(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Biodiversity Heritage Library | Children's 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 
St92 
V.21 
1991 
Cop. 2 



CENTRAL CIRCULATION AND BOOKSTACKS 

The person borrowing this material is 
responsible for its renewal or return 
before the Latest Date stamped below. You 
may be charged a minimum fee of $75.00 
for each non-returned or lost item. 
Theft, mutilation, or defacement of library materials can 
be causes for student disciplinary action. All materials 
owned by the University of Illinois Library are the 
property of the State of Illinois and are protected by 
Article 1 6B of Illinois Criminal Law and Procedure. 
TO RENEW, CALL (217) 333-8400. 
University of Illinois Library at Urbana-Champaign 



0f C j 2m 



When renewing by phone, write new due date 
below previous due date. L162 



Studies in 

The Linguistic Science 



4 1 STX 

St92 

21 :1 SPR 1991 COPY 2 




ZOANN BRANSTINE: Stop/spirant alternations in Spanish: On the 
representation of contrast 

ALI Darzi: Compensatory lengthening in modem colloquial 
Tehrani Farsi 

Barbara J. HANCIN: On the phonology-morphology interaction in 
Brazilian Portuguese vowel harmony 

SEOK KEUN KanG: Moraic phonology and /l/-irregular predicates 
in Korean 

YONGSOONKanG: Coronal: Transparent or opaque? 

HAN-GYU LEE: Plural marker copying in Korean 

NUMA NlARKEE: Toward an integrated approach to language 
planning 

AL MTEN7E (University of Malawi): On Autosegmental feature- 
spreading in phonology: Evidence from Chiyao 

PRAMOD KUMAR PandEY (South Gujarat University, Surat, India) 
Schwa fronting in Hindi 

Reviews 

R. N. Aralikatti (1989). Spoken Sanskrit in India: A study of sen- 
tence patterns. (Hans Henrich Hock) 

Mark R. Baltin and Anthony S. Kroch, eds. (1989). Alternative 
conceptions of phrase structure. (James H. Yoon) 

Deborah Tannen (1990). You just don't understand. 
(Theresa Conefrey) 

Recent Books 



Department of Linguistic 
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: Charles W. Kissebenh 

EDITORIAL BOARD: Eyamba G. Bokamba, Chin-chuan Cheng, Georgia M. 
Green, Erhard W. Hinrichs, Hans Henrich Hock, Braj B. Kachru, Yamuna 
Kachru, Chin-W. Kim, Charles W. Kissebenh, Howard Maclay, Jerry L. 
Morgan, Rajeshwari Pandharipande, and Ladislav Zgusta. 

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. Invited papers by scholars not as- 
sociated with the University of Illinois will also be included. 

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 the Editor, 
Studies in the Linguistic Sciences, Department of Linguistics, University of 
Illinois, 4088 Foreisn Lancuases Buildins, 707 S. Mathews, Urbana, Illinois 
61801. 

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. 

UPCOMING ISSUES: Vol. 21:2: Illinois Studies in Korean Linguistics, 2 (ed- 
ited by Chin-W. Kim); Vol. 22:1: Papers in General Linguistics. 



Price: $7.50 (per issue) 



STUDIES IN THE LINGUISTIC SCIENCES 



Papers in General Linguistics 



EDITOR 
Hans Henrich Hock 



THE LIBRARY OF THB 

FEB 1 2 1993 

A^'l'/^ERSiry OF ILLilNOIS 
URBArvlA-CHA ' PAIGN 



EDITORIAL ASSISTANT 
Amy Cheatham 



VOLUME 21, NUMBER 1 
SPRING 1991 



DEPARTMENT OF LINGUISTICS, UNIVERSITY OF ILLINOIS 
URBANA, ILLINOIS 61801 



CONTENTS 



Preface v 

Zoann Branstine: Stop/spirant alternations in Spanish: On the 

representation of contrast 1 

Ali Darzi: Compensatory lengthening in modern colloquial 

Tehrani Farsi 2 3 

Barbara J. Hancin: On the phonology-morphology interaction in 

Brazilian Portuguese vowel harmony 3 9 

Seek Keun Kang: Moraic phonology and /l/-irregular predicates 

in Korean 5 5 

Yongsoon Kang: Coronal: Transparent or opaque? 6 7 

Han-Gyu Lee: Plural marker copying in Korean 8 1 

Numa Markee: Toward an integrated approach to language 

planning 107 

Al Mtenje (University of Malawi): On Autosegmental feature- 
spreading in phonology: Evidence from Chiyao 125 

Pramod Kumar Pandey (South Gujarat University, Surat, India): 

Schwa fronting in Hindi 147 

Reviews 

R. N. Aralikatti (1989). Spoken Sanskrit in India: A study of sen- 
tence patterns. (Hans Henrich Hock) 1 6 1 

Mark R. Baltin and Anthony S. Kroch, eds. (1989). Alternative 

conceptions of phrase structure. (James H. Yoon) 167 

Deborah Tannen (1990). You just don't understand. 

(Theresa Conefrey) 179 

Recent Books 183 



f 



Preface 



The present issue of Studies in the Linguistic Sciences appears in 
a format slightly revised from previous issues. The 'Times' laser- 
writer substitution font for 'New York' has been employed for easier 
reading and interparagraph spacing has been reduced. 

In addition, this issue contains a new feature, 'Recent Books', 

which provides brief indications of contents or linguistic interest for 

recent books that were sent to us but could not be reviewed in this 
issue. 

Finally, I have the pleasant task of thanking the following facul- 
ty members for refereeing submitted papers. Eyamba Bokamba, 
Jennifer Cole, Braj Kachru, Yamuna Kachru, C.-W. Kim, Charles Kisse- 
berth, and James Yoon (all in Linguistics), and Jose Hualde (Spanish, 
Italian, and Portuguese). The Department of Linguistics also is 
grateful for support from the College of Liberal Arts and Sciences to- 
ward 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, Eileen Sutton, and Amy 
Cheatham for their help in preparing this issue. 

November 1991 Hans Henrich Hock (Editor) 



i 



d 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



STOP/SPIRANT ALTERNATIONS IN SPANISH: 
ON THE REPRESENTATION OF CONTRAST 

Zoann Branstine 

I argue that Spanish has a maximally simple spirant- 
ization rule: 'Insert [+continuant].' The distribution of voiced 
obstruents (stops after homorganic nasal or lateral, spirants 
elsewhere) falls out naturally from the REPRESENTATION, and 
not from explicit restrictions on the RULE. I use Contrastive 
Specification (Steriade 1987) to eliminate target conditions 
[-sonorant, +voice], and Clements's (1987a) version of Fea- 
ture Geometry with an oral cavity node to explain why 
homorganic sonorant-voiced obstruent clusters must share a 
default feature [-continuant]. The non-assimilatory analysis 
presented here has a further advantage: The overall gram- 
mar is simplified by eliminating the need to order default 
rules before phonological ones. 

1. Introduction 

One of the most prevalent topics of discussion in present day 
phonology is the question of just how much information is present in 
the Underlying Representation and how much is to be supplied by 
phonological and default rules — either language-specific or uni- 
versal. Radical Underspecification (Archangeli and Pulleyblank, 1986, 
etc.) seeks to eliminate all redundancy from underlying repre- 
sentations by allowing only one value of a feature to be specified in 
the UR, with the opposite to be filled in by a universal default rule 
(or a language-specific complement rule). Contrastive Specification 
(Steriade 1987, Clements 1987b, Mester and Ito 1989), on the other 
hand, seeks to represent the notion of 'contrast' by allowing both the 
positive and negative values of a feature which minimally distin- 
guishes two phonemes to appear in the UR and eliminating only re- 
dundant feature values. Mohanan (1989) argues that these two ver- 
sions of Underspecification theory are equivalent, since the opposi- 
tion [aF] ~ [ F] is merely a notational variant of the opposition [aF] ~ 
[-aF].i 

However, the stop/spirant alternations in Spanish and other lan- 
guages such as Basque and Moore provide evidence which distin- 
guishes the two theories. The data which will be discussed in this 
paper point toward the need for a formal, structural distinction 



2 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

between segments that contrast for a particular feature and those 
that do not. 'Contrast' is not directly encoded in the UR in Radical 
Underspecification, nor in a theory which calls for full specification of 
features. 

Recent treatments of Spanish stop/spirant alternations have ignored 
the issue of whether a Spirantization (or Fortition) rule should be 
structure-changing or structure-building, considering this question to 
be an unimportant detail. However, precisely that 'detail' is central 
to the analysis I wish to develop here. Contrastive Specification, but 
not Radical Underspecification, provides a unified account for both 
the non-alternation of voiceless obstruents and the [-cont] realization 
of voiced obstruents after a homorganic nasal or lateral. Finally, Con- 
trastive Specification provides a simple and natural account of the 
alternation which is not available in a theory that requires fully- 
specified feature matrices in underlying representation. 

2. Distribution of Stops and Spirants 

An overview of the distribution of the voiced stops and spirants 
in 'Standard Spanish' is given below^. This distribution is most often 
the object of discussion in analyses of Spanish 'Spirantization' (for 
example, see Fernandez 1988:121. Harris (1984) refers to it as the 
'generalizacion normativa', and in fact, this is the distribution reflect- 
ed in the pronunciations given in many dictionaries of Spanish. 

(1) 

Stops occur in absolute initial position and after- homorganic 
nasals or laterals: 



B 


D 


G 


bello 'beautiful' 


dolor 'pain' 


gato 'cat' 


ambos 'both' 


cuando 'when' 
caldo 'broth' 


tarigo 'tango' 



(2) 

Spirants occur after a vowel, after a glide, after a non- 

homorganic lateral, after [r], and after [+cont] obstruents:^ 

hapa 'bean' ha5a 'fairy' haya 'do!' 

vaipen 'sway' trai5or 'traitor' caiya 'fall!' 

eupolia 'propriety' deuSa 'debt' auyusto'august' 

caipo 'bald' — alyo 'something' 

arPol 'tree' ar5e 'burns' amaryo 'bitter' 

a5Perso 'adverse' ap6omen 'abdomen' suPylotal 'subglottal' 

esPozo 'sketch' des5e 'since' rasyar 'scratch' 

The analysis of the spirant/stop alternations is complicated by 
various factors and thus has been subject of a certain amount of 
controversy. One complication is the variation in distribution of 



Branstine: Stop/spirant alternations in Spanish 3 

stops and spirants, according to speech rate and style. It must be 
kept in mind, then, that any analysis, including the present, should 
be seen as a description of tendencies. Another complication is great 
variation in distribution from dialect to dialect. Thus what one may 
say for 'Standard' Spanish cannot be interpreted as describing all 
dialects of Spanish. Finally, in fast speech, Spirantization interacts 
with several other low-level rules. For the most part, these compli- 
cations will not be discussed here though I believe that they could be 
incorporated into the analysis I am developing. 

3. Spirantization or Fortition? 

One of the main points of contention between the various 
analyses of the stop/spirant alternations in Spanish has been that of 
where to place the 'cost' in the grammar, whether in the rule govern- 
ing the alternation or in the redundancy rule. The contexts in which 
stops appear (utterance-initially and after homorganic nasal or 
lateral) are more restricted than those in which spirants appear, so a 
Fortition or stop-formation account maximizes simplicity of rule 
formation but presupposes using the 'marked' default value, [+cont] 
for voiced obstruents. 

In Spirantization accounts, the [+cont] variants are derived, 
usually by left-to-right spreading from a [-i-cont] segment, since the 
alternation appears to be conditioned by material to the left, and not 
the right, of the target. However, some mechanism^ must be invoked 
to block the spreading in just the contexts where Fortition might be 
said to apply. The universal default rule assigns the unmarked 
value, [-cont], to any voiced obstruents which remain unspecified 
after the application of Spirantization. 

If Spirantization is stated as left-to-right spreading, it gets a 
'free bonus' in that the utterance-initial stops receive their [-cont] 
value by default, without any special stipulation. But Fortition 
accounts receive a similar free bonus, since the glides /y/ and /w/ 
appear as affricates in just the same environments in which the 
voiced obstruents appear as stops: sin hielo [sin dyelo] 'without ice', 
son huertas [sot] g^ertas] 'they are gardens' (Lozano 1979:19). Thus, 
the simplicity metric does not appear to be of much help in deter- 
mining the correct analysis. 

This paradox stems from a failure to recognize the intimate way 
in which underspecification and default are tied in with the alter- 
nation. What is desirable is an account that shows how the behavior 
of all consonants is a natural consequence of their general phono- 
logical properties and of the general phonological properties of the 
structures in which they appear. 



4 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

A remark by Master and Ito, having to do with the problems 
incurred by the apparent necessity of a rule spreading [-voice] in 
English, is relevant to the current paper: 'The problem here, we 
believe, can be traced back to a style of phonological thinking which 
strives to mirror every surface alternation by a language-specific 
rule responsible for that alternation' (1989:281-2). Mester and Ito 
argue for the underlying specification of non-redundant but un- 
marked features. Only redundant features are left unspecified. In 
discussing the exclusion of one of the coronal consonants, [r], from 
the class of segments which undergo a certain type of palatalization 
in Japanese, they remark, 

After all, it is only in virtue of the fact that the other 
coronals are SPECIFIED for [coronal] that the segment r 
stands out as UNSPECIFIED. Otherwise the analysis fails to 
derive the palatalization behavior of the segment r from its 
general phonological properties. In this respect, no analysis 
is acceptable that contains any special proviso about r and 
palatalization (p. 276) 

Essentially, this is the point that I wish to argue about the asym- 
metry between voiced and voiceless obstruents in Spanish with 
regard to Spirantization. Only if non-redundant but unmarked [-cont] 
is specified in the UR do the voiced obstruents /B,D,G/ stand out as 
unspecified for [cont]. 

In the analysis I propose here, neither a specific rule of Forti- 
tion, nor an explicit statement of blocking environments for Spiranti- 
zation is needed. For the data which have received the most atten- 
tion in the literature, Spirantization and Fortition accounts are nearly 
completely equivalent, and the question becomes whether an assim- 
ilatory process is at work at all, or whether both [-] and [+] values for 
[cont] are supplied by feature-insertion rules. Further evidence is 
needed to differentiate the two hypotheses, yet the evidence pre- 
sented in the literature is at best contradictory. The fact that in 
certain dialects voiced stops appear not only after a pause and after 
a homorganic sonorant, but also in a variety of additional contexts 
(after non-homorganic nasals, after glides, even after [s] (see Harris 
1986, Fernandez 1988), seems to point to an assimilatory type of 
Spirantization, with dialects varying as to which segments serve as 
triggers to the rule. Yet Hualde (1990) cites examples of voiced spir- 
ants following voiceless stops, something not easily accounted for by 
Continuant Spreading. 

Pending closer investigation of the data which might serve to 
resolve the issue, and given the theoretical problems discussed below 
which are raised by assimilatory accounts of either the Fortition or 
Spirantization variety, I opt for a structure-building (slot-filling) rule 



Branstine: Stop/spirant alternations in Spanish 5 

of Spirantization similar to that proposed in Hualde 1990 which 
freely assigns the feature [+cont].5 Though I refer to this as Spiranti- 
zation in order to be consistent with the name used in the literature, 
I withhold judgment as to whether this is properly seen as a phono- 
logical rule or a default rule. 

4. Some problems with treating Spirantization/Fortition as 
assimilation 

As has been noted by several authors (e.g., Harris, 1986, Hualde 
1988, 1990), Spirantization in Spanish clearly cannot take place in 
the lexical module: it applies both in underived forms and between 
words, and produces segments not present in the UR of the language, 
since voiced obstruents do not have an underlying contrast for 
continuancy. Thus, Spirantization should not be subject to Structure 
Preservation, and should not apply at a stage in the derivation where 
Structure Preservation holds. Nevertheless, 1 suggest that the fact 
that only voiced obstruents participate in the alternation is a result 
of Spirantization being a slot-filling and not feature-changing rule. 
Thus Spirantization must be ordered before the default rule which 
assigns [-cont] to voiced consonants. ^ 

This presents a considerable problem if Spirantization is seen as 
an assimilatory process, since this treatment presupposes the assign- 
ment of the default value [-i-cont] to the rule's triggers, typically 
vowels. While certain authors (e.g., Archangeli and PuUeyblank 
(1986), Borowsky (1986)) have argued that default rules may or 
even must sometimes apply before phonological rules, this particular 
case presents a conceptual problem. It is difficult to think of a plaus- 
ible or principled way in which to allow vowels, for which the 
feature [cont] is completely redundant, to be assigned their default 
value of [+cont] while still withholding the default [-cont] from 
consonants. Other than including a proliferation of extrinsic ordering 
statements between default and phonological rules, there is no way 
to present Spirantization as the spreading of [-t-cont]. 

If the default rule supplying [-i-cont] to vowels is ordered prior to 
the only point in a derivation where the presence or absence of the 
feature is even relevant, this seems identical to specifying vowels as 
[-I-cont] in the UR, a position allowed by neither Radical or Contrastive 
Specification. Though Archangeli and Pulleyblank's (1986) Redun- 
dancy Rule Ordering Constraint allows default features to be supplied 
'as early as necessary' — quite early in the derivation, at the begin- 
ning of the lexical component, in fact — it appears to me that this 
trick usually has the function of supplying a contrastive, rather than 
a redundant, default feature.'' Further, the default rules V -^ [+cont] 
and C -> [-cont] are of different natures. The former is a universal 
cooccurence statement: Vowels are always continuant and are not 



6 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

legitimate bearers of the opposite feature. The latter is a statement 
about markedness. Ordering the vowel default before the consonant 
default does not make sense theoretically. 

A Fortition account of the data, such as that given in Goldsmith 
(1981) or Hualde (1989), in which voiced obstruents receive a [-cont] 
specification by assimilation to a homorganic nasal or lateral, does 
not avoid the problem of extrinsically ordering default rules before 
phonological rules. It merely shifts the question to the specification 
of default [-cont] for sonorant consonants, since spreading implies the 
presence of a feature. Spreading of the redundant feature [-cont] 
from sonorant consonants is no more in line with underspecification 
theory than spreading the redundant feature [+cont] from vowels. 

5. Non-assimilatory accounts 

Below I discuss two different alternative approaches which have 
been proposed to account for the stop/spirant alternations. Both are 
non-assimilatory and thus avoid the problems just discussed. The 
first, Amastae's syllable-final lenition, is attractive but untenable. 
The second, Hualde's [+cont]-insertion rule (actually proposed for 
Basque) is essentially the one I adopt in this paper. However, it 
relies on target conditions and on an explicit statement that the rule 
is blocked in homorganic clusters, both of which can be eliminated in 
the analysis I propose. 

5.1. Spirantization as syllable-final lenition 

Amastae (1986) proposes an account of the stop/spirant alter- 
nations as an instance of syllable-final weakening. This is certainly a 
process which has occurred in Spanish: ^-aspiration, loss of final con- 
sonants such as [d], [r], [1], and nasals, as well as velarization of 
syllable-final nasals are all well-known examples of syllable-final 
weakening. Amastae claims that the 'loose ends' that his analysis 
does not account for show that 'Spanish Spirantization is a still- 
evolving rule' (p. 4). This may well be the case; all languages are 
constantly in flux; however, the basic assumption of generative 
linguistics is that a language, at any point in time, has a learnable 
and coherent grammar. Amastaes's analysis fails on its own account, 
regardless of what kind of historical change is going on in Spanish. 

Though Amastae posits syllable-final weakening, such an analy- 
sis does not correctly characterize the data. In addition to syllable- 
final position (e.g. pare[6], a[P]-5omen) voiced spirants often occur in 
syllable-initial position, (e.g. ha-[p]a, ha-[5Ja, ha-[y]a ap-[S]omen. 

Amastae does not give sufficient evidence for his rule making 
intervocalic consonants^ ambisyllabic (Amastae's figure (6), p. 6): 



Branstine: Stop/spirant alternations in Spanish 



(3) Associate a 



segment with a preceding Rhyme 



+obstr 
+voiced 
.+SI 

(syllable-initial) 

This causes words like dedo 'finger' and hablar 'to speak' to be 
syllabified as follows (Amastae's (7) and (9), pp. 6-7): 

(4) s s s s 

I \ I \ I I \ 

I \ 1 \ I I \ 

R O R R O R 

1 I I I |\ / \ l\ 

I I I I I \l I I \ 

dedo (h)ablar 

Amastae offers as evidence for the ambisyllabicity of inter- 
vocalic consonants his finding that a consonant is more likely to 
spirantize before a stressed vowel than before an unstressed vowel. 
However, it is generally a stressed syllable rather than an unstressed 
one that attracts additional consonants (see Borowsky 1986 p. 258). 
Therefore, the stress of the following syllable is not an argument for 
ambisyllabicity. Further, though syllable-final lenition is a widely- 
attested phenomenon in many languages of the world, including 
Spanish, a look at the principles of Spanish stress assignment shows 
that there is no independent motivation for regarding syllable-initial 
voiced obstruents as ambisyllabic. 

Spanish stress resembles Latin stress. In particular, stress is 
limited to one of the last three syllables. A heavy penult blocks 
stress from being assigned on the antepenult: 

(5) sabana, sabana, BUT caramba, *caramba 

Ambisyllabicity of the consonant preceding a stressed vowel 
would merely make the syllable to the LEFT of the stress heavy. 
Since metrical structure must be assigned moving from right to left 
in Spanish, the structure of syllables to the left of the stress is 
irrelevant. Furthermore, secondary stress, assigned post-cyclically, is 
not quantity sensitive (see Halle and Vergnaud 1988). Thus the 
stress-related facts cited by Amastae do not justify making word- 
internal onset consonants ambisyllabic, since there could be no me- 
trical evidence of whether the newly-created, pretonic heavy syll- 
able was actually heavy. 

What's more, a rule of ambisyllabicity obscures the generali- 
zation that can be made about words with antepenultimate stress: 
None have heavy penults. For instance, words such as *cd.ram.ba, 
*re.ser.va, *de.sas.tre are impossible. If an intervocalic consonant 
must be made ambisyllabic before it can spirantize, words such as 



8 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

pdrpado 'eyelid', ndufrago 'shipwrecked person' would have the same 
structure as the ones which cannot have antepenultimate stress: 
they would have a heavy penult. Although according to Amastae's 
data, these types of words would presumably be less likely to have 
spirants than those where the spirant preceded a stressed vowel, 
nothing in his analysis precludes the spirant from occurring. The 
ambisyllabicity rule would have to be ordered after stress assign- 
ment to keep from incorrectly ruling out words like pdrpado. with 
antepenultimate stress. The addition of extrinsic rule ordering of 
this nature complicates the grammar of the language. 

Ambisyllabicity is undesirable for another reason. Hualde points 
out (personal communication) that in addition, there is evidence that 
prefixes constitute an independent domain for syllabification. Thus 
sub. regional and sub.liminar are divided between the obstruent and 
the liquid, rather than the expected syllabification as in monomor- 
phemic su.blime and so.bre. Amastae's rule would obscure such a 
distinction. 

Even if it were not for the other problems with the ambisylla- 
bicity rule, it is unclear how the Spirantization of a series of two 
voiced obstruents in words such as su[ Pyjlotal could be seen as a 
result of syllable-final lenition. Associating two obstruents to the 
rime of a preceding syllable is contrary to the Sonority Principle, 
thus the syllabification of subglotal could not be subg.lo.tal. 

As I indicated earlier, Amastae's proposal does hold a certain 
attraction. His account is an attempt to find an explanation for the 
Spirantization facts based on linguistic universals. As Hock has 
pointed out, weakening processes such as spirantization usually 
'occur in just two environments: medial intervocalic (or intersonor- 
ant) position and word- or syllable-final environment' (1986:83). 
Unfortunately, Spanish also has spirants in syllable-initial position 
after obstruents, so an appeal to syllable-final weakening as an 
explanation of the facts is simply not possible. 

5.2 Spirantization as continuant-insertion 

Hualde (1990)^ presents a non-assimilatory rule which freely 
inserts the feature [+cont] on voiced obstruents^ 0, except in certain 
'strong' contexts (i.e., homorganic clusters and utterance-initially) 
where it is explicitly blocked. Hualde argues that the mere homor- 
ganicity of the cluster is enough to block Spirantization in Catalan 
and suggests it may be the case for Spanish as welL^^ 

Unfortunately, it does not appear to be true that a voiced 
obstruent must always appear as a stop when preceded by ANY 
homorganic consonant. Spirantization is not blocked in desde [dez5e]. 
Similarly, [r] does not block Spirantization on a following [d] either. 



Branstine: Stop/spirant alternations in Spanish 9 

It could be the case that exact homorganicity is required, down to the 
non-distinct place features, so that dental spirant [5] is permissible 
after the alveolar continuant [z]. However, the examples Hualde 
gives of voiced stop after a homorganic continuant are of labiodental 
[f or v] followed by bilabial [b]. These are not totally homorganic 
clusters, either, but in both the Catalan and the Spanish example, the 
[b] surfaces as a stop. While for Latin American Spanish there is no 
dental continuant phoneme, some dialects of Iberian Spanish do have 
[0] (spelled c or z) which contrasts with alveolar [s]. In words such as 
Mazda (the Japanese car) or phrases such as paz de Dios 'God's peace", 
the /d/ is realized as a spirant [5]. 

Since the facts about what actually happens in homorganic 
obstruent-obstruent clusters remain unclear, I will only concentrate 
on sonorant-obstruent clusters. I propose below that after Nasal and 
Lateral Assimilation takes place, the clusters share not only a Place 
node, but also a feature specification for [cont].^^ Since the sonorant 
member of the cluster cannot be assigned [-i-cont], the cluster must be 
[-cont]. Therefore, as will be seen below, it is not necessary to 
restrict Spirantization by explicit blocking conditions. The fact that it 
does not apply in homorganic clusters is predicted by independent 
factors. 

6. The Contrast/Default analysis 
6.1. Against target conditions 

Any analysis of the stop/spirant alternations in Standard 
Spanish must incorporate the following facts: 

a) THE ALTERNATION ONLY AFFECTS CONSONANTS THAT DO NOT CONTRAST 
FOR CONTINUANCY. That is, voiceless consonants are not affected; they 
contrast both in environments in which voiced obstruents are pre- 
dictably spirants and those in which voiced obstruents are predict- 
ably stops (e.g., cana[p]e vs. ca[f]e, pa[t]a vs. pa[s]a, ta[k]o vs. ta[x]o, 
com[p]adre. vs. con[f]ite, en[t]onces vs. an[s]iosa, ban[k]o vs.fran[x]a, 
fal[t]a vs. fal[s]o). 

b) SONORANT CONSONANTS (n, m, 1, AND r) NEITHER CONTRAST FOR 
CONTINUANCY NOR PARTICIPATE IN THE ALTERNATION. 

Most accounts incorporate (a) and (b) in the form of 'negative 
environment': The target of the rule is stated as [-t-voice, -sonorant]. 
There is nothing unusual about this; after all, the assumption that 
phonological rules affect natural classes of sounds is the backbone of 
phonological theory. However, stating these target conditions ob- 
scures the relationship between the underlying distribution of the 
feature [cont] and the rule of Spirantization. Though the features 
define the natural class which participates in Spirantization, they do 



10 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

not encode the relationship that exists between this particular 
natural class and the feature [cont] in the grammar of Spanish. 

The feature [-sonorant] excludes sonorants from the alternation, 
yet this is less a fact about the rule of Spirantization than it is about 
what sort of segments may occur in natural languages. If nasal frica- 
tives occur, they are very rare and unstable, so clearly the non- 
spirantization of nasals is a consequence of a universal filter *(-i-nasal: 
+consonantal: -(-continuant). 13 There is considerable debate about 
whether the lateral consonant [1] is properly defined as [-i-cont] or 
[-cont]; I take this as evidence that languages do not commonly have 
alternations between a stop [1] and a fricative [1].14 Hualde (1988) 
has argued that, at least for Basque, [1] patterns with [-cont] seg- 
ments; therefore, a similar filter *(-i-lateral : (node): -i-continuant)i5 
could be in effect. Of the sonorant consonants of Spanish, this leaves 
[r]i6. While [r] is often considered to be [-^cont], thus contrasting with 
[1], which is [-cont], actually the feature [lateral], which is needed for 
[1] anyway, distinguishes the two, so /r/ and /I/ are unspecified for 
[cont] in the UR. Lozano (1979 p. 119) discusses evidence that at 
least for some speakers, [r] is produced as a stop. It seems to me, 
then, that the phonetic status of [r] is very similar to that of [1], and 
for the present purposes it can be considered to be redundantly 
[-cont].!'' Since all of the sonorant consonants are predictably [-cont] 
in all environments, I propose a constraint *(-i-sonorant : -i-consonan- 
tal : -i-continuant), which says that in Spanish, sonorant consonants 
may not bear the feature [-(-cont]. 1 8 

The voiced obstruents bear both values for [cont] on the surface, 
but they do not contrast for this feature. Sonorant consonants cannot 
bear [-(-cont], and voiceless obstruents contrast underlyingly for con- 
tinuancy. Nothing prevents voiced obstruents from contrasting for 
continuancy; it is a fact about the phoneme inventory of Spanish that 
they do not. 

Evidence that it is the underspecification for the feature [cont] 
which is relevant to the Spirantization rule and not the features 
[+voice -sonorant] comes from the language Moore, as discussed by 
James Myers (1989). Whereas there is a phonemic contrast between 
b/v, d/z, p/f, t/s, and k/x, there is none between g/y. However, 
there is a surface alternation between [g] and [y].^^ If the Spiran- 
tization rules of Spanish and Moore list target conditions as [-(-voice 
-sonorant] and [-(-voice -sonorant +back], respectively, then the 
relationship between Spirantization and the underlying segment 
inventory of each language is obscured. 

In both Spanish and Moore, ALL segments which do not contrast 
for continuancy and which may legitimately bear both feature values 
for [cont] undergo the Spirantization rule. Thus in both languages the 



Branstine: Stop/spirant alternations in Spanish 1 1 

target conditions on the rule are only duplicating information which 
is represented elsewhere in the grammar.20 if target conditions are 
eliminated, Spanish and Moore share a single rule of Spirantization, 
i.e., 'Insert [+cont]'. Since Spirantization is quite common in languages 
of the world, it is reasonable to speculate that such a rule should be a 
part of UG rather than of individual grammars, similar to Nasal 
Assimilation and Final Devoicing. This is further reason to look only 
at the relevant feature, [cont], rather than at the other features 
which happen to characterize the set of segments which could but do 
not contrast for continuancy. 

Not only do the conditions on the target of the Spirantization 
rule duplicate a generalization already present in the UR, they are 
misleading. A Spirantization rule would be no more formally com- 
plex nor seem less 'natural' if the target conditions referred to 
another natural class, say, [-anterior] or [-i-nasal]. Yet these do not 
describe natural classes which typically undergo Spirantization in 
languages. 

As we have seen, the non-participation of voiceless consonants 
and of sonorants in Spirantization in Spanish is not arbitrary. There 
is an obvious difference between the segments which do spirantize 
and those which do not: those which do spirantize are free to 
associate with either [-i-cont] or [-cont], while the ones which do not 
are restricted. Voiceless obstruents show a phonemic contrast for 
[-•-/-cont] which is not neutralized. Sonorant consonants are consis- 
tently [-cont]. The former is a fact about the underlying phonemes of 
Spanish and the latter, I believe, falls out of universal markedness 
conditions. 

Spirantization, then, though clearly postlexical, is not a feature- 
changing, but rather a slot-filling rule, an option not available in a 
theory requiring full specification in the Underlying Representation. 
Seen in this way, it is not necessary to place conditions on the target 
of the rule. Furthermore, as is shown below, it is not necessary to 
state 'blocking' environments, since the rule of Nasal (and lateral) 
Assimilation creates structures that in and of themselves block 
[+cont] from associating with the obstruent. Spirantization can then 
be reduced to free insertion of the feature [-i-cont]. The cases where it 
does not apply do not need to be stated explicitly in the rule — the 
[-t-cont] value simply does not associate where it cannot. 

6.2 The 'blocking' effect of Nasal/Lateral Assimilation 

The generalization that Spirantization does not apply to seg- 
ments which cannot acquire the feature [-t-cont], in addition to 
accounting for the non-participation of voiceless consonants and the 
sonorants, covers the cases where possible targets, voiced obstruents, 
surface as stops. Essentially the insight of Fortition analyses such as 



12 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Goldsmith 1981 and Hualde 1989 is that the voiced obstruents in rig, 
nd, mb, and Id clusters acquire the [-cont] value from their sonorant 
partner, and the homorganicity of the cluster is somehow crucial. 
However, stating Fortition as an assimilatory process implies spread- 
ing of a redundant value, a theoretically undesirable claim, as has 
been discussed above. Furthermore, as argued in Hualde (1989), 
[+cont] does not appear to be the default value for voiced obstruents. 
Therefore, the 'Fortition' environments should be analyzed not in 
terms of contexts where the feature [-cont] is supplied by rule, but 
rather in terms of contexts where it is impossible for voiced ob- 
struents to receive the feature [-i-cont]. 

6.2.1. Harris's appeal to geminate inalterability 

In order to block Spirantization from applying in just the 
'Fortition' contexts, one might attempt to appeal to some notion of 
geminate inalterability, since the structures created by Nasal Assim- 
ilation are partial geminates. 21 In fact, this is the explanation given 
by Harris, and formalized in his Rule Application Convention (RAC): 

(6) RULE Application Convention: Given a representa- 
tion REP of linked (including merged) matrices and a rule 
RUL of the form [aF] -^ [pF]/SD, RUL applies to REP iff both 
X and Y of REP meet the structural description SD of RUL. 
(1985:132). 

However, the fact that ns, rr)f, and tjx are possible clusters, where 
the values for [cont] differ in a homorganic cluster, points to inde- 
pendence of the continuant and place of articulation nodes. Ad- 
ditionally, note that the RAC refers explicitly to structure-changing 
operations. It cannot apply to slot-filling rules, since the shared 
place of articulation node does not prevent the nasals in mp, nt, and 
r]k clusters from receiving the default feature value [+voice], though 
they are linked to voiceless obstruents which do not meet the SD of 
the default rule [-i-son] -^ [-i-voice]. Harris's account thus requires that 
voiced obstruents be assigned their default [-cont] value prior to the 
application of Spirantization, which is formulated to apply only to 
segments specified as [-cont]. Not only is such an ordering a violation 
of the Elsewhere condition, it once again attributes the difference in 
behavior of voiced and voiceless obstruents to voicing, not to the 
function of the feature [cont]. 

I show below that it is in fact not necessary to state a rule 
separate from the place of articulation assimilation which occurs 
between nasals (and laterals) and a following obstruent. In these 
cases, the important facts are that voiced obstruents themselves are 
unspecified for [cont], unlike voiceless obstruents, and they become 
linked to segments which cannot acquire the feature [+cont]. Where 
a voiced obstruent merely FOLLOWS a sonorant but surfaces as a 



Branstine: Stop/spirant alternations in Spanish 13 

spirant, there is no linking: l[p], l[y], r[y], r[5], r[y]. Thus there are two 
cases where an obstruent may bear the feature [+cont]: (1) if this 
feature is part of its underlying representation {f,s,x) and (2) if the 
obstruent is independent, not linked to a sonorant consonant. 

6.2.2. The oral cavity node and shared default [-cont] 

The question remains of how place and continuancy are related 
in the feature geometry. Clearly, at least for Spanish, homorganic 
clusters do not need to agree for any other non-place feature: they 
may disagree for [sonorant], [nasal], or [cont] and [voice] as in ns, rrjf, 
and rjx. Only clusters which are homorganic and in which both 
members are unspecified for [cont] must share the default value, 
[-cont]. An insightful analysis of the stop/spirant alternations in 
Spanish ought to relate this fact to the way in which the default 
value of [cont] is assigned to clusters which are linked for place of 
articulation. 

Below I give the rule of Nasal and Lateral Assimilation proposed 
by Hualde (1988:161) for Spanish and Basque. 

(7) Nasal and Lateral Assimilation 
Operation: Spread 
Argument: P 
Direction: leftwards 
Target Conditions: [-cont], [+son], rime 

The target condition [-cont] prevents this rule from applying to 
/r/, which in Hualde's analysis is [+cont]. I believe that the feature 
[-cont], being the default for sonorants, is not present at the stage 
where Nasal and Lateral Assimilation applies. Therefore, some other 
device must be used to prevent the rule from applying to /r/, and 
the target conditions will be just [+son], rime. The lateral /I/ only 
assimilates to coronals and not to labials or velars because of the 
positive constraint requiring laterals to have a coronal articulation 
(Hualde 1990, p. 160). 

The application of this rule creates structures which are linked 
on the place of articulation node, represented graphically below (only 
the relevant node structure is represented). 

(8) 

Supralaryngeal Supralaryngeal 



Place 

However, it is still not clear how the shared place node is rele- 
vant for assignment of continuancy. In 'standard' feature geometry 
(Clements 1985, Sagey 1986) these features are not related in any 
clear way. However, Clements (1989) proposes an organization of 



1 4 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

feature geometry with an oral cavity node which dominates only the 
continuant node and the place of articulation node. 



(9) 



Root 



Laryngeal Supralaryngeal 

Oral Cavity [tnasal] 



[±cont] 



Place 



[cor] [lab] ... 

After Nasal and Lateral Assimilation has occurred, I propose that 
in sonorant-voiced obstruent clusters, the Shared Feature Convention 
(Steriade 1982) applies, fusing the oral cavity nodes, since in the 
absence of specification for continuancy they are nondistinct. 

(10) 

SL SL SL SL 

I I SFC ^^^-^ 

OC OC OC 

Place Place 

Thus, in the clusters mb, nd, r]g, and Id, since there is only one 
oral cavity node, there can be only one value for the feature [cont]. 
In other words, there is only one docking site for the feature [cont], 
which is shared by the two consonants. The sonorant member of the 
cluster is not a legitimate bearer of the feature [+cont], so the only 
possible value for the cluster is [-cont]. (This idea is similar to one 
presented in a footnote in Borowsky 1986 on voicing agreement in 
Japanese nasal-obstruent clusters.) 

For ns, etc., the Shared Feature Convention will not apply. No 
fusion can occur since the voiceless fricative carries an underlying 
value of [-cont]. The oral cavity nodes are distinct. Since there are 
two oral cavity nodes, nothing prevents the sonorant and obstruent 
from bearing opposite values for [cont]. 



(11) 



Branstine: Stop/spirant alternations in Spanish 15 

N s N s 



f^""^ oc oc 





[-1-cont] [-cont] [+cont] 

Place Place 

[cor] 

Thus, a voiceless, but not a voiced fricative may follow a homor- 
ganic sonorant. This has nothing to do with the voicing, per se, it has 
to do with the phonemic inventory of Spanish: only voiceless con- 
sonants are specified for [cont] in the UR, because the feature [cont] 
only makes lexical contrasts for voiceless consonants. Clusters such 
as nt, etc. are unproblematic. The fact that both segments are [-cont] 
on the surface is a coincidence. The [t] is specified as [-cont] in the 
UR, and the [n] is assigned that value by default: 

(12) 

N t N t 





[-cont] [-com] [-cont] 

Place Place 

[cor] [cor] 

Conceivably, the SFC would apply to this structure and merge 
the oral cavity nodes, since they dominate identical material, but as 
far as I know nothing crucial depends on the application of the SFC at 
this point. 

In Ig and lb clusters, since there is no linking, the voiced 
obstruents are free to acquire the feature [-i-cont], and nothing will 
prevent Spirantization from applying. Obviously, then [1] is later 
assigned its default value of [-cont]. 



16 



(13) 

L 


G 


OC 
1 


OC 

1 


1 

Place 
1 


1 

Place 
1 


1 

[cor] 


1 
[vel] 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 
L V 1 



OC 
I 



OC 



[+cont] 
Place Place 

I I 

[cor] [vel] 



OC 

[-cont] 

Place 

I 
[cor] 



OC 

[+cont] 
Place 

I 

[vel] 



6.3. Statement of the rule of Spirantization 

I propose a rule of Spirantization adapted from that i Hualde 
(1988), with the crucial difference that neither target condi ns nor 
blocking conditions are necessary. 

(14) Spirantization (Continuant Insertion) 
Operation: insert 
Argument: [+cont] 

Continuant Insertion rather than Continuant Spreading has 
several advantages. First, it eliminates the problem of supplying a 
redundant feature value to certain segments without supplying it to 
others. Second, it has been claimed that redundant features do not 
participate in phonological processes; Continuant Insertion but not 
Continuant Spreading is consistent with this claim. Finally, it is no 
longer necessary to restrict Spirantization to be a slot-filling but not 
a feature-changing rule. Continuant Insertion, by its very nature, 
will only supply the value [-i-cont] to unspecified slots. 

Furthermore, as has been noted by Hualde (also by Myers for 
the language Moore) sometimes a voiced spirant follows a stop. 
Hualde gives the example from Spanish Bonet gano where the /g/ 
surfaces as a spirant [y]. If Spirantization is spreading of the feature 
[+cont], this /g/ should surface as a stop, since the segment to its left, 
[t], is a stop. 

Though it might be tempting to abandon the 'contrastive' speci- 
fication of both [+] and [-cont] for voiceless consonants in the UR and 
claim that this [t] is somehow transparent to the feature [+cont], it is 
not possible to avoid the problem by resorting to radical underspeci- 
fication here. If voiceless stops as well as voiced obstruents are 
underspecified for continuancy, voiceless stops should spirantize. 
Even Structure Preservation (though it shouldn't be operative in the 
postlexical module) cannot block Spirantization from applying to 
voiceless stops, since voiceless continuants occur in the UR. Radical 
Underspecification would require the stipulation of target conditions 
on Spirantization, which I have argued above to be undesirable, since 
the target conditions [+voice -sonorant] really refer to Information 



Branstine: Stop/spirant alternations in Spanish 17 

about phonemic contrasts, which should be encoded elsewhere in the 
grammar and should not be duplicated in phonological rules. 

7.0. Conclusion 

In this paper I have shown that the simplest and most explan- 
atory analysis of the stop/spirant alternations in Spanish is possible 
only within a theory which allows structure-building (slot-filling) 
rules and clearly distinguishes between segments which contrast for 
a particular feature and those which do not. The analysis I have pro- 
posed is possible only in Contrastive Specification. It is not possible 
either in a theory which requires all predictable information to be 
omitted from the underlying representation, nor in a theory like that 
suggested by Mohanan (1989) which has full specification, structure- 
changing rules, and constraints. 

Just like a radically-underspecified account, a fully-specified 
account cannot represent the asymmetry in behavior between voiced 
and voiceless obstruents without stating explicit target conditions on 
the Spirantization rule: since both voiceless fricatives and stops exist 
in the language, no constraint of the type *(-voice +continuant) can 
be invoked to keep the voiceless stops from participating in Spiranti- 
zation, nor, alternatively, can a constraint like *(-voice -continuant) 
keep voiceless fricatives from participating in Fortition. Neither 
Radical Underspecification nor a Full Specification provides a formal 
way to distinguish between segments for which a specific feature is 
distinctive and those for which it is not. 

NOTES 

1 In his paper, Mohanan shows some of the weaknesses of 
Underspecification theory and challenges phonologists to compare 
underspecified accounts with a theory that calls for full specification 
and either constraints or repair strategies rather than structure- 
building default rules. I find such a theory inadequate for dealing 
with the alternations which are the topic of this paper. 

2 The symbols [b,d,g] stand for the voiced bilabial, dental, and 
velar spirants, respectively. Except for the sounds being discussed, 
orthographic form is used. The letter h is not pronounced; v and b 
both represent the phoneme /B/ (with allophones [b] and [b]); the 
digraphs // and ch represent [1] and [c] respectively; unstressed / and 
u become glides [y] and [w] when next to another vowel. 

3 Few words end in non-coronal consonants, so, in the interests 
of parallelism, I have not included examples of word-final environ- 
ment here. However, word-final d is somewhat common and always 
spirantizes: pare[d] 'wall'. For an example of a final non-coronal. 



1 8 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

Lozano 1979 cites the loanword club which is pronounced with a 
spirant [b]. 

4 The 'mechanism' should presumably be some version of the 
Linking Constraint, or Geminate Inalterability, since the clusters in 
which Spirantization is blocked are linked for place of articulation. 
However, Harris's (1986) Rule Application Convention has serious 
problems (see discussion in Hualde 1988 and below in the text) and 
Hayes' (1986) Linking Constraint, 'Association lines in a rule are 
interpreted as exhaustive', does not apply to the structures in which 
Spirantization is blocked, since the rule does not refer to place of 
articulation. 

5 If Spirantization/Fortition were feature-changing, and applied 
to the voiceless obstruents as well, it would be quite easy to decide 
between Spirantization and Forition: If stops and fricatives contrast- 
ed intervocalically, only fortition could be used; if stops and 
fricatives alternated after homorganic nasals, only Spirantization 
could be used. The problem is that in Spanish, voiceless stops and 
fricatives contrast in all environments and voiced obstruents contrast 
in none. 

It appears to me that there is very little difference between an 
account with a Spirantization rule, which freely assigns [+cont] — 
except after a pause or homorganic sonorant — and an account which 
assigns [+cont] by a default rule after some Fortition rule supplied 
the opposite value after a pause or homorganic sonorant. If [-cont] is 
the default value, the rule of Spirantization leaves it very few 
instances in which to apply; what's more, if I am correct in my 
analysis below of homorganic sonorant-obstruent clusters, voiced 
stops get the [-cont] parasitically when the [-cont] feature is assigned 
by a default rule to the sonorant consonants. Thus, a Spirantization 
account leaves only utterance-initial voiced obstruents to receive the 
supposed default [-cont] feature, and it must be stated as an explicit 
condition on the Spirantization rule that it is blocked just in 
utterance-initial position. 

6 This is in direct contradition to the analysis proposed by Harris 
(1986), and does not conform to Archangeli and Pulleyblank's (1986) 
Default Ordering Conventions, since the default [-cont] rule would 
apply 'as early as possible' in its component, that is, at the beginning 
of the postlexical component. 

7 In Contrastive Specification, of course, contrastive features are 
present in the UR and thus such cases are irrelevant to the present 
discussion. 



Branstine: Stop/spirant alternations in Spanish 19 

8 It is not clear how Amastae would syllabify the second in a 
series of two voiced obstruents, for example, in abdomen, subglotal, 
etc. See the discussion in the text below. 

9 This unpublished manuscript was a revision of a chapter of his 
thesis. Hualde subsequently rejected the continuant-insertion analy- 
sis and returned to his earlier Fortition analysis. 

10 This is also the approach adopted by James Myers in his 
account of the spirantization of /g/ in Moore. While Spirantization 
occurs as an insertion of the feature [-i-cont], the blocking conditions 
in Moore are different than in Spanish. In Moore, preceding conson- 
ants are irrelevant. Rather, a preceding [+ATR] vowel is involved in 
the blocking of Spirantization. I will not discuss the exact formulation 
of the blocking here. 

^ 1 For Standard Spanish, it is actually very difficult to decide 
whether Spirantization should be considered a spreading or an 
insertion rule, since both options have drawbacks. A spreading 
account requires that the triggers (most typically vowels, but also 
glides and continuant obstruents, depending on the dialect) be 
assigned their default [+cont] value in order to spread it, while the 
target remains underspecified. An insertion account, however, does 
not provide an easy way to account for the fact that the stop/spirant 
alternation is determined by the nature of the segments to the left 
but not the right of the voiced obstruent, or for the dialectal 
differences in what segments other than vowels may appear to the 
left of a voiced spirant. In Costa Rican Spanish, for example 
(Fernandez 1988), voiced spirants appear only after vowels while 
stops occur not only in homorganic clusters, but also after glides, [1], 
and [s]. Perhaps dialects choose between spreading/insertion. Since 
I am concentrating on Standard Spanish here, and the insertion 
account is consistent both with the facts and with general principles 
of default-rule ordering, I choose to represent Spirantization as 
Continuant Insertion. 

12 This has to do with the feature geometry of the structures and 
with the Shared Feature convention, as will be discussed in the text 
below. 

13 In languages where nasals do undergo Spirantization, the 
nasality tends to be lost or shifted onto a vowel. See Hamp (1974) 
for a discussion of this nasality shift in early Irish, one of the few 
languages which have had nasal continuants. 

I'* Though a fricative lateral occurs in Welsh (for example, in the 
name Lloyd), this sound is also voiceless. It could be that the [+cont] 



2 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

nature of the lateral is a by product of the voicelessness, and thus 
redundant. 

15 If it is indeed necessary to treat the Welsh voiceless lateral 
(see preceding footnote) as having the feature [+cont], the filter could 
be rewritten *(+lateral: +voice: +cont). Since for Spanish all laterals 
are voiced this addition would cause no problems. 

