal analysis the clause-mate constraint 
is regarded as correct, but the structure for (20b) is considered scrambled. The 
NPI amwukesto as the object of the verb mek-e in (20b) can be analyzed like (22), 
where it does not move at all, or it may be treated as a scrambled structure like 
(23), where the NPI moves to the sister of the embedded S. 



12 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(20b) Mary-ka amwukesto mek-e po-ci anh-ass-ta. 

-N anything eat try Neg-P-Dec 

'Mary didn't try to eat anything.' 

(22) Mary-kaj [PROj [amwukesto mek-e]vp]s po-ci anh-ass-ta. 

(23) Mary-kai [amwukestOj [PROj [tj mek-e]vp]s]s po-ci anh-ass-ta. 

If (20b) has a structure like (23), in which the NPI amwukesto is adjoined to the 
embedded S in terms of scrambling, the NPI belongs to the higher S so that it ob- 
serves the clause-mate constraint. Thus the bi-clausal analysis does not need any 
modification for these constructions. 

However, there are examples which show that even though we treat some sen- 
tences as scrambled to make a NPI occur with a Neg within the same clause, these 
sentences cannot be grammatical. The sentence in (24) is unacceptable, even if the 
NPI is scrambled to occur with a Neg within the same clause. Under the bi- 
clausal analysis, in conjunction with the Scrambling solution, (19b) is unaccept- 
able because the NPI in the embedded sentence violates the clause-mate constraint. 
On the other hand, (24) should be acceptable because when the NPI in the embed- 
ded sentence is adjoined to the S in terms of Scrambling, and it belongs to the 
higher S which has a Neg, it does not violate the clause-mate constraint. Yet (24) 
is still unacceptable. 

(19b) * [Mary-ka amwukesto mek-ass-ta-ko] 
-N anything eat-P-Dec-Comp 
John-i (Sue-lul) seltukha-ci anh-ass-ta. 
-N tell-P-Comp Neg-P-Dec 

'John persuaded Sue that Mary ate nothing.' 

(24) *amwukestoi, John-i (Sue-lul) [Mary-ka 
anything -N -A -N 

0i mek-ass-ta-ko] s seltukha-ci anh-ass-ta." 

eat-P-Dec-Comp persuade-Comp Neg-P-Dec 

The Scrambling solution, thus, is also not a good candidate to account for these 
constructions. 

Again, if the bi-clausal analysis is taken to explain the AUX constructions, 
NPI restrictions are necessary, but under the VP analysis such restrictions are not 
necessary. 

2.2.2. The occurrence of caki-ka 

The VP analysis can correctly predict the occurrence of reflexive caki-ka in 
Korean while the bi-clausal analysis cannot. For the bi-clausal analysis claims 
that, as in example (25), the reflexive caki-ka is the subject of the embedded sen- 
tence, where only PRO and the reflexive caki-ka can occur in the X position of 
(17= la). '2 Conversely, the VP analysis claims that the reflexive is not the subject 
of the embedded sentence [X chayk-ul ilk-e ] but an adjunct to modify the subject 
Mary-ka in (17= la). 



Cho: Auxiliary verb constructions in Korean 1 3 

(17=la) Mary-ka [X chayk-ul ilk-e] po-ass-ta. 

-N book-A read-Comp try-P-Dec 

'Mary tried to read a book.' 

(25) Mary-ka [X chayk-ul ilk-e] po-ass-ta. 

a. *John-i 

b. *kunye-ka(=she) 

c. caki-ka(=self-N) 

d. PRO 

On the basis of Sells' 1993 claim that the reflexive caki-ka and PRO can occur in 
the X position as in (25c-d) but a R-expression and Pronominal cannot as in (25a- 
b), the Bi-clausal analysis can claim that the reflexive caki-ka is the subject of the 
embedded sentence so that the phrase [ X chayk-ul ilk-e] constitutes a S as in (26). 
If this is true, the occurrence of caki-ka in the subject position can support the Bi- 
clausal analysis, and the VP analysis, assigning (17) a structure like (27), must 
explain why the VP [chayk-ul ilk-e] in the AUX constructions can have a 
reflexive subject. 

(26) Mary-ka [caki-ka chayk-ul ilk-e]^ po-ass-ta. 

-N self-N book-A read-Comp try-P-Dec 

'Mary herself tried to read a book.' 

(27) Mary-ka caki-ka [chayk-ul ilk-e]yp po-ass-ta. 

-N self-N book-A read-Comp try-P-Dec 
'Mary herself tried to read a book.' 

However, there are examples showing that the reflexive with a subject case 
marker caki-ka can freely occur as an adjunct in a sentence. The examples (28a-b), 
where the reflexive caki-ka occurs as an emphatic expression modifying the sub- 
ject, show that the reflexive with a subject marker need not always be regarded as a 
subject. The subject of cohaha-ess-ta ('liked') in (28a) is Mary-ka, and in (28b) the 
subject of the embedded sentence is Mary-ka and that of the higher S is John-i. The 
reflexive caki-ka in (28) is an adjunct which modifies Mary-ka in (28a) and John- 
i/un in (28b). 

(28) a. [Mary-ka caki-ka John-ul cohaha-ess-ta]s. 

-N self-N -A like-P-Dec 

'Mary herself liked John.' 
b. John-un/i caki-ka [Mary-ka can-ta-ko]s malha-ess-ta. 

-T/N self-N -N sleep-Dec-Comp tell-P-Dec 

'John himself said that Mary slept.' 

To deal with the emphatic reflexive the Bi-clausal analysis must allow both a 
structure like (29), for (25c) where the reflexive caki-ka is the subject of the em- 
bedded clause, and a structure like (30) for (28a). Since the subject of the sentence 
(30) is not caki-ka but Mary-ka, the structure in (30), where caki-ka is an adjunct, 
is necessary. 



14 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(29) [Mary-ka [caki-ka chayk-ul ilk-e]^ po-ass-tajg. 

(30) [Mary-ka caki-ka [John-ul cohaha-ess-ta]vp Is- 

On the other hand, the VP analysis needs only one structure like (27) for 
(25c) and (28), because this analysis regards the reflexive caki-ka in (25c) and 
(28) as an adjunct. 

The difference in grammaticality between (31a) and (31b) shows that the VP 
analysis predicts the correct structure for the AUX constructions with the em- 
phatic reflexive caki-ka. The fact that the multiple occurrences of the emphatic re- 
flexive caki-ka in a sentence with AUX are not possible, as in (31a), shows that 
the string pap-ul mek-e in (31a) cannot be a S. However, the sentences with one 
emphatic reflexive and one reflexive as the subject of the embedded sentence are 
grammatical as in (31b). 

(31) a. *Mary-ka/nun caki-ka caki-ka pap-ul mek-e po-ass-ta.'^ 

-N/-T self-N self-N rice-A eat-Comp try-P-Dec 
'Mary herself tried to have a meal.' 
b. Mary-nun/ka caki-ka caki-ka (kacang) yeypputa-ko 

-T/-N self-N self-N the most pretty-Comp 
malha-n-ta.'4 
tell-Pres-Dec 
'Mary herself says that she is pretty.' 

Under the VP analysis, the structure of (31a) is regarded as (32a) and the structure 
of (31b) must be (32b). So this analysis correctly predicts that (32a) is unaccept- 
able but (32b) is acceptable. The sentence (32a) is unacceptable because both re- 
flexives caki-ka in (32a) are adjuncts modifying the subject Mary, where one of 
them is redundant. But the sentence (32b) is acceptable because the first caki-ka in 
the higher S is an adjunct and the second is the subject of the embedded sentences 
subcategorized by the verb malha- ('say'). 

(32) a. *[Mary-ka/nun caki-ka caki-ka [pap-ul mek-e po-ass-ta]vp]s- 

-N/-T self self rice eat-Comp try-P-Dec 

'Mary herself tried to have meal.' 

b. [Mary-nun/ka caki-ka [caki-ka (kacang) yeyppu-ta-ko]s 
-T/-N self-N self-N the most pretty-Comp 
malha-n-taJs- 
say-Pres-Dec 
'Mary herself says that she is pretty.' 

The Bi-clausal analysis wrongly predicts that both (31a) and (31b) are acceptable, 
because this analysis predicts that the first caki-ka in (31a) and (31b) is an adjunct 
and the second is the subject of the embedded S, as shown in (33). Thus the Bi- 
clausal analysis must specify additional restrictions to explain why (31a) is 
ungrammatical and (31b) is grammatical. 



Cho: Auxiliary verb constructions in Korean 1 5 

(33a)=(31a) *[Mary-ka caki-ka \caki-ka pap-ul mek-e po-ass-ta]s]s- 
(33b)=(31b) [Mary-ka caki-ka [caki-ka (kacang) yeyppe-ta-ko]s 
malha-n-ta]s. 

The VP analysis needs no such restrictions. Therefore, the VP analysis makes 
correct predictions and is preferable, whereas the Bi-clausal analysis does not 
make correct predictions. 

3. Morphological Requirements for the VP 

When an AUX subcategorizes for a VP, the VP has at least two restrictions 
on the suffix form: a restriction on the existence of tense, and a restriction on the 
suffix form for the Comp. 

First of all, the fact that only a VP with the correct suffix for the Comp can 
be grammatical shows that each AUX subcategorizes for the VP with a specific 
suffix for the Comp. The sentence (34a) is acceptable because the AUX po-ass-ta 
requires a VP with the Comp -e and the verb ilk- within the VP has the Comp -e.\ 
(34b-d) are excluded because the requirement for the Comp is not satisfied. For 
example, (34b) is ungrammatical because the verb ilk- has the wrong Comp -eya. 
Similarly, the sentence (34'a) is acceptable because the AUX ha-n-ta requires a VP 
with the Comp -eya and the verb ilk- within the VP has the Comp -eya. But (34'b- 
d) are unacceptable because the requirement for the Comp is not satisfied. 

(34) a. Mary-ka chayk-ul ilk-e po-ass-ta. 

-N book-A read-Comp try-P-Dec 
'Mary tried to read a book.' 

b. *Mary-ka chayk-ul ilk-eya po-ass-ta. 

c. *Mary-ka chayk-ul ilk-ci po-ass-ta. 

d. *Mary-ka chayk-ul ilk-key po-ass-ta. 

(34') a. Mary-ka chayk-ul ilk-0-eya ha-n-ta. 

-N book-A read-Pres-Comp must-Pres-Dec 
'Mary must read the book.' 

b. *Mary-ka chayk-ul ilk-e ha-n-ta. 

c. *Mary-ka chayk-ul ilk-key ha-n-ta. 

d. *Mary-ka chayk-ul ilk-ci ha-n-ta. 

Secondly, the fact that some AUXs require a tensed verb while some do not 
shows that each AUX subcategorizes for a VP but the existence of the tense suffix 
within the VP depends on the AUX. The sentence (35a) is acceptable because the 
verb ilk-e does not have a Tense suffix, whereas (35b) is unacceptable because the 
verb ilk-e has a tense suffix. On the other hand, if the AUX is hanta the verb mek- 
eya in the VP must have a tense suffix as in (36). 

(35) a. Mary-ka chayk-ul ilk-e po-ass-ta. 

-N book-A read-Comp try-P-Dec 

'Mary tried to read a book.' 
b. *Mary-ka chayk-ul ilk-ess-e po-ass-ta. 

read-P-Comp 



16 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



(36) a. Mary-ka pap-ul mek-0-eya ha-n-ta. 

-N rice-A eat-Pres-Comp must-Pres-Dec 
'Mary must have a meal.' 
b. Mary-ka pap-ul mek-ess-eya ha-n-ta. 
eat-P-Comp 

To account for these morphological problems there are at least two ap- 
proaches: the lexical approach and the syntactic approach. The syntactic approach 
assumes that there are several syntactic nodes, such as INFL Phrase (IP), where the 
AUX may or may not subcategorize for a IP, instead of a VP or a S, in order to ac- 
count for the requirement for the tense morpheme. The lexical approach treats the 
Tense requirement in terms of features in the syntax, where the feature and its 
value can play an important role in selecting the correct morpheme from the lexi- 
con. The next section argues for the lexical approach to solve the inflectional prob- 
lem (e.g. tense). '5 

3.2. Analysis of Tense and Comp in HPSG 

If the AUX po-ass-ta subcategorizes for a VP and a NP, we can specify the 
subcategorization information in the SUBCAT of the lexical representation of 
AUX in HPSG, as shown in (37). 



(37) ISUBCAT <NP[1], [2]VP 



IHEAD [3] l> 1 

IMARKING [4] I I 
1SUBCAT<NP[1]> I I 



However, we need to indicate that the verb within the VP must have specific 
suffixes with respect to AUX. One way to ensure the matching of the morphologi- 
cal information between the VP and the verb is to treat tense information as a 
HEAD feature and to treat the information about Comp as a marker in HPSG. For 
example, when the AUX po-ass-ta takes a VP as an argument, the base verb form 
ilk- in the VP must have Comp -e but cannot have tense in grammatical sentences 
like (38). 16 

(38) S 



V[+AUX] 




po-ass-ta 



Mary-ka 

In this case we can specify the information about the tense and the Comp for the 
VP in the subcategorization (SUBCAT) of (37). Once the morphological informa- 
tion is specified for [1]VP in (38), the information about tense must be identical 



Cho: Auxiliary verb constructions in Korean 1 7 

with [2JV in (38). The HEAD Feature Principle (HFP) in HPSG, as shown in 
(39), specifies this condition.''^ 

(39) The HEAD Feature Principle. 

In a headed phrase, the values of HEAD and HEAD-Daughter's HEAD are 
token-identical. 

The information about Comp is also specified on the VP in the SUBCAT in (37) as 
the MARKING feature and its value. If the value of MARKING is specified on 
[1]VP in the tree (38), the value triggers a sort head-marker-structure in terms of 
schema 4 in (40) and the value (Comp) is realized as a marker daughter. 

(40) Schema 4: a phrase with DTRS value of sort head-marker-structure whose 
marker daughter is a marker whose SPEC value is structure-shared with 
the SYNSEM value of the head daughter, and whose MARKING value is 
structure-shared with that of the mother. 

The sort hierarchy and feature declarations related to the information about 
tense and Comp, defined using the HFP and schema 4, as shown in (41-42). If the 
sort head as a value of the attribute HEAD in (42a) is a sort verb as a subsort of 
substantive, the sort verb must have a tense value like P(ast) for the attribute 
TENSE as in (42b). Similarly, if the sort marking as a value of the attribute 
MARKING in (42a) is a sort complementizer as a subsort of the sort marked, the 
sort complementizer must have a subsort like -e. 

(41) a. partition of head: substantive, functional 
h. partition of substantive: noun, verb, adjective, ... 

c. partition of tense: 0, P(ast), Pre(sent), ... 

d . partition of functional: marker, determiner 

e. partition of marking: unmarked, marked 

f. partition of marked.' complementizer (Comp), conjunction 

g. partition of complementizer: -e, -key, -ci, -ko, ... 

(42) a. category: IHEAD head I 

ISUBCAT list(synsem) I 

IMARKING marking I 

b. verb: [TENSE tense ] 

Under the HPSG analysis, including the HFP in (39), Schema 4 in (40), the sort 
hierarchy in (41) and the feature declarations in (42), the new tree diagram (38') 
replaces (38). The tree (38') shows that [1]VP in the head-marker structure has a 
tag [3] as the value of the attribute HEAD, including the tense information which 
is also the value for [2]VP and [4]V, indicating that the value of HEAD of the three 
categories is token-identical. Thus, the structure satisfies the HFP, and the mor- 
phological requirements for tense can be dealt with in terms of the HFP. The mor- 
phological requirement for the Comp can be dealt with in terms of the head- 
marker schema in (40). If the AUX po-ass-ta subcategorizes for a VP with the 
Comp -e, the information for the Comp is specified as the MARKING feature and 
its value as shown in [1]VP of the tree (38') where the value of MARKING is 
structure-shared with that of mother, [IJVP, by the definition (40). 



18 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 




i 



[1]VP HEAD[3]ITENSE0 

|MARKING[5]IComp -e 




V[HEAD[AUX +]] 



HFP 



[2]VPIHEAD[3]ITENSE0] MARKING[5] 



HFP 



Mary-ka 



NP 



chayk-ul 



[4]V[HEAD[3]] 



ilk 



po-ass-ta 



(Subj = Subject daughter, C = Complement daughter, 
H = Head daughter and M = Marker daughter.) 

4. SUBCAT for AUX in HPSG 

On the basis of the VP hypothesis and the morphological information for the 
VP, (43) shows the complete SUBCAT equi (type 1) and raising (type 2) AUXs 
in HPSG. 

(43) AUX: TYPE! (Equi) 

ex: po- ('try'), ... 
TYPE2 (Raising) 
ex: ci- {'become'), po- ('seem'),'^ ha- ('must'), ... 

The TYPE! AUX po- ('try') as an equi-verb has a SUBCAT and a (semantic) 
CONTENT as shown in (44). The Attribute Value Matrix (AVM) (44a) specifies 
that the AUX needs a NP and a VP to be saturated and the values of COMP and 
TENSE for the VP are -e and 0, respectively. In addition, the SUBCAT in the VP V 
indicates that the INDEX of the subject of the AUX and the VP must be identical. 
The AVM (44b) defines the CONTENT of the AUX po- where its RELATION try 
has two arguments. One is the COMMITTOR whose INDEX value is [1] and the 
other is the SOA-ARG whose value is a proposition. 

(44) a. ISUBCAT <NP[1], [2]VP IHEAD ITENSE I l> I 

I IMARKING ICOMP-^ I I I 

I ISUBCAT <NP[1]> I I 



Cho: Auxiliary verb constructions in Korean 1 9 

b. ICONTENT I RELATION po- ('try') I I 

I I COMMITTOR [1] I I 

I I SOA-ARG [2] I I 

When the AUX is in a sentence like Mary-ka pap-ul mek-e po-ass-ta, NP[1] repre- 
sents the NP Mary and the value of SOA-ARG is (45), where the INDEX [1] as a 
value of the argument role EATER refers to the INDEX of Mary and the INDEX 
[3] as a value of EATEN refers to that oi pap-ul ('rice'). 

(45) SOA-ARG: I RELATION mek- ('eat') I 

[2] I EATER [1] I 

1 EATEN [3] I 

The TYPE 2 AUX ci- ('become') has SUBCAT and CONTENT as shown in 
(46). The AVM (46a) specifies that the AUX ci- needs a NP and a AP[+PRD] to be 
saturated, and that the value of COMP for the AP is -e and that of TENSE must be 
0. The SUBCAT of the AP also indicates that the INDEX of the subject of the 
AUX and that of the AP must be identical. The AVM (46b) defines the 
CONTENT of the AUX ci- where its RELATION become has only one argument, 
SOA-ARG. 



(46) a. 



b. 



SUBCAT <NP[1], [2]AP IHEAD ITENSE I l> I 

I IPRD+ I I I 

ISUBCAT <NP[1]> I I 

IMARKINGICOMP-^ I I 

CONTENT I RELATION d- ('become') I I 
I SOA-ARG [2] I I 



When the AUX is in a sentence like Mary-ka yeyppu-e ci-ass-ta ('Mary became 
pretty.'), the INDEX value of the subject NP of the AUX in SUBCAT refers to 
that of the NP Mary and the SOA-ARG in (46b) is as represented in (47). For the 
SOA-ARG in (47), the INDEX value of the argument role INSTANCE and that of 
the NP Mary in SUBCAT are identical. 

(47) SOA-ARG: [2] I RELATION yeypp- ('pretty') I 

I INSTANCE [i] I 

The AUX po- ('seem'), as a member of TYPE2, has a different SUBCAT than 
other members of TYPE 2 like ci- ('become'). Its SUBCAT takes a VP as one of its 
arguments while the AUX ci- ('become') needs a AP to be saturated. The lexical 
representation for the AUX po- ('seem') is as represented in (48). The SUBCAT in 
(48a) specifies that the AUX needs a NP and a VP to be saturated, the value of 
COMP for the VP is -na and the value of TENSE can be Pres(ent) or P{ast). Again, 
the AVM (48b) defines the CONTENT of the AUX po- where its RELATION 
seem has only one argument, SOA-ARG. 

(48) a. ISUBCAT <NP[1], [2]VP IHEAD [TENSE Pres V P] l> I 

I ISUBCAT <NP[1]> I I 

I IMARKINGICOMP -na I I I 



20 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

b. [CONTENT I RELATION po- ('seem') I I 

I I SOA-ARG [2] I I 

When the AUX exists in a sentence like Mary-ka chayk-ul ilk-na po-ta ('Mary 
seems to read a book'), the SOA-ARG in (48b) can be represented as in (49). 

(49) SOA-ARG: I RELATION ilk- ('read') I 

[2] I READER [1] I 

I READED [3] I 

4.1. Evidence against Phrasal analyses 

Cho 1988 proposes an argument against Phrasal analyses on the basis of the 
impossibility of Scrambling of the VP subcategorized for by AUX and the impos- 
sibility of adverbs occurring in front of AUX.'^ The following will show how 
these phenomena can be accounted for in HPSG. 

Cho argues that (50a) can be scrambled in Korean, with the result (50b), 
whereas (51a) cannot be scrambled as in (51b). According to her explanation, 
(51b) is unacceptable because mek-e po-ass-ta in (51a) is a compound verb and the 
NP pap-ul and the verb mek-e cannot be a constituent, so that the string pap-ul 
mek-e cannot undergo Scrambling. 

(50) a. John-un [Suni-ka ka-ass-ta-kojg sayngkakha-ess-ta. 

John-top Suni-N go-P-Dec-Comp think-P-Dec 

'John thought Suni went away.' 
b. [Suni-ka ka-ass-ta-kojg John-un sayngkakha-ess-ta 

(51) a. John-un [pap-ul mek-e] po-ass-ta 

John-top rice-A eat-Comp try-P-Dec 

'John tried to have a meal.' 
b. *[ pap-ul mek-e] John-un po-ass-ta 

The evidence that the S Suni-ka ka-ass-ta-ko in (50) can be scrambled but the S 
pap-ul mek-e in (51) cannot is a crucial argument against the Bi-clausal analysis, 
because the difference in acceptability between (50b) and (51b) must be 
stipulated. 

However, the VP analysis predicts that (50b) is possible while (51b) is im- 
possible. The fact that the sentences, where predicative categories such as VP or 
AP[-I-PRD] or NP[-(-PRD] undergo Scrambling, are unacceptable shows that the 
VP subcategorized for by AUX also cannot undergo Scrambling. The sentence 
(52), which contains a small clause, is ungrammatical because the NP[-i-PRD] 
papo-lul is scrambled with the result in (52b). To deal with the scrambling prob- 
lem in a small clause, Yoo 1993 proposes a Linear Precedence (LP) rule specifying 
that any predicative category cannot precede its subject, as shown in (53). This 
LP prevents the NP Mary-lul and the NP[-i-PRD] papo-lul in (52a) from 
Scrambling. The LP rule states that if there is any predicative category, like a VP, 
which needs only a SUBJECT to be saturated, that predicative category cannot 
precede the SUBJECT. So the independently motivated LP predicts that (50b) is 
possible and (51b) is not. In other words, the constituent Suni-ka ka-ass-ta-ko in 



Cho: Auxiliary verb constructions in Korean 2 1 

(50) is a S so that it has no need to observe the LP whereas pap-ul mek-e is a VP 
under the VP analysis which must. 

(52) a. John-i Mary-lul papo-lulAo mantul-ess-ta. 

-N -A fool-A/-PP make-P-Dec 

'John made Mary a fool.' 

b. * John-i papo-lulAo Mary-lul mantul-ess-ta.^o 
-N -A/PP -A make-P-Dec 

'John made Mary a fool.' 

(53) [1] < [VALENCE1SUBJ<[1]>] 

Under the VP analysis, the difference between (50b) and (51b) naturally follows 
from the LP. Therefore, the evidence in (50-51) is not a counter-example to the VP 
analysis but, rather, supports the claim that AUX subcategorizes for a VP, not a S, 
in Korean. 

Cho 1988 also claims that adverbs like cacwu ('often') cannot occur in front 
of the AUX as shown in (54). On the basis of the fact that the adverb cannot inter- 
vene between the verb mek-e and the AUX po-ass-ta, she claims that the verb and 
the AUX form a compound verb so that the adverb cannot modify the AUX and 
argues that Phrasal analyses are implausible since this problem is difficult to solve 
under these analyses. 

(54) *John-i pap-ul mek-e cacwu po-ass-ta. 

-N -A eat-Comp often try-P-Dec 

'John often tried to have a meal.' 

However, the fact that the adverb can occur between a verb and some AUX 
shows that her argument against Phrasal analyses is untenable. The sentence with 
the causative AUX ha-ta ('cause') is acceptable even if an adverb occurs between 
the two, as shown in (55). 

(55) John-i chayk-ul ilk-key cacwu ha-n-ta. 
John-N book-A read often cause-Pres-Dec 
'John often causes someone to read a book.' 

If sentence (54) is considered unacceptable but sentence (55) is considered accept- 
able, even though both sentences have an adverb modifying AUX, the argument 
for the Compound Verb analysis is not valid. 2' 

Therefore, Cho's argument against Phrasal analyses is not tenable for the VP 
analysis. 

5. Conclusion 

The fact that, in spite of the Compound Verb analysts' claims, AUXs are an 
independent category has been shown through constituency tests such as scope in- 
terpretations, kuray + verb constructions and verbal fronting in sec 2.1. Section 
2.2 showed that AUX subcategorizes for a VP rather than a S by demonstrating 
that against the Bi-clausal analysts' claims, sentences with NPI or the reflexive 
caki-ka can be evidence for the VP analysis. Thus, this supports the claim that the 



22 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

VP analysis is more plausible than both the Compound Verb analysis and the Bi- 
clausal analysis in explaining the AUX constructions. Section 3 claimed that the 
morphological requirements for the VP can be accounted for in terms of the HFP 
and schema 4 only if tense and Comp are regarded as a value of HEAD feature and a 
marker, respectively. On the basis of section 2. and 3. section 4 presented the pro- 
posal for two types of AUXs, equi- and raising-AUX, under the HPSG 
framework. 

Consequently, AUX constructions can be sufficiently accounted for by the 
VP analysis. Furthermore, the Compound Verb analysis must not be assumed to 
be the only hypothesis to explain these constructions in HPSG. Rather, if my 
analysis is chosen, the restrictions for the AUX constructions can be regarded as a 
subcase of the restrictions for the VP. 

NOTES 



* I am grateful to professors Georgia Green, Jerry Morgan and James Yoon 
for their valuable comments. 

1 Following Choi 1971 I will call an auxiliary verb the last one of the 
italicized two sequencing verbs in the data. For example, po-ass-ta will be an 
auxiliary verb in (la). 

2 In this paper I refer to Pollard & Sag 1994 as HPSG 1994 and Pollard &. 
Sag 1987 as HPSG 1987. 

3 In describing Korean sentences, I will use the Yale Romanization System. 

^ N stands for Nominative, A Accusative, Comp Complementizer, P Past, 
Pres Present, Dec Declarative, Neg Negation, Q Question, and T Topic. 

2 
5 Both VP and S are called V in this paper. 

^Yoo 1993 simply adopts Sells' analysis in studying subcategorization in 
Korean. 

"7 My analysis will cover both raising and equi auxiliary verbs, even though 
in this paper I mainly deal with the AUX in (la). 

8 Sells 1991 proposes a semantic analysis of the scope problem. His analysis 
is based on the classification of event types of each AUX. However, it is not clear 
that his analysis can predict the scope problem in coordination. 

9 Kule is a variant form for kuray. 

'0 NM stands for Nominalizing Suffix. 

1' For me, (24) is unacceptable. Regardless of bridge verbs like sayngkakha- 
('think') or non-bridge verbs like seltukha- in (24), the interpretation where the 
NPI negates the embedded sentence is almost impossible. However, I think we can 
get the interpretation where the NPI only negates the higher sentence. For 
example, (24) can have a interpretation like John didn't persuade Sue anything 
about the fact that Mary ate (something). 



Cho: Auxiliary verb constructions in Korean 2 3 

'2 Sells 1993 claims that a Japanese phrase like [ X chayk-ul ilk-e] in ( 17=la) 
is a sub-clause, showing the distributional behaviour of NP for the X position in 
(25). 

'3 The sentence (31a) may improve a little bit with a pause between two caki- 
ka. Still, it is unacceptable for me. On the other hand, (31b) is better in 
acceptability. 

•* The verb malha- takes a NP, a S and an optional PP as its arguments. 

'5 I call my approach the lexical approach in that the information about the 
required suffixes is specified as features. 

•^ This tree is only an abbreviation for Attribute Value Matrix. 

'^ The definition of HFP in (39) is a simplified one to enhance the 
readability. The original definition of the HFP is as follows: 

The HFP. 

In a headed phrase, the values of SYNSEMI LOCAL! CATEGORYI HEAD 
and DAUGHTERSI HEAD-DAUGHTER! SYNSEM! LOCAL! CATEGORYI 
HEAD are token-identical, cf. HPSG (1994:491) 

'8 Even though the raising- AUX po-('seem') has the same phonological base 
form as the equi-AUX po- ('try'), die two AUXs are different words. 

" Cho 1988 uses the term Phrasal analyses to refer to both the VP analysis 
and the Bi-clausal analysis. 

20 The PP stands for Postposition. 

21 The question why (55) is good but (54) is bad is open to further study. 

REFERENCES 

CHO, J. O. 1988. Suffixed verb forms and compound verb constructions. In 

Eung-Jin Baek(ed.). Papers from the Sixth International Conference on 

Korean Linguistics. International Circle of Korean Linguistics 6: 77-106. 
CHOI, H. B. 1971. Korean grammar. Seoul: Chengumsa. 
CHOMSKY, N. 1981. Lectures on government and binding. Dordrecht: Foris 

Publications. 
PARK, B. S. 1990. Contrastive analysis of lexical, morphological and syntactic 

structures for English-Korean machine translation. Kyung Hee Language 

Institute. 
POLLARD, C. and I. Sag. 1987. Information-based syntax and semantics, Vol. 1. 

CSLI, Stanford University. 
. 1994. Head-driven phrase structure grammar. University of Chicago Press 

and CSLI Publications. 
SELLS, P. 1991. Complex verbs and argument structures in Korean. Harvard 

Studies in Korean Linguistics 4: 395-406. 
. 1991. Korean and Japanese morphology from a lexical perspective. MS., 

Stanford University. 
. 1993. The projection of phrase structure and argument structure in 

Japanese. MS., Stanford University. 



24 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

YOO, E. J. 1993. Subcategorization and case marking in Korean. In A. Kathol and 

C. Pollard eds. Ohio-State University Working Papers in Linguistics 42: 

178-198. 
YOON, J. H.-S. 1987. Reconciling lexical integrity with affixation in syntax. In 

J. McDonough and B. Plunkett eds. Proceedings of North Eastern 

Linguistic Society 17: 663-683. 
. 1993. The headedness of inflectional structures in Korean. Seoul 

International Conference on Linguistics 12: 826-838. 



Studies in the Lint^iiistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

CRITICIZING WITH A QUESTION 

Cassandre Creswell 

People in intimate, informal relationships frequently use questions 
as an indirect method to criticize an undesirable situation, particularly 
one that is already in existence and contrary to what the speaker wishes, 
but in accordance with what the speaker believes the addressee wants, as 
in example ( 1 ): 

(1) Are we going this way? (/A/i/j/Zcar/rtg 'This is not a desirable way 
to go.') 

The purpose of this paper is to explain how and why questions of this 
kind can be used to implicate criticisms. 

1. Introduction 

The first section discusses how these questions are distinguished from sin- 
cere questions with no implicature of criticism. I claim the crucial criteria on 
which correct interpretation depends are the beliefs of the addressee, and show 
how these beliefs result in a pattern of inferences that lead to correct or incorrect 
interpretation of the question as a criticism. The second section identifies why a 
speaker's use of these questions is a strategy of politeness. Briefly, these questions 
allow the speaker to refrain from directly performing a face-threatening act in 
order to satisfy the negative face of the addressee (Brown and Levinson 1987), and 
they allow the speaker to offer options by giving the addressee more than one 
possiblity of how to react (Lakoff 1973). Both of these reasons are subsumed by 
the more fundamental desire of the speaker to preserve an informal, intimate 
relationship with the addressee. The final section compares rephrasings of the 
question in alternate forms with a different quality of politeness but with the same 
ability to criticize some state of affairs. This comparison explains why a speaker 
chooses to use a question, rather than some other form, to express criticism. 

Form of the Question 

Although an explanation of how criticism-implying questions function 
should be universally applicable to all questions that implicate criticism, in this 
paper I restrict the class of questions considered to positively phrased, yes-no 
questions that use forms of he or do. This restriction of class allows a more fo- 
cused and effective explanation of the logic of my argument. 

The criterion of positive phrasing eliminates questions with the n 't contrac- 
tion as in (2), but allows questions like (3) where the not occurs in a position other 
than the position immediately following the initial element: 

(2) Aren't you going to get a haircut? 

(3) Are you not going to get a haircut? 



26 Studies in the Linguistic Sciences 23:2 ( Fall 1993) 

Question (2) can make either an implicature that the speaker believes that the ad- 
dressee intends to get a haircut or that the addressee intends to refrain from getting 
a haircut and in either case, the addressee's intention is inconsistent with another 
assumption of the speaker. Question (3), however, can only implicate that the 
speaker believes the addressee intends to refrain from getting a haircut. This dou- 
ble implicature of (2) with its conflicting possible interpretations for exactly what 
state of affairs is being criticized interferes with a clear characterization of the nec- 
essary inferences the addressee makes. Therefore, including negatively-phrased 
questions in this analysis would unnecessarily complicate the characterization. 

No restriction on tense of verb or person of the subject is motivated. Earlier 
examples — (3), using second person, and (1), using first person plural — and fur- 
ther examples (4-6) using third person and a variety of tenses can all be used to 
implicate criticism: 

(4) Does that shirt have to be washed? 

(5) Did that shirt have to be washed? 

(6) Will that shirt have to be washed? 

Variations in person and tense do not affect the usefulness of a question for impli- 
cating criticism. Neither do they complicate the description of the inferences an 
addressee makes, and so they are not restricted in this characterization. 

2. Distinguishing Criteria: Beliefs and Inferences of the Addressee 

Although criticizing with a question is an act of a speaker, the characteriza- 
tion of criticism-implicating questions is explained from an addressee's, rather 
than a speaker's, perspective. Questions can only be successfully used to criticize 
if they are interpreted as implicating criticism. Correct interpretation depends on 
the beliefs of the addressee about the speaker's intentions, not the speaker's inten- 
tions themselves. The addressee must be able to distinguish a critical question 
from a sincere request for information. The difference can only be perceived if the 
addressee holds certain beliefs. The absence of these beliefs will prevent the ad- 
dressee from making the inferences necessary for a correct interpretation of the 
question. Therefore, characterizing the beliefs and inferences of the addressee ac- 
counts for both successful and unsuccessful interpretation of questions that imply 
criticism. 

The addressee must hold two beliefs for interpretation of criticism-implying 
questions. First, the addressee must believe that the speaker believes she' knows 
the answer to the question, and second, the addressee must believe that the speaker 
intends for the addressee to believe that the speaker knows the answer. In the ab- 
sence of these beliefs, misinterpretation occurs. In order to illustrate how different 
interpretations can be generated, different beliefs of the addressee will be matched 
with the use of an example question in a given situation. In the situation in (7), 
John is holding a really ugly shirt. Mary asks John: 

(7) Are you going to wear that shirt? 



Creswell: Criticizing with a question 2 7 

First, if John does not beHeve that Mary believes he does intend to wear that shirt, 
then he cannot correctly interpret the question as a criticism. One possibility for a 
non-criticism interpretation occurs if John believes that Mary believes that he 
does not intend to wear the shirt. The utterance will be almost nonsensical; John 
will have a difficult time thinking of any interpretation of it.2 A second possibil- 
ity for non-criticism interpretation occurs if John believes that Mary holds no be- 
lief about whether or not he intends to wear the shirt. In this case then, the utter- 
ance must be a sincere question because Mary wants information, i.e. she wants to 
know what John intends to do. 

Correct interpretation as implying criticism can only occur if, as stated 
above, the addressee believes the speaker believes she knows the answer. This in- 
ference is made through assessing the relevance of the question. When a speaker 
asks a question with an answer she already knows, she is apparently violating 
Grice's Cooperative Principle that conversational contributions must follow the 
accepted purpose or direction of the exchange (Grice 1975). A question with a 
known answer makes no readily recognized contribution to a conversation. No 
obvious reason exists for the speaker to ask something she already knows. As a ra- 
tional human being engaged in conversation, the addressee assumes that the 
speaker is following the Cooperative Principle and, therefore, a reason does exist 
for her utterance. He then constructs a reason for asking such a question, and in- 
fers what implicature the speaker desired to make through her use of the question. 

The chain of reasoning that leads to an implicature of criticism can best be 
outlined in combination with the use of an example, such as (7) in the above situa- 
tion. This chain begins with John's beliefs: one, that Mary knows the answer to 
(7), that is she already thinks 'Yes, he is going to wear the shirt,' and two, that 
Mary intends for him to believe that she knows this answer. Holding these two be- 
liefs, John cannot regard the question as a sincere request for information because 
Mary apparently already possesses the information. Nonetheless, if Mary is asking 
about the wearing of the shirt, it must have some relevance.-^ Questioning a state of 
affairs, the wearing of a certain shirt in this situation, regarded as definitely true 
could hardly be rational if the speaker is in full support of such a state of affairs. 
On the other hand if the speaker is unhappy with a state of affairs, calling attention 
to the situation through asking an obvious question is perfectly rational because, 
if the addressee is aware of the speaker's unhappiness with a situation, he may try 
to rectify the situation in accordance with the speaker's wishes. So, the next 
inference the addressee makes is that the relevance of the question lies in its ability 
to call attention to a state of affairs and make the speaker's unhappiness known. 
Making one's unhappiness about a state of affairs known is a very simple 
definition of criticism. 

The example situation can demonstrate this final part of the chain of inference 
too. John must interpret the apparent irrelevance of Mary's question about his 
wearing of the shirt as Mary calling attention to his plan to wear the ugly shirt in 
order to implicate her unhappiness with this state of affairs and her wish to make 
this unhappiness known. If John does follow this chain of inferences, he can cor- 



28 Studies in the Linguistic Sciences 23:2 ( Fall 1993) 

rectly interpret Mary's question as an implicature of criticism, specifically the 
implicature in (8): 

(8) 'I don't think you should wear that shirt.' 

The chain of inferences made by the addressee in this particular situation can 
be applied in any other situation in which questions are used to implicate criti- 
cisms. The following example situation will further demonstrate the process of in- 
terpretation. In this case, John has been repeatedly cracking his knuckles for the 
last five minutes. Mary asks John (9). 

(9) Do you have to do that? 

The steps of reasoning John needs to follow to interpret (9) as criticism-implicat- 
ing are very similar to the ones in the situation explicated above. First, John must 
believe that Mary already knows the answer to her question; she thinks 'No, there 
is no compelling reason for John to be cracking his knuckles.' He must also be- 
lieve that Mary intends for him to believe that she knows this answer. Because he 
thinks she knows the answer already, the question cannot be a sincere request for 
information, and he must construct an alternate explanation for Mary's asking it. 
Because the answer to her question is obvious, John must interpret her asking it as 
a way to call attention to the state of affairs she is questioning. While questioning a 
situation that Mary is in full support of would not be regarded as rational, ques- 
tioning a situation that she is displeased with seems reasonable. So, John interprets 
Mary's questioning of the necessity of his cracking his knuckles as implying that 
it bothers her, more directly stated as the assertion in (10): 

(10) 'You don't have to crack your knuckles, and it bothers me that you are doing 
so.' 

The chain of reasoning used in interpreting criticism-implying questions in terms 
of the beliefs of the addressee can be generalized and applied to different uses of 
this type of question in order to explain their implicatures. In addition, this chain 
supports the characterization of this type by explaining how the beliefs of the ad- 
dressee distinguish criticism-implicating questions from sincere information- 
requesting ones. 

3. Motivations for Use 

If criticizing with a question is a politeness strategy, it must be consistent 
with the principles of a general theory of politeness. Politeness in discourse can be 
regarded in a very general way as a means to maintain or change interpersonal rela- 
tions (Green 1989). Because the act of criticizing may disrupt a given level of in- 
terpersonal relations, in order to be polite a speaker will try to minimize this dis- 
ruption. Speakers want to maintain and change relations even within their most 
informal and intimate relationships, the kind in which criticism-implying ques- 
tions are frequently used. 

This desire to preserve the speaker's informal, intimate relationship with the 
addressee motivates two considerations for her. First, she must refrain from di- 
rectly threatening the addressee's negative face by showing respect for the 



Creswell: Criticizing with a question 2 9 

addressee's self-image and desire for freedom of action. Second, she must offer the 
addressee options by allowing the addressee more than one possibility of how to 
react to the criticism. One means of acting in accordance with these two consid- 
erations is the criticism-implying question. 

Threatening the Addressee 

Sincere questions pose a threat to the addressee's face only in that they expect 
him to use his time to answer them and to know what the answer is. Criticism- 
implying questions are much more threatening because they express the speaker's 
doubts and displeasure about a state of affairs the addressee is presumed to be re- 
sponsible for or able to rectify. Because it is non-threatening, a sincere question 
does not require the speaker to apologize when the addressee offers an unantici- 
pated response, although she could offer an apology for imposing upon the ad- 
dressee, as in the dialog in ( 1 1 ): 

(11) Sue: Sorry to bother you, but did you let the cat out? 
Matt: No, I didn't. 

Sue: Oh, okay. Just wondering. 

In contrast, if the question is to implicate criticism, an unexpected answer will 
merit an apology, as in (12), where Sue has discovered the cat outside: 

( 1 2) Sue: Did you let the cat out? (implying 'The cat ought not to have been let 

out.') 
Matt: No, I didn't. 
Sue: Oh, sorry. 

The mistaken criticism results in an apology by the speaker for making an unnec- 
essary threat. An initial apology like that in (II) seems incongruent when 
matched with a question the speaker is using to criticize, as in example (13): 

(13)1 don't mean to bother you, but are you really going out of the house in that 
hideous shirt for the third day in a row? 

The combination of a statement that mitigates threat, / don't mean to bother you, 
with a question that strongly implicates criticism, is self-contradictory and will 
probably result in a conscious attempt by the addressee to assess the speaker's rea- 
son for using such a combination. The assessment he generates may be that the 
speaker's use of the first is entirely insincere and used for a sarcastic effect or, 
along opposite lines, that she is in fact sincerely concerned and desires more in- 
formation about actions she regards as unusual. 

As shown above, the criticism-implying question does threaten the addres- 
see's face to a greater degree than its information-requesting counterpart does. The 
speaker mitigates the threatening aspect of criticism by implicating rather than 
asserting it. Because more than one communicative intention can be inferred from 
a criticism-implicating question, it is done "off-record", a strategy of negative 
politeness behavior, as characterized in Brown and Levinson 1987. An off-record 
strategy does not commit a speaker to a face-threatening act as strongly as one done 



30 Studies in the Linguistic Sciences 23:2 ( Fall 1993) 

on-record, and so the speaker can use this type of question to criticize with less 
risk of disrupting her relationship with the addressee. 

Offering the Addressee Options 

The criticism-implicating question's surface resemblance to an information- 
requesting question means it offers the addressee options. The speaker's implica- 
ture can be ignored if the addressee disregards his own beliefs about the speaker's 
beliefs and responds as if the question was a request for information. When the ad- 
dressee takes this option, it results in a discourse like (14): 

(14) Mary: Are you going to wear that shirt? (attempting to implicate 'That's an 

ugly shirt and I don't think you should wear it.') 
John: Yes, I am. 

With an affirmative response here, Mary must assume either that John did not un- 
derstand what she was attempting to imply with her question or that John ignored 
her implicature deliberately. In either case, in order to attempt to remedy John's 
misunderstanding or to emphasize her unhappiness with the state of affairs, Mary 
might respond with a more direct statement of what she intended to implicate, as 
in (15): 

(15) Well, I don't think you should. It's an ugly shirt. 

The appropriateness of responding to the exchange in (14) with (15) supports the 
claim that although the implicature of (14) can be ignored, intentionally or unin- 
tentionally, its existence can be reaffirmed if the speaker asserts it directly. As 
discussed in the previous section, by offering the option of ignoring its criticism 
the criticism-implying question can function as a politeness strategy, allowing the 
speaker to preserve a relationship and still voice a potentially threatening 
criticism. 

4. Other Forms that Criticize 

A speaker uses a criticism-implicating question as a politeness strategy to 
mitigate threatening criticism. A criticism that differs in form may also differ in 
its politeness. Criticism in question form is more polite, i.e., less threatening of 
others' beliefs and values, than other forms that do not offer the same kinds of op- 
tions in interpretation. As explained above, a question gives the addressee the op- 
tion of ignoring the implicature of criticism and interpreting the question as a 
sincere request for information because the criticism is only implied, not directly 
expressed. When a speaker wants her utterance to be less polite, she will use more 
directness in stating the criticism, as in examples (16-20). The example decrease 
in directness from (16) to (20): 

(16) Inviting John to the reception is wrong and it makes me unhappy. 

(17) I can't believe John is invited to the reception. 

(18) John is invited to the reception?! 

(19) Is John invited to the reception? 

(20) I see John is invited to the reception. 



Creswell: Criticizing with a question 3 1 

Example (16), because it directly states the criticism, allows the addressee no op- 
tions in interpretation. Example (17) could be interpreted literally as a statement 
of the speaker's disbelief, but the high degree of conventionalization of the impli- 
cature of / can't believe as 'I am surprised and/or dismayed that such a state of af- 
fairs exists' makes the literal interpretation unlikely. Although (18), as an excla- 
mation, only implicates the criticism, it cannot be treated as a sincere request for 
information, and as a question, it will be interpreted by way of the same reasoning 
as any other criticism-implicating question. In contrast, when uttered with the 
typical rising intonation of a question, (19) cannot be treated as an exclamation, 
only as a question, making it less directly critical and more polite. The statement 
of fact, (20), can implicate the criticism in much the same way as the question 
(19); by calling attention to the obvious, it will lead to a very similar chain of im- 
plicature. These rephrasings of a single criticism in multiple ways, one of which 
is in the form of the criticism-implicating question, differ in politeness because 
they differ in directness. The fact that each form can express the same criticism 
provides proof that the questions being characterized can be used to criticize. The 
fact that each expresses a different degree of politeness provides a reason for a 
speaker to choose to use one form, such as the question, instead of one of the 
others. 

5. Conclusion 

This paper has demonstrated how positively-phrased, yes-no questions that 
begin with a form of be or do can be used to implicate criticism. First, the specific 
beliefs and inferences of the addressee necessary for interpreting this kind of ques- 
tion as implying criticism rather than requesting information were made explicit. 
Second, the use of this type of question as a politeness strategy was explained. 
Finally, rephrasings of different degrees of politeness were compared with 
criticism-implicating questions. Together these three sections provide a thorough 
characterization of how and why a speaker criticizes with a question. 

NOTES 

' Throughout the paper the speaker will be referred to as female and the ad- 
dressee as male, in accordance with the sex of the participants in the examples 
used. 

- Nonetheless, the natural inclination to interpret speech and in fact human 
behavior in general as rational, i.e., done with a reason, discussed in Green 1993 
means that he will still attempt to construct an interpretation and a reason for her 
apparently nonsensical utterance. 

^ This claim seems quite similar to that of Sperbcr and Wilson 1987, in 
which they assert that "a speaker who asks a question ... indicates that some rele- 
\ant completion of the incomplete thought represented by her utterance is rele- 
\ ant." Disregarding the controversial definition of relevance outlined therein and 
instead thinking of relevance merely as "conforming to Grice's Maxim of 



32 Studies in the Linguistic Sciences 23:2 ( Fall 1993) 

Relevance," this assertion is a good characterization of the chain of reasoning be- 
hind criticism interpretation. 

REFERENCES 

BROWN, Penelope, and Stephen Levinson. 1987. Politeness: Some universals in 

language use. Cambridge University Press. 
GREEN, Georgia. 1989. Pragmatics and natural language understanding. 

Hillsdale, NJ: Erlbaum. 
. 1993. Rationality and Gricean inference. Cognitive Science Technical 

Report, Beckman Institute, University of Illinois at Urbana-Champaign. 
GRICE, H. Paul. 1975. Logic and conversation. Syntax and semantics, vol. 3: 

Speech acts, ed. by Peter Cole & Jerry L. Morgan, 41-58. New York: 

Academic Press. 
LAKOFF, Robin. 1973. The logic of politeness; Or, minding your P's and Q's. 

Papers from the Ninth Regional Meeting of the Chicago Linguistic 

Society, ed. by C. Corum, T. C. Smith-Stark, & A. Weiser, 292-305. 

Chicago: Chicago Linguistic Society. 
SPERBER, Dan, and Wilson, Deirdre. 1987. Precis of Relevance: Communication 

and cognition. Behavioral and Brain Sciences 10: 697-710. 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

VOWEL DELETION IN PULAAR: RIME AND NUCLEAR 
MERGERS AND THE ISSUE OF THE SYNTAX-PHONOLOGY 

INTERFACE* 

Abdul Aziz Diop 

In this paper I am going to analyze vowel deletion (VD, hence- 
forth) in Pulaar, the Mauritanian dialect of Fula. The analysis has two 
parts. First, after some background discussion of Pulaar vowels and 
Pulaar syllables I present some data and suggest a phonological analysis 
in the form of rime and nucleus mergers. Second, I present data that 
suggest that in establishing its domain of application, VD is sensitive 
to syntactic information. In the literature on syntax-phonology inter- 
face we have two main approaches: the direct-access (Clements 1978; 
Kaisse 1987; Odden 1990), and the indirect-access (Hyman 1990; 
Selkirk 1986, 1987, 1990). In this paper I will demonstrate that the 
generalization about the Pulaar data is not consistent with the basic 
tenets of either approach and that an additional statement in the formu- 
lation of the rule, along with either approach, will account for the data. 
There is a third approach in the interface between syntax and phonol- 
ogy that space would not permit to go into, however. It is raised in 
Kenstowicz (1987:229) and has to do with whether the application or 
blockage of (phrasal) phonological rules 'can tell us something about 
the surface syntactic structure — in particular something that we did 
not know already'. 

1. Introduction 

The aim of this paper is to provide an account for all the conditions under 
which vowel deletion and vowel spreading take place in Pulaar. The paper is or- 
ganized as follows: first, after a brief introduction of the vowel system I present the 
phenomenon from a descriptive standpoint in order to show which vowel deletes, 
and what the output of deletion is; then I give some background information per- 
tinent to the topic of the paper. Second, I present my analysis of VD in Pulaar. 
Third, I present two sets of data that seem to be problematic for my analysis in the 
sense of the rule failing to apply. For the first set I show that failure of the rule to 
apply has to do with prosodic information whereas with the second set I propose a 
solution within the syntax-phonology approach to domain definition. I conclude 
by demonstrating that the two major approaches to this theory of domain defini- 
tion, in ihcir current formulation, cannot handle the Pulaar data. 



3 4 Studies in tiie Linguistic Sciences 23:2 (Fall 1993) 

2. Pulaar vowel deletion: a description 

2.1. The vowels of Pulaar 

First, an inventory of Pulaar vowels. Pulaar has five phonemic and seven 
phonetic vowels illustrated below in (la) and (lb), respectively. 

(1) Pulaar vowel inventory 

a. Phonemic: I'll, lul, I el, I o/, and /a/ 

b. Phonetic: III, lul. Id, I eJ, lol, / o/, and /a/ 

Each of these vowels has its long counterpart but the distinction between a 
long vowel and short vowel is unpredictable, as illustrated in (2) where the mean- 
ing of two otherwise similar words only differs because they have one vowel real- 
ized as short in one word and long in the other member of the pair. 

(2) Contrastive vowel length in Pulaar 

nol-de 'to be rotten' iiool-de 'to win' 

lu6-de 'to lend' luu6-de 'to smell bad' 

hir-de 'to be jealous' hiir-de 'to be late (sp.)' 

nan-e 'left-plural' naane 'earlier' 

fere 'expense' fee-re 'manner (spec.)' 

In Pulaar the only environment in which a long vowel is predictable is when 
Ihl or the glottal stop (/?/) is deleted (in coda position) causing the preceding 
vowel to lengthen. (3a-b) illustrate this. There are no complex onsets or codas in 
the language. In (3a) we have nominal roots followed by consonant-initial noun 
class agreement markers. The Ihl or the glottal stop 111 deletes and its mora is as- 
signed to the preceding short vowel, making it realized as a long vowel. In (3b) 
the same roots are used either with consonant-initial noun class agreement markers 
whose initial consonants have been deleted (cf. Paradis (1986, 1992) for a discus- 
sion and an analysis of such initial deletion) (I call these vowel-initial markers for 
expository purposes), or with (vowel-initial) aspectual markers. The Ihl or I II ax t 
then syllabified as onsets, not as codas. They do not delete in this position; there- 
fore the vowel that precedes them does not lengthen. 

(3) Predictable vowel length in Pulaar 

a. Root -I- consonant-initial markers 



/wah-re/ 


[waa-re] 


'beard' 


/yah-re/ 


[yaa-re] 


'scorpion' 


/mah-de/ 


[maa-de] 


'to build' 


/sah-de/ 


[saa-de] 


'to fry' 


/fa? -de/ 


[faa-de] 


'to be headed for' 


/ho7-re/ 


[hoo-re] 


'head' 


/fi7-de/ 


[fii-de] 


'to beat up' 


i. Root + 'vowel' 


-initial markers 




/wah-e/ 


[ba-he] 


'beard'-pl. 


/mah-i/ 


[ma-hi] 


'build'-past 


/sah-i/ 


[sa-hi] 


'fry'-past 


/fa7-i/ 


[fa-7i] 


'be headed for'-past 


/fi7-i/ 


[fi-7i] 


'beat up'-past 



Diop: Vowel deletion in Pulaar 



35 



Long/short vowels can occur freely in the word, as shown by (4). The repre- 
sentation we give for Pulaar vowels is as in (5) (Goldsmith 1990) where (5a) is the 
underlying phonological representation and (5b) the redundancy rule that cap- 
tures the fact that the feature [back] is predictable for Pulaar vowels. 

(4) Distribution 

Beginning Middle End 

ekkaade delep hare 

eewnaade faliima kataa 

(5) Representation of vowels (Goldsmith 1990) 



a. Representation 










[-round] 
1 


[-(-round] 
1 


[-i-round] 
1 


[-round] 
1 




1 
X 


1 
X 




1 

X 

1 


1 

X 

1 


X 

1 








1 
[low] 


1 
[low] 


1 
[low] 


i 


u 







e 


a 


b. Redundancy 


rule 










[a round] 

1 


-» 




[a round] 

1 






1 
X 






1 
X 







[a back ] 

Having gone through the vowel inventory of Pulaar I now present vowel 
deletion in its descriptive form and discuss some of the syllable-related issues that 
are central to the analysis given in this paper. (For the remainder of this paper I 
will not distinguish between [-f-ATR] and [-ATR] vowels orthographically be- 
cause they are irrelevant for the present purposes.) 

In Pulaar, vowel deletion is observed only across morpheme boundaries. It 
does not take place within the morpheme itself because intramorphemically one 
never finds sequences of different vowels. In a configuration where two mor- 
phemes are 'adjacent'; given that the first morpheme ends in a vowel and the sec- 
ond one starts with a vowel the final vowel of the first morpheme deletes provided 
the latter is not a major lexical category, i.e. a verb, a noun, or an adjective. 
Typically, the morphemes in second position are: subject pronouns, conjunctions, 
prepositions, and the vowel-initial possessive pronoun /am/ 'my'. When the final 
vowel of the first morpheme deletes (across morpheme boundaries) it is the features 
of the initial vowel of the second vowel that arc 'kept'. This is the reason why 1 re- 
fer to the phenomenon as deletion rather than coalescence. However, I will not 
dwell on this point as the data will illustrate it even better. In cases where the first 
morpheme (alpha) is consonant-final its final consonant becomes the onset of the 
initial vowel of the second morpheme (beta). In section 2.2., (6)-(8) below I de- 



3 6 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

scribe the output of vowel deletion and the relevant data (the target vowels are in 
bold). 

2.2. The output of vowel deletion in Pulaar 

A long vowel is produced either when the (short) vowels of the two mor- 
phemes involved are identical, when the vowel of the first morpheme is high (or 
high and long), or when final /o/ of the first morpheme is deleted in front of /a/ of 
the second morpheme, as shown by (6a-f). 

(6) a. hannde-e-janngo -> hanndeejanngo 

today-conj -tomorrow 
'today and tomorrow' 

b. Mali-e-Moritani -^ Maalee Moritani 
Mali-conj-Mauritania 

'Mali and Mauritania' 

c. 6ayri-o-yim-at — > bayrooyimat 
since-3sg-sing-Asp. 

'since he sings' 

d. sabu-a-yim-at -^ sabaayimat 
because-3sg-sing-Asp 

'because you sing' 

e. o-yah-ii-e-meere — > ayaheemeere 
3sg-go-Asp-for-nothing 

'he went for nothing' 

f . o-wii-ko-a-yah-ii -^ o wii kaa yahii 
3sg-said-that-2sg-go-Asp 

'he said that you went/left' 

A short vowel obtains when the vowel of the second morpheme is part of a 
closed syllable, as shown by (7a-b). 

(7) a. o-wii-ko-on-njah-ii — > o wii kon njahii 

3sg-said-that-2pl-go-asp *koon 

'he said that you went/left' 

b. o-wii-ko-en-njah-ii -^ o wii ken njahii 
3sg-said-that-lpl-go-asp *keen 

'he said that we went/left' *koon 

We get a vowel-glide sequence when we have the following (mid) vowel 
combinations: /aJ + /z/ = ay, hi + Izl = oy; lal + lol = aw\ Id -t- lol = ew, as illus- 
trated by (8a-d). From these combinations the generalization to be drawn is that 
we get glide formation in Pulaar if a non-high non-low vowel is preceded by an 
non-identical non-high vowel. This generalization is formalized in (8e) within an 
SPE (Chomsky & Halle 1968) type of framework. What (8e) says is that a low 
vowel will become non-consonantal and non- vocalic (therefore, a glide) in the en- 
vironment where it is preceded by another low vowel which differs in its round- 
ing specification. In case the first morpheme is consonant-final the vowel of the 
second morpheme is syllabified with the final consonant of the first word for 



Diop: Vowel deletion in Pulaar 



37 



which it forms a nucleus, as illustrated in (8f) below (a dot indicates a syllable 
break). 