16 The trilled [r] has been argued to be derived from the single 
tap [r]. The two only contrast in intervocalic position. A common 
analysis (which the spelling suggests) is that intervocalic [r] is 
derived from an underlying geminate /rr/, though this would make 
/rr/ the only geminate in the language. Anyway, the alternation 
between the two is not conditioned by the same environments as the 
stop/spirant alternations which are the subject of this paper. For 
example, /r/ is always [r] in word-initial position, not just in phrase- 
initial position. Also, [r] appears after syllable-final consonants: 
Israel [Israel] but desde [desde]. 

17 Actually, it would be nice to find a way to refer to nasals and 
/I/ as a natural class, leaving out /r/, since /r/ does not assimilate to 
a following obstruent but /I/ does (provided the obstruent is 
coronal). Hock has pointed out (personal communication) that a 
phonetic feature such as [-interrupted] can be used to distinguish 
between [r], for which the stop gesture is [-i-interrupted] and the 
nasals and laterals, for which it is [-interrupted]. The constraint 
could be rewritten to say *(-interrupted: -t-continuant) and thus leave 
our [r], which could then be redundantly [-i-cont]. 

18 I have excluded the glides [y] and [w] from this discussion, as 
well as palatal consonants. Lozano (1979) treats glides in onset 
position as obstruents, not sonorants, so they will not be subject to 
the proposed constraint. In a syllable rime, glides are not [+con- 
sonantal], so again, they will not be subject to the constraint. 

19 Though the environment where stop [g] surfaces in Moore is 
somewhat different than that where stop [b,d,g] surface in Spanish, 
Myers shows that the blocking of Spirantization is not accomplished 
by conditions on the rule, but rather falls out from representations 
created by an independent process of the language. This is exactly 
the claim I make for Spanish. 

20 In Tigrinya, discussed in Kenstowicz (1982), Spirantization 
picks out the voiceless velars /k,q/. I do not know the entire 
phoneme inventory of the language; however, Kenstowicz points out 
that Spirantization is not a neutralizing rule in Tigrinya. Therefore 
the feature [cont] is not used contrastively for voiceless velars. If it 
is not used contrastively for other consonants, which do not 



Branstine: Stop/spirant alternations in Spanish 2 1 



spirantize, this would be a counterargument for the stance I take in 
this paper. 

21 In fact, one of the early arguments in favor of the repre- 
sentation of geminates as multiply-linked structures was in 
Kenstowicz's (1982) paper on Spirantization in Tigrinya. Whereas 
single consonants and consonants in ordinary clusters are subject to 
postvocalic spirantizaion, geminates surface as stops. 

REFERENCES 

AMASTAE, Jon. 1986. A syllable-based analysis of Spanish spiranti- 
zation. Studies in Romance Linguistics (LSRL 14), ed. by Oswaldo 
Jaeggli & Carmen Silva-Corvalan, 3-22. Dordrecht: Foris. 

Archangeli, Diana and Douglas Pulleyblank. 1986. The content and 
structure of phonological representations. Department of Ling- 
uistics, University of Arizona and University of Southern Cali- 
fornia, MS. 

BOROWSKY, Toni J. 1986. Topics in the lexical phonology of English. 
University of Massachusetts Ph.D. dissertation in Linguistics. 

Clements, G. N. 1985. The geometry of phonological features. Phon- 
ology Yearbook 2.225-252. 

1987. Phonological feature representation and the description 
of intrusive stops. CLS 23:2 .29-50. 

1987. Toward a substantive theory of feature specification. 

NELS 18:1 79-93. 

Fernandez, Joseph A. 1982 [1985]. The allophones of /b,d,g/ in Costa 
Rican Spanish. Orbis 31:1-2.121-146. 

Goldsmith, John. 1981. Subsegmentals in Spanish phonology: An 
autosegmental approach. Linguistic Symposium on Romance 
Languages 9, ed. by William Cressy & Donna Jo Napoli, 1-16. 
Washington, DC: Georgetown Univ. Press. 

Halle, Morris, and Jean-Roger Vergnaud. 1988. An essay on stress. 
Cambridge: MIT Press. 

Hamp, Eric P. 1974. Reassignment of nasality in early Irish. CLS 
10:2.127-130. 

Harris, James W. 1984. La espirantizacion en castellano y la repre- 
sentacion fonologica autosegmental. Estudis Gramaticals 1.149- 
67. 

1985. Autosegmental phonology and liquid assimilation in 

Havana Spanish. Selected Papers from the Xlllth Linguistic 
Symposium on Romance Languages, ed. by L. D. King & C. A. 
Maley, 127-148. Amsterdam/Philadelphia: John Benjamins. 

Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62.321- 
51. 



22 Studies in the Linguistic Sciences 21:1 (Spring 1991) 



HOCK, Hans Henrich. 1986. Principles of historical linguistics. Berlin: 

Mouton de Gruyter. 
HUALDE, Jose I. 1988. A lexical phonology of Basque. University of 

Southern California Ph.D. dissertation in Linguistics. 

1989. Procesos consonanticos y estructuras geometricas en 

espanol. Lingiiistica 1.7-44. 

1990. Consonantal rule interactions. Department of Spanish, 

Italian, & Portuguese, University of Illinois at Urbana-Cham- 

paign, MS. 
Kenstowicz, Michael. 1982. Gemination and spirantization in Tigrinya. 

Studies in the Linguistic Sciences 12:1.103-122. 
LOZANO, Maria del Carmen. 1979. Stop and spirant alternations: 

Fortition and spirantization processes in Spanish phonology. 

Ohio State University Ph.D. dissertation in Linguistics. 
MasCARO, Joan. 1984. Continuant spreading in Basque, Catalan and 

Spanish. Language sound structure, ed. by M. Aronoff & R. T. 

Oehrle, eds. Cambridge, Mass: MIT Press. 
M ESTER, Armin, and Junko Ito. 1989. Feature predictability and 

underspecification. Language 63.259-293. 
MOHANAN, K. p. 1989. On the bases of underspecification. Presented 

at the Kyung Hee International Conference on Linguistic Studies, 

Kwangneung campus of Kyung Hee University, Seoul, Korea, 

August 1989. 
Myers, James. 1989. 'Spirantization' in Moore. Department of Linguis- 
tics, University of Arizona, MS. 
Sagey, Elizabeth. 1986. The representation of features and relations 

in non-linear phonology. MIT Ph.D. dissertion in Linguistics. 
SCHEIN, Barry, and Donca Steriade. 1986. On geminates. Linguistic 

Inquiry 17.691-744. 
Steriade, Donca. 1982. Greek prosodies and the nature of syllabifi- 
cation. MIT PhD dissertation in Linguistics. 
1987. Redundant values. CLS 23:2.339-362. 



Studies in the Linguistic Science 
Volume 21, Number 1, Spring 1991 



COMPENSATORY LENGTHENING IN MODERN COLLOQUIAL 
TEHRANI FARSI 

Ali Darzi 

This paper deals with compensatory lengthening (CL) 
in the modern colloquial Farsi of Tehran, a process triggered 
by the deletion of glottal consonants. I show that two points 
are crucial to an appropriate analysis of CL in Farsi: (i) 
There is an asymmetry between glottal and non-glottal con- 
sonants as far as moraicity is concerned, only glottals being 
moraic. (ii) In addition to the morale tier, we need a CV tier 
(along the lines suggested in Hock 1986) for representing 
segments in order to properly account for the restrictions 
on CL. These findings undermine recent claims by Hayes 
(1989) concerning both the analysis of Farsi syllable struc- 
ture and the organization of syllabic phonology. 

1.0. The phonological structure of Farsi 

A number of facts about the phonological structure of Farsi will 
be necessary for following the subsequent arguments. Needless to 
state, the present discussion includes only those aspects of the pho- 
nological structure of Farsi that are directly related to CL, in the 
dialect to be described. 

To begin with, I give the consonantal and vocalic system of Farsi; 
cf. (1) and (2). Note that although some of the phonetic symbols 
differ from IPA, their value is the same as their counterpart in that 
system. 



(1) Consonants 














lab 


labi 


o-dent 


dent 


alveo 


alveo-pal pal 


uvul 


glottal 


stop p,b 






t,d 




k,g 


q 


7 


fricative 




f,v 




s,z 


if 


X 


h 


affricate 










90 






trill 








r 








nasal m 








n 








lateral 








1 








glide 










y 







24 Studies in the Linguistic Sciences 21:1 (Spring 1991) 



(2) Vowels 




front 
high i 
mid e 
low ae 


back 
u 
o 
a 



Both traditional grammarians and modern linguists believe that 
in Farsi the vowels /i, u, a/ are longer than /e, o, ae/. However the 
length is not phonologically significant (see Samareh 1977:92). 

There are also some diphthongs in Farsi. Samareh (1985) recog- 
nizes six diphthongs, viz. /ay, uy, oy, aey, ey, ow/, whereas Meshkat 
al-Dini (1985) refers only to /ey/ and /ow/. However, both Samareh 
and Meshkat al-Dini treat diphthongs as a sequence of a vowel and a 
glide. I will not further discuss diphthongs in this paper. 

Since morale structure and syllable structure are interrelated, I 
will briefly discuss the syllable structure of Farsi. The structure of 
the syllable is important in several respects. First, in moraic theory, 
strings of segments are syllabified before moraic structure is as- 
signed to them. Second, McCarthy (1979) has argued that syllabifi- 
cation rules apply during the derivation whenever their structural 
description is met. So, if the deletion of a segment on the segmental 
tier creates the context for syllabification to apply, it will apply. 

There is no unitary view among linguists as to the syllable 
structure of Farsi. Some believe that no syllable can begin with a 
vowel in Farsi. According to this view, the syllable structure of Farsi 
consists of CV, CVC, and CVCC. Some other linguists hold that vowels 
can also begin syllables. In this paper, I will adopt the former view 
which is supported both by native speakers' intuition and by uni- 
versal considerations according to which syllabic elements are first 
linked to the syllable nodes as nuclei, then consonants that precede 
them are attached to the syllable node in accordance with principles 
such as the SONORITY SEQUENCING PRINCIPLE, and finally the remaining 
consonants are projected into the syllable structure. (I should note, 
however, that Farsi does not respect the sonority sequencing prin- 
ciple.) Note that there are no syllabic consonants in Farsi. 

2.0. Moraic phonology 

Autosegmental phonology has developed in the past several 
years into different theories or subtheories such as CV phonology, 
moraic phonology, etc., all of which have centered around a common 
core that distinguishes them from linear phonology such as repre- 
sented in SPE. Although, they all propose multiple levels of phono- 
logical representation, they vary in some respects. 



Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 2 5 

McCarthy (1979) proposed an abstract CV tier through the me- 
diation of which consonants and vowels are organized into syllables. 
This tier characterizes phonological timing relations such as length, 
syllable weight, mora, etc. Levin (1985) took the units of this tier as 
purely timing units, proposing to use one symbol such as X for each 
unit on the CV tier and thus developing what is called X-theory. 
Hyman (1985) and McCarthy & Prince (To Appear) proposed that the 
prosodic tier has just one kind of unit, which represents moras rather 
than segments. According to Hayes, morale theory is preferable in 
two respects. First, it captures the distinction between light and 
heavy syllables, assigning one mora to light syllables and two moras 
to heavy ones. Second, the mora counts as a phonological position 
such that a segment is normally represented as doubly linked. In 
this paper, CL in modern colloquial Tehrani Farsi is analysed within 
the version of morale theory proposed by Hayes (1989). However, as 
I will show later, Hayes's framework and his view regarding the 
morale structure of Farsi must be modified. 

According to the theory, CL is defined as the lengthening of a 
segment upon the deletion or shortening of a nearby segment. This 
process is conditioned both by the position of the segment that 
undergoes deletion within the syllable and by the choice of the 
nearby segment that is to be lengthened. As a result of the process, 
the weight of the syllable in which CL occurs remains intact. It is 
therefore reasonable to assume that CL occurs in languages that have 
syllable weight distinctions. 

According to Hayes, for a string to be syllabified, certain sonor- 
ous morale segments are assigned a syllable node and other seg- 
ments in the string are adjoined to this node to form onset and coda. 
All these are subject to language-specific conditions on syllable well- 
formedness and language-specific rules that specify weight by posi- 
tion. Such rules however only apply to coda consonants. Onsets are 
assumed not to be morale in principle. 

CL, as defined above, is triggered by the deletion or shortening 
of a segment. The theory holds that if a morale segment is deleted, 
the stranded mora will be filled by the spreading of an immediately 
preceding segment, a process which may or may not lead to the de- 
linking of that segment from its own mora. However, if a segment is 
not morale, once it is deleted it is all gone, with no stranded morale 
slot to be filled by spreading. This is the case when a consonant in 
the onset position is deleted, since onsets, as mentioned earlier, are 
not morale. 

Considering the syllable structure of Farsi and its relation to the 
morale structure of syllables, Hayes (1989) proposes that syllables in 



26 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



Farsi may be trimoraic, as Farsi quantitative metrics suggest. He 
puts it this way: 

In this system, the light syllables correspond to a short 
metric position (/-/) and heavy syllables to either a long 
metrical position (/-/) or two shorts {h ^1). Superheavy 
syllables (CVVC and CVCC) are scanned as /- -/. If we 
make the usual assumptions for quantitative metrics (/-/ 
corresponds to two moras, /-/ to one), then the superheavy 
syllables must count as trimoraic. Interestingly, the ultra- 
heavy CVVCC syllables of Persian are scanned as a /--/ as 
well, suggesting that an upper limit of three moras is in 
effect. 

He then refers the reader to Elwell-Sutton (1976) and Hayes 
(1979) and proposes the moraic structure of Farsi as in (3) below, 
where "m" stands for mora. 



(3) light heavy 



superheavy 



ultraheavy 



/^^5:>^ /^^^^^ /^^^:>^ 
/mmm /mmm /mmm 

/ / V / 1 I / V I / 1 I I / ^/l 

be t a baedt a bdaestda/t 
'bad" "swing" "hand" "had" 



3.0. Compensatory lengthening in colloquial Tehrani Farsi 

With this background of moraic theory and of the phonological 
structure of Farsi, let us move to the third part of the paper that 
deals with the process of vowel lengthening in modern colloquial 
Tehrani Farsi. The forms in (4) have been divided into three sets for 
the sake of later discussion. For each set, formal conservative forms 
are given, together with their corresponding colloquial forms. Note 
that a colon is used to stand for length equal to one mora added to 
the inherent moraic length of a vowel. For example, if /o/ is inher- 
ently linked to one mora, then /o:/ would have an extra mora. In 
(4. a) the glottal consonants occur in word-final position. 

(4) a. formal conservative colloquial gloss 

rob? ro:b 'quarter' 

naef? nae:f 'benefit' 

qaet? qae:t 'cut' 

/aem? jae:m 'candle' 

Jey? je:y 'object' 

Jaer? jae:r 'religious law' 

qael? qae:l 'tin' 



Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 



27 



I 



masn? 

fasr? 

su7 

saer? 

sobh 

solh 

kuh 

In (4.b) the glottal consonants 
These words are significant in that 
tricted to word-final position. 

(4) b. qae7r 
lae7n 
/as7n 
laB71 
ro7b 
bo7d 
sae7y 
bae7s 
qaehr 
lashn 
sashm 
baehs 
raehn 
fohl 

In (4.c) the glottal consonants 
within the word. 



mae:n 

fa£:r 

su: 

sae:r 

so:b 

so:l 

ku: 



(4) c. 



tas7mir 

nae71eyn 

jae7fa£r 

nae7na 

tas71iq 

jaehla 

zaehra 

paehna 

vaehji 

maehmud 

me7mar 

7e7zam 

tehran 

behtasr 

behzad 

to7me 

Jo7be 

}o71e 



'prohibition' 

'branch' 

'bad' 

'melancholy' 

'morning' 

'peace' 

'mountain' 



precede a word-final consonant, 
they show that CL is not res- 



qae:r 

las:n 

/as:n 

lae:l 

ro:b 

bo:d 

sae:y 

basrs 

qas:r 

lae:n 

sas:m 

bas:s 

rae:n 

foj 



'bottom' 

'cursing' 

'dignity' 

'spinel ruby 

'terror' 

'dimension' 

'effort' 

(name of a 

'wrath' 

'language' 

'share' 

'discussion' 

'mortgage' 

'foul language 



party) 



occur in syllable-final position 



tae:mir 

naeileyn 

jasrfaer 

naerna 

tae:liq 

Jae:la 

zae:ra 

pae:na 

vaeji 

mae:mud 

me:mar 

7e:zam 

te:ran/te: 

be:taer 

be:zad 

to:me 

Jo: be 

Jo:le 



repair 
'slipper' 
(proper noun) 
'mint' 

'suspension' 
(proper noun) 
(proper noun) 
'width' 
'wild' 

(proper noun) 
'architect' 
'dispatch' 
runi'Tehran' 
'better' 
(proper noun) 
'prey' 
'branch' 
'flame' 



to:fe 


'present' 


so:ba£t 


'talk' 


zo:re 


(proper noun) 


ja:pur 


(proper noun) 


Ja:rox 


(proper noun) 


|a:rud 


(a city in Iran) 



28 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

tohfe 

sohbaet 

zohre 

lahpur 

Jahrox 

Jahrud 

Our first task is to determine the underlying representation of 

the words in the data. If long vowels are treated as a sequence of 

two vowels, then there is an alternation between /?/ and /ae/, /e/, 

and /o/. There is also an alternation between /h/ and /ae/, /e/, /o/, 
/a/, and in a very few cases /u/. 

For the sake of the argument, let us assume that /?/ and /h/ are 
introduced via phonological rules and that the vowels are present 
underlyingly. If so, then we cannot predict in what contexts the 
vowels in question change into /?/ and in what contexts they change 
to /h/, for there are pairs of words in which the contexts for the 
change would be identical such as [qas:r] (the surface phonetic form 
for both /qaehr/, and /qaeVr/), [lae:n] (the surface phonetic form for 
both /laehn/ and /l£e7n/), and /bae:s/ (the surface phonetic form for 
both /bae7s/ and /bshs/). 

Another analysis might be that we take long vowels as under- 
lying and formulate two rules; one to shorten the vowels in question 
and another to introduce a glottal consonant. This approach will not 
work either, since again, we would not be able to predict where to 
introduce /7/ and where to introduce /h/. There are thus two pieces 
of evidence that prevent us from postulating that the long vowels are 
present underlyingly and require us instead to take /7/ and /h/ to 
be present in the underlying representation. 

Further support for the hypothesis that glottal consonants are 
present underlyingly in forms with these long vowels is provided by 
a language game in which the last syllable of a word is transposed to 
its beginning. 2 I asked a native speaker of Farsi to switch the last 
syllable of more than a hundred words that I uttered to the begin- 
ning of the strings. There were a great number of words with long 
vowels among the words that I uttered. As I expected, all the strings 
the informant produced on the basis of the rules of the game con- 
tained a glottal consonant. Below are given the formal conservative 
forms of a few words from the data in (4), their colloquial forms 
which I produced, and the response on the part of the consultant. 
(Dashes indicate syllable boundaries.) 



conservative 


colloquial 


consultant's response 


tae7mir 


t£e:mir 


mir-ta£7 


jas7faer 


jae:far 


f2er-jae7 


zaehra 


zae:ra 


ra-zash 



Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 2 9 



jjehna 


)ae:na 


nae-paeh 


aehia 


s:la 


la-lae h 


me7mar 


me:mar 


mar-me7 


7e7dam 


7e:dam 


dam-7e7 


jo7be 


jo: be 


be-|o7 


to7me 


to:me 


me-to7 


jo71e 


jo:Ie 


le-|o7 


tehran 


te:ran/te:run 


ran-teh/run-teh 


behtaer 


be:taer 


taer-beh 


sohbaet 


so:baet 


bast-soh 


zohre 


zo:re 


re-zoh 


johraet 


jo:raet 


rast-Joh 


lahrox 


ja:rox 


rox-lah 


jahrud 


ja:rud 


rud-Jah 



Now, if there were no underlying glottal consonant in the forms 
that contain a long vowel, how is it that a glottal consonant shows up 
when the informant moves the last syllable of the colloquial forms to 
the beginning of those forms? Even more important, the consonant 
that shows up is identical to the glottal consonant appearing in the 
formal conservative forms. These facts can only be accounted for 
under a hypothesis that posits glottal consonants in the underlying 
representation of the colloquial forms, a consonant that deletes dur- 
ing the derivation. Moreover, the deletion process in question is ac- 
companied by another phonological process that lengthens a vowel in 
specific contexts. 

The loss of the glottal consonants and the lengthening of preced- 
ing vowels have been discussed by several linguists. Samareh (1977, 
1985) treats glottal consonants in coda position as weak consonants 
that are barely perceptible in colloquial speech. According to him, 
they also lead to the lengthening of the preceding vowel. This is in 
accordance with Rostargueva's observation (1975) that the glottal 
stop in Farsi is realized in careful speech as a noticeable stricture, ex- 
cept before a consonant or pause, where it results in lengthening. 

Here, three points are in order. First, neither Samareh nor Ros- 
targueva refers to /h/ as behaving in the same way as /7/ as far as 
vowel lengthening is concerned. This is probably due to the fact that 
the glottal stop has received more attention because of the con- 
troversy over whether it is contrastive in Farsi. Second, Samareh 
(1985) analyzes syllable-initial glottal stops as strong; as such, they 
do not trigger vowel lengthening. However, in his 1975 paper he 
does not rule out the possibility of intervocalic glottal deletion which, 
according to him, does not lead to vowel lengthening. Third, neither 
Samareh nor Rostargueva accounts for the lengthening of vowels 
within a particular phonological theory. 



30 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

I should note that /h/ usually behaves in the same way as /?/, 
not only in coda position, but also in syllable-initial position, such as 
in /maejhaed/ which is realized as /maejaed/ with no vowel lengthen- 
ing. Note however that in the subsequent discussion I will generally 
ignore syllable-initial glottal consonants for two reasons: (a) There 
are many words in which syllable-initial glottal consonants are not 
deleted as in /Jahed/ 'witness', /maher/ 'skillfull', /7asziz/ 'dear', etc. 
This might be why Rostargueva tends not to say anything about the 
possibility of syllable-initial glottal stop deletion, (b) Even if glottal 
consonants in onset position are deleted, they do not result in CL 
because segments in onset position are not morale. 

To account for the data in (4) in a non-linear approach, we need 
first of all a rule such as the one in (6) to delete glottal consonants in 
coda position. Then, if a mora is stranded as a result of glottal de- 
letion, it is either filled by spreading from a preceding segment or, if 
the deleted consonant does not occur next to the nucleus, the spread- 
ing segment is delinked from its mora; cf. (7). These two processes 
are distinguished as Spread and Flop, respectively. (As the deriva- 
tions in (9) [next page] illustrate. Flop must apply before Spread.) 

^^M^ } -> /_ (C)]s 

(7) mm mm 

L,^"^ Spread t*^^ ^^°P 

V C 

Now, while deletion of a glottal consonant in coda triggers CL, in 
words such as /sas7id/, with intervocalic glottal, deletion of the stop 
does not result in CL. This is of course precisely as predicted, since 
onset consonants are not morale, but linked to the following mora. 
Their deletion therefore does not result in either Spread or Flop, and 
no CL results. 

The two different situations are illustrated in Figures 1 and 2 
[next page]. Figure 1 shows how rules (6) and (7) apply to data of 
the type (4a) and (4c) above, with glottal stop or fricative in the 
coda. (To save space, I omit derivations for the type (4b) These are 
substantially the same as those for (4c).) Figure 2 shows that the 
rules fail to apply to forms with glottal stop in onset position. Note 
that the examples in both figures are assumed to have been syl- 
labified already. 



Darzi: Compensatory lengthening i 


n modem colloquial Tehrani Farsi 


/^v\ /^""^^ /^''=>- /^"^ 
/mm immm immm immm 

III Iv 1 / 1 1 

t£7 m i r raB7d naef7 


GD s 

/T" 

t ae 


r £ d n ae f 


Flop 


/ m m m 

n ae f 


Spread s 

Iv 

t X 


/mmm /mmm 

/ y 1 y 1 

r ae d n ae f 


PF [t£B:mir] 


[rs:d] [n£:f] 



31 



Figure 1 



UR s 



/m / m m] / m /mm 

/I /V /I /v 



s ae 7 i d 



s s 

A vl 



7 i d 



s ae Id 

Flop 

Spread 

PF: [saeid] [sae7id] 

Figure 2 



32 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



The analysis proposed so far runs into difficulties once we 
consider the additional data in (8). In these forms a consonant in the 
coda deletes, but its deletion does not result in CL. The forms in (8. a) 
show an alternation between a syllable-final stop preceded by a fric- 
ative and its absence. 

(8) a. 



formal 


conservative 


colloquial 


gloss' 


daest 




daes 


'hand' 


haeft 




haef 


'seven' 


gereft 




geref 


'(he) got' 


loxt 




lox 


'naked' 


mozd 




moz 


'wage' 


bist 




bis 


'twenty' 


saxt 




sax 


'(he) built' 


xolk 




xo/^ 


'dry' 


mojt 




moj 


'fist' 


dozd 




doz 


'thief 


dasstgire 


daesgire 


'handle' 


(8.b) 


show an alternation between 


a [+anter 


final consonant and its 


absence. 




qaend 




qaen 


'sugar' 


7sz 




7ae 


'from' 


fekr 




fek 


'thought' 


jokr 




/ok 


'thanks' 


kond 




kon 


'slow* 



(8) b. 



Now, an insertion account of the C~0 alternations clearly is not 
plausible. We must therefore acccount for the alternations in terms 
of deletion. We can formulate a rule to delete III and /d/ in coda 
when preceded by a fricative; cf. (9). Let us call this rule Consonant 
Deletion (CD). 



(9) 



-con 
-i-cor 
-»-ant 



0/ 



[-son 
+con 



]- 



# Consonant Deletion (CD) 



However, the problem that is still with us is that although a 
consonant has been deleted from the coda, there is no CL in the 
colloquial forms. The analysis proposed earlier, however, predicts 
that deletion of the final consonant of the word for 'hand' should 
result in CL just as does the deletion of final glottal consonants in 
Figure I above. Compare the derivation in (10) [next page]. 



Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 3 3 

(10) UR s s s 

/^^=:^ CD Z^'^- Flop f^=>>- 
/mmm /mmm /mmm 

/ 1 1 1 ^ / 1 I -- l\v 

daest daes daes 



Spread /^^^--^ PF *[dae:s] 

ly I 

d ae s 

One may hypothesize that syllable-final coronals are extra- 
syllabic and therefore not moraic. This is not a plausible hypothesis, 
however, because there are many, many words in colloquial Farsi 
with syllable-final coronals, such as /qaebr/ 'tomb', /faeqr/ 'poverty', 
/hasbs/ 'custody', /jens/ 'material', etc. There are two a priori poss- 
ible alternative analyses for these forms. Under the first analysis, 
we may assume that the forms in (8.b) are exceptional and that /t/ 
and /d/ are not moraic. (I ignore /k/ at this point.) So once they 
delete there would not be a stranded moraic slot to trigger Flop or 
Spread. 

Under the second analysis, we may assume that only glottals are 
moraic in colloquial Farsi. If we do so, we do not need to treat the 
forms in (8.b) as exceptional. Moreover, the second analysis is pref- 
erable to the first one in that it more highly restricts the assignment 
of three moraic slots to just those syllables in Farsi that contain 
glottals in the coda. Given the crosslinguistc markedness of trimoraic 
syllables, this restriction would seem highly desirable. Note that this 
approach assumes that the moraicity of consonants is a language- 
specific phenomenon. But what is important is the fact that such a 
language-specific approach is dictated by the evidence of colloquial 
Farsi: As we have seen, only the deletion of glottal coda consonants 
results in CI, while the deletion of other coda consonants does not. 

If this line of argumentation is correct, then the moraic structure 
of Farsi as proposed by Hayes (cf. (3) above) will have to be modified 
as in (11) so as to capture the range of facts in the colloquial version 
of the language. (Note that following Hayes, I am still assuming that 
non-moraic consonants in the coda are linked to the preceding mora- 
ic slot.) 



34 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(11) light heavy super heavy 

s s s s s 

h /^r\ /V\ ^\ /^^^r^ 

/m /mm /mm /mm /mmm 

// IV\ /M /M\ / V I 

be t ab rob7da|t s u 7 

"to" "swing" "quarter" "had" "bad" 

This analysis makes it possible to account for forms such as the word 
for 'hand', as is clear from the following derivation. 

(12) UR 
CD A Flop & Spread inapplicable 

d ae s t d ae s 

However, if we accept the hypothesis that non-glottal coda 
consonants are not mora-bearing, the analysis in Figure 1 above for 
forms of the type (4. a), with vowel + oral consonant + glottal con- 
sonant (such as /naef?/ -♦ /njerf/) becomes unacceptable. Instead of 
the UR in Figure 1, reproduced as (13. a) below, we now have to pos- 
tulate a structure as in (13.b), in which /f/ is not mora-bearing, but 
merely asociated with the preceding mora-bearing element. 

(13) a. s_ 

/^^^^ 
/mmm 

/I I I 

n ae f 7 

In this form, once the glottal stop is deleted, if /f/ flops to the 
empty moraic slot, there will not be any empty morale slot for the 
vowel to spread to and the vowel length of the surface form /nae:f/ 
cannot be derived. Moreover, if we assume that because it is not a 
mora-bearing consonant, /f/ cannot flop to the empty moraic slot left 
by the loss of the glottal stop, then, according to Hayes (1989), /ae/ 
cannot spread to this empty position, for this would require crossing 
the association line between /f/ and the moraic node or between /f/ 
and the syllable node (if it is linked to the syllable node). 

However, there will be no crossing of association lines if (i) fol- 
lowing Hock (1986), we assume that in addition to a moraic tier, we 
have a CV tier as well, and that (ii) non-moraic consonants are not 
linked to a preceding moraic slot. Once we assume that moras are 




Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 



35 



not constructed directly on top of segments, but rather are on a tier 
separate but linked to the skeleton, the spreading of a segment on 
the morale tier does not cross the association lines between the 
elements of the CV tier and the syllable node or morale slots, because 
they are on two separate but related tiers or planes. Under this 
analysis, we no longer need the flop rule formulated earlier. The 
major problem here is one of presentation, of how to present an 
essentially three-dimensional diagram on a twodimensional sheet of 
paper. An attempt to do just that is presented in (14) below in the 
derivation of the word for 'benefit'. With a certain loss of informa- 
tion, we can represent the derivation of this word even two-dimen- 
sionally, as in (15) below. Note that in both of these representations 
there is no crossing of the association lines. (For ease of exposition, 
the internal structure of the syllable is here ignored.) 



(14) UR 




Spread 




PF: [nae:f] 



(15) UR C V C C 

I I I I 0° 

n ae f 7 "^ 



C V C C V C 

I I I Spread | | | 

n as f ~^ n as f 

I r- 



m PF: [ns:f] 



The stipulation of a CV tier in addition to the moraic tier 
undermines Hayes's argumentation against X-theory and in support 
of moraic theory. As is well known, Hayes supports his argument by 
showing that X-theory would incorrectly predict a change from /a la/ 
to [la:] in Middle English, for once the [9] deletes the /I/ would flop to 



36 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

the empty X-slot, leaving behind another empty X-slot for the final 
vowel to spread to. Compare (16) below, adapted from Hayes 1989. 

(16) mm m m m 

I A A A A 

NON ON ON ON 

III- M-. II- II 

XXX XXX XXX XXX 

III M Ml IN 

# a 1 a # 1 a la la PF: *[la:] 

He then gives the derivation in (17), showing that a morale theo- 
ry is able to derive the correct form [la], since under this analysis the 
spreading of /a/ to the stranded morale slot is impossible due to the 
ban on the crossing of association lines. 



(17) 



s s 

/ / 

/m -^ m /m 

/l /I 



m /m -^ m /m 

I 

# a 1 a # I a PF: [la] 

While an approach couched purely in terms of X-theory thus 
may have to be ruled out, it does not necessarily follow that we must 
adopt a pure morale theory. In fact, as noted above. Hock (1986) 
employs a 'two-tier' approach, combining a skeletal with a separate 
morale tier. Although Hayes has chosen not to adopt Hock's proposal 
he has not ruled it out either, stating merely that he has seen no 
compelling evidence in favor of the two-tier approach. 

By showing that the colloquial Farsi of Tehran requires such a 
two-tier aproach, the present paper should constitute a first step 
toward providing such evidence. 

4.0. Summary 

This paper has presented an investigation of CL in colloquial 
Tehran! Farsi within the morale framework. I have shown that 
morale theory runs into serious difficulties if Hayes's proposed 
account of the morale structure of Farsi is accepted. His proposal 
needs to be modified such that among the consonants, only the glot- 
tals are taken to be morale. This modification, in turn, makes it ne- 
cessary to replace Hayes's exclusively morale approach by a two-tier 
analysis along the lines suggested by Hock (1986), with both a skel- 
etal and a morale tier. 



Darzi: Compensatory lengthening in modem colloquial Tehrani Farsi 3 7 

NOTES 

*I am indebted to Charles Kisseberth, Hans Henrich Hock, and 
Jennifer Cole for their comments. Any shortcoming in this paper is 
mine. 

1 The alternation here is not relevant for the purpose of this 
paper. 

2 My friends and I used to play this language game in our 
neighborhood to call one another when I was 9 years old or so. 
However, the consultant did not know the game before I taught her. 
In teaching her the game, I did not include any of the words that 
involved CL in colloquial Farsi to make sure that she was not given 
any clue as to how to deal with words such as those in (5). 

REFERENCES 

Clements, G. N. 1986. Compensatory lengthening and consonant 
gemination in Luganda. Studies in compensatory lengthening, 
ed. by L. Wetzels & E. Sezer. Dordrecht: Foris. 

, & Keyser, S. J. 1985. CV phonology. Cambridge, MA: MIT press. 

Elwell-Sutton,. 1976. The Persian meters. Cambridge, MA: Cam- 
bridge University Press. 

Hayes, B. 1979. The rhythmic structure of Persian verse. Edebiyat 
4.193-242. 

1989. Compensatory lengthening in moraic phonology. Lin- 
guistic Inquiry 20:2.253-307. 

HOCK, Hans Henrich. 1986. Compensatory lengthening: In defense of 
the concept 'mora'. Folia Linguistica 20.431-460. 

HYMAN, Larry. 1985. A theory of phonological weight. Dordrecht: 
Foris. 

McCarthy,. 1979. Formal problems in Semitic phonology and 
morphology. MIT Ph.D. dissertation in linguistics. 

, & A. Prince. To Appear. Prosodic morphology. MS University of 

Massachusetts, Amherst, and Brandeis University, Waltham, MA. 

MESHKAT AL-Dini, M. 1985. Sakhte Ava'iye Zaban. The Ferdowsi 
University of Mashhad Press. 

ROSTARGUEVA, V. 1964. A short sketch of the grammar of Persian, 
transl. by S. P. Hill and ed. by H. H. Paper. Indiana University 
Publications in Anthropology and Linguistics, 29. (= Supplement 
to the International Journal of American Linguistics.) 

Samareh, Y. 1977. The arrangement of segmental phonemes in 
Farsi. Tehran: Tehran University Publication. 

1985. Ava Shenasiye Zabane Farsi. Tehran: Markaze nashre 
daneshgahi. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



ON THE PHONOLOGYMORPHOLOGY INTERACTION IN 
BRAZILIAN PORTUGUESE VOWEL HARMONY* 

Barbara J. Hancin 

This paper provides an analysis of vowel height 
harmony in Brazilian Portuguese by adopting a unified 
account of segment representation, building from current 
theoretical approaches. It is argued that application of the 
harmony rule is triggered and constrained by morpho- 
logical factors. The harmony rule is shown to be 
noniterative, left-ward spread triggered by suffixation 
and, contrary to harmony processes in other Romance 
languages and dialects, is not directly affected by [ATR] 
specification. Evidence is given for the tautomorphemic 
nature of the OCP. In short, this analysis provides 
significant evidence for the dependence of phonology on 
morphology in Brazilian Portuguese vowel height 
harmony. 

1. Introduction 

The process of vowel height harmony has eluded linguists of 
Brazilian Portuguese (hereafter, BP) phonology for years. Previous 
attempts at explaining the phenomenon have failed to capture the 
significant systematicities underlying the process (Harris 1974, 
Lemle 1974, Lipski 1973, Redenbarger 1981, Viegas and Assis Veado 
1982). The goal of this paper is to show that a principled account of 
vowel raising in BP is, in fact, possible. What seems to have eluded 
previous analyses is the strong dependence of the harmony rule on 
morphological factors. It is argued here that morphological structure 
plays a critical role in rule application and constraint, going against 
claims for the autonomy of phonology and morphology. This paper 
shows that only when we recognize and accept the interaction of 
phonology and morphology can we explain vowel quality variability 
in BP. 

We begin with a discussion of the theoretical approaches 
adopted which, taken together, form a unified account of segment 
structure throughout phonological derivation. In the next section, 
data is presented and an initial harmony rule is formulated. Then, 
transparent and opaque segments are discussed. In the final section, 



40 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

evidence is presented for recognizing a cyclic/noncyclic distinction 
between suffixes of this language. 

2. Theoretical approaches 

This analysis is based on a convergence of three current 
theoretical approaches in phonology, an interaction which, according 
to Goldsmith (1990), has proven to be quite productive. First, with 
respect to segment structure, it is assumed that segments and 
features are represented on separate tiers (Goldsmith 1976) and that 
the distinctive features of a segment are organized in a hierarchical 
structure, each on independent tiers, interconnected under a root 
node (see Clements 1985 and Sagey 1986 for approaches to Feature 
Geometry). In addition, it is assumed that these hierarchical struc- 
tures are not fully specified in their underlying representation. 
Underlyingly, segments only contain unpredictable information on 
feature specification and predictable information is filled in by rules 
during phonological derivation (Archangeli 1984, 1988; Archangel! 
and Pulleyblank 1986, 1987). Finally, as is shown in section 5, BP 
offers evidence for positing a grammar in which phonology and 
morphology interact, similar to the boundary type distinctions made 
by Chomsky and Halle (1968) or to the cyclic/noncyclic distinction 
made in Lexical Phonology (Kiparsky 1982, 1985; Mohanan 1985)i. 
By assuming these three general approaches, we not only can claim a 
unified account of segment representation, but also can emphasize 
the relationship between phonology and morphology in lexical 
derivation. 

3. Data and rule formulation 

Height harmony in BP is characterized by the agreement in 
height between root and initial suffix vowels. As the data in (1) 
show, root vowels alternate between [-hi] and [-f-hi] depending on the 
feature specification of the initial suffix vowel. Non-verbs (A) and 
verbs (B) are presented separately here to show the breadth of rule 
application [For ease of exposition, only vowels are represented 
phonetically]. 

(1) (A) Non-verbs2 

1. gord-oti 'fat person' gurd-iicha 'fat woman' 

2. alegr-eto 'happy way' aligr-ia 'happiness' 

3. tors-edor 'one who cheers' turs-ilhau 'a big cheer' 

4. tors-Su 'a big cheer' turs-imentu 'cheering' 

5. dez 'ten' diz-i-n6vi 'nineteen' 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 4 1 

(B) Verbs 

Infinitive 3rd pers. 3rd pers. past gloss 

sing, past sing. imp. partic. 

1. beb-er beb-eu bib-ia bib-idu 'to drink' 

2. trem-er trem-eu trim-ia trim-idu 'to tremble' 

3. com-er com-eu cum-ia cum-idu 'to eat' 

4. sofr-er sofr-eu sufr-ia sufr-idu 'to suffer' 

Given this data, we can consider either of two rules: (1) [-hi] spread, 
or (2) [+hi] spread. If we posit that [-hi] suffixes are specified 
underlyingly and that [a] is actually specified as [-hi] in the 
underlying representation (see (1)A.4), we could posit a rule of [-hi] 
spread with lexical default values of [+hi]. However, as is shown later 
(Section 4), [a] does not behave similarly to other [-hi] vowels in the 
harmony process, whereas both [+hi] vowels do behave similarly. As 
a working hyypothesis, then, we posit that [+hi] is specified 
underlyingly and a [+hi] suffix vowel predictably spreads its height 
feature to an unspecified root vowel. Adopting a [+hi] spread rule for 
BP is preferred over [-hi] spread because other closely related 
languages have also been shown to have [+hi] spreading rules (cf. 
Vago 1988 for Pasiego Montafies Spanish and Hualde 1989 for 
Tudanca Montaiies and Lena Bable). Root vowels which do not 
receive a height specification by the spreading rule will receive the 
default value [-hi]. The underlying representation for BP vowels is 
given in (2). 

(2) Underlying representation for vowels^ 

i e/e a o/o u 

hi + -I- 

lo + 

back + -I- 

The chart suggests that there are two types of mid vowels: [-ATR] 
and [+ATR]. In fact, these two types do contrast in stressed segments 
(Harris 1974, Major 1985, Pizzini 1982, Redenbarger 1981, Wetzels 
1988). However, it will be argued later that the [ATR] distinction is 
not an underlying property of the segment, rather it is an underlying 
property of a morpheme, as first proposed in Wetzels (1988). Thus, 
the [ATR] feature is not relevant to the above vowel chart. 

As shown in the data in (1), the height of the root vowel 
depends on the height of the initial suffix vowel, so we can posit that 
the application of the harmony rule is triggered by morphological 
affixation. The preliminary rule is formulated in (3). 



42 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(3) Preliminary harmony rule: [+hi] spread"* 

trigger: [+hi] initial suffix vowel 

target: vowels unspecified for height 

domain: 

default : insert [-hi] 

b E b - i a 'to drink- 3rd person sing. 

I I I imperfect' 

X XX 

\ I I 
\ I I 

\ I 1 

[+hi] [+lo] 

The domain of the rule is not yet clear. Because a suffix vowel 
affects a root vowel, we can at least assume left-ward spread. 
Additional data in (4) show that the rule appears to spread more 
than once to the left. 

(4) 1. form6s-a 'beautiful' furmus-ura 'beauty' 

2. prefer-encia 'preference' prifir-ivel 'preferable' 

3. repet-enti 'one who repeats' ripit-idor 'same' 

4. merec-er 'to deserve' miric-idu 'deserved' 

However, when the targets are not identical, the rule only spreads 
once to the left, as the data in (5) show. 

(5) 1. profet-anti 'prophet' profic-ia 'prophesy' 

2. telefon-ema 'phone call' telefun-ista 'phone operator' 

3. promet-er 'to promise' promit-idu 'promised' 

4. oferec-edor 'one who ofiric-imentu 'offering' 

offers' 

The difference between the target root vowels of (4) and those of (5) 
is that the targets of (4) agree in the feature [back] (and are 
underlying not specified for height). It appears, then, that the 
features [back] and [hi] demonstrate a kind of relationship which is 
captured if we assume they are organized on the same articulator 
node tier, namely the dorsal, as Sagey (1986) has done. By adopting 
this representation, then, the target segments in (4) will be identical 
and adjacent on the dorsal tier, a configuration which is disallowed 
by the Obligatory Contour Principle (OCP) (Goldsmith 1976, 1990; 
McCarthy 1986; Yip 1988). Because the OCP prohibits adjacent, 
identical segments, these feature bundles will actually be associated 
to only one segment underlyingly. Due to the linking of adjacent, 
identical hierarchical structures (feature bundles) to one segment, it 
becomes clear that the domain of spread is only once to the left, 
though the target segment may be multiply-linked. 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 43 

There is further clarification, however. The OCP predicts that 
any adjacent, identical feature configurations under the dorsal node 
will be linked to the same segment which can be a target for the 
spreading rule. The harmonizing data in (6), then, are not predicted. 

(6) (A) Non-verbs 

1. fono-l6g-icu 'phonologic' fono-lug-ia 'phonology' 

2. gost-6s-u 'tasty' gost-us-iira 'tastiness' 

3. astro-n6m-icu 'astronomic' astro-num-ia 'astronomy' 

(B) Verbs 



Infinitive 


3rd pers. 
sing, imperf. 


3rd pers. sing, 
future conditional 
(infinitive + I'a) 


gloss 


1. beb-er 


bib-ia 


beb-ir-ia 


'to drink' 


2. tors-er 


turs-ia 


tors-ir-ia 


'to cheer' 


3. comet-er 

4. oferec-er 


comit-ia 
ofiric-ia 


comet-ir-ia 
oferec-ir-i'a 


'to commit' 
'to offer' 



Because the vowels to the left of the trigger are identical, they 
conceivably should be multiply-linked to the same segment which is 
a target for the spreading rule. However, these data show that 
segments are not multiply linked across morpheme boundaries. This 
is evidence that the domain of the OCP is crucially tautomorphemic. 
The example in (6)B.4 shows that two root vowels are linked to one 
segment, but that segment is not affected when there is a morpheme 
intervening between it and the trigger. That the domain of OCP is 
morpheme-bound provides further evidence for the critical role of 
morphology in the phonology of BP. 

At this point, we can summarize what has been presented by 
restating the spread rule and filling in the specification for its 
domain. 

(7) Revised harmony rule: [-t-hi] spread 

trigger: [+hi] initial suffix vowel 

target: segment unspecified for height 

domain: noniterative, left-ward spread 

default: insert [-hi] 

r E p E t - i d O r 'one who repeats' 

\ / I I 

X XX 

\ I 

\ I 

\ I 

[+hi] 



44 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Having posited the basic parameters of the rule, we can now 
turn to particular phenomena which introduce complications for this 
analysis, namely transparent and opaque segments. 

4. Transparent and opaque segments 

The data in (8) show that [+hi] vowels are transparent to the 
harmony process; the rule is allowed to affect a segment beyond (to 
the left of) a [+hi] segment. 

(8) 

1. medic-olegal 'medicolegal' midic-ina 'medicine' 

2. benefic-amenti 'beneficially' binific-ial 'beneficial' 

3. oportun-amenti 'opportunally' opurtun-i-dadi^ 'opportunity' 

According to the theoretical postulates of this analysis, a [+hi] 
segment would not be a target for the spread rule, since it is already 
specified for height. Additionally, the height feature could not 
spread through the underlyingly linked [+hi] segment since that 
would entail an illicit crossover of association lines. Even if it could 
somehow spread past the segment, it would violate the Locality 
Condition, posited by Archangeli and Pulleyblank (1987), which says 
that a rule can only apply if a specified target is adjacent to a 
specified trigger. In order to solve this, the feature [+hi] will be 
allowed to spread to any segment to its left, regardless of whether or 
not the target segment already has a height specification. If [+hi] 
spreads to a segment already specified with [+hi], then the target 
segment gets mobilized to also spread its feature once to the left. 
That is, if a [+hi] suffix vowel spreads to a [+hi] root vowel, it 
reinforces the root vowel so that it can also be a trigger when, prior 
to suffixation, it was not. This process is represented in (9). 

(9) [+hi] feature spread can trigger a root [+hi] to spread 



E d i c- i 


n 


a 


'medicine' 


1 1 1 
XXX 




1 
X 




\ 1 \ 1 




1 




\ 1 \ 1 




1 




\ 1 \ 1 




1 




[+hi] [+hi] 




[+lo] 





By redefining the harmony rule to spread [+hi] to any segment on its 
left, we not only can explain the transparency of [+hi] vowels, but we 
also make the rule more generalizable. This account seems to be the 
simplest one, given the transparency of [+hi] segments in BP.^J 

Up to now, we have accounted for spread to segments 
unspecified for height and to segments specified for [+hi]. What 
remains to be seen is the [a] or (underlyingly) [+lo] vowel. The 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 



45 



current rule predicts that segments specified for [+lo] underlyingly 
will be opaque to [+hi] spread because a segment cannot have a 
contradictory *[+lo, +hi] specification for height on the dorsal tier. 
Thus, if a target segment is already specified for a height which is 
not compatible with [+hi], then [+hi] does not spread. That is exactly 
what the BP data (10) show. 8 



(10) [+lo] vowels do not undergo harmony 

'a bunch of flies' musk-inhu 



mosk-ar-ia 

telefon-ad-inha 

neces-ar-ia-menti 

letarg-u 

moral 

compat-ivel 



'phone call-dim.' telefun-ia 



7. local-iz-ar 



'necessarily' 
'lethargy' 
'moral' 
'compatible' 

'to locate' 



nicis-it-ar-iu 
letarg-ia 
moral-i-dadi 
compat-ibil-i- 

local-i-dadi 



'fly-dim.' 
'phone call' 
'necessary' 
'lethargy' 
'morality' 
dadi 
'compatibility' 
'locality' 



An additional opaque domain in BP is that of a stressed vowel. 
The data in (11) demonstrate that stressed segments, like [+lo] 
vowels, are not affected by, nor are they transparent to [-t-hi] spread. 
The data show that vowel harmony may occur to the left of a 
stressed segment ((11). 1-6) or to the right of a stressed segment 
((11).7-10). 