> 



(8) a. 



b. 



haala-e-kawr-al — > 

talk-conj-agreement-NCagr 

'talk and agreement' 

wuro-e-Fuuta — > 

city-prep-Fuuta 

'a city in Fuuta (Region)' 

ma-o-yah -^ 

future-3sg-go 

's/he will go' 

nde-o-yah-i — > 

when-3sg-go-asp 

'when s/he went/left' 



haalay kawral 
wuroy Fuuta 
maw yaa 
ndew yahi 



e. 



low 

a round 



a round 

-cons 

-voc 



low 

- a round 



ka.ne sehilmum 



f. Kan-e-sehil-mum -^ 
Kan-conj-friend-3sgposs. 
'Kan and his friend' 

Having shown the various patterns attested so far, I am going to discuss some 
of the syllable-related issues that are pertinent to the analysis that would be pro- 
posed for the data in (6)-(8). First, the syllable types. In Pulaar, on the surface, the 
following syllabic types are found: CV, CVV, CVC, and CVVC. The syllable 
template that I assume in this paper is as proposed in Diop (1993) and shown in 
(9) below where aspects of both moraic phonology and X-slot theory are used (cf. 
Diop 1993 for further discussion of this model). 

(9) Pulaar syllable template 




The first syllable-related issue I wish to address here has to do with weight. 
Following Hayes (1987; 1988; 1989) 1 represent vowels (and geminates) with an 
underlying mora whereas single consonants (in coda position) acquire a mora by 



38 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

virtue of the Weight by Position (WBP) Rule (Hayes 1987; 1988; 1989) which, 
on a language-specific basis assigns a mora to a consonant in coda position if such 
mora assignment does not violate the upper-bound limit on the number of moras 
per syllable imposed by the language in question. The motivation for saying that 
WBP applies in Pulaar is as follows. There is a rule in Pulaar that shortens long 
vowels (in certain positions) when they are followed by a heavy syllable (cf. Diop 
(1993); Paradis (1986, 1992); Prunet & Tellier (1984)). This rule applies when 
the syllable following the long vowel is either CVV(C) or CVC, and never before 
CV. This is an indication that with respect to vowel shortening CVC counts as 
heavy (bimoraic) in the same way CVV or CVVC does. However, while CVV 
occurs rather freely, CVVC is more restricted in that in final position its coda 
consonant is either Ivd or Id, as shown in (10a) (where the dots indicate syllable 
breaks). When closed by another consonant in final position a vowel is suffixed to 
that consonant, as shown by (10b). Furthermore, CVVC is never derived by any 
process, phonological or otherwise. 

(10) CVVC in word-final position 

a. mbin. daan 'maid/butler' 
Hal. waar 'name of a village' 
mon. toor 'watch' 

dii. waan 'place/area' 

mi. soor 'headscarf 

ti. su. baar 'prayer time' 

kaf. taan 'type of dress' 

b. tuu. baak-o 'European' 
European-Noun class agr. 

ca. paat-o 'Moor/ Arab' 

moor-Noun class agr. 

Syllable weight and the fact that CVVC is never derived is relevant to VD in 
that in a configuration like that shown in (7a-b) above where the vowel of the sec- 
ond morpheme is closed by a consonant (that consonant is morale by virtue of the 
WBP rule that I mentioned earlier) the rule of deletion is going to avoid deriving a 
long vowel because this will result in a CVVC where the last /C/ is morale; there- 
fore a trimoraic structure, violating thus the upper limit of two moras per syllable 
in Pulaar. This is not to say that CVVC is always trimoraic. In fact it is my claim 
that non-derived CVVC is not trimoraic and here is the evidence. If CVVC were to 
be treated as trimoraic then when its coda consonant is deleted one would expect 
compensatory lengthening to take place and derive a triply long vowel. This does 
not happen. For instance, the Arabic word /Wallaah/ has been borrowed into 
Pulaar; but when the /h/ got deleted (because of the restriction I discussed above 
against coda /h/) what we have is /wallaayi/ instead of */wallaaa/, which is expected 
if CVVC syllables were treated as trimoraic by Pulaar speakers (Cf. Diop (1993) 
for more discussion). 

The next issue is that of glottal insertion. In Pulaar, vowel-initial morphemes 
trigger a glottal insertion. Major lexical category items (nouns, adjectives, ad- 
verbs, and verbs) systematically do so. For example, in the data in (1 la) below the 



Diop: Vowel deletion in I*ulaar 39 

words are underlyingly vowel initial. However, each of them is pronounced with 
a glottal stop before the initial vowel, as illustrated by (lib). 

(11) Vowel-initial words in Pulaar 



ekk-o 


'apprenticeship' 


am-re 


'turtle' 


in-nde 


'name/naming ceremony 


oto 


'car' 


uur-de 


'to smell good' 


7ekko 




7amre 




7innde 




7 oto 




7uurde 





What is then the relevance of this to VD? The answer is a question. If vowel- 
initial words are realized with a glottal stop on the surface, why then would the 
vowels of morphemes like Id, /on/, /a/, /en/, etc. in (6)-(8) above be syllabified with 
the final syllable of the preceding morpheme? Therefore, one would have to say 
that glottal insertion affects only major lexical category items such as nouns, 
verbs, and adjectives. That is not quite adequate because the morphemes whose 
(initial) vowel gets resyllabified with the previous morpheme, when they appear 
in sentence-initial position are also pronounced with a glottal stop. So, /a/ (you- 
singular) in the sentence /a-jal-ii/ 'you-laugh-ed' is pronounced /7a/. Glottal inser- 
tion will then be permitted even with lexical categories other than nouns, verbs, 
and adjectives if those are in sentence-initial position. 

The next and last issue has to do with a proposal (Diop (1993), cf. same ref- 
erence for more discussion) that morale phonology have a rimal tier. Thus, (12a-c) 
show, respectively, the representation of a short vowel (/o/), a long vowel (/oo/), 
and a vowel-glide sequence (/oy/) within the different syllable-internal nodes as- 
sumed in this paper (only the relevant part is shown for each representation). (12c) 
illustrates the fact that 1 treat the vowel-glide sequence as being the direct 
'product' of syllabification whereby the first vowel is syllabified in the nucleus 
whereas the second vowel is syllabified outside it. (12d) will be ruled out as a pos- 
sible representation for /oo/ because it is a violation of OCP (McCarthy 1986; 
Hayes 1986; Odden 1986). 

(12) Syllabification of vowels and vowel-glide sequences 

a. N b. N c. R d. *N 



N (C) 
I I 

I V II II 

o = [o] () = [oo] o e = |oyl o o 



In the foregoing 1 have discussed some background information pertinent to 
the analysis of VD in Pulaar. Here is. again, a summary of the patterns observed so 



40 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

far in (6)-(8) (overlooking the rather unproblematic case of (8e)). We had a pattern 
deriving a long vowel from two short identical vowels. A long vowel was also de- 
rived when we have a vowel preceded by a high (or high and long) vowel. 
Another case where the rule yielded a long vowel is when /a/ was preceded by lol. 

The second pattern was when the vowel of the second morpheme was closed 
by a consonant; syllabifying that vowel with the preceding morpheme produced a 
short vowel (and a consonant) rather than a long one. 

The third pattern was that in which a vowel-glide sequence was created. In 
that pattern, for the A^W/ sequence we had the combinations in ( 1 3a) whereas for 
the rWYI sequence the combinations are illustrated by (13b) below. 

(13) Vowel-glide sequences in Pulaar: a reminder 
a. V-W sequences 

/a/ -I- lol =IssnI 

Id 4- Id =lewl 
h. V-Y sequences 

/a/ -i- Id =/ay/ 

lol + Id =/oy/ 

The challenge presented by the data for any analysis of vowel deletion would 
to predict each of the different outcomes just outlined. Furthermore, in (6e) above 
we have a case where a short vowel (monomoraic) spreads to a position previously 
occupied by a long one (bimoraic). Yet we do not get a trimoraic syllable. What 
we derive is a long (bimoraic) vowel. One has to account for what happened to the 
third mora. This case is somewhat similar to that in (7a-b) where a trimoraic struc- 
ture is also reduced to a bimoraic one. But, as we shall see, they are treated differ- 
ently. Another challenge for any analysis of VD would also be to predict the oc- 
currence of vowel-glide sequences. In the section to follow 1 am going to give the 
solution that I propose for VD. 

3. A phonological analysis of vowel deletion in Pulaar 

In (5) above I gave a representation of Pulaar vowels and a redundancy rule 
that captures the fact that the feature [back] is predictable in Pulaar. Given the 
Pulaar syllable facts I outlined at the beginning of the paper 1 also take it to be the 
case that syllable structure is assigned lexically and post-lexically. Thus, follow- 
ing the operation of certain phonological or morphological rules the output of 
such rules can feed syllabification. Syllabification can take place across morpheme 
boundaries. So, in (8f) for instance, the conjunction lei 'and' is pronounced with 
the final consonant of the previous morpheme (i.e. /n/) as its onset; yet these two 
segments came from two different morphemes. My analysis, however, focuses 
more on the cases where two vowels (instead of a consonant plus vowel) merge. In 
such cases VD takes the form of a rime merger that is illustrated in (14) below. 
Referring to the syllable to account for phonological processes involving vowel 
mergers is not a novel idea. It has been proposed in Schane (1987). Although 
Schane's framework and the one I propose here have some similarities they differ 
in ways that I do not intend to discuss in this paper since it is not my intention to 



Diop: Vowel deletion in Pulaar 4 1 

compare the two frameworks. Nevertheless, I will spend some time summarizing 
Schane's theoretical framework. Schane's framework (which he uses to analyze 
hiatus in Sanskrit and Chicano Spanish) is one in which the autosegmental repre- 
sentations contain three tiers: a syllable tier, a CV tier, and a segmental tier. The 
first tier depicts the number of syllables while the second gives information about 
the quantitative characteristics of phonological units. The quality of the phono- 
logical units is determined by the third tier. In addition to these tiers Schane uses 
the notion of closure (merger) at each of these tiers. In the framework that I propose 
syllable count is not regarded as crucial. The quantitative characteristics of 
phonological units is determined by the mora whereas the rime and nucleus node 
determine the quality of phonological units. So, any phonological unit couched 
within the nucleus is going to be realized as a vowel whereas anything outside the 
nucleus is going to have consonantal status. As I said earlier the rule is formulated 
in terms of a rime and nucleus merger. The first rule (the rime merger rule), re- 
sponsible for glide formation, because it has a more specific environment, is going 
to apply first following Kiparsky's (1973b; 1982a) Elsewhere Condition. It is il- 
lustrated in ( 14a) where it takes the initial vowel of the second morpheme and syl- 
labifies it at the rime node level of the preceding syllable, giving a vowel-glide se- 
quence for reasons that I explained earlier. (14c) is a derivation illustrating (14a) 
whereas in (15) we have the general rule that syllabifies vowels under the nuclear 
node. It accounts for the more general cases, (6)-(7). The delinking of the associa- 
tion line in the second syllable causes that syllable to collapse, making its rimal 
content available for the merger. As a reminder, the combinations that yield glide 
formation in Pulaar are shown in (14b). The generalization was that we get glide 
formation if a non-high non-low vowel is preceded by a non-identical non-high 
vowel. When such is the case the rime merger rule in (14a) syllabifies the second 
vowel within the rime of the first syllable, turning it into a glide. There is no con- 
trast between high and mid glides (at least in Pulaar). Whether we get the labial 
glide /w/ or the palatal glide /y/ depends on the rounding specification of the sec- 
ond vowel. Where that specification is [+] we get /w/; and we get /y/ where it is [-]. 
In (14c) we give a derivation for /Sammbay Zeynabu/ < /Samba e Zeynabu/ 
'Sammba and Zeynabu'. (14d) shows the surface representation of (14c). (Only the 
relevant parts are syllabified for /Sammba/.) 

(14) Pulaar rime merger: a two-step process 

a. The specific rule: glide formation: rime merger 




C V 



42 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

b. Pulaar glide formation 

a + E = /ay/ o + e = /oy/ 

a + = /aw/ e + 3 = /ew/ 

c. Derivation for [Sammbay Zeynabu]: 

a a 




Samm b a e Zeynabu 



Glottal insertion 

rime merger: specific applies 
rime merger: general 



d. Surface representation of (13b) above: 




Samm b a e Zeynabu = [Sammbay Zeynabu] 



Having illustrated how the glide formation takes place I am now going to ac- 
count for the general case: short/long vowels. The merger rule in (15a) below 
(ordered after (14a)) syllabifies the second vowel under the nuclear node. 
Following that we have a deletion rule (affecting the first vowel) and a root node 
spreading rule from the second vowel to the position formerly occupied by the 
first vowel. This second rule affects not only cases where the vowels involved are 
not identical but also those cases where they are in fact identical (in the latter case 
the rule may be considered as having the same effect as OCP). As said earlier, a set 
of vowel features on two (adjacent) moras under the same nuclear node gives a 
long-vowelled nucleus. The deletion and spreading rules are shown in (15b) be- 
low whereas (15c-d) show the derivation for [lewree koode] < /lewru e koode/. 



Diop: Vowel deletion in Pulaar 



43 



(15) The Elsewhere aile: 

a. nucleus merger b. 



o 
J 


a 

1 


1 


1 

R 

1 


1 

N 


1 

N 



Feature deletion and root node 
spreading (FD and RNS) 

N N 

f\ A 

[i \i u \l 

[F] [FJ ^ [F] 



C V V 

(where [F] is the root node) 
c. Derivation for [lewree koode] < /lewru e koode/ 
a o 




e koode 



Specific rule 

General rule applies 

FD and RNS applies 

Syllabification applies 

d. Surface representation of (15c) 




1 w r 



e koode = [lewree koode] 



So far I have accounted for two cases: cases where VD yields a long vowel 
and those which yield a vowel and a glide. I now go on to the third case, i.e. 
where a short vowel results from VD. The relevant data are in items (7a-b) above 



44 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



and they are reproduced here as (16a) for convenience (the target vowels are bold). 
In (16a) we have a monosyllabic (and closed syllable) pronoun /on/. The initial 
vowel of the second morpheme and the final vowel of the first are identical. Still, a 
long vowel is not produced. There are at least two ways in which the data in ( 1 6a) 
can be explained. The first tack we can take is as follows. Assuming that there is 
syllable structure prior to and following VD, also taking into account the fact that 
the WBP rule appHes in Pulaar, the relevant part of (16a) can be represented as in 
(16b). 



(16) a. 



b. 



o-wii-ko-on-njah-ii -^ 
3sg-say-that-2pl-go-Asp 
'he said that you went' 



7o wii kon njahii 
*koon 



R 

N C 



C V 



To the second syllable of ( 1 6b) we cannot apply glottal insertion because the 
morpheme is neither sentence-initial nor a major lexical category item. So, it is 
syllabified with the preceding syllable, causing a trimoraic structure. This struc- 
ture is reduced to two moras (by virtue of a mora deletion rule that applies when- 
ever a trimoraic structure is created) because Pulaar avoids trimoraic syllables (as 
said earlier), and never derives them. 

This solution is not all that desirable (though probably unavoidable in some 
cases; as will be shown later) as it relies on something that does not look quite nat- 
ural in phonological theory, creating a structure and then erasing it in order to ar- 
rive at the correct derivation. 

We can do away with such an intermediary stage that creates three moras. 
The second approach is going to do just that; and for that reason it is the approach 
we adopt here. In this approach the syllable of the second morpheme /on/ (prior to 
the application of the rime merger process) collapses following the delinking of 
the association line from the mora to the nuclear node because a Pulaar syllable can 
only exist if it has a nuclear mora (i.e. a mora dominating a vowel). The mora of 
the nasal, naturally, is 'deleted' because at that stage there is no syllable structure 
and the nasal is no longer in rime position to receive a mora by virtue of the WBP 
and there are no syllabic nasals in Pulaar. The steps just outlined are illustrated in 
(17a-b) below. At this point in the derivation we proceed with syllabification. 
Following the rules that we postulated earlier ((14) above) the vowel of the second 



Diop: Vowel deletion in Pulaar 



45 



morpheme is syllabified in the nucleus node of the final syllable of the first mor- 
pheme. So, it is the general, not the specific one, that applies. Feature deletion and 
root node spreading apply, as illustrated by (17c) below. 

(17) Syllable 'collapse' 



a. 



b. 




I I 

k o o n 




Earlier, I claimed, without elaborating on it much, that in Pulaar a conso- 
nant in coda position must be morale. A piece of evidence for that was that long 
vowels shorten before CVC syllable in the same way they do before CVV(C) ones. 
However, WBP can only apply if its application does not violate the upper bound 
limit of two moras per syllable. Therefore, in (16c) WBP cannot apply. Since 
WBP cannot apply the second mora is then donated to the rimal consonant leaving 
the first vowel with one mora (therefore realized as a short vowel). This way we 
derive a bimoraic single short vowel syllable. In (18a) (next page) we illustrate 
mora donation whereas (18b-c) (next page) show the last and final steps of the 
derivation, including mora donation. 

1 have just given an account of how VD works when the second vowel-initial 
morpheme involved in the process is monosyllabic. Vowel-initial bisyllabic pro- 
nouns show similar behavior to that of monosyllabic pronouns in that when their 
initial is syllabified with the previous morpheme a short, not a long, vowel is de- 
rived. These pronouns, illustrated in (19) next page, are problematic for the 
analysis given for the monosyllabic pronouns in that while in the data in (16-18) 
we had single syllable morphemes (pronouns), in (19) we have bisyllabic 
morphemes as the second morpheme (e.g. /ocf on/). Following the syllabification 



46 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



(18) a. Mora donation 
R 




R 

N C 



■> I 



H \^ 





n = [kon] 



principles (both universal and parametric), the medial consonant in /od'on/ is 
syllabified as the onset of the second syllable instead of the coda of the first. 
Consequently, it is mora-less. Therefore, the first syllable is monomoraic. 
Merging it with the preceding monomoraic syllable should not lead to any mora 
deletion or donation of some sort since bimoraic syllables are accepted in Pulaar. 
The issue, then, is whether the approach used to explain the data in (16-18) can be 
improved to accommodate the data in (19) or whether a whole new approach is 
needed for the new set of data. Clearly the latter is not desirable. We do not want to 
have a multitude of different rules and approaches to account for what might be 
one and the same phenomenon. We adopt the first suggestion then; i.e. improve the 
first approach in order for it to explain the data in (19). (Once again, whether the 
two vowels involved in the process are identical or not is irrelevant because we are 
always going to have a short vowel.) 



(19) Disyllabic pronouns and VD 

a. o-wii-ko-od^on-njar-a 
3sg-say-that-2pl-drink-Asp 
'he said that you drink' 

b . d'o-omo-yah-a-fof 
'pl.Adv-3sg-go-Asp-every' 
wherever s/he goes 

In order to put things into perspective a brief survey of Pulaar pronouns is 
necessary. (20a-c), (21), (22), and (23) illustrate these. 



o wii kod'on njara 
*koocfon 

= d^omo yaha fof 
*d"oomo 



Diop: Vowel deletion in Pulaar 



47 



(20) Pulaar subject pronouns: the one-syllable set 


a. Preposed 




Singular 


Plural 


1 mi 


en / min 


2 a 


on 


3 


6e 


Example 




mi-yah-ii 


'I went' 


Isg-go-Asp 




b. Postposed 




mi 


d^en / en 


cfaa/aa 


d"on / on 


Example 




njah-mi 


'I went' (focus ( 


go-lsg 




c. The two syllable set: never 


postposed 


mid'o 


amin / emin 


acfa 


od^on 


omo 


e6e 


Example 




mid'o-yah-a 


'I go' (habitual) 


Isg-go-Asp 




(21). Object pronouns 




kam/mi 


men /en; min 


ma/ maa 


mon / on 


mo /moo 


6e 


(22). Independentyfocus pronouns 




miin 


minen / enen 


aan/maa 


onon/mon 


kanko/makko 


kam6e/ma66e 


(23). Possessive pronouns 




am 


amen / men 


maa 


mon 


makko / iiko 


ma66e 


/ mum / um 





In Pulaar /on/, /en/, /o/ and /a/ are subject pronouns. They correspond, respec- 
tively, to the second part of the two-syllable pronouns in {20c). The first part in 
each of these pronouns (i.e. lod I, Iccf/, /om/, /ad"/) is neither a pronoun nor an at- 
tested (synchronic) prefix of some sort. All things being equal we take it to be the 
case that in each of the pronouns in (20c) a highly morphologized phonological 
rule suffixes the pronoun to a closed syllable 'morpheme'. That being said, we also 
take it to be the case that the rime merger rule will take place before this process of 
pronoun suffixation. This way the process in (20c) is similar to that in ( 16)-(18) 



48 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



in that we have a closed-syllable vowel-initial morpheme (/ocf /, /ecf/, /om/, and 
/acf/) merging with a vowel-final syllable and the same solution (not repeated here) 
can account for both cases. 

The solutions proposed so far seem to predict and explain the data presented 
thus far on VD in Pulaar. However, there is a set of data that seem to be problem- 
atic for our analysis. In (24a-g) below we have sentences in which VD could take 
place, given that the context of (14) above is met, but indeed does not. (A slash be- 
tween two vowels indicates that the rule fails to apply and that the vowel is pro- 
nounced with a glottal stop.) The data are organized in pairs that show a contrast 
between two cases. On the one hand we have cases where the first morpheme ends 
in a long vowel that does not delete. The next example will show the same vov/el at 
the end of a morpheme where that vowel is short. In that case the second vowel is 
syllabified with the first to form a vowel-glide sequence. This pattern is found, re- 
spectively, from (24a-f) where (a) contrasts with (b), (c) with (d), so on. (24f) il- 
lustrates the fact that long vowels at the end of a morpheme do indeed allow the 
rule to operate. 

(21) Vowel deletion and prosodic information 

a. Muusaa-/-e-debbo-mum 

Moses-coord. -woman-his 

Muusaa and his wife 

=Muusaa 7e debbomum *Muusaay 

b. Rama-e-Abu 

coord. 
Rama and Abu 
= Ramay Abu 

c. njol-d"aa-/-e-oto-makko 
enter-2sg-prep-car-his/her 
you entered (in) his/her car 

= njold"aa7eotomakko *njold"aay 

d. njol-mi-e-oto-makko 
enter- Isg prep car-his 
I entered (in) his car 
= njolmee otomakko 

e. Busoo-/-e-sehil-mum 

coord. -friend-his 
Busoo and his friend 
= Busoo 7 e sehilmum *Busooy 

f. Dono-e-min-um 
coord. -young-sibling-his 
Done and his younger sibling 
= Donoy minum 

g. mi-ar-il-e-meere 

1 sg-come- Asp-prep. for-nothing 
I came for nothing 
= mi aree meere 



Diop: Vowel deletion in F*ulaar 4 9 

Our hypothesis is that prosodic information is responsible for failure of the 
rule to apply in these cases. The explanation for the 'apparent' problem in (24) 
above is as follows. First, notice in Pulaar (as just said) that the initial vowel of a 
second morpheme can be syllabified with the final syllable of the previous mor- 
pheme even when the latter ends in a long vowel (cf. 24g). Second, notice also that 
in all the cases where VD fails to apply (namely, in examples (24a, c, and e) the 
long vowel of the first morpheme is either /a/ or lol. As shown in (8) and (13) 
above, /a/ or lol + Id give, respectively, the sequences /ay/ and /oy/, a closed sylla- 
ble. A sequence of one vowel and a coda consonant is treated as bimoraic in our 
approach to the Pulaar syllable. In the data in (24), where the rule fails to apply we 
have long vowels, instead of short vowels as host, except for (24g). So, in (24a) 
for instance we have /aa/ + /e/; which, in theory, should yield the sequence /aay/ 
(*Muusaay Abu). However, if this was to be the case we create a (superheavy) tri- 
moraic syllable (viz. /saay/). As said earlier this type of syllable in Pulaar is not de- 
sirable (especially 'word'-finally); nor is it derived by any phonological process. 
The prohibition against trimoraic syllables acts then as a filter on all phonological 
operations that involve syllabification to the extent that it prevents one from deriv- 
ing super heavy syllables. So, in (25) we have a case where Id could be brought 
under the rime node subject to being 'rejected' from under it by the moraic filter as 
its syllabification under that rime will create a trimoraic structure. Once it is 
'rejected' from the previous syllable it has to form a syllable on its own, triggering 
thus the glottal insertion rule (mentioned earlier without much formalism) that 
provides a default onset to vowel-initial morphemes. Consequently, we have an- 
other environment where such glottal insertion takes place. First we said it took 
place sentence-initially. We have to add to that another environment; namely, after 
the application of the moraic filter (in phrase-initial position). To end this section I 
am going to present our analysis of the data in (24g). The problem presented by 
(24g) is as follows. The first morpheme has a long (therefore bimoraic) vowel. The 
second morpheme is a single vowel (monomoraic) syllable. Deleting the final 
(bimoraic) vowel of the first morpheme and spreading the second (monomoraic) 
one in this instance will inevitably give rise to a trimoraic structure, especially 
since the mora donation rule in (18a) cannot be applied here as there is no coda 
consonant to yield the third mora to (unlike the case of (17a) above). It looks, then, 
that in deriving (24g) we have to go through a stage at which a trimoraic structure 
is created. However, the difference between this case and the one in (1 7a) is that in 
the latter there was no reason to believe that the nasal carried a mora throughout 
the derivation because when the syllable structure in which it was found col- 
lapsed the mora was then lost because consonants only receive a mora by virtue of 
the WBP, in the theoretical framework that we adopt here. In (24g) this argu- 
ment cannot be made. Instead, we propose the following solution. First, some (by 
now familiar) assumptions. In Pulaar, there are no sequences of three vowels (or 
consonants), identical or not. So, this gap can be expressed in terms of a Pulaar- 
spccific prohibition against ternary branching for vowel (consonant) features. In 
addition, the restriction against trimoraic structures in Pulaar (*a+2)i, i.e. no syl- 
lable can have more than two moras) acts as a filter on all phonological operations 
involving syllabification to the effect that it prevents derivation of syllables that 



50 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

violate the upper bound limit of two moras per syllable. When this limit is vio- 
lated it triggers a deletion of the third mora. In (24g) the issue will be to decide 
which mora to delete and how to make it candidate for deletion. (24g) has been re- 
peated here as (25) as a reminder. 

(25) mi-ar-ii-e-meere =miareemeere 
1 sg-come-Asp-Prep.for-nothing 

I came for nothing 

*areee 

To the vowel Id the glottal insertion rule fails to apply, so the rime merger 
rule (the elsewhere rule) applies syllabifying it under the nuclear node of the syl- 
lable containing the long vowel. Following this we apply the feature deletion rule 
deleting the features of the long vowel I'lil. We have, then, two moras without fea- 
tures. The root node spreading rule applies, spreading the features of the vowel of 
the second morpheme to one of the moras of alpha given that vowel features can 
only be binary branching, at the most. This leaves us with one mora that does not 
dominate any features. Ito (1986) introduced the notion of Prosodic Licensing that 
says that 'phonological material must be incorporated into the next higher level of 
prosodic structure'; otherwise, 'it is deleted by Stray Erasure' (Steriade 1982, 
Harris 1983). As mentioned by Hayes, 'a natural extension of this principle would 
state that higher-level phonological elements, such as moras, are also subject to 
Stray Erasure if they fail to dominate any lower-level element' (Hayes 1989:264). 
One of the moras of the host is exactly in that situation. Its features had been 
deleted by the feature deletion rule but it is not linked to any feature at the end of 
the derivation where syllabification takes place again. Therefore, we have a mora 
that does not dominate any features (vocalic or consonantal); so it is candidate for 
Stray Erasure. The mora deletion rule is illustrated in (26a) whereas the steps dis- 
cussed so far are illustrated in (26b) below in a derivation for [mi aree meere] < /mi 
arii e meere/. Since the analysis here is quite complex (26b) will be done in five 
steps showing the different stages of the derivation and the surface representation. 
What (26a) says in essence is that a mora that does not dominate any segmental ma- 
terial is erased (since it would be impossible for it to be realized phonetically). 

(26) Trimoraic mergers 

a. Mora deletion 






Diop: Vowel deletion in Pulaar 5 1 

b. Derivation for [mi aree meere] < /mi arii e meere/ 

(only the relevant parts /arii e/ are shown) 
Step 1: syllable collapse Step 2: nucleus merger 



R 

I 

N 

t 

I 
[F] 

I 

e a 

Step 3: feature deletion Step 4: Root node spreading 

and mora deletion 







Step 5: Surface representation 




= [a . ree] 

In the foregoing I have shown that the data in (24), which seemed to be prob- 
ematic for my analysis turned out to be easily handled by it. I have, thus, ac- 
counted for what would have otherwise been considered exceptions to VD. 

This is not all, however. We do have further cases where failure of the rule to 
ipply cannot be attributed to prosodic information. I shall demonstrate that syntax 
s the reason for such failure. VD in Pulaar is sensitive to syntactic information in 
establishing its domain of application. In this last section we are going to try and 



52 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

demonstrate that fact. First I present the data, followed by a presentation of basic 
facts about Pulaar syntax; then I present the syntactic structure in which the first 
and second morphemes are found; for instance /V-P/ meaning the first morpheme is 
a verb whereas the second is a preposition. I end this section by testing the data 
against two major approaches to syntax-phonology interface, the direct approach 
and the indirect approach, and showing that the generalization about the data es- 
capes the predictions of either approach. ( 

4. Vowel deletion and syntax 

First, the data (a slash between two vowels indicates that the rule does not 
apply and the glottal stop with which the second vowel is pronounced in this case 
is shown when the example is repeated after the glosses). 

(27) Vowel deletion and syntax 

a. teew-ngu,-/-a-yid"-aa-ngu 
meat-NC-2sg-like-Neg-Respr. 
the meat, you don't like it 

= teew ngu, 7 a yid"aa ngu *ngaa 

b. teew-ngu-a-yid'-aa-ngu 
meat-Rel.-2sg-like-Neg-NC 

the meat (that) you don't like 
= teew ngaa yidaa ngu 

c. rawaa-ndu-ndu-/-e-joom-um 
dog-NCagr-NCDet. -coord. -owner-3sgPoss 
dog the and owner its 

the dog and its owner 

= rawaandu ndu 7ejoomum *ndee 

d. ndu-rawaa-ndu-e-joom-um 
NCDet-dog-NCagr-coord.-owner-3sgPoss 
this dog and its owner 

=ndu rawaadee joomum 

e. ndu-/-e-joom-um 
NCDet-sing-coord-owner-3sgposs 
this (one) and its owner 

=ndu 7 e joomum *ndee 

f. gor-k-o-mo-calmin-mi-/-o 
man-7 -NCagr-Rel-greet- 1 sg-NC 
man that greet I the 

the man I greeted v. 

= gorko mo calmin mi 7 o * moo 

g. gor-k-o-mo-calmin-mi-e-mon-o 

man-7 -NCagr-Rel-greet- lsg-Prep.-2pl. NC 
man that greet I among you the 
the man I greeted among you 
= gorko mo calminmee mon o 



Diop: Vowel deletion in F*ulaar 5 3 

h. o-rokk-i-/-on-jawdi 

3sg-give-Asp-2pl-wealth 

He gave you (pi.) wealth 

= o rokki 7 on jawdi *rokkon 

i. ko-/-enen 

foc.-lpl. 

It's us 

= ko7enen *kenen 

j. oto-am 

car-lsgPoss 

my car 

= otam 
k. mo-/-am 

?of- 1 sgPoss 

that (of) my 

mine 

= mo 7 am *mam 

The data just introduced in (27) present some interesting and challenging 
problems. In the next paragraphs to follow I am going to outHne what those prob- 
lems are. First, (27a) and (27b) show that the morpheme /ngu/, a noun class in the 
first example and a relative clause marker (i.e. head of Comp) in the second, dis- 
plays two different patterns in the sense of its vowel deleting in (27a) but not in 
(27b). (27c) and (27e) are other examples that illustrate a case of a noun class (or a 
determiner) whose final vowel does not delete whereas (27d), the 'mirror' image of 
(27c), shows that the final vowel of the head noun /rawaandu/ deletes. In (270 we 
see that the vowel of the subject pronoun /mi/ does not delete when followed by the 
(open syllable) noun class (determiner of the head noun in the relative clause) 
whereas the same subject pronoun used in front of a following preposition Id (in 
(27g)) shows final vowel deletion. In (27h) the final vowel of the verb does not 
delete when followed by a vowel-initial object pronoun /on/ whereas earlier in 
(26b) we saw that final vowels of verbs can delete. In (27j) and (27k) we have an- 
other interesting alternation. In (27j) the vowel of the first morpheme deletes, al- 
lowing /am/ 'my' to be syllabified with it (cf./otam/) whereas in (27k) the rule is 
blocked between /mo/ and /am/. 

The data presented in (27) show the first and the second morphemes (whose 
vowels are involved in the process of VD), occurring in the following syntactic 
configurations (28). 

(28) Syntactic structures 



Synt. Structi 


jre Example number 


rule applies? 


NP-Conj 


6a-b:8a;8f;21b:24f:27d 


yes 


NP-Conj 


24; 24e 


no 


C-NP 


6c-d; 6f; 7a-b; 8c-d; 19a-b;27b 


yes 


V-P 


6e; 24d; 24g; 27g 


yes 


V-P 


24c 


no 


NP-P 


8b 


yes 



Det-NP 


27a 


Det-conj 


27c; 27e 


V-Det 


27f 


V-NP 


27h 


Focus-NP 


27i 


NP-poss 


27j 


P-poss. 


27k 



54 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

no 

no 

no 

no 

no 

yes 

no 

Having presented the data, the nature of the alternations involved, and the 
syntactic configurations in which the two morphemes involved in vowel deletion 
occur, I am going to first outline some basic facts about Pulaar syntax before get- 
ting into the details of the analyses. In the basic word order of the Pulaar sentence 
subjects precede their predicate, objects follow verbs, as illustrated by (29) below. 

(29) Pulaar basic word order 

mi-yar-ii-kos-am-hanki 
Isg-drink-Asp-milk-NCagr-yesterday 
'I drank milk yesterday' 

The Pulaar noun typically has a root and noun class agreement attached to that root 
that shows what class the noun belongs to (cf. Paradis (1986, 1992); Sylla (1982) 
for detailed studies of Pulaar noun classes). For instance, in (27) below the word 
for dog /rawaa-/ belongs to the /ndu/ class in the singular and to the /d'i/ class in the 
plural. Consequently, it bears /ndu/ or /cf i/ on its root. We can also add to it the 
diminutives /gel/, /kon/, or the augmentative /gal/. 

(30) The Pulaar noun 



rawaa-ndu 


rawaandu 


dog 


rawaa-cfi 


dawaad^i 


dogs 


rawaa-gel 


dawangel 


small dog 


rawaa-kon 


ndawakon 


small dogs 


rawaa-gal 


dawangal 


big dog 



Within the noun phrase the head noun (root + noun class agreement marker) is ei- 
ther in initial position followed by the noun class, as illustrated by (31a-b) below, 
or in phrase-final position preceded by the determiner (e.g. 'this'). (31c-d) is an il- 
lustration. The noun phrase can be null-headed, as illustrated in (31e) where /ndu/ 
is understood to refer to /rawaandu/ 'dog'. These null-headed NPs are represented 
in (31f). 

(3 1 ) The Pulaar noun phrase 
a. Full NPs 

rawaa-ndu ndu = rawaandu ndu 

dog-NCagr NC 
dog the 

the dog 



Diop: Vowel deletion in Pulaar 5 5 

b. Representation 




rawaandu 

c. ndu-rawaa-ndu = ndu rawaandu 
Det-dog-NCagr 

dog this 
this dog 

d. Representation 

NP 



Det N 

I I 

ndu rawaandu 

this dog 

e. Null-headed NPs 

ndu-dog-ii = ndu dogii 

NC(Det)-run-Asp 

it ran (away) 

f. Representation of null-headed NPs 

NP 



N' Det 

I I 

ndu 

Relative clauses are formed by using the noun class of the head noun (as a rela- 
tivizer) or by using /mo/ for nouns which belong to the /o/-class (of humans and 
borrowed words). However, this /mo/ can only appear when we have a negative 
relative clause, as in (32a-b). Positive relative clauses where the head noun is sub- 
ject (instead of agent) do not allow use of the relativizing pronoun, as shown in 
(32c). Instead, the verb shows agreement with the head noun in the sense of bear- 
ing an agreement marker that corresponds to the head noun. 

(32) Relative clauses in Pulaar 

a. gor-k-o-mo-yah-aan-i-o 
man-7-NCagr-Rel-go-Ncg-Asp-NC(Det) 
man who go not the 

the man who did not go 

b. gor-k-o-mo-rokk-u-mi-ndiy-am-o 

man-7 -NCagr-Rel-gi ve-Epen.- 1 sg-water-NCagr-NC(Det) 
man who give I water the 
the man 1 gave water to. 



56 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



c. gor-k-o-naam-cfo-o 

man-7 -NCagr-eat-Agr-NC(Det) 

man eat the 

the man who ate 

* gorko mo naamcf oo 

Verb phrases are somewhat complex. The verb precedes the direct object, indirect 
object, and the prepositional phrase. Verbal complexes may be formed of a root 
plus a number of extensions that are subject to both ordering and co-occurrence 
restrictions. There are two object pronouns /moo/ 'him/her' and /maa/ 'you-sing' 
that are internal to the verbal complex in the sense that they occur before the post- 
posed subject clitic which is also considered part of the verbal complex (cf. 
Paradis (1986, 1992); Prunet & Tellier (1984); Diop (1993) for further discus- 
sion). Other object pronouns are 'outside' the verbal complex in the sense of oc- 
curring after the postposed subject clitic. (33) below is an illustration of the dif- 
ferent verbal complexes just mentioned. 



(33) Pulaar verbal complexes 
a. 



= 7addii 



add-ii 
bring-Asp 
brought 
add-ii-jawdi 
bring-Asp-wealth 
brought wealth 
add-ii-e-jam 
bring- Asp-Prep. -peace 
brought in peace 
add-an-ii 
bring-Ben. -Asp 
brought for 
add-an-oy-ii 
bring-Ben. -Mvt-Asp 
went and brought for 
add-an-oy-moo-mi 
bring-Ben. -Mvt-3sgobj- 1 sg 
I went and brought for him/her 
g. add-an-oy-mi-on 

bring-Ben.-Mvt-lsg-2pl.obj 
I went and brought for you (pi.) 

In focus constructions the focus marker /ko/ appears before the focused NP (cf. 
Sylla (1982) for further discussion). If the NP is a pronoun then it has to be from 
the set given in (20c) or (22) above. Pronouns from those two sets always precede 
the verb. They are never postposed. (34) below illustrates focus constructions. 



d. 



e. 



f. 



= 7addii jawdi 



= addeejam 



= addanii 



= 7 addanoyii 



= ngaddanoymoomi 



= ngaddanoymi 7 on 



(34) Focused NPs in Pulaar 
a. ko aan 
It's you 



= ko 7 aan 



Diop: Vowel deletion in Pulaar 5 7 

b. ko-aan-e-makko = ko aane makko 
foc.-2sing-Conj.-3sg 

It's you and s/he 

c. ko-aan-e-makko-yah-i-e-oto 
foc.-lsg-Conj.-3sg-go-Asp-Prep.-car 
It's you and s/he went in car 

It's you and s/he who went by car 

= ko 7 aane makko njahee 7oto 

Having presented the basic picture of Pulaar word order that is relevant to 
vowel deletion I am now going to discuss the two approaches to syntax-phonology 
interface. As I discuss each approach 1 will test it against the data given so far and 
point to the problematic cases. First, the direct-syntax approach (DSA, hence- 
forth). 

4.1. The direct-syntax approach. (Clements 1978; Kaisse 1985; 1987, 
1990; Odden 1987). 

In this approach an external sandhi rule applies between a sequence of two 
words a and (3 when it is the case that either the two belong to the same X"^ or if 
some c-command relation holds between a and (3, depending on what version of 
c-command one adopts. Two versions of c-command prevail. They are presented 
in (35) below (Sells 1985:39). 

(35) C-command 

a c-commands p iff: 

a. every branching node dominating a dominates |3 

b. every XP dominating a dominates (3 

Under definition (35a) V in (36a) below c-commands NP but not PP 
whereas under (35b) V c-commands both NP and PP. By the same token NP c- 
commands P of PP in (36b) only by virtue of (35b) (as illustrated by the arrows). 

(36) C-command illustrated 

a. 

VP 




58 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



VP 



V 



PP 



NP 



NP 



Given these two notions of c-command I am going to discuss what the predictions 
the direct access approach are going to be with respect to the Pulaar data. First, 
the simple cases. Assuming the c-command notions in (35) (within the DSA) we 
can formulate the rule of vowel deletion as in (37). 

(37) Vowel deletion in Pulaar: preliminary version 

Delete the final vowel of a word when it is followed by a 
vowel-initial word that it c-commands. 

The rule in (37), using either version of c-command, is going to easily predict the 
data illustrating noun coordination where the first NP is not branching, therefore 
not necessarily dominated by XP. The same holds for cases where we have a CP 
whose head (C) is in a c-command relation with the specifier of the following IP. 
These two cases are schematically illustrated in (38a-b), respectively, whereas (39) 
gives a representative sample of the data presented earlier. (Throughout the rest of 
this section of the paper an arrow between two constituents means that the rule op- 
erates between them whereas a barred arrow means that the rule is blocked.) 

(38) Predictions of the DSA 

a. Coordination (Jackendoff 1977:190) 

NP 




a man and a woman 



Data that fit this pattern are presented in (6a-b); 
(8a, f); (24b, 0; and (27d); 



Diop: Vowel deletion in F*ulaar 



59 



b. CPs 




a 

you 
that you did not see 



yiy-aan-i 
see-neg-Asp 



Data that tit this pattern are illustrated in (6c-d, f); 
(7a-b); (8c-d); (19a-b) and (27b). 

The rule in (37) also predicts (using either version of c-command) the data in 
which a verb(al complex) is followed by a preposition. The data are in (6e), (24d, 
g), and (27g) above and the relevant trees are drawn in (39a-b) below where the ar- 
row shows the rule operating. In (39b) the rule operates between the verbal com- 
plex (verb root + postposed subject pronoun) and the following pronoun. 

(39) Further correct predictions of c-command 
a. 




ar-ii e meere 

come-Ap prep. nothing 

come for nothing 




c a 1 ni i II - in i e ni o n 

greet-I prep. you 

I greeted among you 

In addition to these data the rule in (37) correctly predicts that deletion (and 
spreading) do not occur in (27a, c, f)- In all these data c-command does not hold 



60 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



between the two morphemes involved, no matter which of the two versions in (35) 
above one is adopting. So deletion is blocked because the morphemes are not 
within its domain. In (40a-c) below I draw trees for (27a, c, f), respectively, to il- 
lustrate failure of c-command to hold. 

(40) More predictions of the DSA 
a. 

IP 

NP ^ 



( 




N' Det NP 

I l^^l 

teew ngu a yid-aa ngu 

meat the you like-not-it 

the meat, you don't like it 




rawaa-ndu ndu 
dog-NCagr NC 
dog the 

the dog and its owner 



e joom-um 

conj owner-3sgposs 

and owner its 



calmin-mi 



gor-k-o-mo-calmin-mi-o 

man-7 -NCagr-Rel-greet- 1 sg-NC 

the man that I greeted 



Det 




Diop: Vowel deletion in Pulaar 



61 



As shown by the trees in (40a-c) above c-command does not hold in (40a) be- 
tween /ngu/ 'the' and /a/ 'you-sing.' since the first element is dominated by a 
branching maximal projection (NP) that does not dominate the second element /a/. 
In this case neither version of c-command is satisfied; so the two morphemes fall 
outside the domain of vowel deletion. The same reasoning holds for (40b) where 
the determiner /ndu/, daughter of a branching maximal projection, cannot c-com- 
mand the following conjunction Id. In (40c) we see that clearly the pronoun /mi/, 
which is internal to the verbal complex cannot under any of the versions outlined 
above, c-command the determiner lol. So, deletion is expected not to take place 
there either. Another correct prediction that (37) makes is (27j) where we have a 
noun followed by a possessive pronoun /am/. Both of these are within the same 
branching maximal projection; so either version of c-command holds for them, as 
shown in (41) below. 



(41) 



NP 



N" 

I 

oto 



Poss 

I 



However, looking further into the data in (27) we can see that the rule in (37) gets 
in trouble with (27e, h, i, k). In (27e) the final vowel of the first element in an NP 
coordination does not delete whereas it did in (6b), or it fed glide formation, as in 
(8a). (42a) below is a possible representation for (27e). In this representation c- 
command holds between /ndu/ and Id assuming either version; yet the rule fails to 
apply. (42b-d) illustrate trees for cases (24)h, i, and k where the rule also fails to 
apply despite the fact that c-command holds. 

(42) Problematic cases for c-command 
a. 




ndu ^ 

t h i s tI ;i n (J 



joom-um 
o w n c r - i t s 



this (one) and its owner 



62 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 









rokk 


-i \ 




s/h 


e 


give- 


A 


^P 


she 


gave 


you 








Poss 




In all these cases c-command holds but the vowel of the first word is not deleted 
and the initial vowel of the second morpheme is, consequently, realized with an 
initial glottal stop. However, there is a way in which (37) can account for these 
data without any additional stipulations to the rule. It is as follows. In (42a) /ndu/, 
as a determiner, is not the head of the NP. It refers to a noun belonging to the /ndu/ 
class, e.g. /rawaandu/ 'dog'. The head, in this case, is missing. Therefore, what we 
actually have in (42a) is a case of a null-headed NP of the kind introduced in (3 If) 
above. This being the fact, the syntactic representation of (27e) is one in which the 
first member of the coordinate structure is a branching NP that lacks a head, as il- 
lustrated in (43) below. In this case c-command does not hold and the data are ac- 
counted for by the rule in (37), as was the case for (40b) above. 

(43) Further predictions of c-command 

NP 




joomum 



The case illustrated in (42b) above can also be accounted for if we assume the fol- 
lowing facts. As mentioned earlier, (cf. Prunet & Tellier (1984); Paradis (1986, 
1992) for discussion) while postposed subject pronouns are considered part of the 
verbal complex, object pronouns (with the exception of /maa/ and /moo/) are 
viewed as being outside it. One of the arguments used for this analysis of Pulaar 



Diop: Vowel deletion in Pulaar 



63 



verbal complexes by the above-mentioned authors is the fact that subject pronouns 
cause the preceding long vowel within the verbal root to shorten whereas object 
pronouns do not have such effect. Furthermore, they also contend that a high 
vowel from a subject pronoun will cause preceding mid vowels from the preced- 
ing verb root to undergo ATR harmony whereas high vowels from the object pro- 
noun do not cause ATR harmony in verbal roots. So, a verbal complex followed 
by a subject pronoun can be represented as in (44) below. 

(44) Verbal complexes in Pulaar 

VP 



V 

I 
V 



rokk-u-mi 
give-epen- IsgSubj 
I gave 
= ndokkumi 

Since object pronouns do not form a constituent with the preceding verb they 
cannot be represented as in (44) above. So, my analysis is that they were moved 
from that position to a position outside V (by application of the general rule 
Move-a; cf. Chomsky 1986), as shown in (45) below where pro is coindexed 
with the NP object pronoun that has been moved. In this configuration V does not 
c-command the following NP and deletion is not supposed to occur as the two el- 
ements /rokki/ and /on/ fall outside the domain of vowel deletion. 

(45) Pulaar object pronouns: a syntactic representation 
VP 
VP-^ ""-^^ NP 



V 
I 

V 

I 

rokki 



NP 

I 

pro. 



give-Asp 
gave you 



pro 



you-pl 



The arguments that lead to (46b) below are similar to those used to arrive at (43) 
above. Therefore it is not necessary to repeat them all. What is of most relevance 
here is the fact that in a construction like (42d) above the 'complementizer' that is 
used for relative clause formation is used before the possessive pronoun despite the 
fact that a relative meaning is not necessarily implied. That same complementizer 



64 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



is also used in constructions such as 'of + place name' as in (46a) below. So, /mo/ 
whose meaning is something close to 'of is analyzed as a determiner. In this case it 
is part of a branching NP whose head is null, as illustrated in (46b) below. In this 
configuration c-command does not hold and the data are predicted by rule (37) 
above. 

(46) a. gor-k-o-mo-Dimmbee Jooro 

man-7 -NCagr-of-Dimmbee Jooro (name of a city) 
the man from Dimmbee Jooro 




Poss 



am 



The data in (27i), represented in (42c) are a bit different in that we have a focused 
NP. In this case the explanation for failure of the rule to apply could be as simple 
as saying that focused NPs are promoted to the category of major lexical item and, 
therefore, behave like verbs, adjectives, or nouns whose initial vowel never partic- 
ipates in the process of vowel spreading. 

So the rule in (37) seems to be able to account for all the data discussed so far. 
However, (37) along with the c-command versions assumed in (35) above incor- 
rectly predict vowel deletion to be blocked in (27d) which is reproduced here as 
(47a). In (47) we have a branching NP (/ndu rawaandu/ 'this dog') the head of 
which is vowel-final. The rule applies to that head although it does not c-com- 
mand the following conjunction, under either version in (35), as illustrated by 
(47b). 

(47) C-command from branching structures 
a. ndu-rawaa-ndu-e-joom-um 

Det-dog-NCagr-Conj-owner-its 
this dog and its owner 
= ndu rawaandee joomum 




Diop: Vowel deletion in F*ulaar 6 5 

The problem is that Pulaar is full of cases like (47a) above. It is not the case 
that they are just a handful of exceptions that can be dealt with by adding some di- 
acritic to the rule in (37). In fact it is not easy to imagine a proviso that can be 
added to (35) or (37), that is consistent with the different theories of c-command 
in the literature. In (47) there is both a maximal projection and a branching node 
that intervenes between /rawaandu/ 'dog' and Id 'and'. So, neither c-command nor 
government is supposed to hold between these two constituents. Therefore, al- 
though (37) handles all the data presented it cannot account for (27d) and the 
many instances of similar coordination. In such structures the first head noun has 
a determiner and is, consequently, daughter of a branching node. Thus, it cannot 
c-command the following constituent. In the next section I am going to analyze the 
data within the indirect access (also call the end-based theory) framework with a 
view to showing that it too cannot 'straightforwardly' render certain facts of VD 
in Pulaar. First, a look at the basic tenets of the end-based theory. 

4.2 VD in Pulaar and the end-based theory (Selkirk 1984, 1986, 1987) 

Within this approach to domain definition there are two important no- 
tions that play a crucial role in 'the mapping of syntactic representation into that 
hierarchy of prosodic domains which forms the essential constituency of phono- 
logical representation' (Selkirk 1987:152). First, the notion of Designated 
Category (DC). This notion has to do with the idea that for each prosodic category 
P, 'there is a single designated category in the syntax with respect to which 
phonological representation at level Pj is defined' (Selkirk 1987:152). Selkirk ar- 
gues that the basic X-bar levels in syntax, along with Government (more specifi- 
cally L-govemed/non-L-governed; cf. Selkirk (1982) for more discussion) deter- 
mine the different designated category types. So, a designated category could be 
XO, X', or X"/XP (i.e. a maximal projection). The second notion for the end-based 
theory is the End Parameter according to which only one end of a given desig- 
nated category within the X-bar hierarchy is 'relevant in the formation of a 
prosodic constituent Pj: a Pj is claimed to extend from one instance of the appro- 
priate end (R/L) of the DCi to the next (or failing that to the limit of the sentence) 
(Selkirk 1987:152). 

Thus, in (48) below, where it is the designated category is XP and the end 
parameter is left, all three NPs constitute three separate domains on their own 
whereas V is not a domain (because not an XP) though it may be part of the do- 
main of the NP to its left: 

(48) XP/ L 



[NP VP 

[NP (NP V 

If the right edge were to be the parameter then the V would not be within the do- 
main of any of the NPs in (48). It would form a domain on its own. If we change 



66 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



the designated category in (48) above from XP to X and the end parameter to L 
(left), then V would be in a domain separate from that which contains the two NPs 
containing it. If the end parameter is R (right) then V and both NPs preceding it 
would be in the same domain. 

The Pulaar data that I have presented so far establish the fact that when two 
morphemes (the first one vowel-final and the other vowel-initial) are within the 
domain of VD, deletion of the final vowel of the first morpheme is followed by a 
rightward spreading of the initial vowel of the second morpheme. Throughout the 
data one can also establish that the first morpheme within the domain of VD is typ- 
ically the host in the sense of its final vowel undergoing deletion and in terms of it 
hosting spreading from the next morpheme's vowel. Let us also assume for the 
sake of the argument that the prosodic category Pj here is the phonological word. 
The parameters for VD in Pulaar can be set as in (49a) where the designated cate- 
gory is XP and the end-parameter is L (left). What (49a) says is to insert a bracket 
to the left of each maximal projection, in which case anything to its right will be 
included within the same domain (P^ up to the next maximal projection or to the 
end of the sentence. To test (49) against the data presented in this paper I start, 
first, with coordination. Given the representation of coordination that I gave in 
(38) where neither head is dominated by a maximal projection (49a) correctly pre- 
dicts that the first head and the conjunction are going to be within the same do- 
main, as shown in (49b) below. 

(49) Parameters for VD: the left edge 
a. XP/L 
b. 

NP 



N' 



Conj 



gorko e 

[gorkoy debbo] 



N' 

I 
debbo 



(49) also correctly predicts cases where the first noun in a coordinate structure is 
dominated by a (branching) maximal projection, as in (47b) above repeated here as 

(50) below. 



(50) 



NP 



[NP 



Det 
I 

ndu 



Conj 



N' 



N' 

I 

rawaandu 



joomum 



Diop: Vowel deletion in Pulaar 67 

In this and similar cases the first NP and the following conjunction form the same 
domain and VD applies. (49) also correctly predicts cases such as (41) where a 
noun forms a domain w ith the following possessive pronoun because they are both 
dominated by the same maximal projection. Furthermore, (49) also correctly pre- 
dicts VD to be blocked in (40a) because the subject pronoun /a/ is dominated by a 
maximal projection and is, therefore, the beginning of a domain. Consequently it 
is within a separate domain from that containing the vowel-final /ndu/. Other data 
that (49) can easily accommodate are illustrated by (45) where a verb is separated 
from the following object pronoun by two maximal projections at the left of which 
a bracket is inserted to start a domain. VD is expected not to apply in this envi- 
ronment. In (42a) (49) also makes the correct prediction since a bracket is going to 
be inserted to the left of the NP dominating /enen/, putting it in a separate domain 
from the preceding focus marker. 

As one can see (49) can account for a large body of data on VD. However, it 
is not general enough to handle a considerable amount of cases. In the following I 
discuss these. In (38b) the subject pronoun /a/ and the preceding complementizer 
are within the domain of VD because what native speakers say is /maa yiyaani/ < 
/mo-a-yiy-aan-i/ 'that you see-not-Asp'. The subject pronoun is dominated by NP 
and is expected to form the beginning of a domain separate from that containing 
the complementizer; and that is exactly the wrong prediction. Likewise, in (39a-b) 
the algorithm in (49) wrongly predicts that the preposition Id is going to form a 
separate domain from the preceding verb because the preposition is dominated by 
a maximal projection (PP) the left of which is a domain break that puts Id in a sep- 
arate domain from the preceding verb. (49) also wrongly predicts that in (40b) 
/ndu/ (dominated by NP) and the following conjunction are going to be within the 
domain of VD. Likewise, it also predicts that in (40c) the verbal complex 
(dominated by VP) is going to group with the following determiner lol. This is 
wrong. (49) gets in trouble further with (43) where it wrongly groups /ndu/ 
(dominated by the left NP) with the following conjunction within the same do- 
main. The same problem arises with (46b). Therefore, if we maintain (49) we are 
going to have to explain all these exceptions. Instead of doing that let us set the pa- 
rameter as in (51) where the Edge Parameter is set at R (right); everything else is 
going to remain the same as in (49). 

(51 ) Parameter for VD: the right edge 
XP/R 

(51) inserts a bracket to the right of a maximal projection. Everything within that 
maximal projection falls within the domain of VD up to the next maximal projec- 
tion or the end of the sentence. I assume the syntactic representations found in 
(38)-(50) above to be representative of all the data presented in this paper. Given 
that assumption (51), as will be shown shortly, makes the correct predictions for 
all the data except for three cases: (42c-d) and (47). I will first demonstrated the 
non-problematic cases for the right edge, mentioning, where appropriate, the cases 
that do not discriminate between (49) and (51); then I discuss the problematic 
cases. 



68 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

In (38) above rule (51), just like (49), makes the right prediction since the 
same NP dominates /gorko/ and the following conjunction. (51) is able to account 
for the data in both (38b) and (39a-b) whereas these cases were problematic for 
(49). (40a) is unproblematic for both (49) and (51) because they both correctly 
put /ngu/ and /a/ in separate domains. However, (40b-c) are correctly predicted 
only by (51) which puts a bracket to the right of the leftmost NP, and VP, respec- 
tively. (41) is as unproblematic for (51) as it was for (49). As for (42a-b), recall 
that these were said to be the incorrect representation for these sentences. 
Consequently they fall outside the purview of (49) and (51). (43), (45), and (46a- 
b) are also correctly predicted by (51) whereas they were problematic for (49). 

However, as pointed out earlier, (51) also gets in trouble. In particular, it is 
unable to account for (42c-d) and (47). In (42c) the parameter in (51) predicts a 
bracket to the left of the NP dominating /enen/, wrongly putting it and the preced- 
ing fous /ko/ in the same domain. In (42d) where both /mo/ and /am/ are dominated 
by the same maximal projection, (51) just like (49) wrongly puts the two words in 
the same domain. The other problematic case for (51) is in (47). In (47b) a bracket 
to the right of the NP that dominates /ndu rawaandu/ 'this dog' wrongly puts 
/rawaandu/ 'dog' and the following conjunction Id in separate domains whereas 
they should be within the same domain because VD applies in that context. These 
are the three cases that are problematic for (5 1 ) compared to many more problem- 
atic cases for (49). For this reason I am going to choose it to account for VD in 
Pulaar. 

What I have been able to demonstrate so far is the fact that both the direct and 
indirect approach are unable to account for all the data presented in this paper in 
any unified way. For this reason, I propose a rule that has two components. The 
first component is going to ignore any c-command relation (the direct approach) 
or the Edge Parameter (the indirect approach). It targets heads of maximal projec- 
tions and words that are not major lexical category items. It is going to account for 
(42c-d) as well as (47). The second component is going to account for all the re- 
maining data. As I will demonstrate shortly, for the second component both the 
direct and the indirect approach are empirically equivalent. (52a-b) illustrate the 
two different components of the rule. 

(52) Pulaar VD: a final formulation 

a. Delete the final vowel of the final syllable of the head of a maximal 
projection and spread onto that syllable the initial vowel of a 
following word if the latter is not a major lexical category item; 

b. XP/R 

(52a) correctly predicts that in both (42c) and (42d) VD does not take place 
since neither first word in both cases is head of a maximal projection. In (47), 
however, (52a) is going to predict that since /rawaandu/ is a head (of the leftmost 
NP) and the word following it is not a major lexical category item, VD is going to 
apply. (52b) applies after (52a). An ideal situation would have been one in which 
reference is made just to heads of maximal projections and what follows them, as 
indicated in (52a). Most of the data, in fact, could be explained using that refer- 



Diop: Vowel deletion in Pulaar 69 

ence. However, headedness alone is insufficient. The data in (40c), for instance, 
are proof of that. In (40c) we have /calmin-mi/, head of the VP, which does not 
form a domain with the following determiner. (52a) cannot explain that. C-com- 
mand or (52b) above can account for the data there. A further case that would be 
problematic for a solution based solely on the notion of headedness is in (45). In 
this example (52a) would predict /rokki/, head of the VP, to undergo VD. That is 
the wrong prediction. Again in this case, either (52b) or c-command correctly 
predicts VD to be blocked between the verb and the following object pronoun. 
(Since I discussed where c-command holds and where it does not I refer the reader 
to that discussion to better illustrate the fact that c-command, along with (52a) 
predicts all the data presented here in the same way (52a-b) do.) 