(11) 
1. 

2. 
3. 
4. 
5. 
6. 



Stressed segments block vowel harmony 

redond-u 'round' 

benefic-u 'beneficial' 

prop6s-it-u 'proposal' 

form6s-u 'handsome' 

polut-u 'polluted' 



redund-inhu 'round-dim.' 
binific-ial 'beneficial' 

prupus-i9-au 'proposition' 
furmus-ura 'beauty' 
pulu-ir 'to pollute' 

pricip-it-ar(se) 'to rush 

headlong' 



precip-it-i 'in danger 
of falling' 

7. peri6d-icu 'periodic' periud-u 'period' 

8. astro-l6g-icu 'astrologic' astr6-lug-u 'astrologer 

9. iconom-icu 'economic' ic6num-u 'economist' 

10. fosfor-6s-u 'phosphorous' f6sfur-u 'match' 

A stressed segment does not undergo vowel harmony nor does it 
allow the height feature to spread any further to the left (see 
especially examples (11) 2-4 where the stressed segment and the 
one to its left agree in backness). Given the facts that a trigger does 
not have to be stressed and that harmony can occur to either side of 
a stressed segment, we can conclude that stress plays no other role in 
the harmony process than to prohibit a stressed segment from 
undergoing raising. However, accounting for the opacity of a stressed 
segment is quite simply problematic. Even if positing filters were 
desirable, we could not posit the filter *[-(-stress, -i-hi] because BP 



46 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



clearly allows stressed [+hi] vowels (see (11) 1, 4, 5, and 7). Also, 
because stress is a suprasegmental feature which is not present 
underlyingly, but which is admitted to a morpheme by way of 
predictable rules, we cannot follow the same line of argumentation 
that was given for the opacity of [+lo] vowels. For now, then, we will 
simply have to stipulate that target segments are necessarily [stress]. 
The [+hi] feature does not spread to a stressed segment. 

This stipulation presents a problem for adjacent, identical 
vowels which are proposed to be underlyingly linked to the same 
segment. The challenge lies in explaining pairs where root vowels 
are the same in unstressed positions, but are different when one gets 
stress like in (12). 

(12) A B 

1. prupus-it-ar 'to propose' prop6s-it-u 'proposal' 

2. icunum-ia 'economy' ic6num-u 'economist 

3. fosfor-ismu9 'phosphorism' f6sfur-u 'match' 

We will retain the assumption that the OCP links adjacent, identical 
vowels to the same segment underlyingly, since that is proposed to 
be a universal principle. What we propose, however, is that stress 
delinks a vowel from a multiply-linked segment, giving the stressed 
vowel an independent status on the skeletal tier. That will explain 
why like root vowels act in concert when unstressed (12A), but lose 
this property when one gets stressed (12B). 

We may now give a final statement of the harmony rule as in 
(13). 

(13) Final harmony rule: [+hi] spread 

trigger: [+hi] initial suffix vowel 

target: unstressed segment 

domain: noniterative, left- ward spread 

default: insert [-hi] 



[+hi] 

O n O m 

\ / 

\ / 

X 

\ 
\ 



1 
I 
I 
X 

I 
I 
I 

[+hi] 



economy 



i c n 

1 


m-u 

1 1 


1 
X 


1 1 
X X 

\ 1 
\ 1 
\ 1 
[+hi] 


'economist' 





Hancin: Phonology-morphology interaction in Brazilian Portuguese 4 7 

In any discussion of BP vowels, reference must be made to the 
feature [ATR]. As mentioned earlier, [ATR] is contrastive in stressed 
positions, and therefore must be present in some form underlyingly. 
However, the occurrence of [-ATR] is difficult to predict (Cunha 1976, 
Major 1985, Pizzini 1982, Wetzels 1988). Harris (1974) discussed 
the verb paradigm and suggested that root vowels that are 
underlyingly [e] and [o] undergo a combination of rules (like 
harmony, lowering, and neutralization) in the course of derivation to 
exhibit surface forms which are [e] and [o] in unstressed positions 
and [e] and [o] in certain stressed positions, depending on the theme 
vowel and the verb tense/number/person. Pizzini (1982) and 
Wetzels (1988) have also discussed these open and closed vowels 
and have concluded that, aside from a metaphony rule which 
produces sets like [nov-u] 'new-masc. sing.', [n6v-a] 'new-fem. sing.', 
[nov-us] 'new-masc. plur.', [n6v-as] 'new-fem. plur.' and [ispant-ozu] 
'scary-masc. sing.', [ispant-6za] 'scary-fem. sing.', [ispant-6zus] 'scary- 
masc. plur.', [ispant-6zas] 'scary-fem. plur.', some morphemes have 
the [-ATR] specification underlyingly. Wetzels (1988) presents the 
most comprehensive analysis to date and thus will be the one 
adopted here. The following presentation of the feature [ATR] in BP 
will show that this feature does not affect vowel raising, rather it is 
the quality of stress which disqualifies a segment from the harmony 
process. 

Wetzels, in following similar analyses by PuUeyblank (1986, 
1988), Vago (1988), and others, proposes that the feature [ATR] is 
not a quality of a segment, rather it is a quality of a morpheme. That 
is, the feature [ATR] has been extracted out of the underlying 
representation of features and defined as a floating segment in a 
morpheme. Because [ATR] is contrastive in BP, there are two types 
of morphemes as represented in (14). 

(14) A. [-ATR] B. [-hATO] 

The floating feature is not associated to the morpheme in the 
underlying representation. Because [-ATR] only appears in a 
stressed syllable, stress assignment occurs before association of the 
floating feature. After stress has been assigned, the feature gets 
associated to the stressed vowel by what Wetzels calls 'floating- 
feature-to-stress-linking', a rule which he proposes to be part of 
Universal Grammar (1988:20). Vocalic segments which do not get 
associated with this floating feature will receive the unmarked 
default value of [-t-ATR]. 

This analysis suggests that root vowels can be underlyingly 
identical for the features on the dorsal tier. However, when stress is 



48 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



assigned, the stressed vowel will delink from any segment it shares 
with another vowel and will relink as an independent segment on 
the skeletal tier. The floating [ATR] feature gets associated, making 
the segment either [+ATR] or [-ATR]. However, it is not clear that the 
floating feature association is ordered with respect to vowel 
harmony, since it has been shown that [+hi] will not spread to a 
stressed domain. For the purposes of this analysis, then, there is no 
ordering constraint between floating feature association and vowel 
harmony. The examples in (15) and (16) illustrate the phonological 
derivation of relevant words. 



(15) 


[-ATR] 


UR 


fOsfOr-u 

\ / 1 
X X 




[+hi] 


Stress 
assign. 


[-ATR] 

fOsfOr-u 

\ / 1 
X X 




[+hi] 


feature 
assoc./ 
vowel 
harmony 


[-ATR] 

/ 
fOsfOr-u 

1 1 1 
XXX 

\ 1 
\l 

[+hi] 


Default 
rule 


fosfur-u 




1 1 1 
XXX 




[-hi][+hi][+hi] 




'match' 



[-ATR] 

fOsfOr-icu 

\ / II 

XXX 

I I 

[+hi][+hi] 
[-ATR] 

fOsfOr-icu 

I III 
X X XX 

I I 

[+hi][+hi] 

[-ATR] 
I 

fOsfOr-icu 
I I I I 
X X X X 

1 I 
I I 

[+hi][+hi] 



fosfor-icu 
I I I I 
X X X X 

I I I \ 
[-hi][-hi][+hi][+hi] 

'phosphoric' 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 



49 



(16) 





[-ATR] 


[-ATR] 


UR 


prOpOs-it-u 

\ / 1 1 
XXX 


prOpOs-it-ar 

\ / II 
XXX 




[+hi][+hi] 


[+hi][+lo] 


Stress 


[-ATR] 


[-ATR] 


assign. 


prOp6s-it-u 

1 1 1 1 
X X X X 


prOpOs- it- ar 
\/ 1 1 
X XX 




[+hi][+hi] 


[+hi][+lo] 


feature 


[-ATR] 


[-ATR] 


assoc./ 

vowel 

harmony 


prOp6s-it-u 

1 1 1 1 
X X X X 


prOpOs-it-ar 

\ / II 
XXX 

\ 1 1 
\ 1 1 

[+hi][+lo] 




[+hi][+hi] 


Default 






rule 


prop 6s- it- u 

1 1 1 1 
X X X X 


prupus-it-ar 

\/ 1 1 
X XX 

1 1 \ 

[+hi][-^hi][-Ho] 




[-hi][-hi][+hi][-f-hi] 




'proposal' 


'to propose' 



By analyzing the feature [ATR] as a quality of a morpheme 
rather than of a segment, we represent its lack of effect on the vowel 
harmony process. What does affect vowel harmony is stress, which 
delinks any vowel underlyingly linked to a multiply-linked segment 
(OCP). The association of the floating feature is to the stressed 
segment, which is not a target to vowel harmony anyway. In short, 
due to the lack of interaction between the feature [ATR] with the 
feature [+hi], we conclude that there is no evidence that the feature 
[ATR] affects the vowel harmony process. 

5. Evidence for the cyciic/noncyclic distinction 

Not all suffixes which begin with a [-i-hi] vowel trigger the 
harmony process. Examples in (17) show roots which undergo vowel 
harmony with some suffixes, but not with others. 



5 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

(17) 1. pu[k]-inhu 'small-dim.' poc-isimu 'small-super.' 

2. amur-zinhu 'love-dim.' amor-os- 'lovingness' 

i-dadi 

3. ofind-idu 'offended' ofens-ivu 'offensive' 

4. fosfur-u 'match' fosfor-ismu 'phosphorism' 

5. is[k]erd-u 'left' is[k]erd-ista 'leftist' 

Preliminary analysis shows that roots like -ivo, -issimo, -iadj.. 
and -ilho consistently do not trigger harmony, though they have a 
[-i-hi] initial vowel. This fact can be explained by recognizing a 
distinction between different types of derivational morphemes. The 
suffixes may differ either in boundary type as described in Chomsky 
and Halle 1968, or in cyclic vs. noncyclic application, as proposed in 
Kiparsky's 1982, 1985 Lexical Phonology, or generally in being 
triggers or nontriggers of vowel harmony. For the purpose of illus- 
tration, let us adopt a Lexical Phonology approach. 

Kiparsky (1982, 1985) proposes that the lexicon is divided into 
strata on which specific kinds of morphemes can be affixed to a 
word. Within a stratum, phonological rules apply on the morpho- 
logical input if the appropriate contexts are available. These rules 
are subject to the Strict Cycle Condition (SCC) which says that the 
domain of a rule is restricted to a word derived in the stratum in 
which the rule is contained. The rules are divided into cyclic, which 
are contained on one stratum, and noncyclic, which are on a different 
stratum. We suggest, then, that the vowel harmony rule is cyclic and 
is contained on the level where morphemes such as -imento, -inho, 
-it-, -ia are affixed. Suffixes which do not trigger vowel harmony are 
affixed on a stratum which does not contain the vowel harmony rule. 
The interesting point about the noncyclic affixes is that they do not 
change vowel feature specifications set in the previous stratum. The 
examples in (18 [next page]) will make this more clear. We assume 
that bracket erasure and tier conflation are the same processes. 

Placing certain affixes on a separate cycle explains their inability 
to trigger vowel harmony. It also explains why [-ATR] segments 
sometimes DO appear in unstressed positions, such as the case in (18) 
where an adjective is made into an adverbial with the addition of the 
suffix -menti, which induces stress reassignment to penultimate 
position. Compound words, like [mestriskola] 'school master', also 
have [-ATR] segments in unstressed positions, further suggesting that 
some phonological rules occur only at specific states of morphological 
derivation. 

The details of the lexical organization of morphemes in strata are 
far from being clear and complete. Nevertheless, the preliminary 
evidence we have for both vowel harmony and [-ATR] distribution 
offers compelling reasons to pursue a more comprehensive analysis. 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 5 1 

[-ATR] 

a-brEv-i-ar 

a-brEv-i-ar 



(18) 


[-ATR] 


stratum 1 


brEv-i 


stress 


brEv-i 


assignment 




floating feature 




assoc. /vowel 




harmony 


brev-i 


default rules 


brev-i 


bracket 




erasure 


brevi 




'brief 


statum 2 


brevi-mEnti 


stress 


brevi-mEnti 


reassignment 




default rules 


brevi-menti 


bracket 




erasure 


brevimenti 




'briefly' 


6. Conclusion 





a-briv-i-ar 
a-briv-i-ar 

abriviar 

'to make brief 



This paper has presented a principled explanation of a 
phonological process which was previously believed to be 
unpredictable, namely vowel height harmony in BP. The presen- 
tation of the data and rule formulation showed that the harmony 
process is wholly dependent on morphology, both in terms of 
triggering elements and in terms of domain constraints. Transparent 
and opaque segments were accounted for within the theoretical 
approaches adopted in this analysis. Unlike other harmony 
processes, [ATR] specification plays no role in height harmony in BP. 
What does play a major role in BP phonology is morphology. The 
obscurity of the phonological process of height harmony lies in its 
interdependence on morphological affixation. By recognizing this 
interdependence, we were able to show the systematicity of a 
process which was previously not understood. 

NOTES 

* I sincerely thank Jose Ignacio Hualde, Jennifer Cole, and Rakesh 
M. Bhatt for their helpful comments on earlier drafts of this analysis. 
Of course, any errors of omission or commission are my own. 1 also 
thank my patient consultant, Marcelo. 



52 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

1 This analysis does not attempt to provide an argument for 
either approach, that is a theoretical debate beyond the scope of the 
present discussion. Rather this analysis attempts to bring forth 
evidence that certain phonological rules do not apply when certain 
suffixes are added onto a word, thereby suggesting that phonology 
and morphology interact during the derivation of a word. 

2 Most of the data have been compiled from the Dicionario da 
lingua Portuguesa of the Academia Nacional Brasileira (Nascentes 
1961) for the sake of standardization. Verb conjugations and unclear 
cases were checked with a consultant from Belo Horizonte. It should 
be noted that there may be slight variation across dialects (Perrone 
and Ledford-Miller 1985) and across speech styles (Major 1985). 

3 This representation is taken from Archangeli (1988:193). Her 
characterization however does not have both open [e], [o] and closed 
[e], [o] vowels. 

4 Vowels represented in upper case are vowels unspecified for 
height and, therefore, potential targets for the spread rule. 

5 That the initial [o] does not undergo vowel harmony is 
probably due to secondary stress relative to word length. As is 
shown later in this paper, stressed segments are opaque to vowel 
harmony. This is the only example counter to what is predicted. 
Subsequent analyses may want to further explore occurrences of 
initial vowel resilience to harmony in longer words. 

6 This account of transparent segments deviates from the 
proposal in Archangeli and Pulleyblank (1987) and Archangeli 
(1988) which says that the transparency effect is obtained from a 
segment's lack of specification of the spreading feature. However, we 
cannot accept that proposal here since [+hi] is underlyingly present 
and is also transparent in BP. 

"7 Alternate analyses for transparent and opaque segments in 
morphologically-conditioned harmony processes were also 
considered (cf. Cole 1987). 

8 That the [+lo] vowel remains unaffected by the spread rule 
differs from height harmony processes in other Romance languages 
and dialects. In Pasiego Montaiies, [+lo] vowels are transparent to 
[+hi] spread (McCarthy 1984). In Lena Bable (of Asturias), [+lo] 
vowels are affected by [+hi] spread, becoming [-hi, -lo] (Hualde 
1989), which is also the case for the Galician-Leonese dialect of 
Ancares (Fernandez Gonzalez 1981, cited in Hualde 1989). In 
Tudanca Montafies, however, the [+lo] vowel remains unaffected by 
[+hi] spread. Because these harmony processes differ in other ways, 
such as in the conditions on the triggering and target elements 



Hancin: Phonology-morphology interaction in Brazilian Portuguese 5 3 

(necessarily being stressed or unstressed), in the domain of feature 
spread (others exhibiting feature spread across several segments, 
whereas BP [+hi] spreads just once to the left), and in the spreading 
rule's interaction with the feature [ATR], we conclude that there is 
variation across related languages with respect to the nature of their 
[-i-hi] spreading rules; and the fact that [+lo] in BP is not affected by 
[+hi] spread is merely a case in point. 

9 According to what we have already presented, the two root 
vowels in this word should be [+hi]. However, the -ismo suffix does 
not appear to be a trigger in the harmony process. This will be 
discussed further in section 5. 

REFERENCES 

Archangeli, Diana. 1984. Underspecification in Yawelmani 

phonology and morphology. MIT Ph.D. dissertation in 

Linguistics. 
1988. Aspects of underspecification theory. Phonology 5:2.183- 

207. 
, & Douglas Pulleyblank. 1986. The content and structure of 

phonological representations. University of Arizona, Tucson and 

University of Southern California, Los Angeles, MS. 
Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. 

New York: Harper and Row Publishers. 
Clements, George. 1985. The geometry of phonological features. 

Phonology Yearbook 2.223-252. 
Cole, Jennifer S. 1987. Planar phonology and morphology. MIT Ph.D. 

dissertation in Linguistics. 
CUNHA, Celso. 1976. Grammatica do Portugues contemporaneo. Belo 

Horizonte: Benardo Alvares. 
FERNANDEZ GONZALEZ, Jose R. 1981. El habla de Ancares (Leon). 

Universidad de Oviedo, Oviedo. 
Goldsmith, John. 1976. Autosegmental phonology. MIT Ph.D. 

dissertation in Linguistics; published by Garland, New York, 

1979. 
. 1990. Autosegmental and metrical phonology. Cambridge, MA: 

Basil Blackwell, Inc. 
Harris, James. 1974. Evidence from Portuguese for the 'elsewhere 

condition' in phonology. Linguistic Inquiry 5:1.61-80. 
HUALDE, Jose Ignacio. 1989. Autosegmental and metrical spreading 

in the vowel harmony systems of north-western Spain. 

Linguistics 27.773-805. 
KIPARSKY, Paul. 1982. From cyclic phonology to lexical phonology. 

The structure of phonological representations, vol. 1, ed. by H. 

van der Hulst and N. Smith. Dordrecht: Foris. 



54 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

1985. Some consequences of lexical phonology. Phonology 

Yearbook 2.85-138. 
Lemle, Miriam. 1974. Analogia na morfologia: um estudo de um 

caso. Revista Brasileira de Linguistica 1.16-21. 
LiPSKI, John. 1973. Binarity and Portuguese vowel raising. 

Zeitschrift fur Dialektologie und Linguistik 40.16-28. 
Major, Roy C. 1985. Stress and rhythm in Brazilian Portuguese. 

Language 61:2.261-281. 
McCarthy, John. 1986. OCP effects: gemination and antigemination. 

Linguitsic Inquiry 15.575-602. 

1984. Theoretical consequences of Montaiies vowel harmony. 

Linguistic Inquiry 15:2.291-318. 
MOHANAN, K. P. 1982. Lexical phonology. MIT Ph.D. dissertation in 

Linguistics. Distributed by Indiana University Linguistics Club. 
Nascentes, a. 1961. (Academia Brasileira de Letras) Dicionario da 

lingua Portuguesa, vols. 1-4. Brasil: Departamento de Imprensa 

Nacional. 
PERRONE, C, & L. Ledford-Miller. 1985. Variation in pretonic lei in 

Brazilian Portuguese: preliminary studies with popular music of 

the Northeast and Rio de Janeiro. Hispania 68:1.154-159. 
PiZZINI, Quentin. 1982. Portuguese mid-vowels. Papers in Romance 

4:3.197-209. 
PULLEYBLANK, Douglas. 1986. Tone in lexical phonology. Dordrecht: 

Reidel. 

1988. Underspecification, the feature hierarchy and Tiv vowels. 

Phonology 5:2.299-326. 
REDENBARGER, W. 1981. Articulator features and Portuguese vowel 

height. Harvard Ph.D. dissertation in Department of Romance 

Languages and Literatures. 
SAGEY, Elizabeth. 1986. The representation of features and relations 

in non-linear phonology. MIT Ph.D. dissertation in Linguistics. 
Vago, Robert. 1988. Underspecification in the height harmony sys- 
tem of Pasiego. Phonology 5:2.343-362. 
Viegas, M. & R. Assis Veado. 1982. Al9amento de vogais pretonicas. 

Cadernos de Linguistica e teoria da literatura 7.53-70. 
Wetzels, Leo. 1988. Contrastive and allophonic properties of Bra- 
zilian Portuguese vowels, paper presented at the Linguistic 

Symposium on Romance Languages 18, University of Illinois at 

Urbana-Champaign. 
Yip, Moira. 1988. The obligatory contour principle and phonological 

rules: A loss of identity. Lingustic Inquiry 19.65-100. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



MORAIC PHONOLOGY AND /I/IRREGULAR PREDICATES IN 
KOREAN* 

Seok Keun Kang 

This paper reexamines the so-called /l/-irregular 
predicates in Korean in terms of morale phonology. It has 
been asserted that /nal-ini/ 'because (it) flies' has two 
alternations, i.e. [na:ni] and [nallini]. In order to account for 
this, Kim-Renaud (1973) and Ahn (1985) postulated a 
double /ll/ in the underlying representation of the stem, i.e. 
/nail-/. On the other hand, Ahn (1988) asserted that the 
two types of surface forms are idiolectal variants derived 
from two different underlying representation of the same 
word. 

In this paper, I claim that their assertions cannot be 
acceptable for several reasons. I show that [na:ni] and 
[nallini] are not alternants of one and the same word, but 
are two different words. The former, which means 'because 
(it) flies', is derived from /nal-ini/, while the latter, which 
means 'because (we) carry (it)', is derived from /nalli-ini/. 

1. Introduction 

The purpose of this paper is to reexamine the so-called /l/- 
irregular predicates in Korean. The behavior of the liquid is one of 
the most complicated aspects in Korean phonology. A verbstem-final 
/I/ behaves peculiarly before /i/-initial suffix as shown in (1). 

(l)(a) Consonant initial suffix: /u:l - ta/ -^ [u:lda]i 'cries' 

/u:l - ko/ -^ [u:lgo] 'cry and' 

(b) /a /-initial suffix: /u:l - ato/-> [u:r8do] 

'though crying' 

(c) /i/-initial suffix: /u:l - ini/ -^ [u:rini] 'because 

/ [u:ni] of crying' 

In (lb) /!/ weakens to [r] intervocally, but it is optionally deleted 
when followed by /i/-initial suffix as in (Ic). This /l/-deletion is 
viewed as an instance of extreme weakening in the intervocalic 
position (Kim-Renaud 1974). 

For /nal-ini/ 'because (it) flies', Kim-Renaud (1973) and Ahn 
(1985) asserted that it has two alternants, viz. [na:ni] and [nallini]. In 
order to account for this, they postulated a double /ll/ in the 



56 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

underlying representation of the stem, i.e. /nail-/. On the other 
hand, Ahn (1988) asserted that the two types of surface forms are 
derived from two different underlying representations of the same 
word. Proposing that both /nal/ and /nail/ are correct underlying 
representations depending upon the speaker, he said that Speaker A 
with /nal/ would produce [nani], whereas Speaker B with /nail/ 
would produce [nallini]. 

This paper attempts to reconsider the problem in the framework 
of morale phonology. I show that Kim-Renaud's and Ahn's proposals 
cannot be accepted for several reasons. Following Hyman (1985), 
Hock (1986), Hayes (1989) and Zee (1989), I also show that 
compared with the two theories mentioned above, the morale theory 
gives a more natural account. In section 2, Kim-Renaud (1973), Ahn 
(1985), and Ahn (1988) will be reviewed in detail. In section 3, I 
will discuss the problem in the framework of morale phonology. 

2. Previous analyses 

Asserting that /nal-ini/ has two alternants, i.e. [na:ni] and 
[nallini], Kim-Renaud (1973) postulated a double /ll/ in the under- 
lying representation of the stem as in (2). 

(2) /nail - ini/ -^ [na:ni] / [nallini] 'because (it) flies' 

In order to obtain a correct derivation, she proposed the following 
three rules. 

(3) (a) V 1 1 + V 

12 3 4 5 

10 3 4 5 
[+long] 

(b) i -^ / [ Vl]vb-stem [ lAf 

(c) 1^0 / [ V _ ]vb-stem [{n, s, p, 0}]Af 

[na:ni], for instance, is derived as shown below. 

(4) /nail - ini/ 

na:l ini (3a) 

na:l ni (3b) 

na: ni (3c) 

But her analysis cannot be accepted. To begin with, the vowel 
lengthening rule (3a), as Ahn (1985) noticed, does not work in some 
cases. In (5), for instance, the deletion of /I/ is not followed by the 
lengthening of the preceding vowel /a/. 

(5) /nail/ + I it I 'wing' [narae], *[na:rae] 

'device' 

And more crucially, she overlooked the fact that [nallini] and [na:ni] 
are not alternants of the same word. Rather, [nallini] 'because (we) 



S. K. Kang: Morale phonology and /l/-lrregular predicates In Korean 5 7 

carry (it)' should be represented as /nalU-ini/ underlyingly, while 

[na:ni] 'because (it) flies' should be represented as /naMni/, which 
will be discussed in section 3. 

Following Kim-Renaud (1973), Ahn (1985) also claimed that the 

alternants in (2) can be derived by positing a geminate /II/ in the 

underlying representation of the stem. In the framework of lexical 

and CV phonology, Ahn (1985) derived the alternants as shown 
below. 



(6) (a) 


$ 




$ $ 






/ 


1 


\ 


1 / 


\ 






1 


R 

1 


C 

1 \ 


RO 

1 1 


R 

1 


Stratum 4 inflection 


c 

1 


V 


cc 


vc 


V 

1 




1 

n 


1 
a 


I 1 

1 1 


1 1 

i n 


1 
i 




C 

1 


V 

1 


cc 

1 1 


C 

1 


V 

1 


Intersonorant /i/-Deletion 


n 


a 


1 1 


n 


i 




C 

1 


V 

1 


c 

1 


C 

1 


V 

1 


Coda Cluster Simplification 


n 


a 


1 


n 


i 


(simplification of CCC sequence) 


C 

1 


V 

1 


c 

/ 


C 

1 


V 

1 


/l/-Deletion & association of /a/ 


n 


a 




n 


i 


to the empty C 


[na:n 


i] 








(b) CV 


cc 


V C V 




1 


1 


1 1 


1 1 


1 


Stratum 4 inflection 


n 


a 


1 1 


i n 


i 


Intersonorant /4/-Deletion 
(optional) 












C V 


C C V c 


V 




1 


1 


1 1 


1 1 


1 


Resyllabification 


n 


a 


1 1 


i n 


i 


Coda Cluster Simolification 



[nallini] 

In this way, he claims, the two alternants, [na:ni] and [nallini], can be 
explained correctly and the compensatory lengthening (henceforth, 
CL) in (6) can be given a natural account. As mentioned earlier, 
however, [na:ni] and [nallini] should be derived from two different 
words, a fact which will be discussed more fully farther below. 
Besides, in (6) CL allows the preceding vowel /a/ to lengthen, which 
gives an awkward representation of a long vowel as a vowel melody 
linked to VC rather than to VV (Hayes 1989). Finally, the Coda 
Cluster Simplification deletes C of the CCC sequence. Since this rule 



58 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

applies only to coda consonant clusters and since Ahn posits the 
double /ll/ in the underlying representation of the stem, i.e. /nail-/, 
he cannot account for the following in which the underlying geminate 
/ll/ still undergoes degemination in spite of the fact that it does not 
meet the structural description of Coda Cluster Simplification. 

(7) /nall-ae/ 'wing' [naras], *[nalla2] 

'device' 

Under Ahn's analysis, the wrong output *[nallae], not the correct 
output [narae], would be derived. 

A few years later, Ahn (1988) argued that [nani]2 and [nall4ni] 
'because (it) flies' are idiolectal variants derived from two different 
underlying forms of the same word; i.e., the former is derived from 
/nal-ini/ and the latter from /nall-ini/. In order to derive the 
surface forms, he proposed the following two rules. 

(8) (a) Intersonorant /i/-Deletion (optional) 

i -^ / 1 ]v/A [+son, +cons] (domain: inflection) 

(b) /l/-Deletion 

1-^0 / V ]v/A {n, s} (domain: inflection) 

(9) shows how the rules in (8) derive both [nani] and [nalUni]. 

(9) (a) Speaker A (b) Speaker B 

CVC VCV CVCC VCV 

I I I I I I I I I I I I I Inflection 

[[ n a 1 ] i n i] [[ n a 1 1 ] i n i] 'Since (it) flies' 

CVC cv 

I I I I I Intersonorant 

n a 1 n i /i/-Deletion 

CV CV 

II II /l/-Deletion 

n a n i 

/\ /\ /l\ /\ /\ 

CV CV cvccvcv 

II 11 I I 11 I I I Resyllabification 

n a n i n a 1 1 i n i 

[nani] [nallini] 

However, Ahn's account is not acceptable. First, it is unnatural to say 
that alternants of one and the same word are derived from different 
underlying representations. In fact, as will be discussed later, [na:ni] 
and [nallini] mean 'because (it) flies' and 'because (we) carry (it)', 
respectively. Second, Ahn's formulation of Intersonorant HI - 
Deletion wrongly predicts that /nall-ini/ in (9b) may optionally 
undergo the change, which would result in the same output as that in 
(9a). 



S. K. Kang: Morale phonology and /V-irregular predicates in Korean 5 9 

In the following section, I will reexamine the problem in 
question in the framework of moraic theory (Hyman 1985, Hock 
1986, Hayes 1989, Zee 1989). 

3. Alternative analysis 

In this section, moraic phonology is briefly reviewed before the 
problem in question is discussed. 

A. Moraic phonology 

Hyman (1985), Hock (1986), Hayes (1989) and Zee (1989) have 
proposed the mora as a unit in the prosodic tier. It has been 
asserted that the moraic structures of languages vary. A heavy 
syllable is assigned two moras, whereas a light syllable is assigned 
one. In languages with constrastive vowel length, long vowels have 
two moras, and short vowels one, as shown below (Hayes 1989). 

(10) (a) ^ [i (b)^i 

\ / I 

i = /i:/ i = /i/ 

Unlike vowels, short consonants do not bear any mora underlyingly 
as in (11a), while geminates bear one mora as in (lib). 

(11) (a) n = /n/ (b) [i 

I 

n = /nn/ 

In order to account for languages whose closed syllables are heavy, 
Hayes (1989) proposed the rule of 'Weight by Position', which assigns 
certain coda consonants a mora in the process of syllabification, as 
shown in (12). 

(12) Weight by Position 

$ $ 

I I \ 

|j. — > n |i. where $ dominates only |i. 

I I 1 

a p a p 

In addition, no mora is assigned to an underlying glide at all. 

(13) i = /y/ 

B. /l/-irregular verbs in moraic phonology 

In this section, I will show that the proposals reviewed in 

section 2 cannot be accepted for reasons beyond those mentioned 

earlier. Some other examples relevant to the problem in question 
are given in (14). 



60 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(14) (a) [nal-ta] 'to fly' 

[narasa], [narini] / [na:ni], *[nallas3], *[nalUni] 
(a') [nari-ta] 'to carry' 

[nallasa], [nalUni] / [narini], *[na:ni], *[narasa] 

(b) [kil-ta] 'to be long' 

[kiresa], [kirini] / [ki:ni], *[killasa], *[kilUni] 
(b') [kiri-ta] 'to breed' 

[killasa], [killini] / [kirini], *[ki:ni], *[kirasa] 

(c) [mal-ta] 'to roll up' 

[marasa], [marini] / [ma:ni], *[mallasa], *[mallini] 
(c') [mari-ta] 'to dry' 

[mallasa], [mallini] / [marini], *[ma:ni], *[marasa] 

(d) [kkil-ta] 'to pull' 

[kkirasa], [kkirini] / [kki:ni], *[kkillasa], *[kkillini] 
(d') [kkiri-ta] 'to unpack' 

[kkirasa], [kkiUini] / [kkirini], *[kkillasa], *[kkillini] 

The verb stems in (14a, b, c, d) end with /I/, i.e. /nal-/, /kil-/, 
/mal-/, and /kkil-/, while those in (14a', b', c', d') end in vowel /i/, 
i.e. /nali-/, /kili-/, /mali-/, and /kkili-/. For convenience of exposi- 
tion, I will call the first class of verbs /nal-/ type and the second 
/nalli-/ type. 

Verbs of the /nal-/ type work differently from those of the 
/nalli-/ type. First, the former allow only one /I/ in their alter- 
nations as in [kirasa] and no geminate /ll/ as in *[killasa] in (14b), 
whereas the latter have geminate /ll/ when they are followed by 
suffixes beginning with a vowel as in [killasa] in (14b'). Second, 
verbs of the /nal-/ type optionally undergo CL (for example, /kkil- 
ini/ -^ /kkil-ni/ (Intersonorant /i/-Deletion) -^ [kki:ni] (/l/-Deletion 
and CL)), but those of the /nalli-/ type do not (i.e., *[kki:ni]). The 
verb stems underlying (14a) and (14a') thus are phonologically en- 
tirely distinct. Moreover, as the glosses show, they also have entirely 
different meanings. We can therefore conclude that [nallini] and 
[na:ni] are not alternants of the same word but are different phonetic 
forms of different words. [nallini] has the meaning 'because (we) 
carry (it)', whereas [na:ni] means 'because (it) flies'. Therefore, 
neither the assertion that /nal-ini/ 'because (it) flies' has the two 
alternants, [na:ni] and [nallini] nor the assertion that [na:ni] and 
[nallini] are idiolectal variants derived from two different underlying 
forms of the same word and meaning is correct. 

Now let us turn to an account in terms of moraic theory. To 
begin with, consider the alternations in (14a, b, c, d). In order to 
account for them, I will posit only one /I/ in their stem underlyingly. 
The stem of the verb in (14c), for instance, is underlyingly repre- 
sented as follows. 



S. K. Kang: Morale phonology and /IZ-lrregular predicates In Korean 

(15) /mal-/ 

$ 

/l\ 

/ ^i U 

/ I I 

m a 1 - 

(16) shows how the two alternants in (14c) are derived. 

(16) (a) [marini] 'because (we) roll (it) up' 

/ I \ I /I 



/ ^ ^ 


^ / ^ 


— > 


(Intersorant /i/-Delelion 


/ 1 1 


1 / 1 




(optiona 


D) 




m a 1 


-in i 










surface: 


[marini] 










(b) [ma: 


ni] 




'because (we) roll 


(it) up' 


$ 


$ $ 




$ $ 






/l\ 


1 /I 




/l\ /I 






/ |i |i 


H / \i 


— > 


/ H^i / ^i 




(Intersonorant 


/ 1 1 


1 / 1 




/ 1 1 / 1 




/-i/-Deletion) 


m a 1 


-in i 




m a 1 n i 






$ 


$ 






$ 


$ 


/I \ 


/I 






/I \ 


/I 


/ [i- [i- 


/ ^ ' 


[/1/-D 


eletion) — > / 


^i \^ 


/ \i (CL) 


/ \ 


/ 1 




/ 


1/ 


/ 1 


m a 


n i 




m 


a 


n i 



surface: [ma:ni] 

Intersonorant /i/-Deletion, /l/-Deletion and CL in (16) can be 
formulated as in (17a, b, c), respectively. 

(17) (a) Intersonorant /i/-Deletion (optional) 



1 ]v/A - 
(b) /l/-Deletion 
$ 
H I 

I I 

0]v/A {s, n) 



I 
I 
[-i-son, -i-cons] 



62 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(c) Compensatory Lengthening (CL) 

\ / 

\ / 

\ / 

V 

Rule (17a) says that the suffix-initial /i/ is deleted between stem- 
final /I/ and [-t-son] consonants. And when a stem-final /I/ is 
followed by an affix-initial /s/ or /n/, it is deleted obligatorily by 
(17b) (e.g., /til-se/ -^ [tise] 'let's eat', /hsl-ni/ -^ [hani] "(Do you) 
destroy (it)?'). Unlike CV theory, moraic theory derives the 
alternations in (14c) without producing an awkward representation 
in which a vowel is linked to a C slot, as shown above. 

For the verbs in (14a', b', c', d'), a double AV is posited in their 
underlying representations. Compared with /mal-/ in (15), for 
example, the underlying representation of /malli-/ is given in (18). 

(18) /malli-/3 "to dry up' 

$ $ 

/l\ /I 

/ \i[i / [i 

/ I \/ I 

m a 1 i - 

The two alternants in (14c'), [marini] and [malUni] 'because (it) dries', 
for instance, are derived as shown in (19a) and (19b), respectively. 



(19) (a) [marini] 










$ $ 


S 


$ 


$ $ $ 




/l\ /I 


1 


/I 


/l\ /I /I 




/ \i\i / [I 


^i 


/ ^l ^ 


/ H H / H / n 


(/i/-Deletion) 


/ 1 \/ 1 


1 


/ 1 


/ 1 \/ 1 / 1 




mal i - 
—i 


* 


n i 

(Intersonor 


m a 1 i-n i 
ant /4/-Deletion) 





$ $ s 

/I /I /I 

-> / |I / [l / \l (/l/-Degeminaiion (optional)) 

/I / I - / I 

m a 1 i n i 
[marini] 



S. K. Kang: Morale phonology and /V-irregular predicates in Korean 6 3 



(b) [mallini] 

$ $ 


S $ $ $ 


S 


/l\ /I 


1 /I /l\ /I 


/I 


/ \i \i / \i 


\i / \i -^ / \x\x / \i 


/ \l (/4/-Deletion) 


/ 1 \/ 1 


1 / 1 / 1 \/ 1 


/ 1 


m a 1 i - 


i n i m a 1 i - 


n i 



— > (Intersonorant /i/-Deletion) 

— > (/l/-Degemination (optional)) 

[mallini] 

In (19), the vowel /i/ deletes when it is adjacent to another vowel. 
But there is no CL. This /i/-deletion phenomenon can be expressed 
in the following way. 

(20) /i/-Deletion (obligatory) 

I ^ % I 

i [-cons] ]-N - 

This rule says that /i/ obligatorily deletes with its mora either when 
followed by a suffix vowel or when preceded by a stem final vowel. 
'-N' here means that this rule does not apply to noun stems (cf. /ki- 
eykey/ -^ [kkieykey] *[key:key] 'to him", where no deletion applies). 
Some more examples are given below. 

(21) /ka-ini/ -^ [kani] *[ka:ni] 'because (I) go' 
/ssi-9/ -> [ssa] *[ss9:] 'to write' 

/ssi-ini/ -> [ssini] *[ssi:ni] 'because (I) write (it)' 

In (19), the Intersonorant /i/-Deletion (17a) cannot apply, because 
its structural description is not met, which is predicted by the 
following constraint proposed by Hayes (1986). 

(22) Linking Constraint: 

Association lines in structural descriptions are 
interpreted as exhaustive. 

The linking constraint above says that if structures have more 
association lines than the rule requires, then these structures will not 
meet the structural description of the rule, for association lines are 
interpreted as exhaustive. Since the geminate /I/ in (19) is repre- 
sented as linked to two different prosodic tiers (i.e., 'n' and 'S') rather 
than one (i.e., '[i') required by the Intersonorant /i/-Deletion, the 
structural description of the Intersonorant /i/-Deletion is not met 
here. And the geminate /ll/ in (19a) optionally undergoes the /l/- 
Degemination rule. This rule can be formulated as follows. 



64 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(23) /l/-Degemination (optional) 

$ 
/ 

\ / 
\ / 

1 

In (23), the geminate liquid /I/ loses its mora by degemination. 

Finally, let us turn our attention to the data below, which appear 
initially to be baffling. 

(24) (a) [ili-ta] 'to arrive' 

[irini], [irasa], *[i:ni], *[illini], *[ill9sa] 

(b) [ttali-ta] 'to follow' 

[ttar4ni], [ttarasa], *[tta:ni], *[ttallini], *[ttallas9] 

(c) [chili-ta] 'to pay' 

[chirini], [chirasa], *[chi:ni], *[chillini], *[chillasa] 

(d) [phuli-ta] 'to be green' 

[phurini], [phurasa], *[phu:ni], *[phullini], *[phullasa] 

The verbs in (24) seem to be members of the /nalli-/ type above. 
This, however, is not correct, for they cannot have alternations such 
as *[illini], *[ttallini], *[chillini], *[phullini], which ought to be possible 
in verbs of the /nalli-/ type. Since they do not have alternations such 
as *[i:ni], *[tta:ni], *[chi:ni], *[phu:ni], they do not belong to the /nal-/ 
type, either. In order to account for the alternations in (24), I will 
assume that they have an underlying structure which is different 
both from that of the /nalli-/ type and from that of the /nal-/ type 
in that unlike the /nalli-/ type their stems have no double /ll/ in 
underlying representations and unlike the /nal-/ type the stem ends 
in /i/. /ili-/, for example, is underlyingly represented as the follows. 

(25) /ili- / 'to arrive' 

$ $ 

I /I 

^ / Pi 
I / I 
i 1 i - 

Example (26) shows how [irini] is derived. 

(26) /ili-ini/ 'because (we) arrive at' 

I /I I /I 1 /I /I 

li/^i [i / ^ ^ [i / \i / \i (/i/-Deletion) 

1/1 I / I I / I / I 

il i- ini il i-n i 

-> (Intersonorant /i/-Deletion) 

-> (/l/-Degemination) 

[irini] 



S. K. Kang: Moraic phonology and /V-irregular predicates in Korean 6 5 

Since the conditions further applications are not met, neither Inter- 
sonorant /i/-Deletion nor Degemination applies above. By positing 
underlying representations as in (25), we can predict all alternations 
in (24). 

4. CONCLUSION 

I have discussed /l/-irregular verbs in Korean in terms of 
moraic theory. I have shown that compared with both the linear 
theory and the CV theory, the moraic theory accounts for the 
phenomena in a more natural way. 

In Section 2, previous analyses of /l/-irregular verbs, viz. Kim- 
Renaud (1973), Ahn (1985), and Ahn (1988), have been reviewed. I 
have shown that the underlying forms posited by them cannot be 
accepted for several reasons. First, [na:ni] and [nallini] are not 
derived from the same underlying form but are different words; i.e., 
[na:ni] 'because (it) flies' is derived from /naMni/, while [nallini] 
'because (we) carry (it)' is derived from /nalli-ini/. Finally, [irini] 
'because (we) arrive' is derived from the underlying representation 
/iri-ini/, which has a stem-final vowel /i/ but no geminate liquid. 

NOTES 

*I would like to thank Chin W. Kim, Michael Kenstowicz, Jose I. 
Hualde, Jennifer Cole, and an anonymous reviewer for helpful 
comments on an earlier version of the paper. 

1 The non-tense noncontinuant obstruents become voiced 
between voiced segments (e.g., /kut-ini/ -> [kudini] 'as it hardens'). 

2 In contradistinction to Ahn 1985, Ahn 1988 assumes that /!/- 
deletion does not feed CL and accordingly derives [nani], not [na:ni]. 

3(18) is the representation after syllabification applies. In the 
process of syllabification an underlying geminate (one mora) con- 
sonant is syllabified both onto the onset position of the following 
syllable by the onset formation rule and onto the coda position of the 
preceding syllable by the coda formation rule (Hayes 1989). 

REFERENCES 

Ahn, Sang-cheol. 1985. The interplay of phonology and morphology 
in Korean. University of Illinois at Urbana-Champaign disserta- 
tion in Linguistics. 

1988. On /l/-deletion: A lexical approach. Korean Linguistics 
5.101-119. 



6 6 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

FUERSSEL, M. 1986. Glides in Berber and syllabicity. Linguistic In- 
quiry 17.1-12. 

Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62.321- 
351. 

1989. Compensatory lengthening in morale phonology. Linguis- 
tic Inquiry 20:2.253-307. 

Hock, Hans. 1986. Compensatory lengthening: In defense of the 
concept 'Mora'. Folia Linguistica 20.431-460. 

HYMAN, L. 1985. A theory of phonological weight. Dordrecht: Foris 
Publications. 

Kang, Seok-keun. 1989. Morale phonology and compensatory length- 
ening in Korean. Department of Linguistics, University of Illinois 
at Urbana-Champaign, MS. 

Kim, Chin-Woo. 1990. The character of Korean glides. Department of 
Linguistics, University of Illinois at Urbana-Champaign, MS. 

Kim, Jong-mi. 1985. A non-lexical analysis of Korean morphology 
and phonology. University of Southern California dissertation in 
Linguistics. 

KiM-RENAUD, Young-key. 1973. Irregular verbs in Korean. Language 
Research 9.206-225. 

1978. The syllable in Korean phonology. Papers in Korean 
studies, ed. by C.-W. Kim, 85-98. Columbia, SC: Hornbeam Press. 

SOHN, Hyang-sook. 1987. Underspecification in Korean phonology. 
University of Illinois at Urbana-Champaign dissertation in 
Linguistics. 

Zec, Draga. 1989. Sonority constraints on prosodic structure. Stand- 
ford University dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



CORONAL: TRANSPARENT OR OPAQUE? 

Yongsoon Kang 

It has been claimed that coronals in Korean are opaque in that 
they block application of the umlaut rule, contrary to the pre- 
diction of underspecification theory that coronals are trans- 
parent because of their lack of a class or feature node under the 
place node in feature geometry. Reanalyzing the data within the 
framework of Dependency Phonology, I claim that this opacity 
can be ascribed to the Linking Constraint and the Adjacency 
Condition and that this account is not possible within the frame- 
work of underspecification theory and feature geometry. 

1. Introduction 

It has been assumed in underspecification theory (UT) that 
coronal consonants are the most underspecified universally. In fea- 
ture geometry (FG), this is represented as absence of the coronal 
node (Avery & Rice 1989) or place node (Paradis & Prunet 1989). In 
Korean, coronals are also regarded as the most unmarked and 
underspecified segments (Sohn 1987, Kim 1987) and represented as 
totally unspecified phonologically both in UT and in FG, with no 
nodes under the place node. Absence of the coronal or place node for 
coronals has been justified by such phonological processes as neu- 
tralization (Sohn 1987, Kim 1987), epenthesis (Iverson 1989), and 
transparency of coronals (Paradis & Prunet 1989). 

In this paper I argue against the idea that the representation of 
coronals in the underlying structure lacks the coronal or place node, 
on the basis of the umlaut phenomena in Korean in which coronal 
consonants are opaque with regard to feature spreading. In section 

2, the problem of the representation will be discussed, in section 3 I 
will present the representation of Korean segments in terms of 
Dependency Phonology (DP) (Anderson & Ewen 1987), and in the 
final section I will suggest an analysis within the framework of DP. 
Since it is assumed that phonological processes are best explained 
through focus on the representational component rather than on the 
rule component (Archangeli 1984, Avery & Rice 1989), it would be 
natural to evaluate the representational system with respect to the 
phonological process it predicts. 



68 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

2. The representation of coronals 

The representation of coronals in Korean has been suggested as 
follows in the literature of underspecification and FG. 

(l)a. HI (Sohn 1987) b. /t/ (Kim 1987) 

X X 

I I R:Root 

[ ] R L:Laryngeal 

/ \ SL:Supra- 
L SL laryngeal 

/ \ M:Manner 

M P P:Place 

Sohn follows the FG account of Clements (1985) even though she 
makes use of this only for neutralization in Korean, while Kim adopts 
Sagey's account (1986). All the absent features are filled in by 
universal default and complementary rules of the language. The re- 
presentation of coronals is justified by phonological rules like neu- 
tralization, (which is characterized as the delinking of primary or 
secondary place features), and consonantal assimilation, (in which 
coronals assimilate to the following consonant but not vice versa). 
Since the representation of Sohn is not specific for hierarchical struc- 
ture, I will follow Kim's in this paper. 

In other works, similar representations of coronals has been 
suggested; cf. (2). 

(2) a. Avery & Rice 1989 b. Paradis & Prunet 1989 

HI HI 

Root X 

/ I I 

o I Laryngeal RN RN:Root Node 

1 I 

o Supralaryngeal [+cons] 

I 

o Place 

As we see above, the internal structure of coronals is characterized 
by the absence of nodes under the place or root node. Avery and 
Rice (1989:183) claim that if the secondary content node is the only 
distinguishing feature between two segments, then the primary 
feature is activated for the segments distinguished. This is called the 
Node Activation Convention (NAC). Thus the coronal node is activat- 
ed only if there is a secondary content node which differentiates the 
coronal segments. NAC does not apply to Korean because there is no 
secondary content node under the coronal node. The representation 
of coronals by Paradis and Prunet is supported by the transparency 
of coronals in Fula, Guere, and Mau. In those languages, only 



Kang: Coronal: Transparent or opaque? 



69 



coronals are found to be transparent for feature spreading. This can 
be explained by the above representation: Since they do not have 
any nodes or features under the place node, they cannot block the 
spreading of a feature. 

The same behavior is predicted for Korean coronals because 
their representation is the same. Therefore, we should expect that 
coronals are transparent to the spreading of features. However, the 
opposite is true for Korean. 

In Korean, the following regressive assimilation of vowels before 
the high vowel /i/ is found. We will call this rule, which is optional, 
umlaut. 



(3) a. 



/a/ 
/caphita/ 
/makhita/ 
/kamkita/ 
/k'ak'ita/ 
/namkita/ 
/ankita/ 
/api/ 
/nampi/ 
/kalatji/ 
/kacami/ 
/komp'^arji/ 

/9/ 
/maki/ 
/makhita/ 
/paskita/ 
/ami/ 



1^1 
[caep'^ida] 
[maek'^ida] 
[kasmgida] 
[k'aek'ida] 
[naemgida] 
[aerigida] 
[aebi] 
[naembi] 
[karaerii] 
[kajaemi] 
[komp^'aerii] 

[megi] 
[mek^da] 
[pek'ida] 
[emi] 



'to be caught' 

'to be obstructed' 

'to be shut' 

'to be cut' 

'to leave (sth.)' 

'to be embraced' 

'father' 

'pan' 

'crotch' 

'flatfish' 

'fungus' 

'food' 

'to be eaten' 

'to undress' 

'mother' 



This phonological process can be characterized as the fronting of the 
back vowels before the following front vowel /i/. However, if a 



coronal consonant intervenes, 
in the following. 

(4) 

/palphita/ -> 

/allita/ -^ 

/palkhita/ -> 

/mallita/ -^ 

/mati/ -^ 

/kat^i/ -> 

/pallita/ -> 

/apsi/ ^ 

/malli/ -^ 



umlaut does not take place, as is seen 



*[paslp^da] 

*[aellida] 

♦[paelk^da] 

*[maellida] 

*[maeji] 

♦[kasc^i] 

*[pellida] 

*[eps'i] 

*[melli] 



'to be trodden' 

'to let (them) know' 

'to uncover' 

'stop (s.b.)' 

'the first child' 

'together' 

'to stretch' 

'without' 

'far away' 



7 Studies in the Linguistic Sciences 21:1 (Spring 199 1 ) 

According to UT, umlaut can be represented as the spreading of 
[-back] to the preceding vowel because the high front vowel /i/ has 
only the feature [-back] in the underlying representation; cf. (5). 

(5) V /i/ 

X X 

I I 

R R 

/ \ / \ 

L SL L SL 
/ \ / \ 

MP MP 
I I 

dor dor dor:dorsal 

\ I 

\l 
[-back] 

The underlying representation of coronals in UT and FG cannot block 
the spreading of the feature [-back] to the preceding vowels, since as 
we have seen above, no class nodes exist underlyingly under the 
place node. On the other hand, labials and dorsals are represented as 
having class nodes such as [labial] and [dorsal] under the place node 
according to Sagey's framework. Since the spreading of the feature 
[-back] can be assumed to be blocked only by the feature node under 
the dorsal node, labials and dorsals don't prevent the umlaut from 
applying here. Then a problem remains how to explain the opaque- 
ness of coronals in umlaut in Korean. 

Before suggesting my analysis, it would be in order to introduce 
the representational system of DP, which will be used in my analysis. 

3. The representation of Korean segments in DP 

In the previous section, we have seen a representational 
problem of UT and FG in accounting for umlaut in Korean. In this 
section, I will introduce the representation of Korean segments in 
terms of Dependency Phonology (Anderson & Ewen 1987). 

One of the characteristics of DP lies in the idea that a depen- 
dency relation between head (governor) and modifier (dependent) is 
found in every module of grammar. Thus the internal structure of a 
segment is represented as a dependency relation between compon- 
ents which are atoms for representing internal structure of a seg- 
ment. A segment consists of two major gestures, articulatory and 
categorial, and they are connected by an association line. 

Categorial gestures are made up of two components, IVI (Vocalic) 
and ICI (Consonantal), and all the categories needed for Korean are 



Kang: Coronal: Transparent or opaque? 71 

represented by the dependency relation between them, as given 
below. 1 

(6) V V V V:C C 

I I 

vc c 

vowel liquid nasal fricative stop 

For the representation of Korean obstruents, we need one more 
categorial component 101 (Opening glottis). Three different series of 
Korean obstruents (i.e., lax, tense, and aspirated) are then expressed 
in the following way (see Anderson & Ewen 1987 for details). 

(7) lax tense aspirated 

ICklOl ICI 101 

I I 

lOI ICI 

In articulatory gestures, there are four components. They are lil 
(frontness), lal (sonority), lul (roundness), and 111 (linguality). These 
components combine to represent articulatory gestures of vowels 
and consonants. We should note here that vowels and consonants 
can share the same components in articulatory gestures, a fact which 
will turn out to be crucial in representing the consonant and vowel 
interaction phenomena. 

First, the articulatory gestures of Korean vowels can be repre- 
sented as follows. (Following Sohn (1987), I will adopt an eight 
vowel system.) 

(8) (a) 



(b) 



{la;il) 

(lal) 

One of the characteristics of the Korean vowel system is its 
asymmetry between front and back vowels. While two vowels are 
found between /i/ and /a/, only one vowel /o/ occurs between /u/ 
and /a/. 



/i/ HI 


/u, 


/c/ 




/a/ 


lol 


Ix/ 




/a/ 




{lil} {li,ul} 


{lul} 


{li;al} 




{li,a,ul} 


{lu,al} 



7 2 Studies in 4e Linguistic Sciences 21:1 (Spring 1991) 

The following is the inventory of consonants in Korean. 
(9) Consonants in Korean 

labial alveolar palatoalveolar velar 

k, k^ k' 



stops 


p. p\ p 


t, t^ f 


fricatives 




s, s- 


affricates 






nasals 


m 


n 


lateral 




1 



The articulators gestures of consonants (i.e.. places of articulation) 
are represented as follows. 

(10) labial alveolar palato-alveolar velar 

(lull {Oi} (iU) {ll,ul} 

Affricates are interpreted as a combination of stops and frica- 
tives and represented as a dependency relation of the two. in which 
fricatives depend on stops. 

(11) {O} 

\ 
{IC:V!] 

Finally, the velar fricative /h/ is represented as (0) lacking arti- 
culator}' gestures. In DP. each segment consists of two major ges- 
tures. For instance the word ktemi 'ant' is represented as follows. 

(12) /k/ /x/ Iml III 

categorial {G}:{0} {IVl} [W-O] {fVI} 

gesture : : : : 

articulatory {Ml} {la;il} {lul} {til} 

gesture 

There is no dependency relation between gestures. They are con- 
nected by association lines generated by the Universal Association 
Convention. 

4. Umlaut in Korean 

In this section, I will show my analysis of umlaut in Korean in 
terms of the representational system of DP. It will turn out that the 
blocking effect of coronals can be attributed to the Linking Constraint 
(LC> (Hayes 1986) and the Adjacency Condition which restrict the 



Kang: Coronal: Transparrait or opaque? 7 3 

application of spreading and the number of gestures the spreading 
can pass.- 

In DP, umlaut can be characterized as the spreading of the 
component lil. If it spreads to {la!) /a/, it becomes a dependent of lal 
resulting in {la:ii) Isl. If the target of the spreading is {li.a.ul) /a/, lil 
in the trigger strengthens the power of iii in the target. As a result, lil 
becomes a governor dominating either :ai or iul. Since there is no 
segment represented as {ii:u } in Korean, the result naturally would 
be {ii:ai} /e/. This rule only affects the preceding vowel and can be 
represented as follows in the DP framework. 

(13) Umlaut 

Categorial {IVI}{IVI) 

gesture : \ : 

: \ : 
Articulaiory {a} {lil} 

gesture 

However, the spreading of Iii is blocked if there are coronals 
between trigger and target. In the following. I will analyze the data 
case by case to show why. 

First of all. if a palatoalveolar consonant intervenes, umlaut does 
not take place. Following are some e.xamples. 

(14) /kaci/ -^ *[kaeji] 'eggplant' 
/mac''i/ -^ *[maec i] 'as if 
/kac'i/ ^ *[kaec'i] 'value' 
/macimak/ -^ *[maejimak] 'the last' 
/apaci/ -^ *[apeji] 'father' 
/kaci/ -^ *[keji] 'beggar' 

This case can be explained by the Obligaton.^ Contour Principle (OCP) 
and the LC. By the OCP, the word /kaci/ can be represented as 
follows. 

(15) 



/k/ 


/a/ 


Id HI 


{•O} 


(IVI) 


{O} {IVl) 
:\ / 
: \ / 
: \ / 


fll.ul) 


{!al} 


{01} {Bl} 



LC prevents the rule (13) from applying to (15) for it requires the 
trigger to have only one association line to the categorial gesture. 

This fact is not captured in underspecification and feature 
geometry in which umlaut is formalized as the spreading of the 
feature [-back], for the feature [-back] is used only for the character- 



74 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

ization of vowels, not for underlying consonants, and thus there is no 
[-back] present in the underlying representation of consonants. 
Instead, palatoalveolars in Korean have a [-anterior] feature under 
the coronal node. Thus we cannot get structures like (15) in which 
the palatoalveor consonant and vowel /i/ share the same component 
lil. 

In DP, the articulatory components of consonants and vowels are 
not absolutely distinct and so we can see how OCP and LC can apply 
in this case. 

The same effect can result when the preceding coronal conson- 
ant is palatalized before /i/. In Korean, the alveolar consonants /t/ 
and li^l become Id and /c*"/, respectively, before /i/. 

(16) /kat^+i/ -^ [kac'^i] *[ksc''i] 'together' 
/pat^+i/ ^ [pac%] *[pa^c^] 'field (Subj.)' 
/mat +i/ -^ [maji] *[m£eji] 'the eldest (child)' 

+: morpheme boundary 

The palatalization rule can be formalized as follows in the framework 
of DP. 

(17) Palatalization 

(ICI} (IVI) 

:\.. : 

: \ : 
{111} {lil} 

This rule is also characterized as the spreading of the same com- 
ponent lil, but to the preceding consonant not to the vowel. Palatal- 
ization precedes umlaut since the palatalized consonant is more adja- 
cent to the trigger. The result of palatalization can be seen in (18). 

(18) /k/ /a/ /cV HI 

{IVI} {CI} {IVI} 

: :\.. : 

: : \ : 

{lal} {111} {lil} 

Note however that the configuration in (18) which results from pala- 
talization is of the same nature as that in (15) with underlying 
palatoalveolar. As a consequence, here, too, LC prevents umlaut from 
applying, since the potential trigger lil is associated to two categorial 
components. 

Again, this account is not possible under the framework of 
underspecification and feature geometry. Some reasons for this are 
given below. 



Kang: Coronal: Transparent or opaque? 75 

First, since alveolars in Korean would have [-back] by the redun- 
dancy rules, spreading of [-back] to the preceding alveolars does not 
lead to the correct result directly. Therefore, we cannot make use of 
Hayes's Linking Constraint. 

Second, since alveolars in Korean, regarded as the most under- 
specified segments according to UT, have no class nodes under the 
place node, there is no landing site for the spreading feature [-back]. 
To overcome this, we might adopt the Node Generation Convention 
(Archangeli & Pulleyblank Forthcoming), which guarantees the 
missing intermediate class nodes needed for connecting the terminal 
features to the upper nodes. However, the notion of NGC is contrary 
to what Paradis and Prunet (1989) claim concerning the transpar- 
ency property of coronals, for if we accept the NGC then all spreading 
features will land on the place node of coronals instead of passing 
through them. 

The other frequently found coronal in this environment is the 
lateral. It also blocks the application of umlaut. 

(19) /allita/ -^ *[2eRida] [aRida] 'to let (them) know' 
/sallita/ -> *[saelA,ida] [saRida] 'to make (s.b.) alive' 
/c allita/ -> *[caslA.ida] [calXida] 'to be cut' 
/kallita/ -> *[keRida] [kalXida] 'to be caught' 
/pallita/ -> *[peRida] [pslXida] 'to split' 

The examples in (19) are cases of Causative (or Passive) in Korean, in 
which the first verb stem is the head of the word. The phonological 
structure is shown in (20). Since the Causative (or Passive) mor- 
pheme -// appears only after lateral, it is considered total assimil- 
ation of the preceding consonant. 

(20) /a/ /I/ /I/ /i/ 

{IVI} {IV;V:CI} [IV;V:CI} (IVI) 

: \ / : 

: \ / : 

{lal} {111} (lil) 

In (20) the component lil spreads to the preceding coronal resulting 
in the palatalized lateral [X], and this is a postlexical rule because [X] 
is not a phoneme of Korean. Then the structure would be as follows. 

(21) /I/ /I/ /i/ 

{IV;V:CI} {IV;V:CI} (IVI) 
\ / \ ... : 

\ / \ : 

{111} {lil} 

Again in (21), the Linking Constraint prevents the spreading of the lil 
component to the preceding vowel. 



76 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

There are some counterexamples to the assumption that inter- 
vening coronals block the application of umlaut. 

(22) /mati/ -^ [masdi] 'knot' 
/palita/ -> [perida] 'to throw away' 
/calita/ -^ [cerida] 'to be sore' 
/c'^alita/ -^ [c^'aerida] 'to prepare (food)' 
/talita/ -^ [tasrida] 'to iron' 

The examples in (22) can be explained in the following way. They do 
not undergo the palatalization rule because the structural (morpho- 
logical or phonological) conditions for the rule are not met. In the 
first case, palatalization does not occur since mati has no internal 
morpheme boundary. Therefore, the lil component in the trigger 
cannot spread to the coronal consonant. As a consequence, it is free 
to spread to the preceding vowel. 

The lateral consonant of the remaining examples cannot be pal- 
atalized because another phonological condition, requiring two conse- 
cutive laterals before /i/ (i.e., 1 -^X /I i), is not met. Instead they 

are weakened to [r]. As a result, lil can spread to the preceding 
vowel. 

Finally, there are cases in which two different consonants inter- 
vene between the trigger and the target of umlaut. In this case, if 
one of them is coronal, umlaut does not take place. 

(23) a. /namkita/ -> [nasrigida] 'to leave (something)' 

/ankita/ -^ [asrigida] 'to be embraced' 

b. /canti/ -^ *[caendi] 'grass' 
/pantispul/ -> *[paentitp'ul] 'firefly' 
/carjki/ -^ *[caeriki] 'oriental chess' 

c. /palphita/ -^ *[paelp*'ida] 'to be trodden' 
/salphita/ -^ *[saelp''ida] 'to search into' 
/palkhita/ -^ *[paelk^da] 'to uncover' 
/8psi/ -4 *[epsi] 'without' 

d. /capki/ -^ *[ca£pk'i] 'catching' 
/cappi/ -^ *[caEpp'i] 'general cost' 

First of all, it is necessary to explain why umlaut takes place in (23a) 
in spite of two intervening consonants. We should notice here that 
umlaut occurs after nasal place assimilation (n, m -^ tj). Thus the 
resulting structure is as follows. 

(24) {IVI) {IV;CI} {lO} {IVI} 

: \ / : 

: \ / : 

{lal} {ll,ul} {111} 



Kang: Coronal: Transparent or opaque? 



77 



On the basis of (24), I suggest that the spreading of the lil component 
cannot cross over two articulatory gestures. This is a kind of Adja- 
cency Condition. 

Since only one articulatory gesture is between trigger and target 
in (23a) = (24), there is no blocking of spreading lil. However, umlaut 
is not possible in the examples of (23b) even though their structure 
looks the same as that of (24) except for the content of the articula- 
tory gesture. This can be explained by the difference in the two 
structural descriptions. Notice that in (23a) place assimilation 
applied before umlaut, as a result of which the two consonants share 
the same articulatory gesture. In contrast, the phonological structure 
of (23b) is as follows.^ 

(25) /a/ /n/ /t/ /i/ 

{IV;CI} {CI} {IVI} 



{111} {111} {lil} 

Here the Adjacency Condition prohibits the crossing of lil over two 
articulatory gestures, thus blocking the spreading of lil. 

The last example in (23c) can be explained by s-palatalization 

-(s -^ s/ i). The component lil palatalizes the coronal fricative 

yielding the surface form [apsi], with the palatoalveolar fricative. 
This, then, is another example of palatalization in which the LC 
prevents umlaut from taking place. 

The rest of the examples can be explained as violations of the 
Adjacency Condition which prohibits crossing over two articulatory 
gestures. The structure of (23d) can be represented as the sharing of 
a categorial gesture as in (26). 



(26) /a/ 


/p/ /k/ 


/i/ 


{IVI} 


{CI} 

/ \ 
/ \ 


{IVI} 


{lal} 


{lul} {ll,ul} 


{lil} 


Conclusion 







5. 

In this paper, I have argued against the idea that the underlying 
representation of coronals does not have any nodes or features under 
the place or coronal node, as was argued by many linguists who 
favor the theory of underspecification and feature geometry. My 
argument was based on umlaut in Korean in which only coronals 
block the application of the rule. I have shown that within the 
framework of Dependency Phonology, this can be attributed to the 



78 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Linking Constraint and the Adjacency Condition, which does not allow 
the spreading of a component to cross over two articulatory gestures. 

NOTES 

*This paper has benefited greatly from discussions with Chin-W. 
Kim and comments from Jennifer Cole. Any faults are of course 
mine. 

1 In DP, dependency relations are represented by the following 
conventions: 

a : b or a<=>b 'a is equally preponderant with b' 
a ; b or a 'a is preponderant over b' 

I 

b 
a , b 'combination of two components' 

2 Linking Constraint (Hayes 1986:331): 

Association lines in structural descriptions are 
interpreted as exhaustive. 

3 (25) violates the OCP since two consonants, /n/ and /t/, have 
the same articulatory gestures. However, unless this is the correct 
underlying form, there is no way of explaining why [caendi] is not 
possible. 

REFERENCES 

Anderson, J., & C. Ewen. 1987. Principles in dependency phonology. 

Cambridge: Cambridge University Press. 
Archangeli, D. 1984. Underspecification in Yawelmani. Phonology 

and morphology. MIT Ph.D. dissertation in Linguistics. 
, & D. Pulleyblank. Forthcoming. The content and structure of 

phonological representations. Cambridge, MA: MIT Press. 
Avery, P., & K. Rice. 1989. Segment structure and coronal under- 
specification. Phonology 6.179-200. 
Hayes, B. 1986. Inalterability in CV phonology. Language 61.321- 

351. 
IversON, G. 1989. On the category supralaryngeal. Phonology 6.285- 

304. 
KiM, C.-W. 1970. A theory of aspiration. Phonetica 21.107-116. 
1973. Gravity in Korean phonology. Linguistic Research 9:2.274- 

281. 
Kim, K.-H. 1987. The phonological representation of distinctive 

features: Korean consonantal phonology. University of Iowa Ph.D. 

dissertation in Linguistics. 



Kang: Coronal: Transparent or opaque? 79 

KiM-RENAUD, Y.-K. 1974. Korean consonantal phonology. University 
of Hawaii Ph.D. dissertation in Linguistics. 

Paradis, C, & J.-F. Prunet. 1989. On coronal transparency. Phono- 
logy 6.317-348. 

SOHN, H.-S. 1987. Underspecification in Korean phonology. Univer- 
sity of Illinois Ph.D. dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



PLURAL MARKER COPYING IN KOREAN 

Han-Gyu Lee 

When the subject of a Korean sentence is plural, the 
plural marker -tul can be freely suffixed (or copied) to the 
end of any constituent within the predicate. It is an inter- 
esting question whether or not such a freely-copied plural 
marker (CPM) is different from the inherent nominal plural 
marker (IPM). This paper takes the position that they are 
different and that CPM should be treated as an independent 
grammatical entity. To prove this, I demonstrate 
grammatical (morphological, syntactic, and semantic) 
differences between IPM and CPM. A syntactic and 
semantic analysis of CPM within the GPSG framework is 
developed, using the head feature TUL. 

1. Introduction 

Kuh (1986) states that when the subject of a Korean sentence is 
plural, the plural marker -tul can be suffixed to the end of any con- 
stituent within the predicate. This phenomenon is called 'Plural 
Marker Copying'. For the present study, I will refer to the copied 
plural marker as CPM, and the inherent plural marker, which has the 
same grammatical function as the English nominal plural morpheme 
-s, as IPM. In (1), whose subject ai-lul 'children' is plural, for ex- 
ample, the parentheses indicate possible positions that CPM can take. 

(1) ai-tul-i Tom-eykey-() ppang-ul-() manhi-() 

child-IPM-NM to bread-AC a-lot 

cwuesseyo-().i 
gave 
'The children gave Tom a lot of bread.' 

I will argue in this paper that although IPM and CPM have the 
same phonological form, they show different grammatical (morpho- 
logical, syntactic, and semantic) behavior. Morphologically, their dis- 
tributions are different with respect to other suffixes such as case 
markers and postpositions (§2.1); syntactically, CPM requires its 
controller to be plural, while IPM has no such a restriction (§2.2); and 
semantically, the CPM forces a sentence containing it to have a 
distributive interpretation (distributivity between a subject and its 
predicate), while the IPM does not (§2.3). 



82 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

The purpose of this paper is to elucidate the grammatical prop- 
erties of CPM and to provide an account for them within the Gazdar, 
Klein, PuUum, & Sag (= GKPS) 1985 version of GPSG. To explain the 
syntactic number agreement of CPM with its controller, I propose a 
HEAD feature TUL which takes NP[PLU +] as its value, which is 
introduced by the TUL INTRODUCING METARULE (TIM). The TIM 
applies to a VP rewriting ID rule and licenses [AGR NP[-i-PLU]] on the 
VP category so that the controller NP is required to be plural 
according to the CAP. The distribution of the CPM is predicted in 
accordance with the HEAD FEATURE CONSTRAINT (HFC) and the CONTROL 
AGREEMENT PRINCIPLE (CAP (§3.1)). For the morphology of the CPM, I 
will treat the grammatical information supplied by TUL as a 
morphosyntactic feature, distributed by syntactic rules and 
principles but realized as a suffix at the lexical level by a morpho- 
logical rule (§3.1). For the semantics of CPM, I will treat the feature 
TUL as a semantically potent feature and provide a semantic 
interpretation for the distributive reading imposed by the CPM 
(§3.2). 

There are several important implications of this study. First, 
contrary to the general view that CPM is just a copying of the IPM 
occurring on the subject of the sentence, as implied by the trans- 
formation-based notion 'Plural Marker Copying" (Kuh 1986, Youn 
1990), the CPM is a grammatical entity different from the IPM so 
that they should be treated distinctly in grammar: There are two 
tul's in Korean, IPM and CPM. The present approach to CPM makes 
better predictions than an approach where only one tul is recognized. 

Second, the assumption that the CPM is required to agree with 
its clause-mate plural-subject (Kuh 1986, Youn 1990) is abandoned 
in this study (§§2.2, 3.1). There are problematic examples with an 
object-control verb where the CPM is suffixed to a daughter of an 
embedded VP and does not agree with a matrix subject, but with an 
object of the matrix verb which is the semantic subject of the 
embedded VP. So, instead of the term 'subject', I will use the term 
'controller' as defined and used in GKPS 1985: ch. 5. The controller of 
the CPM will be predicted by the CAP. 

Third, as a result of the second implication, this study supports 
the claim in GPSG that the CAP is one of the universal principles. 
Even though Plural Marker Copying seems to be a hard-to-predict 
agreement phenomenon in Korean in that the number-agreement is 
between the CPM (which can attach to any daughter category of VP 
such as NP, VP, S, AP, or PP) and its controller, it will be explained 
through the independently-motivated CAP given in GKPS 1985 
without any modification (§3.1). The work of Kuh (1986), who 
proposes a special agreement mechanism for the Korean CPM, will be 
critically reviewed in §3.4. 



Lee: Plural marker copying in Korean 8 3 

Fourth, this study supports the view that a syntactic phenome- 
non should be treated without recourse to pragmatic factors (Green 
1981, 1982). Pragmatically, CPM has two functions, focus marking 
and indication of controller-plurality. As a controller-plurality indi- 
cator, CPM is used by Korean speakers to inform a hearer that he/she 
is talking about plural controller referents, observing Gricean conver- 
sation principles. This function is crucially related to two charac- 
teristic phenomena in Korean. First, the subject (a potential con- 
troller) of a sentence in Korean can be freely deleted even though the 
verb of the sentence does not carry any grammatical information by 
which the understood subject can be inferred. Second, nominal plu- 
ral marking in Korean is optional although nominal a plural marker 
-tul exists, so that subject nouns not marked for number may be in- 
terpreted as either plural or singular. For these reasons, a Korean 
speaker uses CPM as a subject-plurality indicator to convey informa- 
tion on the number of the subject NP. And as a focus marker, CPM 
takes a position in a sentence that is related to new information. 
Take (1) for example: What position out of four possible ones CPM 
will take is determined only by conversational context, not by 
syntactic constraints. To predict such an occurrence of CPM, we could 
introduce a pragmatic feature [+FOCUS] into syntax and let the 
feature be instantiated on a syntactic category if a speaker thinks the 
constituent conveys new information and he/she is talking about 
more than one subject-referent. But this analysis raises a problem: 
To what extent are pragmatic factors treated in syntax? Until a 
satisfactory solution can be found for this problem, I consider it safer 
not to arbitrarily introduce pragmatic factors into syntax. So, 
following Green (1981, 1982), I will not consider such pragmatic 
factors in a syntactic account of the occurrence of CPM. 

Finally, CPM as a phrasal suffix has scope (precisely speaking, 
scope in relation to its pragmatic interpretation associated with its 
focus function; see the previous paragraph) over the whole phrase it 
attaches to. On the other hand, it morphologically attaches to the last 
item of the phrase. This scope discrepancy between syntax and mor- 
phology is explained by treating the feature TUL as a morpho- 
syntactic feature. 

2. Morphological, syntactic, and semantic properties of CPM 

In this section, I shall discuss the morphological, syntactic, and 
semantic properties of CPM in comparison with those of IPM. The 
discussion will provide evidence for the claim that there are two tul's 
in Korean: IPM and CPM. 



84 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

2.1. Morphological properties 

IPM and CPM attach to different categories. IPM, as a nominal 
plural marker, combines only with a noun category (see (2)), while 
CPM can attach to any syntactic category (see (3)). Tul as an IPM 
pluralizes its preceding constituent as the nominal plural morpheme 
-s in English does, but CPM does not have this function. 2 We can dis- 
tinguish CPM from IPM in (2)-(3) according to this function. If tul in 
(3a) is an IPM and (3a) means 'apples', then the expression is un- 
grammatical; for that meaning, sakwa-tul is used. Further, categories 
other than a noun cannot be pluralized (3b)-(3e). So all the under- 
lined tul's in (3) are CPMs. 

(2) haksayng-tul 
student -IPM 
'students' 

(3) a. sakwa han kay-lul-tul (NP-CPM) 

apple-one-of AC CPM 

'an apple' 

b. ppali-tul (AP-CPM) 
quickly-CPM 

'quickly' 

c. Urbana-eyse-tuI (PP-CPM) 

in-CPM 
'in Urbana' 

d. chinkwu-tul-i Tom-ul [cip-ey ka-tolok]-tul 
friend-IPM-NM AC house-to go CPM 
seltukhayyo.3 (VP-CPM) 
persuade 

'Tom's friends persuade him to go home.' 

e. saram-tul-i [Tom-i aphuta-ko]-tul] malhayyo (S-CPM) 
person-IPM-NM NM is-sick-COMP-CPM say 

'People say that Tom is sick.' 

IPM and CPM exhibit different distributions in relation to other 
nominal suffixes. First, IPM precedes all other nominal suffixes such 
as case markers (4),^ and postpositions (5). 

(4) a. ku haksayng-tul-i/ul/uy (IPM < CM) 

the student -IPM-NM/AC/GEN 

'the students (NM)/the students (AC)/of the students' 

b. *ku haksayng-i/ul/uy-tul 

the student -NM/AC/GEN-IPM 

'the students/the students/of the students' 



Lee: Plural marker copying in Korean 85 

(5) a. ku haksayng-tul-lopwute/eykey (IPM < P) 

the student -IPM-from/to 
'from/to the students' 
b. *ku haksayng-ulopwute/eykey-tui 
the student -from/to - IPM 

'from/to the students' 

When tul comes after CM or P as seen in (4b) and (5b), those expres- 
sions do not have the same meaning as (4a) and (5a). So IPM always 
precedes other nominal suffixes. 

Second, CPM is not in a precedence relation with nominative and 
accusative case markers; it may precede or follow them. 

(6) a. mas-i/ul-tul (NM/AC < CPM)5 

taste-NM/AC-CPM 
'taste' 
b. mas-tul-i/ul (CPM < NM/AC)6 
taste-CPM-NM/AC 
'taste' 

Third, CPM can not cooccur with the genitive case marker -uy 
(7a)-(7b), while IPM can (7d). If CPM cooccurs with -uy, (7a)-(7b) 
should be grammatical and have the same meaning as (7c) where no 
tul occurs, because CPM cannot pluralize its preceding constituent as 
explained in the discussion related to (2)-(3). But if tul in (7b) 
functions as an IPM so that (7b) has the meaning 'the students' book', 
(7b) becomes grammatical as shown in (7d). 

(7) a. *ku haksayng-uy-tui chayk 

the student-GEN-CPM book 

'the Student's book' 

b. *ku haksayng-tul-uy chayk 

the student-CPM-GEN book 

'the student's book' 

c. ku haksayng-uy chayk 
the student-GEN book 
'the student's book' 

d. ku haksayng-tul-uy chayk 
the student-IPM-GEN book 
'the students' book' 

Fourth, CPM comes after all postpositions. 

(8) a. Urbana-eyse/pwute/kkaci-tul (P<CPM) 

in/from/to-CPM 
'in/from/to Urbana' 
b. *Urbana-tul-eyse/pwute/kkaci 
CPM-in/from/to 
'in/from/to Urbana' 



86 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Last, IPM precedes CPM; this means that two tul's can cooccur in 
one constituent. This precedence can be inferred transitively be- 
cause IPM precedes postpositions which in turn precede CPM.^ 

(9) a. haksayng-tul-lopwute-tul (IPM < P < CPM) ' 

student-IPM-from-CPM 
'from the students' 

b. chayk-tul-i/ul-tul (IPM < CM < CPM) 
book-IPM-NM/AC-CPM 

'books' 

c. chayk-tul-tul-i/ul (IPM < CPM < CM) 
book-IPM-CPM-NM/AC 

'books' 

In sum, based on their morphological distributions, we can say 
that IPM and CPM are morphologically different and the transitive 
order between relevant suffixes is as in (10). 8 

(10) Stem # IPM # Postposition # CPM # CM (CM # CPM) 
2.2. Syntactic properties 

As predicted in (3), CPM as a suffix can attach to any major 
category. But its occurrence in a sentence is governed by syntactic 
restrictions which IPM does not obey. First, CPM requires its 
controller to be plural. The subject of a sentence is a potential 
controller in GKPS 1985. So when a subject is plural, CPM can occur 
(see (11)). However, when a subject is singular, the occurrence of 
CPM is not legitimate (see (12)). 

(11) a. haksayng-tul-i yelsimhi-tul kongpwuhayyo. 

b. haksayng-tul-i yelsimhi kongpwuhayyo-tul. 

student-IPM-NM hard -CPM study -CPM 

'The students study hard.' 

(12) *Tom-i yelsimhi-tul kongpwuhayyo. 

NM hard-CPM study 

'Tom studies hard.' 

For the case of object control verbs such as seltukha- 'persuade', the 
object of the verb controls the syntactic distribution of CPM occurring 
within the embedded VP. As illustrated in (13), the controller of 
CPM is an object of the object control verb seltukha- 'persuade', not v 
its subject: 'children' in (13a) and 'Tom' in (13b). In 3.1, I will show 
that the syntactic distribution of CPM is predicted by the CAP 
independently motivated in GKPS 1985. 



Lee: Plural marker copying in Korean 87 

(13) a. Tom-i ai-tul-ul [cip-ey-tul ka-tolok] seltukhayyo. 

NM child-IPM-AC house-to-CPM go-comp persuade 
'Tom persuades the children to go home.' 
b. *ai-tul-i Tom-i [cip-ey-tul ka-tolok] seltukhayyo. 

child-IPM-NM AC house-to-CPM go-comp persuade 
'The children persuade Tom to go home.' 

The second restriction that CPM satisfies is that it should be a 
clause-mate of its controller which is plural. When the matrix 
subject is plural and the embedded subject is singular, the 
occurrence of CPM in the embedded sentence is not permitted (see 
(14)); the opposite cases show the same results as seen in (15). 
When two sentences are conjoined and one of them has the plural 
subject and the other does not, CPM can not occur in the conjunct 
with the singular subject as seen in (16). In a word, the plural 
subject does not license the occurrence of CPM across the clause 
boundary. But IPM is not subject to these syntactic restrictions. As 
illustrated in (17), IPM occurs even when its clause-mate subject 
(controller) is singular. 

(14) a *haksayng-tul-i [Tom-i manhi-tul aphuta-ko]s malhayyo. 
b. haksayng-tul-i [Tom-i manhi aphuta-ko]s malhayyo-tul. 

student-IPM-NM NM a-lot CPM is-sick-COMP say-CPM 

'The students say that Tom is very sick." 

(15) a *Tom-i [ai-tul-i ppali talinta-ko]s malhasseyo-tul. 
b. Tom-i [ai-tul-i ppali-tul talinta-ko]s malhaysseyo. 

NM child-IPM-NM quickly-CPM run-COMP say-CPM 

'Tom said that the children ran quickly.' 

(16) a *ai-tul-i hakkyo-ey ka-ko Tom-i cayo-tul. 
b. ai-tul-i hakkyo-ey-tul ka-ko Tom-i cayo. 

child-IPM-NM school-to-CPM go-and NM sleep-CPM 
'The children go to school and Tom sleeps.' 

(17) Tom-i ku haksayng-tul-ul ttayryessta. 

NM the student-IPM-AC hit 
'Tom hit the students.' 

The third restriction on the syntactic distribution of CPM is that 
CPM cannot attach to its clause-mate subject NP (18); CPM can occur 
right after any constituent other than the controller in the sentence 
where it occurs, if it satisfies the two above-mentioned restrictions. 
In other words, its syntactic distribution is restricted within VP (VP 
Domain Condition). lO But IPM can occur on the controller as well as 
within VP, as seen in (14)-(16b) and (17). 



88 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(18) a. *haksayng-tul-i-tul hakkyo-ey kayo. 

student-IPM-NM-CPM school-to go 
'The students go to school.' 
b. *Tom-i-tul hakkyo-ey kayo. 
NM-CPM school-to go 
'Tom goes to school.' 

Another syntactic difference between IPM and CPM is that IPM 
pluralizes the NP whose lexical head it attaches to, whereas CPM does 
not turn the constituent it attaches to into a plural. In GPSG, 
plurality is represented by the HEAD feature specification [-i-PLU], 
whose distribution is governed by the HFC. So the syntactic behavior 
of IPM is explained by employing the feature specification [-i-PLU], 
but that of CPM cannot. According to HFC, the plurality of IPM is 
percolated up to the NP whose lexical head it attaches to, and [-i-PLU] 
is a part of grammatical information that the NP contains, as seen in 
(19a). 

(19) a. NP[-i-PLU] b. *NP [+PLU] 

I I 

Nl [+PUJ] I 

/ \ I 

/ \ Tom-ul-tul [-1-PLU] 

AP Nl [+?U]] AC-CPM 

I I 

I N [+PLU] 

I 1 

cohun chayk-tul [-t-PLU] 
good book-IPM 

'good books' 

For CPM, however, plurality cannot be treated in the same way as 
that of IPM, because CPM does not pluralize its preceding constituent. 
According to the analysis (19b) of CPM, Korean should allow the 
pluralization of proper nouns. Furthermore, plural forms of adverbs, 
postpositions, and verbs should exist in Korean, because CPM can be 
suffixed to any syntactic category. But this is not true. From the 
syntactic perspective, the plurality of CPM only shows that its clause- 
mate controller is plural. For this purpose, I propose a HEAD feature 
TUL which takes NP[-i-PLU] as its value. Then the expression Tom-ul- 
tul in (19b) is represented as in (20). The analysis of CPM with [TUL 
NP[-i-PLU]] will be provided in section 3. 

(20) NP[TULNP[-^PLU]] 

I 
I 
Tom-ul-tul [TUL NP[-(-PLU]] 



Lee: Plural marker copying in Korean 8 9 

In sum, IPM and CPM show different syntactic behavior; CPM 
has the clause-mate plural-controller constraint and the VP-domain 
condition on its occurrence, whereas IPM does not have such 
constraints. In addition, IPM pluralizes NP, but CPM does not 
pluralize its immediately preceding category. On the basis of these 
differences, I claim that IPM and CPM should be treated distinctly 
and I proposed the head feature specification [TUL NP[+PLU]] for the 
analysis of CPM. 

2.3. Semantic properties 

CPM forces the sentences where it occurs to have distributive 
readings; the distributive relation holds between the subject and its 
predicate. But IPM does not have this function. If each of the 
students bought a balloon, (21a) is true; that is, the number of 
students is equal to the number of balloons they bought. But the 
sentence (21b), which has no CPM, is semantically ambiguous. It has 
both a group reading that there is only one balloon such that the 
students bought it, and the distributive reading (21a) has. To 
capture the contribution of CPM to the distributive reading, I will 
treat TUL as semantically potent in section 3, and provide its 
semantic interpretation. 

(21) a. haksayng-tul-i phwungsen hana-lul-tul sasseyo. 
student-IPM-NM balloon one-AC-CPM bought 

'The students bought a balloon each.' 
b. haksayng-tul-i phwungsen hana-lul sasseyo. 
student-IPM-NM balloon one-AC bought 

'The students bought a balloon.' 

In section 2, I demonstrated the grammatical differences of IPM 
and CPM and showed why CPM and IPM should be treated distinctly. 
Based on this, we can conclude that CPM is a grammatical entity 
different from IPM; i.e., there are two tul's in Korean. 

3. Syntactic and semantic analyses of CPM 

In the previous section, I claimed on empirical grounds that 
there are two tul's in Korean, IPM and CPM, which should be treated 
distinctly in the grammar. In this section, syntactic and semantic 
analyses of CPM will be provided within the GPSG framework. The 
syntactic distribution of CPM and its number-agreement with its 
clause-mate controller will be explained using the HFC, the CAP, a 
HEAD feature TUL which takes NP[+PLU]] as its value, and the TUL 
Introducing Metarule (TIM) (3.1). Treating TUL as a morpho- 
syntactic feature, distributed by syntactic rules and principles such 
as the TIM and the HFC but realized as CPM by a morphological rule, 
I will show that it is possible to explain the scope mismatch between 
the syntax and morphology of CPM: as a phrasal suffix, CPM 



90 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

morphologically attaches to the final item of a phrase. The syntactic 
scope of CPM is represented in terms of the inherited HEAD feature 
TUL. For the semantics of CPM, I treat TUL as a semantically potent 
feature and provide its semantic interpretation which assigns a 
distributive reading to the sentence containing CPM (3.2). I also 
discuss why I treat TUL as a head feature rather than as a FOOT 
feature (3.3). Finally, I critically review a previous analysis of CPM 
(Kuh 1986) in 3.4. 

3.1. Syntactic analysis 

The HEAD feature TUL proposed in 2.3 is introduced by the TUL 
Introducing Metarule (TIM), which constitutes a core part of my 
analysis. 

(22) TUL Introducing Metarule (TIM)ii 

VP ^ W, X 

VP[AGR NP[+PLU]] -* W, X[TUL NP[+PLU]] 

This rule says that, for any ID rule that expands VP, there is another 
rule that permits VP[AGR NP[+PLU]] to dominate the daughters, one 
of which, including the lexical head daughter, has the feature 
specification [TUL NP[+PLU]]. This rule can capture the fact that CPM 
can combine with any major category within VP (see (3)), because 
the TIM permits [TUL NP[+PLU]] to appear on any daughter of VP. 
All the rules in (23)-(25) result from the application of TIM (22) to 
rules (26a), (26b), and (26c), respectively. 

(23) a. VP[AGR NP[+PLU]] ^ NP[TUL NP[+PLU]], PP[eykey], 

AP[+ADV], V 

b. VP[AGR NP[+PLU]] ^ NP, PP[eykey, TUL NP[+PLU]], 

AP[+ADV], V 

c. VP[AGR NP[+PLU]] ^ NP, PP[eykey], 

AP[+ADV,TUL NP[+PLU]]], V 

d. VP[AGR NP[+PLU]] ^ NP, PP[eykey], AP[+ADV], 

V[TULNP[+PLU]] 
([eykey] = [PFORM eykey 'to']) 

(24) VP[AGR NP[+PLU]] -^ NP, S[TUL NP[+PLU]], V 

(25) VP[AGR NP[+PLU]] -^ NP, VP[TUL NP[+PLU]], V 

(26) a. VP -> NP, PP[eykey], AP[+ADV], V 

b. VP -> NP, S, V 

c. VP ^ NP, VP, V 

Furthermore, the TIM in conjunction with the CAP captures the 
clause-mate controller condition on CPM described in (2.2). The 
specification [AGR NP[+PLU]] on the VP in the output of TIM (22) 
requires the VP to take a plural controller because the CAP ensures 



Lee: Plural marker copying in Korean 



91 



the number agreement illustrated in (27), since the plural subject NP 
is the controller of the VP. 

(27) 



NP 



CAP (i) ^ VP [AGR NP [+PLU] ] (= (23c)) 




PP [eykey] NP 



AP [+ADV, V 

TULNP 

[+ PLU] ] 



a. haksayng-tul-i Tom-eykey ppang-ul manhi-tul cwuessta 
student-IPM-NM to bread-AC much-CPM gave 

b. *Sue-ka Tom-eykey ppang-ul manhi-tul cwuessta 

NM to bread-AC much-CPM gave 

a. The students gave a lot of bread to Tom.' 

b. 'Sue gave a lot of bread to Tom.' 

So if a sentence such as (27b) has a singular subject and CPM occurs, 
it violates the CAP. But (27a) is grammatical because its subject is 
plural and it satisfies the CAP. 

The TIM also captures the 'VP-domain condition' on CPM 
discussed in (2.2), which says that CPM occurs only within the VP, by 
introducing [TUL NP[-i-PLU]] only to a daughter category of the VP. 

But my analysis of CPM, in conjunction with the CAP, captures a 
more important fact. The plural subject condition on CPM proposed 
by Kuh (1986), that CPM occurs when a subject is plural, cannot 
predict the syntactic distribution of CPM correctly because, in the 
case of object-control verbs, CPM occurring within an embedded VP 
is not controlled by the subject, but by the object which is the 
semantic subject of the embedded VP, as illustrated in (13). 

(13) a. Tom-i ai-tul-ul [cip-ey-tul ka-tolok] seltukhayyo. 
NM child-IPM-AC house-to-CPM go-comp persuade 
'Tom persuades the children to go home.' 
b. *ai-tul-i Tom-i [cip-ey-tul ka-tolok] seltukhayyo. 

child-IPM-NM AC house-to-CPM go-comp persuade 

'The children persuade Tom to go home.' 



92 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



According to the plural-subject condition, (13a) should be 
ungrammatical and (13b) grammatical, contrary to fact. This shows 
that the controller of CPM is determined by the semantic type of a 
verb. Therefore, I will use the term 'controller' in the way defined 
by the notion 'control' in GKPS (1985:88). If one adopts the notion of 
control, the distribution of CPM will be predicted by the CAP. To 
show this, I will compare two cases of object-control and subject- 
control verbs, where CPM attaches to the embedded VP and where it 
occurs within the embedded VP. 

First, when CPM occurs within the embedded VP, the TIM 
applies to the ID rule (26c) which expands that VP. The output rule 
of the TIM licenses the embedded VP local tree as shown in (28)- 
(29). And, as explained in GKPS 1985, in the case of the object- 
control verb, the embedded VP[AGR NP[+PLU]] agrees with its sister 
plural NP according to the first clause of the CAP (GKPS 1985:89), 
because that NP is the sister controller of the VP, as illustrated in 
(28), where the matrix verb is the object-control verb seltukha- 
'persuade'. (Hereafter, the symbol # will be used for NP[-i-PLU]: so 
[AGR #] and [TUL #] will stand for [AGR NP[+PLU]] and [TUL 
NP[+PLU]], respectively.) On the other hand, for a subject-controlled 
equi verb, the embedded VP[AGR NP[+PLU]] has no sister controller 
NP so that its AGR feature specification should agree with that on its 

(28) 




a) Tom-i ai-tul-ul cip-ey-tul ka-tolok seltukhayssta. 

NM children-AC house-to-CPM go-COMP persuaded 

b) *Tom-i Sue-lul cip-ey-tul ka-tolok seltukhayssta. 

NM AC house-to-CPM go-COMP persuaded 

'Tom persuaded the children/Sue to go home.' 



Lee: Plural marker copying in Korean 



93 



(29) 




CAP (i) 



VP2 a. [AGR#]/ b.*[AGR NP [-PLU] ] 
^CAPTTT) 
VPl [AGR#] V 



PP [TUL#] 



A 



a. ai-tul-i Tom-eykey cip-ey-tul kakessta-ko yaksokhayssta 
children-NM to house-to-CPM will-go-COMP promised 

b. *Sue-ka Tom-eykey cip-ey-tul kakessta-ko yaksokhayssta 

NM to house-to-CPM will-go-COMP promised 

The children/Sue promised Tom to go home.' 

mother category according to the second clause of CAP (GKPS 
1985:89), as illustrated in (29) where the matrix verb is the subject 
control verb yaks oka- 'promise'. 12 VP[AGR#] in (28) requires its 
sister controller NP to be plural, according to the first clause of the 
CAP. So the occurrence of CPM in sentence (a) is legitimate because 
the controller NP is plural, whereas it is not permissible in sentence 
(b) because its controller is not plural and violates the CAP. 

In the case of the subject-control verb yaksokha- 'promise' in 
(29), VPi[AGR#] has no sister controller, so its AGR value is required 
to agree with that of its mother VP according to the second clause of 
CAP. In sentence (b), the AGR values are in conflict, violating the 
CAP. Accordingly, the occurrence of CPM in (29b) is ungrammatical. 
These predictions are exactly the same as our intuitions. 

Next, when CPM attaches to the embedded VP, the matrix 
subject is required to be plural according to the CAP and the output 
rule of TIM, for both the object- and subject-controlled verbs. This is 
illustrated in (30). 

The 'plural-subject condition' (Kuh 1986) produces the same 
predictions for (29)-(30) as my analysis does. However, it predicts 
that (28a) is ungrammatical because the subject is singular. This 
incorrect prediction is not produced if we replace the notion "clause- 
mate plural-subject' by the notion 'control' motivated independently 
in GKPS 1985:88; CPM agrees in number with its controller. 



94 

(30) 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



NP [+PLU] <r- CAP (i) ^ VP [AGR#] 



NP (PP) 



VP [TUL#] 



a. ai-tul-i Tom-ul 
children-NM AC 

b. ai-tul-i Tom-eykey 
children-NM to 



[ka-tolok] vp-tul seltukhayssta.^ 3 
goCOMP CPM persuaded 

[kanta-ko]vp-tul yaksokhayssta. 
go-COMP CPM promised 
'The children (a)persuaded/(b)promised Tom to go.' 

Up to this point I have not mentioned anything about 
instantiations of the feature TUL. Because it is a head feature, its 
distribution in a tree is governed by the HFC. The feature 
instantiations of TUL capture several significant predictions about 
morphological facts. Discussing the morphological properties of CPM 
in 2.1, I mentioned that CPM cannot cooccur with the genitive case 
marker -My (refer to (8)). In my analysis, this is a natural result 
because the possessive NP is a prehead modifier. To see this, 
compare (31a) with (31b). 



(31) 



VP[AGR #] 

I 
NP[TUL #] 



/ \ 
/ 
/ 
NP 
I 
I 
I 

Tom-uy 
of 



Nl [TUL#] 
I 
N [TUL#] 



b. VP[TUL#] 
I 
NP[TUL #] 

/ \ 
/ \ 

/ \ 

NP[TUL#] Nl 
I I 

I N 

I I 

*Tom-uy-tul sakwa-lul 
of-CPM apple-AC 
'Tom's apple' 



sakwa-lul-tul 
apple-AC-CPM 
'Tom's apple' 

The feature TUL on the higher NP node is percolated down to its 
lexical head in (31a) according to the HFC. So the occurrence of CPM 
in (31a) is legitimate. On the other hand, (31b) is ungrammatical 
because TUL is instantiated on the non-head daughter node, violating 



Lee: Plural marker copying in Korean 95 

the HFC; the occurrence of CPM in (31b) is not allowed. In this way 
my analysis predicts how the cooccurrence of CPM and GEN is 
prohibited. 