Therefore. I conclude by saying that the rule that accounts for the data on 
Pulaar VD, necessarily requires the introduction of a statement like (52a), in addi- 
tion to reference to either c-command or the Edge Parameter. I also come to the 
conclusion that the Pulaar data do not discriminate between the direct or the indi- 
rect approach to the syntax-phonology interface because either approach, along 
with the statement in (52a) correctly accounts for the data. 

In this paper I have tried to do the following. After an introduction to the 
vowel system I presented a first set of data illustrating vowel deletion and vowel 
spreading in Pulaar. In that section the output of VD was shown to be: a long 
vowel, a short vowel, and a vowel-glide sequence. In (80 I also showed that syl- 
labification across morpheme boundaries is observed in Pulaar to the effect that the 
final consonant of a word and the following open-syllable word can form a sylla- 
ble. Following this I introduced facts of the Pulaar syllable that are pertinent to 
the discussion and gave a phonological analysis of VD in the form of a two-step 
process: a (specific) rime merger rule that accounts for the vowel-glide sequences 
and a (general) nucleus merger rule for the cases where VD yields long or short 
vowels. These two rules, I suggested, were ordered following Kiparsky's 
Elsewhere Condition. After a series of derivations illustrating the rules at work I 
presented a new set of data ( 19a-b) that was apparently problematic for my general 
rule but I showed that these data are accounted for without changing or adding 
anything to the rule. The next discussion after that also shows how apparently 
problematic data (24) are easily accounted for without changing the rules. In (24) 
I demonstrated that failure of VD to apply in (24a-0 had to do with prosodic in- 
formation, not the way the rule itself was formulated. In section 4 I showed that 
VD is sensitive to syntactic information. After presenting the data and the syntac- 
tic structures in which the two words that are supposed to be within the domain 
of VD appear, 1 introduced background information about the Pulaar noun, noun 
phrase, relative clause, and verbal complexes. In section 4. 1 I presented the direct 
approach whereas the indirect approach was illustrated in 4.2. I demonstrated in 
these two sections that neither approach can handle all the Pulaar data in any uni- 
fied way and that the correct generalization about the data required combining one 
of the approaches with the notion of headedness. 

As I pointed out at the beginning of this paper there is a third approach to the 
syntax-phonology interfaced that 1 do not discuss here. It has to do with the issue 



70 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

of what phonological rules can tell us about surface syntactic structure 
(Kenstowicz 1987). Another somewhat related issue that I am also not going to 
discuss has to do with the issue of whether phonological rules have access to deep 
(syntactic) structure or whether they can apply before or after the application of 
Move-a. The data in (45) above where VD applies after Move-a has moved the 
object NP raises such questions but I will leave them for further research. 

NOTE 

*This paper stems from chapter three of my doctoral dissertation (Diop 
1993). I am indebted to Charles C. Kisseberth, James Yoon, Laura J. Downing, 
Alessandro Zucchi, and Elabbas Benmamoun for invaluable comments on earlier 
versions. All errors contained in this paper are entirely mine. 

REFERENCES 

CHO, Yu Young-Mee. 1990. Syntax and phrasing in Korean. The syntax- 
phonology connection, ed. by S. Inkelas & D. Zee, 47-62. Chicago: 
University of Chicago Press. 

CHOMSKY, Noam. 1986. Barriers. Cambridge, MA: MIT Press. 

, & Morris Halle. 1968. The sound pattern of English. New York: Harper & 

Row. 

CLEMENTS, G. N. 1978. Tone and syntax in Ewe. Elements of tone, stress, and 
intonation, ed. by D. J. Napoli, 21-99. Washington, D.C.: Georgetown 
University Press. 

DIOP, Abdul Aziz. 1993. Aspects of Fula non-linear phonology: Towards more 
structure in morale phonology. Urbana: University of Illinois Ph.D. dis- 
sertation in Linguistics. 

GOLDSMITH, John. 1990. Autosegmental and metrical phonology. Oxford: Basil 
Blackwell. 

HARRIS, J. 1983. Syllable structure and stress in Spanish. Cambridge, MA: MIT 
Press. 

HAYES, Bruce. 1986. Inalterability in CV phonology. Language 62.321-351. 

. 1989. Compensatory lengthening in morale phonology. Linguistic Inquiry 

20:2.253-306. 

HYMAN, Larry. 1985. A theory of phonological weight. Dordrecht: Eoris. 

. 1990. Boundary tonology and the prosodic hierarchy. The syntax- 
phonology connection, ed. by S. Inkelas & D. Zee, 109-126. Chicago: 
Chicago University Press. 

It6, Junko. 1986. Syllable theory in prosodic phonology. Amherst: University of 
Massachusetts Ph.D. dissertation in Linguistics. 

JACKENDOFF, Ray. 1977. X-bar syntax: A study of phrase structure. 
Cambridge, MA: MIT Press. 

KAISSE, Ellen M. 1985. Connected speech: The interaction of syntax and 
phonology. Orlando, EL: Academic Press. 

. 1987. Rythm and the cycle. Chicago Linguistics Society 23:2.199-209. 



Diop: Vowel deletion in Pulaar 7 1 

1990. Toward a typology of postlexical rules. The syntax-phonology con- 
nection, ed. by S. Inkelas & D. Zee, 127-144. Chicago: Chicago 
University Press. 

KENSTOWICZ, Michael. 1987. The phonology and syntax of wh-expressions in 
Tangale. Phonology Yearbook 4.229-241. 

, & Charles Kisseberth. 1990. Chizigula tonology: the word and beyond. 

The syntax-phonology connection, ed. by S. Inkelas & D. Zee. 163-194. 
Chicago: Chicago University Press. 

KIPARSKY, Paul. 1973. Elsewhere in phonology. A festschrift for Morris Halle, 
ed. by S. Anderson & P. Kiparsky, 93-106. New York: Holt, Rinehart and 
Winston. 

. 1982 Lexical phonology and morphology. Linguistics in the morning 

calm, ed. by I.-S. Yang, 3-91. Seoul: Hanshin. 

MCCARTHY, John. 1986. OCP effects: gemination and antigemination. 
Linguistic Inquiry 17.207-263. 

ODDEN, David. 1986. On the role of the Obligatory Contour Principle in phono- 
logical theory. Language 62.353-383. 

. 1987. Kimatuumbi phrasal phonology. Phonology Yearbook 4.13-36. 

. 1990. Syntax, lexical rules and postlexical rules in Kimatuumbi. The syn- 
tax-phonology connection, ed. by S. Inkelas & D. Zee, 259-278. Chicago: 
Chicago University Press. 

PARADIS, Carole. 1986. Phonologic et morphologic lexicales: les classes nomi- 
nales en peul (fula). Universitc de Montreal Ph.D. dissertation in 
Linguistics. 

. 1992. Lexical phonology and morphology: the nominal classes in Fula. 

New York: Garland Publishing. 

PRUNET, J. F., & C. Tellier. 1984. Interaction des niveaux en phonologic: 
I'abregement vocalique en Pulaar. McGill Working Papers in Linguistics 
2:1.65-90. 

SCHANE, Sanford A. 1987. The resolution of hiatus. Parassession on autosegmen- 
tal and metrical phonology. CLS 23:2.279-290. 

SELKIRK, Elizabeth. 1984. Phonology and syntax: The relation between sound 
and structure. Cambridge, MA: MIT Press. 

. 1986. On derived domains in sentence phonology. Phonology Yearbook 

3.371-405. 

. 1987. Government and tonal phrasing in Papago. Phonology Yearbook 

4.151-182 

, & Tong Shen. 1990. Prosodic domains in Shanghai Chinese. The syntax- 
phonology connection, ed. by S. Inkelas & D. Zee. 313-378. Chicago: 
Chicago University Press. 

SELLS. Peter. 1985. Lectures in contemporary syntactic theories. Stanford: 
Center for the Study of Language and Information. 

STERIADE, D. 1982. Greek prosodies and the nature of syllabification. MIT 
Ph.D. dissertation in Linguistics. 

SYLLA, Yero. 1982. Grammaire modcrne du Pulaar. I^akar-Abidjan-Lome: 
Nouvelles Editions Africaines. 



( 



Studies in the Linguistic Sciences 

Volume 23, Number 2. Fall 1993 [publ. October 1996] 

SUBVERSION OR CONVERGENCE? 
THE ISSUE OF PRE-VEDIC RETROFLEXION REEXAMINED* 

Hans Henrich Hock 

I* dv^; I tricot? Id '^^ \j\\rM'->m: (Sankaracarya'sbhasyaonChandogya-Upanisad5.18.l) 

As is well known, members of at least six distinct language fami- 
lies in South Asia have come to converge to a remarkable degree in 
their overall structure through millennia of hi- and multilingual 
contact. Most of the convergence has been in the syntax, but one 
phonological phenomenon, a contrast between dental and retroflex 
consonants (as in Skt. dJ- 'shine' : dl- 'fly'), has the distinction of 
having been noticed earliest (Pott 1833, 1836). It is this phe- 
nomenon which I address in this paper. 

Some scholars (most recently Thomason & Kaufman 1988, 
Kuiper 1991; see also and especially Emeneau 1980) argue that the 
source of retroflexion is Dravidian, for the dental : retroflex contrast 
can be reconstructed for Proto-Dravidian, while the ancestors of the 
other languages are said to have lacked it. Since the contrast is found 
in the earliest attested stage of Indo-Aryan, Vedic Sanskrit, conver- 
gence between Indo-Aryan and Dravidian must therefore have begun 
in the second millennium B.C., in terms of a SUBVERSION (my 
term) of Indo-Aryan by Dravidian. 

In earlier publications (e.g. Hock 1975, 1984) I claimed that the 
arguments for early convergence are not cogent, since the Sanskrit 
dental : retroflex contrast can be explained by internal Indo-Aryan 
developments. Moreover, the contrast appears to be an innovation 
not only in Indo-Aryan, but also in Dravidian. This raises the possi- 
bility that the feature is a JOINT innovation of Dravidian and Indo- 
Aryan, reflecting direct or indirect bilingual contact. 

At the same time, early Dravidian has a TRIPLE contrast, dental : 
alveolar : retroflex (or post-dontal). This difference may be taken to 
cast doubt on the convergence hypothesis. 

In this paper I present a somewhat speculative hypothesis that a 
triple dental : alveolar : retroflex contrast must be postulated for 
early stages of both Indo-Aryan (and Iranian) and Dravidian and fur- 
ther, that this contrast resulted from joint, convergent innovations. I 
support the hypothesis with comparative Indo-lranian evidence and 
the dialectological evidence of early Middle Indo-Aryan (especially 
the Asokan inscriptions). The latter evidence is especially interest- 
ing, since the development of the hypothetical Indo-Aryan alveolars 
to dentals in the more western regions and to rctroflexes in the more 
eastern regions of Indo-Aryan is closely mirrored by corresponding 



74 Studies in tiie Linguistic Sciences 23:2 (Fall 1993) 

developments of the well-established alveolars to dentals in more 
western Dravidian and retroflexes in more eastern Dravidian. 
Alternative explanations of the observed data either are unnecessarily 
complex or are lacking in explanation. 

The finding that the developments are the result of convergence, 
not of subversion, is significant, for it suggests that the social rela- 
tionship between Indo-Aryan and Dravidian speakers in early India 
was not substantially different from what it is today — a relationship 
of (near-)equality, rather that the traditional picture of marauding 
Indo-Aryan invaders suppressing an indigenous Dravidian popula- 
tion and forcing it to learn their language. 

1. Introduction 

South Asia is a paradigm example of a multicultural, multiethnic, multilin- 
gual area. Members of at least six distinct language families coexist: Indo-Aryan, 
East Iranian, Munda, Austro-Asiatic, Dravidian, Tibeto-Burman, plus an 
"unaffiliated" language in the extreme north, Burushaski.' (For a simplified view 
of where these languages are spoken today see Map I, next page.) While only Indo- 
Aryan and Iranian are closely related, and Munda is remotely related to Austro- 
Asiatic, millennia of bi- and multilingual contact have led to a remarkable degree 
of structural convergence between these different language families and their 
members. As a consequence. South Asia has also come to be known as a paradigm 
case of a CONVERGENCE AREA. (For a good synchronic discussion see Masica 
1976.) 

Four features are commonly Usted as characteristic of this area: 

I. An unmarked major constituent order SOV, i.e. subject (S) before object 
(O) before verb (V), as in example (1); 

II. A tendency to use non-finite absolutives, where modem European lan- 
guages might use dependent clauses with finite verbs (2); 

in. The marking of cited discourse by postposed quotative markers and a 

general absence of indirect discourse (3); 
IV. "Retroflexion", i.e. a phonological contrast between dental and retroflex 

consonants (4). 

(1) Hindi mairii (S) kitab (O) parh raha hOm (V) i am reading a book.' 

(2) Sanskrit tatra gatva (abs.) na mucyase 'When you have gone there, you do 

not get free.' 

(3) Sanskrit nakir vakta 'na dad' iti (quot.) 'Nobody will say, "He shall not 

give.'" 

(4) Sanskrit pata- 'flight' : pata- 'portion' 

As is common in convergence areas,2 these features do not cover the en- 
tire region: Kashmiri places finite verbs in second position in main clauses and 
certain dependent clauses. SVO features are found in many Munda/Austro- Asiatic 
languages. Hindi-Urdu, Kashmiri, and many other Indo-Aryan languages, but 



Hock: Subversion or convergence? 



75 



also Dravidian Brahui mark direct discourse by proposed ki, not by postposed 
quotative markers. Assamese and much of Tibeto-Burman lack retroflexion. 




Illllllll 


Burushaski 


1 1 


Dravidian 






KWM 


Indo-Aiyan 

Iranian 

Munda 






I----I 


Other Austro- Asiatic 






mm 


Tibeto-Burman 



Map I: Distribution of modern South Asian languages (simplified) 

The presence of all of the features in Sanskrit, the oldest attested stage of 
Indo-Aryan, suggests that the absence of some of these in some of modern Indo- 
Aryan results from secondary developments. On the origin of the Kashmiri verb- 
second position see Hock 1982a. The use of ki as a marker of direct discourse is no 
doubt due to Persian influence (see Hock 1982b, which needs to be updated; see 
also Marlow Forthcoming). The absence of the retroflexion in Assamese probably 
reflects contact with Tibeto-Burman.^ 

The question of when the linguistic convergence of the South Asian lan- 
guages began and which group is responsible for it is highly interesting for any- 
one concerned with the early linguistic and ethnic history of South Asia. 
Attempts to answer the question, however, have led to some controversy. 

Many scholars (most recently Thomason & Kaufman 1988, Kuiper 1991; see 
also and especially Emeneau 1980) argue that the source is Dravidian, since all of 
the four features can be reconstructed for Proto-Dravidian. while the ancestors of 



76 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

the other languages are said to have lacked them. Since the features are found in the 
earliest attested stage of Indo-Aryan, Vedic Sanskrit, Indo-Aryan must according 
to this view have acquired them prior to that stage, in the second millenium B.C. 

The sociolinguistic setting for subversion is usually considered one of in- 
equality: The Indo-Aryan conquest forced the indigenous Dravidians to learn 
Sanskrit, the language of the Indo- Aryans; and as English has been influenced by 
the modem South Asian languages, the structure of Sanskrit was altered by trans- 
fer of Dravidian features. 

The term used in traditional historical linguistics for such a development is 
"substratum influence". Using the terminology of Thomason & Kaufman 1988, 
the process can be characterized as "shift" of speakers from one language (usually 
of lower power or prestige) to another one (of higher power or prestige). For 
brevity's sake let me use the more compact term SUBVERSION to refer to 
"substratum influence" or the effects of "language shift". 

Following the lead of others, I have claimed (e.g. Hock 1975 and 1984) that 
the arguments for prehistoric subversion are not cogent: The syntactic features (I - 
III) are either inherited from Proto-Indo-European or Proto-Indo-Iranian, or are ty- 
pologically natural in early Indo-European; and the feature of retroflexion can be 
explained by internal Indo-Aryan developments. 

Several recent reinvestigations of early Dravidian and Indo-Aryan/Indo- 
European syntax support the claim that the syntactic features, in their broad out- 
lines (and including at least one other feature, the use of relative-correlative struc- 
tures), were shared by the prehistoric ancestors of Dravidian and Indo-Aryan, go- 
ing back to periods much earlier than the Indo-Aryan migration to South Asia. 
(See Steever 1988, Hock 1988b, 1992a, as well as Hock 1996).4 For these rea- 
sons, and to keep the present paper within manageable limits, I concentrate on 
retroflexion, bringing in other evidence only where relevant to the argument. 

In contrast to most earlier subversionist claims, but also breaking with my 
own earlier counterclaims, I present a speculative argument that retroflexion can 
be explained as resulting from CONVERGENCE, a process different from subver- 
sion, both in its effects and in its social setting. While subversion consists of the 
unidirectional transfer of features from one language to another, under conditions 
of strong inequality and sudden shift, convergence is a more complex, mutual or 
bidirectional development through which languages in long-standing bilingual 
contact come to be more similar in their overall structure. The required extended 
bilingualism is best maintained in a situation of approximate social equality; but it 
can also arise under other conditions, such as a "social imperative" of maintaining 
ethnic, religious, etc. identity by preserving linguistic distinctiveness.^ 

I argue that prehistoric convergence took place under social conditions that 
fostered extended bilingualism, similar to what we find in modern South Asia; 
that it involved Dravidian, Indo-Aryan and East Iranian, and possibly other lan- 
guages as well; and that it led to a triple contrast (at least in Indo-Aryan and 
Dravidian) of dental : alveolar : retroflex, not just the simple dental : retroflex con- 
trast ordinarily postulated for Indo-Aryan. 



Hock: Subversion or convergence? 77 

2. A survey of earlier views on Indo-Aryan retroflexion 

Early scholars such as Pott 1833, 1836 and Caldwell 1855 could simply as- 
sert that Indo-Aryan (Sanskrit) retroflexion results from Dravidian subversion. 
But as time progressed it became necessary to go beyond such sweeping state- 
ments and to state more precisely HOW subversion exerted itself. 

What was especially troubling is that in its general outlines early Indo-Aryan 
retroflexion could be explained by purely internal developments, with parallels in 
other languages (see e.g. Konow 1906. Bloch 1925). Compare the schematization 
in (5). and see also §5.1 below. Similar developments can be found in other Indo- 
European languages, most notably in Swedish and Norwegian dialects. Ever since 
Biihler 1864, anti-subversionists have taken these facts as evidence that we do not 
need to invoke Dravidian subversion to explain Indo-Aryan retroflexion. 
Something like a compromise position was offered by Konow 1906 and Bloch 
1929 who claimed that Dravidian influence may have accelerated or aided in the 
propagation of these developments. 

m rv V 

*lizdha- > *lizdha- > lldha- 'licked' 
*vista- > vista- = vista- 'entered' 
*wiss > *wits > vit 'people, clan' (N sg.) 
*wissu > *witsu > viksu (id.) (L pi.) 

(-♦ post-RV vit-su) 

Emeneau 1956 and Kuiper 1967a introduced a much stronger and more spe- 
cific claim: The presence of retroflexion in Dravidian led to the "redistribution" of 
pre-Indo-Aryan allophones as retroflex phonemes. Kuiper identified these allo- 
phones as Indo-Iranian *i'and *z, elements generally recognized as the "triggers" 
for Indo-Aryan retroflexion, as in the above formulation.^ This is now the stan- 
dard subversionist position and has been accepted in Thomason & Kaufman 1988, 
the major general monograph on linguistic contact. 

Subversionists moreover believe that it is highly unlikely that Indo-Aryan 
retroflexion arose independently from thai of Dravidian, on the same South Asian 
subcontinent. And they consider irrelevant the fact that other Indo-European lan- 
guages have developed retroflexion, since Indo-Aryan is the only EARLY branch 
of Indo-European with this feature (see e.g. Tikkanen 1987: 284). 

Antisubversionists are not convinced of the logic of this argument: While it 
is true that Indo-Aryan retroflexion developed much earlier than retroflexion in 
other Indo-European languages, this does not mean that it must result from sub- 
version. Different languages may exhibit similar phonological changes at different 
rates and at different stages. (Gothic, for instance, virtually leveled out the effects 
of Verner's Law many centuries before the other Germanic languages; but this 
does not mean that Gothic leveling resulted from subversion.) The fact that other 
Indo-European languages were able to acquire the feature demonstrates that 
retroflexion is not such an unusual phonological phenomenon that it must per- 



(5) 


I 




n 




a. 


(*ligh-to- 


>) 


*lizdha- 


> 


cf. 


(*wik-to- 


» 


*wista- 


> 


b. 


(*wik-s 


» 


*wiss 


> 


cf. 


(*wik-su 


» 


*wissu 


> 



78 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



force be attributed to outside subversion — pace Tikkanen 1987:284 or the extreme 
viewof Bhat 1973. 

At least some antisubversionists would however admit that the issue of 
whether it is LIKELY that Indo-Aryan retroflexion arose independently from 
Dravidian is a more serious one. Unfortunately, it is impossible to determine the 
likelihood of two similar phenomena arising independently in languages that 
come to share the same geographic area. As observed by Lyle Campbell (p.c. 
1993), such an event is not impossible, as shown by the case of Brazil: Portuguese 
has come in contact with indigenous languages which, like Portuguese, have a 
contrast between oral and nasal vowels; but we know that the contrast existed be- 
fore contact, in both groups of languages. Now, the Brazilian case merely estab- 
lishes the possibility of chance similarity; it tells us nothing about its statistical 
likelihood. Nevertheless it further supports anti-subversionist reservations about 
the need to attribute Indo-Aryan retroflexion to Dravidian subversion. 

More concrete arguments against Dravidian subversion are based on struc- 
tural evidence.'^ Following Bloch 1925, Hock 1975 and 1984 observes a number 
of differences between early Dravidian and Indo-Aryan: Dravidian has a triple con- 
trast (dental : alveolar : retroflex), while Indo-Aryan is considered to have a simple 
contrast between dental and retroflex (plus post-dental, alveolar r). Dravidian 
permits final retroflex and alveolar sonorants, early Indo-Aryan does not, except 
for the onomatopoetic nonce-form bhan and coined terms of indigenous graiimiar. 
(Sanskrit word-final r is realized as h utterance-finally.) From the earliest times, 
Indo-Aryan has at least one initial retroflex consonant (in Skt. sat '6' and deriva- 
tives); Dravidian initial retroflex consonants are a late innovation. Indo-Aryan has 
retroflex sibilants which are absent in Dravidian, while the latter has a retroflex 
approximant r which is absent in Indo-Aryan.^ These extensive differences, sum- 
marized in Tables 1 and II, are considered difficult to explain if Indo-Aryan 
retroflexion resulted from Dravidian subversion. 

The Brazilian parallel is interesting in this regard: The phonological effects 
of the oral : nasal vowel contrast, which we know to be of independent origin, dif- 
fer considerably: In the indigenous languages, nasal consonants tend to become 
pre- or post-"oralized" next to oral vowels, a phenomenon without counterpart in 
Portuguese. 



SANSKRIT 




DRAVIDIAN 




DENT. 


ALV. 


RETR. 


DENT. 


ALV. 


RETR. 


STOP t 




t 


t 


t_ 


t 


th 




th 








d 




d 








dh 




dh 








SIB. s 




s 








NAS. n 




n 


n 


n 


n 


LIQU. 1 


r 






[ 
r 


1 
r 



Table I: Differences between the early Sanskrit and Dravidian systems 



Hock: Subversion or convergence? 79 

SANSKRIT DRAVIDIAN 
Final retr./alv. sonor. - + 

Initial retr./alv. + 

Idiosyncratic s + - 

Idiosyncratic r - + 

Table II: Other differences between early Sanskint and Dravidian 

Note further that many of the early phonological differences between Indo- 
Aryan and Dravidian disappear toward the modern period (except in the extreme 
south and northwest), as shown by Ramanujan & Masica's 1969 areal study of 
modern South Asian phonology. As argued in Hock 1984, in contrast to the pre- 
historic situation, this development does provide robust evidence for structural 
interaction. But it took place at a considerably later time and it involved conver- 
gence, not subversion. 

In Hock 1975 I suggested that the dental : retroflex contrast may be an inno- 
vation, not only in Indo-Aryan, but also in Dravidian. My claim was based on 
speculative, and in one case clearly premature, attempts to genetically link 
Dravidian with outside languages which do not have the contrast (Uralic and 
Elamite). In a publication not accessible to me at the time, Zvelebil 1970 proposed 
a 'highly speculative' hypothesis that Dravidian consonant clusters and geminates 
result from large-scale assimilatory processes, some of which turned sequences of 
retroflex — or alveolar — sonorant plus dental stop into retroflex or alveolar stops. 

Drawing on Zvelebil 1970 and on my 1975 suggestion, Tikkanen 1987 
claims that retroflexion is innovated both in Dravidian and in Indo-Aryan. He at- 
tributes the impetus for the innovations to two separate substrata (295) and claims 
that convergent developments between Indo-Aryan and Dravidian took place later. 
On the Indo-Aryan side, he believes that the source for subversion was an un- 
known northwestern substratum, which in his view is also responsible for the 
large amount of early Indo-Aryan lexical items that can be traced neither to Proto- 
Indo-European nor to any of the known non-Indo-Aryan languages of South Asia. 
As for Dravidian, he entertains the idea that subversion is attributable to 'some lost 
sub- or adstratum in the pre-Indo-Aryan period' (323). 

Given what we know — or do not know — about the distribution of languages 
in prehistoric South Asia, Tikkanen's proposal cannot be rejected out of hand. In 
fact, the modem presence of the language isolate Burushaski in the northwest may 
be taken as evidence for a prehistoric presence of a non-lndo-Aryan/non-Dravidian 
language in the area. (But see below on the difficulties in trying to draw prehis- 
toric inferences from the modern situation.) Unfortunately, the hypothesis of an 
unknown substratum (or of several such substrata) is methodologically dubious, 
since by definition it is not open to verification or falsification. 

Tikkanen is now doing intensive research on the northwestern languages of 
South Asia.^ It is to be hoped that this research will eventually make it possible to 
identify a likely substratum; but the enormous time difference between the Indo- 



80 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Aryan arrival in South Asia and the first attestation of the northwestern languages 
places formidable obstacles in the way. 

Moreover, the question must remain as to what the relation was, if any, be- 
tween the two separate substrata that gave rise to Indo-Aryan and Dravidian 
retroflexion. Is it likely that the two substrata had developed retroflexion inde- 
pendently? Methodologically, invoking two separate substrata is problematic in 
that it merely projects the issue of Dravidian/Indo- Aryan prehistoric relationship 
to an even more remote — and uncertain — period in prehistory. 

An alternative, and at this stage of our knowledge more feasible, hypothesis 
is that Indo-Aryan (as well as East Iranian) and Dravidian retroflexion and alveo- 
larization are not just parallel innovations due to subversion by different unknown 
substrata, but that they result from CONVERGENT changes. It is this hypothesis 
which I want to support in the present paper, leaving open the question whether 
convergence took place under direct contact or whether it may have been mediated 
by other, intervening languages. 

Before doing so let me briefly discuss some of the subsidiary arguments that 
have been raised in support of the view that Indo-Aryan retroflexion resulted from 
prehistoric Dravidian subversion. An examination of these arguments demon- 
strates the great difficulties facing anyone trying to make inferences about the pre- 
historic linguistic scene in the northwest (or any other part of South Asia) and the 
fact that any hypothesis about prehistoric contacts in this area — and their linguis- 
tic consequences — must by definition be speculative. At the same time, reexami- 
nation of one of the arguments establishes a possible building block for a new hy- 
pothesis. 

3. Subsidiary arguments for prehistoric subversion 

The greatest difficulty in dealing with the linguistic prehistory and early 
history of South Asia is the fact that we have more or less contemporary evidence 
from only one language family, Indo-Aryan; and even for this family the evidence 
is limited, because the early texts are composed in a language, (Vedic) Sanskrit, 
which is quite conservative and puristic. For other languages, we have to depend 
on evidence from much later periods. This is especially true for the non-Indo- 
Aryan languages spoken in present-day northwestern South Asia which are not at- 
tested before the nineteenth century. 

The problems caused by this situation can be illustrated by examination of 
two arguments frequently raised in favor of prehistoric Dravidian subversion of 
Indo-Aryan. 

One argument is based on the presence of a Dravidian language, Brahui, in 
Baluchistan. The geographical isolation of Brahui is taken to establish that it is a 
relic language, especially since migration is believed to normally take place only 
from north(west) to south. These facts are considered to legitimize the assumption 
of a Dravidian presence in the prehistoric northwest of first Indo-Aryan settle- 
ment. Further support is found in the fact that two other Dravidian languages, 
Kurukh and Malto, which with Brahui form the North Dravidian subfamily, are 



Hock: Subversion or convergence? 



81 



spoken fairly to the north (in eastern Central India), suggesting that Brahui was 
part of a Dravidian subfamily which extended over a vast portion of northern 
South Asia. 

A second argument, cited by Thomason & Kaufman 1988, rests on 
Southworth's 1974 attempt to establish a major east-west division of Indo-Aryan 
languages for the time of the Asokan inscriptions (see Map II) and to link this di- 
vision to more recent evidence that in his view suggests greater Dravidian influ- 
ence in the west, i.e. in a region closer to the area of first Indo-Aryan settlement. 



EASTERN 




Map II: Dialect divisions of the Asokan inscriptions according to 
Southworth 1974 (with reference to Bloch 1950) 

Now, some 3,000 years separate the arrival of the Indo- Aryans from the time 
that North Dravidian languages begin to be attested. This fact in itself should give 
us pause. But there are more specific reasons for caution. 

According to their own traditions, the Kurukh (and Malto) people migrated 
to their present locations from Karnataka. via the Narmada valley (Gricrson 1903- 
1928, V. 4; Hahn 1911); see Map III. .Se\oral linguistic facts support this tradi- 
tion: Bloch 1946 points out that the place names in present-day Kurukh and 



82 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



Malto territory are Munda, not Dravidian, in origin. Kuiper 1966 demonstrates 
linguistic influence of Kurukh on Nahali and Kurku, a fact which supports the 
Kurukh tradition of an earlier settlement in the Narmada valley. Bhat 1971 pro- 
duces linguistic evidence that Koraga in South Karnataka (see Map III) is more 
closely related to the North Dravidian languages than to the rest of Dravidian. In 
short, we have cumulative evidence that connects North Dravidian Kurukh and 
Malto to the south, not to the extreme northwest of Brahui. 






h * * * 




Burushaski 

Dravidian 

Indo- Aryan 

Iranian 

Munda 

Other Austro- Asiatic 

Tibeto-Burman 



Map III: Northern Dravidian languages and migrations 

In fact, Bloch (1911, see also 1925, 1929) has suggested that Brahui. too, 
may have a southern origin, since according to their own traditions, the Brahuis 
have migrated to the area in which they live now.'*' 

Such a northward migration would in fact not be unusual. As is well 
known, several Indo- Aryan groups likewise have followed this route, or migrated 
even farther. These include Gandhari or Niya Prakrit in early medieval Khotan 
and farther east; modem Dumaki in northwestern South Asia; the Parya who came 
to modem Uzbekistan via Afghanistan (Comrie 1981); and the 'Gypsies' or Dom 
who, via Central Asia, have spread all over Eurasia. 



Hock: Subversion or convergence? 8 3 

The present-day linguistic distribution, some 3000 years "after the fact" thus 
cannot be taken as cogent evidence for a prehistoric Dravidian presence in the 
northwest. 

In all fairness, however, it must be admitted that the possible southern origin 
of modern North Dravidian does not preclude a Dravidian presence in the prehis- 
toric northwest. We know that the just mentioned transplanted Indo-Aryan lan- 
guages "remigrated" northward from areas well to the south. Thus, Comrie 1981 
with references shows that the language of the Parya is closely affiliated to 
Hindi/Panjabi, and Kuiper 1966 adduces evidence for Kurukh influence on Dom 
in the Narmada valley. At the same time, we also know that the Indo-Aryan lan- 
guages originally moved into South Asia from the northwest. Given the Indo- 
Aryan precedents for southward migration and subsequent remigration to the 
north, it is possible that there were Dravidians in the northwest when the Indo- 
Aryans came to South Asia, that these Dravidians moved southward under Indo- 
Aryan pressure (or that their languages died out), and that the present-day location 
of Brahui results from remigration to the north. But it is just as possible that the 
Dravidians. if they originally came from the north, had already departed from the 
northwest to the south by the time of Indo-Aryan arrival, and that only later did 
Brahui and the other Dravidian languages move north again. 

The problem is, we simply do not have any reliable independent evidence 
that would permit a choice between these different possibilities. 

From the chronological perspective, Southworth's dialectological division 
of the Asokan inscriptions rests on firmer grounds, since only a little more than a 
thousand years separate Asoka from the time of Indo-Aryan arrival in South Asia. 
But his dual division is not supported by Bloch 1950. to whom he refers, who in- 
stead suggests a triple division: Center and East vs. [South]West vs. Northwest. A 
recent reexamination of the treatment of (syllabic and nonsy liable) r + dental stop 
in the Asokan inscriptions suggests a different division (superseding Hock 1991). 
If we exclude developments limited to specific lexical items which may be sus- 
pected of being borrowings, we can distinguish four different areas (see Map 
IV"): 

a. A northwestern area with almost exclusively retrotlex outcomes (beside 
cluster representations: Shahba/garhi mainly /r etc., Mansehra /retc); 

b. Southwestern Girnar with predominant dental; 

c. North-central Kalsi with a retrotlex : dental ratio of about 4:1; 

d. An eastern area (Dhauli and Jaugada) with almost exclusive retroflex. 

However, given the proximity of Kalsi to the east, its relatively high 
retrotlex ratio may be attributed to eastern intluence.'- Under this assumption, it 
is possible to resolve Asokan dialectology into three areas: 

a. The northwest (predominant retroflex beside cluster representations); 

b. A central area that originally includes Girnar and Kalsi (predominant 
dental); 

c. The extreme east (predominant retroflex). 

See Map IV (next page). 



84 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 




rt etc. > predomi 
nantly retroflex 
(retr. 20+ : dent, 
rt etc. > retroflex 
beside dental 
(retr. 4 : dent. 1 ) 
rt etc. > predomi- 
nantly dental 
(retr. 1 : dent. 10) 



Map IV: Development of r + dental stop in the Asokan inscriptions 

Significantly, this distribution agrees well with the Modern Indo-Aryan 
outcomes of r + dental stop discussed by Turner 1926 with 1921, 1924, a fact 
which suggests that the Asokan inscriptions, at least on this count, offer a reliable 
window on developments in archaic Middle Indo-Aryan.'-* 

By contrast, the evidence of the Asokan inscriptions does not support 
Southworth's east- west division of Indo-Aryan and the concomitant claim that 
western Indo-Aryan exhibits stronger influence from Dravidian — whatever may 
be the merits of his findings for modem South Asia.''* 

Interestingly however — and surprisingly — the triple north-south division of 
Indo-Aryan as regards dental vs. retroflex outcomes of r -i- dental stop lines up 
amazingly well with a Dravidian areal division between languages in which alveo- 
lar stops have turned into dental vs. retroflex '5 (making allowances for soine dis- 
tributional irregularity in the transition area between dental and retroflex out- 
comes). See Map V (next page).'*' 

To lay the foundation for an account that explains this parallelism as the re- 
sult of convergent phonological developments, it is useful to briefly reexamine the 



f 



Hock: Subversion or convergence? 



85 



available evidence regarding the prehistoric and early historic social relationship 
between Indo- Aryans and non-Indo-Aryans. 




MapV: 



t(t) (etc.) retained 
^ tt>t(t) 

^3 ii>t(i) 

tj_>t(t)and t_>d 
t merges with t 



Development of r + dental stop in the Asokan inscriptions and Modern 
Indo-Aryan (according to Turner) compared with the Dravidian 
development of alveolar stops (mainly geminates). 



4. The social relationship 

Explicit, or at least implicit, in the subversionist view of early Indo- 
Aryan/Dravidian contact is the assumption of unilateral inlluence of Dravidian on 
Indo-Aryan. (No mention is ever made of prehistoric Indo-Aryan inlluence on 
Dravidian.) Such a unilateral development requires the assumption that the prehis- 
toric social relationship between Indo-Aryans and Dravidians was one of consider- 
able inequality. Dravidian speakers therefore would have had to speak Indo-Aryan 
and, just as happened to F.nglish in modern South Asia, in shifting to Indo-Aryan 
they transferred structural features of their own language(s). This interpretation of 
Indo-Aryan/Dravidian interaction is apti\ suniniari/ed by limeneau 1956: note 4: 

... it was to their [the Indo-Aryans" | advantage, political, economic, re- 
ligious, to have subjects and proselytes. Absorption, not displacement 



86 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

is the chief mechanism in radical language changes of the kind we are 
considering. 

This view is often considered supported by the belief that the Rig-Vedic 
Indo- Aryans made a strong ethnic distinction between themselves and the indige- 
nous population, the dasas or dasy us, frequently depicting the latter as 'infidels' 
(adeva), and characterizing them as 'black-skinned' in contrast to their own 
lighter hue. While not all subversionists accept this view (Emeneau, for instance, 
does not), it pervades much of the literature on the linguistic and general prehis- 
tory and early history of South Asia, as can be gauged from the following incom- 
plete list of references: Zimmer 1879 (apparently the first propagator of the view); 
Macdonnell & Keith 1912: s.vv. dasa and varna; Chatterji 1960:7 and 32; 
Elizarenkova 1995:36; Gonda 1975:129; Hale 1986:147 (see also 154); Kuiper 
1991:17 (vs. ibid. 3-4); Kulke & Rothermund 1986:35; Mansion 1931:6; Rau 
1957:16; Parpola 1988:104-106, 120-121, 125; see also Deshpande 1979:260, 
1993:216-127. 

Examination of the textual evidence of the Rig- Veda and general considera- 
tions regarding the early interaction of different ethnic groups suggest that, like 
several other aspects of early South Asian society,'"' this picture of radical inequal- 
ity needs to be redrawn. (For a fuller discussion see Hock 1996.) 

The Rig-Vedic passages in which adjectives meaning 'black' or 'dark' are 
used in reference to human enemies are of two kinds. One refers to the forts of the 
enemies, especially their 'womb', a term which may simply refer to their dark in- 
terior. The other passages seem to use the adjective in an ideological/metaphorical 
sense, contrasting the 'dark' world of the ddsas and dasyus with the 'light' world 
of the aryas. As far as I can tell, there is no unambiguous evidence for an awareness 
of color-related "racial" differences in the Rig-Veda. In fact, the notion "race" is a 
problematic invention of the colonial period, quite inappropriate, I believe, in the 
ancient world. 

It is, I think, similarly inappropriate to project the supremacist ideology of 
modem colonial powers like the British into ancient and prehistoric times. True, 
those defeated in war often suffered a cruel fate, even extinction. At the same time, 
both "civilized" empires (such as the Roman one) and "barbarian" ones (such as the 
Huns) were multiethnic, multilingual, and multicultural. War-time alliances kept 
shifting and could pit members of the same ethnic group against each other (such 
as Germanic tribes aUied with the Huns, and with the Romans). In fact, according 
to Classical Sanskrit political theory, alliances were to be made with people living 
on the other side of one's enemy, who would often be ethnically closer to the ene- 
my's party than one's own. 

Most important, the Rig-Vedic evidence suggests a fluid situation of this 
type, in which ethnicity played a relatively minor role. A dasa, Balbutha Taruksa, 
is mentioned as patron of a Vedic seer. Numerous passages refer in one breath to 
dasa and arya enemies and in one of these, both types of enemies are referred to as 
ddeva 'godless, infidel'. Especially instructive is the famous "Battle of the Ten 
Kings" (RV 7:18 with 7:33, 7:83; cf. also Kuiper 1991): On both sides of the bat- 



Hock: Subversion or convergence? 8 7 

tie we find people with "Aryan" names (such as Vasistha vs. Bharata) AND with 
names that sound "non-Aryan" (such as the Srfijayas vs. Simyu). 

The picture that emerges is rather different from the one commonly drawn: 
While there was hostility and warfare between Indo-Aryans and their dasa/dasyu 
opponents, there was no social chasm comparable to that between the British colo- 
nialists and the South Asian people(s) they subjugated. Whatever the ethnic and 
linguistic differences, they did not prevent aryas and dasas/dasyus from making 
shifting alliances with each other, requiring them to interact bi- or multilingually 
on a continuing basis. 

If, then, there was indeed some kind of (direct or indirect) contact between 
Indo-Aryans and Dravidians, the linguistic consequence should be expected to 
have been bidirectional convergence, rather than the unidirectional subversion 
commonly assumed. This is the hypothesis which I will try to support in the re- 
mainder of this paper as regards the origin of Indo- Aryan and Dravidian retroflex- 
ion and alveolarization. 

5. Retroflexion and alveolarization as convergent developments 

As observed in §3, Indo-Aryan clusters of r -i- dental stop underwent a dual 
development in the Asokan inscriptions and in later Indo-Aryan, either to dental 
or to retroflex stop. (Rig-)Vedic evidence shows that this development goes back 
to Old Indo-Aryan times: As is well known, although the Rig- Veda, and Sanskrit 
in general, ordinarily retain r + dental stop, we find occasional Rig-Vedic forms 
of the type (6), usually explained as borrowings from a contemporary "Vedic 
Prakrit".'*^ Note especially (6b) vs. (6c) with retroflex vs. dental outcomes of the 
same element, krta-. 

(6) a. RV karta 'cavity, hole' > RV kata 'cavity, depth' 

b. RV vikrta 'changed; misshapen' > RV vikata 'hideous, terrible' 

c. *krta-vat "having the lucky throw in gambling' > RV kitava 'gambler' 

In the remainder of this section I want to advance the hypothesis that these 
dual outcomes of r -t- dental stop, which we find from Old Indo-Aryan, through 
the time of Asoka, to the present day, go back to an intermediate earlier stage with 
alveolar stops, and that the prehistoric developments giving rise to these alveolar 
stops, as well as to retroflex stops, were convergent with similar changes in Proto- 
Dravidian. 

In support of this hypothesis, recall that the geographical distribution of 
dental vs. retroflex outcomes of r -i- dental in the Asokan inscriptions is remark- 
ably similar to the distribution of dental vs. retroflex outcomes of (geminate) alve- 
olar stop in Dravidian (except in the extreme south which tends to retain the alveo- 
lars); see Map V above. 

If we assume that r + dental changed directly into dental or retroflex in Indo- 
Aryan, this similarity in distribution would be accidental; but if we hypothesize 
that the development took place via an intermediate alveolar, then we can explain 
the similarity as resulting from convergent changes that eliminated alveolar stops 
in both Indo-Aryan and Dravidian (except in the deep south), in favor of either 



88 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

dental or retroflex, depending on geographical region, along north-to-south 
lines. '9 (The fact that "rich" systems with a triple contrast dental : alveolar : 
retroflex may be relieved by merger of the alveolar with either dental or retroflex 
is demonstrated by most of Dravidian. Similar developments have taken place in 
dialectal Norwegian and Swedish; cf. Steblin-Kamenskij 1965.) 

Although many of the specific arguments for the hypothesis are circumstan- 
tial and speculative, and although there are possible chronological problems, I be- 
lieve that the Dravidian/Indo-Aryan parallelism, combined with the Vedic evi- 
dence on the prehistoric and early historic social relationship between Indo- 
Aryans and non-Indo-Aryans, makes it worth while to develop this convergence 
hypothesis so that it can be tested by other scholars. 

§ 5.1 addresses the Indo- Aryan and East Iranian evidence in favor of the hy- 
pothesis. §5.2 deals with the more controversial issue of explaining Dravidian 
alveolar and retroflex stops as early Dravidian innovations. §5.3 draws on §§5.1 
and 5.2 to set out the hypothesis that the Indo- Aryan/East Iranian and Dravidian 
developments are a common innovation. §5.4 discusses certain difficulties regard- 
ing the origin of the retroflex "triggers" for the Indo-Aryan and Dravidian devel- 
opments. §5.5 is concerned with problems of chronology. §5.6 discusses alterna- 
tives and the consequences that arise from not accepting the convergence hypothe- 
sis. 

5.1. Indo-Aryan 

Let me begin with arguments and evidence that make it possible to support 
the "alveolarization hypothesis" that r + dental stop first changed into alveolar stop 
in some variety of Old Indo-Aryan and that this alveolar subsequently merged 
with either dental or retroflex, depending on the dialect. To do so it is necessary to 
remove several possible obstacles and, in the process, to examine relatively arcane 
aspects of early Vedic phonology, as well as parallels in the early East Iranian lan- 
guage, Avestan. The "fringe benefit" of this undertaking is that the alveolarization 
hypothesis raises interesting questions about both the early phonological history 
of Indo-Aryan and the dialectological or area-linguistic relationship between Indo- 
Aryan and ancient East Iranian. 

As noted earlier, Indo-Aryan retroflexion is considered an innovation by all 
scholars, whether they attribute the change to subversion or to internal develop- 
ments. The formulation of the changes in (5), repeated for convenience, is based on 
Hock 1975 and 1979 (with references ).20 

(5) I n ffl lY V 

a. (*ligh-to- >) *lizdha-> *lizdha- >*lizdha- > lldha- 'licked' 
cf. (*wik-to- >) *wista- > *vista- > vista- = vista- 'entered' 

b. (*wik-s >) *wiss > *wiss > *wits > vit 'people, clan' (N sg.) 
cf. (*wik-su >) *wissu > *wissu > *witsu > viksu (id.) (L pi.) 

(-» post-RV vit-su) 

For present purposes, nothing depends on the specific formulation of the 
changes. Under any formulation, however, the changes between the last three 



Hock: Subversion or convergence? 8 9 

Stages are the phonologically most important: From III to IV, two different pro- 
cesses introduce relrotlex stops. Of these, the one in (5a) consists of a fairly ordi- 
nary assimilation of dental stop to preceding retrotlex sibilant, a change with par- 
allels in Swedish/Norwegian (except that here retroflexion is introduced after 
"dark" [{]; see Steblin-Kamenskij 1965); the change in (5b) is more 
"idiosyncratic", involving a pre-Indo-Aryan dissimilation of geminate sibilants 
(see Hock 1987-'). The crucial next step is that from IV to V: Some of the 
"triggers" for the change are lost; as a consequence retrollex stops are no longer 
fully predictable and therefore become phonologically significant. 

The alveolarization hypothesis can a priori be formulated as involving 
changes entirely parallel to those in (5a); and this parallelism may be considered an 
element in favor of the hypothesis. Dental stop assimilates to preceding alveolar 
r,~~ becoming alveolar; loss of the "trigger" r makes the alveolar stop (t etc.) un- 
predictable and hence phonologically significant. See the formulation in (7). 

(7) I n UI 

*krta > *krt_a > *ki/at_a (hence RV kita- : -kata-) 

The hypothesis, however, runs into some empirical obstacles. First, Vedic. 
and following it. later Sanskrit, normally retains r + dental stop. If there was in 
fact assimilation of dental t to alveolar r, the change should only have progressed 
to stage II. 

This particular obstacle can be taken care of by adopting the common as- 
sumption that forms of the type kitava and vikatu are borrowings from "Vedic 
Prakrits", more vernacular varieties of Old Indo-Aryan that coexisted with the 
puristic language of the Vedic poets and which we need to posit on independent 
grounds. 23 All we need to assume is that, in addition to other changes characteris- 
tic of the later Prakrits, the Vedic Prakrits carried out the change(s) in (7). The 
dental or retroflex of "puristic Vedic". then, can be accounted for in one of two 
ways: Either the Vedic Prakrits had already changed the alveolar into dental or 
retroflex, depending on dialect; or in the borrowing process the Vedic Prakrit 
alveolar was nativized variably as dental or retroflex. (The second alternative 
might receive support from the fact that river names which can plausibly be con- 
sidered Tibeto-Burman in origin show suffix variation between -ta or -ta (Wii/el 
1995) which may result from different nativizations of a Tibeto-Burman element 
-ta, with alveolar; see note 20 abo\e.) 

Even so, we might expect puristic Vedic to at least show traces of stage II, 
with alveolarization still predictable because the trigger for the change is still 
present. The Vedic Pratisakhyas, however, make no mention of an alveolar ar- 
ticulation of dentals after r, even though they observe all kinds of other fine 
phonetic details. 

Fortunately, this difficulty, loo, is amenable to explanation; but signifi- 
cantly, the explanation raises interesting issues for early Indo-Iranian dialectol- 
ogy, as well as for any contact-induced account of retroflexion and alveolarization, 
whether of the subversion or of the convergence variety. 



90 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

As George Cardona has reminded me (p.c. 1991), according to Atharva- 
Pratisakhya 1.101-102 and Rik-Pratisakhya 6.13-14, a svarabhakti vowel is reg- 
ularly inserted in Vedic recitation between r and consonant. ^4 As a consequence, 
dental stops would not be directly preceded by r in this variety of Old Indo- Aryan 
and thus would not become alveolar. 

Svarabhakti actually had a more general motivation than just to keep alveolar 
r apart from dental stops: As is well known, r is the weakest of the Old Indo- Aryan 
consonants. (For instance, it is the only consonant that is not permitted as a gemi- 
nate.) If we assume that it was especially weak in syllable-coda position, we are 
able to account not only for the fact that the (Vedic) Prakrits lost it in this position 
(with compensatory length on the preceding vowel if r was non-syllabic) but also 
for the fact that r is the most pervasive trigger for the gemination of neighboring 
consonants in puristic Vedic (a compensation for the weakening of r and its loss of 
mora-bearing ability). Svarabhakti then can be seen as an alternative to r-weaken- 
ing employed in careful recitation: Insertion of the vowel places r into the onset of 
a syllable and thus preserves it from weakening. (See Howell 1991 for Germanic 
parallels to this dual behavior of r.) 

The Rig- Vedic cooccurrence of forms like kartd- and katd- in example (6a) 
above, then, can be explained as reflecting two different traditions — one being the 
puristic tradition of mainstream Sanskrit which in Vedic times pronounced kartd- 
as [kar^ta] and in so doing preserved the [r] as well as the dental articulation of /, 
the other a Prakritic tradition which did not have svarabhakti and which therefore 
permitted rt to develop to an alveolar stop (with loss of [r], except for compen- 
satory lengthening). 

As it turns out, the early East Iranian language of the Avesta exhibits a simi- 
lar dual treatment of r -i- consonant. Here, too, we find a general tendency to insert 
a vowel, generally a, as in (8a). But in combinations of r plus voiceless dental stop 
we fmd the alternative outcome s, as in (8b). With other voiceless stops we find a 
similar variation, as in (8c) vs. (8d). 

(8) a. karata 'done' (cf. Skt. krta) 

baratar 'carrier' (cf. Skt. bhartr) 
b. asa 'truth' (cf. Skt. rta) 

x^asa 'food' (from x^ar 'eat') 
cf. c. vahrka 'wolf (cf. Skt. vrka) 
vs. d. Gath. maraka vs. YAv. mahrka 'destruction' 

Of specific interest for present purposes is the dual development of r -i- r clus- 
ters, because it is highly reminiscent of the relationship between the 'puristic" 
Rig-Vedic type kartd [karata] and the 'Prakritic' type kdtd, see (9).--'' That is, in 
both languages, combinations of r (or r) -i- 1 either are broken up by svarabhakti or 
are fused, as it were, into a new sound. 

(9) 



Old Indo- Aryan 


Avestan 


kata (from kata) 


asa 


karta [karata] 


baratar 



Hock: Subversion or convergence? 9 1 

But there is more than simple parallelism: Hoffmann 1958/67. 1971 inter- 
prets Av. .s-'as a voiceless, perhaps retrollex. lateral, comparable to Pashlo retrotlex 
r from r?.26 A priori, of course, it is possible that .v' designates an alveolar, rather 
than a retroflex. In that case the assimilation in (7) may have been shared by Indo- 
Aryan and East Iranian. In this regard it may be significant that the "retroflex" 
consonants of Pashto and other northwestern languages are commonly described 
as (post-)alveolar, not retroflex. 27 

Support for considering the two phenomena to be related comes from the fact 
that there is a remarkable shared idiosyncracy as regards svarabhakti. All four 
Pratisakhyas (AP 1.101, RP 6.13-14, TP 21.15-16, VP 4.16) rule out 
svarabhakti in the context between r and a sibilant + stop cluster; of. (10a). Gatha- 
Avestan normally does not have a-insertion in the same context; cf. (10b); occa- 
sional forms with a-insertion, such as aihi.darasia can be accounted for as ana- 
logical on the model of related forms with legitimate a-insertion. such as the root 
y/daras- 'see'. Even more important, where *r + voiceless sibilant is not followed 
by t, a-insertion is absolutely regular. (All of the more than 27 Gatha-Avestan oc- 
currences of such forms have a-insertion.) 

(10) a. Vedic: No svarabkakti between r/r and Sib. -t- Stop 
b. GAvest.: Normally darstoisca (Y 33.6) etc. 

Occasionally aibl.darasta (Y 21.2, 50.5) motivated by 
Vdaras- 

The lack of vowel insertion in (10) is especially noteworthy since the clus- 
ters involved arc more complex than those in which insertion does take place. It is 
therefore highly unlikely that this restriction on svarabhakti — and svarabhakti it- 
self — are independent phenomena in Vedic and Avestan. 

We can thus conclude that early Indo-Aryan and East Iranian share a dual 
treatment of r + consonant, one with svarabhakti, the other without. While the 
precise conditions for the choice between these two treatments is not entirely clear 
in Avestan, in Indo-Aryan it appears to be socially conditioned: Svarabhakti is a 
feature of puristic Vedic, its absence a feature of more vernacular Vedic Prakrits. 

In the case of r + t this variation is responsible for a dual development. 
Svarabhakti permits retention of both r and /; its absence results in an interaction 
between the two consonants. The outcome of this interaction most likely was an 
alveolar, given the evidence for a retracted articulation of Avestan .s-'and the fact 
that the variation between dental and retrollex in puristic Vedic can be explained 
as reflecting an earlier alveolar in the Vedic Prakrits. 

The present account raises interesting questions regarding prehistoric lin- 
guistic contacts in northwestern South Asia and neighboring areas. But its signif- 
icance and fruitfulness extend farther: By pointing out parallel phenomena which 
ignore the boundary between Indo-Aryan and Iranian and by providing a 
UNIFIED explanation for these phenomena it raises important questions for early 
Indo-Iranian dialectology and/or areal linguistics. 



92 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

5.2. Dravidian 

Most Dravidianists would accept that morphophonemic alternations of the 
type (11) show that certain instances of Dravidian alveolar and retroflex stops re- 
sult from secondary, assimilatory developments; see e.g. Zvelebil 1970 and 
Krishnamurti In Press (as well as Tikkanen 1987). 

(11) Dravidian retroflexion and alveolarization (data from Tamil) 

a. cen 'go' + -t- + -en : centen 'I went' 
un 'eat' -i- -t- -i- -en : unten 'I ate' 
kol 'kill' + -t- + -en : konten 'I killed' 
al 'rule' -i- -t- -i- -en : anten 'I ruled' 

b. kal 'stone' -1- tun 'pillar' : kattun 'stone pillar' 
kal 'booze' -t- tantan 'gave' kattantan 'gave booze' 

Following Krishnamurti 1961, Zvelebil 1970:178-18028 attempts to extend 
this explanation to account for root variations of the type (12a), with final alveolar 
or retroflex sonorant alternating with alveolar or retroflex stop. As shown in 
(12b), the stops of these forms can be derived from the alternating sonorants by 
"fusions" parallel to those responsible for the alternations in (lib). Subsequent to 
fusion, the resulting forms evidently were reinterpreted as simple roots in their 
own right, ending in alveolar or retroflex stop. (See Krishnamurti In Press for a 
comprehensive discussion of this reinterpretation, its pervasive nature in 
Dravidian, and its consequences for Dravidian morphology.) 

(12) a. Tam. kal 'air, wind' : kattu (id.) 

Tam. urul 'to roll (itr.) : uruttu (id., trans.) 

b . kattu < kal-tu 
uruttu < urul-tu 

The discussion of Krishnamurti (In Press, see also 1995) almost exclusively 
deals with verbal roots; and perusal of DEDR yields ample evidence for verbal root 
alternations that can be explained along the lines of (12b). But Zvelebil' s Tam. kal 
: kattu shows that alternations also occur in nominal roots; and while such alterna- 
tions do not appear to be as numerous, the examples in (13) illustrate that they are 
not limited to just one or two words. (The examples are drawn from Zvelebil 1970 
and from DEDR; numbers in parentheses indicate the entry in DEDR.) 

(13) Tam. il 'house' (494) : itai 'inside of a roof, eaves of a house ...' (528) 
Tam. al 'man ...' (399) : atti 'woman ..." (400) 

Tam. kal 'air, wind' : kattu (id.) 

Tam. oil 'some, few, small' (1571) : citu 'small, etc' (1594) 

Tam. col 'fine rice' (Zvelebil) : cotu 'boiled rice' (2897) 

Tam. neru-nal 'yesterday' (3578) : neUu 'recently' (ibid.) 

Tam. pan 'song, melody' (4068) : patu "sing, chant ...' (4065 

Tam. palli 'hamlet' (4018) : pati 'town, city, hamlet" (4064) 

Tam. purai 'ulcer, fistula' (4297) : putj^u 'anything scrofulous or cancerous' 

(4336) 

Tam. purai 'hole, tube ...' (4317) : putti 'flask, bottle' (4265a) 

Tam. peru, per 'great' (441 1) : peUam 'greatness' (4425) 



Hock: Subversion or convergence? 93 

Tarn, perai 'box, chest' (4442) : petti 'box, chest ../ (4388) 

Tam. pol 'hollow object' (4604b) : potai 'hole, hollow' (4604a) 

Kan. mala 'other, next' (4732) : Ta. matu 'another, other, next' (4766) 

Tam. val 'thong, lash' (5305) : valam 'cable, cord, bowstring ...' (5220) 

Tam. vlr/virutu 'aerial root' (5431) : vitutu (id.; ibid.) 

For Zvelebil, accounting for alternations of this type actually was of minor 
significance. His major claim is the 'highly speculative and hypothetic' proposal 
that many (though not all) consonant clusters and geminate consonants of 
Dravidian can be 'further analysed ... as results of assimilations' (178). 

It remained for Tikkanen 1987:285 to interpret Zvelebil's account as sup- 
porting my earlier, rather poorly substantiated claim that all of Dravidian 
retroflexion and alveolarization is the result of secondary developments. As 
rikkanen states it. 