Likewise, we can explain why CPM does not occur within 
prenominal APs and quantifiers, as illustrated in (32). 

(32) a. *yeppun-tul kay-lul b. *sey-tul sonyen-ul 

pretty-CPM dog-AC three-CPM boy-AC 

I assume that [TUL NP[+PLU]] is morphologically realized as tul 
at the lexical level by a morphological rule.i'^ Apparently, the 
surface realization of TUL shows that CPM morphologically attaches 
to the head noun, for example, in (31a). However the scope of CPM 
as a phrasal suffix is the whole NP. This scope mismatch can be 
explained this way: Although TUL on the lexical head is realized as 
tul, the inherited TUL on, for example, the NP node in (31a) 
syntactically guarantees the scope of CPM. The occurrences of CPM 
in (33a)-(33b) show the scope mismatch more explicitly. 

(33) a. haksayng-tul-i [Tom-i aphuta-ko]s-tul malhayssta. 

student-IPM-NM NM is-sick-COMP-CPM said 

'The students said that Tom was sick.' 
b. Tom-i [haksayng-tul-i aphuta-ko]s-tul malhayssta. 
NM student-IPM-NM is-sick-COMP-CPM said 

'Tom said that the students were sick.' 

In both sentences, CPM morphologically attaches to aphuta-ko 
'is-sick-COMP'. But the CPMs in (33a)-(33b) have different scopes, S 
for (33a) and V for (33b); that is, they agree with the matrix plural 
subject in (33a) and the embedded plural subject in (33b), 
respectively. This follows from the fact that, although TULs are 
morphologically realized at the same surface position in (33), the 
inherited TULs as shown in the trees in (34) determine the scopes of 
CPMs. 



(34) a. S 


b. S 


/ \ 


/ \ 


NP VP[AGR #] 


NP VP 


/ \ 


/ \ 


S[TUL#] V 


S V 




/ \ 




NP VP[AGR#] 

1 




V[TUL#] 



96 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

If CPM in (33b) has V as its scope, then we might expect it to 
occur before the COMP -ko. But a morphological restriction in Korean 
does not allow this; CPM cannot come immediately after the mood 
marker -ta (see (35)). As the result of this restriction, CPM follows 
COMP -ko in (33b). 

(35) *haksayng-tul-i aphu-ta-tul. 

student-IPM-NM is-sick-DEC-CPM 
'The students are sick.' 

Finally, CPM can occur multiply in a sentence such as (36). The 
multiple occurrence of CPM can be explained if we revise TIM(22) to 
enable more than one daughter of a VP to have the feature 
specification [TUL NP[+PLU]]. Then the TIM may allow as many 
[TUL #]s as the number of CPMs in a sentence, which results in 
licensing trees such as (37). 

(36) haksayng-tul-i Tom-eykey-tul ppang-ul-tul 
student-IPM-NM to-CPM bread-AC-CPM 

manhi-tui cwuesseyo-tul. 
much-CPM gave-CPM 

'The students gave a lot of bread to Tom.' 

(37) VP[AGR#] 



PP[TUL#] NP[TUL#] AP[TUL#] V[TUL#] 

3.2. Semantic analysis of CPM 

CPM forces the subject of a sentence containing it to have a 
distributive relation over its predicate (VP). As explained in (2.3), 
sentence (21a) is true if each of the students has the property of 
buying a balloon. To represent this interpretation, I propose to treat 
the feature specification [TUL NP[-i-PLU]] as semantically significant. 
According to the definition of 'semantically potent' given in GKPS 
(1985:224), the semantically significant feature at the highest point 
of occurrence in a tree is semantically potent. In (38), the tree of 
(21a), the circled feature specification [TUL NP[-i-PLU]] is semantically 
potent, which is licensed by the output rule of TIM (22). The 
distributive relation is formed over VP with respect to the referent 
of the NP which agrees with the value of TUL, NP[-t-PLU]. So the IL 
translation TUL' is defined as in (39) and interpreted as in (40).!^ 



Lee: Plural marker copying in Korean 



97 



(38) 



NP 



VP [AGR#] 



NP([ TUL# ]^ 



haksayng-tul-i phwungsen hana-lul-tul sassta 

student-IPM-NM balloon one-AC-CPM bought 

'The students bought a balloon each.' 

(39) TUL' = Vx ?c vVP X P X X [ P(x) ^ "vVP (x*)] 

(40) TUL' denotes the function / such that for any 
NP extension a, VP extension b, and x e a, 
/(b)(a)= 1 iff Vx/(b)(x) = 1. 

The definition and the interpretation say that TUL' takes as its 
argument any type of VP which combines with the type NP whose 
denotations satisfy the extension of the VP. So according to (39)- 
(40), if we assume that the students in (38) are Tom, Sue, and Mary, 
(38) is true when Tom bought a balloon. Sue bought a balloon, and 
Mary bought a balloon. This interpretation is equal to our 
intuition. '6 

3.3. TUL as a Foot Feature? 

We could obtain almost the same predictions as our analysis of 
CPM in 3.1 if we treated TUL as a foot feature rather than as a head 
feature. But I reject the treatment of TUL as a foot feature for two 
reasons, theoretical and empirical. In GPSG, foot features are devised 
to capture the unbounded dependency relations which reflexives, 
WH-words, and slash categories show. But the number agreement of 
CPM is confined, simply speaking, clause-internally, as discussed in 
2.2. On an empirical basis, the foot-feature analysis needs feature 
cooccurrence restrictions such as (41) or other devices to explain the 
ungrammaticality of (31b) and (32), which the head-feature analysis 
does not need. Relying on Occam's Razor, I reject the foot-feature 
analysis. 



(41) 



FCR 50: ~[TUL & GEN] 



98 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

3.4. Kuh's (1986) analysis 

Kuh (1986) develops a special agreement mechanism in the 
GPSG framework to explain the syntactic number agreement of CPM 
with its controller, as an alternative to the CAP of GKPS (1985). The 
core of his analysis consists of two statements on agreement pairs in 
Korean; 

(42) a. <NP, VP> is an agreement pair in Korean, 
b. <V, XP> is an agreement pair in Korean. 

The first element of each agreement pair is a controller and the 
second element is a controllee. Controller and controllee are sisters. 
(42a) says that a subject NP shares an agreement feature with its 
predicate VP, and (42b) says that a lexical verb shares an agreement 
feature with any one of its sister XPs. His special agreement 
mechanism works directionally, different from the CAP in GKPS 
(1985), which is symmetric. So, for the analysis of CPM, [AGR 
NP[+PLU]] of VP and XP is inherited from the number feature of NP 
and [AGR NP[+PLU]] of V, respectively. 

This analysis faces many empirical problems, because Kuh 
(1986) does not provide any device to limit the instantiations of 
[+PLU] on the sisters of the controller NP or V by his agreement pairs. 
First, in the case of simple sentences such as (43), [+PLU] must be 
instantiated on every daughter node according to his agreement 
pairs (42). He assumes that [+PLU] on the preterminal node is mor- 
phologically realized as tul. As far as I understand, Kuh (1986) does 
not mention anything about what controls the morphological real- 
ization of the feature specification [+PLU]. His analysis could explain 
the agreement of CPM with its plural subject in (43) [next page], but 
it cannot account for why all [-i-PLU]'s on the daughters of VP are 
realized as tul in (43a), why none of them is in (43b), why one of 
them, only on PP but not on NP or AP or V, is in (43c). 

Putting aside such a problem, Kuh's analysis faces more serious 
problems when his agreement mechanism (42) applies to sentences 
with embedded clauses. As illustrated in (44), the embedded VP and 
its daughters have two conflicting features, [+PLU] and [-PLU], which 
is what his agreement pairs predict. If CPM agrees with [**] (which 
stands for [+PLU]) in (44), then his analysis would predict correctly; 
(44) is ungrammatical. But if CPM agrees with [*] (which stands for 
[-PLU]), then his analysis would predict (44) to be grammatical 
against our intuition. That is, one theory produces two completely 
conflicting predictions on one sentence at the same time. 



Lee: Plural marker copying in Korean 



99 



(43) 




a. ai-tul-i Sue-eykey-tul ppang-ul-tul manhi-tul cwueyo-tul. 

b. ai-tul-i Sue-eykey ppang-ul manhi cwueyo. 

c. ai-tul-i Sue-eykey-tul ppang-ul manhi cwueyo. 
children-NM to-CPM bread-AC-CPM much-CPM give -CPM 
'The children give a lot of bread to Sue.' 



(44) 



NP [-t-PLU] 




NP [-PLU] — (42a)^ VP [*,**] 

HFC 

PP [*,**] <-(42b)—V [*,**] 



^ 



*ai-tul-i Sue-ka hakkyo-ey-tul kanta-ko malhayssta. 
children-NM NM school-to-CPM go-COMP said 

'The children said that Sue went to school.' 



100 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

In the case of object-equi verbs, a similar problem arises. Let 
me take the grammatical sentence (28a) for example. According to 
Kuh's agreement pairs (42), the occurrence of CPM in an embedded 
VP should be ungrammatical, because the matrix subject NP is sin- 
gular. The upshot is that his analysis is too strong in that it produces 
ungrammatical sentences such as (44), and too weak in that it cannot 
produce all the grammatical sentences. 

To solve the problems mentioned, some restriction on the ap- 
plication of Kuh's (1986) agreement mechanism would be required. 
His problems arise from his too special agreement mechanism. My 
analysis never faces these problems. 

4. Summary 

In this paper I developed syntactic and semantic analyses for 
CPM on the basis of the assumption that there are two tul's in 
Korean, IPM and CPM. This assumption was supported in section 2 
by demonstrating the morphological, syntactic, and semantic prop- 
erties of CPM which are different from those of IPM. For the syn- 
tactic analysis, in 3.1, I proposed the head feature TUL and the TUL 
Introducing Metarule (TIM), and showed how they predicted the 
syntactic agreement of CPM with a plural controller in harmony with 
the CAP and HFC. Treating TUL as a morphosyntactic feature, I ex- 
plained the scope discrepancy between syntax and morphology of 
CPM as a phrasal suffix. For the distributive reading forced by CPM, 
I assumed that TUL is semantically potent, and proposed its semantic 
definition and interpretation. And I critically reviewed Kuh's (1986) 
analysis in 3.4. 

NOTES 

*I would like to express my gratitude to Professors Georgia 
Green and James Yoon for their valuable comments and suggestions. 

1 The following abbreviations are used in this paper; 

NM = Nominative Case Marker, IPM = Inherent Plural Marker 

AC = Accusative Case Marker, CPM = Copied Plural Marker 

GEN = Genitive, P = Postposition 

CM = Case Marker, DEC = Declarative ending 

TIM = TUL Introducing Metarule, DAT = Dative 

Q = Question ending, TP = Topic marker 

2 Actually, this is a semantic property. However, I put it down 
here to show how to distinguish CPM from IPM. 

3 CPM in the position of (3d)-(3e) is potentially ambiguous 
because it can also be controlled by the semantic subject of the 
embedded VP (3d) and the subject of the embedded clause (3e), if 



Lee: Plural marker copying in Korean 101 

'Tom' is replaced by a plural NP such as 'Tom and Sue'. For this kind 
of scope ambiguity of CPM, refer to the discussion relating to (33)- 
(34). 

4 By saying CM, I refer to NM, AC, and GEN. I will treat the 
Korean dative case marker (DAT) -eykey (to) as a postposition 
because it, as a nominal suffix, shows behaviors which are different 
from those of CM but similar to those of postpositions. First, CMs 
have no consistent thematic roles so that their roles could be changed 
according to verbs. On the other hand, DAT keeps consistent the- 
matic roles (agent in the passive or recipient in non-passives) as 
postpositions do (e.g., -pwute 'from: (source)', -lo/ulo 'with: 
(instrumental)). Second, CMs are freely and easily deleted (i), while 
DAT are not, like postpositions (ii). Third, CMs are not compatible 
with the topic marker (TP) or delimiters (iii), while DAT and 
postpositions are (iv). Finally, as Kuh (1986) observed, CMs, on the 
one hand, and DAT and P, on the other hand, have different behavior 
with respect to their interaction with the nominal conjunctor -kwa 
(and). CMs are suffixed only to the second conjunct (v), while DAT 
and P can be suffixed not only to the second conjunct but to both 
conjuncts (vi). Based on this empirical evidence, I will treat DAT as a 
P, not as a CM. 

(i) Sue- Tom- ppang- mekkess-ni? 

NM GEN bread-AC ate-Q 

'Did Sue eat Tom's bread?' 

(ii) a *Sue-ka Tom- malhayssta. 

NM DAT talked 

'Sue talked to Tom.' 

b. *Sue-ka payk peici- kongpwuhayssta. 

NM 100 page-from studied 

'Sue studied from page 100.' 

(iii) a. *Tom-i/ul-(n)un b. *Tom-i/ul-man c. Tom-un/man 
NM/AC-TP NM/AC-only TP/only 

(iv) a. Tom-eykey-nun/man b. Tom-ulopwute-nun/man 

DAT -TP/only from -TP/only 

(v) a. [Tom-kwa Sue]Np-ka/lul b.*Tom-i/ul-kwa Sue-ka/lul 

and NM/AC NM/AC-and NM/AC 

(vi) a. [Tom-kwa Sue]NP-eykey/lopwute 
and DAT/from 

b. Tom-eykey/lopwute-wa Sue-eykey/lopwute 
DAT/from -and DAT/from 

5 Some Korean linguists, e.g. Kuh (1986), treat (6a) as un- 
grammatical. (Actually, Kuh does not consider the cases of NP-NM- 



102 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

CPM such as (i), which we can find in the Korean Double Nominative 
Construction where the first NP[NM] is plural.) But many Koreans 
show different intuitions on (6a)-(6b); some of them judge (6a) and 
(6b) equally grammatical, some prefer (6a) to (6b), and some say 
that (6b) is preferable. I think Kuh's (1986) intuition on the 
precedence between CPM and CM (here CM = NM/AC) is 
overgeneralized because I often hear sentences containing such 
expressions as (6a) in Korean movies and dramas. In my opinion, 
CPM basically follows CM in that it always follows all sorts of suffixes 
such as postpositions (see (8)) and verbal suffixes (ii). I think the 
reason why CPM may also precede CM in an NP can be inferred from 
the fact that CPM has the same phonological form as IPM so that 
people tend to use CPM in the position of IPM in an NP (that is, 
before CM) when the NP doesn't have IPM. 

(i) haksayng-tul-i him-i-tul/-tul-i seyyo. 

student-IPM-NM power-NM-UPM/-UPM-NM be-strong 

'Students are strong.' 

(ii) ip-hi-si-ess-eyo-tul 

wear-Causative-Honorific -Past- Mood -CPM 

6 The expression (6b) is ambiguous with respect to the 
grammatical status of tul. If it has the function of IPM, because IPM 
precedes CM as discussed in (4)-(5), (6b) means 'various kinds of 
taste'. But when it is a CPM, (6b) means just 'taste' as its gloss shows. 

"^ According to Kuh (1986), (9b)-(9c) are ungrammatical; two 
different tul's cannot occur in one NP, but only one kind of tul can. 
His explanation for the ungrammaticality of (9b) is that CPM occurs 
unlawfully because it should precede CM. But this is not true; refer 
to (6a) and note 3. In case of (9c), he tries to explain its un- 
grammaticality either by positing a phonological constraint, keeping 
two tul's from occurring back to back, or, following Zwicky (1985a), 
by formulating the morphological realization rule of the plural 
marker in such a way that two tul's occupy the same slot. But (i) is 
grammatical for some speakers with the right intonation. 

(i) (nehi-tul) chayk-tul-tul kaciko-wass-ni? 
you(plu)-IPM book-IPM-CPM bring-came-Q 
'Did each one of you bring your books?' 

8 Korean has abundant examples such as (i) where case markers 
follow postpositions. 

(i) a. Tom-i Seoul-ey-lul kassta. 
NM to-AC went 

'Tom went to Seoul.' 



Lee: Plural marker copying in Korean 103 

b. Seoul-ey-ka saram-i manhta. 

in-NM people-NM to-be-many 
'In Seoul, there are a lot of people.' 

9 Subject-to-object raising cases such as sayngkakha-'to think' 
and mi t -'to believe' are included in the category of object-control 
verbs, because their sister object controls the occurrence of CPM 
within the embedded VP as seen in (i). 

(i) a na-nun haksayng-tul-ul [kyosil-ey-tul issta-ko] mitnunta. 
I-TP student-IPM-AC classroom-in-CPM be-comp believe 

'I believe that the students are in the classroom.' 

b. *haksayng-tul-un na-lul [kyosil-ey-tul issta-ko] mitnunta 
student-IPM-TP I-AC classroom-in-CPM be-comp believe 

'The students believe that I am in the classroom.' 

However, some group of verbs allows CPM to occur even though 
the subject is not plural. Such a verb has no sister VP, but its plural 
object may control CPM occurring in its other sister category, except 
the verb, as illustrated in (ii)-(iii). Of course, if the subject of the verb 
is plural, it can license CPM to occur in any constituent within its 
predicate in the predictable way. In Korean, lexical causatives and 
some verbs such as ponay-'to send (= to cause to go)' and cwu-'lo 
give (= to cause to have)' which potentially have causative meanings 
belong to this group of verbs. I will not deal with these exceptional 
cases here, leaving them for future study. 

(ii) a. na-ka ai-tul-ul pap-ul-tul mekyesseyo. 

I-NM child-IPM-AC rice-AC-CPM cause-to-eat 

'I made the children eat rice.' 
b. *na-ka ai-tul-ul pap-ul mekyesseyo-tul. 

I-NM child-IPM-AC rice-AC cause-to-eat-CPM 

'I made the children eat rice.' 
(iii) a. na-ka ai-tul-ul kyosil-ey-tul ponaysseyo. 

I-NM child-IPM-AC classroom-to-CPM sent 

'I sent the children to a classroom.' 
b. *na-ka ai-tul-ul kyosil-ey ponaysseyo-tul. 

I-NM child-IPM-AC classroom-to sent-CPM 

'I sent the children to a classroom.' 

10 The VP Domain Condition remains in effect even when CPM 
occurs in a topicalized category as seen in (i), because the topicalized 
NP in (i) and its trace share exactly the same feature information 
according to the CAP and FFP. 

(i) chayk-ul-tuli ku haksang-tul-i [ ti ilkesseyojvp. 

book-AC-CPM the student-IPM-NM read 

'A book, the students read.' 



104 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

11 The TIM (22) allows only one CPM to occur within a VP. 
However, more than one CPM can appear in a sentence as seen in 
(36). We can predict this kind of multiple occurrence of CPM within 
a VP if we revise TIM (22) to enable more than one daughter of a VP 
to have the feature specification [TUL NP[+PLU]]. 

12 In Korean, the object-control verb seltukha- 'persuade' and 
the subject-control verb yaksokha- 'promise' subcategorize VP[COMP 
tolok] and VP[COMP ko], respectively, as shown in their ID rules (i)- 
(ii). For simplicity, I do not specify the COMP features in the trees 
(28)-(29). 

(i) VP -> NP, VP[COMP tolok], V[18] 
(ii) VP ^ PP[PFORM eykey], VP[COMP ko], V[19] 

13 CPM in this position is potentially ambiguous as mentioned in 
note 3. This ambiguity will be explained in the discussion relating to 
the data (33)-(35). 

I'* I assume that IPM and CPM are realized by the following 
morphological rules. As observed in (2)-(3), (i) applies only to a 
noun category, while (ii) applies to any category. 

(i) IPM Affixation: N => N-tul 
(ii) CPM Affixation: X => X-tul 

15 Korean has a distributive marker -ssik 'each'. Different from 
CPM, ssik cooccurs only with numerical expressions. Even though 
CPM and ssik both induce the distributive readings, there seems to 
be some difference in that the distributivity of CPM is rather weaker 
than that of ssik. It would be interesting to investigate why they are 
different, but a more systematic comparison of CPM and ssik must be 
left for future study. 

16 The exceptional occurrences of CPM such as (ii)-(iii) of note 9 
also contribute to the distributive interpretation of a sentence where 
it occurs. But I think that such distributivity is not triggered 
grammatically, but pragmatically; distributivity over events with 
respect to the subject referent. In other words, in a discourse, a 
speaker presupposes that there are multiple events and the subject 
referent participates in each event. So, for example, (iiia) of note 9 
has an interpretation that there were multiple sending-students-to- 
a-classroom events and in each event, T sent students to a class- 
room. It seems that this event-based distributive interpretation for 
the exceptional cases of CPM could be expanded to the general cases 
of CPM such as (38); in that case, (38) can be interpreted in the way 
that there were multiple buying-a-balloon events and all members 
of the students participated in a buying-a-balloon event each. If this 
pragmatic analysis of distributivity of CPM is successful, I think we 
could get a uniform and general way of interpreting CPM without 



Lee: Plural marker copying in Korean 105 

consideration of the semantics of CPM such as (39)-(40). But to 
arrive at this conclusion, we need more systematic study of the 
distributivity of CPM. 

REFERENCES 

Cho, Euiyon. 1988. Some interactions of syntax and pragmatics in 

Korean. University of Illinois Ph.D. dissertation in Linguistics. 
ChOE, Hyunbae. 1961. Urimalbon. Seoul. 
Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, & Ivan Sag. 1985. 

Generalized phrase structure grammar. Cambridge, MA: Harvard 

University Press. 
Gil, David. 1987. Definiteness, noun-phrase configurationality and 

the count-mass distinction. The representation of (in)definite- 

ness, ed. by E. J. Reuland & A. G. B. ter Merlen, 254-269. 

Cambridge, MA: MIT Press. 
Green, Georgia M. 1981. Pragmatics and syntactic description. 

Studies in the Linguistic Sciences 11:1.27-37. Urbana, IL: 

Deptartment of Linguistics, University of Illinois. 
1982. Linguistics and the pragmatics of language use. Poetics 

11.45-76. 
HUKARI, Thomas E. 1989. Reflexivization in English. Linguistics 

27:2.207-244. 
Kendall, Sue Ann, & James Yoon. 1986. Morphosyntactic interaction 

with pragmatics and sentence particles. CLS 22:2.54-66. 
Kuh, Hakan. 1986. Plural copying in Korean. Harvard Studies in 

Korean Linguistics 2.239-250. 
LlEBER, R. 1983. Argument linking and compounds in English. 

Linguistic Inquiry 14.251-285. 
MORGAN, J. L. 1978. Two types of convention in indirect speech acts. 

Syntax and semantics 9: Pragmatics, ed. by P. Cole, 261-280. 

New York: Academic Press. 
Selkirk, L. 1982. The syntax of words. Linguistic Inquiry Mono- 
graph, 7. Cambridge, MA: MIT Press. 
SUGIOKA, Yoko. 1984. Interaction of derivational morphology and 

syntax in Japanese and English. University of Chicago Ph.D. 

dissertation in Linguistics. 
YOUN, Cheong. 1990. A relational analysis of Korean multiple 

nominative constructions. State University of New York at 

Buffalo Ph.D. dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



TOWARD AN INTEGRATED APPROACH TO LANGUAGE 
PLANNING* 

Numa Markee 

In this paper i) discuss the administrative ecology within 
which Language Planning (LP) processes occur; ii) define 
the term 'diffusion of innovations'; iii) link the concept of 
relative advantage to cost-benefit analysis; and iv) show 
how this diffusionist perspective on LP provides an 
integrated framework for managing the diffusion of 
communicative innovations at different levels and 
perhaps between different foci of planning. 

1. Introduction 

Language planners have tended to focus on the language-related 
problems of large aggregates, such as a community, society, or nation 
(see Das Gupta 1973, Jernudd & Das Gupta 1971, Neustupny 1983, 
Karam 1974, Fishman 1974 & Okonkwo 1977). However, Fishman 
(1974), Kloss (1977), Thorburn (1971), Tollefson (1981) and Brown 
(1989) either explicitly recognize or imply that there are macro and 
micro levels of Language Planning (LP), the latter being exemplified 
by LP for such target groups as factories or educational institutions. 

More recently. Cooper (1989) has defined LP as 'the efforts to 
influence the language behavior of others with respect to the acquisi- 
tion, structure, or functional allocation of their language codes.' In 
effect, when Cooper refers to the language behavior of others without 
specifying who these 'others' might be, he is arguing that an 
exclusive concern with large aggregates is overly restrictive. 
Furthermore, this definition expands the scope of the discipline as 
this is understood by most other authorities^ in that it claims that 
the traditional defining foci of LP, namely corpus and status 
planning, must be complemented by what he calls 'acquisition 
planning.' (1987; 1989). This additional focus of interest must be 
incorporated into LP because educational systems and personnel, in 
particular language educators, are instrumental in diffusing com- 
municative innovations. 

This paper will i) discuss the administrative ecology within 
which LP processes occur, which entails recognizing at least six levels 
of decision-making that may be involved in the diffusion of a 



108 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

language or other communicative innovations: the government, the 
ministry (of education), the region, the institution, the department, 
and the classroom; ii) define the term 'diffusion of innovations'; iii) 
link the concept of relative advantage to cost-benefit analysis; and 
iv) show how this diffusionist perspective on LP provides an 
integrated framework for managing the diffusion of communicative 
innovations at different levels and perhaps between different foci of 
planning. 

2. The levels of planning 

We may conceptualize macro and micro level LP in terms of a 
decision-making structure that is constituted as shown in Figure 1 
(Kennedy 1982). There are three clarifications I wish to make about 
this diagram. First, it goes without saying that the number of levels 
of planning posited here is not exhaustive; the diagram merely 
attempts to show the kind of administrative system within which 
communicative innovations must diffuse. Other levels are easily 
identifiable. For instance, as Tauli (1974) implies, the scope of LP 
can be international. An example of this type of macro level LP is 
the Council of Europe's Modern Languages Project. This project seeks 
to facilitate the movement^ of individuals within the member states 
of the European Economic Community (EEC) by developing common 
communicative goals for instruction in the Community's twelve 
languages. These goals are known as the Threshold level, which 
represents the minimum worthwhile educational target learners 
should achieve. At the other end of the spectrum, the activities of 
Eliezer Ben Yehuda (who promoted the use of Hebrew among 
members of his family; see Fellman 1974 and Cooper 1989) 
represent a type of micro level LP whose scope is even more 
restricted than that of the classroom. And it is doubtless easy also to 
identify further intermediate levels of planning within the range 
identified above. 

Level 
Macro LP 1 Government 

2 Ministry 

3 Regional 
Authority 

4 Institution 

5 Department 
Micro LP 6 Classroom 

Figure 1: Macro and micro levels of LP (Kennedy 1982:268) 



Markee: Toward an integrated approach to language planning 109 

Second, the types of language-related problems addressed at the 
various levels of planning are qualitatively though not quantitatively 
similar. If we equate the notion of planning with what Neustupny 
(1983) terms language correction, LP at all the levels identified in 
Figure 1 is concerned with decoding a problem, developing a design 
for its removal and implementing the design. 

This position implies that there is no distinct cut-off point 
between macro and micro levels of planning. How then can we 
validly distinguish between the two? We might make an arbitrary 
decision that macro LP shades into micro LP somewhere between 
levels 3 and 4. But this is hardly satisfactory. In this regard. Cooper 
(personal communication) suggests that the distinction between 
macro and micro LP is best motivated if it is couched in terms of the 
potential for interaction between individuals. 

One criterion for deciding whether a given bit of LP is micro or 
macro might be whether all or most of the individuals constituting 
the target are in at least occasional interaction with one another. 
According to this criterion, LP for a school, a parish church, a 
neighborhood community center, a family, a department store, a 
company in the army, etc. would be micro planning, whereas LP for a 
school system, a religious denomination, a city, a department store 
chain, an army, etc., would be macro planning. 

Third, Figure 1 does not imply that all levels of planning will 
necessarily be found in all LP contexts; nor does it reflect a predilec- 
tion for a centralized, bureaucratic approach to LP, in which the role 
of lower levels is merely to implement prior decisions made by high- 
er levels of decision-making. In relatively decentralized societies 
such as the United States, for example, there is no central LP agency 
akin to the Academie Fran9aise, nor are most language education 
programs federally-funded. Thus, (with the exception of bilingual 
education paid for by Title VII funds, where the full range of levels 
displayed in Figure 1 obtains), most LP in the US is 'done' at levels 3 
through 6. Furthermore, planning is not only a top-down pheno- 
menon (though certainly Figure 1 does not exclude this possibility); it 
can also involve a bottom-up process of decision-making. 

Again, LP of this kind is most likely to occur in relatively decen- 
tralized societies. For example, the Graded Objectives phenomenon in 
Britain is a grassroots movement by language teachers which has 
been quite successful in changing the national examination system. 
This movement's goal is to ensure that foreign language instruction 
and testing should contribute to Britain's continuing integration with- 
in a multilingual EEC. However, as Hurreiz (1968) shows, bottom-up 
decision making can also occur in more centralized societies such as 
the Sudan. More specifically, in an assertion of nationalist pride, 



110 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



secondary school teachers took the lead in 1965 in Arabicizing an 
educational system that had retained English as the medium of 
instruction after independence from the former colonial power. 

In summary, even in highly centralized decision-making 
systems, macro level decisions serve as input for more detailed 
planning on qualitatively similar problems at micro levels of decision 
making. Thus, the major difference between macro and micro level 
planning is that the scope of the latter is more restricted than that of 
the former. And this normative view of planning as a cycle of 
complementary decisions that must be acceptable to all participants 
involved in the process leads us to consider (a) how innovations 
diffuse; and (b) what factor(s) impact most on the successful 
diffusion of a given language policy. 

3. The diffusion of innovations 

Practical attempts to diffuse or spread languages have a long 
history (see Bokamba 1984). However, in its modern technical sense, 
the term 'diffusion of innovations' has been introduced into the LP 
literature by Cooper (1979, 1982, and In Press), who borrows it from 
Everett Rogers, a leading scholar in the field of rural sociology. 
Following Rogers (1983:10), we may define diffusion as 'the process 
by which 1) an innovation 2) is communicated through certain 
channels 3) over time 4) among the members of a social system.' 
This process of adoption is typically described by an S-shaped curve 
(see Figure 2), which (a) shows the adoption rate for an innovation in 
terms of the percentage of adopters who take up the innovation 
within a specific time-frame; and (b) specifies the characteristics of 
adopters themselves. Thus, depending on where they are placed 
along the S-shaped curve, individuals may be categorized as ranging 
from relatively early to late adopters. 




Time 
Figure 2: An S-shaped curve 



Markee: Toward an integrated approach to language planning 111 

An innovation is 'an idea, practice, or object that is perceived as 
new by an individual or other unit of adoption' (Rogers 1983:11). 
From this perspective, the purpose of corpus planning is to develop 
and diffuse communicative innovations such as new technical 
terminology in, say, Arabic, Hebrew, or French among a unit of 
adoption, whose size may range from a nation-state to an individual. 
In French, for example, such innovations as ordinateur and logiciel 
are objectively new coinings which avoid the necessity of borrowing 
the English words computer and software respectively; see 
Benhamida 1989 and Thogmartin 1989 for further discussion of this 
issue in relation to French, and Alloni-Fainberg 1974 who discusses 
the processes of planned lexical innovation in modern Hebrew. But a 
phenomenon need not be objectively new to 'count' as an innovation. 
They can also be subjectively new to potential adopters. To give a 
status planning example, a language may enjoy official status in a 
given country (such as English in India, Singapore, or Kenya), even 
though it is the native language of few if any citizens of that country. 
This situation necessitates diffusing this second language through the 
formal educational system; and in such a context, even though 
English has objectively existed for some 1500 years, it is still a 
subjectively new innovation to the individual learners studying and 
eventually adopting it. 

As Cooper (1979) points out, the advantage of viewing LP in this 
light is that we can compare the spread of these communicative 
innovations to that of any other types of innovations, such as a new 
toothpaste, detergent, or vehicle. But a diffusionist perspective on LP 
also suggests that decision-making at all the levels identified in 
Figure 1 is essentially concerned with planning the adoption of a 
given communicative innovation. 

4. Cost-benefit analysis and relative advantage 

A useful technique in this respect is cost-benefit analysis, a 
procedure which has already been widely used in macro LP to help 
administrators narrow down policy options. But cost-benefit analysis 
can also be used to determine at micro levels of LP whether potential 
clients perceive adopting a given communicative innovation as being 
advantageous or disadvantageous to them. As we will see in Section 
5 of this paper, the development of an integrated perspective on LP 
involves rationalizing potential conflicts revealed by cost-benefit 
analysis at both macro and micro levels of LP. And perhaps this can 
also lead to a better understanding of areas of tension between 
different foci of planning. 

Thorburn (1971:256) defines cost-benefit analysis as 'an at- 
tempt to state the differences between two exactly defined alterna- 
tives in Language Planning' and illustrates this with a status planning 



112 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

example. More specifically, he shows how this technique can be used 
to select either a Language of Wider Communication (LWC) or an 
indigenous language as the most appropriate official language for an 
unspecified developing country. As Thorburn points out, cost-benefit | 
calculations in LP differ from ordinary economic calculations in that 
costs and benefits cannot be stated in exclusively economic terms. A 
good example of this problem is the inherent conflict between LP 
that is oriented to promoting economic development and/or nation- 
building and LP which seeks to cultivate cultural or religious 
authenticity (Fishman 1968). The former type of LP (exemplified by 
Singapore's choice of English as an official language) tends to promote 
an LWC, often at the expense of an indigenous language or languages. 
This solution assumes that economic and political benefits will offset 
the costs of individual and/or national alienation that this choice 
might provoke. And the latter form of LP (illustrated by neighboring 
Malaysia's adoption of Bahasa Malaysia as its official language) tends 
to choose an indigenous language, often to emphasize its cultural 
uniqueness and separateness from a former colonial power. The 
question for language planners therefore consists of deciding which 
factors should be included in the calculation, and how much 
importance should be assigned to variables such as linguistic 
nationalism that are difficult to quantify. For these reasons, as 
Fishman (1974) notes, language is a particularly difficult resource for 
cost-benefit analysis to handle well. It is at this juncture that a 
diffusionist view of LP can provide language planners with some 
useful insights. And perhaps these insights can lead to a better 
understanding of areas of tension between different foci of planning 
also. 

A diffusionist view of LP recognizes that meeting the actual j 
needs and wants of clients is crucial to designing a product that will I 
be attractive to customers. Thus, cost-benefit analysis must ask the 1 
question: 'What range of qualities should communicative innovations 
possess to ensure their successful spread among potential adopters?' 

Following Rogers (1983), there are at least five such qualities: 
relative advantage; compatibility; complexity; trialability; and ob- 
servability. Compatibility is the degree to which there is congruence 
between the cultural aspects of an innovation and the value systems 
of its potential adopters. Where such congruence is lacking, the | 
innovation will likely not be adopted. For example, efforts to diffuse 
Arabic in Southern Sudan have failed because Arabic is seen as an 
instrument of Islamization by predominantly Christian Southerners. 
Complexity is the degree of difficulty associated with adopting an 
innovation; the more complicated an innovation is perceived by 
potential adopters to be, the less likely it is that it will be adopted. 
Thus, the choice of a script for unwritten language X is often 



Markee: Toward an integrated approach to language planning 113 

constrained by what script is used for language Y with which 
language X is in contact. Trialability is the extent to which it is 
possible to try out an innovation on an incremental basis. If an 
innovation need not be adopted wholesale, there is more likelihood 
of it diffusing successfully. For example, the period of time set aside 
to implement an orderly change in the official status of one language 
in relation to another allows adopters to try out the innovation at an 
acceptable pace. Finally, observability is the extent to which an 
innovation is visible. The more visible an innovation is, the more 
likely it is that it will diffuse. Thus, new technical terminology that 
is not highly visible in the linguistic marketplace (i.e. in widely read 
journals and other technical publications) is unlikely to diffuse 
successfully. 

Other factors to be considered in relation to the diffusion or non- 
diffusion of an innovation also include individual personality traits of 
adopters and systemic constraints. In this paper, however, I will 
concentrate on the first of these variables, which Rogers (1983:15) 
defines as 'the degree to which an innovation is perceived as better 
than the idea it supersedes'. Like Thorburn, Rogers notes that the 
advantages conferred on those adopting an innovation are often 
economic; however, adoption may also result in adopters enjoying 
less tangible rewards of an affective nature, such as feelings of 
increased social prestige, convenience or personal satisfaction. 

5. Toward an integrated perspective on LP: An acquisition 
planning solution to a status planning-related problem 

How might language planners use the notion of relative 
advantage in cost-benefit analysis to select a particular innovation or 
innovations and promote their diffusion at macro and micro levels of 
planning? Let us assume that the language-related problem to be 
resolved consists of selecting either English or Arabic as the medium 
of instruction for a technological university in an Arabic-speaking 
developing country. ^ Significant constraints to be considered in 
relation to this traditional problem of status planning include i) an 
Arabic-medium secondary education system that produces students 
with extremely low entry levels of communicative competence in 
English; and ii) on-going pressure primarily orchestrated by Muslim 
fundamentalist groups to Arabicize tertiary education in order to 
promote cultural authenticity. These factors are considered as 
constraints at micro levels of planning because in the original context 
on which this discussion is based, they emerged as important 
variables after the formulation (and indeed implementation) of the 
policy at the macro level of policy-making. In this example, there- 
fore, they become particularly significant problems for middle level 
planners to solve. 



114 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



Figure 3 shows how a simplified cost-benefit analysis using four 
variables might be used to analyze the relative advantages and 
disadvantages of using English as the medium of instruction from the 
perspective of macro and micro levels of LP. Note that the analysis is 
simplified to show how such an analysis might work. It does not 
explicitly consider factors related to issues of language maintenance, 
which are particularly important in bilingual education, for example. 



Access to 

graduate 

science Economic education window on 

and benefits ^" English- ^^e world 

speaking 

countries 



Access 
to 



technology 



Macro 
LP 



Micro 
LP 



+ + + + 



Figure 3: Relative advantages and disadvantages 
of using English as a medium of instruction 

From a macro level status planning perspective (in this example, 
levels 1-4 in Kennedy's diagram in Figure 1), English scores 
positively on all four variables. That is to say, it provides access to 
science and technology, since English is the most widely used 
language of publication in scientific journals. It also provides 
economic benefits for individuals and the country as a whole, since 
the economic development of developing countries is contingent in 
large part on the ability to access science and technology through the 
medium of English. English-medium instruction also increases the 
possibility of pursuing graduate studies in English-speaking countries 
later on in students' careers, particularly when the students' home 
institution is the recipient of aid packages from donor countries such 
as the United States or Britain. Furthermore, universities in the US 
and elsewhere have stringent language proficiency requirements for 
non-native-speakers of English wishing to study for graduate 
degrees. Consequently, the more practice learners have with using 
the language, the more likely they will be able to use it adequately in 
native-speaker contexts. And finally, English can also provide a 
window on the world which broadens the learners' world view.^. 5 



Markee: Toward an integrated approach to language planning 



115 



However, from a micro level status planning perspective (in this 
example, levels 5-6 in Kennedy's diagram), the picture is quite 
different: English only scores positively in terms of providing a 
window on the world. With respect to providing access to science 
and technology, the low entry level of students' communicative 
competence in this language makes it extremely difficult for them to 
understand, much less produce English. And as regards the economic 
benefits and access to graduate education in English-speaking 
countries which learners might expect to enjoy, these are deferred 
advantages which have little or no importance in the short term. 

Let us now turn to the relative advantages and disadvantages of 
using Arabic as the medium of instruction for scientific subjects^ 
from the perspective of macro and micro levels of status planning as 
shown in Figure 4. 



Access to 
graduate 

science Economic education window on 
and benefits ^" English- ^^e world 

speaking 
countries 



Access 
to 



technology 



+ - - - 



Macro 
LP 

Micro 
LP 



Figure 4: Relative advantages and disadvantages 
of using Arabic as a medium of instruction 

From a macro level perspective, Arabic scores negatively on all 
four variables. More specifically, little if any original research in 
science and technology is published in this language. Furthermore, 
even science text books written in Arabic are difficult to obtain in 
many disciplines. Consequently, there are few economic benefits to 
be derived from studying in Arabic. Studying in this language also 
makes it more difficult for students to study for graduate degrees in 
English-speaking countries later on because they will have had less 
opportunity to use this language than students who have studied in 
English-medium institutions. And finally, although Arabic provides a 
means of communicating and identifying with the citizens of other 
Arabic-speaking countries, it does not provide a window on how 
other cultures think and act to nearly the same extent as English 
does (see note 4). On the other hand, the ideological dimension of 



116 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

this judgement should be recognized. As a REGIONAL language of 
communication, Arabic would have to be assigned a [+] value rather 
than a [-] value. And from the perspective of the proponents of 
Arabicization, the use of this language as medium of instruction 
might be seen as providing a window on the more desirable, less 
corrupt world of fellow Muslim nations. 

From a micro level status planning perspective, the use of 
Arabic as a medium of instruction scores positively in terms of 
providing access to science and technology but negatively in terms of 
the other variables. It scores positively with respect to the first 
factor because it is not necessary to use English to 'do' science and 
technology adequately at the undergraduate level. Consequently, 
rather than impeding science education, the use of the mother tongue 
facilitates the learners' access to science at this level of discoursal 
complexity. Indeed, it is probably associated with A HIGHER 
PROBABILITY OF ACADEMIC SUCCESS IN THE SHORT TERM. As in the case of 
English, the economic benefits and the possibility of attending 
graduate school in an English speaking country are deferred benefits 
which are not important in the short term. And finally, the use of 
Arabic does not open any windows on the world at this level of 
planning either. However, we must again note the ideological nature 
of this statement and acknowledge that if a higher value is placed on 
promoting students' regional, national, and cultural identity as Arabs, 
then we would have to change the [-] value assigned to this variable 
to a [+] value also. 

On the basis of this analysis, it would seem from a macro level 
status planning perspective that English represents a clearly better 
choice than Arabic as a medium of instruction. However, from a 
micro level perspective, it might be argued that Arabic is a 
marginally better choice than English. Since the primary justification 
for selecting one language rather than the other is instrumental, 
within the micro LP paradigm, Arabic is preferable because it 
enables students to understand the content of their lessons more 
easily and efficiently, at least in the beginning stages of instruction. 
Clearly, if we seek to develop an integrated perspective that will help 
us to resolve language-related problems at both macro and micro 
levels of decision-making, these conflicting conclusions must be 
reconciled. The question, therefore, is how this goal might be 
achieved. 

If for macro level status planning reasons, English is chosen as 
the medium of instruction, it is inevitable that students will have 
great difficulty in doing their course work. As one learner at 
Khartoum Polytechnic wrote: 



Markee: Toward an integrated approach to language planning 117 

Any one in his live must go to the infront, and comes 
from stage to the other one ... I came to the polytechnic in 
this year and I am afired from the study in the 
polytechnic is pure English Langutish and in the high 
secondry school the study with Arabic and the English 
Langutish is neglgable ... I came to the polytechnic and 
immediately study with English, we nearly about a month 
do not know any thing, after that you know what 
teachers said. (Markee 1986b) 

This sample of student writing is not offered in a spirit of 
ridicule but to show that the learners themselves were to some 
extent aware of the language-related problems they faced. However, 
as the last sentence demonstrates, they did not fully understand the 
true magnitude of their problems. Thus, it is clear that learners must 
be helped with their linguistic difficulties in a way that takes into 
account the limitations imposed by local constraints and clients' 
perceptions of their problems. 

The conflicting conclusions reached by different planners can 
only be resolved by recasting the original status planning problem in 
acquisition planning terms. Thus, macro level acquisition planning 
can provide part of the answer by making provision for appropriate 
resources to be allocated for this purpose; at the micro level of 
acquisition planning, curriculum designers and teachers must plan an 
effective language teaching program. This objective can be realized 
most efficiently by providing English for Specific Purposes (ESP) 
instruction. The advantages of adopting such a solution are two-fold. 
First, the relevance and the immediate advantages of an ESP 
approach are immediately apparent to the students. Second, the 
inherent flexibility of ESP, which is not committed in principle to the 
exclusive use of English as a medium of instruction, provides for the 
implementation of a dual medium of instruction policy that explicitly 
acknowledges the important role of local constraints such as the low 
entry level competence of students. More specifically, as discussed 
in Markee (1986a), this approach allows for the partial use of Arabic 
for classroom activities involving student-student interaction and 
English for instructor-student interaction. This solution has the 
advantage of reflecting actual patterns of communication in different 
domains of language use within the institution. But, in addition, it 
has the advantage of allowing micro level planners to concentrate on 
improving students' study skills in reading English, the area of com- 
municative competence with which learners get least help from their 
subject teachers. Thus, by focusing on reading rather than listening, 
a skill that learners perceive themselves to be able to cope with 
quite rapidly, ESP instructors can teach to real short and long term 
needs. 



118 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Of course, ESP solutions to problems of communicative incom- 
petence are hardly new. But what is new is that the micro level 
acquisition planning solutions outlined above are not developed in 
complete isolation from macro level status planning input. 
Normatively speaking, in a completely integrated approach to LP, 
tensions between different types of planning need to be resolved 
more efficiently through a better understanding of the interplay 
between different areas of planning. Nonetheless, the use of cost- 
benefit analysis influenced by the notion of relative advantage 
represents the beginnings of an attempt to rationalize contradictions 
between different levels and foci of planning which will lead to a 
truly integrated perspective on LP. Thus, to develop further a claim 
initially made in Markee (1986a) and further articulated in Markee 
(1989), it is in this sense that ESP may be seen as a language 
planning solution to language planning problems. 

6. Summary and conclusions 

This paper has outlined some of the levels of planning that may 
be involved in the diffusion of communicative innovations and 
concluded that the most important difference between macro and 
micro LP is quantitative, not qualitative. That is, macro LP decision- 
making affects large aggregates, while micro LP targets individuals 
who potentially have the opportunity to interact with each other. 
Furthermore, it has defined three related concepts, namely the 
diffusion of innovations, cost-benefit analysis, and relative advan- 
tage. And finally, it has shown how these notions can be used to 
integrate decision-making at different levels of planning. Ultimately, 
this integrated perspective on LP must also be capable of resolving 
tensions between the different foci of LP in a principled fashion. 

We should also expect a diffusionist framework to provide 
interesting insights in other areas of applied linguistics, particularly 
into what we may call (in contradistinction to traditional Second 
Language Acquisition research) the Sociology of Second Language 
Learning and Teaching (SSLLT); see Spolsky 1989 for similar 
arguments. For example, it would be interesting to establish whether 
the S-shaped curve that describes the diffusion of other innovations 
also describes the rate of learning (in the non-technical sense of this 
word) of morphosyntax and other communicative innovations by 
second language learners. The next step would be to investigate 
whether the five qualities that promote or inhibit diffusion 
mentioned in this paper are sufficiently powerful to explain language 
acquisition in diffusionist terms. In this regard, these five qualities 
would subsume many of the conditions for second language learning 
identified by Spolsky. Translated into empirically-testable 

hypotheses, these conditions for learning would provide applied 



Markee: Toward an integrated approach to language planning 119 

linguists interested in the macro and micro level diffusion of 
communicative innovations with a ready-made research program. 
This program would draw on two complementary research traditions, 
and would potentially contribute valuable insights to both fields. 
Confirmation of these and other related hypotheses would lend 
strong support to Rogers's claim that the same processes of diffusion 
obtain irrespective of the type of innovation that is diffusing, since 
none of the studies he mentions focus on language. The program of 
research is an exciting one, therefore, but is still very much in its 
infancy. The basic facts must still be established. And this can only 
be accomplished by launching a program of empirical research on 
these questions. 

NOTES 

* This is an expanded version of a paper presented at the 24th 
Annual Mid America Linguistics Conference, October 7 1989, 
University of Northern Iowa, Cedar Falls, lA. My thanks to Peter 
Strevens, Eyamba Bokamba and Steve Gaies for commenting on 
previous drafts of the paper. Of course, final responsibility for its 
contents remain my own. 

1 However, see also Neustupny 1983 and Prator's views on this 
issue cited in Cooper (1987, 1989) . 

2 The stated aim of the Council of Europe is to facilitate 
movement for both work and leisure in order to improve the quality 
of life of citizens and to promote the growth of intercultural contacts 
through tourism. 

3 This example is based on the author's experience at Khartoum 
Polytechnic, Sudan. See Jernudd (1979) and Mahmud (1983) for 
information on the language situation in Sudan, Yokwe (1984) for a 
discussion of recent Arabicization policies at the macro level of LP, 
and Markee (1986a) for more detailed discussion of the solutions 
reported in this paper. The analytical technique used is similar to 
the one utilized in Kennedy (1986a). 

'♦ Given its continuing internationalization, English is a resource 
for appreciating a broad range of cultures, not just those societies 
where it is the native language of the great majority of the 
population (Kachru 1985). 

5 Steve Gaies (personal communication) points out that, from a 
macro level perspective, an LWC will always score positively in 
terms of providing access to technology, promoting economic bene- 
fits, and giving access to graduate education in countries where that 
LWC is spoken as a native language. Conversely, an indigenous 
language will always score negatively on these variables. Conse- 



120 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

quently, he argues that these three variables should be collapsed into 
one because in reality it is impossible to tease them apart. This is 
ultimately an empirical issue. Meanwhile, we may concede that 
Gaies is probably correct when the choice consists of selecting be- 
tween an LWC and an indigenous language. However, the values 
need not automatically all be either [+] or [-] for these variables when 
the choice consists of selecting between two LWCs (such as English 
and French). In the face of competition from English, the importance 
of French as a language of science is receding worldwide, particularly 
in Francophone Africa (see Hamouda 1984). But it is receding even 
inside France. For example, the Institut Pasteur, a leading inter- 
national institution in AIDS research, has recently switched to using 
English instead of French as its language of publication. Thus, it is 
possible to envision that an AIDS researcher from Francophone 
Africa wishing to study at the Institut Pasteur would find that 
French-medium education in his/her country did not provide the 
best access to understanding the latest research findings about AIDS. 
Similarly, French might not be the best linguistic resource for doing 
basic research on AIDS at this institution. But given the close 
economic and linguistic ties that still obtain between France and 
Francophone Africa, a knowledge of French might provide the surest 
route to personal economic advancement, in that newly independent 
countries still tend to look primarily to the former colonial power for 
economic and other assistance. In this hypothetical situation, there- 
fore, French would score negatively with respect to access to science 
and technology and positively with respect to opportunities for per- 
sonal economic advantage. Conversely, English would score positive- 
ly with respect to access to science and technology and negatively 
with respect to opportunities for personal economic advantage. 

6 Of course, the justification for mandating English as the 
medium of instruction for non-scientific subjects is much weaker. As 
Eyamba Bokamba (personal communication) suggests, while English 
is the language associated with upward mobility in the outer world, 
the indigenous language (be this Arabic or Swahili in Tanzania, for 
example) is obligatory for internal upward mobility in almost all 
practical spheres of life: education, employment, politics etc. 

REFERENCES 

Alloni-FAINBERG, Yafa. 1974. Official Hebrew terms for parts of the 
car: A study of knowledge, usage, and attitudes. Advances in 
language planning, ed. by J. Fishman, 79-102. The Hague: 
Mouton. 

Benhamida, Laurel. 1989. Translators and interpreters as adopters 
and agents of diffusion of planned lexical innovations: The Fran- 



Markee: Toward an integrated approach to language planning 121 

cophone case. University of Illinois at Urbana-Champaign dis- 
sertation in SLATE. 

BOKAMBA, Eyamba. 1984. French colonial language policy in Africa 
and its legacies (Part 1). Studies in the Linguistic Sciences 14:1- 
36. 

Brown, Kimberley. 1989. Language planning in higher education: 
The ITA curriculum. Paper presented at the 24th Annual Mid 
America Linguistics Conference, October 7 1989, Cedar Falls, lA: 
Northern Iowa University. 

Cooper, Robert. 1979. Language planning, language spread, and lan- 
guage change. Language in public life, ed. by J. Alatis and R. 
Tucker, Georgetown University Round Table on Linguistics '79, 
23-50. Washington, DC: Georgetown University Press. 

1982. A framework for the study of language spread. Lan- 
guage spread: Studies in diffusion and social change, ed. by R. 
Cooper, 5-36. Bloomington, IN: University of Indiana Press. 

1987. Planning language acquisition. Language spread and 
language policy: Issues, implications and ideas, ed. by P. Lowen- 
berg, Georgetown University Round Table on Linguistics '87, 
140-151. Washington, DC: Georgetown University Press. 

1989. Language planning and social change. Cambridge: 
University Press. 

Das Gupta, Jyotirindra. 1973. Language planning and public policy: 
Analytical outline of the policy process related to language 
planning in India. Report of the Twenty-Third Annual Round 
Table Meeting on Linguistics and Language Studies, ed. by R. 
Shuy, Georgetown University Round Table on Linguistics '73, 
157-165. Washington, DC: Georgetown University Press. 

Fellman, Jack. 1974. The role of Eliezer Ben Yehuda in the revival 
of the Hebrew language: An assessment. Advances in language 
planning, ed. by J. Fishman, 427-456. The Hague: Mouton. 

FISHMAN, Joshua. 1968. Nationality-nationalism and nation- 
nationism. Language problems of developing nations, ed. by J. 
Fishman, C. Ferguson, and J. Das Gupta, 39-51. New York: John 
Wiley and Sons, Ltd. 

1974. Language modernization and planning in comparison 
with other types of national modernization and planning. 
Advances in language planning, ed. by J. Fishman, 79-102. The 
Hague: Mouton. 

Hamouda, Hassan. 1984. Some aspects of English for specific 
purposes in Tunisian higher education. English for Specific 
Purposes in the Arab world, ed. by J. Swales and H. Mustapha, 
29-150. Birmingham: Language Studies Unit, University of 
Aston in Birmingham. 

HURREIZ, Salih. 1968. Linguistic diversity and language planning in 
the Sudan. African Studies Seminar, paper 5. Sudan Research 



122 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Unit, Faculty of Arts, University of Khartoum. Khartoum: The 

Sudan. 
Jernudd, Bjom. 1979. The Language Survey of Sudan. Umea Studies 

in the Humanities 22. Umea: Sweden. 
, & Das Gupta, Jyotirindra. 1971. Towards a theory of language 

planning. Can language be planned? ed. by J. Rubin and B. 

Jernudd, 195-215. Honolulu, HI: East-West Center, The Univer- 
sity of Hawaii Press. 
Kachru, Braj. 1985. Standards, codification, and sociolinguistic 

realism: The English language in the outer circle. English in the 

world, ed. by R. Quirk and H. Widdowson, 11-30. Cambridge: 

Cambridge University Press. 
Karam, Francis. 1974. Towards a definition of language planning. 

Advances in language planning, ed. by J. Fishman, 103-124. The 

Hague: Mouton. 
Kennedy, Chris. 1982. Language planning. Language Teaching, 

15:264-284. 

1986. Costs, benefits, and motivation in ESP. Paper presented 

at the National Conference in Higher Education, University of 

Algiers, Algeria, March 1986. 
KlosS, Heinz. 1977. Ten types of language planning. In Language 

planning and the building of a national language, ed. by B. 

Sibayan and A. Gonzalez, 152-159. Manila, The Philipines: 

Linguistic Society of the Philipines and Language Study Center, 

Philipine Normal College. 
Mahmud, 'Ushari. 1983. Arabic in the southern Sudan: History and 

spread of a pidgin-creole. Khartoum: Khartoum University 

Press. 
Markee, Numa. 1986a. The importance of sociopolitical factors in 

communicative course design. The ESP Journal 5:1.3-16. 

1986b. Unpublished transcripts of Khartoum Polytechnic 

examinations. Khartoum: The Sudan. 

1989. ESP within a new descriptive framework. World 

Englishes 8:2.133-146 
NeuSTUPNY, Jiry. 1983. Towards a paradigm for language planning. 

Language Planning Newsletter 9:4.1-4. 
Okonkwo, Emeka. 1977. Towards an integrated theory of language 

planning. SUNY dissertation in Linguistics. 
ROGERS, Everett. 1983. The diffusion of innovations. New York: 

Academic Press. 
Spolsky, Bernard. 1989. Conditions for second language learning. 

Oxford: Oxford University Press. 
TAULI, Valter. 1974. The theory of language planning. Advances in 

language planning, ed. by J. Fishman, 49-69. The Hague: 

Mouton. 



Markee: Toward an integrated approach to language planning 123 

Thogmartin, Clyde. 1989. Franglais and lexical planning in France. 
Paper presented at the 24th Annual Mid America Linguistics 
Conference, October 7 1989, Cedar Falls, lA: Northern Iowa 
University. 

THORBURN, Thomas. 1971. Cost-benefit analysis in language 
planning. Can language be planned? ed. by J. Rubin and B. 
Jernudd, 253-262. Honolulu, HI: East-West Center, The Uni- 
versity of Hawaii Press. 

TOLLEFSON, James. 1981. The role of language planning in second 
language acquisition. Language Learning 32.2.337-349. 

YOKWE, Elizai. 1984. Arabicization and language policy in the Sudan, 
Studies in the Linguistic Sciences 14:2.149-170. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



ON AUTOSEGMENTAL FEATURE-SPREADING IN PHONOLOGY: 
EVIDENCE FROM CHIYAO* 

Al Mtenje 
(University of Malawi) 

In recent autosegmental studies abundant empirical 
evidence has been given to justify the claim that 
assimilation processes are best expressed through feature- 
spreading rather than feature-copying operations. A fur- 
ther claim involving autosegmental spreading has been 
made by Clements (1985) to the effect that in multi-tiered 
tree structures, only the feature(s) characterizing a single 
node can spread in assimilation processes. 

The present study provides evidence from Chiyao to 
support Clements' view. It shows that in this language, an 
assimilation rule which involves the features {-continuant] 
and {-f-voiced} and appears to require those features to 
spread from two different autosegmental nodes does not 
provide a genuine argument against the single-node 
spreading hypothesis. 

The claim that spreading affects single nodes is then 
shown to be more restrictive and compatible with a more 
constrained theory of assimilation. 

1:0 Introduction 

It has become increasingly clear from many recent auto- 
segmental studies on the nature of phonological representations that 
there is a need to develop a phonological model (or models) of multi- 
tiered feature representation in which some kind of hierarchical 
organization of features is recognized. In such a model, some tiers 
would be allowed to dominate other tiers. This line of argumentation 
is well illustrated in recent work by Clements (1985), Hayes (1986b), 
Mascaro (1983), McCarthy (1986), Mohanan (1983), and Sagey 
(1986), among others. 

Another area of recent research concentration has been the 
characterization of assimilation within a phonological model 
involving a hierarchy of feature representation such as that referred 
to above. The claim that has gained overwhelming support is that 
rules of assimilation involve feature-spreading rather than feature- 



126 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

copying, (cf. Clements 1985, Hayes 1986b, Schein & Steriade 1986, 
and others). There is, however, a further question that arises from 
this conception of assimilation, namely, given a phonological model 
with hierarchically organized tiers, how is spreading in assimilation 
constrained? Does feature-spreading occur on several autosegmental 
nodes or is it constrained in such a way that it only affects single 
nodes in tree structure? Clements (1985) has proposed and de- 
fended the strong and more restrictive position that spreading only 
affects single nodes. 

The present study makes two points. Firstly, it argues that at a 
time like the present when many substantial issues of feature 
geometry remain unclear and unresolved (and thus await more 
research) the most profitable strategy to adopt pre-theoretically is to 
abide by the scientific principle of keeping our assumptions about 
feature spreading to the minimum. That is, unless proven otherwise 
by empirical evidence, the tendency to multiply the number of nodes 
and tiers from where feature-spreading occurs should be avoided. 

Secondly, and more importantly, the paper provides evidence 
from Chiyao (an East-Central African Bantu language spoken in 
Malawi and other neighbouring countries) in support of the view that 
feature-spreading affects single nodes in tree structure. 

The paper is organized as follows: Section 2.0 presents a brief 
sketch of Clements' (1985) model of phonological feature geometry 
and how it relates to recent views on assimilation. Section 3.0 
presents the relevant rules from Chiyao and shows their 
compatibility with Clements' hypothesis on feature-spreading in 
assimilation. 

2.0 On feature organization 

Recently, it has been shown in several studies (cf. Clements 
(1985), Hayes (1986b), Sagey (1986) etc.) that some sets of 
phonological features consistently function as a unit with respect to 
certain phonological features (e.g. assimilation) while other 
imaginable sets do not function in such a unitary fashion. This 
observation has been taken as evidence for the need to group such 
features simultaneously as a unit at some level of phonological 
organization. 

Building on this functional unity of some phonological features 
and the multi-tiered representational approach independently made 
available by autosegmental theory, Clements (1985) proposed a 
model of feature representation in which the interesting relationship 
between simultaneous feature grouping and phonological processes 
like assimilation is said to be more naturally expressed. 



Mtenje: Autosegmental feature-spreading in phonology 



127 



Within this model, individual features are organized under 
hierarchically superordinate autosegmental nodes which Clements 
refers to as CLASS NODES. The class nodes are themselves dominated 
by yet a higher-level class node which is referred to as the ROOT NODE 
which is in turn linked to the CV tier. 

The range of class tiers is then defined as including the root, 
laryngeal, supralaryngeal, place, and manner tiers. The class nodes 
dominate one another in the following order: The root tier imme- 
diately dominates the laryngeal and supralaryngeal nodes. The for- 
mer immediately dominates such features as {voiced}, {spread}, and 
{constricted}, while the latter immediately dominates manner and 
place features. Under the manner features are included those con- 
cerned with the degree and manner of constriction in the oral tract 
which include {consonantal}, {sonorant}, {continuant}, {lateral}, and 
{strident}. On phonological criteria, the feature {nasal} is also 
assigned to the manner tier. The place features are those features 
which distinguish place of articulation in consonants and vowels, and 
they include {coronal}, {anterial}, {distributed}, {high}, {back}, and 
{rounded}. 

The hierarchical organization of these tiers is thus as shown in 
(1). 



(1) 



Laryngeal 
tier 



Root tier 




Supralaryngeal 
tier 



place tier 



{nasal 

{continuant} 

{consonantal 



{distributed} 
etc. 



128 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

According to Clements, each feature in the diagram above 
characterizes every node that dominates it (with the root and CV 
nodes being characterized by virtually all the features of the 
representation). For example, the manner node is characterized by 
the features {nasal}, {continuant}, {consonantal}, and {sonorant}. A 
phonetic segment in this model is thus defined as any element of the 
CV tier together with all the features characterizing it (i.e. all the 
features dominated by the C or V slot).i 

3.0 Assimilation and spreading in Clements' model 

As noted above, one of the major motivating factors for the 
simultaneous grouping of phonological features into separate 
hierarchical levels of representation as proposed by Clements is the 
fact that certain common types of phonological and phonetic 
processes exhibit some kind of phonological (functional) 
independence in that such processes may affect only one set of 
features to the exclusion of other logically possible sets. For instance, 
it has been commonly observed that phonological processes can 
affect laryngeal features without affecting supralaryngeal features. 
Rules of voice assimilation and (de)aspiration are typical examples. 
Conversely, phonological processes can affect supralaryngeal features 
without affecting laryngeal features. This is common in cases of 
partial assimilation where only place or manner features of a 
segment may be involved. One interesting implication of the 
observations made above is that assimilation processes may be 
expressed (in some way) within the phonological model proposed by 
Clements. Before we examine Clements' suggestions on how 
assimilation ought to be expressed in his model, a brief discussion on 
recent views on assimilation in autosegmental theory is in order. 

The linear model of phonology presented in Chomsky & Halle 
1968 views assimilation (both partial and total) as a process 
whereby one segment is altered in its feature values so as to become 
more similar to a neighbouring segment. This view essentially 
considers assimilation as a feature-copying process. In other ver- 
sions of phonology, assimilation by copying can be achieved by either 
feature-specifying or feature-changing mechanisms or both, depend- 
ing on certain assumptions about the underlying specification nature 
of the target segment. 

Recent research in CV phonology, however, suggests that 
assimilation is better expressed by feature-spreading than feature- 
copying mechanisms (cf. for instance Halle & Vergnaud 1980, 
Goldsmith 1981, Steriade 1982, Steriade & Schein 1986, McCarthy 
1986, Clements 1985, Hayes 1986a,b among others). 

Assimilation as spreading involves expanding the temporal 
domain of autosegments by adding association lines and often 



Mtenje: Autosegmental feature-spreading in phonology 129 

deleting those autosegments which have been displaced. 
Representations showing differences between spreading-cum- 
delinking and simple spreading within this model are given below in 
(2) where the double-crossed line represents delinking and the 
broken line shows relinking of the second segment to the first 
segment. 

M N M N 



(2) 


CV tier 


(a) 






Melody tier 




CV tier 


(b) 






Melody tier 



■>N 



P Q 



M N M N 



=> 



I \l 
P Q 



Here, figure (2a) shows that all the features of P are severed and the 
remaining timing slot M is then reassociated with Q yielding a 
geminate while (2b) simply shows the spreading of the features of Q 
to the timing slot M without any delinking involved. 

There are several arguments in favour of assimilation as a 
spreading process. Since these arguments are readily available in 
the literature, only a brief summary will be presented here. Firstly, 
while the view treating assimilation as feature-copying in principle 
allows for any feature or set of features to undergo assimilation, a 
spreading account of assimilation constrains assimilation rules more 
sensibly by predicting that only certain (sets of) features can 
undergo assimilation. This view thus makes it possible to articulate a 
more constrained and predictive theory of phonology. 

Secondly, the view that assimilation involves spreading accounts 
for certain behavioral properties of long segments derived by 
assimilation rules. This cluster of properties can be conveniently 
described under what Hayes (1986a) terms 'Ambiguity', 'Integrity', 
and 'Inalterability'. Ambiguity refers to the well-known case where 
long segments behave like a single segment with respect to quality- 
sensitive rules and like two segments with respect to quantity- 
sensitive rules. This property has been observed to extend to 
segments derived by total assimilation rules (for examples, see Hayes 
1986a, Clements 1986, and the references cited there). 

By assuming that geminates are single segments linked to two 
timing slots on the CV tier the property of ambiguity is easily 
accounted for. That is, total assimilation rules involving spreading 



130 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

yield multiply linked geminates. Rules affecting the melody tier will 
treat the segment as a single unit, while those affecting quantity and 
applying on the CV tier will have to affect the two slots to which the 
segmental material is linked. 

Integrity is a case where long segments including 
tautomorphemic geminates derived by assimilation rules cannot be 
broken up by epenthesis rules (cf. Abu-Salim 1980, Guerssel 1978, 
Hayes 1986a for examples). As pointed out by Schein (1981), 
Kenstowicz (1982), Steriade (1982), and McCarthy (1986), this 
property can be explained within CV phonology by the assumption 
that geminates derived by assimilation involving spreading affect 
multiply linked structures on the CV tier. Inserting a segment (e.g. a 
vowel) between them would result in the crossing of association lines 
which is forbidden by the well-formedness condition. 

Finally, Inalterability refers to the failure of segments forming 
halves of a geminate (including tautomorphemic geminates from 
assimilation) to undergo a rule they would otherwise be expected to 
undergo (for examples see Schein 1981, Kenstowicz 1982, Hayes 
1986a, and others). Hayes (1986a,b) and Schein and Steriade (1986) 
have proposed general principles that predict cases of Inalterability 
automatically for those rules that display it. Hayes's 'Linking 
Constraint' for example proposes that association lines be treated 
exhaustively for purposes of rule application. That is, a rule whose 
structural description refers to multiply linked structures cannot 
apply to segments which are singly linked and, conversely, a rule 
whose structural description refers to structures with single linkage 
cannot affect multiply linked segments. Now, this constraint 
provides a diagnostic for the mechanisms of assimilation as involving 
spreading. 

The logic is simple. If a rule of total assimilation is due to 
spreading, then it must create doubly-linked structures and 
according to Hayes's constraint, such structures, like all true 
geminates, will be inalterable by any rule that crucially refers to 
singly linked structures in its structural description; and this is 
precisely what happens. Thus we see that Ambiguity, Integrity, and 
Inalterability support the claim that (total) assimilation involves 
spreading. 

Interestingly enough, cases of partial assimilation involving 
spreading are also supported by principles like the Linking 
Constraint. Hayes (1986b), for instance, presents cases of rules of 
partial assimilation in Toba Batak, an Austronesian language, which 
obey Inalterability as predicted by the Linking Constraint. He shows 
that a number of structures resulting from rules of partial 
assimilation fail to undergo a rule of glottal formation which is 



Mtenje: Autosegmental feature-spreading in phonology 131 

formulated as affecting only singly-linked consonants. Hayes thus 
concludes that rules involving partial assimilation must be 
considered as yielding multiply-linked structures, explaining why 
they fail to undergo glottal formation. 

Having briefly reviewed the arguments for treating assimilation 
as spreading, let us now consider Clements's claims about the 
characterization of such spreading in the multi-tiered hierarchical 
representations proposed in his model. Clements has argued that 
assimilation processes only involve single nodes in tree structure. 
That is, only those phonological features dominated by a single node 
can spread to neighbouring nodes to effect the relevant assimilation 
changes. Clements's hypothesis on spreading predicts that in 
structures like (3) where spreading occurs from more than one node, 
(the letters stand for an arbitrary set of features dominated by the 
relevant nodes) the processes are independent and ought not be 
represented in a single rule. 

(3) Laryngeal tier A B 



---^.. + 



Root tier C JTD 



Supralaryngeal tier E F 

Clements (1985) discusses apparent counterexamples to this 
otherwise interesting constraint on the nature of spreading in 
assimilation and shows how they are accounted for by other general 
factors in the languages concerned. Here, we will only review the 
examples Clements cites from Kikuyu, a Bantu language of Kenya. In 
this language, there is a general process which assigns the features 
{-continuant} and {-i-voiced} to post-nasal obstruents. The relevant 
data are presented below (from Clements 1985:244). 

(4) Imperative 1st singular Stem (gloss) 

Imperfect 

(iur - a m-bur-eete 'lop off 

tem - a n-dem-eete 'cut' 

reh - a n-deh-eete 'pay' 

Cin - a n-jin-eete 'burn' 

Kom - a ri-gom-eete 'sleep' 

yor - a rj-gor-eete 'buy' 

Here, it can be noted that the obstruents following the nasal 
consonants become voiced non-continuants. Within the framework 
proposed by Clements, this assimilation process can be accounted for 



132 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

by assuming that the feature {-continuant} and {+voiced} of the 
preceding nasal spread on to the following obstruent as illustrated 
below. 2 (Here, as well as in subsequent discussions, we follow 
Clements in assuming that the feature {nasal} is a manner feature 
dominated by the supralaryngeal node). 

(5) Laryngeal tier {-i-voiced} {•} 



Root tier 



Supralaryngeal tier {+nasal} {-cont} {+obst} 

But, as Clements remarks, spreading in (5) involves two 
independent nodes, the feature {-continuant} belonging to the 
supralaryngeal tier and the feature {-i-voiced} belonging to the 
laryngeal tier. Thus, there appears to be no way within the 
framework under consideration in which this assimilation process 
could be expressed in terms of the spreading of a single node. This 
set of data, therefore, appears to constitute a counterexample to the 
single-node spreading hypothesis. 

Clements, however, proceeds to show that the Kikuyu data cease 
to be recalcitrant once other factors of Kikuyu phonology are 
considered. Particularly, he argues that the feature {voiced} is re- 
dundant in Kikuyu and therefore need not be specified underlyingly. 
This therefore implies that in a tree structure such as (5) the feature 
{voiced} need not be included, which means that it cannot spread. 
The argument that voicing is redundant in Kikuyu is based on the 
following factors: Firstly, all sonorants in Kikuyu are voiced, 
secondly voicing is also predictable in obstruents: Stops are voiced 
after nasals, otherwise they are voiceless; fricatives are also always 
voiced. This would then seem to suggest that the assimilation rule in 
(5) only involves the feature {-continuant} and a redundancy rule 
assigns the feature {+voiced} later to the segments to yield the 
correct output. The Kikuyu assimilation process therefore does not 
provide a genuine counterexample to the claim that only single nodes 
assimilate in tree structure. 

Clements goes on to suggest that this interesting intersection 
between phonological redundancy on the one hand and assimilation 
and feature representation on the other may turn out to be a vital 
clue in explaining some cases of apparent exceptions to his views on 
feature-spreading in assimilation in the sense that one or more of the 
features involved may not yet be present in representations, such 
features being added later by redundancy rules. 



Mtenje: Autosegmental feature-spreading in phonology 133 

It is my conviction here that Clements's hypothesis defines a 
more constrained theory of assimilation and, unless challenged by 
empirical evidence, it should be considered a viable hypothesis. 

In this paper, I use evidence from Chiyao to support the single- 
node spreading hypothesis. I argue that a series of assimilation rules 
in Chiyao involving the same features as in Kikuyu (i.e. {-continuant} 
and {-i-voiced}) which appear to require spreading to affect more than 
one node present further supporting evidence for the view that most 
of the apparent exceptions to the single-node spreading hypothesis 
can be explained in terms of one of the features being redundant. 
That is, the notion of redundancy provides considerable insight into 
the nature of feature spreading in multi-tiered representations and 
assists in our efforts to preserve a more constrained theory of 
phonology. 

4.0 Assimilation in Chiyao: Apparent counterevidence 

Chiyao has a class of assimilation processes which are, in many 
respects, similar to the Kikuyu case. 3 There are several assimilation 
rules in this language (a full discussion of such rules is available in 
Mtenje In Preparation) but the rules that will concern us in this 
paper are the following: post-nasal stop formation I, consonant- 
voicing, and post-nasal stop formation II. We will first formulate 
these rules individually before collapsing them into one general 
assimilation rule. 

4.1 Post-nasal stop formation I 

This rule changes /I/ to /d/ when it occurs after a nasal 
consonant in the perfective tense. Consider the following forms 
where /ku/ is the infinitive marker and the /a/ at the end of the 
verb is the final vowel characteristic of Bantu languages. (I omit 
tone details here and in all subsequent data because they are 
irrelevant to the present study). 

(6) a) ku-lapit-a 'to lick' 

b) ku-lirig-a 'to try' 

c) ku-lokot-a 'to pick up' 

d) ku-lila 'to cry' 

Now, the perfective tense in Chiyao triggers interesting 
phonological processes (see Mtenje In Preparation for more 
examples). One such process is the change of /I/ to /d/ in post-nasal 
environments. A characteristic way of expressing this tense is by 
prefixing a subject marker (SM) to the root and then suffixing either 
-e or -He to the root as the tense marker (TM). The phonological 
alternations triggered through this process are illustrated in (7) 
below (syllabic nasals are noted by subscript dots). 



134 Studies in the Linguistic Sciences 21:1 (Spring 1991) 



(7) 


SM - 


Root - 


TM 






a) 


n 


lapit - 


e ^ 


ndapite 


'I have licked' 




m - 


lapit - 


e -» 


mlapite 


'You (singular) 
have licked' 




a 


lapit 


e -> 


alapite 


'He/she/they 
have licked' 




tu - 


lapit - 


e -* 


tulapite 


'We have licked' 


b) 


n 


liTlg 


ile-^ 


ndiiijile 


'I have tried' 




m 


liTlg 


ile^ 


rnliiijile 


'You (singular) 
have tried' 




a 


liTlg 


ile^ 


alinjile 


'He/she/they 
have tried' 




tu - 


liTlg 


ile^ 


tuliiijile 


'We have tried' 


c) 


n 


lokot - 


e -* 


ndokwete 


'I have picked' 




m - 


lokot - 


e -» 


mlokwete 


'You (singular) 
have picked' 




tu - 


lokot - 


e ^ 


tulokwete 


'We have 
picked' 


d) 


n 


HI 


ile-» 


ndisile 


'I have cried' 




m 


HI 


ile^ 


mlisile 


'You (singular) 
have cried' 




a 


HI 


ile^ 


alisile 


'He/she/they 
have cried' 



Here, I am mainly interested in the alternation between /I/ and 
/d/ shown root-initially in the forms ndapite, ndinfile and ndokwete. 
(For changes which are irrelevant to the present study such as the 
palatalization of /g/ to /j/ before /i/ in (7b), the change of ko to kwe 
in (7c) and the /l/~/s/ alternation in (7d), see Mtenje In Preparation 
for details. Note that /I/ changes to /d/ only when the preceeding 
nasal is tautosyllabic with it. When that nasal belongs to a different 
syllable as in the forms mlapite, mlokwete, mlinfile and mlisile, the 
change fails to occur. 

Within the framework assumed above, where assimilation in- 
volves spreading, this process can be accounted for by a rule such as 
that given in (8) where the feature {-continuant} characterizing the 
nasal consonant spreads on to the following lateral. (Rule (8) shows 
the relevant tiers only).'* 



Mtenje: Autosegmental feature-spreading in phonology 135 

(8) Post-nasal stop formation I 

Manner tier {-i-nasal} {-cont} {-i-lateral} 

Supralaryngeal • "^^ 



Place tier {•} {•} 

4.2 Consonant voicing 

This rule voices a consonant which follows a nasal consonant. 
Consider the following data: 

(9) a) ku - pel - a 'to be tired' 

b) ku - kat - a 'to cut' 

c) ku - timb - a 'to beat' 

Now consider what happens when the verb occurs in the perfective 
tense as shown in (10). 

(10) SM ROOT TM 

ile -*■ mbesile 'I am tired' 
ile -> Tigatile 'I have cut' 
ile -> ndimbile 'I have beaten'5 

Here, the post-nasal stops are voiced.^ It may be worth pointing 
out that like post-nasal stop formation I in (8), consonant voicing 
applies only to consonants which are tautosyllabic with the nasal. 
Note that the syllabic bilabial nasal /rn/ does not trigger this rule as 
shown in (11). 

(11) m - pel - ile -* mpesile 'You are tired' 
ile -* rnkatile 'You have cut' 
ile -> mtimbile 'You have beaten' 

This voicing process can be accounted for within the spreading 
framework by assuming that the feature {voiced} from the nasal 
spreads on to the following consonant as shown below. (I assume the 
tautosyllabicity condition on the relevant segments.) 



a) n 


pel 


b) n 


kat 


c) n 


timb 



m - 


pel 


m - 


kat 


rn 


timb 



136 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

(12) Consonant Voicing 

Laryngeal tier {+voiccd} 



-voiced} 



Root tier 



Supralaryngeal tier 



+nasal] 



+consonantar 



Let us now consider post-nasal stop formation II. 

4.3 Post-nasal stop formation II 

This rule changes the labial glide /w/ into the voiced bilabial 
stop /b/ in post-nasal environments. This change is shown in (14) 
where the initial glides in (13) occur in the appropriate post-nasal 
environments. 



(13) a. 


wugul - a 'open' 




b. 


walarig - a 'read' 




c. 


wug - a 'cook' 




d. 


wik - a 'arrive' 




e. 


wecet - a 'talk' 




(14) a. 


a - n - wugul - ile ^ ambugulile 
you - me - open - for 


- 'You open for me 


b. 


n - walarig - ile ^ mbalasile'' 
I - read - perfective. 


'I have read' 


c. 


n - wug - ile -^ mbusile^ 
I - cook - perfective 


'I have cooked' 


d. 


n - wik - e ^ mbice^ 
I - arrive - perfective 


'I have arrived' 


e. 


n - wecet - e -* mbecete 
1 - talk - perfective 


'I have talked' 



Again, like Post-nasal Stop Formation 1 and Consonant Voicing, 
this rule also applies to tautosyllabic segments only. This can be 
observed in the following forms involving the syllabic bilabial nasal 
where the rule fails to apply. 

(15) a. a - rn - wugul - ile -♦ amwugulile 
He/she - you - open - for 
'He/she should open for you (singular)' 
b. m - walarig - ile -* rnwalasile 
you - read - perfective 
'You (singular) have read' 



Mtenje: Autosegmental feature-spreading in phonology 137 

c. m - wug - ile -» mwusile 
you - cook - perfective 

'You (singular) have cooked' 

d. m - wecet - e -* rnwecete 
'You (singular) have talked' 

From the point of view of spreading, we can account for this 
process by assuming that the manner features of the glide are 
severed and the feature {-continuant} characterizing the nasal 
spreads to the supralaryngeal node of the glide, thus appropriately 
changing it into a corresponding non-continuant with the same place 
of articulation as /w/, namely /b/. The rule achieving this result can 
be formulated as follows (assuming, again, the tautosyllabicity 
condition). 



(16) Post-nasal stop formation II 

Manner tier {+nasal} {-cont) 



(syll V 
L-cons J 

. + 



Supralaryngeal 

Place tier {•} f+ant 



anti 

cor J 



4.4 Formulating a General Rule 

Now, although rules (8), (12), and (16) are formulated as 
separate and independent rules, it can easily be noted that they 
share several elements in common. Firstly, in virtually all cases the 
segments affected occur in post-nasal environments. Secondly, in 
two of the rules ((12) and (16)), the spreading feature is {-con- 
tinuant}. Finally, in all cases the process involves changing a con- 
sonant into a voiced non-continuant obstruent. It is thus obvious that 
stating these properties in three separate rules misses the crucial 
generalization that all the rules involve the feature {4-voiced} and {- 
continuant}. One general rule is therefore needed to capture all these 
generalizations. The general rule, given below in (17), specifies that a 
post-nasal consonant becomes a voiced non-continuant through the 
spreading of the features {-i-voiced} and {-continuant}. 



138 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(17) Voiced non-continuant formation 
Laryngeal tier {+voiced} {•} 

Root tier • "^^ 



Supralaryngeal tier {+nasal} (-cont) {-syllabic} 

At this point, let us review Clements's hypothesis on feature- 
spreading in assimilation processes and assess the extent to which it 
is compatible with rule (17) above. It may be recalled that Clements 
argues that spreading in assimilation involves only single nodes, i.e. 
only a feature or features dominated by a single node can spread. 
Rule (17) above which is in many respects similar to the Kikuyu 
assimilation rule presented in (5) requires the spreading of at least 
two features {-continuant} and {+voiced} from two different nodes, 
namely, the laryngeal and the supralaryngeal nodes. 

Now, the crucial question that can be asked in relation to 
spreading in rule (17) is whether the rule is really incompatible with 
Clements's hypothesis and ipso facto indicates that the hypothesis is 
too strong and ought to be relaxed to allow spreading to affect more 
than one node in tree structure. The main claim of this paper is to 
demonstrate that, on the contrary, the Chiyao data do support 
Clements's single-node spreading in assimilation. I turn to this issue 
immediately. 

4.5 Redundancy in Chiyao 

It may be recalled that Clements accounts for the apparent 
contradiction to his hypothesis by arguing that in the Kikuyu data 
one of the features is redundant. I wish to argue here that a closer 
look at the phonology of Chiyao also shows that some of the features 
involved in rule (17) above are predictable and redundant and 
therefore need not be included in the formulation of that rule. 

4.5.1 Voicing in Chiyao 

The distribution of the feature {± voice} in Chiyao is interesting 
in a number of respects. To start with, note that the feature is 

underlyingly distinctive for non-continuant obstruents in V V and 

# V positions as shown in the data below. 1 2 



Mtenje: Autosegmental feature-spreading in phonology 139 



(18) 



/p/ 






Ihl 






ku-pela 


'to 


get tired' 


ku-barigula 


'to 


roar' 


ku-panda 


'to 


plant' 


ku-bindicila 


'to 


stay indoors' 


ku-ponda 


'to 


tread' 


ku-bendula 


'to 


break' 


III 






/d/ 






ku-tama 


'to 


sit' 


ku-dandaula 


'to 


worry' 


ku-timba 


'to 


beat' 


ku-delela 


'to 


despise' 


ku-tota 


'to 


sew' 








/k/ 






/g/ 






ku-kata 


'to 


cut' 


ku-ganda 


'to 


be thin' 


ku-kulukutala 


'to feel 


ku-gomba 


"to 


play music' 






rough' 


ku-gumba 


'to 


mould' 


Icl 






/J/ 






ku-cacuka 


'to 


rush' 


ku-Jaliwa 


'to 


be blessed' 


ku-coma 


'to 


burn' 


ku-Jembecela 


'to 


wait' 



This would appear to suggest that the feature {± voice} is 
nonredundant in crucial respects and needs to be underlyingly 
specified for all obstruents. However, notice that there are other 
environments (apart from derived post-nasal positions where rule 
(16) predicts the voicing) in which voicing redundancy is observed. 
There are underived (underlying) contexts in which the sequence 
Nasal-i-Voiced Stop (e.g. nd,mb,ng) is attested. '3 In these 
environments, the sequence Nasal + Voiceless Stop (e.g. *nt, *mp, and 
*nk) does not occur. This therefore shows that the {± voice) contrast 
is non-distinctive in that context. The implication is that in this 
position, the voicing value of the post-nasal stop need not be 
specified underlyingly. A redundancy rule, which would essentially 
be a feature-specifying rule, would fill in the missing value at a later 
stage. 

On the other hand, in derived environments we do get a neutral- 
izing/feature-changing effect where an originally voiceless obstruent 
becomes voiced after being attached to a nasal consonant. These are 
the cases which are handled by rule (17) above. We thus get into a 
dilemma where on the one hand, (in underived contexts) voicing is 
considered to be redundant, and where on the other hand, (in 
derived environments) the same feature is regarded as non- 
redundant. Obviously, positing two separate rules for the two 
contexts (one a redundancy and the other a feature-changing rule) is 
unattractive and unnecessary. 

While a satisfactory solution to this problem still needs to be 
found, the case, nevertheless, forces one to speculate that before a 
feature becomes completely redundant it goes through a stage of 



140 



Studies in the Linguistic Sciences 21:1 (Spring 1991) 



partial redundancy where its occurrence is predictable in some 
context(s). As the contexts of redundancy become wider, the 
feature's degree of redundancy also increases until it becomes 
predictable in all contexts and thus rendered completely redundant. 
I believe that this change from (feature) distinctiveness or total 
contrast to complete redundancy is already affecting the feature 
{voice} in Chiyao. It would thus not be surprising if at a later stage 
in the development of the language, this feature were to become 
entirely redundant and not be required in rules like (17) above. 

The implication of this observation for the spreading of voicing 
in rule (17) is that although the feature is contrastive in cases like 
(18), there is still a nonnegligible element of redundancy in that 
feature and given a thorough and more refined theory of feature 
representation (such as that proposed by Sagey (1986) and Clements 
(1987)) redundancy of this type can be appropriately characterized 
in ways which may probably not require voicing to be underlyingly 
specified and needed in assimilation rules like (17). 

There is another theoretical issue which needs to be addressed 
in relation to voicing in clusters of the type N+C. In most Bantu 
languages, clusters like these are represented as in (19b) rather than 
(19a). 



(19a) 



O 



(19b) 



O 





That is, the combination of a nasal and consonant constitutes a 
complex (contour) segment. Now, the question which arises is 
whether we want our theory of feature representation and geometry 
to allow for all kinds of contour segments including those which start 
out as {+voice} and end up {-voice}. The crucial fact worth noting is 
that such complex segments, (as far as I know) have never been 
attested in any natural language, although the theory does not 
exclude such cases (cf. for instance Lieber 1987:20 for a similar case). 

The point being made here then is that an adequate theory of 
phonological feature representation should predict that cases like 
post-nasal consonant voicing in Chiyao which result in voicing 



Mtenje: Autosegmental feature-spreading in phonology 141 

agreement within a complex segment are not surprising since they 
follow naturally from a general and universal effect characterizing 
feature composition and relations within such types of segments. 
That is, the theory ought to predict, in some way, the universal effect 
that no complex segment starts out as {+voice} and ends up as 
{-voice} and that the application of voice agreement (rules) simply 
avoids that unnatural situation. But notice now that if it is indeed 
true that such voice agreement phenomena are a result of a 
universal effect governing feature co-existence in complex segments, 
as it appears to be the case, then it is possible that formal rules like 
post-nasal consonant voicing which were needed to achieve this 
effect could very well turn out to be redundant and unnecessary, 
since those universal effects could be stated through some formal 
devices such as default rules, redundancy statements, filters, or 
formally equivalent mechanisms. Such a view of post-nasal voicing 
processes would thus render the inclusion of the feature {±voice) and 
its eventual spreading in the assimilation rule (17) unnecessary and 
ipso facto get rid of the two-node spreading problem. 

4.5.2 The feature {continuant} 

One issue which also undermines the two-node spreading 
problem and supports Clements' proposal is the status of the feature 
{ ±continuant} in obstruents. In the preceding discussion on 
assimilation in Chiyao, we assumed that this feature is underlyingly 
distinctive and therefore it had to be represented on a separate tier 
in rule (17) from where it spreads. However, close examination of 
the phonology of the language shows that this feature is redundant 
and need not be included in rule (17). The facts leading to this 
conclusion are as follows: In its inventory of obstruents, Chiyao has 
only one fricative sound, namely /s/. (Notice, as a matter of interest, 
that voicing in fricatives is therefore redundant since there is only 
one fricative sound in the language whose {-voice} value could be 
supplied by default.) Now, and this is the crux of the argument, the 
entire range of obstruents can be adequately distinguished by using 
the feature {±strident}. The (only) fricative /s/ can be specified as 
{-(-strident) and the remaining set of obstruents, i.e. stops and 
affricates, will automatically be characterized as {-strident}. The 
feature {-continuant) which also characterizes the nonstrident 
obstruents, is then effectively rendered redundant and can be filled 
in by a default rule later, after the main assimilation process which 
marks all {-strident} obstruents as {-continuant}. 

The observation made above about the feature {±continuant) 
being redundant now effectively gets rid of the two-node spreading 
problem presented by rule (17). That is, the fact that this feature is 
redundant means that it need not be included in rule (17) and 



142 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

therefore it cannot spread in the manner indicated in that rule. Thus 
even if we were to assume (for the sake of argument) that the 
feature {± voice} is not sufficiently redundant to be excluded from 
rule (17), the Chiyao assimilation process does not offer counter- 
evidence against the hypothesis that spreading in assimilation only 
affects single nodes in tree structure, since only the feature {±voice} 
would spread from the laryngeal tier (with the feature {-continuant} 
being supplied by default later in the derivation). Thus Chiyao 
provides interesting supporting evidence for the strong and more 
restrictive position that only single nodes can spread during 
assimilation. Furthermore, it supports Clements's claim that the 
intersection between phonological redundancy and assimilation on 
the one hand and feature representation on the other provides one of 
the most reliable clues in explaining cases where, contrary to the 
single-node spreading hypothesis, spreading appears to affect two 
different autosegmental nodes. A thorough investigation of the 
phonology of the language(s) concerned may very likely show one or 
more of the features involved to be redundant and thus not available 
for spreading at the time when one of the features assimilates. 

5.0. Conclusion 

This paper has argued that the notion of phonological 
redundancy provides considerable insight into problems related to 
feature spreading in assimilation. It has been" shown that when 
thoroughly investigated, feature spreading which appears to 
contradict Clements's single-node spreading hypothesis could be 
easily explained in terms of one of the features being redundant and 
thus not being available in tree structures at the time when 
assimilation occurs. 

It has been demonstrated that the spreading of the feature 
{-continuant} and {-i-voice} in Chiyao, for instance, which appears to 
affect two different autosegmental nodes does not constitute 
counter-evidence against the hypothesis that only single nodes 
spread, because one of the features {-continuant} is redundant and 
thus does not feature in the main assimilation rule. Besides, there is 
some indication that even the feature {±voice} is redundant to some 
degree (although not completely) and, given an elaborate theory of 
redundancy and feature representation, may turn out to be 
unnecessary too in the assimilation rule. 

Finally, the discussion has also helped to show that it is a priori 
desirable to maintain the strongest possible hypothesis consistent 
with the scientific requirement of the simplicity hypothesis such as 
the single-node spreading, rather than to start with a weak 
hypothesis which fails to impose the strongest possible constraint on 
the phenomena being studied. In terms of spreading, the weak 



Mtenje: Autosegmental feature-spreading in phonology 143 

position is the one that would allow spreading to affect virtually any 
number of nodes. The need for strong constraints is particularly 
pertinent when the area being investigated — such as that involving 
tier structure and composition — still requires much research work to 
be done. 

NOTES 

*This paper was originally prepared for presentation at the 19th 
Annual Conference on African Linguistics, Boston University, April 
1988. I would like to thank Tom Likambale and Hari Kambwiri for 
the Chiyao data. I would also like to express my gratitude to John 
Harris for making very helpful comments and suggestions on an 
earlier draft. The nature of argumentation presented here directly 
reflects his input. None of these people should shoulder the blame 
for any errors in the paper. 

1 Other versions of phonological feature representation have also 
been proposed. Hayes (1986b) for instance accepts the general 
geometric view proposed by Clements but differs from his theory in 
terms of tier composition and representation details. 

2 The voicing quality of the obstruent does not appear to be 
crucial in the formulation of the rule since there is no evidence that 
the obstruent must be initially voiceless. In fact, /r/ in /reha/ would 
normally be considered voiced and would still undergo the rule, 
albeit vacuously. On these grounds, then, the voicing value of the 
obstruent can be left unspecified. (This is symbolized as {.} in the 
diagram). In subsequent discussion, the symbol {.} will be used to 
represent features which need not be specified or stated in a rule 
because they are not crucial. 

3 These phonological processes are commonly attested in many 
Bantu languages, suggesting a common source, probably in Proto- 
Bantu or one of its branches. 

^ I also assume the following conditions: (i) The two segments 
are tautosyllabic, (ii) the change occurs in the perfective tense (cf. 
Mtenje In Preparation for discussion). 

5 There is a general rule in Chiyao (and in Bantu in general) of 
nasal assimilation which assimilates nasals to their following con- 
sonants in place of articulation. 

6 The rule is general and is not restricted to any particular tense. 
Note that even forms from the progressive tense undergo the rule as 
shown below: 



144 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

n - ku - kata -» Tigukata "I am cutting' 
I-prog-cut 

n - ku - timba -» 'ngutiniba 'I am beating' 
I-prog-beat 

7 There is an independent rule which changes stops into 

fricatives in the context {+nasal} {i} in the perfective tense. This 

rule is responsible for the change of /g/ to /s/ in this form (cf. 
Mtenje In Preparation for details). 

8 An alternative form used in other dialects is {mbujile}. This is 
obtained through the application of a general rule of palatalization 
which changes velar stops into their corresponding palatal affricates 
before front vowels, hence the alternation between /g/ and /J/ in 
this form. For these dialects, palatalization is ordered before the 
/g^s/ rule, while the dialects using {mbusile} use the reverse order. 

9 The same palatalization referred to in footnote (8) above is 
responsible for the {k]~ {c} alternation here. 

10 It may be argued that the feature {syllabic} is phonetically 
not a manner feature but just a class feature referring to the function 
of segments. Here the feature {syllabic} is being considered as a 
major class feature like the features {consonantal}, {sonorant}, etc. 

11 The feature {-vocalic} can also be used in place of {-syllabic} 
to cover the class of obstruents, laterals, and the glide /w/. 

12 AH the voiced stops are implosives. 

'3 One finds forms like mboga "type of food', ndawi 'time', rjguku 
'chicken', ligarjga 'stone', likambale 'type of fish', lindanda 'egg' etc. 

REFERENCES 

Abu-SALIM, I. 1980. Epenthesis and geminate consonants in Palestin- 
ian Arabic. Studies in the Linguistic Sciences 10:2.1-11. 

Chomsky, N., & M. Halle. 1968. The sound pattern of English. New 
York: Harper & Row. 

Clements, G. N. 1985. The geometry of phonological features. Phono- 
logy Yearbook 2.223-250. 

1986. Consonant gemination and compensatory lengthening in 

Luganda. Studies in compensatory lengthening, ed. by E. Sezer & 
L. Wetzel. Dordrecht: Foris. 

1987. Toward a substantive theory of feature specification. To 
appear in NELS 18. University of Toronto. 

Goldsmith, J. 1976. Autosegmental phonology. MIT Ph.D. dissertation 
in Linguistics. 



Mtenje: Autosegmental feature-spreading in phonology 145 

1981. Subsegmentals in Spanish phonology: An autosegmental 

approach. Linguistic symposium on Romance languages, ed. by 
W. Cressey & D. J. Napoli, 9.1-16. Washington: Georgetown Uni- 
versity Press. 

GUERSSEL, M. 1978. A condition on assimilation rules. Linguistic Anal- 
ysis 4.225-254. 

Halle, M., & J. R. Vergnaud. 1980. Three-dimensional phonology. 
Journal of Linguistic Research 1.83-105. 

Hayes, B. 1986a. Inalterability in CV phonology. Language 62.321- 
351. 

1986b. Assimilation as Spreading in Toba Batak. Linguistic In- 
quiry 17:3.467-500 

KensTOWICZ, M. 1982. Gemination and spirantization in Tigrinya. 
Studies in the Linguistic Sciences 12:1.103-122. 

Lieber, R. 1987. An integrated theory of autosegmental processes. 
New York: State University of New York Press. 

Mascaro, J. 1983. Phonological levels and assimilatory processes. 
(Ms, Universitat Autonoma de Barcelona.) 

McCarthy, J. 1986. OCP effects: Gemination and antigemination. Lin- 
guistic Inguiry 17:2.207-263. 