Both alveolarization and retroflexion of dental stops in [P]roto- 
Dravidian are ... reflections of the same coarticulative process, i.e. the 
retraction of the point of articulation after retroflex and alveolar sono- 
rants (with or without subsequent merger) ... (285) 

The processes involved are summarized in (14). As in Indo- Aryan, the first 
step (stage 1 to stage II) consists in straightforward assimilations. The loss of some 
:^f the triggers for the change (II to III) then makes the alveolar and retroflex stops 
jnpredictable and therefore phonologically significant.-'^ 

Ill 



:i4) 




1 


II 




I 


a. 


Retroflexion: 


*nt > 


nt 










*lt > 


It 


> 


t 






*rt > 


rt 


> 


t 


b. 


Alveolarization: 


*nt > 


nt 










*lt > 


iV 


> 


t_ 






*rt > 


rt 


> 


t 



Dravidianists like Zvelebil, Krishnamurti, and Emeneau acknowledge that 
many instances of Dravidian alveolar and retroflex stops can be explained along 
the lines of (14), but they are evidently not prepared to accept the view that, in 
principle, ALL Dravidian alveolar and retroflex stops are amenable to such an ex- 
planation. For instance, Krishnamurti 1995 distinguishes between two types of 
[ilveolars and retroflexes, the "alternating" type (12)-(13) and another type for 
which there is no evidence of alternation. Only the alternating type is considered 
to result from the changes in (14), within the linguistic history of Dravidian; the 
non-alternating type, by contrast, is believed to be directly inherited from Prolo- 
Dravidian. 

This is indeed a possible interpretation of the evidence. But a simpler inter- 
pretation would be that ALL alveolar and retroflex stops result from changes of the 
type (14). 

The alternations in (12)-(I3) do not seem to be confined to any particular 
subgroup of Dravidian and therefore must be considered a feature common to all of 



94 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Dravidian. No evidence requires the assumption that they arose in a post-Proto- 
Dravidian stage. It is therefore entirely possible that, just like the "non-alternat- 
ing" type, they go back to Proto-Dravidian. 

Moreover, lack of alternation does not guarantee different origin from the al- 
ternating type. It is certainly possible that "non-alternating" forms have the same 
origin as alternating ones and that the two types merely differ in terms of whether 
or not the original forms with root-final alveolar or retroflex sonorant happen to 
have been preserved (in meanings that are still relatable to those of the derived 
forms). In this regard note that a large number of early Indo- Aryan (Vedic) forms 
with voiced retroflex stop are synchronically "non-alternating"; it is only because 
we have access to earlier, pre-Proto-Indo-Aryan stages that one can propose for 
some of them the same historical derivations as for synchronically "alternating" 
ones (see Mayrhofer 1986- : 1: 69, 187, 204, 313, 385 (with Vine 1987), 413, 
415; 2: 49, 136, 326, 357, 387). Even with this access to earlier stages, a number 
of "non-alternating" voiced retroflex stops (and other retroflex consonants) remain 
unexplained. Significantly, however, it is because we have access to these earlier 
stages that we know Indo- Aryan retroflexion to be an innovation. 

A possible counterargument is that, in contrast to early Indo-Aryan, the 
number of Dravidian "non-alternating" forms is very large. But given the rela- 
tively late attestation of the Dravidian languages, the large number of "non-alter- 
nating" forms may simply result from the fact that over the centuries and millen- 
nia, many of the sonorant-final base forms have become obsolete, or that their 
meanings have diverged too much to permit linking them to roots in alveolar or 
retroflex stop. If our knowledge of Old Indo-Aryan had to be derived solely from 
Middle or even Modern Indo-Aryan sources, the number of "non-alternating" 
retroflexes would no doubt be much greater, too. 

Moreover, once phonologically significant alveolar and retroflex stops have 
arisen, it is possible to extend these stops to new contexts. In Indo-Aryan, for in- 
stance, Hoffmann 1941 argues that retroflex -nd- is common in words belonging 
to two semantic categories, of "roundness" and of "breaking, crushing", and may 
have been analogically introduced in many of these words because of their mean- 
ing; see note 20 above. Indo-Aryan retroflex consonants are also commonly used in 
newly created onomatopoeia (Hoffmann 1956). And they are found in many sus- 
pected borrowings. Similar developments may have introduced some of the "non- 
alternating" alveolar and retroflex stops in Dravidian. Note in this regard that 
retroflex consonants are very common in Emeneau's 1969 collection of Dravidian 
onomatopoeia. 

The hypothesis that all Dravidian alveolar and retroflex stops are an innova- 
tion along the lines of (14) moreover provides a motivation for the often-noted ab- 
sence of these stops in initial position: The clusters that gave rise to them, i.e. se- 
quences of sonorant followed by dental stop, are highly unlikely to have occurred 
in initial position. For similar reasons alveolar and retroflex stops are barred from 
initial position in Norwegian and Swedish, and the Rig- Veda has initial retroflex- 
ion only in one word {sat 'six' and derivatives) where it results from distant assim- 
ilation. 



Hock: Subversion or convergence? 95 

5.3. The "convergence hypothesis" 

Accepting the hypothesis that all of Dravidian alveolari/.ation and retrotlcx- 
ion is an innovation and that varieties of early Indo-Aryan and East Iranian 
changed r + dental stop to alveolar stop has important consequences. 

As shown by the comparison in (15) of the major changes^" that gave rise to 
Dravidian and Indo- Aryan/East Iranian alveolar and retroflex stops, these changes 
arc remarkably similar to each other. 

(15) Dravidian Old Indo-Aryan 

a. RETR. *nt > nt *st(h) > st(h) 

*h > It > t *zd(h) > zd(h) > d(h) 

b. ALV. *nt > nt_ *rt(h) > rt_(h) > Ved. Pkt. t_(h)) 

*lt > It > t *rd(h) > rd(h) > Ved. Pkt.'d(h) 

The parallelism of these changes is striking enough to require a reassessment 
of the prehistoric linguistic interaction between Dravidian and Indo-Aryan/East 
Iranian. 

As we have seen in §2, the early Dravidian and Indo-Aryan synchronic sys- 
tems, as they are usually posited, are sufficiently different that antisubversionists 
may doubt the cogency of the Dravidian subversion hypothesis and would there- 
fore attribute the presence of retroflex consonants in both groups to chance. 

Such a chance explanation becomes extremely difficult to justify if the early 
Dravidian and Indo-Aryan/East Iranian systems result from the changes in (14); 
for the parallelism of the developments is simply too great. On both sides, dental 
stops assimilate to the same class of preceding consonants — alveolars and 
rctroflexes. On both sides, the preceding consonants are higher in sonority than 
the dental stops — alveolar and retroflex sonoranls in Dravidian. alveolar sonoranl 
and rcironex continuant in Indo-Aryan/East Iranian. And on both sides, the same 
thing happens to make the results of assimilation unpredictable and therefore 
phonologically significant — loss of some of the triggers for the changes. 

Even the differences between the two early systems becomes smaller, since 
under the present hypothesis, an alveolar series is found, not only in Dravidian, 
but also in Indo-Aryan/East Iranian. Any remaining differences, such as idiosyn- 
cratic Dravidian r vs. idiosyncratic Indo-Aryan .s/c. are of comparatively minor 
significance and can be attributed to preexisting differences between the two 
groups. (But see also ^5 A below.) 

Now, the changes in (14) are innovations in both Dravidian and Indo- 
Aryan/East Iranian. Moreover, there is no evidence that would force us \o locate 
the origin of the changes in one or the other linguistic group and to assume that 
they spread to the other group by subversion. Rather than making an arbitrary 
choice it is preferable to consider the changes to reflect CONVERGENCE between 
the two groups. This interpretation finds support in the fact that, as we have seen 
in §4, the evidence of the Rig- Veda is compatible with assuming a social situation 



96 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

favoring the continuing bilingualism that would encourage convergence. 
Moreover, as shown in Hock 1988a, it is often difficult in convergence areas to 
pinpoint a particular language that may have been the source for a given shared 
innovation. 

The convergence hypothesis, if correct, is significant, for it suggests that the 
social relationship between Indo- Aryan and non-Indo-Aryan speakers in early In- 
dia was not substantially different from what it is today — a relationship of (near-) 
equality that encouraged continuing bilingualism, rather than the traditional pic- 
ture of marauding Indo-Aryan invaders suppressing an indigenous Dravidian 
population and unilaterally forcing it to learn their language. 

Note further that convergence does not require direct contact. As shown in 
Hock 1988a, languages in a convergence area behave very much like dialects in a 
monolingual situation; innovations can spread from language to language, eventu- 
ally covering a vast territory. The convergent changes in (14) therefore could have 
resulted from mediated contact, possibly involving some ancestral form of 
Burushaski, Tikkanen's unknown northwestern substratum, and yet other lan- 
guages. As a consequence, the convergence hypothesis does not depend on a reso- 
lution of the — highly controversial — question of whether there was a Dravidian 
presence in the prehistoric northwest (see §3) — a clear advantage of the conver- 
gence hypothesis over the subversion account. 

A quasi-dialectological view of convergence further makes it possible to ac- 
count for a fact that has so far been glossed over: While early Indo-Aryan has both 
retroflexion and alveolarization, Avestan offers no evidence for an assimilation of 
dental stop to preceding retroflex sibilant. In fact, most varieties of Middle and 
Modern East Iranian likewise do not exhibit such a change. Exceptions are 
Middle Iranian Saka (hista 'sent', related to Skt. ista 'sent') and Modern Pashto 
(with lar 'ache' < *diizdah) and SanglTci-Iskasml (with t< st < st); see Emmerick 
1989, Skjaerv0 1989a,b, Payne 1989, and the discussion and references in 
Tikkanen 1987: 289.31 But the Pashto development is limited to voiced sibilant -t- 
dental stop; voiceless *si results in dental t; and the Sangllci-Iskasmi retroflex out- 
come t alternates with dental t. Even Nuristani, somewhat intermediate between 
Iranian and Indo-Aryan, only has variable traces of this development; see Tikkanen 
1987:287-288 (with references). In fact, the earlier change of f to s, which pro- 
duced the trigger for retroflexion, likewise shows more limited or variable distri- 
bution in Nuristani and East Iranian. 

Under the convergence hypothesis, the more limited and variable distribu- 
tion of the change s> s and of the retroflexion of dental stop after s in Nuristani 
and East Iranian finds a ready explanation as a peripheral, transition-area phe- 
nomenon. The core area of the change must have been in South Asia proper, from 
which the change spread only incompletely to the Nuristani and East Iranian lan- 
guages on the northwestern periphery, before coming to a complete halt in geo- 
graphically even more remote Iranian territory. 

Alveolarization after r evidently was more "vigorous" in the northwest. ^2 
Plausible effects of the change are found in Saka (Emmerick 1989: 215), modem 



Hock: Subversion or convergence? 9 7 

Sanglicl-Iskasmi, Yidgha, Pashto, and Parachi (Skjairv0 1989), as well as of 
;ourse in ancient Avestan. In Avestan. Saka, and Yidgha, however, the change is 
■estricted to clusters of r + voiceless dental stop. This restriction would, again, be 
:onsonant with the view that East Iranian was on the periphery of South Asian 
;onvergence and therefore only partly affected by it. 

The convergence account of alvcolari/ation and retroflexion, thus, proves to 
3e multiply fruitful. Not only does it explain the otherwise inexplicable paral- 
elism in the changes that gave rise to alveolar and retroflex stops in Dravidian and 
!ndo-Aryan/East Iranian, it also provides a principled account for the more vari- 
ible effects of these changes in East Iranian and other languages on the northeast- 
ern periphery. Moreover, it does so without requiring the highly controversial as- 
iumption of direct prehistoric contact between Dravidian and Indo-Aryan — a 
:lear advantage over the subversion hypothesis. Finally, it is consonant with the 
i^ig-Vedic evidence for the social relation between Indo-Aryans and non-Indo- 
\ryans, a relationship that does not differ significantly from what we encounter in 
ater, historic times. 

5.4. The triggers for retroflexion and alveolarization 

In spite of all its advantages, however, the convergence hypothesis encoun- 
ers some problems of its own. The most important of these concerns the origin of 
he retroflex and alveolar continuants and sonorants that triggered the develop- 
iient of retroflex and alveolar stops. 

On the Indo-Aryan side the picture is reasonably clear: Indo-Aryan /■ no 
joubt was alveolar to begin with (Hock 1992c). Indo-Aryan retroilex s/z contin- 
jes Proto-Indo-Iranian s/z. and as argued in Hock 1975, 1984. the change of ,v7£'to 
■etroflex can be motivated by the principle of polarization (for which see the more 
general discussion in Hock 1986/1991), to maintain the contrast with distinc- 
ively palatal .s'from Proto-Indo-Iranian c. Tikkanen 1987: 289 argues against this 
explanation by observing that Proto-Indo-Iranian r does not change to palatal .v'in 
Nurisiani, which nevertheless has retroflex .v. But as he himself notes, the 
Vuristani counterparts of Proto-Indo-Iranian .v7z'show considerable fluctuation be- 
;ween retrollex, a sibilant marked s, and even dental sibilant. ^-^ (In fact, Nuristani 
raises difficulties for Indo-Iranian comparative linguistics, including the usual re- 
:onstruction s/z\ see e.g. Morgenstierne 1975a.) This variability is compatible 
with the peripheral position of Nuristani noted in §5.3. It is therefore possible 
Lhat the occasional Nuristani retroflex counterparts of .v/'£' result from the spread of 
[he Indo-Aryan change s/z>s/z. In that case, there would be no problems, since as 
noted, polarization is well motivated in Indo-Aryan. 

But what arc the sources for the Dravidian alveolar : retrollex contrasts in the 
sonorants — r vs. r, / vs. /. n/n vs. /)'.' 

As far as I can tell, there is nothing in the comparative Dravidian literature to 
suggest that this contrast is secondary, comparable to the one between Old Indo- 
Aryan retroflex .v, palatal .s', and dental .v.^-* However, examination of the data in 
DEDR yields a number of semantically relatable entries whose major difference 



98 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

lies in the presence of a retroflex sonorant in one entry and a corresponding alveo- 
lar sonorant in the other; see the data in (16). If these, and perhaps other, similar 
doublets, should indeed turn out to be related, then the retroflex : alveolar contrast 
of the Dravidian sonorants must be secondary, resulting from some kind of phono- 
logical split (whose conditions may no longer be recoverable). In that case, the 
prehistoric parallelism between Indo-Aryan (and East Iranian) and Dravidian 
might extend even farther than envisaged in this paper. 

(16) Tam. anai 'approach...' (120) : Tam. anuppu 'send (off)' (329) 
Tam. en 'thought' (793 ) : Tam. en 'say so, utter ...' (868) 
Tam. inai 'join together' (457) : Tam. inam 'class, group'(531) 
Kan. gal(i) 'air, wind' (1499) : Tam. kal 'air, wind' (1481) 

Tam. kar 'blackness, blemish ...' (1494) : karu 'black' (1278a), kar 

'blackness ...' (1278c) 
Tam. kol 'strike, hurt' (2152) : Tam. kol 'kill, murder' (2132) 
Tam. nal 'night' (3621) : Tam. nallam 'blackness' (3613) 
Kan. talisu 'pound, beat' (3130) : Tam. tallu 'beat, crush' (3105) 
Tam. pariccu 'praise' (4003) : Tam. paracu 'praise' (3951); cf. parattu 

(4092) 'applaud' 
Tam. purai 'hole, tube' (4317) : Tam. purai 'tube ...' (4197) 
Tam. muru 'all, entire' (4992) : Tam. muraiicu 'be full, abundant' (4970) 
Tam. muraiiku 'roar ...' (4989) : Tam. mural 'make noise ...' (4973) 
Tam. varahku 'move ...' (5292) : Tam. var- 'come ...' (5270) 
Tam. val 'lustre, splendor' (5377) : Tam. val 'whiteness, purity' (5364) 

While to my knowledge the alternations in (16) have not been noted in earlier 
Dravidianist literature, an alternation that can be linked with this type has re- 
ceived attention, although it also has been subject to some disagreement. As 
Subrahmanyam 1983:350 reports, Krishnamurti 1961 'on the basis of a small 
number of examples, talks about alternation of *t_ with */ ...'. The alternations are 
given in fuller detail in Zvelebil 1970:98 and 102 (see also 178-179); see example 

(17) which also includes references to DEDR. As Zvelebil notes, the alternation is 
especially common in Telugu, 'where verb bases with *t have transitives with */. 
In To[da] there are also traces of this alternation' (102). 

(17) Tam. kati 'chew, bite ...' (1390) : kati 'bite, bite off (1 124) 
Tam. vata 'dry up, shrink' (5320) : vatu 'wither, fade' (5342) 
Toda piry 'dust' etc. (with */ according to Zvelebil) (4481) : Tam. poti 

'powder, dust ...' (ibid.) 
Toda kwidy 'a family of children' (1655)'''' : Tam. kuti 'house, family' 

(ibidO 
Tam. citu 'small, etc' (1594) : cittu 'anything small' (2513) 

While Krishnamurti and Zvelebil evidently are convinced that alternations of the 
type (17) are not just accidental, Subrahmanyam considers the connection doubt- 
ful 'since the two sounds are kept distinct in numerous etymologies' (1983: 
350). And Burrow and Emeneau 1984 in their introduction to DEDR consider 
the explanation of the alternation 'still uncertain' (xv). 



Hock: Subversion or convergence? 99 

The last item in (17) makes it possible to argue that the connection is valid 
(even though the explanation may be different from Krishnamurti's). What is rel- 
evant here is that the two stop-final roots coexist with a third root which ends in 
alveolar sonorant: Tam. cil 'some, few, small' (1571); compare perhaps also Tam. 
til 'small piece; potsherd ...' (1577). 

The relationship between cil and citu is, of course, of the type (13) above, 
where ciijt can be derived from earlier *cil-tu, with alveolar root-final sonorant 
followed by dental suffix. (In fact, cil : citu is included in the examples under 
(13).) Given the evidence in (16) it is now possible to account for the form cittu as 
derived from a parallel form *cil-ttu, with root-final retroflex sonorant followed 
by dental suffix. By extrapolating from this well-supported case it is possible to 
account for the other pairs of forms in ( 17) under the following assumptions: 

• Proto-Dravidian had an alternation of root-final alveolar and retroflex sono- 

rants (whose origin is at this point obscure). 

• Just like other root-final alveolar and retroflex sonorants these alternating 

sonorants could be extended by dental stops and could thus yield alveo- 
lar and retroflex stops. 

• As no doubt happened with many roots in "non-alternating" alveolar and 

retroficx stops, the original sonorant-final root may have become obso- 
lete, thereby making it appear that in most cases the alveolar : retroflex 
stop alternation in (17) is primary, rather than secondarily built on an 
original alveolar : retroflex sonorant alternation. 

The fact that the hypothesis of a Proto-Dravidian root-final alveolar : retroflex al- 
ternation thus helps explain the alternations in (17) shows that the hypothesis is a 
fruitful one. 

What is most significant for present purposes, however, is that the hypothe- 
sis raises an interesting question regarding the prehistory of South Asian alveolar- 
ization and retroflexion. Given that alveolar /■ and / arc less "marked" than their 
retrofiex counterparts, it is reasonable to explain the alternation as the result of a 
change from alveolar to retrofiex sonorant (under as yet unknown conditions). The 
triggers for stop retroflexion, then, are the result of an innovation. Now, the Indo- 
Aryan trigger for stop retroflexion, the retroflex sibilant (whether voiced or voice- 
less), likewise results from an innovation. What, then, is the likelihood that the 
Dravidian and Indo-Aryan innovations were independent from each other? Should 
we conclude that these changes, too, were convergent? And if so, what arc the im- 
plications for the chronology of South Asian alveolariz.ation and retroflexion? 

1 do not have any answers to these questions, and perhaps it will never be 
possible to give a satisfactory reply. But the fact that the hypothesis advanced in 
this paper encourages such questions may be taken as a further element in its fa- 
vor. 

5.5. Problems of chronoloj^y 

Beyond the somewhat hypothetical chronology problems raised toward the 
end of the preceding section, there is a much more concrete chronological prob- 



100 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

lem. Recall that the starting point for this paper's hypothesis consisted in the sim- 
ilar geographic distribution of dental and retroflex outcomes of alveolar stops in 
Dravidian and r + dental stop clusters in Indo-Aryan. This similarity was ex- 
plained under the assumption that Indo-Aryan r -v dental stop did not directly 
change to dental or retroflex, but that it did so via an intermediate stage with alve- 
olar stop. The alternative would have been to consider the similarity to be acciden- 
tal. 

From the geographical perspective, this line of argumentation is quite rea- 
sonable. But the chronology creates greater problems: Even if we assume that 
Vedic forms like the kitava and \ikata of example (6) are to be explained as differ- 
ent nativizations of Vedic Prakrit forms with alveolar stops (see §5.1), the evi- 
dence of the Asokan inscriptions shows that by the third century BC, the dialec- 
tally differentiated merger of alveolar stop with dental or retroflex had been com- 
pleted. In the oldest stages of the literary Dravidian languages, however, distinct 
alveolar geminates are maintained not only in Old Tamil, but also in the earliest 
records of Kannada, i.e. as late as the ninth to tenth century AD. Only Old Telugu 
(7th century AD) no longer distinguishes alveolar from retroflex geminates. (See 
Zvelebil 1970:100.) The evidence of Kannada suggests that the Dravidian merger 
took place some thousand years after the Indo-Aryan one. 

Does this mean that we have to consider the geographical similarities in the 
distribution of dental and retroflex outcomes to be accidental? If so, do we have to 
abandon the idea that Indo-Aryan r -i- dental stop changed to dental or alveolar stop 
via an intermediate alveolar stop? And what are the effects for the convergence hy- 
pothesis of this paper? 

The chronological difference is indeed troublesome. But there are possible 
ways of getting around this difficulty. One possibility is that the Old Kannada 
texts reflect a conservative form of the language which retained the dental : alveo- 
lar distinction, while the popular language had long abandoned it — an early stage 
of diglossia. 

A second possibility lies in taking a closer look at the geography: Within the 
literary Dravidian south, the merger of geminate alveolars with dental or retroflex 
is even later in Tamil than in Kannada; and Malayalam still preserves geminate 
alveolars. That is, the merger appears to have been spreading from north to 
south — and at a fairly slow pace. If we back-project the direction of the spread, we 
will eventually reach Indo-Aryan territory; and we may hypothesize that the 
change originated there. Now, except for "transplanted" texts like those of 
Yerragudi, the Asokan inscriptions come from locations considerably to the north 
of the literary Dravidian languages. If the spread of the merger was as slow-paced 
in its early Indo-Aryan stages as it was later in southern Dravidian, it is possible to 
speculate that the time difference between the Kannada merger and the Asokan 
merger results from the interaction between geographical distance and the slow 
pace of spread. 



Hock: Subversion or convergence? 101 

5.6. Alternatives and their consequences 

If the explanations in the preceding section for the lime difference between 
Dravidian (Kannada) and Indo- Aryan (Asokan) merger of alveolar with dental or 
retroflex are not accepted, it becomes necessary to examine alternative accounts 
and their consequences. 

An obvious consequence of rejecting the explanations would consist in rejec- 
tion of the significance of the geographical alignment between the dental and 
retroflex outcomes of Dravidian alveolars and Indo- Aryan r + dental stop clusters. 
Such a rejection would entail an alternative interpretation of the geographical evi- 
dence as being the result of chance. 

Now. the geographical alignment was an important building block for the 
Indo-Aryan/East Iranian "alveolari/ation" hypothesis. If the alignment is consid- 
ered to be due to chance, one might be tempted to reject the "alveolarizalion" hy- 
pothesis, too, and claim instead that r + dental stop changed directly to Indo- 
Aryan dental or retroflex, without an intervening alveolar stage. The Avestan de- 
velopment of /•/ to s", then, might either be an unrelated phenomenon or, if related, 
simply another instance of r + dental stop directly going to retroflex. 

If the "alveolarization" hypothesis is rejected, then of course the prehistoric 
parallelism between Dravidian and Indo-Aryan/East Iranian is diminished. As a 
consequence, the "convergence" hypothesis might be rejected, too. Additional rea- 
sons for such a rejection might come from the Dravidianist side, by insisting that 
"non-alternating" alveolar and retroflex stops are inherited from Proto-Dravidian, 
rather than the result of assimilations between alveolar or retroflex sonorants plus 
dental stops. 

In that case, we might have to return to earlier subversionist accounts of 
Indo-Aryan and East Iranian retroflexion. Moreover, we would have to choose be- 
tween the simple Dravidian subversion hypothesis favored in traditional accounts 
and Tikkanen's hypothesis of an unknown northwestern substratum. 

What would be the consequences of these various alternatives to the hypothe- 
ses presented in this paper? 

Most obviously, return to subversionist accounts would mean a return to all 
the difficulties that have been observed for such accounts. To my mind the most 
important among these is the fact that unilateral subversion is not what we would 
expect, given the Rig-Vedic evidence on the social relationship between Indo- 
Aryans and non-Indo- Aryans, as well as the general uncertainty as to the identity 
of these non-Indo- Aryans (were they Dravidians, Mundas, or speakers of yet other 
languages?). 

Beyond that, several steps in the arguments against the hypotheses oi' this 
paper call into the question the very foundations of subversionist claims: 

If we attribute the geographical alignment of dental vs. retroflex outcomes of 
Dravidian alveolars and Indo-Aryan/East Iranian r + dental to chance, by what 
right, then, do we decide that the similarities between early Dravidian and Indo- 



102 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Aryan retroflexion can NOT be due to chance? We would have to develop a much 
better theory of chance similarities before we can make such a decision without 
appearing to be arbitrary or self-serving. (True, there is a chronological problem 
with the geographical alignment; but for all we know, there may have been similar 
chronological problems as regards Indo-Aryan retroflexion. The fact that we do 
not have access to relevant information on the chronology of retroflexion does not 
necessarily give us license to assume that there were no problems.) 

Rejecting the "alveolarization" hypothesis likewise raises questions about 
chance: Given the alveolar articulation of Indo-Aryan r, is it likely that combina- 
tions of r -I- dental stop directly went to dental or retroflex, rather than to alveo- 
lar — especially in light of the fact that Dravidian and Norwegian/Swedish furnish 
precedents for dialectally differentiated merger of alveolar with either dental or 
retroflex? Moreover, rejection of the hypothesis makes the early Indo-Aryan 
phonological system more different from that of Dravidian, and thereby reduces 
the plausibility of hypotheses that want to link Indo-Aryan retroflexion to 
Dravidian, whether by subversion or by convergence. 

Rejection of the hypothesis that all Dravidian alveolar and retroflex stops re- 
sult from assimilations of dental stops to preceding alveolar or retroflex sonorants 
has consequences, too, since it rejects a simple, general account in favor of a more 
complicated one, deriving some alveolar and retroflex stops by assimilation, but 
others by inheritance from Proto-Dravidian. Opting for the more complex account 
calls into question a fundamental assumption of subversionists like Emeneau 
1971b and Thomason & Kaufman 1988 that the Dravidian subversion hypothesis 
should be accepted because it is the SIMPLEST account. True, this is not necessar- 
ily an argument against subversion, since as noted in Hock 1996, there are 
independent reasons for doubting the cogency of claims based entirely on simplic- 
ity; but it does constitute a problem for subversionist argumentation. (Ultimately, 
the issue is not merely one of simplicity, but of "Occam's Razor" which states that 
elements in an argument should not be multiplied WITHOUT NECESSITY. While 
the "necessity rider" clearly is relevant, it also opens the way for disagreement 
over when a more complex argument accounts better for the data than a simpler 
one.) 

At numerous points in this paper I noted that the hypotheses advanced in the 
paper are fruitful in that they explain interesting linguistic issues that go beyond 
the question of convergence or subversion. Rejection of these hypotheses would 
require offering alternative accounts for these issues. 

In some cases, this should not be too difficult. For instance, my claims con- 
cerning the alternations in (16) and (17) can be maintained, even if the hypothesis 
is rejected that all Dravidian alveolar and retroflex stops are the result of innova- 
tion. But note that examples such as the last one in (17) suggest that the number of 
lexical items with secondary, rather than "non-alternating, inherited", alveolar or 
retroflex stop may be much larger than is commonly assumed. And this fact may 
raise questions about the claim that there were such "non-alternating, inherited" 
stops. 



Hock: Subversion or convergence? 103 

6. Summary and conclusions 

This paper proposes a set of related hypotheses concerning South Asian 
retroflexion that differ significantly from earlier views. In contrast to earlier 
Dravidian subversion explanations and in contrast to simple rejections of such ex- 
planations, I argue for CONVERGENT changes which introduced not only 
retrotlexion. but also alvcolarization, both in Dravidian and in Indo-Aryan/East 
Iranian. As common in convergence areas, the changes lost momentum on the pe- 
riphery, in East Iranian and Nuristani. Later convergent developments led to the 
merger of alveolar stops with dental or retroflex stops in Indo- Aryan and most of 
Dravidian. The hypothesis of convergence, rather than subversion, finds support 
in the Rig-Vedic testimony regarding the social relationship between Indo-Aryans 
and non-lndo- Aryans. 

My claims and findings are significant on several counts. First, they suggest 
that the prehistoric relationship between Indo-Aryans and non-lndo-Aryans was 
not substantially different from what we find in observable history — a relation- 
ship that encouraged extended bilingual interaction with bidirectional linguistic 
consequences, rather than the usually assumed forced shift from non-Indo-Aryan 
to Indo-Aryan with unilateral linguistic consequences. By drawing on the evi- 
dence of Old Iranian Avestan, I expand the horizon for convergence (or subver- 
sion) hypotheses to East Iranian and, in so doing, raise interesting questions about 
early dialectal or bilingual interactions in Indo-Iranian. The assumption of con- 
vergence, rather than subversion, makes it possible to provide an explanation of 
this relationship (in terms of the peripheral location of East Iranian). Moreover, 
because convergence does not require direct contact, the hypothesis avoids the dif- 
ficulty encountered by subversion hypotheses that independent evidence for pre- 
historic Dravidian/Indo-Aryan contact is highly controversial and that vocabulary 
evidence favors Indo-Aryan contact with neither Dravidian nor Munda, but pos- 
sibly with some unknown northwestern language (Tikkanen 1987, 1988). 
Finally, in the process of developing the convergence hypothesis 1 advance a 
number of subsidiary arguments which shed an interesting light on Indo-Aryan 
and Dra\ idian historical phonology. 

While reactions to earlier versions of this paper by advocates of Dravidian 
subversion suggest that they will not be convinced by my claims, I hope that they 
will consider the arguments presented in this paper to be worthy of serious dis- 
cussion. Whatever the outcome of the discussion, if it is supported by alternative 
explanations and new data, it is bound to advance our understanding of the prehis- 
tory and early history of South Asia and ol historical linguistics in general. 



Notes 

* This paper grows out of continuing research on the issue of prehistoric and 
early historic South Asian convergence. The present paper is a thorough revision 
of Hock 1995, which itself is a revised version of a paper read at the 1992 Annual 
Meeting of the American Oriental Society. A related paper has been presented on 



104 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

numerous occasions, including the 1993 Linguistic Institute at Ohio State 
University, lectures at the Universities of Hamburg and Freiburg, and most 
recently at the November 1994 International Seminar on 'Ideology and Status of 
Sanskrit in India and Asia', International Institute for Asian Studies, Leiden (NL). 
(See Hock 1996.) Part of the research has been supported by grants from the 
University of Illinois Research Board and a spring 1995 sabbatical leave. I am 
indebted to Rahul Peter Das for kindly making a copy of Hoffmann 1941 available 
to me. I am also grateful for comments I received on earlier versions of this paper 
and related papers, especially from Lyle Campbell, George Cardona, Jan Houben, 
Murray B. Emeneau, Bh. Krishnamurti, and Sarah Thomason. I know that the 
three last-mentioned scholars do not agree with many of the claims in this paper; 
but I sincerely hope that our disagreement will stimulate further fruitful 
discussion. As usual, the responsibility for any errors and omissions rests with 
me. 

' Nahali might constitute the remnant of yet another language family (Kuiper 
1966 with references). Witzel 1995 further adds Kusunda in central Nepal, as 
well as possibly other languages, including that of the Veddas. 

2 See e.g. Masica 1976 and Hock 1988a. 

3 Interestingly, in other areas of close contact, Tibeto-Burman has converged 
with Indo-Aryan, by acquiring the contrast. 

4 For Emeneau 's 1974 lexical-syntactic arguments regarding Skt. api : Drav. 
-um, see Hock 1975 and Gil 1994 (apparently independent of Hock 1975). For 
Abbi's 1992 monograph on 'reduplicated' structures, see the review in Hock 
1993b. 

5 See Hock 1986/1991: Chapter 16 for general discussion. The notion 
"convergence" and "convergence area". Germ. Sprachbund, was introduced by 
Jakobson 1931 and Trubetzkoy 1931. Emeneau has introduced and popularized 
an alternative term, "linguistic area". For treatments of more recent South Asian 
convergence, both "global" and more localized, see e.g. Emeneau 1989, Gumperz 
& Wilson 1971, Krishnamurti 1991, Pandharipande 1982. 

6 The precise manner in which Indo-Iranian *.f and *£'gave rise to Indo- 
Aryan (Vedic Sanskrit) retroflexion, the conditions under which the development 
took place, and the extension of retroflexion beyond its original domain are still a 
matter of controversy. For earlier views and literature see Wackernagel 1896, 
espcially pp. 164-177, and Debrunner's supplement of 1957. More recent 
literature is found, and referred to, in Kuiper 1967b. Hock 1975 (with 1974), 
Hock 1984 (with 1979, 1987), Kuiper 1991, see also Hock 1991, as well as note 
20 below. 

7 Deshpande 1979 claims that the Sanskrit dental : retroflex contrast 
developed in post-Rig-Vedic. If correct, this would be another argument against 
prehistoric subversion. However, as noted in Hock 1979, the Rig- Veda offers 
evidence for a highly patterned, rule-governed DEGENERALIZATION of retroflex 



Hock: Subversion or convergence? 1 05 

sandhi across word boundary, an early phase of a change that gets virtually 
completed in the Classical period. We must therefore assume that retrollexion was 
introduced prehistorically, before the attested Rig-Vedic texts. 

^ Dravidian r is occasionally written z, but Krishnamurti 1969:318, n. 18 
notes that there is no strong empirical evidence for this phonetic interpretation. 
Typologically, a system with a voiced obstruent not matched by a corresponding 
voiceless one is rare enough to require more than cursory justification. Note 
further that in the traditional Tamil alphabetical arrangement retroOex /■ holds the 
same position relative to retroflex / as alveolar r to alveolar /. 

^ See for instance Tikkanen 1988. 

'" The value of the tradition is weakened by the claim that the Brahuis came 
from Aleppo, in present-day Syria [!], but this element may reflect a later 
"Islamization" of an earlier tradition according to which the Brahuis are 
immigrants to the area. 

' ' The extreme southeastern inscriptions from Yerragudi are most similar to 
those of north-central Kalsi. Since their language is clearly transplanted (the 
inscriptions are found deep in Dravidian territory), they cannot be relied on for 
dialectological judgments. The evidence of Gandhari Prakrit (with predominant 
dental) must likewise be ignored, since the language is transplanted, too. (The 
dental outcome might suggest an original affiliation closer with southwestern 
Girnar than with northwestern Gandhara, since the northwestern Asokan 
inscriptions have predominant retroflex.) 

'- Note in this regard that unlike the south- and northwestern inscriptions, 
Kalsi does not substitute /• for the / of the eastern inscriptions in words like 
/J/V7//v7/<7'king". 

'■^ As I became aware only after having examined (and reexamined) the 
As'okan inscriptions, Turner did in fact connect the modern distribution to the 
Asokan northwest : central : eastern distribution advocated here. 

'•* Southworth's modern retroflex distribution is based on text frequencies. 
The highest retroflex : dental ratios are found in Sindhi, Gujarati, Marathi and the 
Dravidian south, the lowest ratios in Panjabi, Hindi, the Bihari languages, and 
Bangla, with the remaining areas having an intermediate ratio. Unfortunately 
Southworth does not indicate the texts on which his statistics arc based. 

Examining versions ol the "Prodigal Son", representati\c ol the different 
languages (and major subdialects) in Grierson 1903-1928, I arrive at rather 
different text frequencies and distributions: The highest retroflex : dental ratios (1 
: 1-2.5) are found in a discontinuous Indo-Aryan area comprising Sindhi. 
Rajasthani. and Pahari dialects, and in Malayalam. Among the major languages, 
Kashmiri, Nepali, and Bangla have the lowest ratios (I : 30 for Kashmiri, 1 : 9-15 
for Nepali and Bangla). Intermediate ratios of 1 : 3-8 are found in most of South 
Asia, including the northwest and most of the Dravidian south. The northwest has 



106 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

a number of pockets with significantly lower retroflex ratios (beside Kashmiri, 
note e.g. Burushaski with 1 : 33, Ormuri with 1 : 13.5, and Khowar with 1 : 
58.5); there are similar pockets in the central area around Nahali (Gondi of Mandla 
with 1 : 13.5, Kurku with 1 : 34; see also Kuiper 1962: 255). 

Whatever these geographical distributions may indicate about the history of 
South Asian languages, they do not support Southworth's grouping of Gujarati / 
and Marathi with Sindhi and the entire Dravidian south, a grouping which is V 
crucial for his claim that there was a strong prehistoric Dravidian presence in 
present-day western Indo- Aryan. 

15 See Subrahmanyam 1983, Zvelebil 1970. Northwestern Brahui does not 
offer any conclusive evidence for geminate alveolar rr; but as Emeneau 1971b 
observes, single alveolar t_ merges with retroflex /, not with the dental; that is, its 
outcome is parallel to the dominant retroflex outcome of r + dental stop in 
northwestern Indo-Aryan. 

16 For most of Dravidian, only the geminate alveolar stop is considered; 
single alveolar stop generally changes to a liquid, commonly an [r]-sound. For 
Tamil, the conservative, literary retention of n is assumed. (The dialectology of 
non-conservative, colloquial, and vernacular Tamil is quite complex and also, to 
my knowledge, not yet fully investigated. It appears that different varieties prefer 
dental or retroflex outcomes.) The geographically easternmost Dravidian 
languages have assibilated outcomes of geminate alveolar stops; these are not 
included in Map V. 

'"7 See for instance Hock 1991 and 1993a on early Indo-Aryan. 

18 Lengthening of the vowel preceding the cluster appears to depend in these 
early attestations on whether the r-sound was non-syllabic or syllabic. In the 
former case, r added a mora in the coda of the syllable so that its loss resulted in 
compensatory lengthening; in the latter case, there was no compensatory 
lengthening. (In later attestations, the loss of non-syllabic r in coda more 
commonly resulted in compensatory lengthening of the following consonant.) 

'9 As I hope to show elsewhere, a similar north-to-south aHgnment between 
Indo-Aryan and Dravidian can be observed in the modem distribution of retroflex 
vs. dental (or rather, alveolar) nasals and laterals. 

20 See also note 6. More problematic is the question of "sponteneous 
retroflexion", as in RV atati 'wanders' vs. later atati. Developments of this type, t 
too, have been attributed to subversion (e.g. Kuiper 1967a, Emeneau 1974); but 
alternative solutions have been proposed. 

Some retroflexes have been explained by sporadic internal developments 
(e.g. dissimilation), others as borrowings from "Vedic Prakrits" or as 
anticipations of changes that become regular in Middle Indo-Aryan; cf. the 
discussion and references in Wackernagel 1889, Hoffmann 1941, Hock 1975, 
1984, and 1991, Vine 1987. 



Hock: Subversion or convergence? 1 07 

If some of the developments should have been the result of contact, languages 
other than Dravidian might furnish alternative, or even more plausible sources: As 
argued in Hock 1984, since Dravidian has a contrast dental : retroflcx, it is 
'difficult to see how [the substitution of retrotlex consonants for dentals] could be 
attributed to the mistakes made by Dravidians trying to speak a Sanskrit with 
undifferentiated dentals (cf. Emeneau 1974). Rather, just as in the case of modem- 
day contacts between Westerners and South Asians, I would expect speakers 
LACKING the contrast to make mistakes in trying to speak a language which has 
it. (Perhaps speakers of early forms of Munda, or of Tibeto-Burman, might be 
involved?)' In fact, if Indo-Aryan "dentals" had really been post-dental/alveolar, 
one would have expected the substitution of Dravidian alveolars; this is precisely 
what we find in Malayalam in nativizations of English words with alveolars. 

Support for Tibeto-Burman provenience of some lexical items with 
"spontenous retroflexion" may be found in the fact that, except where it has 
undergone South Asian influence, Tibeto-Burman has undifferentiated alveolars 
which could be nativized either as dentals or as retroflexes in languages like Indo- 
Aryan which already had a contrast dental : retroflex (whether that contrast was 
due to subversion or not). That this is not just a thought experiment is suggested 
by the evidence in Witzel 1995 for river names ending in -ta or -ta (with apparent 
dental : retroflex variation) at the Himalayan border of Vedic Sanskrit, i.e., an area 
where a Tibeto-Burman presence is most likely. Interestingly, kirata. the name of a 
non-Aryan people mentioned in the Rig-Veda and tentatively identified as Tibeto- 
Burman by Witzel, has a Pali variant with retroflex, kirata. 

Hock 1991 adds the further possibility that some "spontaneous" changes of 
dental to retroflex may have resulted from inner-Indo- Aryan differences suggested 
by the Priltisilkhyas (such that a 'tooth-root' / of one variety of Vedic could be 
reinterpreted as postdental and therefore rctroHex in another variety whose / was 
interdental). 

We should also consider the possibility that retroflexes replaced earlier 
dentals through sporadic analogical developments. In this regard note Hoffmann's 
1941 observation that most Sanskrit words with retroflex -nd- belong to one of 
two semantic categories, that of "roundness" and that of "breaking, crushing". As 
Hoffmann correctly notes, this fact makes it possible that -nd- was secondarily 
extended to words belonging to one or the other of these categories. 

s 21 ^(^ I realized when rereading Kuiper 1967b for this paper, my 19K7 

' account is similar, even though by no means identical, to that of Kuiper. 1 lake 
this opportunity to add the reference to Kuiper's article to my 1987 paper. 

22 On the Vedic articulation of r sec Hock 1992c. 

2-^ Kuiper's 1991 rejection of this explanation ignores the well-known 
independent evidence for the existence of Vedic Prakrits (for which see the 
discussion and references in Hock 1991 ). See also the evalution by Oberlies 1994. 



108 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

24 The Vajasaneyi- and Taittirlya-Pratisakhyas specify svarabhakti only 
between r and sibilant (4.16 and 21.15-16, respectively). This is the context for 
which the Atharva- and Rik-Pratisakhyas teach a fuller (half- or quarter-mora) 
version of svarabhakti, while before stop they recognize a shorter (1/8-mora) 
version. One suspects that the Vajasaneyi and Taittirlya-Pratisakhyas do not 
describe a different 'dialect', but merely overlook the shorter, less noticeable 
variety of svarabhakti. (See further below for a restriction on svarabhakti even 
before sibilant, if followed by stop.) 

25 Earlier discussions of the Avestan situation that I am aware of do not 
consider this parallel with Old Indo- Aryan. Miller 1968 argues that structures of 
the type (8a), with r + a + voiceless dental stop, are morphological renewals 
which, by undoing the conditioned development in (8b), reassert the 
morphological transparency of synchronically analysable forms. Kellens 1989 
and Beekes 1988 ignore Miller's position and reassert the old claim that the 
difference between (8a) and (8b) is accentually conditioned. After a careful survey 
of all relevant Gatha Avestan forms, Beekes is able to maintain this view only by 
claiming that Iranian accentuation differed in a number of forms from the one of 
Vedic, even though the conditions for accent shift are no longer discernible. (He 
further claims that the change to s' was post-Gathic, but I do not find his 
arguments convincing.) None of these approaches offer a satisfactory account for 
the fluctuation in (8d). 

26 The view that the sound designated by the symbol «s» is some kind of 
retracted sound is supported by the fact that it reflects an earlier cluster involving 
POST-DENTAL, alveolar r and that in early Avestan it is distinguished from two 
other ^'-sounds of different, non-alveolar origins: a plain «s» (reflecting ordinary 
Pllr. s) and a palatal «s» (resulting from palatal c + y). (See Hoffmann 197 1 .) 

27 Could the alveolar articulation of retroflex stops in many varieties of 
modern Hindi-Urdu be due to the influence of the Muslim conquerors, at least 
some of whom came from the northwest? Other, apparently less urban (or urbane) 
varieties have strong retroflex articulation, a fact which suggests that alveolar 
pronunciation is an urban overlay. 

28 See also p. 175, note 5. 

29 The fact that in (11a) the triggers are not lost may be cause for concern. 
But their presence can be accounted for as the result of analogical restoration of the 
root-final sonorants; in synchronically opaque structures such as (12) and (13), 
non-nasal triggers regularly are lost. The difference between geminate and non- 
geminate retroflex and alveolar stops in (12) and (13) can be explained with 
Krishnamurti (In Press) as reflecting the difference between geminate and simple 
dental stop in the input. In example (12b), taken from Zvelebil, the inputs 
therefore should be rewritten as *kal-ttu and *urul-ttu. 

30 The (retroflex) sibilant dissimilation in (5b) is an idiosyncracy of Indo- 
Aryan and is therefore not included in the comparison. 



(1 



Htx:k: Subversion or convergence? 1 09 

^' Tikkanen adds Yidgha, for which Skjaerv0 does not hsl the variant with 
retroflex stop. 

-^2 There is also a wide-spread and robust distribution of .v resulting from 
earlier s + r or r + s in the northwest, found in Nuristani, Middle and Modern 
East Iranian, and even northwestern Middle and Modern Indo-Aryan; see 
Tikkanen 1987:287-289. Morgenstiernc 1947:234-235, von Hinuber 1986:28- 
29. The antiquity of the phenomenon is not clear; in Indo-Aryan, the input .v may 
reflect earlier palatal *.s'. Since *s > s is a Middle Indo-Aryan innovation, it 
appears that the change of s + rorr + s may be a relatively recent phenomenon. 

^^ For another source of 5 see the preceding note. 

^"^ Zvelebil 1970:177, however, notes in passing the correspondence Tam. 
annial "nearness, being near' : anpu 'love", which is obviously related to the first 
item in (16) below, but he does not discuss its significance. A further exception 
might be Levitt 1989; but 1 find the arguments of the paper difficult to penetrate. 

35 The DEDR puts a question mark next to the Toda word. 

References 

DEDR = Burrow & Emeneau 1984. 

ABBl, Anvita. 1992. Reduplication in South Asian languages: An areal, typologi- 
cal, and historical study. New Delhi: Allied Publishers. 

BEEKES, Robert S. P. 1988. A grammar of Gatha-Avestan. Leiden: Brill. 

BHAT, D. N. S. 1971. The Koraga language. Poona: Deccan College, 
Postgraduate and Research Institute. 

. 1973. Retroflexion as an areal feature. Stanford University Working Papers 

on Language Universals 13: 27-67. 

BLOCH, Jules. 1911. Review of Denys de S. Bray, The Brahui language, 1. 
Journal asiatique, lOe serie. 17: 1: 162-167. (Repr. in Bloch 1985.) 

. 1925. Sanskrit et dravidien. Bulletin de la Societe de Linguistique de Paris 

25: 1-21. (Repr. in Bloch 1985.) 

. 1929. Some problems of Indo-Aryan philology. Bulletin of the School of 

Oriental Studies (London) 5: 719-756. (Repr. in Bloch 1985.) 

. 1946. Brahui et tsiganc. Journal of the Royal Asiatic Society 1946: 199- 

201. (Repr. in Bloch 1985.) 

. 1950. Les inscriptions d'Asoka. Paris: Les Belles Lettres. 

. 1985. Recueil d'articles de Jules Bloch (1906-1955). ed. by C. Caillat. 

Paris: College de France, Institut de Civilisation Indienne. 

Bi'HLER, George. 1864. On the origin of the Sanskrit linguals. Madras Journal 
of Literature and Science, No. 1, Third Series, 1 16-186. [1 have not been 
able to sec this article; reference in Wackernagel 1896: 165 and elsewhere.] 

BURROW, Thomas, & Murray B. Emeneau. 1984. A Dravidian etymological dic- 
tionary. 2nd ed. Oxford: Oxford University Press. 



1 1 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

CALDWELL, Robert. 1856. A comparative grammar of the Dravidian or South 

Indian family of languages. First edition. London. (Third edition repr. 

1974, New Delhi: Oriental Books.) 
CHATTERJL Suniti Kumar. 1926. The origin and development of the Bengali 

language. Calcutta: University Press. (Repr. 1970, London: Allen & 

Unwin. Also distributed by Motilal Banarsidass, Delhi.) ^ 

COMRIE, Bernard. 198L The languages of the Soviet Union. New York: ^ 

Cambridge University Press. 
DESHPANDE, Madhav. 1979. Genesis of Rgvedic retroflexion: A historical and 

sociolinguistic investigation. Aryan and non-Aryan in India, ed. by M. 

Deshpande and P. E. Hook, 235-315. Ann Arbor: Center for South and 

Southeast Asian Studies, University of Michigan. 
. 1993. Aryans, non-Aryans, and brahmanas: Processes of indigenization. 

Journal of Indo-European Studies 21: 215-235. 
ELIZARENKOVA, Tatyana J. 1995. Language and style of the Vedic rsis. Albany: 

State University of New York Press. 
EMENEAU, Murray B. 1954. Linguistic prehistory of India. Proceedings of the 

American Philosophical Society 98: 282-292. (Repr. in Emeneau 1980.) 
. 1956. India as a linguistic area. Language 32: 3-16. (Repr. in Emeneau 

1980.) 
. 1969. Onomatopoetics in the Indian linguistic area. Language 45: 274-299. 

(Repr. in Emeneau 1980.) 
. 1971a. Dravidian and Indo- Aryan: The Indian linguistic area. Symposium 

on Dravidian civilization, ed. by Andree F. Sjoberg, 33-68. Austin & New 

York: Jenkins Publishing & Pemberton Press. Repr. in Emeneau 1980. 
. 1971b. Kodagu and Brahui developments of Proto-Dravidian */;. Indo- 

Iranian Journal 13: 176-198. (Repr. in Emeneau 1994.) 
. 1974. The Indian linguistic area revisited. International Journal of 

Dravidian Linguistics 3: 92-134. (Repr. in Emeneau 1980.) 
. 1980. Language and linguistic area: essays selected by A. S. Dil. Stanford, 

CA: University Press. 
. 1989. The language of the Nilgiris. Blue Mountains: The ethnography and 

biogeography of a South Indian region, 133-143. (Repr. in Emeneau 

1994.) 
. 1994. Dravidian studies: Selected papers, ed. by Bh. Krishnamurti. Delhi: 

Motilal Banarsidass. 
EMMERICK, Ronald E. 1989. Khotanese and Tumshuqese. Schmitt 1989: 204- 

229. % 

GELDNER, Karl Friedrich. 1951. Der Rig-Veda ..., 3 vols, posthumously edited 

by Charles R. Lanman. Cambridge, MA: Harvard University Press. 
GIL, David. 1994. Conjunctive operators in South-Asian languages. SALA XV: 

Papers from the Fifteenth South Asian Language Analysis Roundtable 

Conference 1993, ed. by A. Davison and F. M. Smith. Iowa City: South 

Asian Studies Program, University of Iowa. 
GONDA, Jan. 1975. Vedic literature. (A history of Indian literature, 1:1.) 

Wiesbaden: Harrassowitz. 



Hock: Subversion or convergence? Ill 

GRIERSON, George. 1903-1928. Linguistic survey of India, 11 vols. Calcutta. 

Repr. 1968, Delhi: Motilal Banarsidass. 
GUMPERZ, John J., & Robert Wilson. 1971. Convergence and creolization: a case 

from the Indo-Aryan/Dravidian border. Pidginization and creolization of 

language, ed. by D. Hymes, 151-168. Cambridge: University Press. 
HAHN, Ferd. 1911. Kurukh grammar. Calcutta: Bengal Secretariat. Repr. (no 

date), Delhi: Mittal. 
HALE, Wash Edward. 1986. Asura in early Vedic religion. Delhi: Motilal 

Banarsidass. 
HINUBER, Oskar von. 1986. Das iiltere Mittelindisch im Uberblick. 

(Osterreichische Akademie der Wissenschaften, Phil. -hist. Klasse, 

Sitzungsberichte 467.) 
HOCK, Hans Henrich. 1974. Historical change and synchronic structure: The 

case of the Sanskrit root nouns. Toward tomorrow's linguistics, ed. by R. 

W. Shuy & C.-J. N. Bailey, 329-42. Georgetown University Press. (Repr. 

in International Journal of Dravidian Linguistics 4: 215-28 (1975).) 
. 1975. Substratum influence on (Rig-Vedic) Sanskrit? Studies in the 

Linguistic Sciences 5: 2: 76-125. 
. 1979. Retroflexion rules in Sanskrit. South Asian Languages Analysis 1: 

47-62. 
. 1982a. AUX-cliticization as a motivation for word order change. Studies in 

the Linguistic Sciences 12:1: 91-101. 
. 1982b. The Sanskrit quotative: A historical and comparative study. Studies 

in the Linguistic Sciences 12: 2: 39-85. 
. 1984. (Pre-)Rig-Vedic convergence of Indo-Aryan with Dravidian? 

Another look at the evidence. Studies in the Linguistic Sciences 14: 1: 89- 

107. 
. 1986/1991. Principles of historical linguistics. Berlin: Mouton de 

Gruyter. (Second revised edition, 1991, ibid.) 
. 1987. Regular contact dissimilation. A Festschrift for Henry Hoenigswaid, 

ed. by G. Cardona & N. Zide, 143-53. Tubingen: Narr. 
. 1988a. Historical implications of a dialectological approach to convergence. 

Historical dialectology, ed. by J. Fisiak, 283-328. Berlin: Mouton de 

Gruyter. 
. 1988b. Review article: Finitencss in Dravidian. (Sanford B. Steever (1988): 

The serial verb formation in the Dravidian languages.) Studies in the 

Linguistic Sciences 18: 2: 211-31 
-. 1991. Dialects, diglossia, and diachronic phonology in early Indo-Aryan. 

Studies in the historical phonology of Asian languages, ed. by W. G. Boltz 

and M. C. Shapiro. Amsterdam & Philadelphia: Benjamins. 
. 1992a. Reconstruction and syntactic typology: A plea for a different ap- 
proach. Explanation in historical linguistics, ed. by G. W. Davis & G. K. 

Iverson, 105-121. Amsterdam & Philadelphia: Benjamins. 
. 1992b. Review of Tikkanen 1987. Kratylos 37: 62-68. 



112 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

. 1992c. Were r and / velar in early Sanskrit? Vidya-Vratin: Professor A. M. 

Ghatage felicitation volume, ed. by V. N. Jha, 69-94. (Sri Garib Dass 
Oriental Series, 160.) Delhi: Sri Satguru Publications. 

. 1993a. A critical examination of some early Sanskrit passages alleged to in- 
dicate dialectal diversity. Comparative-historical linguistics: Indo- 
European and Finno-Ugric: Papers in honor of Oswald Szemerenyi 111, ed. 
by B. Brogyanyi & R. Lipp, 217-232. Amsterdam & Philadelphia: 
Benjamins. 

. 1993b. Review of Abbi 1992. Studies in the Linguistic Sciences 23: 1: 

169-192. 

. 1995. Subversion or Convergence? The issue of pre-Vedic retroflexion re- 
examined. Symposium on Language and Prehistory in South Asia, 
University of Hawaii, March 1995. 

. 1996. Pre-Rgvedic convergence between Indo-Aryan (Sanskrit) and 

Dravidian? A survey of the issues and controversies. Ideology and status of 
Sanskrit: Contributions to the history of the Sanskrit language, ed. by J. E. 
M. Houben, 17-58. Leiden: Brill. 

HOFFMANN, Karl. [1941]. Die alt-indoarischen Worter mit -nd-, besonders im 
Rgveda. Unpublished doctoral dissertation, Miinchen. 

. 1956. 'Wiederholende' Onomatopoetika im Altindischen. Indogermanische 

Forschungen 60: 254-264. (Repr. in Hoffmann 1975.) 

. 1958/67. Altiranisch. Handbuch der Orientalistik 1: 4: 1: 1-19. Leiden: 

Brill. (Repr. in Hoffmann 1975:58-76.) 

. 1971. Zum Zeicheninventar der Avesta-Schrift. Festgabe deutscher 

Iranisten zur 2500Jahrfeier Irans, 64-73. (Repr. in Hoffmann 1975:316- 
326.) 

. 1975. Aufsatze zur Indoiranistik, vol. 1, ed. by J. Narten. Wiesbaden: 

Reichert. 

HOWELL, Robert B. 1991. Modern evidence for ancient sound changes: Old 
English breaking and Old High German vowel epenthesis revisited. 
Staefcraeft: Studies in Germanic linguistics, ed. by E. H. Antonsen & H. H. 
Hock, 65-73. Amsterdam: Benjamins. 

JAKOBSON, Roman. 1931. Uber die phonologischen Sprachbiinde. Travaux du 
Cercle Linguistique de Prague 4: 234-240. 

KELLENS, Jean. 1989. Avestique. Schmitt 1989: 32-55. 

KONOW, Sten. 1906. (Discussion in vol. 4: 279 of Grierson 1903-1928.) 

KRISHNAMURTI, Bh. 1961. Telugu verbal bases: A comparative and descriptive 
study. Berkeley & Los Angeles: University of California Press. [In spite of 
repeated attempts 1 have not been able to secure a copy of this book. My ci- 
tations are based on references in other works.] 

. 1969. Comparative Dravidian studies. Current trends in linguistics, ed. by 

T. A. Sebeok, 5: 309-33. The Hague: Mouton. 

. 1991. The emergence of the syllable types of stems (C)VCC(V) and 

(C)VC(V) in Indo-Aryan and Dravidian: Conspiracy or convergence?. 
Studies in the historical phonology of Asian languages, ed. by W. G. Boltz 
and M. C. Shapiro, 160-175. Amsterdam & Philadelphia: Benjamins. 



Hock: Subversion or convergence? 113 

. 1995. Patterns of sound change in Dravidian. Symposium on Language and 

Prehistory in South Asia, University of Hawaii. March 1995. 
. In Press. The origin and evolution of primary derivative suffixes in 

Dravidian. Historical, Indo-European, and lexicographical studies for 

Ladislav Zgusta in honor of his 70th birthday, ed. by H. H. Hock. Berlin: 

Mouton do Gruyter. 
KUIPER, F. B. J. 1962. Nahali: A comparative study. (Mededelingen der 

Koninklijke Nederlandse Akademie van Wetenschappen, Afd. Lelterkunde, 

N.R., 25: 5.) Amsterdam: Noord-Hollandsche Uitgevers Maatschappij. 
. 1966. The sources of Nahali vocabulary. Studies in comparative 

Austroasiatic linguistics, ed. by N. H. Zide. The Hague: Mouton. 
. 1967a. The genesis of a linguistic area. Indo-lranian Journal 10: 81-102. 

(Repr. in International Journal of Dravidian Linguistics 3: 135-153 

(1974).) 

. 1967b. The Sanskrit nom. sing. vit. Indo-lranian Journal 10: 103-125. 

. 1991. Aryans in the Rigveda. (Leiden Studies in Indo-European, 1.) 

Amsterdam and Atlanta, GA: Rodopi. 
. 1992. Rigvedic loanwords. International Journal of Dravidian Linguistics 

21: 1-49. 
KULKE, Hermann, and Dietmar Rothermund. 1986. A history of India. Repr. 

1990, New York: Dorsett. 
LEVITT. Stephan Hillycr. 1989. The alternation of r and / in Dravidian. Indian 

Linguistics 50: 130-147. (Published in 1991.) 
MACDONNELL, Arthur Anthony, & Arthur Berriedale Keith. 1912. Vedic index 

of names and subjects, 2 vols. Repr. 1958, Delhi: Motilal Banarsidass. 
MANSION. Joseph. 1931. Esquisse d'une histoire de la langue sanscrite. Paris: 

Geuthner. 
MARLOW, Patrick. Forthcoming. Origin and development of the Modern Indo- 

Aryan quotative and complementizer: an areal approach. University of 

Illinois Ph.D. dissertation in Linguistics. 
MAYRHOFER. Manfred. 1986-. Etymologisches Worterbuch des 

Altindoarischen. Heidelberg: Winter. 
MILLER. D. Gary. 1968. r/-clusters in Avestan. Language 44: 274-283. 
MORGENSTIERNE. Georg. 1947. Metathesis of liquids in Dardic. (Det norske 

videnskapsakadcmi i Oslo, Skriftcr. 1947: 2.) (Repr. in Georg 

Morgenstiernc 1975b.) 

. 1975a. Die Stellung der Kafirsprachen. Morgenstiernc 1975b: 327-343. 

. 1975b. Irano-Dardica. Wiesbaden: Rcichert. 

OBERLIES. Thomas. 1994. Review of Kuiper 1991. Indo-lranian Journal 37: 

333-349. 
PANDHARIPANDE. Rajeshwari. 1982. Counteracting forces in language change: 

convergence vs. maintenance. Studies in the Linguistic Sciences 12: 2: 97- 

116. 
PARPOLA. Asko. 1988. The coming of the Aryans to Iran and India and the cul- 
tural and ethnic identity of the dasas. International Journal of Dravidian 

Linguistics 17: 2: 85-229. 



114 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

PAYNE, John. 1989. Pamir languages. Schmitt 1989: 417-444. 

POTT, August Friedrich. 1933, 1836. Etymologische Forschungen auf dem 

Gebiete der indogermanischen Sprachen, 1 and 2. Lemgo: Meyer. 
RAMANUJAN, A. K., & Colin Masica. 1969. Toward a phonological typology of 

the Indian linguistic area. Current trends in linguistics, ed. by T. A. 

Sebeok, 5: 543-577. The Hague: Mouton. ^ 

RAU, Wilhelm. 1957. Staat und Gesellschaft im alten Indien nach den Brahmana- ( 1 

Texten dargestellt. Wiesbaden: Harrassowitz. 
SCHMITT, Riidiger (ed.). 1989. Compendium linguarum iranicarum. Wiesbaden: 

Reichert. 
SKJ/ERV0, Prods O. 1989 (a). Modern East Iranian languages. Schmitt 1989: 

370-383. 

. 1989 (b). Pashto. Schmitt 1989: 384-416. 

SOUTHWORTH, Franklin C. 1974. Linguistic stratigraphy of North India. 

International Journal of Dravidian Linguistics 3: 201-223. 
STEBLIN-KAMENSKIJ, M. I. 1965. Om alveolarer og kakuminaler i norsk og 

svensk. Norsk Tidskrift for Sprogvidenskap 20: 18-27. 
STEEVER, Sanford B. 1988. The serial verb formation in the Dravidian lan- 
guages. (MLBD Series in Linguistics, IV.) Delhi: Motilal Banarsidass. 
SUBRAHMANYAM, P. S. 1983. Dravidian comparative phonology. 

Annamalainagar: Annamalai University. 
THOMASON. Sarah Grey, & Terrence Kaufman. 1988. Language contact, cre- 

olization, and genetic linguistics. Berkeley & Los Angeles: University of 

California Press. 
TIKKANEN, Bertil. 1987. The Sanskrit gerund: A synchronic, diachronic, and 

typological analysis. (Studia Orientalia, 62.) Helsinki: Finnish Oriental 

Society. 
. 1988. On Burushaski and other ancient substrata in northwestern South 

Asia. Studia Orientalia 64: 303-325. 
TURNER, R. L. 1921. Gujarati phonology. Journal of the Royal Asiatic Society. 

Repr. in Turner 1973 = 1985: 88-145. 
. 1924. Cerebralization in Sindhi. Journal of the Royal Asiatic Society. Repr. 

in Turner 1973 = 1985: 206-227. 
. 1926. Middle Indian -d- and -dd-. Festgabe Hermann Jacobi. Repr. in 

Turner 1973 = 1985: 239-250. 
. 1973 = 1985. Indo-Aryan linguistics: Collected papers, ed. by J. Brough. 

Repr. 1985, Delhi: Disha Publications. ^ 

VINE, Brent. 1987. Vedic kuldyati and "spontaneous cerebralization" in Sanskrit. A^ 

14th International Congress of Linguists, Berlin, 1987. (Published in the ^ 

Proceedings, ed. by W. Bahner, J. Schildt, and D. Viehweger, Berlin: 

Akademie- Verlag . ) 
WACKERNAGEL, Jakob. 1896. Altindische Grammatik, 1. Gottingen: 

Vandenhoeck & Ruprecht. Second edition, 1957, with a new introduction 

by L. Renou and Nachtrage by A. Debrunner, ibid. 



Hock: Subversion or convergence? 115 

WITZEL. Michael. 1995. The linguistic situation in 1000 B.C. Symposium on 
Language and Prehistory in South Asia, University of Hawaii, March 
1995. 

ZIMMER, Heinrich. 1879. Altindisches Leben. Berlin: Weidmann. 

ZVELEBIL, K. 1970. Comparative Dravidian phonology. The Hague: Mouton. 



yn, 
:r? 



^ 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

REQUEST FORMATION IN ECUADORIAN QUICHUA 

Joni Kay Hurley 
Clemson University 

This study analyzes request formation in Ecuadorian Quichua. fo- 
cusing specifically on verb forms, morphological softeners, and lex- 
ical softeners. Perceptions of the degree of politeness conveyed by 
the various grammatical forms are presented, followed by a discus- 
sion of the influence of Spanish on Quichua. The findings will then 
be reviewed within the framework of language contact. 

1. Introduction 

The study of Quichua' has traditionally focused on the writing of grammar 
books and dictionaries (such as those of Catta Quelen 1987, Cordero 1989, 
Grimm 1896, Guzman 1920, Leonardi 1966, Pan's 1961, Stark and Carpenter 
1973, and Vasquez 1990). Pragmatics, which studies language use to accomplish 
conversational goals such as requests, invitations, and offers, has apparently not 
been studied in Quichua. This paper is part of a larger research project to deter- 
mine request formation in the Spanish and Quichua spoken in the Otavalo area of 
Ecuador, and how request formation in each of these languages may have been al- 
tered as a result of the language contact situation. The present study analyzes re- 
quest formation in Quichua. the differing degrees of politeness conveyed by dif- 
ferent grammatical structures, and possible Spanish influence. 

2. Review of the literature: The study of requests 

The study of requests originated with Austin 1962 and Searle 1976, 1979 
and their Theory of Speech Acts. They were the first to relate grammatical form to 
the purpose of the utterance in the conversation. Originally they classified utter- 
ances into categories based upon the type of verb used. For example, / request that 
you come would be classified as a Directive in Searle's classification due to the 
presence of the verb request. The inherent defects in this classification system be- 
came apparent when Searle 1979 attempted to classify both direct and indirect 
speech acts. In direct speech acts there is a one-to-one correspondence between the 
syntactic form of a sentence and its illocutionary meaning. For example, if an ut- 
terance contains an imperative then it conveys a request. Indirect speech acts in- 
volve more than one possible interpretation. The syntactic form may convey one 
type of speech act, but the utterance is being used to execute a different speech act. 
Proper interpretation of the utterance is dependent upon background knowledge, 
the context in which the utterance is said, and the roles of the speaker and the 
hearer. The large number of utterances that Searle listed as forms of indirect speech 
acts caused him to label them as idiomatic since each conveys more than one 
meaning. 



1 1 8 Studies in tlie Linguistic Sciences 23:2 (Fall 1993) 

Language specific studies of request formation have been conducted on 
English (Searle 1976, Ervin-Tripp 1976, Wardhaugh 1985, Blum-Kulka, House, 
and Kasper 1989, Eraser and Nolen 1981), Portuguese (Koike 1986, 1989, 
Wherritt 1983), Athapaskan (Rushforth 1985), Tzeltal (Brown 1979), and 
Spanish (Blum-Kulka, House, and Kasper 1989, Eraser and Nolen 1981, 
Haverkate 1979, Wilson 1965). The results of these studies show that different 
grammatical formations result in requests that are perceived by speakers of these 
languages to be more, or less, polite. (The concept of politeness has been elaborated 
at length by Brown and Levinson 1978). Politeness in requests is determined by a 
combination of grammatical/lexical form and patterns of use in a given 
language/culture. 

Blum-Kulka, House, and Kasper 1989 have conducted an ambitious study of 
requests and apologies across five languages (Spanish, German, Hebrew, English, 
and Erench). They found that all the languages in question have direct and indi- 
rect request strategies, but that each language may form these strategies with dif- 
ferent grammatical forms. In addition, perceptions of politeness- and patterns of 
use for the different request strategies can vary from language to language. 

3. Study site 

The research for this study was conducted in Otavalo and neighboring small 
towns in Ecuador over a period of seven months from 1989 to 1992. Otavalo is lo- 
cated about two hours north of Quito in the Andes mountains in the province of 
Imbabura. Residents of Otavalo speak either Spanish or Quichua as their maternal 
language. Many native Spanish speakers have some knowledge of Quichua since it 
is heard daily in the marketplace, but they do not normally acquire the ability to 
speak it. All Indians speak Quichua as their native language and most are also 
bilingual in Spanish to some degree.-^ The only Indians who remain monolingual 
in Quichua are those who live in isolated areas. 

4. The Quichua language 

Quichua is a member of the Andean-Equatorial language family and may be 
closely related"* to the Aymara language which is spoken in areas of Peru and 
Bolivia (Escobar 1986). Quichua (Quechua) is spoken by some seven million peo- 
ple from Ecuador to northern Argentina (Eromkin and Rodman 1988). Quichua 
does not have a standardized orthography and has been declared an official lan- 
guage only in Peru. This language is largely connected with Indian culture and is 
therefore highly stigmatized in Andean countries. 

Quichua is an agglutinating language in which the accumulation of suffixes 
conveys grammatical relations that are expressed in Indo-European languages by 
syntactic means. The following example illustrates the use of suffixes in Quichua: 

(1) Raimicunapica 

Raimi -i- cuna + pi -t- ca 

holiday / plural / in, on / topic marker 

'On holidays' 



Hurley: Request formation in Ecuadorian Quichua 119 

It is sometimes difficult to determine the meaning of a Quichua suffix, especially 
when it serves a function unknown in Indo-European languages, such as the topic 
marker -ca and the validation suffix -mi. These suffixes require further study. 

5. Methodology and sample 

Both elicited data and naturally-occurring requests were tape recorded in the 
Otavalo area. A questionnaire to elicit role-play (based upon the model of Blum- 
Kulka, House, and Kasper 1989) was developed in Quichua with the help of a 
bilingual Indian. Various situations were devised in which requests would 
commonly occur, such as: 

(2) Shuc mamaca, imashinata chunga ishcai huatata charic churita maiian, 
cuyman sarata jihuata carachun? 

How does a mother ask her 12 year old son to give corn and grass to the 
guinea pigs? 

(3) Imashna shuc tiuca, quilcana cashpita, paipac mashita maiian? 

How does a man ask his friend for a pen? 

Seventeen such situations were developed, and were mixed with elicitations of 
thirteen other speech acts (greetings, offers, expressions of gratitude, complaints, 
etc.) so that the informants would not realize that requests were being solicited. 

Another bilingual Indian tape-recorded the interviews and answers to infor- 
mant profile questions (age, occupation, education, language ability, etc.) with 75 
Quichua-speaking males between the ages of 20 and 50.'' Each interview lasted 
from 20 to 45 minutes. 

Six and a half hours of natural conversations were recorded in a store in 
downtown Otavalo that sells ponchos, blouses, and blankets to the Indians. At 
least 56 different speakers were represented. 

6. Corpus 

A total of 1,873 requests were recorded in Quichua, of which 1,803 were 
produced in interviews and 69 occurred in natural conversations. The small 
number of requests in natural conversations is due to the extensive use of Spanish 
in commerce transactions (Hill and Hill 1986). 

7. Data analysis 

Native speakers of Quichua listened to the tapes and transcribed what was 
said. These data were then entered into LOTUS 123 spreadsheet computer pro- 
grams which contained columns dedicated to the interview number, the number 
of the question, and the utterance. All the data were coded for grammatical and 
lexical forms that would have a bearing on the politeness of the requests.'' This al- 
lowed the data to be sorted in a variety of ways to determine patterns. In a follow- 
up study, Quichua speakers ranked a series of requests (selected from the inter- 
views) from least to most polite.'' 



1 20 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

8. Findings 

This analysis of Quichua requests will be divided into verb forms (modal 
verbs, imperatives, and other verb forms) and morphological and lexical softeners 
(diminutives, politeness suffixes, courtesy expressions, interjections, and voca- 
tives). This will be followed by a discussion of the perception of politeness in 
Quichua, and possible Spanish influence on the Quichua language. 

8.1 Verb forms 

8.1.1 Modal verbs — carana 'to give' 

The only modal verb in this study is carana 'to give', which is used with the 
gerund to convey softened requests.^ This structure translates as do me the favor of 
or please and can only be used with transitive verbs (i.e. verbs that permit a direct 
object) that do not clearly indicate benefit to the speaker.^ As a result, carana as a 
softener is of low frequency, occurring in 11.2 percent (n=209) of the elicited 
data, and did not occur in the natural conversations due to the scarcity of transi- 
tive verbs. An example of this structure is: 

(4) Papaguta randishpa carahuay. 

Papa -t- gu -I- ta randi -i- shpa cara -i- hua -I- y 
potato / dim.'" / ace. / buy / -ing / give / me / imp. 
'Do me the favor of buying (me) some potatoes.' 

In a few cases (4 out of 209 occurrences) another verb meaning 'give', cuna, is 
used in the same way: 

(5) Ashtahuan, ribajashpa cuhuay. 
Ashtahuan ribaja -i- shpa cu + hua -i- y 
too much / reduce / -ing / give / me / imp. 

(That's) too much, do me the favor of reducing (the cost). 

Both these verbs are used in Ecuadorian Quichua with the meaning of 'to 
give' and are used as softeners with the gerund -shpa (-ing) (Albor 1973, Catta 
Quelen 1987, Toscano Mateus 1953). Studies of the Cuzco dialect of Quechua 
describe the verbal suffix -cu as conveying cordiality and personal interest when 
used with imperatives (Gutierrez 1990, Sola and Yupanqui 1970). However, the 
meaning added by -cu is not translated in their examples," such as: 

(6) Kapuliyta ranticuhuay. 

Kapuli -I- y -I- ta ranti + cu -i- hua + y 
cherry / my / ace. / buy / me / imp. 
Buy from me my cherries. 

As this example illustrates, cuhuay could easily have been separated from the verb 
ranticuhuay 'buy from me' and used as a separate verb, equivalent in form to 'give 
me', cuhuay. This usage has apparently been transferred to the other verb for 'to 
give', carana, in Ecuadorian Quichua. 



Hurley: Request formation in Ecuadorian Quichua 121 

8.1.2 Imperatives (Commands) 

8.1.2.1 Present imperatives 

Present imperatives, formed in Quichua by suffixing -y to the verb root as in 
shamuy 'come', were the most common means of conveying requests in Quichua. 
They occurred in 73.4 percent (n= 1,325) of the elicited requests and in 52.2 per- 
cent (n=36) of the naturally-occurring requests. These commands are used when 
the addressee is expected to carryout the request immediately ( Leonard] 1966). An 
example of such a command is: 

(7) Manachiy lapizguta escribingapac. 
mananchi -i- y lapiz -i- gu + ta escribi + ngapac 
lend / imp. / pencil / dim. / ace. / write / in order to 
Lend (me) a pencil in order to write. 

8.1.2.2 Future imperatives 

The future imperative (formed by suffixing -n^id to the verb root) was used 
in 10.6 percent (n=192) of the elicited data and in 29 percent (n=20) of the 
naturally-occurring data. The future imperative is used for commands that are to 
be executed at a time subsequent to right now (Leonardi 1966, Mugica no year). '2 
This can be clearly seen in the data in which two commands occur together, as in: 

(8) Shamuy, randipangui, caipi yapachishpa cusha. 
Shamu -i- y randi -i- pa -i- nguicai -i- pi yapachi + shpa 

come / imp. / buy / please / fut.imp. / this / in / to give one extra / -ing / 

cu + sha 

give / 1 will 

Come (pres. imp.), please buy (fut. imp.) here, I will give (them to you) 

giving (you) one extra. 

As this example illustrates, a present imperative command to come (shamuy) is fol- 
lowed by a command in the future imperative, randipangui 'please buy* convey- 
ing what is to be done after the addressee comes. 

The future imperative is also used to convey politeness (Carpenter 1982). 
This was supported by research conducted by the author of this paper in 1990, in 
which the average ranking of various request structures by forty-eight Quichua 
Indians indicated that requests in the future imperative are considered to be more 
polite than requests in the present imperative. 

8.1.3 Other verb forms 

The remaining 15.4 percent of the elicited requests and 18.2 percent of the 
naturally-occurring requests were formulated with six other grammatical strate- 
gies. The frequencies of these strategies are presented in Table 1. 



122 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

TABLE 1. FREQUENCIES OF OTHER VERBAL STRATEGIES 

Grammatical Frequency of Frequency of 

Structure Occurrence in Occurrence in 

Elicited Data Natural Data 

l.Questions'3 6.8% (n=122) 8.7% (n=6) 

2. Statements (not 'need') 2.3% (n=42) 5.8% (n=4) d 

3. -shun 'let's' 4.2% (n=76) 2.3% (n=2) ^ 

4. 'Need' statements 1.8% (n=33) 0% 

5. -chun (subjunctive)'^ .3% (n=5) 0% 

6. Softener only 0% 1.4% (n=l) 

An example of each of these request structures is provided below. 

a. Questions 

(9) Nachu chai puchata charingui? 

Na + chu chai pucha + ta chari + ngui 

neg. / quest. / that / yam / ace. / have / 2nd pers. 

Don't you have that yam (that I am wanting to borrow)? 

b. Statements other than need 

(10) Pero chai preciupacca na ushani. 

Pero chai preciu + pac + ca na usha + ni 
but / that / price / for / topic / neg. can / 1st pers. 
But for that price I can't (buy it, so reduce the price). 

c. 'Let's' ... 

(11) Jacu futbulta pucllashun. 
Jacu futbul + ta puclla + shun 
let's go / soccer / ace. / play / let's 
Let's go, let's play soccer. 

d. 'Need' statements 

(12) Por Diosmanda, sacota ahuangapac, trabajangapac munani. 

Por Dios + manda saco + ta ahua + ngapac trabaja + ngapac muna + ni 
Please / by / sweater / ace. / weave / to / work / to / need/want / 1st pers. 
Please, I need to work, (I need) to weave sweaters. 

e. Subjunctive 

(13) Ricungui, cunan charini shuc carruguta, cunan como can yachangui 
manejanaca, munani que can trabajachun iiuca carrupi.'^ 
Ricu + ngui cunan chari + ni shuc carru + gu + ta cunan como can 
look / fut.imp. / now / have / 1st pers. / a / car / dim. / ace. / now / since / you/ 
yacha + ngui maneja + na + ca muna + ni que can trabaja + chun 

know / 2nd pers. / drive / inf. / topic / want / 1st pers. / that / you / work / subj./ 

nuca carru + pi 

my / car / in 

Look, I have a car now, and since you know how to drive, I want you to 

work (for me) in my car. 



Hurley: Request formation in Ecuadorian Quichua 123 

f. Softener only 

(14) Por Dios. cumari. 
Por Dios, cumari 
please / godmother 

Please, godmother (sell it to me cheaper). 

8.2 Morphological softeners 

8.2.1 Diminutives 

There is much variation in the diininutives used in the various dialects of 
Ecuadorian Quichua. and they include -hua, -cu, and -lla (Jara [no year]). 
Diminutives were used in 61 percent (n= 1,101) of the elicited requests, and in 
24.6 percent (n=17) of the naturally-occurring requests, and were found primar- 
ily on direct objects (55.6 percent [n=612] of the diminutives) and on vocatives 
(14.2 percent [n=156] of the diminutives). Three diminutives were used in this 
sample of Otavalo Quichua: -cu (and its voiced variant -gu), -lla. and the Spanish 
diminutive -ito. 

The most common diminutive was -ciil-gii, representing 84.7 percent 
(n=932) of all diminutives in the elicited data, and 41.2 percent (n=7) of the 
diminutives in the naturally-occurring data. Examples include: 

(15) esferucuta 
esferu + cu + ta 
pen / dim. / ace. 
a pen (dim.) 

(16) tandagu 
tanda + gu 
bread / dim. 

a little bread 

There appears to be no difference in meaning or usage between -cu and -gu, and 
the voiced variant is not due to the phonological environment. Catta Quelen 1987 
reports the distribution to vary according to geographic region. In the present 
study, usage could be closely linked to town of residence. For example, all infor- 
mants from Peguche used only -^w, except on the word uiiia 'father, sir', which 
always contained the diminutive -cu. Informants residing in Otavalo used only 
-cu. The only informants who would use both diminutives had been born in one 
town and were now living in another. It is not clear why the variant -cu is the 
only one used on taita 'father, sir" regardless of the diminutive used in a particular 
informant's town of residence. 

The diminuti\c -lla was much less common, representing 10 percent 
(n=l 10) of the diminutives used in the elicited data, and 41.2 percent (n=7) of the 
diminutives used in the naturally-occurring data. The use of this diminutive is 
largely restricted to specific lexical items, especially the Spanish loanwords hurato 
'cheap', tio 'uncle, sir' and amo 'master'. An example is: 



124 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(17) baratulia 
baratu + 11a 
cheap / dim. 

a little cheap 

The two diminutives -cul-gu and -lla can be combined on the same word to 
strengthen the minimizing effect: 

(18) ratugulla 
ratu + gu + lla 
while / dim. / dim. 
a very little while 

Forty Quichua-speaking Indians ranked nouns with diminutives as more polite 
than nouns without diminutives, but there was little difference in perceived po- 
liteness between -cul-gu and -lla. 

The last diminutive found in this sample was the Spanish -ito, which repre- 
sented 5.4 percent (n=59) of the diminutives used in elicited requests, and 17.6 
percent (n=3) of the diminutives used in naturally-occurring requests. A careful 
analysis of the use of this diminutive reveals that it occurred only on Spanish 
loanwords, principally vocatives. For example: 

(19) amiguito 
amigo -I- ito 
dear friend 

8.2.2 The suffixes -pa, -lla, and -ya(ri) 

The suffix -pa is attached to verb forms to convey courtesy and respect on the 
part of the speaker toward the addressee (Quintero and Cotacachi 1986, Carpenter 
1982) and is often loosely translated as 'please' (Jara [no year], Catta Quelen 
1987). This suffix occurred in 25.4 percent (n=458) of the elicited requests, and 
in 1 1.6 percent (n=8) of the naturally-occurring requests. An example is: 

(20) Nucapac carrupi trabajangapac shamupay. 

Nuca -I- pac carru + pi trabaja + ngapac shamu -i- pa -t- y 
I / of / car / in / work / to / come / please / imp. 
Please come to work in my car. 

The suffixes -lla and ya(ri) are placed on imperatives to achieve opposite ef- 
fects. The suffix -lla is used to soften verbs and is usually translated as 'just' 
(Mugica [no year], Quintero and Cotacachi 1986, Carpenter 1982, Stark and 
Carpenter 1973). It was used in only .1 percent (n=2) of the elicited requests, and 
in 10.1 percent (n=7) of the naturally-occurring requests. An example of this 
suffix is: 

(21) Shamuylla.'6 
Shamu -t- y + lla 
come /imp. /just 
Just come (on over). 



f 



Hurley: Request torniation in Ecuadorian Quicluia I 25 

It is possible thai this is simply the diminutive suffix -lla placed on verb forms. 
Further support for this idea is found in the use of nomas 'just' in Andean 
Spanish, which serves as a softener for both nouns and imperatives (Naula Gaucho 
1975, Stratford 1989). It is claimed that this use ofnoimis in Spanish is due to the 
influence of the Quichua suffix -lla (Quintero and Colacachi 1986, Catta Quelen 
1987). In many instances there were strong parallels in the data gathered in this 

) study between the use of -lla in Quichua and nomas in Spanish, as is illustrated in 

■ the following examples: 

(22) Compadrito, caiman shamuylla. 
Compadre + ito cai + man shamu + y + lla 
godfather / dim. / this / to / come / imp. / just 
Godfather, just come (on) over here. 

(23) Ya, ya, bueno, bueno, vendra nomas. 

Okay, okay, fine, fine, just come (on) (fut. imp.). 

When -lla is used in commands there is a distinctive intonational contour that is 
not found when it is placed on nouns. The pitch rises suddenly to a higher level 
on the suffix -lla. and then drops off rapidly. 

The suffix -ya(ri). also translated as 'just' or 'come on' (Gutierrez 1990). is 
attached to commands to make them more emphatic (Centra de Investigaciones 
para la Educacion Indigena 1983). The longer form, -yari. is considered to be more 
emphatic than the shorter form -ya (Catta Quelen 1987, Quintero and Cotacachi 
1986). Gutierrez 1990 reports that -yari can also serve to emphasize pleading, as 
in come on in English or ya pues in Spanish. The suffix -yari was used in 1.5 per- 
cent (n=26) of the elicited requests, and in 10.1 percent (n=7) of the naturally- 
occurring requests. For example: 

(24) Cuatrupac cuhuayyari. 
Cuatru + pac cu + hua + y + yari 

four / for / give / me / imp. /just (emphatic) 

Come on, give (it) to me for four (thousand sucres). 

This suffix serves as the pattern for the use of pues 'just, come on" as a suffix in 
Andean Spanish, which emphasizes utterances (Quintero and Cotacachi 1986). As 
is true for the suffix -lla in Quichua, pues is placed after the word or words that 
the speaker wishes to emphasize: 

(25) Venderame'" a 50 pues. 

Sell (future imperative) (it) to me for just 50 (sucres). 

In this example, pues is placed after 50 conveying that what is being emphasi/ed 
is the reduced price of 50 sucres, compared to the asking price. 

8.3 Lexical softeners 

S.3.1 Courtesy expressions 

The sample gathered in this study contains six expressions that convey 
please' and all are either borrowed directly from Spanish, translated from 



Frequency of 


Frequency of 


Occurrence in 


Occurrence in 


Elicited Data 


Natural Data 


20.2% (n=364) 


1.4% (n=l) 


3.0% (n=55) 


0% 


.2% (n=4) 


0% 


.4% (n=7) 


0% 


.7% (n=12) 


0% 


.3% (n=6) 


0% 



126 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Spanish, or a combination of both. These expressions were used in 24.8 percent 
(n=448) of the elicited requests, and in only 1.4 percent (n=l) of the naturally oc- 
curring requests. The lexical items and their frequencies are presented in Table 2. 

TABLE 2. LEXICAL SOFTENERS AND THEIR FREQUENCIES 
Courtesy Expression 



Por dios(manda) 'By God' 

Ama shinagu cashpa 'Don't be that way' 

Favor 'Please' 

Por favor 'Please' 

Favorta rashpa 'Doing the favor' 

Favorta shinashpa 'Doing the favor' 

The first expression, por Dios(manda) 'by God', was used in Old Spanish 
and is presently the most frequent means of conveying 'please' in the Quichua spo- 
ken in Otavalo. In many cases the Quichua suffix for 'by' is added, forming por 
Diosmanda 'by God by'. Ama shina cashpa 'don't be that way','** is the translation 
of the Spanish politeness expression no sea(s) malito 'don't be bad' which has been 
documented in Ecuador, Bolivia, and Mexico. Both /a vor and por /avor 'please' 
have been borrowed directly into Quichua, and the last two expressions listed in 
Table 2 are loanblends, combining the Spanish word favor with the Quichua 
verbs (ru)rana and shinana meaning 'to do'. The resulting expressions are equiva- 
lent to the Spanish hdga(me) el favor de... 'do (me) the favor of...'. 

8.3.2 Interjections and vocatives 

Interjections were used in 6.4 percent (n=l 16) of the elicited requests, and in 

4.3 percent (n=3) of the naturally occurring requests. Only two interjections, ya/« 
'look' QXidjaica 'take', are purely Quichua. The others were either borrowed or 
translated from Spanish, such as: bueno 'well', oye 'listen', uyay (from the 
Spanish oye, 'hsten'), and ricuy 'look'. The most common interjection wasjala 
'look', representing 59.5 percent (n=69) of the interjections in the elicited data, 
and 66.6 percent (n=2) of the interjections in the naturally occurring data. 

The vocatives that occurred in the corpus were primarily terms of family re- 
lationship or friendship, and 75 percent were of Spanish origin. They were used 
in 26.2 percent (n=472) of the elicited requests, and in 20.3 percent (n=14) of the 
naturally-occurring requests. Vocatives in this sample included: compadre 
'godfather', comadre 'godmother', pana 'sister', guambra 'guy', taita 'father/sir', 
amigo 'friend', and senor 'sir'. 

8.4 Request strategies and perceptions of politeness 

Forty-eight Quichua-speaking Indians were asked to rank a series of requests 
to come in order of politeness (without considering factors of the situation). Their 
ranking is presented in Table 3. The resulting scale demonstrates that: the future 



Hurley: Request fomiation in Fxuadorian Oui<-"hua 1 27 

imperative is more polite than the present imperative, the future imperative is 
more polite than the sulTix -pa 'please', por Diosmanda "by God' is the strongest 
courtesy expression, and the most polite request formation consists of the future 
imperative, -pa, and por Diosmanda 'by God". Gutierrez 1990 reports that the first 
example, shamuy 'come', is a command, and that the most polite request, Por 
Diosmanda. shamupangui 'By God, please come [future imperative]', conveys 
almost pleading. 

TABLE 3. RANKING OF QUICHUA REQUEST FORMS IN ORDER 
OF POLITENESS (least polite to most polite) 

Shamuy. Come (pres. imp). 

Shamuylld. Just come (pres. imp.). 

Shamupay. Please come (pres. imp.). 

Shamungui. Come (fut. imp). 

Shamupangui. Please, come (fut. imp.). 

Por Diosmanda. shamuy. Please, come (pres. imp.). 

Por Diosmanda, shamupay. Please, please come (pres. imp.). 

Por Diosmanda, shamungui. Please, come (fut. imp.). 

Por Diosmanda, shamupangui. Please, please come (fut. imp.). 

A similar ranking of requests to buy potatoes demonstrates that carana 'to 
give' makes requests more polite and that por favor is the least polite expression 
for 'please", Wwh por Diosmanda being considered the most polite. This ranking is 
presented in Table 4. 

TABLE 4. POLITENESS RANKING INVOLVING THE VERB CARANA 
'TO GIVE' (least polite to most polite) 

Papaguta randihuay. 
Buy me potatoes (dim.). 

Papaguta randishpa caray. 

Do (me) the favor of buying (me) potatoes (dim.). 

Papaguta randishpa carahuay. 

Do me the favor of buying (me) potatoes (dim.). 

Por favor, carahuay randishpa papaguta. 

['lease, do me the favor of buying (me) potatoes (dim.). 

Ama shinagu cashpa, papaguta randishpa carahuay. 

Don't be that way, do me the favor of buying (me) potatoes (dim.). 

Por Diosmanda, papaguta randishpa carahuay. 

By God, do me the favor of buying (mc) potatoes (dim.). 

Finally, forty Quichua Indians ranked vocative terms with diminutives as more 
polite than vocatives without diminutives, but there was very little difference in 
politeness noted between -lla and -cu. 



128 Studies in tiie Linguistic Sciences 23:2 (Fall 1993) 

8.5 The influence of Spanish on Quichua 

In the elicited data the most obvious influence of Spanish on Quichua is the 
large percentage of loanwords, which represent from 7 to 49 percent of the vocab- 
ulary used, depending upon the informant. However, a careful analysis of the 
words themselves reveals that the 2,828 occurrences of Spanish words represent 
only 300 different words. The majority refer to items and concepts brought by the 
Spanish such as carro 'car', trabajar 'to work for money', llamada 'telephone call', f 
and bautizar 'to baptize'. Also frequently borrowed are connecting words and 
phrases such as pero 'but', y 'and', o sea que 'or rather', and entonces 'then' 
which often replace the Quichua suffixes with similar meanings. In addition, the 
only lexical items (as opposed to suffixes) that are used to soften requests in these 
data are apparently all of Spanish origin. 

Spanish influence is also seen in the borrowing of one, and possibly two, 
suffixes into Quichua. The Spanish suffix -dor 'the person who' is occasionally 
used with this meaning on Quichua words, replacing the Quichua equivalent -c. 
Examples include /?Mc//aJor 'ball player' and ahuador 'weaver'. It is also possible 
that the Quichua diminutive suffix -cu/-gu is from the Spanish diminutive -ico, 
although this has not been documented in any of the Quichua grammars. Support 
for this idea is found in Bolivian varieties of Quechua, which use the Spanish 
diminutive -ito as the primary diminutive, as in jamp'atitu 'little toad' (Urioste 
1955:21). In the Otavalo area, -ito is used only on Spanish loanwords. 

In naturally-occurring conversations the most obvious influence of Spanish 
on Quichua is language mixing: either code-switching between Quichua and 
Spanish, or media lengua 'middle language' - Quichua syntax with approximately 
90 percent Spanish vocabulary (Muysken 1981). Media lengua has been reported 
by Muysken 1981 in the southern dialects of Ecuadorian Quichua. This mixed 
language is described as a combination of the Quichua grammatical system with 
the majority of the lexicon of Spanish origin. Media lengua occurs to a limited ex- 
tent in the Otavalo area. In the example provided below, the vocabulary of Spanish 
origin is written in capital letters, and Quichua words and suffixes are written in 
lower case letters: 

(28) CUCINA URA ISQUINAcupi; VINTANAcuna, SILLAcuna, tianmi. 
CUCINA URA ISQUINA + cu-i-pi; VINTANA + cuna 
kitchen / now / comer / dim. / in / window / pi. 
SILLA + cuna tia + n + mi 
chair / pi. / exist / 3rd pers. / validator 

The kitchen now (is) in the corner (diminutive); there are windows and | 
chairs. 

In this example, all the vocabulary except tian 'there are' is of Spanish origin. 
Word order and grammatical relations are completely Quichua. 

9 Conclusions 

The grammatical strategies used in the formulation of requests in the 
Quichua data recorded for this study are summarized in Table 5. The elicited data 
is characterized by more extensive use of softeners than is the case in the naturally- 



Hurley: Request formation in Ecuadorian Quichua 



129 



occurring conversations. This is most likely due to the fact that the interview sit- 
uation is more formal and there is no true relationship between the people in the 
hypothetical situations. In addition, the natural conversations dealt with com- 
merce, in which the banter between the customer and the vendor is relatively di- 
rect and to the point. In both types of data the primary verb form is the imperative, 
both present (-v) and future i-ngui). Elicited speech contained many Spanish 
loanwords (ranging from 7 to 50 percent of the vocabulary used by each individ- 
ual), whereas the naturally-occurring conversations were characterized not only 
by many loanwords but by code-switches to Spanish as well. 



TABLE 5. SUMMARY OF QUICHUA REQUEST STRATEGIE 


Strategy 


Frequency of 


Frequency of 




Occurrence in 


Occurrence in 




Elicited Data 


Natural Conversations 


Modal verbs: 






Carana "to give' 


11.2% (n=209) 


0% 


Verb forms: 






Present imperative 


73.4% (n=l,325) 


52.2% (n=36) 


Future imperative 


10.6% (n=192) 


29.0% (n=20) 


Questions 


6.8% (n=122) 


8.7% (n=6) 


-shun 'let's' 


4.2% (n=76) 


2.3% (n=2) 


Statements (not 'need') 


2.3% (n=42) 


5.8% (n=4) 


'Need' statements 


1.8% (n=33) 


0% 


Softener only 


0% 


1.4% (n=l) 


-chim (subjunctive) 


.3% (n=5) 


0% 


Morphological softeners: 




-Cll/-gU 


51.7% (n=932) 


10.1% (n=7) 


-lla (added to nouns) 


6.1% (n=110) 


10.1% (n=7) 


-ito (from Spanish) 


3.3% (n=59) 


4.3%(n=3) 


-pa (politive) 


25.4% (n=458) 


11.6% (n=8) 


-lla (added to verbs) 


.l%(n=2) 


10.1% (n=7) 


Lexical softeners: 







Courtesy Expressions: 

Por Diosmanda '"Qy God' 20.2% (n=364) 
Ama shinagu cashpa 



.4%(n=l) 



'Don't be that way' 
Favor 'please' 
Por favor 'please' 
Favorta rashpa 
'Doing the favor' 
Favorta shinashpa 
'Doing the favor' 

Interjections 
Vocatives 



3.0% (n=55) 
.2% (n=4) 
.4% (n=7) 

.7%(n=12) 

.3% (n=6) 

6.4% (n= 11 6) 
26.2% (n=472) 



4.3% (n=3) 
20.3% (n=14) 



130 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

When the data collected in Quichua is compared to that collected for Spanish 
in the Otavalo area (Hurley 1992), important observations can be made concerning 
language contact and request formation. Both data sets support the idea that indi- 
rect request strategies in Quichua and Spanish have been greatly reduced (as com- 
pared to other varieties of these languages) and replaced with a higher frequency of 
direct strategies that translate easily from one language to another, specifically 
present and future imperatives softened with lexical expressions and diminutives.^ 
This shared pragmatic system can be clearly seen in Table 6. 

Table 6. REQUEST STRATEGIES IN THE SPANISH AND QUICHUA 
OF OTAVALO, ECUADOR 

Grammatical Form Frequency in Spanish Frequency in Quichua 

^Puede..? 'Can you..r 2.2% (n=63) 0% 

Present imperative 61.9% (n=l, 807) 72.7% (n=l,361) 

Future Imperative 7.7% (n=226) 11.3% (n=212) 

'Give' as a softener 10.2% (n=298) 11.2% (n=209 

Diminutives 36.8% (n=l,133) 59.7% (n=l,118) 

Lexical softeners 23.2% (n=714) 24.0% (n=449) 

Interjections 20. 1 % (n=6 1 8) 6.4% (n= 1 1 9) 

Vocatives 28.4% (n=875) 25.9% (n=486) 

To reach this point of shared pragmatics, the following changes have appar- 
ently occurred in Otavalo Quichua: a decreased reliance on the use of politeness 
suffixes and the borrowing of Spanish lexical courtesy expressions, word order 
changes so that softening suffixes in Quichua (such as -lla) occupy the same syn- 
tactic slot as their Spanish equivalents, the adoption of at least one (-ito) and possi- 
bly all three Spanish diminutives {-lla [from the Spanish -illol] and -cu/-gii [from 
the Spanish -icol]), and a preference for direct verbal request strategies (i.e. imper- 
atives). Changes in Otavalo Spanish (as compared to the findings of Blum-Kulka 
1989) include: the virtual abandonment of poder 'to be able' as a request softener, 
an increased use of imperatives, the adoption of both a present and a future impera- 
tive, the use of the future imperative to signal compliance at a future time or in- 
creased politeness, the loan translation of the Quichua modal verb carana 'to give', 
and the use of nomas 'just' and piies (emphatic) to express shades of meaning that 
are conveyed in Quichua through suffixes. 

The solution to cross-cultural communication problems in a language con- 
tact situation lies in the development of a shared set of pragmatic strategies. In or- 
der to reach this point, strategies that are used by both linguistic groups are used |l 
with greater frequency (such as imperatives in Quichua and Spanish), and those ^ 
which are not shared are either borrowed (such as the borrowing of give as a 
modal verb into Spanish and of lexical courtesy expressions into Quichua) or dis- 
carded (such as the modal verb poder 'to be able' in Spanish). Since the shared 
pragmatic system used in the Quichua and Spanish spoken in Otavalo is based 
upon direct request strategies (present and future imperatives plus softeners) 
which can be used in all conversational situations, there is very little possibility 
of being misunderstood. 



Hurley: Request formation in Ecuadorian Quichua 131 

NOTES 

' Quichua is referred to as Quechua outside of Ecuador. 

2 There have been many studies on the concept of politeness, beginning with 
Goffman 1967 and Brown and Levinson 1978. They determined the basic princi- 
ples of politeness, which are considered universal, such as 'saving face'. However, 
each language and culture possesses a variety of linguistic forms that are consid- 
ered more, or less, polite by members of that culture. There may be some overlap 
between cultures in the grammatical structures and the relative degree of polite- 
ness they are perceived to convey. Many language-specific studies have been con- 
ducted on request formation and perceived politeness (such as Searle 1976, Fraser 
and Nolen 1981, Koike 1986, 1989 and Brown 1979). In order to determine 
which grammatical structures are considered more polite by speakers of a given 
language, they are commonly asked to rank a series of requests that vary in gram- 
matical/lexical choice from least to most polite (such as Fraser and Nolen 1981 and 
Koike 1986, 1989). While this method does not associate grammatical structure 
with actual patterns of use, it does shed light upon what linguistic features convey 
increased politeness. 

^ Bilingualism is regarded as a continuum, ranging from knowing a few 
words and phrases in the second language to being a fluent speaker of two 
languages. 

■^ There is disagreement among linguists as to whether or not 
Quichua/Quechua and Aymara are genetically related. 

5 Only males were used due to the limited amount of time spent in Ecuador 
and the desire to have as homogeneous a group as possible. 

^ The determination of what grammatical/lexical categories are important in 
request formation was based upon the coding manual developed by Blum-Kulka 
(1989:273). As proposed by Blum-Kulka, a request can contain the following 
components: the head act (the minimal requesting unit), alerters (vocatives, inter- 
jections), the directness of the request (grammatical moods such as the imperative, 
'want' statements, hints, etc.), syntactic downgraders (interrogative form, tense, 
and aspect), and lexical and phrasal downgraders (politeness expressions such as 
'please', hedges, cajolers, etc.). 

'' A set of eight index cards were presented to each informant. Each card con- 
tained the same request, but worded differently. They were asked to order them 
from least to most polite. A similar ranking process was used by Koike 
(1989:195) for Portuguese and by Fraser and Nolen (1981:106) for Spanish. 

^ This structure is not found in Southern Peruvian varieties of Quechua 
(Gutierrez 1990). 

'^ The use of ^ive as a modal verb has been translated into the Spanish spoken 
in the Ecuadorian Andes, and is used like its Quichua counterpart. An example 
would be: Dame ahriendo la ventana 'Do me the favor of opening the window'. 



132 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

'0 The abbreviations in this paper are the following: 

dim. diminutive 

ace. accusative case, marking the direct object 

imp. imperative, or command form 

pres. present 

fut. future 

neg. negative i 

quest. question 

pers. person, as in the form of the verb 

inf. infinitive ending 

subj. subjunctive mood 

pi. plural 

11 Gutierrez 1990 reports that the suffix -cu in Peruvian Quechua is used in- 
stead of the words allichu and ichii, which are commonly used to soften requests. 
She provides the following example: Allichu tantata ruwapuway? 'Would you 
please make bread for him/her?'. In this case, allichu is translated by 'would you 
please...?'. 

'2 The same tense usage was observed in the Spanish sample from the Otavalo 
area. An example is: Deja por ahora - huscards algiin rato 'Leave (it) (present im- 
perative) for now, look for (it) (future imperative) some other time)'. 

13 Questions are commonly used to convey requests, especially questions of 
ability and availability. In this study, questions are considered requests when 
they are generated in response to a stimulus using the requesting verb mahana 'to 
ask' in the elicited data. In naturally-occurring conversations, questions are con- 
sidered requests when they expect the hearer to comply by performing an action. 

1"^ This strategy was only used by one informant whose speech was highly 
influenced by Spanish. 

15 This example also shows the heavy syntactic influence of Spanish in this 
speaker's Quichua. The word order is completely Spanish (subject, verb, object) 
and this sentence could be translated into Spanish merely by substituting Spanish 
vocabulary (with the exception of the final preposition -pi 'in'). 

16 Gutierrez 1990 reports that in Southern Peruvian Quechua -lla is affixed 
to words before the tense/aspect marker -y, as in shamulldy 'Just come on'. This 
was also reported for the Quichua spoken in the Ecuadorian jungle (Catta Quelen 
1987). The positioning of -lla in word final position is apparently typical of the .» 
Quichua spoken in Imbabura (Stark and Carpenter 1973). It could be theorized iL 
that -lla was moved to word final position in this area to "match" the equivalent 
structure in Spanish: Venga nomas 'just come', Shamuylld 'Just come'. 

1"^ The future tense is used in the Spanish of Otavalo to formulate requests 
that are to be executed at a time subsequent to the present. This is the same tense 
usage as in Quichua. 

18 In Southern Peruvian Quechua the equivalent expression is ama hina 
kaychu (Gutierrez 1990). 



Hurley: Request formation in Ecuadorian Quichua 133 

REFERENCES 

ALBOR, Hugo. 1973. Da + gerundio, ^un quechuismo? Hispania 56.316-318. 

AUSTIN, John. 1962. How to do things with words. London: Oxford University 
Press. 

BLUM-KULKA, Shoshona, Juliane House, & Gabriele Kasper. 1989. Cross-cul- 
tural pragmatics: Requests and apologies. Norwood, N.J.: Ablex Publishing 
Corp. 

BROWN, Penelope. 1979. Language, interaction, and sex roles in a Mayan com- 
munity: A study of politeness and the position of women. University of Cal- 
ifornia Ph.D. dissertation. 

, & Stephen Levinson. 1978. Universals in language usage: Politeness phe- 
nomena. Questions and politeness, ed. by E. Goody, 56-288. Cambridge, 
Mass.: Cambridge University Press. 

CARPENTER, Lawrence. 1982. Ecuadorian Quichua: Descriptive sketch and 
variation. University of Florida Ph.D. dissertation. 

CATTA QUELEN, Javier. 1987. Gramatica del quichua ecuatoriano. Quito, 
Ecuador: Ediciones Abya-Yala. Centre de investigaciones para la educacion 
indigena. 

CENTRO DE INVESTIGACIONES PARA LA EDUCACION INDIGENA. 1983. 
Nucanchic llactapac shimi. Vol. I. Quito, Ecuador. 

CORDERO, Luis. 1989. Diccionario quichua. Quito, Ecuador: Corporacion 
Editora Nacional. 

ERVIN-TRIPP, Susan. 1976. Is Sybil there? The structure of some American 
English directives. Language in Society 5:1. 25-66. 

ESCOBAR, Alberto. 1986. Types and stages of bilingual behavior: A socioprag- 
matic analysis of Peruvian bilingual Spanish. State University of New York 
at Buffalo Ph.D. dissertation. 

ERASER, Bruce, and Nolen, William. 1981. The association of deference with 
Unguistic form. International Journal of the Sociology of Language 27.93- 
103. 

FROMKIN. Victoria. & Robert Rodman. 1988. An introduction to language. 
Fourth edition. Chicago: Holt, Rinehart and Winston. 

GOFFMAN, Irving. 1967. Interaction ritual: Essays on face-to-face behavior. 
Garden City, N.Y.: Anchor Books. 

GRIMM, Juan. 1896. La lengua quichua: dialecto de la Repiiblica del Ecuador. 
Quito, Ecuador: Proyecto de educacion intercultural. 

GUTIERREZ, Salome. 1990. Personal communication. 

GUZMAN, Manuel. 1920. Gramatica de la lengua quichua. Quito, Ecuador: 
Prensa catolica. 

HAVERKATE, Henk. 1979. Impositive sentences in Spanish. New York: North- 
Holland Publishing Co. 

HURLEY, Joni K. 1992. A cross-cultural pragmatic study of Spanish and 
Quichua request strategies as influenced by language contact in Otavalo, 
Ecuador. University of Pittsburgh Ph.D. dissertation. 

HILL, Jane, & Kenneth Hill. 1986. Speaking Mexicano: Dynamics of syncretic 



134 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

language in central Mexico. Tucson: The University of Arizona Press. 
JAKE, Janice. 1979. Validation suffixes in Imbabura Quechua. Papers from the 

fifteenth regional meeting of the Chicago Linguistic Society, ed. by P. 

Clyne, W. Hanks, & C. Hofbauer, 172-184. 
JARA, Fausto. No year. Morfologia quichua. Quito, Ecuador: Ediciones Mundo 

Andino. 
KOIKE, Dale. 1986. Differences and similarities in men's and women's directives ' 

in Carioca Brazilian Portuguese. Hispania 69:2.387-394. 
. 1989. Requests and the role of deixis in politeness. Journal of Pragmatics 

13.187-202. 
LEONARDL P. Jose. 1966. Lengua quichua (dialecto del Napo) gramatica y 

diccionario. Quito, Ecuador: Editora 'Fenix'. 
MUGICA, P. No year. Aprenda el quichua: Gramatica y vocabularios. Quito, 

Ecuador: CICAME. 
MUYSKEN, Pieter. 1981. Halfway between Quechua and Spanish: the case for 

relexification. Historicity and variation in Creole studies, ed. by A. 

Highfield & A. Valdman, 52-78. Ann Arbor: Karoma Publishers Inc. 
NAULA GAUCHO, Juan. 1975. Bosquejo gramatical del quichua de Chimborazo. 

Quito, Ecuador: Universidad Central. 
Nucanchic Uactapac shimi. Vol. 1.1983 Quito, Ecuador. 
ORR, Carolyn. 198L Algunos rasgos caracteristicos del discurso en el quechua 

del Napo. Revista latinoamericana de estudios lingiiisticos 1.135-175. 
PARIS, J. 1961. Gramatica de la lengua quichua. Quito, Ecuador: Editorial Santo 

Domingo. 
PARKER, Gary. 1964. Gramatica del quechua ayacuchano. Lima, Peru: 

Universidad Nacional Mayor de San Marcos. 
QUINTERO, Maria, & Maria Cotacachi. 1986. Quiquinllatac quichua shimita 

yachacupai. Vol. 4. Quito, Ecuador: Ediciones Abya-Yala. 
RUSHFORTH, Scott. 1985. Some directive illocutionary acts among the Bear 

Lake Athapaskans. Anthropological Linguistics 27:4.387-411. 
SALOMON, Frank. 1983. El quichua de los Andes ecuatoriales: algunos aportes 

recientes. Revista Andina 1:2.393-405. 
SEARLE, John. 1976. A classification of illocutionary acts. Language in Society 

5.1-23. 
. 1979. Expression and meaning: Studies in the theory of speech acts. 

Cambridge, Mass.: Cambridge University Press. 
SOLA, Donald, & D. Yupanqui. 1970. Quechua hablado. New York: Cornell 

University. 
STARK, Louisa, & Lawrence Carpenter. 1973. El quichua de Imbabura: Una 

gramatica pedagogica. Otavalo, Ecuador: Instituto Interandino de Desarrollo. 
STRATFORD, Billie. 1989. Structure and use of Altiplano Spanish. University of 

Florida Ph.D. dissertation. 
TOSCANO MATEUS, Humberto. 1953. El espaiiol en el Ecuador. Madrid, Spain. 
URIOSTE, Jorge. 1955. Gramatica de la lengua quechua y vocabulario. La Paz, 

Bolivia: Editorial Canata. 



Hurley: Request formation in Ecuadorian Quichua 135 

VASQUEZ, P. Victor. 1990. Runa shimita yachacushun: Aprendamos quichua. 
Quito, Ecuador: Ediciones Abya-Yala. 

WARDHAUGH, Ronald. 1985. How communication works. New York: Basil 
Blackwell. 

WHERRITT, Irene. 1983. Directives in Brazilian Portuguese: Mother-child inter- 
action. Spanish and Portuguese in social context, ed. by J. Bergen & G. 
Bills, 105-118. Washington, D.C.: Georgetown University Press. 

WILSON, Robert. 1965. Polite ways to give orders. Hispania 48:1. 



i 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

ADVANCEMENT IN SOME ASIAN AND AFRICAN LANGUAGES* 

Nkonko M. Kamwangamalu 
University of Swaziland 

'Advancement', a Relational Grammar rule which promotes a 
nominal bearing a given grammatical relation in a clause to a higher re- 
lation in the same clause (Perlmutter 1983), has been one of the central 
themes in Relational Grammar (RG) for the past twenty years or so. In 
RG, examples of advancement include such traditional rules as dative 
movement, raising, and passive. This paper discusses advancement of 
accusative, dative, and locative nominals in passive constructions in 
some South Asian and African languages, with a focus on Hindi and 
Ciluba. The paper is especially concerned with the claim in RG that 
'the relational network of every passive clause in any human language 
has a nominal bearing the 2-relation and the 1 -relation in successive 
strata (Perlmutter & Postal 1983:17). The data presented not only 
challenges this claim, but also has far-reaching implications for the 
relational laws resulting therefrom, viz. the Agreement Law, the 
Chomeur Law, and the Stratal Uniqueness Law. The implications of 
the data for relational concepts such as 'Terms' will also be discussed. 
It will be suggested that RG modify its claim, laws, and concepts to 
accommodate the data presented here and elsewhere in the literature on 
South Asian (e.g. Y. Kachru et al. 1976, Pandharipande 1981, Hock 
1982, Mohanan 1990) and African (e.g. Dalgish 1976) languages. 

1. Introduction 

The subject of this paper is 'advancement', a Relational Grammar (RG) rule 
that promotes a nominal bearing a given grammatical relation in a clause to a 
higher relation in the same clause (Perlmutter 1983). The paper aims to discuss 
advancement of accusative, dative and locative nominals in passive constructions 
in some Asian and African languages, and in Hindi and Ciluba in particular. More 
specifically, the paper addresses the claim in RG that the 'relational network of 
every passive clause in any human language has a nominal bearing the 2-relation 
and the 1-relation in successive strata (Perlmutter & Postal 1983:17). Before I dis- 
cuss this claim and the conditions or laws resulting therefrom, I shall, by way of 
background, first present a brief introduction to RG theory. Subsequently, I shall 
discuss accusative advancement in Hindi, and accusative, dative, and locative ad- 
vancement in Ciluba, with a focus on how these nominals achieve subjecthood in 
passive constructions. This will be followed by a discussion of the implications of 
the Hindi and Ciluba data for the RG's claim under consideration. It is worth not- 
ing here that the discussion of advancement in Hindi will draw heavily from pre- 
vious works in which this topic has received extensive coverage (e.g. Y. Kachru 



138 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

1980, Pandharipande 1981, Hock 1982, Mohanan 1990). This discussion will be 
limited to accusative nominals only because these are the only ones that can ad- 
vance to subject in passive constructions in Hindi and related languages (e.g. Y. 
Kachru et al. 1976, Hock 1982). 

2.0 Relational Grammar 

2.1 Background 

Central to RG theory is the notion of grammatical relations in a clause. RG 
views a clause as consisting of a network of grammatical relations such as 
SUBJECT (SU), DIRECT OBJECT (DO), INDIRECT OBJECT (lO), LOCATIVE 
(Loc), INSTRUMENTAL (Ins), BENEFACTIVE (Ben), etc. These are referred to as 
primitives of syntactic theory. The primitives are divided into two main cate- 
gories: central relations and oblique relations. Central relations include SU, DO, 
and 10, known as TERMS or as the 1 -relation, the 2-relation, and the 3-relation, re- 
spectively. Oblique relations include the remaining relations, viz. Loc, Ins, Ben, 
etc. These are known as NON-TERMS. 

Also central to RG is the notion of linguistic levels. RG argues that multiple 
syntactic levels must be recognized in the analysis of clause structure. This is be- 
cause in a clause a nominal may bear a range of relations to the predicate at differ- 
ent syntactic levels and also because certain syntactic phenomena are sensitive to 
some grammatical relations but not to others. For example, in (la) below the term 
banana bears the 2-relation to the predicate, while in (lb) it bears the 1 -relation. 
Similarly, in (2a) the term Paul, for instance, bears the 3-relation to the predicate, 
whereas in (2b) and (2c), it bears the 2-relation and the 1 -relation, respectively. 
Related to the question of Unguistic levels is the distinction in RG between initial 
and final grammatical relations. For instance, in (la) the term child is an initial 1, 
while the term banana is an initial 2. In (lb), however, the term child bears the 
chomeur relation to the predicate, while the term banana bears the final 1 -relation. 

(1) a. The child ate the banana. 

b. The banana was eaten by the child. 

(2) a. John gave food to Paul 

b . John gave Paul food 

c. Paul was given food by John 

Similar examples can be drawn from Asian languages, e.g. Malayalam and 
Hindi, or from African languages, e.g. Ciluba and Lingala, as shown in (3)-(4). In 
(3a) the highlighted terms each bear the 2-relation to the predicate, while in (3b) 
they bear the 1 -relation. The data in (4a) shows that the term Paul bears the 3-rela- 
tion in the initial stratum, while where applicable in the final stratum in (4b) it 
bears the 1 -relation. 

(3) a. 'The child ate the banana.' 

(M=Malayalam, H= Hindi, C=Ciluba, L= Lingala) 

M: kutti param tunnu 

child-N / banana-N / eat-PT 



vj 



Kamwangamalu: Advancement in some Asian and African languages 139 

H: bacce-ne kelaa k^aayaa 

child-Erg / banana-N / eat-Perf 

C: mu-ana u-aku-di-a ci-bota 

pf-child / Ag-PT-eat-FV / pf-banana 

L: mu-ana a-li-aki e-tabi 

pf-child / Ag-eat-PT / pf-banana 

b. 'The banana was eaten by the child.' 

M: kuttiyaal param tinnappetu. 

child-Ins / banana-N / eat-PSV-PT 

H: bacce-dvaaraa kelaa k^aayaa gayaa 

child-through / banana-N / eat-Perf go-Perf 

C: ci-bota ci-aku-di-ibwa kudi mu-ana 

pf-banana / Ag-PT-eat-PSV / by / pf-child 

L: e-tabi e-li-am-aki na mu-ana 

pf-banana / Ag-eat-PSV-PT / by / pf-child 

(4) a. 'John gave food to Paul (gave Paul food).' 