Mester, R.-A. 1986. Studies in tier structure. University of Massa- 
chusetts Ph.D. dissertation in Linguistics. 

MOHAN AN, K. P. 1983. The structure of the melody. (MS, MIT and Uni- 
versity of Singapore.) 

Mtenje, A. D. In Preparation. Autosegmental spreading and Chiyao 
phonological rules. University of Malawi, Zomba. 

Sagey, E. 1986. The representation of features and relations in non- 
linear phonology. MIT Ph.D. dissertation in Linguistics. 

SCHEIN, B. 1981. Spirintization in Tigrinya. MIT Working Papers in 
Linguistics 3.32-42. 

SCHEIN, B., & D. Steriade. 1986. On geminates. Linguistic Inquiry 17: 
4.691-744. 

Steriade, D. 1982. Greek prosodies and the nature of syllabification. 
MIT Ph.D. dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



SCHWA FRONTING IN HINDI* 

Pramod Kumar Pandey 
(South Gujarat University, Surat, India) 

This paper presents an analysis of Schwa Fronting (SF) in 
standard Hindi. I show that SF is an ongoing change in the 
language, especially in its eastern varieties. Moreover, it is 
subject to lexical restrictions. In lexical phonology terms, 
then, it is a process of the lexical component. Two syn- 
chronic analyses are possible, one in terms of an assim- 
ilatory process, the other in terms of underspecification. 
After discussing the relative advantages — and problems — 
of the two approaches, I examine the question of motivation 
of SF. I argue that a motivation can be found in a dynamic 
approach to linguistic description which conceives of evolu- 
tion as inherent to linguistic creativity. 

0. Introduction 

In this paper, I present an account of Schwa Fronting in stand- 
ard Hindi with a two-fold purpose: to describe an ongoing change in 
standard Hindi, and to show the relevance of this change to the issue 
of naturalness in phonological theory. With regard to the latter, I ar- 
gue that innovative processes such as Schwa Fronting in Hindi, which 
do not show a universal regularity of alternation, are difficult to ex- 
plain within the prevailing static approach of the universalist theory 
of generative phonology. The latter defines 'naturalness' of phono- 
logical rules on the basis of finding a correlation between alternating 
segments and their distribution. I shall try to show that within a 
dynamic approach to linguistic description, which conceives of evo- 
lution as inherent to linguistic creativity, the 'naturalness' of such 
processes can be better appreciated. 

1. Analysis 

1.1. The data for the present description were elicited from speak- 
ers coming from different speech areas of standard Hindi. In all, fif- 
teen speakers were interviewed. Six of them represented the west- 
em variety, as spoken in Delhi, Mathura, and Etah. Another group of 
six represented the eastern variety, as spoken in Varanasi, Jaunpur, 
and Rewa. The remaining three represented the variety spoken in 
the central region, Kanpur.i 



148 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

The first six speakers, representing the two varieties, eastern 
and western (three each), were asked to read in normal tempo a 
minimal list of 91 words containing 39 words having /a/ in different 
environments. The other nine speakers (three for each region) were 
given an additional list of 29 words to test the hypotheses formed 
following the elicitations from the first six speakers. Instances of 
Fronting were recorded by two persons, including myself. In case of 
differences between the two recordings (which were rather limited), 
the speakers were asked to pronounce the forms again, until an 
agreement between the two records was reached. 

1.2. Schwa in standard western Hindi is fronted before /h/, which 
may be followed by a consonant or a schwa, as shown in the ex- 
amples below. 2 

(1) a. [k8<hna:] 'say' (Inf.) b. [S8<h8r] 'city' 

[pa<hna:] 'wear' (Past) [Th8<h9r] 'stop' (Imp.) 

[g3<hna:] 'jewelry' [na<h8r] 'canal' 

[m8<hka:] 'smell+past' (Intr.) [b8<h8n] 'sister' 

[p8<hla:] 'first' [p8<h8r] (part of day) 

[d8<hla:] 'No. 10 card' [8<har]ka:r] 'pride' 

[l3<hga:] 'skirt' (N) [[c8<h8l][pa<h8l]] 'cheerful 

[pra<hla:d] (a name) movement' 

[8sa<hy] 'intolerable' 

[[ga<hma:] [ga<hmi:]] 'commotion' 

c. [ka<h] 'say' (Imp.) 

[sa<h] 'bear' (Imp.) 

[suba<h] 'morning' 

[t8r8<h] 'like' (Adj.) 

Exceptions to the above generalization are to be found in numerals, 
cf. e.g. the examples in (2). 

(2) a. [bahattar] 'seventy-two' b.3 [gyairah] 'eleven* 

[satahattar] 'seventy-seven' [ba:rah] 'twelve' 

-[satattar] [te:rah] 'thirteen' 

[sThahattar] 'seventy-eight' [c3:d9h] 'fourteen' 

~ [aThattar] etc. 

Schwa systematically does not front before a consonant other than 
/h/, or if /h/ is followed by a vowel other than schwa: 



(3) a. 




b. 




(i) [ksmra:] 


'room' 


[raho:] 


'stay' (Imp.) 


[laRka:] 


'boy' 


[mahila:] 


'lady' 


[patii:] 


'thin' (Fem.) 


[baha:dur] 


'brave' 


[gamla:] 


'flower pot' 


[pahuca:] 


'reached' 


[n8mki:n] 


'salty' 


[sahuiliyat] 


'facility' 


[ham] 


'we' 


[S8he:li:] 


'female friend 


[hal] 


'plough' 







Pandey: Schwa fronting in Hindi 149 

(ii) [brsmmha:]'* 'Lord Brahma' 

[brammhaiND] 'universe' 

[ellhaR] 'innocent' 

[nannha:] 'small' (Masc.) 

[k8T(9)hal] 'tapioca' 

[arhar] (a type of pulse) 

1.3. We must rule out possible but untenable explanations for the 
process. Note that the change is not affected by the prosodic struc- 
ture of words, as the process of Hindi Schwa Deletion is (cf. Pandey 
To Appear). Although the fronted /a/ in (1) is stressed ((a) and (c)) 
or stressable (i.e., tending to be stressed), as in (b), there are clear 
cases of unstressed [a<], as for example in (4). 

(4) [pa<hca:n] 'acquantance, identity' 
[ma<hattw] 'importance' 
[ma^hatta:] 'importance' 
[Ta<halna:] 'to stroll' 

Moreover, even when /a/ is stressed, it is not fronted before /h/ if 
the latter is followed by a vowel other than /a/; cf. e.g. /mahila:/ 
'lady' in (3b) above. 

1.4. Clearly, then, one of the factors influencing /a /-fronting is its 
melodic sequential occurrence. The rule of /a /-fronting can be in- 
formally stated as follows. 

(5) Schwa Fronting (informal version): 

/a/ is fronted before /h/ if the latter is not followed 
by a vowel other than /a/. 

(5) may be formally stated as (6), ignoring segmental feature speci- 
fications. 

(6) Schwa Fronting 

/a/ -* [-back] / _ /h/ 

Condition: /h/ is not followed by a vowel other than /a/. 

It is not immediately apparent as to why Schwa Fronting (SF) 
should include the condition regarding the following vowel being 
other than /a/. But I shall not go further into a formal investigation 
of the process at this stage. I shall return to it in the next section to 
show that phonological theory must treat processes such as SF as ex- 
pectedly idiosyncratic, rather than 'unnatural', on account of their 
relatedness with the diachronic dimension. 

2. Descriptive relevance 

2.1. Spread in eastern standard Hindi 

The fronting of schwa is a characteristic feature of western 
Standard Hindi. This fact has not found mention in the literature on 



150 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Hindi phonology. Kelkar (1968) notes it as characterizing the 'micro- 
lects' Urdu and Hindi-Urdu, in contrast to Hindi which has non-front 
/9/. Kelkar specifies the environinent for SF in the former microlects 
as a following /h/ which is followed by any vowel (i.e., not only by 
/9 /). While Urdu certainly has a predominance of fronted schwa, 
Hindi-Urdu and Hindi are found, in the present investigation, to have 
dialectal differences within them, which are being gradually lost. 

Two main varieties of Standard Hindi(-Urdu) are recognized — 
eastern and western. 5 I find Schwa Fronting to be a hall-mark of the 
western standard. ^ Speakers of the eastern variety show a range of 
manifestations, depending upon many factors. 

Within the Eastern Standard, two main types of speakers are 
found. One are diglossic speakers, who use the standard variety in 
formal situations, while for informal interaction they use a regional 
variety. The other type speak only the standard variety. Speakers of 
both types show a range from total absence of Fronting to its sys- 
tematic presence. The latter type of speakers tend to have Fronting 
more frequently in their speech. The productivity of the rule seems 
to involve many factors, including age, mobilility, exposure to the 
western standard, etc.'' which need to be more closely investigated. 

In short, SF in eastern Standard Hindi is a case of sound change 
in progress. 

In the present investigation, with its limited scope, two of the 
six eastern Hindi speakers were found to show no trace of a fronted 
schwa in a total of 39 forms, and one has it in only four forms: 
[t8<hkha:na:], [t9<hki:ka:t], [ma<hfil], [d9S8<hra:]. The other three have 
it in 14, 23, and 34 forms, respectively. The results are presented in 
Table 1 below. 





Nil 


1-10 


11-20 


21-30 


31-39 


No. of speakers: 


2 


1 


1 


1 


1 


No. of FS forms: 





4 


14 


23 


34 


Perc. of FS forms: 


0% 


10% 


36% 


60% 


87% 



Table 1: Fronted schwa in eastern Standard Hindi: 
Sound change in progress. 

As compared to eastern Hindi, the western Hindi data show a 
high percentage of the stable occurrence of fronted schwa. Five of 
the six western Hindi speakers were found to have common fronted 
schwas in 35 forms (90%) — the individual differences among them 
involve only 4 of the forms (10%). I guess the differences in non- 
occurrence of fronted schwas in these instances are attributable to 
the fact that the forms are in infrequent use in colloquial speech, e.g. 
[pr9<h9r] (part of the day). The three speakers from the central re- 
gion (Kanpur) testify to the presence of fronted schwas in this area; 



Pandey: Schwa fix)nting in Hindi 151 

but the number of common fronted schwas in their speech is smaller 
— 27 (69%), with individual scores of 28, 30, and 35. 

The preceding brief discussion of SF spread in eastern Standard 
Hindi makes it clear that we are dealing with a change in progress 
and that the change is taking place slowly and irregularly, by lexical 
diffusion. 

This fact is consonant with the predictions of lexical phonology 
(Kiparsky 1988): The theory of lexical phonology (Kiparsky 1982, 
1985; Mohanan 1986) distinguishes between two modular applica- 
tions of rules — lexical and postlexical. Broadly speaking, lexical rules 
apply within words, can be cyclic, have exceptions, and involve lex- 
ically distinctive features. Postlexical rules have the opposite charac- 
teristics: They apply across the board, are non-cyclic and exception- 
less, and can introduce novel features. This modular distinction has 
been found to have interesting implications for resolving the neo- 
grammarian controversy (for which see Labov 1981, Hock 1986). It 
is now well-known that sound change can take place both by lexical 
diffusion (cf. e.g. Wang 1977) and exceptionless change, as claimed 
by the neogrammarians and attested by recent sociolinguistic 
investigations of sound change in progress (cf. e.g. Labov 1981). 
According to Kiparsky (1988), the difference in the two mechanisms 
of sound change spread is predicted by the existence of two types of 
rule applications. Since lexical rules involve distinctive features and 
can have exceptions, it follows that 'lexical diffusion must be a redis- 
tribution of phonemes among lexical items and cannot create any 
new phonological contrasts.' Postlexical rules which do not admit of 
exceptions are expected to undergo exceptionless or 'neogrammarian' 
change. 8 

Now, returning to Schwa Fronting in Hindi, the evidence from 
the exceptions in (3) suggests that it must have lexical application. 
Moreover, it involves lexically distinctive features. The lexical dif- 
fusion of the process in eastern Hindi thus is consonant with Kipar- 
sky's claims. The underspecification account of Schwa Fronting in the 
following section provides another reason, as we shall see, for treat- 
ing it as a lexical rule. 

2.2. Rule formalism 

2.2.1. Schwa Fronting, as stated in (6) (and (5)), must contain a stip- 
ulated condition. Even otherwise, there is no natural correlation ap- 
parent between the change and the environment in the rule. For an 
explanation of the rule, therefore, it must be shown either that there 
is a correlation between the fronting of schwa and /h/, or (keeping 
the conditioning in mind) that schwa has an exceptional property, not 
shared by the other vowels in the language. The former is the 
standard strategy in phonological analyses of explaining a rule as 



152 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

plausible and natural on the basis of relating the alternating seg- 
ments with their distribution. The latter approach is currently fol- 
lowed in the underspecification theory (cf. e.g. Archangeli 1984, 
Pulleyblank 1988), which explains the exceptional behavior of a seg- 
ment to rules of alternation in terms of its being underspecified for a 
feature in the underlying representation. 

Let us briefly turn to the first alternative of finding a distribu- 
tion-driven explanation for (6). Note that Chomsky & Halle (1968) 
assign the feature [+low] to [h] and to pharyngeal consonants. Indian 
phoneticians (e.g. Allen 1951) consider [h] and [a] to be homorganic. 
If, then, Hindi /9/ can be treated as a low vowel, we have a way of 
relating /a/ to /h/ by revising (6) as (7): 

<^' [loTg]^l-bac«/- [,L] 

Condition: /h/ is not followed by a vowel other than f -i-low 1 

L -long J 

The formulation in (7) is the only form by which SF can be shown to 
have a natural relation between the alternating segments and the 
environment. (7), however, still does not provide much by way of an 
explanation of the condition by which it is restricted. Without that 
condition, we might perhaps be able to consider SF to be an 
assimilatory process, which assimilates the vowel /a/ to the [-back] 
feature of the homorganic following /h/ — provided that /h/ can 
indeed be classified as [- back].9 (See also 2.2.2 below.) But the 
condition on (7) does not provide any support for an assimilatory 
account. 

An explanation for the exceptional behavior of /a/ in Hindi is 
possible within the underspecification approach: /a/ can be shown to 
be the only underlying Hindi vowel whose features are totally un- 
specified, whereas all the other vowels are specified for at least one 
feature. The distinctive feature specifications for Hindi vowels are 
given in Table 2. Table 3 provides a tentative underspecified repre- 
sentation of the vowels underlyingly. 



[high] ++------++ 

[low] ---+ + + -+-- 

[back] ----+ + + + + + 

[long] - + -I- -I- - + + -I- - + 

Table 2: Distinctive features for Hindi vowels 



(8) 


a. 
b. 
c. 
d. 


[+low] ■ 




e. 


[+low] 




f. 




Notice 


that Table 



Pandey: Schwa fronting in Hindi 153 

i i: e: £: a a: o: o: u u: 

[high] + + + + 

[low] - - - .... 
[back] .... 

[long] + + + + + + + 

Table 3: Underspecified representation of Hindi vowels 

Following Archangeli (1984), I assume that the features for the 
underspecified segments are filled by Readjustment rules of the fol- 
lowing type: 

[-high] 

[+low] 

[-high] 

[+back] 

[+back] 

[-long] 

shows only /a/ to be fully unspecified. All 
other vowels are underspecified for at least one feature. There is 
some independent evidence to support the above observation 
regarding /a/: Only unstressed /a/ undergoes deletion, subject to 
certain conditions (cf. Ohala 1983, Pandey To Appear). Secondly, 
vowel epenthesis between consonant clusters is a wide-spread pro- 
cess in the regional dialects of Hindi. The epenthesized vowel norm- 
ally is [a], as in (9) below. lO 

(9) [pradi:p] ^ [pardi:p] (a name) 
[kle:s]^ -» [kale:s] 'sorrow' 

[mle:ks] -♦ [maleiks] 'a foreigner' (obsolete) 

[jayadrath] ^ [jay(a)darath] (a name) 

Considering the special underspecified property of /a/ in Hindi, the 
condition in (5) appears quite reasonable. (5) may thus be alterna- 
tively rewritten as (10): 

(10) Schwa Fronting (informal version 2): 

/a/ is fronted before /h/, if the latter is not followed by a 
vowel specified for some feature. 

The formal version of (10) is as follows: 

(11) Schwa Fronting (formal version 2): 
/a/ -* [-back] / _ /h/ 

Condition: /h/ is not followed by a vowel specified for 
some feature. 

In fact, given the nature of underspecification theory, it is poss- 
ible to reformulate the process even more insightfully, as in the in- 
formal version of (12). 



154 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

(12) Schwa Fronting (informal version 3): 

/a/ is fronted before /h/, which may optionally be followed 
by a consonant on the SEGMENTAL tier. 

Since schwa is completely unspecified on the segmental tier, it is in- 
visible with respect to SF, and a consonant following it does not block 
fronting. A more formal version of (12), then, is as follows: 

(13) Schwa Fronting (formal version 3): 
/a/ - [-back] / _ /h/ (C) 

The environments in which SF applies are the following:^! 

(14) a. : b. : : c. : : 

Root: • • • • • 

I II II 

Skeleton: XX XXX XXXX 

ah ahC ahaC 

Note that rules (11)/(13) and (7) contradict each other. Whereas 
the assimilation rule (7) has /a/ specified for the feature Low, (11) 
and (13) must assume it to be underlyingly unspecified. The ques- 
tion that the alternative formulations of Schwa Fronting raise is, 
which of them is preferable? I am not aware of any grammar-in- 
ternal considerations that would lead to a choice between the al- 
ternative formulations. The underspecification account has the ad- 
vantage of elegance; but to carry conviction, it would require a more 
thorough analysis of the phonological structure of the language. The 
account in (7) is attractive in that it may provide a motivation for 
fronting — if an analysis of /h/ as [- back] can be justified (cf. also 
below). If the problems with either account can be solved, it might 
even be possible to combine the two accounts. 

2.2.2. A dynamic approach to explanation 

It is my contention that within a dynamic approach to linguistic 
analysis, a rule such as SF is expected to have idiosyncratic pro- 
perties as regards comparative phonological considerations. From the 
dynamic point of view, we must focus not on the alternating 
segment, i.e. /a/, but on the TRIGGER that induces the alternation, i.e. 
the following /h/, for reasons that will be apparent in a moment. 

/h/ is known to play a significant role in inducing phonological 
change in Indian languages (cf. e.g. Chatterji 1960, Vajpeyi 1981). 
Note that it is the only sound which is stated in two 'pratya haras' by 
Panini — in one of them, the last, it occurs alone. 12 The loss of medial 
[h] and of 'aspiration' in voiced sounds in Punjabi has given rise to 
substantial change in Punjabi phonology (cf. Chatterji 1960:113-14 
and Hock 1986 with references). In East Bengali, /h/ has become a 
glottal stop, among other related changes (Chatterji 1960:112-13). 



Pandey : Schwa fronting in Hindi 155 

Gujarati has acquired murmured vowels under the influence of me- 
dial [h] (which is lost) and of voiced aspirates (cf. Pandit 1957). Be- 
sides, [h] has led to other types of alternations, accompanied by its 
loss, in other languages. Thus in Chaddho, an American Indian lan- 
guage described in Chafe 1968, [h] is lost if followed by a sequence of 
two consonants, and a preceding vowel gets stressed. Similarly, in 
the Gitskan dialect of Tsimshian (Anderson 1974:175-178), [h] is 
deleted intervocalically, leading to the lengthening of short vowels. 

The cloudiness concerning the motivation and condition for (7) 
might clear up if we look at (7) as a prelude to a possible sound 
change in the language. Is the language preparing itself for a loss of 
/h/? If so, then /h/ is about to be lost after /a/ word-finally, before 
consonant, and before /a/, but not before other vowels. Fronting of a 
preceding schwa may then be a way of leaving a trace of its loss in 
the language, just like murmur in Gujarati, tone in Punjabi, stress in 
Chaddho, and vowel lengthening in Gitskan Tsimshian. 

Considering the comparative evidence of other languages, this is 
not an implausible hypothesis. In this regard, it is interesting to note 
that the loss of /h/ after a fronted schwa, or rather after [e] < /a/, is 
a common feature in some varieties of western Hindi, and of 
Rajasthani, giving rise to forms such as [p£ :1a:] for /pahla:/ 'first', 
[k£:na:] for /kahna:/ 'to say', etc. 

What is more, the loss of /h/ is already taking place in standard 
Hindi in stray cases, as in [che:], [che:] < /chah/ 'six'. And as observed 
in note 1, the optional loss of word-final /h/ in the numeral forms of 
(2b) leads to the lowering of /a/ to [a]. 

The loss of /h/, combined with Schwa Fronting, can be expected 
to lead to a fully front vowel /e :/ or /e:/. Notice that this is a tense 
and long vowel. The loss of /h/ thus may be a case of compensatory 
lengthening. 

3. Conclusion 

The alternation between [a] and its fronted variant [a<] or [&:]/ 
[e:], which appears to be governed by unnatural conditions in a static 
conception of linguistic structure, is accepted as being naturally 
indeterminate within a dynamic conception of linguistic structure, in 
which a rule integrates not only with the rest of the synchronic 
structure (cf. Chomsky 1964:22), but also with its evolving structure 
(cf. Humboldt 1933 (also Robins 1987), Harris 1966:18-19, Vachek 
1967, Stockwell & Macaulay (eds) 1973). 13 In tying up Schwa Front- 
ing with a possible future loss of its trigger /h/, on the basis of its 
indeterminate role in inducing alternations and change in various 
languages, and of some evidence of its loss in the language and its 
dialects, we observe a complex dynamic picture of a change already 



156 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

effected in western Hindi, taking place in the eastern Standard, and 
leading to a further change in the language. 

A conclusion that seems to emerge from this investigation is that 
we need to distinguish between segments that alternate, and seg- 
ments that induce alternation. The latter, such as /h/, may lead to 
indeterminate alternations, which are not subject to 'naturalness' ex- 
planations based on distributional facts. The only plausible explan- 
ation of their naturalness may lie in a dynamic conception of the 
organization of linguistic structure. 

A dynamic model of linguistic structures is expected to dis- 
tinguish, for the Evaluation Metric, phonological processes that in- 
tegrate with the synchronic structure of a language from those that 
integrate with its evolution. Let us call them Type I and Type II 
processes, as the terms 'synchronic' and 'diachronic' do not ade- 
quately distinguish them. Note that Type II processes may be pro- 
ductive and synchronic, such as SF in Hindi, but still be involved in 
phonological change. They are either a result of a phonological 
change, or an instrument to it, and, unlike Type I processes, add to 
the cost of the grammar. Type II processes are thus not subject to 
the Simplicity Metric. As some of them are not easily distinguishable 
from Type I processes on grounds of productivity and distribution, 
they constitute an important area for further investigation within the 
research program of Universal Grammar (cf. Chomsky 1986). The 
Simplicity Metric, which takes into account only those phenomena 
that belong to the Core and excludes others as being a part of the 
Periphery, will find it diffficult to provide a principled account of 
some of the Type II processes, unless the latter are properly inves- 
tigated and defined. The investigation of such processes is expected 
to show the need to internalize the temporal dimension in linguistic 
theory, to whatever degree of idealization. 

NOTES 

*I wish to thank D. M. Joshi for comments on the paper. 

1 The reason why the same number of speakers were not chosen 
for the central region as for the western and eastern regions is that 
the focus of the present investigation is the eastern variety, where 
the process of change is beginning to take shape. Speakers from the 
western and eastern regions were consulted for comparative data in 
the first phase. Speakers from the central region were included in 
the second phase of data elicitation for the purpose of attesting the 
wide-spread presence of the process in Standard Hindi. 

2 Here as elsewhere, [b<] indicates a fronted [a]. 



Pandey: Schwa fronting in Hindi 157 

3 In Colloquial Standard these forms have the variants [gya:ra:], 
[ba:ra:], [terra:], [cD:da:]. In all of these, the final /h/ is deleted, and 
the schwa is lowered and lengthened. 

'^ The [CCh] forms of these examples are underlyingly /Ch/, cf. 
Kelkar 1968. 

5 For a little more on this, cf. Pandey 1989:78. 

6 A bilingual speaker of Kashmiri and Hindi, Vijay Koul, has 
interesting data on this. In his speech, forms which belong to the 
common vocabulary of Kashmiri and Hindi have different pronun- 
ciations of schwa in the environments affecting Fronting. Schwa 
Fronting applies in his Hindi, but not in his Kashmiri. Some examples 
of his Hindi and Kashmiri pronunciations of the related forms are: 



Hindi 


GLOSS 


Kashmiri 


[k9<h9r] 


calamity 


[kahar] 


[la<h9r] 


wave 


[lahar] 


[t9<hki:ka:t] 


inquiry 


[t9hki:ka:t] 


[m9<hfil] 


a party 


[m9hfil] 



"7 Notice that education is invariably associated with Standard 
Hindi in the eastern region. The uneducated or not so well-educated 
normally speak the local dialect, or a non-standard form of KhaRi 
Boli. 

8 For a revision of Kiparsky's claim, cf. Pandey 1990. 

9 The question as to whether /h/ can be classified as [- back] is 
an interesting one, especially for the theory of feature geometry and 
the distinction between C and V features. It certainly deserves fuller 
investigation. For the time being, see Allen 1951 on diachronic evid- 
ence from Indo-Aryan that might provide further support for the 
view that /h/ may have properties affiliating it with front vowels. 
This evidence involves the change of final -as to -e (as in *daiwas > 
Pali deve 'god'), presumably via -ah > -ai. Similarly Latin final -s has 
changed into -/ in Italian (as in post > poi ■after(wards), then'), 
presumably via *-h (Hock 1986 with references). On the other hand, 
affiliation of /h/ with [+ back] is suggested by the normal Sanskrit 
development of -as to -o [-5] in voiced contexts (as in *daiwas > devo 
'god'), presumably via *-ah > *au; cf. again Allen 1951 with referen- 
ces to similar developments in Iranian. 

10 Non-standard eastern Hindi has also an epenthetic [i] preced- 
ing word-initial /sC/ clusters, e.g. 

Non-standard Forms Standard Forms Gloss 
[iste:s9n] [ste:san] station 

[isku:l] [sku:l] school 

[isth9l] [sth9l] place 



158 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

However, /-epenthesis before word-initial /sC/ clusters is a natural 
process, in the sense of Stampe 1973, and perhaps not peculiar to 
Hindi. Notice that in case of vowel epenthesis into initial /sC/ clust- 
ers, in certain dialects of Hindi, especially western ones, the epen- 
thetic vowel is [9], e.g. [s9te:S9n] 'station'. 

1 1 This refinement of the underspecification analysis has been 
suggested by Jennifer Cole. 

12 To the best of my knowledge, there is no generally agreed 
upon explanation for the occurrence of /h/ in the two pratyaharas in 
Panini. 

13 Post-Chomskian linguistics has seen an upsurge of dynamic 
approaches to linguistic analysis. Chief among these are the works of 
sociolinguists (e.g. Labov 1972), developmentalists (e.g. Bailey & Har- 
ris 1985), and those following the general systems approach (e.g. 
Mohanan 1989). 

REFERENCES 

Allen, W.S. 1951. Phonetics in ancient India. London: Oxford Uni- 
versity Press. 

Anderson, Stephen R. 1974. The organization of phonology. New 
York: Academic Press. 

Archangeli, Diana. 1984. Underspecification in Yawelmani phono- 
logy and morphology. MIT Ph.D. dissertation in Linguistics. 

Bailey, Charles-James N., & Roy Harris. 1985. Developmental mech- 
anisms of language. Oxford: Pergamon Press. 

Chafe, Wallace L. 1968. The ordering of phonogical rules. Inter- 
national Journal of American Linguistics 34.115-36. 

Chatterjee, Suniti K. 1960. Indo-Aryan and Hindi. Calcutta: Firma 
Mukhopadhyaya. 

Chomsky, Noam. 1964. Current issues in linguistic theory. The Hague: 
Mouton. 

, & Morris Halle. 1968. The sound pattern of English. New York: 

Harper and Row. 

Harris, Zellig. 1966. An introduction to mathematical linguistics. 
Cambridge, MA: MIT Press. 

HOCK, Hans Henrich. 1986. Principles of historical linguistics. Berlin: 
Mouton de Gruyter. 

Humboldt, Wilhelm von. 1903-30. Gesammelte Schriften, ed. by A. 
Leitzman et al. Berlin: Behr. 

Kelkar, Ashok R. 1968. Studies in Hindi-Urdu, I: Word phonology. 
Poona: Deccan College. 



Pandey: Schwa fronting in Hindi 159 

KIPARSKY, Paul. 1982. From cyclic phonology to lexical phonology. 
The structure of phonological representations, ed. by H. van der 
Hulst & N. Smith, 1.131-75. Amsterdam: Foris. 

. 1985. Some consequences of lexical phonology. Phonology Year- 
book 2.55-138. 

1955. Phonological change. Linguistics: The Cambridge survey. 

Vol. I: Linguistic theory: Foundations, ed. by F. W. Newmeyer, 
363-415. Cambridge: Cambridge University Press. 

Labov, William. 1972. Sociolinguistic patterns. Philadelphia: Uni- 
versity of Pennsylvania Press. 

1981. Resolving the neogrammarian controversy. Language 57. 

267-308. 

MOHANAN, K. P. 1986. Lexical phonology. Amsterdam: Reidel. 

1989. Universal attractors in phonology. (MS, Stanford Uni- 
versity.) 

OHALA, Manjari. 1983. Aspects of Hindi phonology. Delhi: Motilal 
Banarsidass. 

PANDEY, Pramod. K. 1989. Word accentuation in Hindi. Lingua 77. 
37-73. 

1990. Optionality, lexicality, and sound change. (Unpublished 

MS, South Gujarat University, Surat.) 

. To Appear. Hindi schwa deletion. Lingua 82. 

PANDIT, Prabodh B. 1957. Nasalization, aspiration, and murmur in 
Gujarati. Indian Linguistics 17.165-72. 

PULLEYBLANK, Douglas. 1988. Vocalic underspecification in Yoruba. 
Linguistic Inquiry 19.233-70. 

STAMPE, David. 1973. On chapter nine. Issues in phonological theory, 
ed. by M. Kenstowicz & C. W. Kisseberth, 44-52. The Hague: 
Mouton. 

Stockwell, Robert P., & Ronald K. S. Macaulay (eds). 1973. Linguistic 
change and generative theory. Bloomington: Indiana University 
Press. 

Vachek, Josef. 1967. The non-static aspect of the synchronically 
studied phonological system. Phonologic der Gegenwart = Wie- 
ner Slavistisches Jahrbuch VI.79-87. Graz-Wien-Koln: Bohlau. 

Vajpeyi, Kishori Das. 1981 [1949]. Hindi Nirukta. Delhi: Vani Pra- 
kashan. 

Wang, William S.-Y. (ed.) 1977. The lexicon in phonological change. 
The Hague: Mouton. 



i 



i 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



REVIEW 

R. N. Aralikatti: Spoken Sanskrit in India: A study of sen- 
tence patterns. (Kendriya Sanskrit Vidyapeetha, Tirupati 
Series No. 53.) Tirupati (India): Kendriya Sanskrit Vidya- 
peetha, 1989. Pp. (xii +) xxiv, 278, + appendix of transcripts 
(pp. 1-172), indices, bibliography, and errata (i-xxviii). 

Hans Henrich Hock 

While there is a vast and ever-increasing literature on Vedic and 
Classical Sanskrit, i the use of Sanskrit in modern India has received 
little attention. In fact, the common assumption is that, like Latin, 
Sanskrit now is a dead language. The author of the monograph 
under review is one of the few linguists who has demonstrated that 
this assumption is erroneous: Sanskrit not only survives as a written 
language in which every year thousands of publications are 
produced, but also as a spoken language; cf. in addition to the book 
under review, Aralikatti 1976, 1980, 1981, 1982, 1991. In this 
respect, his work converges with my own (cf. Hock 1981, 1983, 
1988); but my publications paint a bleaker picture of spoken 
Sanskrit, viz. as a dying language. 

In addition, however, to establishing that Sanskrit is still alive, 
Aralikatti (A) has made it his life's goal to describe the nature of the 
modern spoken form of Sanskrit and to foster its continuous use, by 
developing materials for the teaching of spoken Sanskrit. 

The present monograph, a slightly revised version of the au- 
thor's 1983 Ph.D. dissertation at Sri Venkateswara University in 
Tirupati, is the most significant step in this direction and represents 
the results of years of research. It is based on transcriptions of 
taped telephone conversations, lectures, and conversations of 
speakers from all areas of India, as well as one foreigner, namely me. 
As the title suggests, the major emphasis of the volume is on 
sentence structure. Two chapters, however, are devoted to the 
question of how to preserve Sanskrit as a spoken language and what 
pedagogical approaches to direct toward that goal. 

A brief introduction (1-13) is followed by Chapter II (14-42) 
which provides a preliminary view of Sanskrit sentence structure. 
Topics covered include a classification of sentences, null subjects, 
verb agreement, interrogative sentences, participial structures, and 
argument structure. Chapters III (43-60) and IV (61-83) cover the 



162 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

structure of subject and object NPs. Chapter V (84-95) is devoted to 
'Complement structures'. Chapters VI (96-103), VII (104-121), and 
VIII (122-165) take up the use of adjectives, adverbs (including sen- 
tential particles), and verbs, respectively. 'Minor and simple sen- 
tences' are dealt with in Chapter IX (166-186), and 'Complex and 
compound sentences' in Chapter X (188-211). Chapter XI (212-232) 
briefly deals with other issues, such as pronunciation and loanwords, 
as well as regional variations in spoken Sanskrit. Chapter XII (236- 
252), 'Guidelines for a programme of spoken Sanskrit', outlines 
pedagogical applications of A's research. The concluding chapter, 
XIII (253-264), advocates and attempts to justify the continued 
cultivation of Sanskrit as a spoken language. An extensive set of 
notes, transcriptions, indices, references, and errata round out the 
volume. 

The book is useful especially for its wealth of data on the syn- 
tactic structure of modern spoken Sanskrit. Its information on pro- 
nunciation and lexicon is understandably less complete. 

In addition, the attentive reader can find in the volume a large 
amount of information on regional differences in modern Sanskrit. 
Some of these are expressly mentioned by A. These are mainly 
peculiarities of northern Indian usage. Interestingly, a number of 
other differences go unmentioned, viz. those characteristic of 
southern usage and found prominently in A's personal speech. (A is 
a native speaker of Kannada, a southern, Dravidian language.) 

Among the peculiarities of northern speech cited by A (196) is 
the common practice of marking cited discourse by a preposed part- 
icle yad, as in (1). (Formally, this is the nominative/accusative 
neuter of the relative pronoun.) A is no doubt correct in attributing 
this usage to the influence of Hindi, where direct discourse similarly 
is introduced by a particle, viz. ki; cf. (1'). This contrasts with the 
normal traditional pattern of using a quotative marker iti, placed 
AFTER the direct discourse; cf. (2). And again, A is correct in noting 
that this traditional pattern is fully preserved in the south; cf. (2'). 
What he fails to mention is that the usage finds strong support in the 
corresponding structures of the southern languages, which regularly 
exhibit a postposed quotative particle; cf. the Tamil example in (2"). 
Regional influence, therefore, may be going in both directions. 

(1) bharatavidyasamupasakah jananty eva 
yat [asyarh vidyayarh sa vidhih vartate ...] 

Direct Discourse 
'Devotees of Indian science know "In this science there 
is the rule ..." ' (Northern speaker; p. 40 of A's monograph) 

(!') bharat vidya ke samupasak jante hairh 

ki [us vidya mem yah vidhi hai ...] (Hindi version of (1)) 
Direct Discourse 



Hock: Review of Aralikatti 163 

(2) kathitam avalokitaya [madanodyanam gato madhava] iti 

Direct Discourse 
'Avalokita said: "Madhava has gone to the garden of the 
God of Love." ' (Malat. 1:11) 

(2') [samyag eva asit] iti te kathayanti 
Direct Discourse 
'They say, "It was good indeed." ' (Southern speaker; p. 4-5) 

(2") [nan varuven] enru avan sonnan 
Direct Discourse Quotative Marker 
'He said "I will come." ' 

That this interpretation is on the right track is suggested by the 
structure of yes/no-questions. Classical Sanskrit used a variety of 
(optional) particles, but the most common one was kim (nominative/ 
accusative singular of the interrogative pronoun). In his discussion 
of yes/no-questions (179-182), A duly takes note of these structures 
in modern Sanskrit, but adds (182), without comment, the structure 
in (3) below, in which a particle vd follows the clause-final verb of 
the question. This usage is without direct precedent in traditional 
Sanskrit, where vd is a coordinating conjunction, normally meaning 
'or'. 2 My observations show this structure to be the exclusive 
property of southern Sanskrit. And in the south, the usage can be 
accounted for as a caique of the normal Dravidian pattern of yes/no 
questions, in which a phonetically quite similar particle, -d, follows 
the clause-final verb; cf. the Tamil example in (3'). Here, then, it is 
the southerners who have innovated, while the northerners preserve 
the older pattern (which, again, is supported by the structure of 
Hindi). 

(3) kevalam ajfiatva purvatanah evarh upasanam akurvan va 

Fin. Verb 
'Did (our) ancestors worship (the moon) ignorantly?' 
(Southern speaker) 

(3') avan pustakattai kotuttan-a 
Fin. Verb 
'Did he give the book?' 

The fact that A does not consider this use of vd worthy of 
further discussion shows the extent to which it has become part and 
parcel of 'southern Sanskrit'. At the same time, however, his 
inability to take note of such southern regionalisms diminishes the 
value of his book to some extent. 

Additional problems arise from the fact that his transcriptions 
are not always accurate. Now, transcribing taped conversations is a 
formidably difficult undertaking, and some minor mistakes are 



164 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

bound to occur. However, the transcription of a conversation involv- 
ing me (pp. 140-147) differs extensively, and in many places pro- 
foundly, from the taped original, a copy of which is in my possession. 
I became alerted to this problem when 1 found A in his transcript 
attributing to me the use of vd as question particle, a usage I was 
sure is not part of my spoken Sanskrit. While I do not have access to 
the taped originals of A's other texts and thus cannot judge the 
accuracy of their transcription, the evidence available to me suggests 
that A's primary data must be used with some caution. Fortunately, 
the overall picture is not affected; the structures that A considers are 
all very common in modern spoken Sanskrit. The only aspect of 
modern Sanskrit that IS affected is the issue of regional, or personal, 
variation and peculiarities. 3 

A further difficulty arises from the fact that the book is marred 
by a large number of misprints, as well as errors in the translitera- 
tion of Sanskrit examples, only a small part of which is corrected in 
the Errata. One especially egregious example consists in the fact that 
in the excerpt cited on p. 214 as an example of 'stress accent' in my 
spoken Sanskrit, the promised accent marks are consistently absent. 
(Most of these errors, of course, are attributable to the printers and 
publishers.) 

In spite of such difficulties, however, A's book presents a valu- 
able contribution to the study of modern spoken Sanskrit, the only 
ancient Indo-European prestige language that has remained in spok- 
en use to the present day. 

NOTES 

^ For syntax, note for instance the bibliography of Hock & Desh- 
pande 1991. 

2 There is, to be sure, a special use of vd as emphatic particle in 
Classical Sanskrit questions, as in mdnusisu katham vd sydd asya 
rupasya sambhavah (Sak. 1) 'How indeed might there be such a 
beauty among human females?' However, this use is quite different 
from that of vd as question marker in 'southern Sanskrit'. 

3 Thus, A's transcriptions in his appendix (114-115) contain a 
few examples of vd in yes/no-questions of native Marathi speakers. 
Rajeshwari Pandharipande (p.c, fall 1990) informs me that such use 
is not found in the Sanskrit of Marathi speakers. Are A's trans- 
criptions here inaccurate? Or are there some Marathi-speaking in- 
dividuals who have picked up the southern pattern of vd in their 
spoken Sanskrit (perhaps because they learned to speak Sanskrit in 
the south)? 



Hock: Review of Aralikatti 165 

REFERENCES 

Aralikatti, R. N. 1975. Simplifying the verbal system in spoken 
Sanskrit: A propos of kta participles. Sarhskrta-Vimars'ah 
3.142-146. 

1980. Sentence patterns in formal speeches in modern spoken 
Sanskrit as recorded on tapes: A note on equational sentences. 
Summaries of papers. Proceedings of the 30th All-India Oriental 
Conference, 125. 

1981. A note on the complement structures in sentences in 
modern spoken Sanskrit as recorded on tapes. Proceedings of 
the Fifth World Sanskrit Conference, ed. by R. N. Dandekar & P. 
D. Navathe, 16-26. New Delhi: Rashtriya Sanskrit Sansthan. 

1982. A note on object structures in modern spoken Sanskrit. 
Proceedings of the 31st All-India Oriental Conference, 358-60. 

1991. A note on word order in modern spoken Sanskrit and 
some positive constraints. Studies in Sanskrit syntax, ed. by H. 
H. Hock, 13-17. Delhi: Motilal Banarsidass. 
Hock, Hans Henrich. 1981. Sambhasitasarhskrtasyadhuniki sthitih. 
(The present-day status of spoken Sanskrit.') Bhasvati, 
Research Journal of the Sanskrit Department, Kashi Vidyapeeth, 
Varanasi, 6.61-4. 

1983. Language-death phenomena in Sanskrit: Grammatical 
evidence for attrition in contemporary spoken Sanskrit. Studies 
in the Linguistic Sciences 13:2.21-35. 

1988. Spoken Sanskrit in Uttar Pradesh: A sociolinguistic 
profile. Lokaprajfia: Journal of Orientology 2.1-24. (Puri, India). 

, & Madhav Deshpande. 1991. A bibliography of writings on 

Sanskrit syntax. Studies in Sanskrit syntax, ed. by H. H. Hock, 
219-241. Delhi: Motilal Banarsidass. 



i 



i 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



REVIEW 

Alternative conceptions of phrase structure, edited by Mark 

R. Baltin and Anthony S. Kroch. Chicago & London: The 

University of Chicago Press, 1989; pp. xi, 315. $19.95 pa- 
per/ $60.00 cloth. 

James H. Yoon 

0. The volume 'Alternative conceptions of phrase structure" 
(ACPS hereafter) is a collection of papers from a conference held 
under the same name during the 1986 Linguistic Institute at New 
York University. Some of the papers presented at the conference are 
not included since they have been published elsewhere (e.g. 
Grimshaw 1988, Kuroda 1988). In addition, the volume contains one 
paper (Steedman's) that was not presented at the conference. In all, 
there are twelve papers with an introduction by the editors 
containing a helpful summary of current issues concerning phrase 
structure, as well as the contents of individual contributions. 

1. A fundamental fact about natural language syntax is that it 
is structured. This raises the following questions about syntactic 
structure: What motivates structure? What is structure like? What 
is the best way to represent structure? 

The motivation for structure comes both from our intuition and 
linguistic tradition, and is felt most palpably in the existence of 
structural ambiguities. The answer to the second question provided 
by Chomsky (1955/75, 1957) is that structure involves the sub- 
grouping of linguistic elements (TERMINAL elements) into 
'constituents' (the DOMINANCE relation) and 'syntactic categories' (the 
LABELING relation), with 'order' (the PRECEDENCE relation) defined 
among constituents. The formal means Chomsky employed to 
represent structure was context-free rewriting rules. Adopting 
rewrite rules commits one to a view of syntactic structure as 
determinate, continuous, single-rooted graphs where dominance and 
precedence relations are complementary (Wall 1972). Thus, there 
are no 'virtual' categories, multiply dominated, discontinuous, or 
dismembered constituents. In addition, from early on, Chomsky was 
committed to the belief that functional information (such as 'subject- 
of, 'object-of, etc.) could be reduced to structural information 
available in trees. 



168 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

However, it was recognized immediately that these restrictions 
on an unadorned Phrase Structure Grammar (PSG) made it a clumsy 
tool for representing the full range of dependencies found in natural 
language syntax. Chomsky's answer to this problem was not to give 
up PSG as the tool of syntactic representation but to think of syn- 
tactic representations instead as consisting of sets of trees given by a 
PSG which are related via a different rule type — transformations. 

Subsequent developments in generative grammar largely 
assumed the correctness of Chomsky's answer. In fact, the propon- 
ents of an influential modern syntactic theory, Government & Bind- 
ing (GB) theory, still operate on the same assumption. Thus, the GB 
papers in the volume do not really deal with ALTERNATIVE concep- 
tions of phrase structure, if 'alternative' is taken to imply 'other than 
PSG'. However, one must grant that the potential benefit of not 
probing into the mathematical foundation of PSG is that the 
innovations in GB are driven more by a desire to describe new and 
emerging empirical generalizations, elevating them, where possible, 
into explanations. 

In contrast, several of the non-GB papers seek to provide proof 
that a new formalism is adequate and capable of describing the stan- 
dard constructions in every syntactician's stock. This is true espe- 
cially of theories embracing newer formalisms. However, quite often, 
the adoption of newer formalisms leads to interesting and elegant 
solutions to older problems that received cumbersome treatments 
within standard transformational grammar. 

Five of the papers in the collection represent various aspects of 
work in GB theory on phrase structure and will be discussed in 
section 2. Section 3 is devoted to papers written in other frame- 
works. Section 4 concludes the review. 

2. Baltin's paper is an exercise in taking to the limit the gener- 
alization captured by X-bar theory (Chomsky 1970, Jackendoff 1977) 
that natural language structures are endocentric, or LEXICALLY AN- 
CHORED, an idea that is also expressed in the PROJECTION PRINCIPLE of 
Chomsky 1981. B argues that subcategorization (or selection) is al- 
ways for a lexical (zero-level) category and not for a maximal pro- 
jection. Thus, there are no instances where a head that occurs alone 
is exhaustively dominated by a maximal projection. One-bar and 
two-bar (maximal) projections are licensed by X-bar theory only 
when there are complements and specifiers respectively, a proposal 
which is somewhat reminiscent of categorial grammar which also 
does not distinguish lexical and phrasal categories. Thus, B proposes 
to replace the usual X-bar schema in (la) with that in (lb). 



Yoon: Review of Baltin & Kroch 169 

(1) a. X" -^ Spec(Zmax)X' 

X' ^ X ymax 

b. X" -4 Spec X' 
X' ^ X Y 

In addition, B proposes a principle which is similar to the Projection 
Principle that says a maximal projection of X can occur where the 
head X can occur. 

He shows that this proposal solves a number of problems. For 
example, the fact that verbs select an embedded complementizer is 
no longer problematic if Comp is the head of S'. Idioms like make 
headway are no different from verb phrases like make a cake since 
both involve selection of N by V. The only difference is that in the 
former, the verb 'make' selects a specific N, viz 'headway'. The dis- 
junctions in certain principles of grammar which refer to a maximal 
projection or a head (government, ECP) can also be eliminated. 

Marantz's paper comes closest among the GB-oriented papers to 
proposing an alternative (or a supplement) to PS rules as the repre- 
sentational tool of syntactic structure. M's paper deals specifically 
with the problem that clitics pose for the standard X-bar theoretic 
view of phrase structure and draws some interesting conclusions 
about the role of phrase structure constraints in grammar. The prob- 
lem that M addresses is this: The surface structure of clitic construc- 
tions defies adequate characterization in terms of X-bar schema. For 
example, a principled labelling of the French surface constituent du 
below is impossible in orthodox X-bar theory since it is a merger of a 
preposition and an article. 

(2) [pp [-du] [garcon]] 

of-the boy 

The dominant analysis of such constructions is to derive them via a 
cliticization rule from a structure where precedence, dominance, and 
labelling consistent with X-bar schema hold, such as the one in (3). 

(3) [pp de (np le garcon]] 

of the boy 

Basing himself on earlier work (Marantz 1984), M proposes to derive 
clitic constructions DIRECTLY from S-structures in which linear order 
is undefined. In M's work, S (and D) structures are unordered lists of 
constituents and the relations holding between them. Although the 
relations and constituents may be represented in trees, the relations 
are not determined FROM them. A principle called the MAPPING PRIN- 
CIPLE in Marantz 1984, 1988 and which is termed the EXTENDED PRO- 
JECTION Principle in this work, ensures that relations between con- 
stituents at one level must map onto a well-defined set of relations 



170 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

between the corresponding constituents at another level. Thus, while 
Chomsky's Projection Principle requires constancy of lexical prop- 
erties as reflected in tree structure from level to level, M's version 
requires that relations at D-structure be preserved at all other syn- 
tactic levels. 

The trees representing the relations at D and S structure, for 
example, conform to X-bar theory. But M argues that X-bar theory 
has no independent status as a well-formedness condition on syn- 
tactic representations, since X-bar theoretic constraints are complete- 
ly determined by the combinatorial potential of categories as in cate- 
gorial grammar. 