M: John Paul b^aksanam kotu»u. 
John-N / Paul-D / food / give-PT 

H: John Paul k^aanaa diyaa 

John-E / Paul-D / food-N / give-Perf 

C: Jean u-aku-pa *ci-akudia Paul (Paul ci-akudia) 
John / Ag-PT-give / pf -food / Paul (Paul / pf-food) 

L: Jean a-pes-aki *bi-lei Paul (Paul bi-lei) 

John / Ag-give-PT / pf-food / Paul (Paul / pf-food) 

b. 'Paul was given food by John' 

M: X (no equivalent) 

H: Paul-ko John-dvaaraa k^aanaa diyaa gayaa 

Paul-D / John-through / food-N / give-Perf / go-Perf 

C: Paul u-aku-p-ibw-a ci-akudia kudi Jean 

Paul / Ag-PT-give-PS-FV / pf-food / by / John 

L: Paul a-pes-am-aki bi-lei na Jean 

Paul / Ag-give-PSV-PT / pf-food / by / John 

Considering data such as (l)-(4), the question is how does RG explain the fact that 
the term Paul, for instance, which is an initial 3 in (2a), turns out to be a final 1 in 
(2c). This is where the notion of 'advancement' comes into the picture, a point to 
which I turn below. 



140 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

2.2 Advancement and passive in RG 

In view of data such as (l)-(4), and in particular the English data in (1 ) and 
(2), Perlmutter (1983:17) makes the claim given in (5) about advancement in pas- 
sive clauses not just in English but in human languages in general: 

(5) the RN of every passive clause in any human language has a nominal 
bearing the 2-relation and the 1 -relation in successive strata 

In line with this claim Perlmutter 1983 defines passive as: 

(6) a rule which sanctions 1-hood in an immediately successive stratum for 
a nominal which is a 2 of a clause at a stratum in which some nominal is a 1 . 

In other words, what both (5) and (6) mean is that in passive constructions noth- 
ing can be a final 1 (i.e. subject) which was not a 2 (i.e. direct object) in a preced- 
ing stratum. Advancement in passive constructions, as defined above, is governed 
by certain restrictions or laws in RG terminology, including the following 
(Perlmutter & Postal 1983:88-101, Frantz 1981:71): 

(7) a. The AGREEMENT LAW: 

'only Terms can trigger verb agreement. That is, only a nominal bear- 
ing Term relation in some stratum may trigger verb agreement.' 

b. The CHOMEUR LAW: 

If some nominal N^, bears a given Term relation 'n' in a given stratum 
Sj, and some other nominal N5 bears the same Term relation in the fol- 
lowing stratum, Sj+i, then N^ bears the chomeur relation (n) in S^+]. 

c . The STRATAL UNIQUENESS LAW: 

Each Term bears one and only one grammatical relation to the predi- 
cate. 

It is the above claims and laws that I shall be concerned with in this paper. It 
should be pointed out that these claims and laws have been challenged in recent 
literature on the syntax of Hindi and related languages (e.g. Pandharipande 1981). 
My concern here is to determine to what extent the claims and laws are applicable 
in passive constructions involving accusative, dative and locative nominals in 
Ciluba. First, I shall argue that contrary to the RG view of passive, in Ciluba da- 
tive and locative nominals may passivize directly from their initial grammatical 
relation as 3 or loc to the 1 -relation, and that attempting to advance these nominals 
to 1 via 3>2 or loc>3>2 would yield ungrammatical sentences. Second, I shall 
show that in Ciluba the distinction between terms and non-terms does not hold 
since, contrary to the Agreement Law, non-terms do also trigger verb agreement 
in this language. Third, I shall show that the facts of Ciluba receive support from 
previous works on languages as distant as Asian languages, such as Hindi. For in- 
stance, there is evidence from the literature on Hindi syntax that shows that 
contrary to one of the RG laws referred to earlier, viz. the Stratal Uniqueness Law, 
in Hindi a term may simultaneously bear two grammatical relations to the predi- 
cate, the subject relation on the one hand, and the direct object relation on the other 
(e.g. Y. Kachru 1980, Pandharipande 1981). 



Kamwangamalu: Advancement in some Asian and African languages 141 

3.0 Hindi and Ciluba 

3.1 Background 

Hindi is an Indo-Aryan language spoken on the Indian subcontinent. Ciluba 
is a Bantu language spoken in the Republic of Zaire. Both languages differ in 
many important respects. Here I shall highlight some of the features that are rele- 
vant to this paper. In terms of word order, Hindi is an SOV language, while 
Ciluba is an SVO language. In Hindi direct daughters of S can scramble freely, 
but this is not allowed in Ciluba, the latter being a strict word order language. 
Hindi has a case-marking system whereby in a clause the syntactic function of a 
given nominal is signaled. In the clause Ninaa-ne bacce-ko kitaab dii 'Nina gave 
the child a book', the nominal Ninaa carries the ergative case while the nominal 
bacce carries the dative case, as signaled by the clitics -ne and -ko, respectively. In 
Hindi syntax, a nominal that does not bear a clitic, such as kitaab 'book' is con- 
ventionally assumed to bear a nominative case (Y. Kachru et al. 1976, 1977; Y. 
Kachru 1980; Pandharipande 1981, 1990; Mohanan 1990). In terms of agree- 
ment, in Hindi a verb agrees in number, gender, person with its subject if it is 
nominative. And if the subject is not nominative, the verb agrees with the object if 
that is nominative (Mohanan 1990:14). 

In Ciluba, as in most Bantu languages (e.g. Bresnan & Kanerva 1989), a fi- 
nite verb must agree with its subject noun in person, number and noun class by 
means of an agreement prefix. To ensure subject-verb agreement, each Bantu lan- 
guage, and Ciluba is not an exception, has a noun class system whereby each noun 
consists of two basic morphemes, a noun prefix and a noun stem. In the noun ba- 
ana 'children', for instance, ba- is the noun prefix, and -ana the noun stem. The 
noun prefix provides a clue to determining the type of agreement that must obtain 
between a subject noun and a verb (Kamwangamalu 1985:110). For instance, in 
the clause ha-ana ba-di ba-dila 'the children are crying' the prefix ba- in ba-ana en- 
sures that whatever verb comes after the noun ba-ana 'children' must bear this 
same prefix for agreement, as evidenced by the presence of the prefix ba- in ba-di 
'are' and ba-dila 'crying'. 

Hindi and Ciluba may be different from each other in many other important 
respects, but describing such differences is beyond the scope of this paper: ad- 
vancement of accusative nominals in Hindi, and accusative, dative and locative 
nominals in Ciluba. 

3.2 Accusative/dative/locative nominals and subjecthood in Hindi 

and Ciluba 

It is generally agreed that a nominal that bears an accusative case ranks 
higher in the subject accessibility hierarchy. The questions I would like to raise in 
this section concern mainly accessibility to subject of dative and locative nomi- 
nals. First, can dative and locative nominals advance to 1 -relation (i.e. subject) in 
passive constructions in Hindi and Ciluba and, if they can, how is this advance- 
ment process done? Is it the case that a dative/locative nominal that advances to 1 
does so in one step, that is from its initial grammatical relation as 3/loc to the sub- 



142 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

ject relation; or does it achieve subjecthood through intermediate stages, such as 
exemphfied in (2a-c) above? 

Let us first address the question of subjecthood of accusative/ dative/locative 
nominals in Hindi and Ciluba, digressing briefly on the concept of subject. 1 shall 
start with Hindi, drawing heavily on the works of Y. Kachru 1980, 1981, 1990, 
Pandharipande 1981, and Mohanan 1990. According to the works just cited, in 
Hindi there are two types of nominals that are considered canonical or unmarked 
subjects, viz. the ergative subjects and the nominative subjects. However, such 
nominals are not the only ones that can function as grammatical subject in a Hindi 
clause. Other nominals that behave like subject include those I am concerned with 
in this paper, viz. the accusative, dative and locative nominals. Determining the 
subjecthood of these nominals is not a straightforward affair in Hindi. To deter- 
mine the subjecthood of these or any other nominals most Hindi grammarians usu- 
ally appeal to syntactic phenomena such as case-marking, agreement, word order, 
pronominal coreference, passivization, gap control, reflexive binding, conj- 
unction reduction, etc. Here I shall refer to few of these phenomena, as discussed 
in recent works on Hindi syntax (Y. Kachru 1990, Mohanan 1990). In their 
works, Y. Kachru and Mohanan are of the opinion that in Hindi a nominal that is 
claimed to be a subject must behave like one that is, it must have the properties 
associated with subject in the language, including the following, among others: 

i) it must be able to control reflexivization 

ii) it must be able to control conjunction reduction 

iv) it must be able to control equi-NP deletion 

No universality is claimed for these conditions on subjecthood. That is, a nominal 
that meets these conditions and therefore qualifies for subjecthood in Hindi, for in- 
stance, may not necessarily qualify as subject in other South Asian languages and 
vice versa. It is not surprising, then, that in languages such as Maithili, for in- 
stance, dative nominals are treated as objects rather than subjects (e.g. Mishra 
1990). 

Unlike Maithili, there seems to be enough evidence from recent works on 
Hindi syntax that in Hindi accusative/dative/locative nominals also behave like 
subjects (e.g. Pandharipande 1981, Mohanan 1990). While accusative nominals 
may function as subject with any class of predicate, there are in Hindi certain 
classes of predicates which govern dative/locative subjects. For instance, Y. 
Kachru 1990 notes that predicates that denote a set of 'inherent properties' such as 
utsaah 'enthusiasm', dhairy 'patience', himmat 'courage', etc. require a locative 
subject, while those that denote perception (e.g. dikhaaii denaa 'to be visible'), 
liking (e.g. pasand aanaa 'to like'), knowledge (e.g. maaluum honaa 'to come to 
know'), etc. require a dative subject. In what follows 1 present data which show 
that accusative, dative and locative nominals do indeed have properties associated 
with subject, for they meet the above and other diagnostics for subjecthood in 
Hindi. 



diO 



Kamwangamalu: Advancement in some Asian and African languages 143 

3.2.1 Accusative/dative/locative subjects in Hindi 

3.2.1.1 The reflexive apnaa binding 

According to Kachru & Bhatia 1977 and Pandharipande 1981, in Hindi the 
reflexive apnaa can take as its antecedent a subject, grammatical or logical, but no 
other argument. In the literature this phenomenon is also known as reflexive 
binding. The data in (8)-(9) is illustrative. In both (8a) and (9a) the dative nomi- 
nals Rita and Vijay are the logical subjects in their respective structures and, 
therefore, they qualify as antecedent of the reflexive apnaa. (8b) and (9b) show that 
in contrast to the reflexive, a pronoun cannot be coreferent with the subject of its 
minimal clause (Mohanan 1990). The dative subjects in the (b) sentences in (8)-(9) 
therefore cannot be coreferential with the pronoun uske. 

(8) a. ritaa-ko apnaa ghar bahut yaad aa rahaa thaa 

Rita-Dat / self s / home / much / memory / coming / was 
'Rita; was missing self Sj home very much.' 
(Y. Kachru 1990:70) 
b . ritaa-ko ghar uska bahut yaad aa rahaa thaa 

Rita-Dat / home / pron / much / memory / coming was 
'Rita; was missing herj/*j home very much.' 

(9) a. vijay-ko kitaab apnee g^ar-me milii 

Vijay-Dat / book-N / self-Gen house-L / fmd-Perf 
'Vijay; found the book in self;/*; house.' 
(Mohanan 1990:197) 
b. vijay-ko kitaab uske g^ar-me milli 

Vijay-Dat / book-N / pron / home-L / fmd-Perf 
'Vijayj found the book at hisj/*; home.' 

Reflexive binding, as described above, holds not only for dative nominals 
but also for locative nominals, as can be seen in (10). (10a) shows that the locative 
nominal, niina-me, is the only eligible antecedent of the reflexive apnaa. In the 
Hindi grammarians' view, this suggests that either the logical subject, namely the 
locative niina-me is the subject, or that there is no subject at all in (10a). It is noted 
that the facts of pronominal coreference support the former alternative. Pronouns 
cannot be coreferent with the subject in their minimal finite clause. This is borne 
out in (10b), where it is shown that the pronoun uskii is not coreferent with the 
locative nominal Ninaa-me. This suggests that the latter is indeed the subject in 
both (10a) and (10b) (e.g. Mohanan 1990:235-36). (The list of abbreviations used 
in the data below is given in the footnotes section'). 

(10) a. niinaa-me apnii mausii-ke liye badii mamtaa h 

Nina-Loc / self-Gen / aunt-Gen / for / much / affection-N / be-pres 
'Ninaj has a lot of affection for self S| aunt.' 
b. niinaa-me uskii mausii-ke liye badii mamtaa hai 

Nina-Loc / pron-Gen / aunt-Gen / for / much / affection / be-pres 
'Ninaj has a lot of affection for her*j aunt.' 

The facts of reflexive binding presented in (8)-(10) obtain also in construe- 



144 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

tions with accusative nominals, as illustrated in (11). Note that (11a) is the active 
counterpart of the passive construction in (lib). In (11a) the ergative nominal 
John, the unmarked subject, is obviously the eligible antecedent of the reflexive 
apne, as required in Hindi. In (lib), however, the ergative nominal under consid- 
eration has been demoted from its initial grammatical relation of subject to the 
chomeur relation as a result of passive, thus leaving the initially accusative nomi- 
nal, Paul, as the binder of the reflexive apne. Since the latter can only have a sub- 
ject, logical or grammatical, as its antecedent, it is correct to assume that the nom- 
inal Paul is the grammatical subject, and it is, in the passive construction in 
(1 lb). It is worth pointing out here that in addition to (1 lb), there is an alternative 
passive to the construction in (11a). This alternative, which I shall discuss later, 
is given in (lie). This construction differs from (1 lb) in terms of case-marking: 
in (1 Ic) the nominal Paul is case-marked, while in (1 lb) it is not case-marked. 

(11) a. John-ne Paul-ko apne kamre me dekhaa 

John-Erg / Paul-Acc / self / room / in / saw-Perf 
'Johnj saw Paul; food in selfi/*j home.' 

b . Paul apne kamre me dekhaa gayaa 
Paul / self / room / in / seen / was-Perf 
'Paul; was seen (by Johnj) in selfj/(*j) home.' 

c . Paul-ko apne kamre me dekhaa gayaa 
Paul-Acc / self s / room / in / seen / was-Perf 
'Pauli was seen in self S; room.' 

3.2.1.2 Conjunction reduction^ 

The data in (12)-(13) shows that the dative subject behaves like a subject be- 
cause it controls conjunction reduction, as in (12), though it does not undergo 
this process, as can be seen in (13) (e.g. Y. Kachru 1990:63). 

(12) tasviir dekh kar use gussa aayaa 
picture / see / CP / him / Dat / anger came 
'Hci became angry _i/having seen the picture.' 

(13) * gussa aa kar us-ne sab ko bahut DaaTaa 

anger / come I CP I he-Erg / all / DO / much / scolded 
'HCi scolded everyone_*i having become angry.' 

3.2.1.3 Equi-NP deletion 

In Hindi, like subject the dative nominal both controls and undergoes equi, 
as shown in (14) and (15) (Y. Kachru 1990: 64). 

(14) larke-ko film dekhnaa pasand hai 
boy-Dat / film / viewing / liking / is 
'The boy likes to view films.' 

(15) larke-ne film pasand aane kii carcaa nahii kii 
boy-Erg / film / liking / coming of / mention / not / did 
'The boy did not mention (his) liking the film.' 



Kamwangamalu: Advancement in some Asian and African languages 145 

As can be seen from the data presented above (e.g. (8)-(17)), dative, locative 
and accusative nominals prove to function as subject in Hindi, a point that is 
demonstrated at length by Y. Kachru and Mohanan. Rather than pursue this point 
any further, I shall assume the correctness of the conclusions reached by Y. 
Kachru and Mohanan and others regarding the subjecthood of the above- 
mentioned nominals in Hindi and will, instead, focus on how these nominals 
achieve their status as subject in this language. But first, a word on the 
subjecthood of accusative, dative and locative nominals in Ciluba. 

3.2.2 Accusative/dative/locative subjects in Ciluba 

We have seen that in Hindi one needs a number of diagnostics to show that 
accusative/dative/locative nominals can behave like subjects. In Ciluba, however, 
the situation is much simpler. Compared to Hindi, in Ciluba it simply takes one 
test to determine the subjecthood of not just accusative/dative/locative nominals, 
but of any nominal that claims subjecthood in a Ciluba clause. The most common 
test is agreement: In Ciluba, as in related Bantu languages (e.g. Swahili, Lingala, 
Kikongo), the verb must agree in person, number and noun class with nothing 
else but the subject, as can be seen from (16)-(17). In (16) and (17a) the verb agrees 
with the nominative nominals mwana 'child' and bibota 'bananas', respectively, 
while in (17b) the verb agrees with the inverted locative pa-mesa 'on the table.' 

(16) mu-ana u-aku-di-a bi-bota 

pf/sg-child / Ag-PTs-eat-FV / pf.pl-bananas 
'The child ate the bananas.' 

(17) a. bi-bota bi-di pa-mesa 

pf.pl-banana / Ag-are. / Loc. on-table 
'The bananas are on the table.' 
b. pa-mesa pa-di bi-bota 

Loc. on-table / Ag-are / pf.pl-banana 
'Lit: On the table is (are) bananas.' 

Agreement, as shown in (16)-(17), obtains also in passive constructions with 
accusative, dative and locative nominals, as can be seen in (18)-(19). Note that in 
the active clause in (18a), the verb -pa 'give' agrees with the subject John by 
means of the (singular) agreement prefix u-. Note also that in (18) both the ini- 
tially accusative nominal, ci-akudia 'food', and the initially dative nominal, ba- 
ana 'children' each can be passivized, as shown in (18b) and (18c), respectively. 
In the passive construction in (18b), the verb agrees with the initially dative nom- 
inal ba-ana 'children', which in this case is the grammatical subject of the clause 
under consideration. Here agreement is done by means of the (plural) agreement 
prefix ba-. In (18c), the verb agrees with the initially accusative nominal ci- 
akudia 'food' by means of the agreement prefix ci-. In (19a) agreement is the same 
as in (18a). In (19b), which is the passive counterpart of (19a), the verb agrees 
with the locative nominal mu-cikuku 'in the kitchen' by means of the locative 
prefix mil-. 



1 46 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(18) a. Jean u-aku-p-a ba-ana ci-akudia 

John / Ag-PTs-give-FV / pf.pl-child / pf-food 
'John gave the children food/food to the children.' 

b . Ba-ana ba-aku-p-ibw-a ci-akudia kudi Jean 
pf.pl-child / Ag-PTs-give-PSV-FV / pf-food / by / John 

'The children were given food by John.' > 

c. Ci-akudia ci-aku-p-ibw-a ba-na kudi Jean V 
pf.sg-food /Ag-PTs-give-PSV-FV / children / by / John 

'Food was given to the children by John.' 

(19) a. Jean u-aku-p-a ba-ana ci-akudia mu-cikuku 

John / Ag-PTs-give-FV / pf.pl-child / pf-food / Loc.in-kitchen 
'John gave the children food in the kitchen.' 
b. mu-cikuku mu-aku-p-ibw-a ba-ana ci-akudia kudi Jean 

Loc.in-kitchen / Ag-PTs-give-PSV-FV / pf-child / pf-food / by / John 
Lit: 'In the house was given the children food by John.' 

In addition to the facts presented in (18)-(19), elsewhere I have shown that in 
Ciluba, accusative, dative, and locative nominals behave like subject-Terms not 
only in terms of their ability to govern agreement on the verb, but also in terms of 
other properties associated with Terms, such as the ability to passivize, to rela- 
tivize, to incorporate onto the verb, to cleft, and to topicalize (e.g. Kamwangamalu 
1985) 

Having shown that in both Hindi and Ciluba accusative/dative/locative nom- 
inals may also function as subjects, 1 shall now move on to the other concern of 
this paper, viz. how these nominals achieve their status as subject in passive con- 
structions in the languages under consideration. 

4. Accusative/dative/locative advancement to subject in Hindi/Ciluba 

It was observed earlier that in Hindi, dative and locative subjects are base- 
generated rather than derived through processes such as advancement. Therefore, 
they will not be included in the discussion of advancement that follows. As back- 
ground for this discussion, let us recall the claim in (5) regarding RG's concep- 
tion of the relational network of a passive clause. Again, RG claims that the rela- 
tional network of every passive clause in any human language has a nominal bear- 
ing the 2-relation and the 1 -relation in successive strata. Applying this claim 
about passive to Hindi and Ciluba, the following analyses can be envisaged for ac- 
cusative (in addition to dative/ locatives for Ciluba) advancement in passive con- ^ 
structions in these languages. | 

One analysis, which follows directly from and is consistent with the above- 
stated claim of RG, is that in Hindi and Ciluba, an accusative nominal behaves 
like a subject that has undergone 2 to 1 advancement. Following this analysis, 
Ciluba nominals such as locative/dative, for instance, cannot be promoted to sub- 
ject unless they have first undergone loc>3>2 / 3>2 advancement, respectively. 

The other analysis, one that I shall suggest in this paper, is that for Ciluba, 
locative/dative nominals do not have to undergo loc>3>2>l / 3>2>1 advance- 



Kamwangamalu: Advancement in some Asian and African languages 1 47 

ment, and that they undergo loc/3>l advancement instead. For Hindi, the Ht- 
erature (e.g. Pandharipande 1981, Mohanan 1990) suggests that an accusative 
nominal may behave like subject in a given construction without necessarily 
having undergone 2>1 advancement. This analysis conflicts with the claim of 
Perlmutter and others (e.g. Johnson 1981), but it is consistent with the data of 
Hindi and Ciluba presented thus far in this paper. For the sake of illustration, let 
us look again at the passive constructions given earlier in (1 1) for Hindi and in (4) 
for Ciluba, repeated here below as (20) and (21), respectively. 

Regarding Hindi, it is clear that in (20b) the accusative nominal Paul has 
advanced to 1 , as can be concluded from the absence of the accusative case on the 
nominal under consideration. In (20c), however, there is no evidence that ad- 
vancement has taken place. The presence of the accusative case on the nominal 
Paul attests to this conclusion. Of crucial importance regarding (20c) is that in 
this construction the accusative nominal Paul is the only eligible antecedent of the 
reflexive apnee. Recall that in Hindi apnee can have nothing else but a subject as 
its antecedent. It follows that the accusative nominal Paul, the only antecedent of 
apne, is the subject of the passive construction in (20c). In a sense, then, it can be 
concluded that in (20c) the subjecthood of the accusative-marked nominal Paul is 
not dependent on its promotion to 1 , and that promotion of this nominal to 1 is ac- 
tually optional. Hock (1985:66) draws similar conclusions regarding advance- 
ment in Sanskrit of non-terms and terms to direct object and subject, respectively. 
He notes (p. 66) that ... 'if a non-term, adverbial constituent shows case variation 
between, say, locative and accusative, promotion to direct object status is possible 
only if there is no other direct object ... and that even under these conditions, 
promotion of that accusative-marked NP to subject of the passive is only optional.' 
The point here is to show that the facts of Hindi presented in (20c) are not an iso- 
lated case, and that they obtain in other Southeast Asian languages as well, such as 
Sanskrit. While these facts accord well with Hindi syntax they, obviously, con- 
flict with the 2>1 analysis as well as with some of RG laws presented earlier in 
this paper, such as the Stratal Uniqueness Law. Again, by virtue of this law, each 
term bears one and only one grammatical relation to the predicate (Perlmutter 
1983:88). Now, reconsider the construction in (20c). As was pointed out above, 
in this construction the term Paul bears not one but two grammatical relations to 
the predicate: First, Paul is a direct object because of its case, it bears the ac- 
cusative case; second, Paul is the grammatical subject in the construction under 
consideration because it is the only eligible antecedent of the reflexive apne: in 
Hindi, only a nominal that is a subject can be the binder of the reflexive apne. 

(20) a. John-ne Paul-ko apne kamre me dekhaa 

John-Erg / Paul- Ace / self / room / in / saw-Perf 
'John, saw Paul, food in self;/*; home.' 
b . Paul apne kamre me dekhaa gayaa 

Paul / self / room / in / seen / was-Perf 
'Paul; was seen (by John;) in selfj/(*|) home.' 



148 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

c . Paul-ko apne kamre me dekhaa gayaa 

Paul- Ace / self s / room / in / seen / was-Perf 
'Paulj was seen in self Sj home.' 

For Ciluba, the data in (21)-(22) suggest that accusative nominals can un- 
dergo 2>1 advancement, much as they can in English and other languages. For 
dative nominals, however, the data show that when such nominals advance to 
subject, they do so in one leap only that is, from their initial relation as dative to 
subject relation, and not through 3>2>1 advancement. Any attempt to advance a 
dative nominal for instance to 2 first and then to 1 results in an ungrammatical 
structure, as can be concluded from (2 lb). 2 

(21) a. Jean u-aku-p-a ba-ana ci-akudia 

John / Ag-PTs-give-FV / pf pl-child / pf-food 
'John gave the children food/food to the children.' 

b . *Jean u-aku-p-a ci-akudia ba-ana 

Jean / Ag-PTs-give-FV / pf-food / pf pl-child 

c . Ba-ana ba-aku-p-ibw-a ci-akudia kudi Jean 
Pr.pl-child / Ag-PTs-give-PSV-FV / pf-food / by / John 
'The children were given food by John.' 

Similar conclusions obtain also for locative advancement in this language, as 
shown in (22). Here, note that the locative mu-cikuku 'in the kitchen' advances 
directly to 1, as in (22d), and that attempting to advance it to 1 via Loc>3>2>l 
advancement would yield unacceptable sentences, as evidenced by (22b,c). 

(22) a. Jean u-aku-p-a ba-ana ci-akudia mu-cikuku 

John / Ag-PTs-give / pf pl-child / pf-food / Loc.in-kitchen 
'John gave the children food in the kitchen.' 

b. *Jean u-aku-p-a ba-ana mu-cikuku ci-akudia 

John / Ag-PTs-give / pf pl-child / Loc.in-kitchen / pf-food 
'John gave the children in the kitchen food.' 

c. *Jean u-aku-p-a *mu-cikuku ba-ana ci-akudia 

John / Ag-PTs-give / Loc/in-kitchen pf/pl-child / pf-food 
*'John gave in the kitchen the children food.' 

d. mu-cikuku mu-aku-p-ibw-a ba-ana ci-akudia kudi Jean 
Loc.in-kitchen / Ag-PTs-give-PSV-FV / pf pl-child / pf-food / by / Jean 
'Lit: In the house was given the children food by John.' 

The fact that locatives in Ciluba can advance to 1 via lool rather than 
Loc>3>2>l advancement is not an isolated case. Dalgish 1976 makes a similar 
claim with respect to Olutsootsoo, a Bantu language of Kenya, and so does 
Kimenyi 1974 with respect to Kinyarwanda, a Bantu language of Rwanda. Both 
Dalgish and Kimenyi show respectively that in Olutsootsoo and Kinyarwanda ad- 
vancement to 1 is not limited to terms and that locatives can advance to 1 as well. 
That locatives can advance to 1 is not unique to Bantu languages, but it is also at- 
tested to in non-Bantu languages. For instance, quoting Bell 1974 on advance- 
ment in Cebwano, a language of the Philippines, Perlmutter and Postal (1984:90) 
acknowledge that Cebwano allows with great freedom advancement^ to 1 not only 



n 



Kamwangamalu: Advancement in some Asian and African languages 149 

of 2s and 3s but also of instrumentals, locatives, benefactives, temporals, etc. This 
freedom of advancement of both terms and non-terms to 1 is also evident in 
Ciluba, as can be seen from the data in (21) and (22). But what are the implica- 
tions of such advancement of non-terms to 1 for the relational distinction between 
terms and non-terms, and for relational laws such as the Agreement law. 

Consider, for instance, the Agreement Law. According to this law, which 
^ was stated earlier in (7a), only Terms can trigger subject-verb agreement in a 
clause. Now, consider agreement in (22d), above. This clause shows that the loca- 
tive nominal mu-nzubu 'in the house' agrees with the verb -pa 'give' by means of 
the locative agreement prefix mu-. The question that arises here is whether loca- 
tives should be treated as Terms. Based on the data presented here, I would like to 
suggest that the scope of termhood in RG be extended so as to include locatives in 
languages such as Ciluba, since locatives are shown to behave like subject and es- 
pecially so with respect to the Agreement Law. 

5. Conclusions 

In this paper I have been concerned with one of central claims in RG regard- 
ing passive, viz. the claim that the relational network of every passive clause in 
any human language has a nominal bearing the 2-relation and the 1 -relation in 
successive strata. While this claim receives support from languages such as 
English and other languages around the world, it fails to accommodate data from 
some Asian and African languages, and from Hindi and Ciluba in particular. 

The literature on Hindi provides evidence that in Hindi, there are cases 
where an accusative nominal can be the subject of a passive clause without neces- 
sarily having undergone 2>1 advancement. As a result, contrary to the Stratal 
Uniqueness Law, in Hindi it is possible that a term bear two grammatical relations 
to the predicate: the direct object relation on the one hand, and the subject relation 
on the other, as illustrated in (20c). 

Unlike Hindi, accusative advancement in Ciluba accords well with the 
above RG claim about the relational network of a passive clause. However, the 
challenge to this claim comes from dative/locative advancement. I have shown that 
in Ciluba passivization of dative/locative nominals is a one-step process, 3>1 
/lool advancement, and that these nominals do not need to undergo 3>2 
/loc>3>2 advancement prior to advancing to 1, as claimed in Relational Grammar. 
Since in Ciluba it is not just dative nominals that passivize directly from 3 to 1, 
and locatives behave the same way as well, the question is whether taking into ac- 
) count RG laws such as the Agreement Law passivizing locatives should be treated 
as Terms. In light of the available evidence I have suggested that they should: In 
Ciluba and related Bantu languages locative nominals behave like Terms not only 
in terms of their ability to govern subjccl-verb agreement but also, as 1 have shown 
elsewhere (e.g. Kamwangamalu 1985), in terms of their ability to do other things 
that Terms can do, including the ability to relativize, the ability to passivize, the 
ability to cleft, to list just a few. 



150 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

NOTES 

An earlier version of this paper was presented at the 13th South Asian 
Languages Analysis (SALA) Roundtable, University of Illinois at Urbana- 
Champaign, Illinois, U.S.A., 25-27 May 1991. I would like to thank the National 
University of Singapore, where I was then working as a Visiting Lecturer, for 
sponsoring my participation in the SALA Roundtable. Also, I would like to thank 
Yamuna Kachru and Tara Mohanan for providing me with the Hindi data which 
have made this paper possible. Finally, I would like to acknowledge with grati- 
tude the comments of Yamuna Kachru and Hans H. Hock on an earlier version of 
this paper. I alone am responsible for any remaining errors of interpretation or 
analysis of the data presented in this paper. 

' Below are the abbreviations used in this paper: 



Dat = dative; 


Erg 


= ergative 


Nom 


= nominative 


Loc = locative; 


Pf 


= prefix 


Perf 


= perfective 


PSV = passive 


Pl 


= plural 


sg 


= singular 


PTs = past tense 


Gen 


= genitive 


Ag 


= agreement 


FV = final vowel 


pres 


= present tense 





CP = conjunctive participle 

2 Besides, it should be pointed out that in Ciluba a dative (i.e. indirect) ob- 
ject has prominence over an accusative (i.e. direct) object. This explains why in 
ditransitive constructions a dative object must always be close to the verb, regard- 
less of whether the accusative object is animate or inanimate. For further details, 
see Kamwangamalu 1985. 

3 Perlmutter and Postal 1984:90 do not specify whether in Cebwano ad- 
vancement of instrumental, locatives, benefactives, temporals and other non- 
terms to subject is done via intermediate stages, such as Inst/Loc/Ben/Temp 
>3>2>1, or whether it is done in one leap, e.g. Inst/Loc/Ben/Temp >1, as is the 
case in Ciluba and related Bantu languages (e.g. Lingala, Swahili, Kikongo). 

REFERENCES 

BRESNAN, Joan, & J. M. Kanerva. 1989. Locative inversion in Chichewa: a case 

study of factorization in grammar. Linguistic Inquiry 20:1.1-50. 
DALGISH, Gerard M. 1976. Passivizing locatives in Olutsootsoo. Studies in the 

Linguistic Sciences 6:1.57-68. 
FRANZ, Donald G. 1981. Grammatical relations in universal grammar. 

Monograph, Summer Institute of Linguistics: University of Lethbridge. 
HOCK, Hans Henrich. 1982. The Sanskrit passive: Synchronic behavior and di- 

achronic development. South Asian Review 6.127-37. 
. 1985. Sanskrit Double-Object Constructions: Will the real object please 

stand up? Indian Journal of Linguistics 12.50-70. 
KACHRU, Yamuna. 1980. Aspects of Hindi grammar. New Delhi: Manohar 

Publications. 
. 1981. Transitivity and volitionality in Hindi-Urdu. Studies in the 



Kamwangamalu: Advancement in some Asian and African languages 151 

Linguistic Sciences 11.2.191-93. 

. 1990. Experiencer and other oblique subjects in Hindi. Experiencer 

subjects in South Asian languages, ed. by M. K. Verma and K. P. Mohanan, 
59-75. Stanford: The Center for the Study of Language and Information. 

, Braj B. Kachru, & Tej K. Bhatia. 1976. The notion 'subject': A note on 

Hindi-Urdu, Kashmiri and Panjabi. The notion of subject in South Asian 
Languages, ed. by M. K. Verma, 79-108. University of Wisconsin, 
Madison: South Asian Studies. 

, & Tej K. Bhatia. 1977. On reflexivization in Hindi-Urdu and its theoretical 

implications. Indian Linguistics 38:1. 21-38. 

KAMWANGAMALU, Nkonko. 1985. Passivization in Bantu languages: implica- 
tions for relational grammar. Studies in the Linguistic Sciences 15:1.109- 
133. 

KIMENYI, Alexandre. 1976. A relational grammar of Kinyarwanda. UCLA Ph.D. 
dissertation in Linguistics. 

MISHRA, Mithilesh K. 1990. Dative/experiencer subjects in Maithili. The notion 
of subject in South Asian Languages, ed. by M. K. Verma, 105-117. 
University of Wisconsin, Madison: South Asian Studies. 

MOHANAN, Tara W. 1990. Arguments in Hindi. Stanford University Ph.D. dis- 
sertation in Linguistics. 

PANDHARIPANDE, Rajeshwari. 1981. Syntax and semantics of the passive con- 
struction in selected South Asian languages. University of Illinois at 
Urbana-Champaign Ph.D. dissertation in Linguistics. 

PERLMUTTER, David. 1983 (ed.). Studies in relational grammar 1. Chicago: The 
University of Chicago Press. 

, & P. Postal. 1984. The 1 -Advancement Exclusiveness Law. Studies in rela- 
tional grammar 2, ed. by D. Perlmutter and Carol G. Rosen, 81-125. 
Chicago: The University of Chicago Press. 



<1 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

ANIMACY DISTINCTIONS IN AKAN GRAMMAR* 

Emmanuel Kweku Osam 
University of Oregon 

Animacy distinctions has never been considered to be one of the out- 
standing features of the grammar of Akan. However, based on the form 
and distribution of nominal prefixes in the language and the nature of 
the pronominal system, it is concluded that the notion of animacy dis- 
tinction is relevant to the grammar of Akan. 



0. Introduction 

The purpose of this paper is to show that the grammar of Akan is sensitive to 
animacy distinctions, and to some limited extent, we could even talk about the 
presence of animacy hierarchy in the language. Animacy distinctions in Akan ap- 
pear mainly in nominal affixes and the forms and behavior of pronouns. 

1. Nominal affixes 

One of the areas in which the distinction between animate and inanimate 
nouns is shown is in the nominal affixes in the language. The fact that Akan has a 
nominal prefix system has long been recognized (see for example Christaller 1875, 
Balmer and Grant 1929, Akrofi 1935, Welmers 1971, 1973, Essilfie 1977, 
Dolphyne 1988, Dolphyne and Dakubu 1988, Osam 1993 and 1994). In Osam 
1993 and 1994, in agreement with Welmers 1971 and 1973, the argument is 
made that the current noun prefixes are the historical remains of the old noun class 
system that must have existed in Proto-Akan. The prefixal system as it exists cur- 
rently in the language is illustrated in (1). 

(1) Class 1 



Class 2 



Class 3 



Class 4 



o-nua 


'sibling' 


o-nipa 


'person/human being' 


o-birempon 


'a great person' 


o-baa 


'woman' 


o-bodom 


'dog' 


o-ponko 


'horse' 


a-bofra 


•child' 


a-noma 


'bird' 


a-berewa 


'old woman' 


e-wi 


'thief 


i-dua 


'tree' 


i-kur 


'sore' 


i-dan 


'building' 


e-boo 


'stone' 


o-dan 


'building' 


e-woo 


'honey' 



154 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Class 5 



Class 6 



n-dua 


'trees' 


n-kura 


'mice' 


n-twer 


'frogs' 


n-dowa 


'bees' 


m-bofra 


'children' 


m-bowa 


'animals' 


m-fe 


'years' 


a-ka 


'debts' 


a-dan 


'buildings' 


a-hoho 


'visitors' 


e-kunyin 


'great men' 


e-din 


'names' 


e-kuw 


'clubs/associations' 



As indicated in (1), the prefixes are either vowels or homorganic nasals. 
Prefixes involving vowels, with the exception of the prefix labeled Class 4, are 
paired on the basis of vowel harmony (2). The first member of each pair has the 
advanced tongue root feature (+ATR), and the second member is minus the ad- 
vanced tongue root feature (-ATR). The nominal prefixes are also distinguished 
according to number. The prefixes marked Classes 1-4 indicate singular nouns 
whereas those marked Classes 5 and 6 identify plural nouns. 

(2) SINGULAR PLURAL 
Class 1 o-/o- Class 5 n- 
Class 2 e-/a- Class 6 e-/a- 
Class 3 i-/i- 

Class 4 e- 

The noun prefixes currently in the language, in various ways, reflect the distinc- 
tion between animate and inanimate nouns. 

1.1 Singular prefix 

One of the features of the prefixal system is that there is a semantic motivation 
associated with the nouns that take a particular prefix. For example, in the singu- 
lar, the nominal prefix o-/o- can be found, with some exceptions', on animate 
nouns. On the other hand, the prefix e-/t- goes on inanimate nouns only. This dis- 
tinction is illustrated in (3). In (3a), the nouns which have o-/o- are animate. 
However, in (3b), only inanimate nouns are shown to have the prefix e-/e-. The 
main point here is that whereas the o-/o- are predominantly animate prefixes, the e- ^m 
/e- are inanimate prefixes without exception. In other words, only inanimate ^^ 
nouns occur with e-/e- . 

(3) a. 



o-panyin 


'elder' 


o-hoho 


'visitor' 


o-kodee 


'eagle' 


3 -koto 


'crab' 



Osam: Animacy distinctions in Atcan grammar 



155 



b. 



e-boo 
e-dan 
e-woo 
e-tuo 



'stone' 
'building' 
'honey' 
'gun' 



1.2 Loss of nominal prefixes 

In Osam 1993 and 1994, the observation is made that one of the reasons for 
considering the noun class system in Akan as a decayed one is the loss of nominal 
prefixes. This loss may affect the singular only, or in some instances affects both 
the singular and the plural. In (4), the loss has affected only the singular nouns, 
but in (5) there is complete loss of the nominal prefixes in the singular as well as 
the plural. 

(4) 



(5) 



SINGULAR 


PLURAL 




tetea 


n-tetea 


'ant' 


prako 


m-prako 


'pig' 


pataku 


m-pataku 


'hyena' 


bokiti 


m-bokiti 


'bucket' 


dadewa 


n-dadewa 


'nail' 


siw 


e-siw 


'hill' 


SINGULAR 


PLURAL 




koraa 


koraa 


'calabash' 


kyew 


kyew 


'hat' 


sundzi 


sundzi 


'pillow' 


dwomba 


dwomba 


'pestle' 


wodur 


wodur 


'mortar' 


kuntu 


kuntu 


'blanket' 



The complete loss of nominal prefixes as illustrated in (5) is another evidence that 
the grammar of Akan is sensitive to animacy distinctions. The observation here is 
that inanimate nouns are more likely to lose their nominal prefixes than animate 
nouns. There are definitely more inanimate nouns in the language without prefixes 
than animate nouns without them. 

1.3 Double plural marking 

As stated in Osam 1993:100, one of the characteristics of the decayed noun 
class system of Akan is that certain nouns have double plural marking. This is the 
process in which certain nouns mark their plurals by using a prefix and a suffix at 
the same time. When we examine those nouns which behave this way, we find that 
they are all human nouns. Non-human nouns do not undergo double plural mark- 
ing. This process, therefore, serves to distinguish human nouns from non-hutiian 
animate nouns as well as from inanimate nouns. 



156 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(6) 



SINGULAR 


PLURAL 




o-hen 


a-hen-fo 


'chiefs' 


o-saman 


n-saman-fo 


'ghosts' 


o-banyin 


m-banyin-fo 


'men' 


o-panyin 


m-panyin-fo 


'elders' 


a-sew 


n-sew-nom 


'in-laws' 


o-nua 


e-nua-nom 


'siblings' 



1.4 Numeral modifiers 

Modification by numerals is not part of the nominal affixes but since it has to 
do with noun modification it is appropriate if it is discussed under this section. 
The behavior of numeral modifiers in Akan distinguishes human nouns from non- 
human ones. Before discussing this difference I give the numerals from 1 to 9 in 
Akan using the Fante dialect (7). 

(7) 



FANTE 


ENGLISH 


kor 


one 


ebien 


two 


ebiasa 


three 


anan 


four 


enum 


five 


esia 


six 


esuon 


seven 


awotwe 


eight 


akron 


nine 



In all the dialects of Akan, when the numerals from 1 to 9 are used to modify hu- 
man nouns, the form of the numeral is different from when they are used to modify 
non-humans. When these numerals modify non-human nouns, the forms are the 
same as given in (7). But when the modified noun is a human noun, the prefix ba- 
is attached to the numeral. This prefix derives from the noun ba 'child'. I should 
also point out that the vowel of the prefix will harmonize with the vowel of the 
root numeral. This distinction is illustrated in (8-10) with examples from the 
Fante dialect. As (9b) and (10b) show, it is wrong to put the prefix ba- on a nu- 
meral that modifies non-human nouns. On the other hand, when a numeral modi- 
fying a human noun does not have the ba- (8b), the result is only a questionable 
construction. It is possible to predict that over a period of time speakers may drop 
the ba- and regularize the form of the numeral modifiers. 

(8) a. Nyimpa ba-anan 

people child-four 
Four people, 
b. ?Nyimpa anan 
people four 
Four people. 

(9) a. N-dua anan 

CLASS 5-tree four 
Four trees. 



Osam: Animacy distinctions in Akan grammar 157 



(10) a. 



N-dua 


*ba-anan 


CLASS 5-tree 


child-four 


Four trees. 




M-bodom 


anan 


CLASS 5-dog 


four 


Four dogs. 




M-bodom 


*ba-anan 


CLASS 5-dog 


child-four 


Four dogs. 





It is also possible to use the human noun numeral modifier as the head of a noun 
phrase. This is illustrated in (11). These sentences are taken from the New 
Testament of the Bible. In this usage it is impossible to replace the (/jo-i-numeral) 
with the plain numeral. So whereas in its function as the modifier the numeral 
could possibly occur without the ba- (see 8b), when it functions as the head of the 
noun phrase it is grammatically wrong to have a bare numeral. In (11), therefore, 
obaako 'one (person)' and baakron 'nine (people)' cannot be replaced with kor 
'one' and akron 'nine' respectively. 

(11) a. Na o-baa-ko no bua-a no de 

and CLASS 1 -child-one the reply-COMPL 3SG OBJ COMP 
"Ana i-nn-suro Nyankopon?" (Luke 23:40) 

why 2SG SUBJ-NEG-fear God 

And the (other) one replied "Don't you fear God?" 
b. Na ba-akron no wo hen? (Luke 17:17) 

and child-nine the be where 
And where are the (other) nine (people)? 

2. Pronouns 

Another source of evidence for the animate-inanimate distinction is the 
pronominal system of the language. The evidence is based on the forms and the 
behavior of pronouns. 

2.1 Subject pronouns 

The subject pronouns in the language are distinguished on the basis of ani- 
macy. This distinction exists only in the Twi dialects, and it is relevant only in 
the 3SG subject pronoun. The 3SG subject prefix for animate nouns is 0-/0-, but 
that for inanimate nouns is e-/c-. This is illustrated by examples (12) and (13). 

(12) a. Abofra no be-yera 

child the FUT-be lost 
The child will be lost, 
b. -be-yera 

s/he-FUT-be lost 
s/he will be lost 



158 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

(13) a. Dua no be-yera 

tree the FUT-be lost 
The tree will be lost, 
b. e -be-yera 

it-FUT-be lost 
it will be lost 

It is clear that these prefixes derive from the old noun class system (Osam 1993). y^ ^ 
The animate subject prefix is a reanalysis of the old noun Class 1 marker whereas 
the inanimate subject prefix derives from the old noun Class 4 prefix. In the Twi 
dialects, the distinction in animacy as reflected in the 3SG subject prefixes is 
strictly maintained. In Fante, however, this distinction is neutralized since the 
same pronominal form is used irrespective of the animacy status of the antecedent 
noun. The Fante equivalent of (12b) and (13b) would be (14). Note that Fante uses 
the lexical form yew 'be lost' in place of the Twi verb yera 'be lost'. So the verbs 
yew and yera 'be lost' are dialect variants. 

(14) -be -yew 
s/he/it-FUT-be lost 
s/he/it will be lost 

2.2 Lack of number distinctions 

A further evidence of the grammatical differentiation between animate and 
inanimate nouns is that whereas anaphoric animate pronouns distinguish between 
singular and plural, inanimate pronouns do not make such distinctions. This dis- 
tinction applies more in the case of Asante and Akuapem and related dialects than 
in Fante, even though as I will show below, in the speech of some Fante speakers 
this distinction is available. In (15a), the subject is a plural animate noun and in 
(15b) it is replaced by the anaphoric pronominal prefix, wo 'they'. In (16a), the 
subject is an inanimate plural noun but the pronominal prefix replacement in 
(16b) is the same form used for singular as illustrated earlier in (13b). 

(15) a. Mbofra no be-yera 

children the FUT-be lost 
The children will be lost. 
b. Wo -be -yera 

3PLU-FUT-be lost 
They will be lost. 

(16) a. Ndua no be-yera 

trees the FUT-be lost 

The trees will be lost, 
b. e -be -yera 

3PLU-FUT-be lost 
They will be lost. 

The relationship between the notion of animacy and subject prefixes can be sum- 
marized as in (17). 



ii 



Osam: Animacy distinctions in Akan grammar 



159 



(17) 



Animate 


Inanimate 


3 


e 


VO 



Singular 



Plural 



This lack of number distinction in the inanimate pronominal prefixes reflects the 
presence of animacy hierarchy in the language. In his discussion of animacy hier- 
archy, Comrie states the following: '... having distinct singular and plural forms 
is again a characteristic of noun phrases with high animacy ... another opposition 
that correlates closely with animacy is the existence versus non-existence of a 
number distinction, the split invariably being that noun phrases higher in ani- 
macy have the distinction while those lower in animacy do not.' (1981:180, 182) 
Even though the evidence for such a hierarchy in Akan is minimal compared to 
languages like Russian, Chukchi, Dyirbal and many others, we still have some- 
thing that validates the argument that the notion of animacy hierarchy is relevant 
in the language. 

2.3 Possessive pronouns 

The distinction between animate and inanimate nouns is also demonstrated, 
in the Twi dialects, in the marking of certain possessive constructions. These are 
the constructions in which the possessed noun indicates some kind of relation, for 
example self, inside, bottom. In Akan and other languages, these are the nouns 
which are the sources of postpositions. In such constructions, when the possessor 
noun is animate, a full pronoun is used; but when it is inanimate we only get a 
pronominal prefix which incidentally is of the same form as the subject pronomi- 
nal prefix. This difference is shown in (18) and (19). 

(18)a. 



(19) a. 



b. 



Kofi ho 


a-ye fi 


Kofi self 


PERF-be dirty 


Kofi is dirty. 




Ne 


ho a-ye fi 


3SG POSS 


self PERF-be dirty 


He is dirty. 




Adaka no 


ho a-ye fi 


box DEFself PERF-be dirty 


The box is dirty. 


e-ho a-ye 


fi 


it-self PERF-be dirty 


It is dirty. 





2.4 Behavior of 3SG object pronoun 

Another evidence that the grammar of Akan is sensitive to animacy has to do 
with the behavior of the 3SG object pronoun. An aspect of Akan grammar that has 
been noted by various writers (including Christaller 1875, Stewart 1963, Boadi 



160 Studies in the Linguistic Sciences 23:2 (Fail 1993) 

1976, Lord 1982, Saah 1988, 1992) is that if the antecedent of the 3SG object 
pronoun is an inanimate noun, the pronoun is not overtly coded. Boadi 1976 
refers to this as the "Pronoun-3-Object Deletion Rule". Examples (20) and (21) il- 
lustrate this phenomenon. In (20a) the direct object is an inanimate NP, whereas in 
(21a) it is an animate NP. If the inanimate NP of (20a) is replaced by a pronoun, 
even though the sentence is grammatical, semantically, it implies that the an- 
tecedent of the object pronoun is an animate entity, not an inanimate one. The only t 
way (20b) can be formed to mean that the direct object is an inanimate entity is to 
have a zero pronoun in object position, as shown in (20c). Similarly, in 
(21c)where the object pronoun is covertly coded, the implication is that the direct 
object is an inanimate entity. In order to have the meaning that the direct object as 
an animate entity, the object pronoun has to be overtly coded as in (21b). 

(20) a. 



b. 



Kofi bo -ton dua no 




Kofi FUT-sell tree the 




Kofi will sell the tree. 




Kofi bo -ton no 




Kofi FUT-sell 3SG 




*Kofi will sell it. 




Kofi bo -ton 




Kofi FUT-sell 3SG 




Kofi will sell (it). 




Kofi bo -ton abofra 


no 


Kofi FUT-sell child 


the 



(21) a. 

Kofi will sell the child. 

b. Kofi bo -ton no 
Kofi FUT-sell 3 SO 
Kofi will sell him/her. 

c. Kofi bo-ton 
Kofi FUT-sell 3SG 
*Kofi will sell him/her. 

Even though Akanists are aware of this process, not enough has been done in 
terms of an explanation. One paper that has tried to deal with the issue is Saah 
1992. This paper sets the rule in Akan within the framework of the Government 
and Binding Theory by treating it as an example of null object in Akan. Another 
paper that also attempts to explain the process is Boadi 1976. Boadi's approach is 
to determine the historical source of the phenomenon. His conclusion is that this 
phenomenon must have been borrowed into Akan from Ga, a neighboring Kwa il I 
language spoken in coastal Ghana. ^^ 

Irrespective of the historical source of this phenomenon, it is my opinion that 
it has important functional implications in synchronic Akan. This view becomes 
relevant when we try to answer the question: Why is the object pronoun overt 
when its antecedent is an inanimate noun? The answer to this question is that it 
does so in order that the hierarchical ordering of animate and inanimate nouns is 
not subverted. It has to be understood that the form of the 3SG object pronoun is 
the same irrespective of the animacy status of the antecedent noun. Furthermore, as 



Osam: Animacy distinctions in Akan grammar 161 

mentioned above, since inanimate nouns do not make number distinctions in the 
pronoun, this same pronoun, no, is used for singular and plural antecedent inani- 
mate nouns. The covert coding of the object pronoun when its antecedent is inan- 
imate is built into the language to avoid the danger of hearers confusing an inani- 
mate noun with an animate noun. It is a way of telling the difference between ani- 
mate and inanimate nouns. In another sense the fact that animate nouns get re- 
I placed by pronouns but inanimate nouns are replaced by zero demonstrates that in 
a hierarchical ordering, animate nouns occupy a higher level than inanimate ones. 
It is for this reason that we can talk of animacy hierarchy in Akan. 

Having discussed this process and the motivation for it, it is necessary to 
point out that there are two conditions under which this process is compromised. 
As it will be shown below, these exceptions can be functionally accounted for. 
One of the exceptions is that when the direct object in the sentence is followed by 
an adverbial element indicating time or location, the rule does not apply. In other 
words, for a third person pronoun whose antecedent is inanimate to be covert, as 
Boadi 1976 puts it, the inanimate direct object has to "occur utterance finally". 
This process is illustrated in (22). As shown in (22c), the presence of the adver- 
bial okyena 'tomorrow' requires that the object pronoun be overtly coded. That 
the overtness of the pronoun is conditioned by the presence of the adverbial ele- 
ment is supported by the fact that when the adverbial item is fronted in a focus 
construction as in (22d) so that the direct object is in utterance final position, the 
pronoun is not overt as expected. I should mention that when uttered without a 
context, the animacy status of the antecedent noun of the object pronoun in (22b) 
is not clear; it could refer to an animate or inanimate noun. Similarly in (22e), the 
sentence is ungrammatical if the utterance final pronoun has an inanimate noun as 
its antecedent. 

(22) a. Kofi bo-ton dua no okyena 

Kofi FUT-sell tree the tomorrow 

Kofi will sell the tree tomorrow. 

b. Kofi bo -ton no okyena 
Kofi FUT-sell 3SG tomorrow 
Kofi will sell it tomorrow. 

c. *Kofi bo-ton okyena 
Kofi FUT-sell 3SG tomorrow 
Kofi will sell (it). 

d. okyena na Kofi bo-ton 

, tomorrow FOC Kofi FUT-sell 3SG 

It is tomorrow that Kofi will sell (it). 

e. *okyena na Kofi bo-ton no 
tomorrow FOC Kofi FUT-sell 3SG 
It is tomorrow that Kofi will sell (it). 

We can offer a functional explanation as to why the presence of an adverbial ele- 
ment requires the inanimate object pronoun to be overtly coded. This explanation 
has to do, specifically, with the pragmatic notion of topicality. It has been estab- 
lished that, at the clausal level, the NP that codes the subject relation is more topi- 



162 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

cal than any other entity in the clause; this is followed by the NP that codes the di- 
rect object relation (Givon 1984). In the Givonian functional framework, the 
subject is the "primary clausal topic" while the direct object is the "secondary 
clausal topic". Topicality hierarchy involving grammatical relations can be repre- 
sented as follows: 

Subject > Direct Object > Adverbial. 

In Akan, the immediate postverbal position is one of the crucial defining charac- 
teristics of direct object. This means that an NP which bears the direct object rela- 
tion necessarily has to occur immediately following the verb. 

The reason the presence of an adverbial element in the post object position as 
in (22a) triggers the presence of the inanimate object pronoun is that since the di- 
rect object is more topical than an adverbial item, and since the immediate 
postverbal position defines direct objecthood in Akan, if the pronoun is not 
overtly present it would create the impression that the adverbial element is more 
topical than the direct object NP. It is as if the inanimate object pronoun finds its 
topicality status threatened and so it has to make a physical appearance in order to 
assert its status. With this explanation it is understandable why when there is no 
adverbial in sentence final position the pronoun is covert. Under that condition, 
there is no threat to its topical status. 

The second condition which dictates the overt coding of the inanimate object 
pronoun is that there are a class of verbs which when used in the clause requires 
the presence of the pronoun. Example (23) illustrates this. In (23c) the absence of 
the inanimate object pronoun changes the meaning of the sentence. It should be 
noted that (23b) is ambiguous. If it is uttered in isolation from a context, the refer- 
ent of the pronoun no could be either animate or inanimate. 

(23) a. Kofi be-hyew edziban no 

Kofi FUT-burn food the 

Kofi will burn the food. 

b. Kofi be-hyew no 
Kofi FUT-burn 3SG 
Kofi will burn it. 

c. Kofi be-hyew 
Kofi FUT-burn 3SG 
*Kofi will burn (it). 

d. Kofi be-hyew 
Kofi FUT-burn 
Kofi will get burnt. 

Other verbs in this class are: see 'destroy', bu 'break', hyew 'burn', kyea 'bend', 
tsen 'straighten', tsew 'tear', moa 'crumple', yew 'lose', koa 'bend', monkyem 
'crumple', butuw 'overturn', bo 'break'. One feature of these verbs is that they be- 
long to the class of middle verbs, that is those verbs which normally take Theme 
direct object in transitive clauses, but also permit Theme subject in intransitive 
constructions. Why these verbs condition the overt coding of the inanimate object 
pronoun is not very clear to me at this stage. However, a possible reason may be 



(' 



Osam: Animacy distinctions in Akan grammar 163 

because they allow Theme subject in intransitive constructions. In such construc- 
tions, the subject entity is the one which undergoes the event indicated by the 
verb. Now, if the clause is supposed to have an object entity but this entity is not 
overtly coded, the only interpretation we can assign to such a clause is that it is 
the subject, the Theme, which undergoes the change in state. So in (23c), without 
the overt pronoun, the sentence cannot be interpreted as 'Kofi will burn it'. With 
no pronoun following the verb hyew 'burn', we have to interpret it as being used 
intransitively (23d). There is a pragmatic constraint on this analysis and it is that 
the analysis is legitimate only where the verb is capable of taking animate Theme 
subjects. For example, the verb bo 'break' cannot be given the same analysis be- 
cause by its semantics animate entities cannot be its Theme. Another possible rea- 
son is that these are change of state verbs which have drastic effect on the state of 
the entities which undergo the change of state. Since the change is drastic a way 
has to be found out to show the entity that has been so affected. In terms of notion 
of transitivity, the extent of the affectedness of the Theme NPs of these verbs make 
clauses that involve these verbs very high in transitivity. This is because the ex- 
tent of the affectedness of a Theme entity is one of the indicators of high transitiv- 
ity (Hopper and Thompson 1980). 

3. Conclusion 

In this paper I have shown that the conceptual distinction between animate 
and inanimate nouns in Akan has reflections in the grammar of the language. As 
mentioned at the beginning of the paper, Akan is not one of the languages noted 
for having an animacy hierarchy. Nevertheless, there is an extent to which we can 
say that the animacy distinctions instantiated in the language form a basis for such 
a hierarchy. This comes out in the pronominal coding of animate and inanimate 
entities. I have shown that whereas animate entities have coding forms which re- 
flect differences in number, the same is not true of the coding forms of inanimate 
nouns. Furthermore, the system of plural suffixation in addition to the regular 
prefixation, sets human animate nouns apart from nonhuman animate and inani- 
mate entities. The strongest manifestation of an animacy hierarchy is in the behav- 
ior of the inanimate object pronoun. Based on the evidence, we can say that in 
Akan, the following hierarchy exists: 

Human > Animate Nonhuman > Inanimate. 
This hierarchy, as has been shown above, is manifested in various aspects of the 
grammar. 

NOTES 

* An earlier version of this paper under the title Animacy distinctions in 
Akan was presented at the 25th Annual Conference on African Linguistics, 
Rutgers University, March 25-27, 1994. 1 appreciate comments made on earlier 
drafts of this paper by Colette Craig, Florence Dolphyne, and Kofi Saah. 

' Even though the nouns which occur with the prefix o-/o- are predomi- 
nantly animate entities, there are some inanimate entities that take this prefix. 
Examples include obotan 'rock', Oman 'country, nation', odan 'house', o tan 



164 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

'hatred'. There are other cases like own 'death', ohia 'poverty', Osaman 'ghost' 
which are biologically not animate but are considered as such based on cultural 
beliefs. 

REFERENCES 

AKROFI, C. A. 1937. Twi kasa mmra [Twi grammar] (1965 edition). Accra: 

Waterville Publishing House. 
BALMER, W. T., & F. C. F. Grant 1929. A grammar of the Fante-Akan language. 

London: Atlantis Press. 
BOADI, L. 1976. A note on the historical antecedents of the obligatory pronoun 

-3- deletion rule in the Akan dialects. Acta Linguistica Hafniensia 16:1.1-10. 
CHRISTALLER, J. 1875. A Grammar of the Asante and Fante language called 

Twi. Basel: Basel Evangelical Missionary Society. 
COMRIE, Bernard. 1981. Language universals and Unguistic typology. Chicago: 

University of Chicago Press. 
DOLPHYNE, Florence. 1988. The Akan (Twi-Fante) language: Its sound systems 

and tonal structure. Accra: Ghana Universities Press. 
, & M. E. K. Dakubu. 1988. The Volta-Comoe languages. The languages of 

Ghana, ed. by M. E. K. Dakubu, 50-90. London: Kegan Paul International. 
ESSILFIE, T. 1977. Serialisation in Fante. Leeds University Ph.D. dissertation in 

Linguistics. 
GIVON, Talmy. 1984. Syntax — a functional-typological introduction. Amster- 
dam: John Benjamins. 
HOPPER, Paul., & Sandra. Thompson. 1980. Transitivity in grammar and dis- 
course. Language 56.251-299. 
LORD, C. 1982. The development of object markers in serial verb languages. 

Studies in transitivity, ed. by P. Hopper & S. Thompson, 277-299. New 

York: Academic Press. 
OSAM, Emmanuel. K. 1993. The loss of the noun class system in Akan. Acta 

Linguistica Hafniensia 26.81-106. 
. 1994. Aspects of Akan grammar — a functional perspective. University of 

Oregon Ph.D. dissertation in Linguistics. 
SAAH, Kofi. K. 1992. Null object construction in Akan. MIT Working Papers in 

Linguistics 17.219-244. 
. 1988. WH- questions in Akan. Journal of West African Languages 18:1.17- 

28. 
STEWART, J. M. 1963. Some restrictions on objects in Twi. Journal of African 

Languages 2.145-149. 
UNITED BIBLE SOCIETIES 1982. AhyEmu Fofor [The New Testament]. Accra: 

The Bible Society of Ghana. 
WELMERS, W. 1971. The typology of the Proto-Niger-Kordofanian noun class 

system. Papers in African Linguistics, ed. by C.-W. Kim, & H. Stahlke, 1- 

16. Edmonton: Linguistic Research, Inc. 
. 1973. African language structures. Berkeley: University of California 

Press. 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 

THE INTERPLAY BETWEEN TONE, STRESS, AND 
SYLLABIFICATION IN THAI' 

%Uthaiwan Wong-opasi 
University of Illinois at Urbana-Champaign 
and Syracuse University 

In this study, tone is analyzed as an autosegment closely related to indi- 
vidual phonetic segments and suprasegments such as stress and syllable 
structure. Tone assignment in Thai is cyclic throughout word forma- 
tion, observing two well-formedness conditions, namely — syllable 
structure and the one-tone-only association. The hypothesis that tone is 
an autosegment, separated from, but correlates with stress and syllable 
structure will be proven via a historical linguistic account of tonogene- 
sis. Finally, the influence of Thai stress on tone distribution and sylla- 
ble structure will be treated for the first time in the literature via an ex- 
amination of language games involving foreign loanwords. 

1. Background 

Thai has FIVE lexical tones in the unmarked cases (cf. (1)) 

(1) Mid: khaa2 'a grass (imperata cylindrica)' 

Low: khaa 'galangal, a rhizome' High: khaa 'to engage in trade' 
Falling: khaa 'to kill; a servant' Rising: khaa 'a leg' 

However, not all five tones occur freely and various researchers, of both the 
segmental and autosegmental theories, have attempted to describe the tone occur- 
rence restrictions. We will first outline some of the arguments supporting the 
segmental analyses (Henderson 1949; Leben 1971, 1973; Candour 1974) which 
Yip 1980, 1982 rejected in favor of an autosegmental approach. 

Leben argues for a segmental nature of Thai tones because tone may be de- 
pendent on vowel length. He points out one tone change phenomenon in Thai 
which is triggered by a vowel shortening rule. Assuming contour tones to be de- 
rived (Leben 1971; Candour 1974), from the evidence that as a result of vowel 
shortening, the contour tone in the first element is neutralized to mid in the com- 
^^ binative speech style (cf. (2a)); while tone remains unaltered when vowel shorten- 
^^ ing does not apply (2b)3, Leben concludes that there is a vowel shortening rule 
(VV->V) and a convention that simplifies HL or LH to a "compromised" mid tone, 
phonetically. Furthermore, in the cases of underlying level tone, e.g. a high level 
tone, no tone change results despite vowel shortening (2c). 

(2) a. thii nay — > thi' nay 'where?' 

sii khaaw — > si' khaaw 'white' 

saaw saaw -^ saw saaw 'young girls' 

waarjwaar) — > war) waar) 'at your leisure' 



166 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

b. thaw ray — > thaw ray 'how much?' 

c. naam chaa — > nam chaa 'tea' 

According to Yip, if tone is segmental, when a vowel deletes, tone should be 
deleted as well (LH->L or H) rather than a compromised mid tone (*M). She also 
cites Candour's observation that elsewhere in the language, LH is converted to H, 
as in chan -^chdn T; phom -^phom 'I, male', and khaw -^khdw (third person 
pronoun). Moreover, upon a close examination of Henderson's data. Yip found 
examples of HL tone on short vowels followed by a glottal stop or zero in sentence 
intonation such as bd(7), IdC?). Furthermore, Yip cites counterexamples which 
include cases of tone change accompanying vowel shortening with underlying 
LEVEL tones. 

(3) yaar) rai — > yar) gai 'how?' 

yaar) nii ^ yar) oii 'this way, like this' 

Due to the gaps in Henderson's data for the discussion in her article. Yip 
cannot offer a firm conclusion but suggests that the tone change in (3) may be 
morphological rather than phonological or it may be stress conditioned (not by 
complete absence of stress, like the neutral tone, but rather by secondary as op- 
posed to main stress). This view is supported by Henderson's discussion of the 
neutral tone in unstressed syllables (1949:37, fn. 27) 

The actual pitch of the neutral tone may vary according to context, but 
is most commonly mid level. 

Also, Henderson 1967 affirms that the stress in forms like (2a), such as thii nay 
-^thi ' nay, falls on the second syllable. 

At this point, we would like to suggest treating the vowel shortening and 
tone change in (2) - (3) as postlexical phenomena which are caused by the stress 
pattern in Thai. Specifically, Thai is stress-final within a given phonological do- 
main, i. e. a word, a compound, a phrase, or a sentence. Although the two pro- 
cesses usually cooccur as they usually take place in unstressed syllables, they are 
independent from each other as we may find tone change without vowel shorten- 
ing (4a) and vice versa (4b). 



(li 



a. n a 


SWi 




-^ 


n a Q 


s«i« 




'book' 


tha7 


lee 




-^ 


tha 


lee 




'sea' 


ma7 


ra7 




-> 


ma 


ra7 




'bitter melon' 


kra7 


cok 




-^ 


kra 


cok 




'mirror' 


thoo 


ra7 


that 


-^ 


thoo 


ra 


that 


'television' 


kaa 


la7 


mee 


— > 


kaa 


la 


mee 


'a kind of sweet' 


pray 


sa7 


nii 


— > 


pray 


sa 


nii 


'post office' 


b. taa 


plaa 




— > 


ta' 


plaa 




'caIlous( lit. eye+fish)' 


paak 


kaa 




-> 


pak 


kaa 




'pen(lit.mouth-i-crow)' 


yaar) 


nan 




-> 


yaar) 


gan 




'that way, like that' 


yaar) 


nii 




-^ 


yao 


Qii 




'this way, like this' 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 167 

The last example, ycicuj nii —>yaij tjii, is a variant of yacuj nii -^ yaij ijii 
according to Henderson (cf. (3) above). It is given as a counterexample in Yip 
1982. 

The preceding discussion of tone neutralization is disapproved of by 
Candour who claims that acoustic results showed no evidence of such neutraliza- 
tion (though he eventually admits it in Candour 1979:140). However, native 
speakers of Thai (Warotamasikhadit 1967; Surintramont 1973) can hear the tone 
change which approximates the mid tone, although not exactly identical to under- 
lying mid tone, and the shortened vowels which are not exactly equivalent to un- 
derlying short vowels, either. Civen the nature of these postlexical rules which 
can be stylistic variants, and therefore optional, the inconclusiveness of the data is 
expected. Therefore, it may be safer to consider, as a point of departure, the nature 
of Thai tones in isolative speech. That is, we should concentrate on the tone on 
each syllable when pronounced in isolation or emphatically since these tones are 
truly lexical tones which occur at the lexical level in Lexical Phonology's terms. 

We turn now to isolative speech style. Candour 1974:138 proposes that the 
tone domain for each tone is a sonorant segment in syllable rime. That is, a syllable 
with CV(V)Cs structure, where Cs = a sonorant coda (traditionally called LIVE 
syllable), can have all 5 tones, while syllables checked with a stop (henceforth, 
CHECKED or DEAD syllables) have the tone distribution, formulated in Yip 
1982:89, as follows. 

(5) CVC H *HL *LH L *M 
CVVC *H HL *LH L *M 

A segmental analysis like Candour's clearly explains the absence of LH and 
HL on short CVC, and the presence of HL on CVVC. However, he attributes the 
absence of LH on CVVC as due to universal performance reasons. Yip, on the 
other hand, contests that HL on short CVC, and CVVC with high level tone are 
found in words other than loanwords and onomatopoeia (although they are less 
common), and that LH on CVVC is found in other languages. 

(6) khaat 'card' (Eng. loanword) but also poot 'a woman's nickname' 
khlak 'crowded, tightly packed' 

The only real gaps, according to Yip, are LH and M. (She also notes that the 
few apparent examples of LH are all intonationally derivable.) It must be noted that 
neither Candour nor Yip can account for *M. Regarding the absence of LH, Yip 
proposed a universal condition that prohibits the configuration *LH7 which ap- 
plies to Thai, Zahao, and Cantonese. She claims a laryngeal feature such as sylla- 
ble-final glottalization as the (p, t, k, 7) stops in Thai involve a glottal closure. 
Initially, these stops and (c) are glottalized. According to Yip, a segmental analy- 
sis cannot account for the existence of words like khldk which bear two tones on 
one vowel. Neither can the segmental approach describe the lack of three tone 
sequences in forms with three sonorant segments. To circumvent this fact. 
Candour posits a requirement that either the first and second segments be the 
same, or the second and third. In an autosegmental solution, however, the restric- 
tion is stated with less complications — only two tones are permitted per syllable. 



168 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

Another argument for a segmental account given by Leben (1973:34) is the 
fact that tones move with their vowels. Evidence is drawn from a Thai word game 
where rimes interchange between syllables. 

(7) kon yay -^ kay yon 'big bottom' 

Yip (1980:11), however, argues for floating tones which remain stable de- 
spite segmental change, i.e. transposition of rimes, in another Thai word game 
(following Candour's observation). 

(8) kluay hoom -^ kloom huay 'banana' 
t6n ram — > tam ren 'dance' 

A segmental suggestion will fail to account for the lack of tone movement. 
An autosegmental proposal, on the other hand, can account for both phenomena. 
In (7), rime movement occurs before the mapping of the suprasegmental tones, 
whereas (8) shows segmental change after tone association. 

Although two distinct rule orderings can give rise to two different language 
games (as in (7) and (8)), upon a scrutiny of more data, we find tone patterns that 
cannot be accounted for by Leben's or Yip's proposal since they show different 
tone patterns from those existing in the base forms. 

(9) 



dl 



chi khaa koo 


-> a. 


•choo 


khaa 


ki 


' 'Chicago' 


H M HL 




*HL 


M 


H 


(rime change, then tone 
mapping) 




b. 


*ch6o 


khaa 


kj 








H 


M 


*HL (tone mapping, then rime 












change) 




c. 


choo 


khaa 


ki 








M 


M 


L 


(resulting tones not in base 
form) 


khaa buu ki 


-> a. 


khi 


buu 


kaa 


'Kabuki' 


M M L 




L 


M 


M 


(rime change, then tone 
mapping) 




b. 


*khi 


buu 


kaa 








*M 


M 


*L 


(tone mapping, then rime 
change) 




c. 


khi 


buu 


ka7 








H 


M 


L 


(resulting tones not in base 



form) 

Following Burzio 1991, who proposes a well-formedness condition for 
English in cases where the ordering of certain phonological rules cannot yield a 
correct result, the situation in Thai here can be explained by resorting to tone 
well-formedness conditions (cf. details in section V below). The proposed well- 
formedness conditions on tone advocated in the present study, in turn, find sup- 
port in tonogenesis and syllable structure (cf. sect. 111). 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 169 

2. Tonal onsets 

Tumtavitikul 1991 argues for a combined segmental and suprasegmental ap- 
proach. She demonstrates the inadequacy of positing floating tones alone by prov- 
ing the fact that certain Thai consonants, in fact, carry underlying tones."* This is 
illustrated by suffixed examples such as leekh + aa -^leekhaa 'secretary' vs. rookh 
+ aa -^rookhaa. 'disease, illness'. The two forms have identical syllable structures 
in the base forms, i.e. a long syllable closed by an aspirated stop, plus an aa 
suffix. The two forms also exhibit tonal differences when suffixed by a short 
vowel plus a stop consonant, e.g. leekh -f- a7^>leekha7 vs. rookh + a?-^rookhd7. 
According to her, even though we may posit a floating H and L tone on the two 
forms, respectively, at the right edge of the base morpheme, or at the left edge of 
the suffixes, we still cannot account for such variations. Thus, she spells out tone 
assignment rules in (10) (present author's emphasis). 

(10) a. For ALL voiceless fricative onsets and SOME voiceless aspirated 
stop onsets. Rising tone surfaces on unchecked syllables. 

b. Elsewhere, unchecked syllables surface with Mid tone. 

c. For all resonant onsets and SOME voiceless aspirated stop 
onsets, short checked syllables surface with High tone. 

d. Elsewhere, checked syllables surface with Low tone. 

e. Rising tone in (a) is derived. 

The problems which arise in the above analysis are that various issues are left 
out unanswered. First of all, Tumtavitikul does not account for the fact that the 
long and short versions of the suffix, i.e. aa vs. a7, are merely the Thai phono- 
logical variants of the Indie short nominal suffix -a. Secondly, the tonal rules in 
(10) seem to be arbitrary. Moreover, there are no explicit ways to distinguish the 
two types of voiceless aspirated stop onsets in (10a) and (10c) or two distinct 
classes of voiceless fricative onsets. Failure to define these distinctions results in 
arriving at wrong predictions. For example, we would not have a monosyllabic 
minimal pair such as/a/i 'to dream, a dream' vs. fan 'tooth'; or a tonal contrast on 
the second syllable in multisyllabic words such as cha7 niian 'an electric in- 
sulator' vs. chd7 nuan 'a slate for writing on, a fuse', and cha.7 liar) 'a verandah, 
a balcony' vs. cd7 riaij 'woman's name'. Finally, the application of too 
many complicated phonological rules which interact with each other does not 
facilitate language acquisition. 

Alternatively, an analysis which is based on phonetic-phonological correla- 
tions and tonogenesis as well as the morphology-phonology interface, utilizing 
well-formedness conditions on syllable structure and tone at each stage of word 
formation will be argued for to which we now turn. 

3. Tonogenesis and phonetic-phonology correlations 

It could well be argued for that each Thai word is morphologically marked 
for tone in the lexicon. However, regarding loanwords from nontonal languages 
such as Pali, Sanskrit, English, etc., how the borrowed words get tones is of par- 



170 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

ticular interest. An examination of the Indie and Thai consonantal systems reveals 
the following facts : 

i) There was a loss in the voicing contrast so that both the Indie voiceless 
aspirated stops and voiced unaspirated stops were neutralized to voiceless aspi- 
rated stops, i.e. kh, g ->kh; ch, j-^ch; th, d^th; th, d— >th; ph, b— >ph. The 
original distinction between the two classes of consonants, however, is pre- , ' 
served in the resulting distinct tonal consonant classes (cf. appendix and v _ 
details below). 

ii) Voiced aspirated stops were lost in Thai pronunciation, i.e. gh, jh, dh, 
dh, bh^ 0. 

iii) The asymmetrical distribution of Thai voiced unaspirated stops, i.e. *g 
but d, b can be explained as due to a later Thai creation and thus, was not 
involved in tonogenesis. 

iv) Both the retroflex and the dental consonants are neutralized to dentals, t, t 
-^ t; th, th ^ th; d, d ^ d (^ th); dh, dh -^ dh (^th); ii, n ^ n. 
v) Moreover, laboratory experiments show that tone change can be induced 
by onsets: 
....voiceless oral obstruents produce high tone (or a higher variant of a 
tone) on the following vowel, whereas voiced oral obstruents produce 
low tone (or a lower variant) on the following vowel. (Haudricourt 
1961, Cheng 1973, quoted in Ohala 1978:25) 

....high air flow after voiceless, especially voiceless aspirated, obstru- 
ents, and low air flow after voiced obstruents caused the high and low 
pitches, respectively. (Ohala 1973, and Ohala 1978:26)5 

The available data suggest that pitch following voiced stops is substan- 
tially similar to that following sonorants and that it is the pitch follow- 
ing voiceless stops that is perturbed upwards. (Lea 1972, Hombert 
1975, Jeel 1975, quoted in Ohala 1978:29) 

The lost voicing and aspiration contrasts are preserved in the different tones 
assigned to the consonants in question. Thus, during the evolution from a voiced 
unaspirated stop to a devoiced aspirated counterpart, two features were encoded as 
one extrapolated tone, namely, [-voice] -^ H and [+spread] -^ H, equivalently 
HH— > H. To distinguish original voiceless aspirated stops from devoiced aspirated 
stops which were already assigned a high tone (in addition to unmarked [-vc, 
-sprd] onsets -^ default M), the single feature [-i-spread] of the original voiceless 
aspirated stops must take on a low tone as they are a lower variant between the two g. 
types of consonants. This is as outlined in (1 1) and (12), respectively. (^| 

(11) [-voice] [-(-spread] ^ HH ^ H 

\ / 
O 

That is, high tone in Thai onsets was caused by the devoicing and the aspiration 
of voiced unaspirates (e.g. g -^ kh). This also applies to sonorants since sonorants 
pattern like original voiced stops in tone assignment (cL point (v) above), and 
because of the loss of voicing contrast in voiced and voiceless sonorants in Thai. 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 171 

(12) [+spread] -^ L 

I 
O 

(12) should include also all aspirated voiceless segments, e.g. [kh], sibilants of 
Indie origin [s, s, s ^ s], and segmentless voiceless glottal fricative [h]. 

Contrary to Hombert 1978, Hock (1985:98) states that "Both onset and post- 
nucleus consonants can induce tone, but onset consonants have a greater effect." 
The Thai data confirms Hock's claim in unmarked cases. For marked cases such as 
in checked syllables, we postulate that the glottalized feature of final stops (p, t, k, 
7) in Thai has an effect on tone because these stops involve a glottal closure. 
Syllable-initially, these stops and (c) are glottalized while at the end of a syllable, 
they are unreleased stops, and therefore can increase syllable weight which in 
turn, can cause a low tone.^ The formulation is as follows. 

(13) [-continuant] -^ L 

I 
R 

4. Objectives 

This paper seeks to answer the following questions via the use of phonetic- 
phonological explanations in the Feature Geometry Theory (Clements 1985; 
Sagey 1986; Bao 1990; Duanmu 1990), the Autosegmental approach, and 
Burzio's 1991 Well-formedness Condition as an amendment to previous research. 

i) How is tone influenced by syllable structure? Specifically, why only 
heavy (or checked) syllables have tone occurrence restrictions as in (5), 
reproduced in (14). 

(14) CVC H *HL *LH L *M 
CVVC *H HL *LH L *M 

In addition to Yip's counterexamples to *HL on CVC and *H on CVVC, 
explanations will be given as to why these occurrences are marginal. 
Moreover, Yip's universal prohibition of the *LH7 configuration will be ex- 
plained. Finally, *M will also be accounted for with the same principle. 

ii) Why is a modular approach to Thai tonology preferred? This is because 
within Feature Geometry Theory, the segmental analysis can be restated in 
autosegmental terms. With each phonetic feature constituting a separate tier 
from, but interactive with, the segmental and the tonal tiers, we can achieve an 
optimal analysis of tone assignment and tone change in the formation of native 
Thai words along with foreign loans, language games, and sentential tonology. 
That is. Thai tone must be assigned and reassigned according to the syllable 
structure of Thai words at every stage of derivation, from isolated lexical 
words to units larger than words or postlexical units. 

iii) How can stress interact with tone assignment? Tone assignment and 
reassignment will be seen as an adaptation of Thai lexical and prosodic stress to 
accommodate different stress patterns of the donor languages. 



172 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

5. Well-formedness condition, syllable structure and tone 
assignment 

In this analysis, Thai syllables are proposed to be of two types: — LIGHT vs. 
HEAVY syllables — with the former containing [+continuant] rime, and the latter, 
[-continuant]. All short vowels are treated as followed by a glottal stop and long 
vowels are represented as sequences of two identical vowels. It will be argued for 
here that the tone bearing units (TBU's) in Thai can be any segment in the sylla- \ 
ble. Moreover, unlike previous proposals which utilize a noncyclic approach, this 
study will show that tone assignment in Thai is cyclic both at the syllabic and the 
morphological levels (cf. (15) and section 6). That is, tone is first assigned ac- 
cording to the phonetic properties of the onset of a syllable which constitutes the 
first tonal domain, represented here by a right bracket. Unspecified onsets induce 
mid tone which is also an unspecified or default tone. Devoiced aspirated onsets, 
on the other hand, raise the tone of the syllable to H (rule (11)) whereas original 
voiceless aspirated onsets are assigned L tone (rule (12)). Upon incorporation of 
each new segment in the syllable, a subsequent tonal domain is created (i.e. from 
tonal domain 1, 2... to Tn ), while previously assigned tones are erased and a new 
tone is assigned in accordance with the [+/- continuant] feature of the new rime 
segment. That is, in marked cases, a rime-final [-cont] segment induces L tone 
(rule (13)) while M is the default tone for unmarked [+cont] rime. The generation 
and regeneration of tone is seen in (15) where the association line is truncated and 
readjoined between the tonal tier (T's) and the segmental tier (X's) when each ele- 
ment is integrated in the syllable, from onset (O) to rime (R), thereby creating a 
gradually larger tonal domain. We show the tone assignment of a Thai native word 
raa 'mold' in (16) (next page), [r] is first assigned a H tone. Then, [a] is integrated 
with a new tone M assigned because [a] is [-i-cont]. The preceding L tone is thus 
dissociated. When the final segment is incorporated into the syllable, the second 
tone is replaced by the final tone. 

(15) Tone Assignment : T — > T 

/ I \ 

T, T2....Tn) 

I I I 
) 



X, 

1 


X2. 
1 


,...X, 
1 


1 

...][ 

1 


1 

\ 


1 

] [. 

/ 







R 


\ 


a 


/ 



Hence, for light syllables which have unmarked syllable structures, the default 
tone is mid unless specified otherwise in the lexicon. Consequently, the unmarked 
syllable structure allows its association with any one of the five lexical tones, the 
choice of which is lexically governed, and is reflected in the use of an explicit 
tone marker in the language. Heavy syllables, on the contrary, are subject to a 
well-formedness condition in Thai which prohibits mid tone (cf (17a)). This is 
due to the fact that unlike sonorant codas which pattern like vowels, stop codas 



/ 


1 


\ 




H 

1 


M 

1 


M) 
1 




r 


1 

a 


1 

a) 




[+son] [+cont][+cont] 


1 


\ 


/ 







R 






\ 


/ 

G 







Wong-opasi: The interplay between tone, stress, and syllabification in Thai 173 

constitute heavy syllables and for the purpose of tone assignment, this heavy 
weight exerts itself in a low tone. Another well-formedness condition (17b) pro- 
hibits that no Thai segments contain more than one tone. 

(16) T ^ T -^ T 
I I \ 
H) H M) 
I I I 
r) r a) 

[+son] [+son] [+cont] 

I I I 

O OR 

\ / 
o 

(17) Well-formedness Conditions : 

a. Prohibition of Mid Tone: *M ) 

I 
[-cont] ) 
I I / 
O R 
\ / 

CT 

b . Tone Restriction:only ONE tone can be associated to a segment : *T T ) 

\ / 
X 

The fact that contour tones exist in Thai syllables is due to the preservation 
of previously assigned tone when the onset contains either the dual [-voice] and 
[+spread] feature or [-f-spread] alone. That is, the bracket erasure convention does 
not apply in these cases and therefore, two tones are assigned, one to the onset, and 
the other eventually to the coda. For low tone, since both the onset and the final 
segment bear non-distinct low tones, the two low tones are fused into one — LL -^ 
L (cf. (18a) on the next page). For rising tone, there is an additional contour tone 
exaggeration effect whereby the LM tone sequence is enhanced to LH for maximal 
audibility (cf. (18b) on the next page). Thus, the cyclic tone assignment (15) and 
the well-formedness condition (17b) account for contour tones in short and long 
syllables. 

We turn now to the high tone in CVC and the falling tone in CVVC struc- 
tures. High tone is first assigned to a devoiced aspirated onset. Because of the 
highly marked feature [-voice, +spread], the already assigned high tone is kept 
while integration of the remaining segments of the syllable generates an additional 
low tone in agreement with the stop coda. The outcome is HL but due to the short- 
ness of the heavy syllable, the final tone is not integrated, except in very few into- 
nationally derivable lexical words. However, in long heavy syllables, either op- 
tion is available. With the preservation of the final low tone, the syllable contains a 
falling tone as found in most Indie loanwords. In most English loans, in contrast. 



174 



Studies in thie Linguistic Sciences 23:2 (Fall 1993) 



\ 

L) 

I 

t) 

I 



/ 



/ 



L) M 



in response to the need to distinguish English words from already established 
Thai words, including Indie loans, the final low tone is simply dropped. 

(18) a. Low Tone in CV(V)C Structures : Tone Conflation: LL ^ L. 

i. khat 'scrub' ii. khaat 'broken' 

T 
/ I 
L) M 
I I 
kh)a 
I I 
[+sprd] [+cont] [-cont] 
I \ / 
O R 
\ / 

CT a 

b. Rising Tone in CV(V)C Structures : Contour Exaggeration: LM -> LH. 
i. khan 'bowl or basin used for dipping water' ii. khaan 'to call in reply' 



I 
kh) 



\ 

M 



\ 

L) 
I 
t) 



[+spread] [+cont] [+cont] [-cont] 
I \ I / 

O R 

\ / 



( 



/ 

L) 
I 



\ 



M M) 
I I 



kh) a n) 



[-i-sprd] [-i-cont] [-i-cont] 
I \ / 

O R 

\ / 

o 

(19) a. High Tone : 

i. khat 'select' 
T 
/ I \ 

H) M L) 



kh) 



I 

t) 

I 



/ 

L) 

1 


/ 

M 

1 


\ 

M 
1 


\ 

M) 

1 


kh) 

1 


a 

1 


a 

1 


n) 

1 


[-i-spread][+cont][ 
1 \ 


\ 


+cont][+cont] 
1 / 
R 

/ 




a 






b. Falling Tone: 






ii. khaat 

/ 

H) 

1 


'to gird' 
T 

/ 

M 

1 


\ 

M 

1 


\ 

L) 

1 


kh) 
1 


a 

1 


a 

1 


t) 

1 


[-VC, -i-sprd] 
1 


\ 


[+cont] 

\ 

a 


[-I-cont] 
1 
R 

/ 


[-cont] 

/ 



[-vc,-i-sprd] [-I-cont] [-cont] 
I \ / 

O R 

\ / 

o 

6. Evidence 

Tone forms an autosegmental tier as it can be separated from the segmental 
tier. However, the tone autosegment is linked to individual segmental features as 
in the Feature Geometry Theory. Evidence is found both on phonological and 



Wong-opasi: The inteqjlay between tone, stress, and syllabification in Thai 175 

morphological grounds. We will first look at three types of cluster onsets:a) clus- 
ters with a segmentless tonal onset [h]; b) true cluster onsets; and c) unparseable 
cluster onsets. 
6.1 Phonological evidence 

Thai has one segmentless tonal onset which is a voiceless glottal fricative [h] 
with the feature [+spread]. Like all other aspirates which are [+spread], it bears a L 
tone. If tone were segmental, it would be impossible to convert the high tone to 
low in short heavy syllables with a sonorant onset just by adding this segmentless 
[h] to the onset, e.g. mat (H) 'to tie' vs. hmat (L) [mat] 'fist' because this segment 
does not manifest itself phonetically. A rising tone obtained by adding this 
segmentless onset to a sonorant within the initial cluster would be impossible, too, 
e.g. naa (M) 'farm' vs. hnaa (LH) [naa] 'thick'. "^ (20) shows the elimination of the 
stranded consonant [h] from an unpronounceable Thai cluster [hn]. Despite the 
suppression of [h] from the segmental tier, its tone is not deleted and has the 
spreading effect onto the following onset. The well-formedness condition (17b) 
ensures that only one tone is assigned to the sonorant after spreading; thus, the 
tonal feature [H] of the sonorant must be dropped 

a. Segmentless Tonal Onset : 



(20) a. naa 


'farm' 




vs. 


b. naa 'thick' (LM 


^LH) 




T 






T 






/ 


1 


\ 




/ / 


\ 


\ 


H 

1 


M 
1 


M) 
1 




L) H 

1 \ 1 


M 

1 


M) 
1 


n 

1 


a 

1 


a) 

1 


-^ 


(h) n) 
1 \ 1 


a 

1 


a) 
1 


[+son] 


[-i-cont] 


[-i-cont] 




[-t-sprd] [+son] 


[-I-cont] 


[-I-cont] 


C 


V 


V 




C C 


V 


V 


1 


\ 


/ 




\ / 


\ 


/ 





R 









R 




\ 


/ 






\ 

a 


/ 




b. True cluster onsets: True cluster onsets are those which share the 



slot underlyingly and therefore, do not permit vowel insertion to break the 
cluster. As such, only the first member in the cluster can attract tone while no 
tone can be assigned to the second member. In (21a) (next page), high tone is 
assigned to [kh] according to the features [-voice, -^spread]. Because [r] is not 
provided with an independent C-slot, it cannot bear tone but receives the high 
tone from its partner through spreading. Likewise, the feature [-i-spread] triggers 
low tone in (21b) (next page) and spreads onto [r]. If tone were segmental alone, 
there would be no way to explain the single tone assigned to true cluster onsets. 

c. Unparseable cluster onsets : For consonants that cannot form an initial 
cluster in Thai, then, either the second member is dropped, e.g. sr — > s; cr — > c, 
or a default vowel [a7] is inserted to break the cluster to accommodate the Thai 
pronunciation, e.g. chn — > cha'7-n. Before vowel epenthesis, however, the 
features of the first member of the cluster are collocated onto the second member, 



176 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



i.e. [-voice, +spread] of [ch] on [n] in (22a) and [+spread] in (22b), respectively. 
Thus, H and L tones are assigned to (22a) and (22b), in that order. 



(21) a. khrua 

/ 
H 

1 \ 
kh r 
1 \ 1 
[-vc+sp][+son] 
\ / 
C 
1 

o 

\ 

Derived Tones: 


'kitchen' 

T 

1 

M 
1 


vs. 

\ 

M) 
1 


b. khriia^ 'a^ 

/ 

L) 

1 \ 

kh r) 

1 \ 1 

[+sprd] [+son] 

\ / 

C 

1 



\ 

LM 


;ed Buddhist priest' 
T 
1 \ 

M M) 

1 1 


1 

u 
1 
[+cont] [ 
1 

V 
\ 

R 
/ 
c 
M 


a) 
1 

+cont] 
1 

V 
/ 


1 

u 

1 
[+cont] 

1 

V 

\ 

R 
/ 
o 
->LH 


a) 
1 
[-i-cont] 

1 

V 
/ 


(22) a. chnuan 

a. T 
/ 1 \ 
H MM 
1 \ 1 1 
ch (n) u a 
1 \ 1 1 1 

[-vc+sp] [+sn] [+ct] [+ct] 
1 III 


— > 

— > 

\ 

M) 
1 


cha7 nuan 'fuse 
T 

/ 1 \ 
H) M L) 

1 1 1 


/ 

H 

1 


T 

1 \ \ 

M M M) 

1 1 1 


1 

n) 

1 

[+ct] 
1 


1 [- 


1 1 1 

ch ) a ? ) 

1 1 1 

vc, +sp][+ct][-ct] 

1 1 1 


1 III 

n u a n ) 

1 1 1 1 

[-vc,-i-sp] [+ct] [-I-ct] [-I-ct] 
1 1 1 1 


1 1 

C C 

\ / 

o 

\ 


V V 

\ 1 
R 

/ 


1 

C 

/ 




1 1 1 

C V C 

1 \ / 
R 

\ / 


1 

C 

1 


\ 


V V c 

\ 1 / 

R 

/ 




Derived Tones: 


M 






a 
HL^H 




a 

M 


b. chnuan 
b. T 

/ 1 \ 
L) MM 
1 \ 1 1 
ch ) (n) u a 
1 \ 1 1 1 
[+spread] [+sn] [-i-ct] [+c1 
1 III 


\ 

M) 
1 


-4 
— > 


cha7 nuan 
T 

/ 1 \ 
L) ML) 

1 1 1 


'insulator' 
T 

/ 1 \ \ 
L) M M M ) 
1 ill 


1 

n) 
1 

:][+ci 
1 


t] 


1 1 1 

ch ) a ? ) 

1 1 1 

[+sprd][+ct][-ct] 

1 1 1 


1 III 
n ) u a n ) 
1 1 1 1 

[-hSp] [-HCt][-f-Ct][+Ct] 
1 III 


1 1 

C C 

\ / 


\ 


V V 

\ 1 
R 

/ 


1 

C 

/ 




1 1 1 
C V c 

1 \ / 

R 

\ / 


1 1 

C V 

1 \ 
o 

\ 


V C 

1 / 
R 

/ 


a 
Derived Tones: 


LM^ 


LH 




a 
LL-^L 


a 
LM^LH 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 177 

6.2 Morphological evidence 

Tonal consonants clearly manifest themselves in derived words through the 
infixation of am (n) (23), lexical derivation (24-26), and language games (32). 

6.2.1 am (n) infixation 

A limited number of monosyllabic words become bisyllabic via am (n) in- 
fixation. Such an infixation is no longer operative in Thai and its idiosyncrasies 
must be recorded in the lexicon since the base for the am (n) infixation can be of 
any lexical category and the derived form usually maintains a synonymous or re- 
lated meaning to that of the base form while the lexical category of the base may or 
may not be altered in the derived form. Examples are: s(r)et 'to finish'-^ s+am+ ret 
'to succeed'; t(r)uat 'to check'^ t+am+ ritat 'police'; dadn 'to walk'— » d+am + 
119911 'to proceed; (Royal vocabulary) to walk'; khrop 'to complete'— > k+am+ rop 
'time (as in "cycle")'; praap 'to get rid of^ h+am+racp 'to subdue'; prutj 'to 
improye'—^h+an+rutj 'to support'; and trat 'to speak (Royal vocabulary)' -> 
d+am+rat 'to speak (Royal vocabulary)'. For the first example, the cluster [sr] 
does not form true cluster onsets in Thai since they each take a separate C-slot. 
Because the [r] is deleted from the cluster, it allows the spreading of [-i-spread] 
from the first member of the cluster onto the second member before am(n) 
infixation. The tone in the infixed form is thus as predicted. (The changes p— >b 
and t^d are explained in section III pt. (iii) above.) 



(23) i) s(r)et 

/ 

L) 

1 \ 
s) (r) 
1 \ 1 
[-f-sprd][-i-son][ 
1 1 


T 

1 
M 

1 


\ 

L) 
1 


s -1- a m -1- r et 
T 

/ 1 \ 
L) MM) 
1 1 1 


'succeed' 

T 
/ 1 
L) M 
1 1 




\ 

L) 
1 


1 

e 

1 
+cont] 

1 


1 

t) -> 
1 
[-cont] 

1 


1 1 1 

s ) -I- a m ) 
1 1 1 
[-(-sprd][-t-cnt][+cnt] 
1 1 1 


1 1 

+ r ) e 
1 1 
[+sprd][+cont]| 
1 1 


1 

t) 
1 
[-cont] 
1 


I 1 
C C 
\ / 


\ 


1 

V 
\ 
R 

/ 


1 

C 

/ 


1 1 1 

C V c 

1 \ / 

R 

\ / 


1 1 

C V 

1 \ 

o 

\ 


R 

/ 


1 

C 

/ 


a 
Derived Tones: 


: LL 


-^L 


a 
LM-^LH 


a 
LL-> 


L 





Only through an autosegmental analysis which permits spreading can we ar- 
» rive at a consistent explanation for retaining the tone of the base forms in the sec- 
ond syllable of the derived words while the first syllable acquires a new tone ac- 
cording to its phonological constructs. It also allows for the lack of tone sandhi in 
forms such as sraij — > s+rni+niaij 'voice' and sriicm — > s+am+ruaix 'laugh'. 
In the latter cases, we can claim that am(n) infixation applies first and therefore, 
bleeds spreading. Finally, for words that are clearly marked with a lexical tone via 
the use of a tone marker such as cciay 'pay' — > c+am+ naay 'distribute', it is 
evident that since the lexical tone is not part of the syllabic structure, the lexical 



t 



178 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

tone must then be retained in the second syllable, leaving the first syllable free to 
obtain tone as conditioned by its own syllable structure. 

6.2.2 Derived lexical items 

Most Indie loanwords suffered the following morpho-phonological changes: 
i) final vowel (plus nasal) truncation^ (followed by cluster simplification, when 
applicable, and final stop neutralization to unreleased p, t, k stops; or 1, r — > n); 
ii) otherwise, the final vowel is retained, either as a long vowel or as a short vowel 
accompanied by a glottal stop, primarily in suffixed forms and compounds. 

In general, the Indie -a suffix is truncated in Thai. The following examples 
are from Pali (unless indicated otherwise) although the same set of rules applies 
also to Sanskrit:/:«/+a (^ kul) — > kun 'family'. (The intermediate stage of 
derivation is given in parentheses.) The distinct tone patterns are regulated by 
Thai syllable structure. Specifically, tone evolved from the loss of certain Indie 
phonetic properties of the onset and the new rime structures in the Thai language 
as discussed above, e.g. ratth+am-^ rat 'country'; raastr+a (Skt.)— > rdat 
'country'; saastr+a — > saat 'science/art'; narak+a -> na rok 'the great heir;yVw+a — > 
chon 'the public; people', the last two examples exhibit an additional vowel 
change [a-^o] word-finally in most Indie loans. (24) shows the effect of syllable 
structure on tone in simple loanwords with final vowel truncation. 

i) Final Vowel Truncation : 

(24) a. rattham'O (Pah) -^ rat 'country' 

a. T T ^ T 

/ I \ / I \ / I \ 

H) M L) L) M M) H) M L) 

I I I I I I III 

r) at)th) am) r) a t) 

I I I I II III 

[+son][+ct][-ct][+spr] [+ct][+ct] [4-son] [+ct] [-ct] 

I \ / I \ / I \ / 

R O R OR 
\ / \ / \ / 

o (5 a 

Derived Tones: HL^H and LM ^ LH vs. HL^H 

b. raastra (Skt.) -^ raat 'country; citizen' 

b. T T -^ T 

/ / \ \ / \ \ / / \ \ 

H) M M L) M ML) H) M M L) I 

1 I I I I \ I I I I I I 
r) a as) t r a 7) r)aa t) 
I III I \ I I I I I I I 

[+son] [-i-ct] [-i-ct][-ct] [...][+sn][+ct] [-ct] [+son][-i-ct] [-Kct] [-ct] 
I \ I / \ / \ / I \ I / 

OR OR OR 

\ / \ / \ / 

a o <3 

Derived Tones: HL and L vs. HL 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 179 

The preservation of the suffix -a is less common and requires various ad- 
justments. Despite the fact that Thai does not have comparable stress to the type of 
stress found in stress languages like English, Pali, and Sanskrit, Thai stress can be 
detected solely via its influences on stress-related phenomena such as tone and 
vowel length. Stated differently, given that the tone and the vowel length of each 
word in Thai are lexical as they constitute semantic differences, these tone and 
vowel length contrasts must be encoded in the lexicon. In addition, each lexical 
word has a potential to bear stress in a slow, emphatic speech. The capability to 
bear tone, vowel length contrast, and stress applies both to mono- and poly- 
syllabic words, in the optimal cases. By implication, this means that each syllable 
in poly-syllabic words can bear the tonal, vocalic, and stress distinctions as well. 
However, in normal speech, not all syllables or words are pronounced with equal 
force or stress as regulated by the prosodic rhythmic patterns. We postulate that 
within a phonological domain. Thai primary stress falls on the last syllable, and 
the secondary stresses alternate with unstressed syllables. When a non-final sylla- 
ble is destressed, it can cause vowel shortening and/or mid-tone neutralization as 
seen in (4a) above. The shortening of the short vowel [a] which, in Thai, must be 
accompanied by a glottal stop, is carried out by the deletion of the glottal stop. 
Our hypothesis that Thai is stress-final within a given phonological domain is 
corroborated by the fact that a glottal stop can be deleted in non-final, unstressed 
position, as in ma? ra?-> ma rd7 (with an additional mid-tone neutralization on 
the unstressed syllable) but it can never be dropped word-finally as shown in the 
ill-formedness of ma? ra?^ *ma7 ra or *ma ra. In effect, stress in Thai falls on 
the final position of a phonological domain, i.e. on the final syllable of a word, at 
the lexical level where lexical words are being constructed, or on the last syllable 
of a phrase at the sentential or postlexical level. This is due to the fact that the final 
syllable of the stress domain in Thai is heavy by position. The final stress in Thai 
overrides the stress patterns in the donor languages, which for Pali and Sanskrit 
may fall either on the stem or the suffix; and for English, either on the final or non- 
final syllable. As such, the domain-final syllable in Thai must contain at least two 
moras or a two-segment rime, so that the final syllable can bear stress. To render 
the short Indie suffix with a plain short vowel in Thai would violate the syllable- 
structure requirement. Consequently, the Indie short vowel must be lengthened 
or a glottal stop must be added to preserve the heavy weight assigned to the 
stressed position. This effect is attested in many doublets in Thai. For 
example :ra«/-i-a -^ raa ch+aa/raa ch+d7(oT rdat) 'king'; kaay+a -^ kiia y+aa / kaa 
y-i-a?(or kaay) 'body'; geh+a -^ khee h+aa / khee /z-l-a? 'house'. The choice 
between the final vowel truncation, its lengthening, and the glottal stop insertion 
is arbitrary and thus, must be marked in the lexicon. In certain instances, the 
resultant distinct forms are all employed to denote different lexical meanings, e.g. 
likh+a (cf. likh+ati) 'to write' -^ lee kh+aa 'secretary (=one who writes)' vs. leek 
'arithmetic (=written numbers)'; sukh+a 'happiness' — > sii kh+ua 'toilet (=happy 
place)' vs. suk 'happiness'. Word-final glottal stop is retained only in emphatic 
speech or in Indicized contexts, e.g. su kha? in the Buddhist blessing, ?aa yii 
wan nd? su kha? pbd7 Id? 'longevity, growth, happiness, and health'. 

The effect of suffixal vowel lengthening is illustrated in (25). 





T 




T 


/ 


1 \ 


/ 


1 \ 


-I 


M M) 


H 


M M) 



180 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

ii) Indie Suffixation and Vowel Lengthening : 

(25) a.leekh + aa (Skt.) 'secretary' vs. raaj+ aa (Skt.) 'king' 

b.sukh + ii (P) '(the) happy, blessed (one)' vs. rookh + ii (P) '(the) sick' 

T T 

/ I \ / I \ 

L) M L) L) M M) 

I I I I I I III I II 

s) u 7 ) kh) i i) r o o) kh i i) 

I I I I I I III I II 

[+spr] [+ct] [-ct] [+spr] [+ct] [+ct] [+son] [+ct] [+ct] [- vc ,+sp] [+ct] [+ct] 

I \ / I \ / I \ / I \ / 

OR OR OR OR 

\ / \ / \ / \ / 

a a a a 

Derived Tones: L and LM -^ LH vs. M and M 

The distinct tonal patterns caused by different syllable structures are most 
evident in the Thai reading of original Indie suffixed forms, e.g. PaVv.khattiya —> 
khat. til' yd? (m.) 'man of the warrior caste' vs. khattiyaa -^ khdt. ti7 yaa, 
khattiyaanii -^ khat. ti? yaa nii (f.) 'woman of the Khattiya clan'; naaga -^ naa 
khd? (m.) 'cobra, elephant; iron-wood tree; noble person' vs. naagii -^ naa khii, 
naaginii — > naa khi? nii (f.) 'female cobra, female elephant; noble woman'. 

Our postulation of syllable weight received from word-final stress is further 
corroborated by the fact that when simple words are compounded, neither an 
epenthetic long vowel nor a glottal stop are needed since the final syllable of the 
first member of the compounds is no longer stressed as the new stress domain is 
the last syllable of the entire compound. As a result, the tone of the unstressed syl- 
lable is neutralized to mid as observed by previous researchers (cf. (2-4) above). In 
emphatic speech, however, where all syllables are stressed, the glottal stop is re- 
tained and triggers an appropriate tone assignment according to the new syllable 
structure, e.g. jan+a + pad+a (people+path) 'country-side' -^ chon+d + bdt 
(emphatic speech) vs. chon+a +bdt (normal speech); or when the second member 
of the compound nouns begins with a vowel, e.g.jal+a (Skt.) 'water' + aalay+a 
(Skt.) 'dwelling, house' -^ chd? laa lay (emphatic speech) vs. cha laa lay (normal 
speech). 

iii) Indie Compounds : 

Thai has borrowed heavily from the Indie vocabulary through Thai nativiza- 
tion processes. We will first demonstrate simple word borrowing. Examples are 
provided in (26a-b). The adaptations of certain phonemic sounds to fit the Thai 
phonetic inventory, e.g. dental and sibilant neutralizations, aspiration and devoic- 
ing of stop consonants (cf. section III above), where applicable, are all shown in 
step of (26a-b) below, since no ordering relation is assumed among these phono- 
logical rules. For each borrowed lexical item, the Indie syllabification is shown in 
step 1 . Being a tonal language. Thai assigns tone according to the syllable struc- 
ture and the phonetic properties of both the onset and rime (step 2). Final nasals 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 181 



are deleted (step 3). When a segment is deleted from a syllable, several adjustments 
are underway. First, as noted by Thai researchers, all syllables in Thai are closed 
by either a sonorant or a stop, including a glottal stop. This glottal stop may be 
deleted in unstressed position. However, since the stress domain of Thai words is 
the final syllable, a bare short vowel may not surface in word-final position. 
Consequently, no tone can be placed or retained on the ill-formed syllable struc- 
ture (cf. the output after the word-final nasal is deleted *tha in rat. *tha in step 3). 
To remedy the situation, a glottal stop is inserted (step 4) which in turn, calls for 
tone reassignment since the stressed, checked syllable conditions a low tone in 
unmarked cases (step 5), unless the onset is marked as [-voice, -i-spread]. 



0. 


rattham 


1. 


rattharn 


2. 


rat.tharn 


3. 


rat.tharn 


4. 


rat.*tha 


5. 


rat.*tha7 



li) 


-> rat 


'country' 


-» 


rattharn 


Dental Neutralization 


-^ 


rat.tharn 


Syllabification 


-^ 


rat.tharn 


Tone Assignment 


-^ 


rat.*tha 


Final Nasal Deletion 


-^ 


rat.*tha7 


7-Insertion 


-^ 


rat.tha? 


Tone Reassignment 



The derivation in (26b) is in a similar fashion. Step denotes the neutraliza- 
tion of all sibilants. Syllable structure is assigned in step 1 followed by tone as- 
signment in step 2 which succeeds only in the first syllable while the second syl- 
lable is devoid of any tone since its syllable structure is not permitted in Thai. 
Step 3 illustrates the non-applicability of the nasal deletion rule since the word in 
question does not contain a word-final nasal. Syllable adjustment by the insertion 
of a glottal stop takes effect in step 4 so that tone can be placed on the final syllable 
in step 5. 





b. saastra 


(Skt.) 


-> saat 


'science/art' 


0. 


saastra 


-^ 


saastra 


Sibilant Neutralization 


1. 


saastra 


-> 


saas.*tra 


Syllabification 


2. 


saas.*tra 


-^ 


saas.*tra 


Tone Assignment 


3. 


- 




- 


Final Nasal Deletion 


4. 


saas.*tra 


— > 


saas.tra? 


?-Insertion 


5. 


saas.tra? 


— > 


saas.tra? 


Tone Reassignment 



The outputs at this stage of derivation serve as inputs to compounding. Step 
6 illustrates the juxtaposition of two or more simple words. The glottal stop of the 
first member of the compound is deleted in step 7 because it no longer stands at the 
edge of the word since the word boundary is now extended to cover the entire 
compound. Consequently, the previously-assigned tone is also removed as a con- 
sequence of loss of stress in non-final position. The elimination of glottal stop and 
mid-tone neutralization are also found at the postlexical or sentential level where 
words arc put together to form phrases or sentences. 

c. rat.tha? + saas.tra?— > rat.tha saat 'political science' 

6. rat.tha?+saas.tra?— > rat.tha? saas tra? Compound Formation 

7. rat.tha?-i-saas.tra?^ rat.tha saas tra? Non-final ? Deletion 

8. rat.tha?-i-saas.tra?-^ rat.tha saas tra? Mid-Tone Neutralization 



182 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 



(26') a. 


ratthafn (Pali) 


-^ rat 


9. rat.tha? 


-^ 


rat.th 


10. rat.th 




-> 


ratth. 


1 1 . ratth 




— > 


rat 


12. - 






- 


13. - 






- 



c 



b. saastra (Skt.) — > saat 

9. saas.tra? — > saas.tr 

10. saas.tr — > saastr. 

1 1 . saastr — > saas 

12. saas — > saat 

13. - 



Most simple words (cf. (26' a-b)) and compound lexical items (cf. (26' c) en- 
tered the Thai lexicon with the truncation of the final vowel (and glottal stop) 
(step 9), followed by the resyllabification of the remaining phonetic material (step 
10), cluster simplification (step 11), final stop neutralization (step 12), and tone 
reassignment in accordance with a new syllable structure (step 13) in case a differ- 
ent syllable structure is derived. 

'country' 

Final Vowel (and 7-) Truncation 

Resyllabification 

Final Cluster Simplification 

Final Stop Neutralization 

Tone Reassignment 

'science/art' 

Final Vowel (and 7-) Truncation 

Resyllabification 

Final Cluster Simplification 

Final Stop Neutralization 

Tone Reassignment 

c. rat.tha7 + saas.tra7— > rat.tha saat 'political science' 

9. rat tha saas.tra7 -» rat tha saas.tr F. V (& 7-)Truncation 

10. rat tha saas.tr -^ rat tha saastr. Resyllabification 

1 1 . rat tha saastr -^ rat tha saas F. Cluster Simplification 

12. rat tha saas — > rat tha saat F. Stop Neutralization 
13.- - Tone Reassignment 

6.2.3 Language games involving foreign loans 

Interactions between morpho-phonological structures and stress are most ev- 
ident in tone assignment to foreign loanwords in language games. We will attempt 
at a schematic characterization of Thai tone in language borrowing in the para- 
graph immediately below. (For a full discussion of Thai tonal adaptation of loan- 
words from both tonal languages such as Chinese and non-tonal languages such as 
English and Japanese, the reader is referred to Wong-opasi (in preparation).) 

7. Tones in foreign loanwords 

Gandour 1979 offers an insightful analysis of Thai tones in English loan- 
words in which he attributes the tonal development in English loans to the inter- 
actions between syllable structure and stress patterns of both English and Thai, ill 
Following Gandour, we exemplify the Thai tonal patterns in three classifications ^^ 
of words:monosyllables, bisyllables, and polysyllables. 

7.1 Monosyllables 

Tone is assigned based on the underlying syllable structures of English. 
That is, unchecked syllables receive the M tone while checked syllables acquire 
the H tone, e.g. cream -4 khriim; share — > chee; free -^ frii vs. jet -^ cet; soup —> 
Slip; card -> khdat; cake — > kheek. The underlying structure of English syllables 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 183 

is crucial to understand the counterexamples of a H tone on surface unchecked 
Thai syllables such as bank -^ hetj; pump -^ pan and pipe -^pdy;mouse (as 
in 'computer mouse') -^ maw. In effect, tone is assigned first according to the 
English syllable structures before cluster simplification to avoid violations of 
Thai syllable structure constraints which prohibit word-final clusters containing 
two consonants in a row (*CVCC) or a diphthong followed by a consonant 
I (*CVGC) as the above examples reveal. 

7.2 Bisyllabies 

The intervention of stress in tone assignment is more obvious in words with 
more than one syllables. For penultimate stress which produce a falling tone on 
the final syllable, e.g. visa -^wii sda; party -^ paa tii; doctor — > dok t93, Candour 
provides an explanation in (27). 

(27) The stressed-unstressed English pattern correlates with a falling pitch 
contour. Since Thai rhythm requires that the last syllable in a phrase be 
stressed, it would appear that the falling pitch contour has been pre- 
served in the Thai pronunciation, but that the point of the fall has been 
shifted to the final syllable in accordance with Thai rhythmic con- 
straints. (Candour 1979:137) 

However, the HL pitch pattern is rendered in two different ways in final 
checked syllables, namely, as L or H, e.g. credit -^ khree dit; passport — > phdat sa 
poot vs. bonus — > boo ndt; sandwich — > seen wit. To account for these 
discrepancies. Candour attributes them to competing strategies in tone 
assignment, phonetic vs. non-phonetic motivations like ENGLISH orthographic 
influences. 

In contrast, the correlations between ultimate stress and the mid tone, e.g. 
shampoo -^ chem phuu; hotel — > hoo ten, are described as follows: 

(28) These English source words have an overall rising stress pattern, the 
second syllables being comparatively higher in pitch than the first, and 
longer. If these words were to be adapted with the falling tone of the 
second syllable, the resultant tonal pattern would be considerably dif- 
ferent from the perceived stress pattern. Thus, the final syllables of 
these bisyllabic words are assigned a mid tone which results in a closer 
approximation to the English stress pattern. 

The tonal adaptation in (28) is not without exceptions as given in (29). These 
exceptional tonal behaviors are cited without explanations in Candour 
(1979:139). (The comments in parentheses are made by the present author.) 

(29) soda -^ soodaa *soodaa 
billiard -^ bin liat *bin liat, bin li'at 

necktie -^ nek thay *nek thay (but also nek thay) 

Christmas — > khrit sa maat *khrit sa maat; *khrit sa maat 



1 8 4 Studies in the Linguistic Sciences 23:2 (Fall 1 993) 

7.3 Polysyllables 

Polysyllabic words are assigned tone primarily according to English sylla- 
ble structure, irrespective of the pitch patterns of English stress. 

(30) The rules for tonal assignment are based strictly on the interpretation of 
English syllable structure. Those syllables interpreted as smooth re- 
ceive the mid tone in non-final position, the falling tone in final posi- 
tion; those syllables interpreted as checked receive the high tone in 
non-final position, the low tone in final position. Short open syllables 
in English source words that occur between a primary stressed syllable 
and a following syllable are assigned a mid tone in accordance with the 
tone reduction rule in Thai. Since the tonal patterns remain fixed in the 
adaptation of variable stress patterns found in polysyllabic English 
words, we cannot attribute the resultant tonal patterns to perceptual in- 
terpretation of the variable pitch contours associated with the English 
stress patterns. (Candour 1979:140) 

Again, alongside predictable tonal patterns in computer — > khom phiw Wd 
Chicago — > chi'7 khaa koo; hamburger -^ hem h99 k99 ; lottery -^ lo ta rii, we 
also encounter exceptions such as names of certain countries pronounced with the 
M tone on the final syllable, e.g. America -^ la mee ri kaa; Switzerland — > sa wit 
sa leen. According to Candour, these exceptions are, again, due to a conventional- 
ized reading pronunciation of ENGLISH orthography (present author's emphasis). 

In spite of Candour's keen observations of the influences from the donor lan- 
guage, we would like to elaborate an analysis which gives precedence to the stress 
pattern and syllable constructs of the host language. As seen in section VII above, 
for every generalization made on tone caused by stress and syllable structure of the 
mono-, bi-, and poly-syllabic English loans, there are exceptions which, follow- 
ing Candour, suggests competing strategies in tone assignment. We will first 
comment on the influence of stress. 

Concerning the stress patterns of English, although pre-final stress actually 
corresponds to the falling tone on final unchecked syllables in certain words, no 
conclusive evidence is found as we find both the HL and M tones assigned to the 
final unchecked syllable of bi- and poly-syllabic English words with non-final 
stress, e.g. visa -^wii sda vs. soda -^ soo daa (*soo dda); Chicago — > c hi 1 khaa koo 
vs. America — > ?a mee ri kaa (*7a mee ri kaa). Neither can we find a uniformed 
tone assignment on final unstressed checked syllables, e.g. credit -> khree dit vs. 
sandwich — > seen wit. On the contrary, words with final stress in the English 
source always get the M tone on unchecked syllables, e.g. shampoo -^ chem phuu 
(*chem phitu) while the L tone is invariably assigned to checked syllables, e.g. 
promote —> proo moot. This is due to the fact that the Thai final stress pattern is in 
agreement with that found in English, as pointed out by Candour. We take this to 
reconfirm our already-proven claim for Indic loans that the domain-final syllable 
in Thai is heavy, and that only a long vowel or a short vowel with a glottal stop 
and L tone can surface in the unmarked cases. From this hypothesis, the 
discrepant tone assignments in English follow directly. A digression to a 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 185 

discussion of tone in monosyllables within the framework of the present analysis 
is necessary at this point. 

We claim in section V that the default tone for light or unchecked syllables is 
M and that the unmarked tone for heavy or checked syllables is L. This generaliza- 
tion covers both Thai native words and Indie loans as it is supported by 
Candour's 1982 study of the frequency of tones in Thai words, ranking the fre- 
I quency percentage as follows: 

(31) M (39.98%) L (20.72%) F (17.33%) H (11.81%) R (10.16%) 

(from a total of 61,222 tone occurrences in 25,000+ entries taken from 
Haas' 1964 dictionary) 

The default tone assignment can also be extended to Non-Indic loanwords 
like English and Japanese. Examples are:(Eng.) beer -^ bia; mile — > may; film — > 
film; (Jap.) zen -^sen; yen — > yen: (Eng.) date — > deet: gate — > keet: vote —> wool. 
(Japanese checked monosyllablic loans are rare in Thai.) 

However, there is an additional psychological factor in tonal development in 
foreign loans. That is, speakers of a borrowing language may employ a special tone 
to indicate that the word is clearly a foreign item in the lexicon of the donor lan- 
guage as reported in Kiu 1977. In the case of borrowing in Thai, unlike Indie 
loans which entered the Thai lexicon at a very early stage during the formation of 
the Thai language and have thus become an integral part of Thai vocabulary, for 
later borrowings from non-Indic sources, the Thai speakers feel the need to signal 
the foreign nature of these lexical words. Hence, there is a tendency to assign more 
marked tones to non-Indic loans. We hypothesize that these marked tones are the 
high and the rising tones (cf. the tone frequency scale in (31)). The H tone is an 
across-the-board assignment to all checked syllables as an attempt to neutralize the 
variable tones (i.e. L, HL, and H) conditioned by the underlying syllable struc- 
tures in English, and by the effects of tone sandhi, i.e. tone change due to neigh- 
boring sounds, in the Chaozhou dialect of Chinese (see below). The rising tone, on 
the other hand, applies to unchecked syllables in Chaozhou, as another indicator, 
to distinguish the Chinese source since Thai and Chinese syllable structures are 
fairly similar. 

Apart from external linguistic factors, the marked H tone is also based on 
phonetic grounds. As a result of tonogenesis from the Indie phonetic inventory, 
the majority of Thai consonants belong to the Low-consonants class, bearing the 
underlying [-vc, +sprdj phonetic features (cf. section III and the appendix). From 
) the point of view of syllable structure in (5), we witnessed that both the CVC and 
CVVC structures can carry the L tone, while other tones are more restricted. The 
M tone is impossible due to the well-formedness condition in (17a), while the se- 
quence LH never results in checked syllables because the tone imposed by the 
[-continuant] feature of the coda is the L tone (cf. (13)). The HL tone is barred from 
CVC syllables because the vowel is too short to bear a sequence of two tones in 
Thai. This leaves two possibilities, either the H or the L tone. While the Low-class 
consonants, including sonorants, have more restrictions than the other two classes 
of consonants, that is, thev cannot bear a L tone when closed by a stop due to the 



186 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

underlying [-vc, +sprd] features of the onset (cf. (11)). Thus, the H tone is as- 
signed uniformedly to all checked monosyllables despite the three options L, HL, 
and H, e.g. jet — > cet; soup — > siip; card — > khdat; cake — > kheek. This 
psychological inclination also competes with the natural phonetic motivation, as 
evidenced in the conflicting tone assignments, L vs. H, in AIDS — > '7eet. vs. Ed 
— > et; vote — > woot vs. Vogue —> wook. The tendency for the high tone <- 
assignment to non-Indic loans is also reflected in Chinese loans, e.g. pe? -^ pe ? f | 
'eider paternal uncle'; kok -^ kok 'country', despite a possible direct tone transfer 
by preserving the original tone in Chinese. This is due to the fact that Chinese 
lexical words are subject to tone sandhi. Consequently, kok 'country' is 
pronounced as kok in isolation but as kok in kok udij 'king' (<- country+king). 
Regarding unchecked monosyllables, since some English syllable structures, e.g. 
peculiar onsets, fr-, fl-, sp-, st-, sk-, v-, or coda, -r, -1, -nk, -nt, -mp, etc., are 
sufficiently foreign to the Thai ears, the M tone is assigned without further needs 
to mark the non-native source. However, this strategy may compete with the 
preference for the H tone and both tones are found idiosyncratically, e.g. cent -^ 
sen vs. saint — > sen (in 'Saint John School'), but also sen (in 'Saint Joseph 
School'). 

The strategy to differentiate foreign loans with a distinct tonal pattern from 
those found in the majority of words in the borrowing language persists in poly- 
syllabic words as well. However, in polysyllabic words, there is another phonetic 
requirement, namely, stress. Contrary to Candour, we claim that the Thai stress 
pattern plays a major role in tone assignment than the English stress patterns. 
Specifically, English final stress can fit into the Thai final stress pattern easily, re- 
sulting in M tone for CVV structure and L for CV(V)C, and no tone adjustment is 
needed. Pre-final English stress, in contrast, calls for certain adaptations. This is 
because unstressed syllables in English are normally shortened to a schwa. 
However, the final stress pattern in Thai requires that the vowel of the final sylla- 
ble may not be reduced. As in the Indie cases, the vowel in question must be 
lengthened and assigned the M tone as regulated by the syllable structure re- 
quirements, e.g. India -^ 7in dia, or be imposed a falling tone because of the 
psychological need to signal the foreign stress pattern, e.g. chi7 khaa koo, with 
the HL sequence combined on the last syllable. Otherwise, the shortened vowel in 
English must be assigned a glottal stop which entails the L tone, e.g. khaa buu 
ki7, with the split HL effect on two syllables, since the perceived overall pitch 
contour ML of the word approximates HL. The two competing strategies 
sometimes coexist as we find two pronunciations for proper names such as Ithaca 
-> 7it tha kda vs. 7it tha kd7; Amko -> 7am koo vs. 7am kd7, and for other not U, 
well-established loanwords. The proposition that the stress contour in English is 
of minor importance to the Thai stress pattern is also corroborated by the fact that 
the English stress pattern must be prohibited when it is in opposition to Thai 
phonology. Specifically, the HL pitch contour of English cannot be retained 
when the last syllable is CVC because of the shortness of the vowel. Therefore, 
only the L tone is retained (which still preserves the overall falling pitch of 
English), e.g. promote -^ proo moot, or in the case of a sonorant onset, only the H 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 187 

tone is integrated on CVC, e.g. lennis -^ then nit, while HL is not ruled out on 
CVVC, e.g. Christmas — > khrit sa mdat, as in the case of Indie loans (of. (19)). 

8. Evidence from language games involving foreign loans 

It is clear that tone is governed by syllable structure, stress, and the phonetic 
properties of each syllabic segment as some resulting tones in the language games 
may not be there underlyingly. In (32), all of the syllables in the source words are 
assigned tone according to rules (30). It is possible that the H M HL sequence of 
chi7 khaa koo be inverted following rime movement (i.e. choo khaa ki7). 
However, HL tone preservation is impossible when the rime is turned into a short 
heavy syllable {choo khaa *ki7). To preserve the tones, syllable rimes must be 
adjusted (cho? khaa kii). However, the preferred form is choo khaa /:?? since both 
the new tones and rimes are in agreement with the Thai stress patterns. Likewise, 
khaa huu ki7 may be rendered khi7 hiiii kaa. Nevertheless, the mid tone can never 
be retained on short heavy syllables (*^/z/? /?«« *kaa) according to the well- 
formedness condition (17a). The preferred syllable structures for bearing M tone 
is a long light syllable (khii), and for L tone in final position, it is a short heavy 
syllable {ka7), i.e. khii huu ka7. The best preferred form, however, is khi7 huu 
ka7. 



(32) 


a. chi7 khaa koo 




-^ choo khaa ki7 








T 


T 


T 




T 




T 


T 


/ 


1 \ / 


1 \ 


/ 1 


\ 


/ 1 \ 


/ 


1 \ 


/ 1 \ 


H) 

1 


M L) H 

1 1 1 


M M) 
1 1 


M M 

1 1 


M)^ 

1 


>H M M)H 

1 1 1 1 


M M)M M L) 

1 1 1 1 1 


1 
ch) 


1 1 1 
i 7) kh 


1 1 
a a) 


1 1 
k o 


1 

0) 


1 1 1 
ch o) 


kh 


1 1 
a a) 


k i 7) 


[-vc][+ct][-ct][-vc][ 


+ct][+ct] 


[...][+ct][+ct] 1 


-vc][+ct][+ct] 1 


[-vc][+ct][+ct] 


[...][+ct][-ct] 


1 


\ / 1 


\ / 


1 \ 


/ 


1 \ / 


1 


\ / 


1 \ / 


O 


R O 


R 


O 


R 


R 


O 


R 


O R 


\ 


/ \ 


/ 


\ / 




\ / 


\ 


/ 


\ / 


( 


J 


o 


o 




a 




a 


o 


Derived Tones: 


H M (M->) HL (by sentential stress) -^ 


•M M 


L 




b . khaa buu kiV 




-^ 


khi7 buu kaa 








T 


T 


T 




T 




T 


T 


/ 


1 \ / 


1 \ 


/ 1 


\ 


/ 1 \ 


/ 


1 \ 


/ 1 \ 


H 

1 


M M) M 

1 1 1 


M M)M M 

1 1 1 1 


D^H) M L) 

1 III 


M 

1 


M M)M M L) 

1 1 1 1 1 


1 
kh 


1 1 1 
a a) b 


1 1 
u u) 


1 1 
k i 


1 ill 
7) kh) i 7) 


1 
b 


1 1 
u u) 


1 1 1 
k a 7) 


[-VC] [+ct][+ct] [...][+ct][+ct] [...][+ct][-ct] 1 


[-vc][+ct][-ct] 


[...]| 


[+ct][+ctl 


[...l[+ct][-ct] 


1 


\ / 1 


\ / 


1 \ 


/ 


1 \ / 


1 


\ / 


1 \ / 





R O 


R 


O 


R 


O R 


O 


R 


O R 


\ 


/ \ 


/ 


\ 


/ 


\ / 


\ 


/ 


\ / 




a 


o 


a 




a 




a 


o 


Derived Tones: 


M 


M 


L 


-^ (HL^) 


H 


M 


L 



188 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

9. Conclusion 

We hope to have presented an extensively convincing analysis of tone as- 
signment in Thai, under the assumptions that tonal assignment in Thai is modu- 
lar, having a cyclic application, and that it must conform to the well-formedness 
conditions (17) at each stage of derivation. The modular tone assignment accounts 
for tone change at morpheme junctures, word boundaries, and within syllables. A\ 
Through the cyclic application. Yip's stipulation of the prohibition of three-tone Vlj. 
syllables can be dispensed with. The present study translates the three canonical 
tonal consonant classes (i.e. M, L and H) into morphophonological marking of 
underlying phonetic features resulting from historical sound change. We point 
out the importance of the recognition of tonal consonant classification as it facili- 
tates the acquisition of tonal patterns in Thai. Without such marking in the lexi- 
con, we would have no way to derive varying tonal patterns in Indie loanwords. 
We would also have to devise various complicated tonal rules which at times may 
fail to yield the correct results. Moreover, this morphophonological marking is 
needed independently for all lexical items in Thai since Thai tones are lexical but 
are constrained by the segmental properties and the structure of each syllable, and 
for loanwords, the stress patterns of both the donor and the borrowing languages 
as well, although the stress pattern of the borrowing language takes precedence. 

Regarding syllable structure, the relevance of syllable weight in the opera- 
tion of phonological rules is proven in the literature (for stress, see Halle & 
Vergnaud 1987; Wong-opasi 1987, etc.) For tone, the light syllable weight al- 
lows the syllable more flexibility to carry any of the five tones, including contour 
tones. Because syllables closed with a stop segment involve glottalization, it cre- 
ates extra weight on these syllables, and hence it restricts HEAVY syllables from 
carrying certain tones. Unless intervened by some special phonetic properties of 
the onset, heavy syllables are assigned L tone; thereby, excluding the default M 
tone (*M). The shortness of vowels in heavy syllables accounts for their inability 
to accommodate both the extra syllable weight, imposed by the Thai final stress, 
and a contour tone (i.e. *HL or *LH on stressed CVC syllables), except when the 
contour tone is intonationally derived or in some marginal lexical words. Further, 
postulation of the [-cont] feature as causing L tone explains the absence of rising 
tone (*LH) on all Thai heavy syllables, long or short, despite its presence in other 
languages. This is due to the fact that the resulting tone would have the L tone, 
never the H tone, at the end of the contour tones. Aside from morphophonological 
factors, psychological factors do influence tone assignment as well since Thai has 
special tone patterns to mark English, Chinese, and Japanese loanwords from n\ 
Indie loanwords and Thai native words. ^ 

NOTES 

' I wish to thank Jack Candour for kindly sending me his various papers on 
Thai. Special thanks are also due to Martha Ratliff for editing an earlier version of 
this paper. All errors of interpretation, however, are my own. Last, but not least. 



Wong-opasi: The inteqDlay between tone, stress, and syllabification in Thai 189 

this version of the paper has included substantial changes since its presentation at 
the first annual meeting of SEALS. 

~ The phonetic transcription in this study may be different from those em- 
ployed in the papers being discussed. Specifically, off-glides are represented as [y, 
w] whereas on-glides are written as [i] and [u], respectively. All long vowels are 
transcribed as sequences of two identical vowels while short vowels are single 
vowels followed by a glottal stop except when eliminated by some phonological 
rule. All tones in the phonetic transcription are written on the first vowel irre- 
spective of the gliding or vocalic properties of the high segment. Such tone values 
as M, L, H, HL, and LH on the examples are given as necessary. 

^ Another example of Leben's is tStj kaan — > tS) kaan 'want' is ques- 
tionable since underlyingly, the first lexical item contains a long vowel and it is 
shorthened tooij kaan -^ totj kaan. 

"^ For more details, the reader is invited to read the full discussion in her pa- 
per. 

5 The development of a high tone as in (12) is also attested in the evolution of 
Punjabi from Sanskrit (Bhatia 1975; Hock 1985). 

^ Phoneticians are invited to check out the validity of this hypothesis from 
laboratory analyses of the actual Thai isolative speech style. 

"7 Along this line of hypotheses, David Strecker (in Comrie (1987:753)) dis- 
cusses the development from Proto-Tai to the modern Tai languages. He states that 
the Thai words for 'face' and 'mother's younger sibling' had the same tone but dif- 
ferent initials, namely a voiceless vs. voiced nasal, respectively, whereas in mod- 
em Thai they have the same initial but different tones as falling vs. high tone, in 
that order. 

8 As predicted, the bracket of the first tonal domain is erased in light sylla- 
bles with devoiced aspirated onsets while the first tone is retained in light sylla- 
bles with original aspirated onsets. This is due to the fact that tone is assigned in 
the first case before devoicing and aspiration apply while in the second case, it is 
assigned directly to the intrinsic [-i-spread] property of original aspirated onsets. 

9 These final vowels (plus an applicable nasal) are Indie declension endings. 
They were deleted in the majority of cases in the Thai adoption (Phanthumetha 
1975:26-47). 

'OThis lexical word is given with a final [m] in HRH. Prince Kitiyakara 
Krommaphra Chandaburinarunath's dictionary. However, according to 
Phanthumetha (1975:68), the ending for Pali is [13] while it is [m] for Sanskrit. 



190 



Studies in the Linguistic Sciences 23:2 (Fall 1993) 
Appendix: From Indie Consonant Inventory to Thai 



ITonal Consonants 


Mid 
(-ve, -spr) 

n 

k 
k 


1 High 1 
l(-vc, +sprd)l(+vc 
1 \i 1 
1 kh 1 
1 kh 1 


Low 
, -sprd)l(+vc, +sprd)l 
R 1 7J 

g 1 gh 
kh 


1 


1 stops: 

j Thai orthography: 
1 Indie velars: 
J Thai velars: 


(+nas) 1 
1 
1 
1 


1 Thai orthography: 
1 Indie palatals: 
1 Thai palatals: 


c 
e 




eh 
eh 




XI 
J 


1 

1 

ch 


QJ 
jh 


fD i 

h 1 

y 1 


1 Thai orthography: 
1 Indie retroflexes: 
1 Thai dentals: 


t 
t 




1 

th 

th 




n 

d 


1 

1 

th 


RJ 
dh 


Hi 1 
n ] 
n ] 


1 Thai orthography: 
1 Indie dentals: 
1 Thai dentals: 


PI 

t 
t 




th 

th 




d 


1 

1 

th 


dh 


U 1 
n 1 
n i 


1 Thai orthography: 
1 Indie labials: 
1 Thai labials: 


u 
p 
p 




ph 
ph 




Ul 
b 


1 

1 

ph 


/I 
bh 


U 1 
m 1 
m 1 


1 Fricatives: 1 
1 1 Hig 
1 Thai ortho.l f\ M 
1 Indie 1 s s 
1 Thai 1 


+strident) 

il 1 Low 
r( 1 M 
s 1 - 
s 1 s 


1 (+glottal/laryngeal) 
1 High 1 Low 

1 w 1 e 

1 h 1 - 
1 h 1 h 


(+labiodental) ! 
High 1 Low 1 

cJ 1 vJ 1 

- 1 - 1 
f 1 f 1 


1 Glides: 1 
1 Thai ortho.l ti 
1 Indie 1 y 
1 Thai 1 y 


Low 

V 

w 




1 
1 
1 
1 


Liquids: 

r r rh 
r 


Li 


3W 

) U 
1 1 

1 


1 
1 
1 
1 



Thai invention: <Q R (d) from Indie (t), (t), respectively; 6 (b) from(p); 
and syllable initial t) (7) 



REFERENCES 

BAO, Z. M. 1990. On the nature of tone. Massachusetts Institute of Technology, 

Ph. D. dissertation in Linguistics. 
BHATIA, Tej. 1975. The evolution of tones in Punjabi. Studies in the Linguistic 

Sciences. 5:2.12-24. 
BUDDHADATTA, A. P. 1968. Concise Pali-English dictionary. Colombo: The 

Colombo Apothecaries. 



Wong-opasi: The interplay between tone, stress, and syllabification in Thai 191 

BURZIO, Luigi. 1991. Principles in phonology. Paper presented at the annual 

meeting of the Linguistic Society of America, Chicago. 
CHENG, C. C. 1973. A quantitative study of tone in Chinese. Journal of Chinese 

Linguistics. 1:93-110. 
CLEMENTS, George. 1985. The geometry of phonological features. Phonology 

Yearbook 2:225-252. 
DUANMU, San. 1990. Formal study of syllable, tone, stress and domain in 

Chinese languages. Massachusetts Institute of Technology Ph. D. disserta- 
tion in Linguistics. 
CANDOUR, Jack. 1974. On the representation of tone in Siamese. UCLA 

Working Papers in Phonetics 27:1 18-146. 
. 1979. Tonal rules for English loanwords in Thai. Southeast Asian 

Linguistic Studies 4:131-144. 
. 1982. The relative frequency of tones in Thai. Pacific Linguistics, Series A- 

No. 62: Papers in South-East Asian Linguistics 8: Tonation. David Bradley 

(Ed.), 155-159. 
HAAS, M. R. 1964. Thai-English student's dictionary. Stanford: Stanford 

University Press. 
HALLE, Morris, & K. N. Stevens. A note on laryngeal features. R.L.E. Quarterly 

Progress Report. 101:198-213. 
, & Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge: The MIT 

Press. 
HAUDRICOURT, A. G. 1961. Bipartition et tripartition des systemes de tons dans 

quelques langues d'Extreme Orient. Bulletin de la Societe Linguistique de 

Paris 56:163-180. 
HENDERSON, E. J. A. 1949. Prosodies in Siamese. Asia Major (New Series) 

1:189-215. (Reprinted in Prosodic Analysis , F. R. Palmer (Ed.). London: 

Oxford University Press, 1970). 
. 1967. Grammar and tone in south east Asian languages. Wissenschaftlische 

Zeitschrift der Karl-Marx-Universitiit. Leipzig 16:171-8. 
HOCK, Hans H. 1986. Principles of Historical Linguistics. Amsterdam: Mouton 

de Gruyter. 
HOMBERT, J.-M. 1978. Consonant types, vowel quality, and tone. Tone: A lin- 
guistic survey. Victoria Fromkin (Ed.). N.Y.: Academic Press, 77-1 1 1. 
JEEL, V. 1975. An investigation of the fundamental frequency of vowels after var- 
ious Danish consonants, in particular stop consonants. Annual Report of the 

Institute of Phonetics, University of Copenhagen. 9:191-21 1. 
KIU, K. L. 1977. Tonal rules for English loan words in Cantonese. Journal of the 

International Phonetic Association. 7/1:17-22. 
LEA, W. A. 1972. Intonational cues to the constituent structure and phonemics of 

spoken English . Purdue University Ph.D. dissertation in Linguistics. 
LEBEN, W. 1971. On the segmental nature of tone in Thai. R.L.E. Quarterly 

Progress Report. 101:221-4. 
. 1973. Suprasegmcntal Phonology . Massachusetts Institute of Technology, 

Ph.D. dissertation in Linguistics. 



192 Studies in the Linguistic Sciences 23:2 (Fall 1993) 

OHALA, John J. 1973. The physiology of tone. In L. Hyman (Ed.), Consonant 
types and tone. Southern California Papers in Linguistics. 1:1-14. 

. 1978. Production of tone. Tone: A linguistic survey. Victoria Fromkin 

(Ed.). N.Y.: Academic Press, 5-39. 

PEYASANTIWONG, Patcharin. 1986. Stress in Thai. Papers from a Conference on 
Thai Studies in Honor of William J. Gedney. Robert J. Bickner, et al. (Eds.), 
211-230. (Originally presented at the 13th International Conference on Sino- 
Tibetan Languages and Linguistics, University of Virginia, October 1980.) 

PHANTHUMETHA, Banchop. 1975. Bali Sanskrit nai phasaa Thai. Bangkok: 
Ramkhamhaeng University Press. 

SAGEY, Elizabeth. 1986. The Representation of Features and Relations in 
Nonlinear Phonology. Massachusetts Institute of Technology, Ph.D. disser- 
tation in Linguistics. 

STRECKER, David. 1987. Tai languages. The World's Major Languages. Bernard 
Comrie (Ed.). Oxford: Oxford University Press, 747-775. 

SURINTRAMONT, A. 1973. Some aspects of underlying syllable structure in 
Thai:evidence from KHAMPUAN-a Thai word game. Studies in the 
Linguistic Sciences. 3:1.121-142. 

HRH. Prince Kitiyakara Krommaphra Chandaburinarunath. 1969. Pali-Thai- 
English-Sanskrit Dictionary. Bangkok: Mahaamakutaraajavidyaalaya King 
Mahaamakutaraajavidyaalaya. 

TUMTAVITIKUL, A. 1991. Perhaps, the tones are in the consonants? Paper pre- 
sented at the annual meeting of the Linguistic Society of America, Chicago. 

WAROTAMASIKHADIT, U. 1967. Some phonological rules in Thai. Journal of the 
American Oriental Society. 87:4.541-574. 

WONG-OPASI, U. 1987. Lexical Phonology and the Spanish Lexicon. 
Bloomington: Indiana University Linguistics Club (1990). 

. (Forthcoming). An autosegmental account of tone and stress in language 

contact. MS. 

YIP, Moira. 1980. The tonal phonology of Chinese. Bloomington: Indiana 
University Linguistics Club. 

. 1982. Against a segmental analysis of Zahao and Thai: A laryngeal tier pro- 
posal. Linguistic Analysis 9:1.79-94. 



Studies in the Linguistic Sciences 

Volume 23, Number 2, Fall 1993 [publ. October 1996] 



ADJECTIVAL AND DETERMINATIVE MEASURE PHRASES 
AND NP INTERPRETATIONS IN MANDARIN CHINESE* 

Mary A. Wu 

In Mandarin Chinese (MC), the order of prenominal elements 
such as measure phrases (MP) and modifying phrases (MOD-de) corre- 
lates with different readings of the NP relative to the definite and indef- 
inite distinction (Annear 1965; J. Huang 1982, 1983). In this paper, I 
show first that definite and indefinite readings of NPs correlate not only 
with the order of MP and MOD-de but also with the syntactic category 
that combines with de. Secondly, I argue that the order-related definite- 
ness facts fall out from Partee's 1987 assumption that numerals, hence 
MPs, may be adjectival or determinative, and from the assumption that 
NPs in MC may have a null determiner which, in principle, can be in- 
terpreted as definite or as indefinite. 

1. Introduction: order-related deflnite interpretations 

Mandarin Chinese (MC) lacks a word for the definite article. Definiteness of 
MC NPs may be signaled by the order of words (more precisely, of constituents) 
within the NPs. The relation between word order and definiteness in MC NPs is 
the subject of this paper. 

In MC NPs, measure phrases such as liang-tou 'two head (of)' occur 
prenominally.' MC NPs may also contain prenominal modifying phrases marked 
by the particle de, like for example hen xin de 'very new de . I will refer to 
modifiers of this form as MOD-de and Measure Phrases as MP. MP and MOD-de 
may be ordered as in (1) and (2) below. 

(1) MP MOD-de N' 

(2) MOD-de MP N' 

The order of MP and MOD-de appears to correlate with different readings of 
the NP (Chao 1968; Annear 1965; J. Huang 1982:152-153, 1983).2 For example, 
in (3), MP precedes MOD-de and the NP may be interpreted as indefinite or defi- 
nite, but in (4), MOD-de precedes MP and the NP has a definite reading only. (c7 = 
CLASSIFIER in the gloss.) 

(3) [np MP MOD-de N' ] 

Liang-tang shangwu qu Badaling de lieche yijing fachuqule. 

two-cl morning go Badaling de train already dispatched 

'Two/The two trains going to Badaling in the morping have already been 
dispatched.' 



194 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

(4) [np MOD-de MP N' ] 

Shangwu qu Badaling de liang-tang Heche yijing fachuqule. 

morning go Badaling de two-cl train already dispatched 

The two trains going to Badaling in the morning have already been 
dispatched.' 

While the correlation of the definite reading with the order MOD-de preced- 
ing MP has been observed before in the literature, it has been described as a ten- 
dency (J. Huang 1983:51-52) or associated with a demonstrative in the post-MOD- 
de MP (Chao 1968, Annear 1965). There is evidence, however, that the correlation 
may be more than just a tendency. It systematically holds for NPs containing cer- 
tain types of MOD-de but not for certain others, and it is not limited to cases with 
demonstratives in the MP. For instance, the order MOD-de MP is associated with 
the definite interpretation in (4), in which a VP-de precedes the MP. But indefinite 
readings are quite common where MOD-de precedes MP if MOD-de contains a 
possessive NP or an NP/oc , i.e., an NP followed by a locational particle such as // 
'in', shang 'on', etc. (Li & Thompson 1981:391).^ As in (5a) and (5b), both indefi- 
nite and definite readings of the larger NP are possible even though MOD-de 
precedes MP. 

(5) a. [np MOD-de MP N' ] 

dashan li de liang-ge xuesheng 

mountain in de two-cl student 

'two/the two students in the mountain(s)' 
b. [np MOD-^e MP N' ] 

Zhangsan de liang-zhi shouzhi 

Zhangsan de two-cl finger 
'two/the two fingers of Zhangsan's' 

An adequate account of Mandarin NPs should thus explain not only why the 
order MOD-de MP correlates with the definite-only reading in (4) but also why the 
same order fails to produce a definite-only reading for the NPs in (5). 

Example (6) with a bare Noun shu 'book(s)' shows that in the absence of an 
MP and MOD-de, both definite and indefinite readings are in principle possible. 
(part = SENTENTIAL PARTICLE in the gloss.) 

(6) Lisi maile [np shu ] le 
Lisi bought book part 
'Lisi has bought a book/some books.' or 
'Lisi has bought the book(s).' 

Sentence (7) says that Lisi is informed of the fine quality of cake in the 
dining-hall by a student/some students/the student(s) who work there, which 
shows that an NP containing a MOD-de but no MP may have both definite and in- 
definite readings as well. 



Wu: Adjectival and determinative measure phrases 195 

(7) [np Zai canting dagong de xuesheng ] gaosu Lisi nali de dangao bucuo. 

at dining-hall work de student tell Lisi there de cake not-bad 
'A/Some student(s) working at the dining-hall told Lisi that cake there is 
pretty good.' 

'The student(s) working at the dining-hall told Lisi that cake there is pretty 
good.' 

MOD-de may also precede or follow quantificational phrases (QP) contain- 
ing quantificational words such as mei 'every', as in (8), but the difference in the 
order of MOD-de and mei-tang 'every-cl', on a superficial examination, does not 
seem to correlate with a difference in truth-conditions. 

(8) a. 



[np QP 


MOD-de 




N'] 




mei-tang 


shangwu qu 


Badaling de 


Heche 


zai Shahezhen 


every-cl 


morning go 


Badalingde 


train 


at Shahezhen 


ting wu 


-fenzhong. 








stop five-minute 








'Every train going to Badaling 


in the morning stops at Shahezhen for five 


minutes.' 










[np UOD'de 




QP 


N'] 




Shangwu 


qu Badaling de 


mei-tang 


lieche 


: zai Shahezhen 


morning 


go Badaling de 


every-cl 


train 


at Shahezhen 



ting wu-fenzhong. 

stop five-minute 
'Every train going to Badaling in the morning stops at Shahezhen for five 
minutes.' 

While the focus of this paper is on the interpretation of NPs containing nu- 
merals and MOD-^^, an adequate account needs to be able to account for cases 
like (6), (7) and (8) as well. 

In the following I first show that when MOD-de precedes MP in the NP, 
definite-only readings of the NPs pattern with a number of syntactic categories in 
MOD- J^, whereas when MP precedes MOD-de, both definite and indefinite read- 
ings are in principle possible with all categories in MOD-de. I then propose an 
analysis to account for the patterns observed. I need to point out that the general- 
izations argued for in this paper do not carry over to NPs in predicative positions 
or with emphatic stress. For example, an NP like a student in Lisi is a student, or 
NPs with emphatic stress, e.g., shangwu qu Badaling de LIANG-TANG lieche 
'TWO trains going to Badaling in the morning', may not fit the patterns here. 

2. The syntactic category combining with de and definite- 
only readings of the NPs 

2.1. [np MOD-de MP N'] with definite-only readings 

The examples in (9) illustrate the kind of contrasts used in determining 
whether an NP has only a definite reading or not. By (9a), there are four trains go- 
ing to Badaling in the morning. Adding (9b) to (9a) means two out of the four 
trains going to Badaling in the morning have already been dispatched. But adding 



196 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

(9c) results in an infelicitous discourse. The only reading for (9c) is that there are 
just two trains going to Badaling in the morning, and this is at odds with (9a), 
which states that there exist four such trains. The infelicity resulting from continu- 
ing (9a) with (9c) makes it clear that the NP in (9c) with WP-de preceding MP has 
a definite reading only. 

(9) a. Shangwu you si-tang Heche qu Badaling. 
morning have four-cl train go Badaling 
'In the morning there are four trains going to Badaling.' 

b. [np MP MOD-de N' ] 

Liang-tang shangwu qu Badaling de lieche yijing fachuqule. 

two-cl morning go Badaling de train already dispatched 

'Two trains going to Badaling in the morning have already been 
dispatched.' 

c. #[np MOD-de MP N' ] 

Shangwu qu Badaling de liang-tang lieche yijing fachuqule. 

morning go Badaling de two-cl train already dispatched 

'The two trains going to Badaling in the morning have already been 
dispatched.' 

If we have (9a') instead of (9a), then both (9b) and (9c) may follow (9a') fe- 
licitously. The subject NPs in both (9b) and (9c) may be understood as anaphori- 
cally related to the two trains mentioned in (9a'), i.e., they may have the definite 
reading 'the two trains going to Badaling in the morning'. This again shows that 
NPs with the order [f^ WP-de MP N'] have definite readings. It also shows that 
NPs with the order [np MP WP-de N'] may have definite readings as well. 

(9) a'. Shangwu you liang-tang lieche qu Badaling. 

morning have two-cl train go Badaling 
'In the morning there are two trains going to Badaling.' 
b. [np MP MOD-de N' ] 

Liang-tang shangwu qu Badaling de lieche yijing fachuqule. 

two-cl morning go Badaling de train already dispatched 

'The two trains going to Badaling in the morning have already been 
dispatched.' 

In (10), MOD-de is an NP-de which is not a possessive. According to (10a), 
five of the ten birds in the cage have black tails. (10b) may follow (10a) felici- 
tously, but (10c) is infelicitous after (10a). (10c) may be used felicitously only if 
the number of black-tailed birds that are salient in the context is just two, i.e., it 
has a definite reading only. 

(10) a. Longzi h you shi-zhi niao, qizhong wu-zhi weiba shi hei de. 

cage in have ten-cl bird among-which five-cl tail be black de 
'In the cage are ten birds, among which five have tails that are black.' 
b. [np Liang-zhi hei weiba de niao ] hai hen xiao, bu hui fei. 
two-cl black-tail de bird still very young not can fly 
'Two birds with black tails are still very young and cannot fly.' 



Wu: Adjectival and determinative measure phrases 197 

c. #[np Heiweiba de liang-zhi niao ] hai hen xiao, bu hui fei. 
black-tail de two-cl bird still very young not can fly 
'The two birds with black tails are still very young and cannot fly.' 

If we replace (10a) with (10a'), then both (10b) and (10c) follow (10a') felici- 
tously. In particular, it is possible to understand the NPs as referring back to the 
two black-tailed birds mentioned in (10a'). Therefore, [np ^P-de MP N'] has only 
a definite reading while both definite and indefinite readings are possible with [np 

M?NP-deN']. 

(10) a'. Longzi li you shi-zhi niao, qizhong liang-zhi weiba shi hei de. 

cage in have ten-cl bird among-which two-cl tail be black de 
'In the cage are ten birds, among which two have tails that are black.' 
b. [np Liang-zhi heiweiba de niao ] hai hen xiao, bu hui fei. 
two-cl black-tail de bird still very young not can fly 
'The two birds with black tails are still very young and cannot fly.' 

2.2. [np MOD-de MP N'] with deflnite and indefinite readings 

2.2.1. NP-de MP N' (with possessive NP-de) 

Example (1 1) is similar to (10) in the sense that MOD-de is NP-de. Unlike in 
(10), however, NP-de is now possessive. Sentence (11a) says that five of the ten 
birds in the cage are Zhangsan's. Both (lib) and (lie) may follow (11a) felici- 
tously with the interpretation that there are two birds that are Zhangsan's that are 
still very young and cannot fly. In either case, Zhangsan is assumed to have more 
than two birds, given (11a). Therefore, unlike the cases of [np NP-de MP N'] ob- 
served in (10) with a non-possessive NP-de, [np NP-de MP N'] with a possessive 
NP-de has a reading that is not definite here. 

(11) a. Longzi li you shi-zhi niao, qizhong wu-zhi shi Zhangsan de. 

cage in have ten-cl bird among-which five-cl be Zhangsan de 
'In the cage are ten birds, among which five are Zhangsan's.' 

b. [np Liang-zhi Zhangsan de niao ] hai hen xiao, bu hui fei. 

two-cl Zhangsan de bird still very young not can fly 
'Two birds that are Zhangsan's are still very young and cannot fly.' 

c. [np Zhangsan de liang-zhi niao ] hai hen xiao, bu hui fei. 

Zhangsan de two-cl bird still very young not can fly 
'Two birds that are Zhangsan's are still very young and cannot fly.' 

If we replace (11a) with (11a'), then both (lib) and (1 Ic) may follow (11a') 
felicitously with a definite reading referring back to the two birds that are 
Zhangsan's in (1 la'). 

(11) a'. Longzi li you shi-zhi niao, qizhong liang-zhi shi Zhangsan de. 
cage in have ten-cl bird among-which two-cl be Zhangsan de 
'In the cage are ten birds, among which two are Zhangsan's.' 
b. [np Liang-zhi Zhangsan de niao ] hai hen xiao, bu hui fei. 
two-cl Zhangsan de bird still very young not can fly 
'The two birds that arc Zhangsan's are still very young and cannot fly.' 



198 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

c. [np Zhangsan de liang-zhi niao ] hai hen xiao, bu hui fei. 
Zhangsan de two-cl bird still vei^ young not can fly 
'The two birds that are Zhangsan's are still very young and cannot fly.' 

Cases like (12) provide further evidence that [np NP-de MP N'] with the pos- 
sessive NP-de may have indefinite interpretations. (12a) with [np MP NP-de N'] 
seems to convey a sense of contrast, e.g., Zhangsan's fingers compared to someone 
else's; (12b) [np ^P-de MP N'] lacks this contrastive implication and sounds more 
natural than (12a) in non-contrastive contexts. 

(12) a. #[np San-zhi Zhangsan de shouzhitou ] youdian zhong. 

three-cl Zhangsan de finger somewhat swollen 

'Three fingers that are Zhangsan's are somewhat swollen.' 
b. [np Zhangsan de san-zhi shouzhitou ] youdian zhong. 
Zhangsan de three-cl finger somewhat swollen 

'Three fingers of Zhangsan's are somewhat swollen.' 

More needs to be said about the contrastive implications of (12a). For the 
purpose of the present discussion, the point is that (12b) may be used felicitously if 
Zhangsan has ten fingers as people normally do, whether any of them are salient in 
the context or not. This should not be possible if [np NP-de MP N'] had a definite 
reading only. Such a reading would require Zhangsan to have exactly three fingers 
or to have three that are salient in the context, as is the case with the English 
possessive in (13). 

(13) John's three fingers were injured in an accident. 

In (13), John's three fingers is definite and may be used felicitously only if 
John has exactly three fingers or three of his fingers are salient in the context. 
Therefore, [np NP-de MP N'] with a possessive NP-de has a reading that is not 
definite in (12b). 

2.2.2. NPioc-de MP N' 

In (14), MOD-de is NPioc-de. According to (14a), there are six pots of 
flowers in the yard and six in the house. Both (14b) and (14c) may follow (14a) 
felicitously. This shows that both [np MP NP i„,-de N'] and [np NP loc-de MP N'] 
may have an indefinite reading. 

(14) a. Yuanzi li he wuzi li ge you liu-pen hua. 

yard in and house in each have six-cl flower 

'There are six pots of flowers in the yard and in the house respectively.' 

b. [np Wu-pen yuanzi li de hua ] hen haokan. 

five-cl yard in de flower very pretty 
'Five pots of flowers in the yard are very pretty.' 

c. [np Yuanzi li de wu-pen hua ] hen haokan. 

yard in de five-cl flower very pretty 
'Five pots of flowers in the yard are very pretty.' 

If we replace (14a) with (14a'), then (14b) and (14c) may both follow (14a') 
felicitously with a definite reading referring back to the five pots of flowers men- 



Wii: Adjectival and determinative measure phrases 199 

tioned in (14a'). Therefore, both definite and indefinite readings are possible for 
[np ^Pioc-de MP N'] and [np MP NP i^c-de N']. 

(14) a'. Yuanzi li he wuzi li ge you wu-pen hua. 

yard in and house in each have five-cI flower 

'There are five pots of flowers in the yard and in the house respectively.' 

b. [np Wu-pen yuanzi li de hua ] hen haokan. 

five-cl yard in de flower very pretty 
'The five pots of flowers in the yard are very pretty.' 

c. [np Yuanzi li de wu-pen hua ] hen haokan. 

yard in de five-cl flower very pretty 
'The five pots of flowers in the yard are very pretty.' 

Applying the same kind of tests to NPs containing MP and various MOD- 
de's in subject and object positions, we get the results in Table 1, which we will 
get to shortly. 

3. Readings for NPs with MOD-de and mei 'every' 

As already mentioned, an NP may contain a MOD-de and a QP headed by a 
quantificational word like mei 'every'. MOD-de may precede or follow QP, as in 
(8) repeated below.'^ NPs with QP preceding MOD-de and those with QP follow- 
ing MOD-de appear to have similar interpretations. However, a careful examina- 
tion reveals that there are some meaning differences between the two orders, 
which I illustrate in the subsections below. 

(8) a. [ 



b. 



NP QP MOD-de 




N'] 


mei-tang shangwu qu 


Badaling de 


Heche zai Shahezhen 


every-cl morning go 


Badalingde 


train at Shahezhen 


ting wu-fenzhong. 






stop five-minute 






NP MOD-de 


QP 


N'] 


Shangwu qu Badaling de 


mei-tang 


Heche zai Shahezhen 


morning go Badaling de 


every-cl 


train at Shahezhen 


ting wu-fenzhong. 






stop five-minute 







3.1. Partitive and non-partitive readings 

Sentences like (15a) and (15b) bring out a difference with respect to partitive 
and non-partitive readings of the NPs containing MOD-de and QP. Both (15a) and 
(15b) may mean Lisi sympathizes with every student wearing glasses within a 
particular group, i.e., the NPs may have the partitive reading 'every one of the stu- 
dents wearing glasses'. However, while (15a) may also mean Lisi sympathizes 
with any student who wears glasses, (15b) lacks such a reading. 

(15) a. [np QP VP-J^ N'] 

Lisi tongqing mei-ge dai yanjing de xuesheng. 

Lisi sympathize every-cl wear glasses de student 

'Lisi sympathizes with any student wearing glasses.' 
'Lisi sympathizes with every one of the students wearing glasses.' 



200 



Studies in the Linguistic Sciences 23:2 (Spring 1993) 



b. [np VP-Je QP N'] 

Lisi tongqing dai yanjing de mei-ge xuesheng. 

Lisi sympathize wear glasses de every-cl student 

'Lisi sympathizes with every one of the students wearing glasses.' 

3.2. Distributive and non-distributive readings 

Sentences like (16a) and (16b) show a difference in the availability of dis- 
tributive readings and collective readings. In (16a), where QP precedes MOD-de, 
there is only a distributive reading, i.e., each puppy performed two numbers. But 
(16b) is ambiguous. While a distributive reading is possible, the sentence may also 
mean the puppies jointly performed two numbers. 

(16) a. [np QP AP-de N' ] 

Mei-zhi tiaopi de xiaogou biaoyanle 
every-cl naughty de puppy performed 
'Every naughty puppy performed two numbers.' 
b. [np AP-de QP N' ] 

Tiaopi de mei-zhi xiaogou biaoyanle 
naughty de every-cl puppy performed 
'Every one of the naughty puppies (jointly) performed two numbers.' or 
'Every one of the naughty puppies performed two numbers.' 



liang-ge jiemu. 
two-cl number 



liang-ge jiemu. 
two-cl number 



4. Summary 

The patterns of definite and indefinite readings of NPs vis-a-vis the order of 
MOD- J^, MP and QP are summarized in Table 1 . 

Table 1 . Patterns of Definite and Indefinite Readings of the NP 
vis-a-vis the Order of MOD-de, MP, and QP (every) 

bare Nouns: def./indef. MOD-de N': def./indef. 



MOD-de 


MP MOD-de 

N' 

def./indef. 


MOD-de MP 

N' 

def. only 


QP MOD- ^6- N' 


MOD-de QP N' 


WP-de 


-pt./pt., distr. 


pt., distr ./coll. 


AP-de 


def./indef. 


def. only 


-pt 


/pt., distr. 


pt., distr./coll. 


PP-de 


def./indef. 


def. only 


-pt 


/Pt., distr. 


pt., distr./coll. 


SP-de 


def./indef. 


def. only 


-pt 


/pt., distr. 


pt., distr./coll. 


Sf^Pobj-de 


def./indef. 


def. only 


-pt 


/pt., distr. 


pt., distr./coll. 


MP-de 


def./indef. 


def. only 


-pt 


/pt., distr. 


?? pt. 


]<\P-de [-poss] 


def./indef. 


def. only 


-pt 


/pt., distr. 


pt., distr./coll. 


nP-de [+poss] 


def./indef. 


def./indef. 


-pt 


/pt., distr. 


pt., distr./coll. 


^P>.r-de 


def./indef. 


def./indef. 


-pt 


/pt., distr. 


pt., distr./coll. 



pt.=partitive; -pt.=non-partitive; distr. =distributive; coll.=collective; 
SP=sentential predicate phrase, e.g., yanjing hao 'eye(s) good'; 
S/NPoi,y=clause missing an object. 

Both definite and indefinite readings are possible for bare Nouns and NPs 
containing MOD-de but no MP or QP. NPs with the order MP-MOD-Jf-N' have 
both definite and indefinite readings, whereas those with the order MOD-^e-MP- 



Wu: Adjectival and determinative measure phrases 



201 



N' have only definite readings except when MOD-de is a possessive NP or NP/,„., 
in which case both definite and indefinite readings are possible. There are meaning 
differences between NPs containing QPs with quantificational words like mei 
'every' as well. While both QP-MOD-Jf-N' and MOD-Jf-QP-N' have partitive 
readings, MOD-J^'-QP-N' lacks the more general reading of 'every' available for 
QP-MOD-Jf-N'. Also, both orders have a distributive reading, but QP-MOD-J^'- 
N' is only distributive, whereas MOD-Jf-QP-N' allows also a reading that is 
collective. 

5. Towards an analysis of the patterns 

5.1. Structural realizations of the different word orders in Mandarin NPs 

We already showed that MOD-de may precede or follow MP or QP. Since 
MOD-de is plausibly regarded as an N' modifier, it is also plausible to assume that, 
if MOD-de precedes MP or QP, then MP and QP are dominated by N' and the de- 
terminer position is phonologically not realized.-'' This gives us the structures in 
(17) corresponding to the order MOD-de MP/QP, where MP and QP occur under 
N' and the determiner phrase (DetP) is phonologically null. 

(17) a. ^Ji^ ^- ^P 

DetP ^^N' DetP' ""N' 





I assume, however, that, if MP or QP occur NP-initially, then MP and QP 
may occur in the DetP. Since MP and QP may occur under N', and the determiner 
need not be phonologically realized, it follows that the order MP/QP MOD-de 
does not univocally determine the structural position of MP and QP, since this or- 
der is also compatible with these phrases occurring under N'. Structures corre- 
sponding to the order MP/QP MOD-de are therefore represented as in (18) and 
(19). In (18a) and (19a), MP and QP are DetPs. In (18b) and (19b), they occur 
under N'. and DetP is phonologically null. 

(18) a. 



NP 



NP 



DetP 



DetP 



MP MOD-de 



■N' 



(19) a. 



NP 



DetP 

I 
QP 



N' 



DetP 



MOD-de 



MP 



NP 



QP' 



■N' 



MOD-de 



N' 



MOD-de 



'N' 
I 
N 



202 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

5.2. The semantic import of MP in different positions 

The view that from a syntactic standpoint numerals can either be determiners 
or be dominated by N' is not new (Partee 1988). As suggested in Partee 1987, nu- 
meral words may have a double life. For instance, three may be an adjective or a 
determiner, and the two three's have different semantic types (1987:130). Partee's 
analysis is motivated to explain the diversity of NP interpretations on the bases of 
general syntactic and semantic principles (1987:115). Following Partee, I assume 
that numerals can be either adjectives or determiners. Accordingly, MPs can either 
be adjectival MPs, hereafter MP[+Adj], which combine with an N' to form an N', or 
be determiner MPs, hereafter MP[+d], which are determiner phrases (DetPs) that 
combine with an N' to form an NP. Semantically, MP[+Adj] marks the cardinality of 
the N' it combines with. For example, the function of the adjectival MP 'three' in 

[n' [MP[+Adj] three] [n' men] ] 

is to restrict the set denoted by [n' men] to a set whose members are groups of 
three. On the other hand, MP[+d] denotes a relation between two sets and is in this 
sense quantificational. For example, the function of the determiner 'three' in 

[s [np [oetp three] [n' men] ] [vp left] ] 

is to indicate that the intersection of the set denoted by the N' 'men' and the set de- 
noted by the VP 'left' is not empty and has three members. The distinction between 
cardinal and quantificational MPs by itself does not give us the distinction between 
definite and indefinite readings of the NP. However, given this distinction and the 
assumption that when numerals occur under N', DetP is still present, though 
phonologically null, the definite and indefinite readings of the NP may then follow 
from the realizations of the DetP. 

5.3. The realizations of the null determiner 

I want to suggest that when the determiner is phonologically null, it may still 
be semantically active. In particular, the phonologically null det can in principle be 
realized as [+def] or [-defj. This means that, in the absence of grammatical factors 
constraining the realization of the null determiner, an NP whose determiner 
position is phonologically null may have the structures in (20): 

(20)a. NP b. NP 



DetP ^N' DetP ^N' 

I I 

[+def] [-def] 

I am assuming, however, that the availability of the [+def] or [-def] interpre- 
tation of the null determiner is subject to the same constraints that limit the distri- 
bution of the determiners some and the. It may be noticed that while the definite 
determiner the may co-occur with (adjectival) numerals in English, the indefinite 
determiner some cannot (I am ignoring the 'about' reading of (21b)). The same 
may be said of the Chinese examples in (22): (22a) is fine, but (22b) is not. 

(21) a. the three men 
b.?? some three men 



Wu: Adjectival and determinative measure phrases 



203 



(22) a. Nei san-ge xuesheng zoule. 

that three-cl student left 
'Those three students left.' 
b. *Yixie san-ge xuesheng zoule. 
some three-cl student left 

The exact nature of the constraint responsible for the unacceptability of (21b) and 
(22b) deserves further investigation. For the purpose of this paper, I will simply 
assume the existence of a grammatical constraint which rules out structures like 

(23) (whether or not DetP is phonologically realized): 

(23) *[np [oetp -def] y [MP[+Adj) num ] N'] ] 

(23) says cardinal adjectives may not co-occur with an indefinite determiner in an 
NP. Given (23), the contrasts in (21) and (22) are expected. I now turn to some 
predictions this analysis makes for MC NPs. 

6. Some predictions 

6.1. The interpretation of NPs without MPs or QPs 

With the analysis proposed here, when no numeral or overt quantificational 
element is present in DetP, DetP can in principle be realized as [+def] 'the' or 
[-def] 'some', as in (24) and (25). This accounts for the indefinite and definite read- 
ings found with bare Nouns and with NPs containing MOD-de but no MP or QP, 
as in (6) and (7), respectively. 



(24)a. 




b. 



NP 



shu 

book 



(6) Lisi maile [np shu ] le 
Lisi bought book part 

'Lisi has bought a book/some books.' or 
'Lisi has bought the book(s).' 



DetP 
1 


N' 
1 


1 

D 
1 


N 
1 


[-def] 


1 

shu 
book 


'a/some book(s)' 



(25) a. 



DetP 




[+def] zai canting dagong de xuesheng 

at dining-hall work de student 
'the student(s) working at the dining-hall' 



DetP 



[-def] 




zai canting dagong de xuesheng 
at dining-hall work de student 
a/some student(s) working at the 
dining-hair 



204 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

(7) [np Zai canting dagong de xuesheng ] gaosu Lisi nali de dangao bucuo. 
at dining-hall work de student tell Lisi there de cake not-bad 
'A/Some student(s) working at the dining-hall told Lisi that cake there is 
pretty good.' 

'The student(s) working at the dining-hall told Lisi that cake there is pretty 
good.' 

6.2. The interpretation of NPs with numerals 

By the analysis proposed here, the null determiner can in principle be real- 
ized either as [-i-def] or as [-def], which are responsible for definite and indefinite 
readings of the NP. We observed however that in the case of (4), where MOD-de 
precedes the MP, only a definite reading is possible. 

(4) [np MOD-de MP N' ] 

Shangwu qu Badaling de liang-tang lieche yijing fachuqule. 

morning go Badaling de two-cl train ab^eady dispatched 

'The two trains going to Badaling in the morning have already been 
dispatched.' 

Why? Given our assumptions about the possible realizations of the null 
determiner, the subject NP in (4) has the structure in (26). 

(26) 

DetP 




[+def] shangwu qu Badaling de liang-tang lieche 

morning go Badaling de two-cl train 

'the two trains going to Badaling in the morning' 

The [-def] reading is ruled out by the same co-occurrence restriction which is 
responsible for ruling out expressions like *Some three boys left in English. Due to 
the co-occurrence restriction on determiners and cardinal adjectives in the NP, 
when MP follows MOD-^^^ and therefore is adjectival, DetP can only be realized 
as [-t-def], and the NP has a definite interpretation only. On the other hand, when 
MP precedes MOD-de, the MP may be under N' hence be adjectival, in which case 
DetP is [-(-def] again and the NP is definite. The other possibility is that the MP 
may be in DetP, in which case the NP is indefinite on the standard assumption that 
in the determiner interpretation numerals are existential quantifiers. This explains 
why cases like (4) with MOD-de preceding MP have a definite reading only and 
accounts for cases like (3), where MP precedes MOD-de and both definite and 
indefinite readings are possible. 



Wu: Adjectival and determinative measure phrases 205 

(3) [np MP MOD-de N' ] 

Liang-tang shangwu qu Badaling de Heche yijing fachuqule. 

two-cl morning go BadaHng de train already dispatched 

'Two/The two trains going to Badaling in the morning have already been 
dispatched.' 

In (3), where MP precedes MOD-de, the MP may be in DetP or under N', as 
in (27) and (28) respectively. Since the determiner 'two' is indefinite, the NP in 
(27) has an indefinite reading. But the NP in (28) has no overt determiner, and its 
DetP is [+def| due to the co-occurrence restriction on determiners and cardinal ad- 
jectives in the NP. Therefore, the NP in (28) has a definite reading. And between 
(27) and (28), we have the two possible readings for (3). 

(27) NP 

DetP _ 

I MOD-de ^^N' 

MPi+D] I N 

liang-tang shangwu qu Badaling de lieche 
two-cl morning go Badaling de train 



(28) 



'two trains going to Badaling in the morning' 



DetP 




f+defl liang-tang shangwu qu Badaling de lieche 
two-cl morning go Badaling de train 

'the two trains going to Badaling in the morning' 

The analysis proposed here accounts for the patterns of definite and indefinite 
readings in Table 1 except for the cases where MP follows a possessive NP-de or 
an NPloc-de. In these cases, both definite and indefinite readings are possible. We 
turn to such cases now. 

6.3. Possessive NP-rfe and NPioc-de 

A plausible analysis for the readings found with NPs containing possessive 
NP-de and NPi„c-de would be that possessive NP-de and NP/,«-J(' have syntactic 
and semantic properties that set them apart from other types of MOD-J^'s. But 
what may these properties be'? 

In L & T, NP-Je is treated as an associative phrase (ASSOCP), which com- 
bines with an NP to form another NP ( 1981:1 13) or with a noun to form an NP 
(1981:126). In our terms this means that possessive NPs can occupy the positions 
in the trees in (29): 



206 



Studies in the Linguistic Sciences 23:2 (Spring 1993) 



(29) a. 



NP 



NP 



NP-de 



[+poss] 



NP 



DetP 

I 
NP-cle [+POSS] 



As Huang points out, however, possessive NPs in MC can receive both defi- 
nite and indefinite interpretations. For example, wo de shu (Lit. 'I de book') stands 
for 'my book(s)' as well as 'book(s) of mine', which in English are definite and in- 
definite respectively. Also, possessive NP-J^ may co-occur with a demonstrative 
prenominally, as in wo de neiben shu 'my that book' (= that book of mine). Huang 
suggests that NP-<ie is therefore not a determiner and has no definitizing function 
as determiners do (1987:252). If this is correct, then the structure (29b) above is 
ruled out for possessive NPs, and we are left with structure (29a). L & T also claim 
that ^V-de only occurs NP-initially (1981:124). Sentence (30a), however, is a 
counterexample to this claim which incorrectly rules out the reading in (30c). 

(30) a. liang-ge Beida de xuesheng 
two-cl BU de student 

b. [nv-de liang-ge Beida de ] xuesheng 

two-cl BU de student 
'students of two BU's' 

c. liang-ge [NP-^e Beida de ] xuesheng 
two-cl BU de student 
'two students of BU' 

Following our assumption that MPs can either be DetPs or be cardinal adjec- 
tives, we are thus led to recognize the following possible structures as licensed. 
MP precedes the possessive NP in (31 ) and follows it in (32). 



(31)a. 



NP 



DetP 

I 
MP[+D] NP-^e [+POSS] 



N' 
I 

N 



b. 



DetP' 

I 

D 

I 

[+def| 




[+Adj] 

NP-t/e [+POSS] 



(32)a. NP 

N?-de [+POSS] 



DetP 
I 



MP, 



•NP 



[+D] 



N' 
I 

N 




Given our analysis of the possible realizations of the null DetP, this means 
that while (33a) will allow the structures in (33b) and (33c), (34a) will allow both 
those in (34b) and (34c). 



Wu: Adjectival and determinative measure phrases 207 

(33) a. liang-ben Zhangsan de shu 

two-cl Zhangsan de book 
b. [np [oetp two-cl] [n' [NP-de[+possi Zhangsan de ] [n' book] ] ] 
c- [np [oetP +def] [w [mp two-cl] [n' [NP-(if[+possi Zhangsan de ] 

[n' book] ] ] ] 

(34) a. Zhangsan de liang-ben shu 

Zhangsan de two-cl book 

b. [np [wp-^d+poss] Zhangsan de] [np [oetP two-cl] [n' book ] ] ] 

c. [np [NP-rft'[+poss] Zhangsan de] [np [oetP +def] [n' two-cl book ] ] ] 

This predicts correctly that both orders should allow both a definite and an 
indefinite reading. The analysis proposed here for possessive NP-de requires addi- 
tional investigation, in particular independent evidence is needed for the structures 
I am assuming. This analysis provides, however, a plausible lead that could ac- 
count for the exceptional behavior of possessive NPs. The analysis may work for 
NPioc-de as well. Assuming that these NPs can occupy the same positions as pos- 
sessive NPs, then again definite and indefinite readings are predicted for both 
orders of NPioc-de and MP. 

7. The interpretation of NPs with MOD-de and QP 

As shown in (8), where a quantificational word mei 'every' is present, the 
different orders of the QP mei-ge 'every-cl' and MOD-J^ are still available. 

(8) a. 



[np QP 


MOD-Je 




N'] 




mei-tang 


shangwu qu 


Badaling de 


Heche 


zai Shahezhen 


every-cl 


morning go 


Badalingde 


train 


at Shahezhen 


ting wu 


-fenzhong. 








stop five-minute 








'Every train going to Badaling 


in the morning stops at Shahezhen for five 


minutes.' 










[np MOY^-de 




QP 


N'] 




Shangwu 


qu Badaling de 


mei-tang 


Heche 


: zai Shahezhen 


morning 


go Badaling de 


every-cl 


train 


at Shahezhen 



ting wu-fenzhong. 

stop five-minute 
'Every train going to Badaling in the morning stops at Shahezhen for five 
minutes.' 

Two issues arise in the analysis of NPs containing MOD-i/e and QP here. 
First of all, although syntactically QP in (8b) does not have scope over MOD-de, 
semantically QP takes scope over the entire NP whether it precedes or follows 
MOT)-de in the NP. Namely, (8b), like (8a), is true only on condition that every 
train going to Badaling in the morning stops at Shahezhen for five minutes. So 
what is the relation between the syntax and the semantics of NPs containing 
MOD-Je and QP? Secondly, we observed that the difference in order correlates 
with a difference in partitive and non-partitive readings of the NPs. I will not at- 



208 Studies in the Linguistic Sciences 23:2 (Spring 1993) 

tempt a full treatment of these two issues here but will sketch some possible 
solutions in the sections below. ^ 

7.1. Scope relations 

Quantificational words like every are inherently relational, i.e., they always 
denote a relation between two sets. For example, in every man runs, every ex- 
presses a relation between the set of individuals with the property MAN and the set 
of individuals with the property RUN. In particular, every says the first set is in- 
cluded in the second. In comparison, numerals, as we saw, need not be relational 
in this sense, but may be cardinality predicates. For example, in three men run, 
three may simply mark the cardinality of sets of individuals that are men. In order 
to capture the relational nature of quantificational words like every, we may follow 
Heim (1982, 1990) and assume that in addition to Quantifier Raising (QR), there is 
a Quantifier Construal Rule (QC) which maps SS onto LF. QR is an operation that 
raises every non-pronominal NP out of S and adjoins them to S. QC is an opera- 
tion that then attaches every quantifier as a leftmost imme