M's analysis of clitics crucially utilizes the possibility inherent in 
his theory of grammar of an S-structure relation being able to be 
mapped onto several distinct surface structure (or PF) relations, to- 
gether with the ASSOCIATIVITY of the S-structure relation of ADJACENCY 
(notated as '*'), as argued for in Marantz 1984 and Sproat 1985. By 
the associativity of adjacency, the effects of 'rebracketing' necessary 
in clitic constructions and other instances of bracketing paradoxes 
follows simply as one of the allowable mappings from S-structure 
relations to surface structure relations. There is no need to invoke 
actual rebracketing of constituent-structure trees. 

(5) a. [X* [YPY*Z]] 
de le garcon 

b. ( [ X * Y ] * Z ] 

de le garcon = du garcon 

The requirement that de be left-adjacent to le is satisfied in both 

structures. The surface bracketing in (5b) differs from that in (5a) 

solely because a clitic, unlike other syntactic primitives, has a mor- 
phological requirement that it is a prefix. 

Williams's paper is also on morphosyntax, but the ideas in the 
paper are at best tangential to the stated theme of alternative 
conceptions of phrase structures. W addresses a potentially interest- 
ing issue that has been clouded by a lot of controversy and misun- 
derstanding: Are word structure and phrase structure the 'same' 
simply because (i) both can be described with a PSG, and (ii) prin- 
ciples which are purportedly 'syntactic' apply both in and outside of 
words? (Cf. Yoon 1989) While this is doubtless an interesting ques- 
tion, it does not bear on the limits of PSG or whether alternatives to 
PSG should be sought in this particular domain. W comes up with the 
conclusion that even though PSG can be used to characterize both 
word and phrase structure (and many other things, in fact), the two 
types of structure are not the same, because the notion of 'maximal 
projection' (XPs) is absent from word structure. W's strategy in the 



{ 



Yoon: Review of Baltin & Kroch 171 

paper is to attribute things like theta roles, predication. Case, and 
others as exclusive properties of XPs. If so, then their 'absence' at 
the word (internal) level can be explained if XPs are lacking at that 
level. 

In a sense, this conclusion is almost trivially true. The system of 
'bar' levels, which yields the notion of 'maximal' projection (XP), is 
(phrasal) syntactic. With it, we characterize 'constituents' and con- 
stituent structure, which are irretrievably (phrasal) syntactic notions. 
Words are the lowest levels (X^) in this hierarchy, simply because 
under standard conceptions they are the atoms/primitives of the 
phrasal rule system, not because they are inherently incapable of 
being 'maximal' in any epistemologically prior sense of the term. The 
hypothesized difference between 'words' and 'phrases' is encoded in 
this distinction. To claim then that the lack of XPs is the source of 
difference between words and phrases is simply a restatement of the 
hypothesis that words are not phrases. What is needed is indepen- 
dent support, outside the domain of the assignment of bar-levels in 
X-bar theory, for the proposed distinction. 

Saito deals with a problem that arises from his earlier (1985) 
analysis of Scrambling in Japanese as A'-movement. The case for the 
A'-movement analysis of Scrambling proposed in that work received 
considerable support from Hoji 1985. However, if Scrambling is A'- 
movement, it should behave in all respects like A'-movement. S 
shows in this paper that there are instances where it does not. More 
such differences between Scrambling and Wh-movement have been 
discovered recently by Mahajan (1990) and Webelhuth (1989), 
among others. As an example, while Wh-movement shows Weak 
Crossover (WCO) effects. Scrambling does not, as seen below, even 
though the requisite WCO configuration is present in both. 

(5) Wh-movement: 

*?Whoi does hisj mother love tj? 

Scrambling (Korean): 

Johni-ul kuj-uy emeni-ka ti salanghanta 
John-ACC hc-GEN molhcr-NOM loves 
'John, his mother loves.' 

S's proposal in this paper is that while Scrambling is an A'-move- 
ment, it can be freely 'undone' at LF, since it does not form an 
operator-variable chain relevant for interpretation. He then suggests 
that the reason an A'-position (adjoined position) in a language like 
Japanese doesn't behave like a typical operator position is that in 
Japanese there are constructions where at D-structure nominals are 
base-generated in these adjoined positions. S calls these 'D-positions' 
(on the analogy to A-positions) and speculates that the difference 
between Wh-movement in English and Scrambling in Japanese might 



172 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

be attributed to whether or not movement is to a D or a D'-position. 
Comp is a D'-position and movement to it cannot be 'undone', while 
adjoined positions in Japanese are D-positions and as such allow A'- 
movement to be 'undone'. 

StOWELL deals with the question of the correspondence between 
syntactic categories and semantic categories on the basis of an inves- 
tigation into the syntax (and semantics) of noun phrases. In partic- 
ular, he finds that there is evidence that referential noun phrases 
should be analyzed as the category 'DP' (cf. Abney 19870, although 
certain predicative uses of noun phrase should be analyzed as 'NP's. 
S examines data on extraction from predicative vs. referential noun 
phrases and the governability of the subject position of each type 
and concludes that there is evidence for maintaining a (restricted) 
version of the DP hypothesis. 

An apparent problem for this is that certain instances of DPs 
(NPs headed by determiners in languages like English) appear in 
clearly predicative contexts. 

(6) Bob called Stan *(a) fool 

This would appear to challenge the conclusion that there is a corres- 
pondence between the semantic type of noun phrases and their syn- 
tactic instantiations. S argues, however, that the nonreferential uses 
of DPs are tied to the expression of 'membership in a kind' (Carlson 
1977). If this is feasible, then, once again, there is a nontrivial cor- 
respondence between syntactic categories and semantic categories. 

The question of whether there is a strict 'correspondence' be- 
tween syntactic and semantic categories is an interesting one, and is 
certainly one of the better uses to which the 'extended' clause/noun 
phrase structure proposals (Chomsky 1989, Pollock 1989, Abney 
1987) can be put, since most proposals on 'articulated' clausal struc- 
ture have been based either on distributional generalizations, con- 
ceptual 'arguments", or the need to account for certain word order 
variations. Regardless of the ultimate correctness of S's conclusion, 
practitioners of GB would do well to turn their sights to the question 
that S dealt with in his paper — the relation between syntactic and 
semantic categories. 

Travis, in the tradition of 'eliminating' PS rules, proposes that 
word order generalizations that are expressible by the head-peri- 
pherality parameter (i.e., head-first vs. head-last) of standard X-bar 
theory are inadequate and proposes to supplement the parameter 
with two additional parameters — directionality of theta-role assign- 
ment and directionality of Case-assignment. These two, in addition 
to a default setting for the head parameter, provide far greater free- 



Ycxjn: Review of Baltin & Kroch 173 

dom in expressing certain sub-generalizations about word orders not 
expressible in the old system. 

T's primary data comes from Mandarin Chinese, and is aug- 
mented with some data from Kpelle. In Mandarin VPs, only those 
elements (NPs or PPs) that are direct arguments of the V can appear 
postverbally, subject to other conditions on their co-occurrence. PP 
adjuncts appear preverbally, and objects of V's may do so as well, if 
they are flanked by the preposition ba. T accounts for this state of 
affairs with the following parametric settings: Chinese is (i) default 
head-final, and (ii) Left-to-Right theta-marking. 

Mandarin word order is a bit more complex than what T des- 
cribes and is open to several alternative analyses (Li 1985, Huang 
1988, Yoon 1989, Tang 1990, inter alia). Yet the idea that the basic 
ingredients of word order typology concern three primitives, instead 
of one, is interesting in that while flexible, it still makes predictions 
about the types of partial word order generalizations that should not 
exist. An extensive typological investigation of a large number of 
languages seems to be the logical next step for this program. 

3. Beginning perhaps with Relational Grammar in the seventies, 
several linguists began to realize that Chomsky's answer to the 
'inadequacy' of a simple PSG is not the only, or the most plausible, 
way of remedying its defects. Both Relational Grammar and Lexical- 
Functional Grammar seek the direct expression of FUNCTIONAL/ 
RELATIONAL information in the syntactic representation. The absence 
of any offering from RG in this collection appears to be a major omis- 
sion. LFG is represented by a paper of KAPLAN and Zaenen (KZ) which 
proposes a new analysis of long-distance dependencies in functional, 
rather than constituent-structural terms. 

The paper begins by showing that there are instances of argu- 
ment-adjunct asymmetries with respect to extraction which cannot 
be adequately characterized in structural terms. Ever since Huang 
1982, there has been intensive research on the differing locality con- 
ditions on adjunct vs. argument extraction (and extraction from ad- 
juncts vs. arguments). Even within a structure-based theory like GB, 
the current concensus is that the argument-adjunct asymmetry can- 
not be adequately described with structure alone but must make ref- 
erence to functional/relational notions such as 'L/Theta-marking' 
(Chomsky 1986). The same conclusion is supported in KZ. However, 
the similarity between the two ends here. 

Employing the concept of 'functional uncertainty', the new 
analysis of LD dependencies in KZ takes on the following rough form: 



174 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

(7) S' ^ X Y 

(TDF) = i 

(T DF) = (T body bottom) 

DP stands for discourse functions such as TOP(ic). The 'body' con- 
strains the function of constituents that can occur 'between' the filler 
and the gap (bottom). The 'bottom' constrains the grammatical func- 
tion of elements that correspond to the 'gap'. (8) is the Topicalization 
rule for English given in this format. 

(8) S' -^ XPorS' S 

(TTOP) = i 

(T TOP) = (T {COMP, XCOMP}* (GF-COMP)) 

(8) says that topics in English can bear any GF except COMP (finite 
clause complements) and that the body of the dependency can cross 
a potentially infinite number of COMPs and XCOMPs (nonfinite com- 
plements). 

One thing to note is the treatment of locality constraints. In this 
revision of LEG, the theory of 'islands' is built directly into the LD de- 
pendency rules for each individual language — in the form of spe- 
cification of possible functional paths in the body of the rule intro- 
ducing the dependency. This stands in contrast to recent attempts 
within GB circles that attempt to define possible islands in universal 
terms. This is where the two part ways. 

Stipulating locality into the rules is no doubt more 'economical' 
as KZ claim, but this move appears to be a step back into construc- 
tion-specificity which is being avoided as much as possible these 
days. The paper also includes interesting discussion on whether cat- 
egorial (and by hypothesis, c-structural) information is ever relevant 
in the syntax of LD dependencies and demonstrates the usefulness of 
multi-domination in f-structures introduced by different types of 
functional dependencies. 

Karttunen's contribution is a categorial-unification grammar 
(CUG) fragment of Finnish. CUG adapts a classical Categorial Grammar 
with (directional) Functional Application to the unification formalism 
currently used in several other theories. With the novel assumption 
that nouns are functor categories and verbs are basic categories in 
free word order languages like Finnish, K demonstrates that the facts 
of free word order, including 'long-distance' (non-clause bounded) 
Scrambling which either requires more powerful combinatorial oper- 
ations such as Type Raising and Functional Application (cf. Steed- 
man's contribution) or unbridled use of transformational adjunction 
(Saito 1985 and Saito's contribution to the volume) can be handled 
with only Functional Application. 



Yoon: Review of Baltin & Kroch 175 

KroCH's paper represents Tree Adjoining Grammar (TAG) in the 
volume. TAG does not offer a new formalism for PS trees. The in- 
novation in TAG is in the treatment of embedding. Instead of stating 
possibilities of embedding directly in PS-rules, TAG defines an 
operation of Adjunction which maps elementary trees onto elemen- 
tary trees to derive complex structures, somewhat in the manner of 
embedding transformations of early Transformational Grammar 
(Chomsky 1955/75, 1957). The subject matter of the paper is the 
apparent exception to island constraints (Subjacency) that have been 
noted in the literature and how TAG can deal with them. 

This question is of importance to TAG because in TAG GLOBAL 
conditions on complex structures like Subjacency can be made to fall 
out as a theorem when reasonable assumptions are made about 
constraints on its elementary structures. Thus, the fact that WH 
extraction out of a 'WH-island' is impossible can be made to follow 
from the simple fact that multiple WH fronting is impossible in 
simple clauses in English. This is because when a tree is adjoined to 
an already ill-formed tree, the resulting tree is also ungrammatical. 

The problem with this approach is two-fold: First, there are cer- 
tain instances of Subjacency-like violations that cannot be accounted 
for straightforwardly under this derivation of Subjacency; cf. (9). 
Second, there are selective violations of WH-islands which must be 
allowed, in particular, the fact that arguments violate WH-islands 
much more freely than adjuncts when extracted; cf. (10). 

(9) *Whoi does he think that ei left? 
Wheni does he think that we left Ci? 

(10) ?Whati were you wondering howj to say Ci Cj? 
*Howjwere you wondering what, to say e; Cj? 

K solves these problems by importing GB conditions such as proper 
government into TAG, formulating them with the notion of auxiliary 
tree sets. K concludes that the flexibility of the TAG formalism in ac- 
counting for a range of cross-linguistic variations with regard to is- 
land-related phenomena makes it a serious candidate as a formalism 
for describing natural language syntax. 

MCCawley's contribution does not propose an 'alternative' to 
phrase structural accounts of syntax. Nor does it contain a specific 
proposal concerning formal aspects of PSG. It is noncommittal and 
yet it probes into areas of natural language syntax where the stan- 
dard mathematical strictures (i.e., P-marker axioms) appear to run 
into problems. 

Dubbing them 'individuation' problems, M draws on the wealth 
of his knowledge and astute intuitions and investigates various 
constructions that have direct bearing on the standard P-marker 



176 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

axioms. He questions the validity of Single-Rootedness on the basis 
of sentences with parentheticals and S, therefore S sequences. He 
raises the question of whether or not succeeding strata in a de- 
rivational model of syntax constitute a single or multiple repre- 
sentation(s). The validity of the ban on Multiple Domination in P- 
markers is questioned on the basis of the study of the syntax of 
'restructuring' in Japanese. Issues arising from morpho-syntactic in- 
dividuation (French du and other hybrids) and categorial/selectional 
uncertainty round out the paper. 

Sag and POLLARD's paper is one among several co-authored 
papers the duo have presented over the years in which the theory of 
Head Driven Phrase Structure Grammar (HPSG) is expounded. By 
now familiar to a wider circle of syntacticians, HPSG differs from its 
predecessor GPSG in not adopting P-markers as the formalism for 
syntactic (and semantic) description, opting instead for feature 
structures (attribute value matrices) and employing the operation of 
unification as the fundamental tool for expressing various kinds of 
syntactic dependencies, as opposed to the PS relations of precedence, 
dominance, and labeling. The virtue of adopting unification is that it 
allows a more DIRECT encoding of linguistic information than in a PS- 
based approach, and that it is declarative and monotonic. These are 
properties that are taken to be virtues from a computational 
perspective. The approach is also akin to LFG in directly encoding 
certain types of grammatical functions. The coverage of a fragment 
of English syntax in the paper is quite extensive. Questions of 
subcategorization, constituent order, grammatical function, and 
semantic role selection are dealt with, among other topics. 

HPSG has going in its favor a winning combination of (i) a level 
of rigor needed to make claims precise and to test their predictions, 
and (ii) a flexibility that enables the formalism to serve potentially 
as a 'lingua franca' within which various syntactic traditions can be 
formalized and tested. 

Steedman's contribution is the second Categorial-Grammar (CG) 
paper in the volume. In contrast to the stated aim of Karttunen's 
paper, S employs the full gamut of operations available in 'extended' 
CG — (directional) Application and Composition and Type Raising. S 
deals with a range of 'problematic' constructions — nonconstituent 
coordination, right node raising, gapping, and unbounded dependen- 
cies. The attractiveness of S's analysis is that it allows one to simply 
do a left-to-right parse of the syntactic string for these constructions 
and still get their syntax and semantics to come out right. However, 
the theoretical 'cost' in the process is the extended use of Type Rais- 
ing and the formation of some unorthodox "constituents'. 



Yoon: Review of Baltin «fe Kroch 177 

While the simplicity and elegance of this extended CG treatment 
of 'problematic' constructions are impressive, there is a certain sense 
in which the rather pervasive derivational ambiguity allowed by the 
system, even if harmless from a mathematical point of view (as S 
notes), seems unnatural. For example, in S's account, a string like 
(11) can be parsed in two ways (at least). 

(11) Apples are good 

In one derivaton, the predicate are good (S\NP) combines by Ap- 
plication with apples (type NP) to yield an S. Another possible deriv- 
ation is one where the subject has been Type Raised (S/(S\NP)) and 
combines via Composition with the predicate to yield S. Both deriva- 
tions, however, yield equivalent semantic interpretations. 

4. Given the theoretical fragmentation of the field of syntax, it 
might be impossible to come up with a less diversified collection of 
papers than this one. A minor objection must be raised against the 
title. The title of the volume engenders false hopes that there is 
more coherence in the selection and theme of individual papers than 
actually exists. However, the individual papers are in general in- 
formative and stand up on their own, even when not addressing the 
topic of 'alternative conceptions' of phrase structure. 

NOTES 

' Abney (1987) popularized the idea that the structure of noun 
phrases is like the structure of clauses currently assumed in GB 
theory; i.e., there is an internal lexical projection (NP or N') whose 
outer layer is a functional projection, headed by the category D(et). 
As S notes, Abney's proposal for articulated noun phrase structure is 
based mainly on distributional considerations. S, on the other hand, 
seeks to 'ground' the DP hypothesis in a theory of semantic types of 
nominals. 

REFERENCES 

Abney, Steven. 1987. The noun phrase in its sentential aspect. MIT 
Ph.D. dissertation in Linguistics. 

Carlson, Greg. 1977. Reference to kinds in English. University of 
Massachusetts Ph.D. dissertation in Linguistics. 

Chomsky, Noam. 1955/75. Logical structure of linguistic theory. 
Chicago: University of Chicago Press. 

1957. Syntactic structures. The Hague: Mouton. 

1970. Remarks on nominalization. Readings in English trans- 
formational grammar, ed. by R. Jacobs & P. Rosenbaum, 184- 
221. Boston: Ginn and Company. 



178 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

1981. Lectures on government and binding. Dordrecht: Foris 

1986. Barriers. Linguistic Inquiry Monograph. Cambridge: MIT 
Press. 

1989. Some notes on the economy of derivation and represen- 
tations. MIT Working Papers in Linguistics 10.43-74. 

Grimshaw, Jane. 1988. Adjuncts and argument structure. Lexicon 
Project Working Papers 21, MIT 

HOJI, Hajime. 1985. Logical form constraints and configurational 
structures in Japanese. University of Washington Ph. D. disser- 
tation in Linguistics. 

Huang, James C-T. 1982. Logical relations in Chinese and universal 
grammar. MIT Ph.D. dissertation in Linguistics. 

1988. Wo pao-de kuai and Chinese phrase structure. Language 

64.274-311. 

JACKENDOFF, Ray. 1977. X-bar Syntax. Linguistic Inquiry Monograph. 
Cambridge: MIT Press. 

KURODA, Sige-Yuki. 1988. Whether we agree or not: A comparative 
syntax of English and Japanese. Papers from the Second Inter- 
national Workshop on Japanese Syntax, ed. by W. J. Poser, 103- 
144. Stanford: Center for the Study of Language and Informa- 
tion. 

Ll, Audrey. 1985. Abstract Case in Chinese. University of Southern 
California Ph.D. dissertation in Linguistics. 

Mahajan, a. 1990. On the A-A" distinction. MIT Ph.D. dissertation 
in Linguistics. 

Marantz, Alec. 1984. On the nature of grammatical relations. 
Linguistic Inquiry Monograph. Cambridge: MIT Press. 

1988. Clitics, morphological merger, and the mapping to 
phonological structure. Theoretical morphology: Approaches in 
modern linguistics, ed. by M. Hammond and M. Noonan, 253- 
270. San Diego: Academic Press. 

Pollock, Jean-Yves. 1989. Verb movement, universal grammar, and 
the structure of IP. Linguistic Inquiry 20.365-424. 

SaitO, Mamoru. 1985. Some asymmetries in Japanese and their 
implications. MIT Ph.D. dissertation in Linguistics. 

SprOAT, Richard. 1985. On deriving the lexicon. MIT Ph.D. disser- 
tation in Linguistics. 

Tang, Jane. 1990. Chinese phrase structure and the extended X-bar 
theory. Cornell University Ph.D. dissertation in Linguistics. 

Wall, Robert. 1972. Introduction to mathematical linguistics. 
Englewood Cliffs: Prentice-Hall 

Webelhuth, G. 1989. Syntactic saturation phenomena in the Ger- 
manic languages. University of Massachusetts Ph.D. dissertation 
in Linguistics. 

YOON, James Hye-Suk. 1989. A restrictive theory of morphosyntactic 
interaction and its consequences. University of Illinois Ph.D. dis- 
sertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



REVIEW 

Deborah Tannen: You just don't understand. New York: 
William Morrow and Company, Inc., 1990; pp. 330. 

Theresa Conefrey 
(Division of English as an International Language) 

Deborah Tannen's latest book is an attempt to explain mis- 
communication between the sexes. In her 1986 book, That's not what 
I meant, she devoted the eighth chapter to gender differences in 
conversational style. This chapter was the one that apparently re- 
ceived most interest from her readers and the one which forms the 
basis of her most recent book. The basic tenet of that chapter and the 
present book is that sons and daughters are treated differently from 
birth onwards. Although they may grow up in the same family, so 
different are the lessons they learn that their upbringing is the 
equivalent to an upbringing in two different cultures. Language is 
used differently about them and by them and they learn to assign 
different meanings to the same interaction. When men and women 
interpret each other's conversation in terms of their own systems and 
are not aware that another interpretation exists, inexplicable and 
painful misunderstandings can arise. It is toward raising awareness 
and reducing miscommunication that this book is dedicated. 

As indicated above, inculcation in a conversational style begins at 
an early age. Tannen draws on the work of Malz & Borker (1982), 
Dorval (1990), and others to highlight the different worlds that boys 
and girls inhabit. Whereas little girls play in cooperative, egalitarian 
groups, evidence suggests that little boys play in groups where there 
is continuous competition and jostling for status. Girls spend more 
time talking and exchanging personal information, while boys are 
more physically active and use talk to try and prove that they are 
superior, rather than equal, to their peers. Although conflicting 
desires for independence and involvement are present in all child- 
ren's interactions, the focus of girls' talk is on involvement while the 
focus of boys' talk is on independence. It is not surprising that 
problems arise in later life, because the stage has already been set at 
an early age for different uses and expectations of verbal communi- 
cation. 

Men and women carry into adulthood the same conflicting 
desires for involvement and independence in interactions, as well as 
the mistaken belief that thcir's is the one and only way to speak and 



180 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

to listen. This means that they can interpret the same conversation 
differently even if there is no apparent misunderstanding. Whereas a 
woman might offer a man advice as a way of showing concern, he 
may interpret this as a threat to his right to make his own decisions. 
Similarly, while a man might intend to make his wife feel better by 
telling her that his problems are worse than hers, she might interpret 
his behavior as invalidating her experiences and trying to distance 
himself from her worries. Thus, as well as having to balance dual 
needs for involvement and independence, men and women have to 
cope with different ways of expressing those needs. Tannen claims 
that the differences between male and female communication styles 
are significant enough to view male-female communication as cross- 
cultural communication. If we accept this claim, that is, that the 
differences between men's and women's ways of speaking and using 
speech are as great as those between speakers of different ethnic 
backgrounds, we can more readily accept that differences between 
men's and women's talk are indeed considerable. 

A commonly held belief is that women are the more loquacious 
of the sexes. This myth seems counter-intuitive when we consider 
study after study that shows that men talk more than women in the 
classroom, in conferences, at work and at social gatherings. However 
if we consider Tannen's discussion of the purposes of women's and 
men's talk, there are some possible explanations. Offices and schools 
are public places. The participants in the interactions are 'on show'. 
For men talk is a way of preserving independence and maintaining 
status in a hierarchical order. The more they talk and display 
knowledge the more attention and esteem they receive. In the public 
sphere, therefore, there is great incentive to talk and occupy the floor. 
Turns taken and utterance length of women's speech are over- 
estimated and women are perceived as talking much more than they 
actually do because, according to Spender (1980), they are not 
expected to be talking at all. Talk in the home, however, is a very 
different matter. In the private sphere men feel they can relax; they 
do not have to compete so they do not have the same incentive to 
impress through verbal skill. Women frequently complain that they 
have trouble getting men to talk at all. Women talk a dispropor- 
tionate amount of the time in order to draw men into conversation 
because talk is a means of establishing intimacy. By way of illus- 
trating the effort involved, Tannen comments on the many jokes 
about the insurmountable morning newspaper at the breakfast table. 
In the private sphere men think women talk a lot because they talk 
in situations where men would not. 

Dividing speaking arenas into 'public' and 'private' also throws 
light on the mystery of the woman who rarely opens her mouth at a 
meeting yet talks freely in the house, and the husband who is 'the life 



Conefrey: Review of Tannen 181 

and soul of the party', but has practically nothing to say to his wife in 
the home. The husband interprets the wife's attempts to 'make 
conversation', to establish a connection between them, as attempts to 
overwhelm him with inconsequential talk and restrict his freedom to 
spend his leisure time as he wants. The woman for her part inter- 
prets the man's behavior as unfriendly and inconsiderate. Tannen 
introduces the paired terms 'report-talk and rapport-talk', 'contest 
and community', and 'lecturing and listening' to characterize these 
differences in men's and women's conversational styles. 

Since it is not just academics but also laypersons who are faced 
with having to interact with the opposite sex, it is appropriate that 
this book is aimed at the general reader. For this reason the text is on 
the one hand lively and entertaining, but on the other hand overly 
dramatic and simplified. Although Tannen suggests that both men 
and women should be aware of differences in each others' style and 
should both seek to accommodate one another, she notes that male- 
female conversations are more like men's conversations than they are 
women's. There is much less incentive for men to become familiar 
with women's genderlect than there is for women to become fluent in 
men's. Tannen implies, however, that the different styles of men and 
women are equally valid. As long as men use talk primarily to nego- 
tiate and maintain status in a hierarchical society, they will continue 
to dominate women. As long as women use talk primarily to nego- 
tiate intimacy and focus on relationships rather than status, they will 
continue to be dominated by men. Though Tannen's book goes far in 
explaining how and why miscommunication arises, it unfortunately 
does not come up with viable ways of improving women's 
experiences of cross-sex communications. Nonetheless there is much 
of interest for men and women scholars and laypersons alike, and the 
path is cleared for further in-depth study in this hitherto neglected 
field. 

REFERENCES 

DORVAL, Bruce (ed.). 1990. Conversational coherence and its develop- 
ment. Norwood, NJ: Princeton University Press. 

Maltz, Daniel N., & Ruth A. Borker. 1982. A cultural approach to 
male-female miscommunication. Language and social identity, 
ed. by J. J. Gumperz. 196-216. Cambridge University Press. 

Spender, Dale. 1980. Manmade language. London: Routledge and 
Kegan Paul. 

Tannen, Deborah. 1986 That's not what I meant! How conver- 
sational style makes or breaks your relations with others. New 
York: William Morrow. 



i 



Studies in the Linguistic Sciences 
Volume 21, Number 1, Spring 1991 



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. 

C. Anthony Anderson & Joseph Owens, eds. Propositional atti- 
tudes: The role of content in logic, language, and mind. (Lec- 
ture Notes, 20.) Stanford, CA: Center for the Study of Language and 
Information, 1990, pp. xvi, 342. $16.95 (Paper), $37.50 ('Library 
cloth"). 

'The papers in this volume treat issues involved in formulating a logic of pro- 
positional attitudes and consider the relevance of attitudes to the continuing 
study of both the philosophy of language and the philosophy of mind.' (Cited 
from the jacket.) Contributors are: K. Fine, H. Kamp, E. Lepore & B. Locwer, T. 
Burge, R. Stalnakcr, J. Owens, H. Wallace & H. E. Mason. K. S. Donnellan. N. Sal- 
mon, S. Schiffcr, J. R. Scarle, and K. Gunderson. 

Edna Andrews. Markedness theory: The union of asymmetry 
and semiosis in language. (Sound and Meaning: The Roman Jakob- 
son Series in Linguistics and Poetics.) Durham, NC: Duke University 
Press, 1990, pp. ix, 220. 

The notion 'markedness' was introduced to linguistics by Jakobson. The pres- 
ent book attempts to relate the notion to Peircean semiotics. Chapter One out- 
lines 'The principles of Jakosonian markedness theory' and considers, among 
other topics, deixis and 'shifters' in the Russian verb. Chapter Two deals with 
'Peirce and Jakobson revisited: a reconciliation.' Chapter Three discusses 'Mar- 
kedness theory as mathematical principle.' Chapter Four addresses 'Myths 
about markedness' and considers issues such as statistical frequency, neutral- 
ization, and substitutability. Chapter Five applies markedness theory to 'The 
category of grammatical gender in Russian. Serbo-Croatian, and Modern 
Greek.' 

John Baldwin & Peter French. Forensic phonetics. London & New 
York: Pinter Publishers, 1990, pp. (viii,) 141. $47.50. 

Phonetic evidence is increasingly being used in legal proceedings. This book 
is intended to make relevant information available to persons in the legal 
profession and the police who have no prior knowledge of phonetics, as well 
as to provide information on issues of English criminal law relevant to foren- 
sic phoneticians. 



184 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

Susan Bassnett & Andre Lefevre, eds. Translation, history, and 
culture. London & New York: Pinter Publishers, 1990, pp. viii, 133. 
$49.00. 

This book offers twelve essays concerned with translation studies, with em- 
phasis on cultural and social factors relevant in and for translation. Contrib- 
utors are: S. Bassnett, D. Delabatista, B. Godard, M. Hjort, P. Kuhiwczak, A. Le- 
fevre, V. Macura, M. Sengupta, S. Simon, M. Snell-Homby, A. Tabakowska, M. 
Tymoczko, and P. Zlateva. 

Philipe E. Bennett &. Graham A. Runnalls, eds. The editor and the 
text. Edinburgh: Edinburgh University Press, 1991, pp. xiv, 175. 
$15.00. 

This volume addresses the issue of editing medieval French text, which invol- 
ves drawing on a large number of different disciplines, including paleo- 
graphy, historical linguistics, lexicography, and most important, philology. 
Contributors are: E. Baumgartner, P. E. Bennett, C. Corley, T. Hunt, A. Kennedy, 
J. C. Laidlaw, A. Lodge, J. H. Marshall, P. Menard, G. Roques, G. A. Runnalls, W. 
van Emden, and K. Varty. 

Christine Cheepen & James Monaghan. Spoken English: A prac- 
tical guide. New York & London: Pinter Publishers, 1990, pp. (viii,) 
215. $47.50. 

The book is intended as a textbook for students of language and linguistics, 
with focus on spoken, conversational English and the encoding of topic, 
speaker status, and speaker orientation. The book concludes with exercises, a 
bibliography, an appendix of transcribed conversations, and an index. 

D. S. Clarke, Jr. [ed.]. Sources of semiotic: Readings with com- 
mentary from antiquity to the present. Carbondale & Edwards- 
ville: Southern Illinois University Press, 1990, pp. xvi, 208. 

As suggested by the title, this book is a reader of writings on the nature of the 
sign, from antiquity (Aristotle, Quintilian etc.), through the Middle Ages (e.g. 
St. Augustine, Ockham, Hobbes), via early modem philosophers (e.g. Locke, 
Arnauld, Kant), to more recent approaches (including Carnap, Chomsky, 
Hjelmslev, Ogden & Richards, Osgood, Quine, Russell, de Saussure, Skinner, and 
such clearly semiotically oriented authors as Peirce, Barthes, and Sebeok). The 
book concludes with readings on 'recent philosophical developments' con- 
cerned with 'Criticisms', 'Communicative intent', 'Convention', 'Reference', and 
'Iconic representation'. 

Susan D. Fischer & Patricia Siple, eds. Theoretical issues in sign 
language research. Chicago: The University of Chicago Press, 1991, 
pp. ix, 338. $29.95 (Paper), $55.00 ('Library cloth'). 

The book offers articles on sign language in the 'four traditional core areas of 
phonology, morphology, syntax, and semantics.' (p. 1) Most contributions deal 
with American Sign Language (ASL), but four cover other sign languages (of 
Brazil, New Zealand, Sweden, and Taiwan) and one provides a contrastive ana- 
lysis of ASL, Chinese Sign Language, and three sign languages created by 'iso- 
lated deaf adult signers in the absence of all internal linguistic input models' 
(363). Contributors are: L Ahlgren, L. F. Brito, M. Collins-Ahlgren, G. R. Coulter, 



Recent Books 185 

S. K. LiddcU, D. Lillo-Martin & E. S. Klima. C. Lucas & C. Valli. R. P. Meier. D. M. 
Pcrlmutter. W. Sandler, W. H. Smith. T. Supalla, R. B. Wilbur, and S. Yau. 

Jean Mark Gawron & Stanley Peters. Anaphora and quantifica- 
tion in situation semantics. (Lecture Notes, 19.) Stanford, CA: 
Center for the Study of Language and Information, 1990, pp. xi, 187. 
$15.95 (Paper), $37.50 ("Library cloth"). 

'This book is an investigation into the semantics of quantification and ana- 
phora with third person singular pronouns.' (p. 1) Chapter 2 sketches the 
general theory of meaning and semantic interpretation adopted; Chapter 3 
provides a fragment of a semantics for NPs; Chapters 4 and 5 specifically deal 
with anaphora; Chapter 6 turns to the 'theory of circumstances' and a 
circumstance-based account of scope, concluding with a brief presentation of 
Binding Theory; Chapter 7 compares the present approach to those of 
Montague and Kamp & Hcim, and accounts of binding conditions such as 
Reinhart's. 

Patrick de Gramont. Language and the distortion of meaning. 

(Psychoanalytic Crosscurrents.) New York & London: New York Uni- 
versity Press, 1990, pp. xi, 292. $45.00. 

Drawing on evidence from infant observations and linguistics, as well as in- 
formation theory, de Gramont attempts to show how language 'distorts mean- 
ing' and our perception of reality. The focus of the book is psychoanalytical. 

Claude Hagege. The dialogic species. (European Perspectives.) New 
York: Columbia University Press, 1990, pp. xii, 288. $35.00. 

A translation of L'homme du paroles: contribution linguistique aux sciences 
humaines (\9S5, Librairie Arthcme Fayard), this volume offers an opportunity 
to become familiar with the thoughts on human language by Hag6ge, one of 
the most productive and influential French structuralists, but relatively un- 
known outside France. The book draws on H's extensive work with a large 
variety of European and non-European languages. (The latter include Tikar 
and Mbum of the Cameroon, Palau of Micronesia, and Comox Laamen of British 
Columbia.) H's goal is to present a synthesis on the issues of linguistic struc- 
ture and development, typology, the social function of language, innateness, 
and the relation of language to human nature in general. 

William F. Hanks. Referential practice: Language and lived 
space among the Maya. Chicago & London: The University of Chi- 
cago Press, 1991, pp. xxiv, 580. $27.50 (paper), $65.00 (cloth). 

Referential practice is an anthropological study of language use, based on ex- 
tensive fieldwork with Maya speakers in Yucatin. The central concern of the 
book is dcixis, defined as a 'cultural construct' that links language with the 
physical space of speakers' bodies, their immediate surroundings, and their 
ritual and conceptual world. Part I addresses the 'Social foundations of refer- 
ence'. Part II, 'Person, participation, and perception'. Part 111, 'Space and spa- 
tial reference', and Part IV, 'Structure in referential practice'. 

Hans Henrich Hock. ed. Studies in Sanskrit syntax: A volume in 
honor of the centennial of Speijer's Sanskrit syntax. Delhi: 
Motilal Banarsidass, 1991. pp. xi. 244. Rs. 180/- (cloth), 95/- (paper). 



186 Studies in the Linguistic Sciences 21:1 (Spring 1991) 

The volume contains papers originally read at Sanskrit Syntax Symposia of the 
1986 and 1987 South Asian Languages Analysis Roundtables (University of Illi- 
nois, and Cornell University & University of Syracuse). Topics include word or- 
der, the interaction between phonology and syntax, and the use and function 
of grammatical categories. A wide variety of different approaches are present- 
ed, ranging from historical-comparative to synchronic-theoretical. Many pa- 
pers deal with Vedic, but the classical language is presented as well, and one 
contribution focuses on modern spoken Sanskrit. Contributors arc: A. Akluj- 
kar. R. N. Aralikatti, V. Bubenik. M. M. Deshpande, H. H. Hock. S. W. Jamison. B. 
D. Joseph. J. S. Klein. K. Meenakshi. S. SchSufele. and B. Tikkanen. The volume 
concludes with a bibliography on Sanskrit syntax compiled by H. H. Hock and 
M. M. Deshpande. 

Sharon Inkelas & Draga Zee, eds. The phonology-syntax connec- 
tion. Chicago & London: The University of Chicago Press, 1990, pp. 
XV, 428. 

This volume grew out of a workshop on the phonology-syntax connection, 
held in May 1988 at Stanford University. The purpose of the workshop was to 
bring together linguists concerned with the interaction between phonology 
and syntax and the role of prosodic hierarchy in this interaction. Contributors 
arc: L. Bickmore, M. Y. Chen, Y.-M. Y. Cho. C. Condoravdi, B. Hayes. L. Hyman. S. 
Inkelas. E. M. Kaissc. J. M. Kancrva. I. Kcncsci. M. Kenstowicz. L. Kidima, C. W. 
Kisseberth. B. McHugh, M. Ncspor, D. Oddcn. W. Poser. K. Rice. E. Selkirk. T. 
Shcn. I. Vogel. D. Zee. and A. Zwicky. 

Thomas Amis Lyman. Grammar of Mong Njua (Green Miao): A 
descriptive linguistic study. Sattley, CA: The Blue Oak Press, 
1979, pp. ix, 100. 

Mong Njua (or Green Miao) is a language of northern Thailand. This volume 
provides information on the ethnography, phonology, and morphology of ihc 
language, plus thirty-two pages of tcxis and a bibliography of earlier work on 
the language. 

P. A. Messelaar. La confection du dictionnaire general bilin- 
gue. Leuven: Peelers, 1990, pp. 109. 

According to the preface, this book can be considered an aide-memoire for the 
head of a bilingual dictionary project or a manual for beginning bilingual le- 
xicographers. The areas covered include elements of lexicography, semantics, 
stylistics, communication theory, and translation techniques. Illustrations are 
drawn from a variety of Romance and Germanic languages. 

Yves Charles Morin & Etienne Tiffou. Dictionnaire complemen- 
taire du bourouchaski du Yasin. Paris: Peeters/Selaf, 1989, pp. 
(vi), 58. 

Etienne Tiffou & Jurgen Pesot. Contes du Yasin. Paris: Peelers/ 
Selaf, 1989, pp. (vi), 163. 

Burushaski. a language isolate in the extreme north of South Asia, is of con- 
siderable interest to linguists concerned with linguistic convergence in South 
Asia, as well as to typologists. Of special interest to typologists are the ergative 
construction, considerable restrictions on finiteness, and a complex morpho- 
logical system of both prefixal and suffixal verb agreement. Until recently, the 



Recent Books 187 

only major sources on the language were D. L. R. Lorimer's The Burushaski 
language, 3 vols. (Oslo: Institutet for Sammenlignende Kulturforskning, 1935- 
1938) and H. Berger's Das Yasin-Burushaski (Wiesbaden: Harrassowitz, 1974). 
The two volumes under consideration are welcome additional sources. The 
volume by Tiffou & Pesot offers far more than its title (Contes du Yasin 'Tales of 
Yasin') might suggest: Pages 7-80 contain an extensive outline of Yasin-Buru- 
shaski grammar, including two sections (pp. 53-71 and 73-75) concerning syn- 
tactic issues, a topic that has received short shrift in earlier publications. The 
volume by Morin and Tiffou provides a fuller account of the lexicon than 
available in earlier publications. 

Geoffrey Nunberg. The linguistics of punctuation. (Lecture Notes, 
18.) Stanford, CA: Center for the Study of Language and Information, 
1990, pp. ix, 141. $14.95 (Paper), $35.00 ('Library cloth'). 

In contrast to widely held views on the relation between written and spoken 
language, Nunberg claims that written language is equivalent to and, in large 
measure, independent from spoken language, and therefore worthy of serious 
linguistic study in its own right. Specifically, he attempts to demonstrate that 
punctuation is 'a linguistic subsystem and hence to be considered as part of the 
wider system of the written language ...' (6) This system has arisen enirely 
within the written medium, 'as a response to the particular communicative re- 
quirements of written language texts, and as an exploitation of the particular 
expressive resources that graphical presentation makes available.' (7) 

Paul Postal & Brian D. Joseph, eds. Studies in Relational Gram- 
mar, 3. Chicago & London: The University of Chicago Press, 1990, pp. 
xii, 390. 

This is the third and final volume in a series of Studies in Relational Grammar 
published by the University of Chicago Press. (The earlier two volumes, edited 
by D. M. Perlmutter and D. M. Perlmutter & C. G. Rosen, appeared in 1983 and 
1984, respectively.) The present volume contains 'a selection of papers which 
not only draw on and support the insights, analyses, and theoretical devices 
developed in the earlier collections but also provide various refinements and 
modifications.' (viii) Among these are 'the recognition of a broader class of 
primitive relations', 'specification of universal restrictions' on verb agree- 
ment, and new proposals regarding clause union, (ibid.) Some of the papers 
are formulated in the framework of Arc Pair Grammar. Contributors are J. L. 
Aissen, B. J. Allen, A. Berinstein, S. Dubinsky, D. G. Frantz, D. B. Gardiner, D. B. 
Gerdts, J. D. Gibson, N. Gonzalez, B. D. Joseph, D. M. Perlmutter, and P. M. Postal. 

Chris Sinha. Language and representation: A socio-naturalis- 
tic approach to human development. New York: Columbia Uni- 
versity Press, 1988, pp. xix, 235. $35.00. 

This book attempts to provide a synthesis between linguistics, philosophy, se- 
miotics, and biology as regards linguistic and cognitive development. 'The 
book offers a psycho-semiotic analysis of "context"; relating this discussion to 
controversial issues in the acquisition of word meaning, and providing experi- 
mental evidence for its account.' (Cited from the jacket.) 



188 Studies in the Linguistic Sciences 21:1 (Spring 1 99 1 ) 

William A. Smalley, Chia Koua Vang, & Gmoa Yee Yang. Mother of 
writing: The origin and development of a Hmong messianic 
script. Chicago & London: The University of Chicago Press, 1990, pp. 
xii, 221. 

In 1959, Shong Lue Yang, called 'Mother [= Source] of Writing' by his follow- 
ers, began developing an alphabet for two quite unrelated languages, one of 
which is Hmong. The alphabet he produced efficiently conveyed all phonolog- 
ical contrasts (including tones), but its structure was notably different from 
any other writing system the authors of this book were able to locate. The 
authors trace the development of this writing system through the historical 
record created by Shong Lue Yang's disciple, Chia Koua Vang, one of the co- 
authors of the book. 

H. L. Somers. Valency and case in computational linguistics. 

(Edits, 3.) Edinburgh: Edinburgh University Press, 1987, pp. x, 328. 

The aim of this book is to provide an application of linguistic case and valency 
theories to 'computational language-processing tasks'. Part One deals with the 
theories (Chapter 1: Valency; Chapter 2: Fillmore; Chapter 3: Anderson: Localist 
case; Chapter 4: Chafe, Cook, Longacre: Verb features; Chapter 5: Starosta's 
Lexicase). Part Two is concerned with 'Some classical problems for case' (Chap- 
ter 7: Defining the cases: Specificity and multiplicity; Chapter 8: One-case-per- 
argument: Dual roles; Chapter 9: One-case-per-clause: 'Inner' versus 'outer' 
roles). Part Three addresses the issue of 'Case and valency in language pro- 
cessing' (Chapter Eleven: Case in computational linguistics and artificial in- 
telligence; Chapter Twelve: Case and valency in machine translation). 

Loreto Todd & Ian Hancock, eds. International English usage. 
New York: Columbia University Press, 1990, pp. vii, 520. $20.00 
(paper). 

International English usage is a compendium in dictionary-form, covering not 
only, as its title suggests, issues of English usage. It also has entries of the type 
acquisition of language, a summary of psycholinguistic views on language 
acquisition, or Anglo-Romani , a brief account of Romani, the language of 
the Gypsies, as well as its offshoot, a variety of English heavily code-mixed 
with Romani. The focus of the volume is INTERNATIONAL English, i.e.. not only 
the language used by native speakers of English, but also established regional 
varieties (e.g. Indian English) which are used by native speakers of other lan- 
guages. 



^ 



I 



STUDIES IN THE LINGUISTIC SCIENCES 

The following issues are available: 



Vol. 16:2 Fall 1986 

Vol. 17:1 Spring 1987 

Vol. 17:2 Fall 1987 

Vol. 18:1 Spring 1988 

Vol. 18:2 Fall 1988 

Vol. 19:1 Spring 1989 

Vol. 19:2 Fall 1989 



Vol. 20:2 



Fall 1990 



Vol. 20:3 Spring 1991 

IN PREPARATION: 

Vol. 20:1 Spring 1990 



Illinois Studies in Korean Linguistics 
(Editor: Chin-W. Kim) 

Papers from the 1986 South Asian 
Languages Analysis Roundtable 
(Editor: Hans Henrich Hock) 

Papers in General Linguistics 

Papers in General Linguistics 

Papers in General Linguistics 

Papers in General Linguistics 

The Contribution of African Linguistics 
to Linguistic Theory, Vol. 1 
(Editor: Eyamba G. Bokamba) 

Linguistics for the Nineties: 
Papers from a Lecture Series in Celebration 
of the Department's Twenty-Fifth Anniversary 
(Editor: Hans Henrich Hock) 

Meeting Handbook: Thirteenth South Asian 
Languages Analysis Roundtable, 25 - 27 May 
1991, University of Illinois at Urbana-Champaign 



The Contribution of African Linguistics 
to Linguistic Theory, Vol. 2 
(Editor Eyamba G. Bokamba) 



$6.00 
56.00 

$6.00 
$6.00 
$6.00 
$7.50 
$7.50 

$7.50 



$5.00 



$7.50 



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 



Studies in 
The Li 



Sciences 



)hl 21, No. 2 
^all 1991 



ILUNOIS STUDIES IN KOREAN UNGUISTICS, II 

Preface v 

Sang-Cheol Ahn: Vowel deletion and epenthesis: The vowel i 1 

Euiyon Cho: Notes on some tests for subjecthood in Korean 19 

Seikyung Cho: The acquisition of English reflexives by Korean 

ESL learners 31 

Yeon Hee Choi: Discourse reference in written Korean folk tales 69 

Seok Keun Kang: Morale representation of ambisyllabicity: 

Evidence from Korean 89 

Yongsoon Kang: The Locality Condition of tonal systems: With 

special reference to North Kyungsang dialect in Korean 101 

Chin W. Kim and Hyoung-Youb Kim: The character of Korean 

glides 113 

Hyoung-Youb Kim: Prosodic phonology of Korean 127 

Han-gyu Lee: The pragmatics of the pragmatic morpheme com 

*a little' in Korean 143 

Virginia K. McClanahan: The pragmatics of negation in Korean 167 

James Hye Suk Yoon: Inflectional structures in Korean and 

headedness 179 



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: Eyamba G. Bokamba, Chin-chuan Cheng, Jennifer S. 
Cole, Georgia M. Green, Hans Henrich Hock, Braj B. Kachru, Yamuna Kachni, 
Chin-W. Kim, Charles W. Kisseberth, 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 especially students of the Department of Linguistics, 
University of Illinois, Urbana-Champaign. Invited papers by scholars not as- 
sociated with the University of Illinois will also be included. 

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 the Editor, 
Studies in the Linguistic Sciences, Department of Linguistics, University of 
Illinois, 4088 Foreign Languages Building, 707 S. Mathews, Urbana, Illinois 
61801. 

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

UPCOMING ISSUES: Vol. 22:2: Twenty-Five Years of Linguistic Research at 
the University of Illinois at Urbana-Champaign: Post-graduate research by 
Doctoral and Master's degree students in the Department of Linguistics. Editors: 
Braj B. Kachru and Frances Vavrus; Vol. 23:1: Papers in General Linguistics. 



Price: $7.50 (per issue) 



» 



STUDIES IN THE LINGUISTIC SCIENCES 
Illinois Studies in Korean Linguistics, II 



EDITORS 

Chin W. Kim 
Jerry L. Morgan 
James H-S. Yoon 



EDITORIAL ASSISTANT 
Amy C. Cheatham 



VOLUME 21, NUMBER 2 
FALL 1991 



DEPARTMENT OF LINGUISTICS, UNIVERSITY OF ILLINOIS 
URBAN A, ILLINOIS 61801 



Preface 



Korean linguistics at The University of Illinois at Urbana- 
Champaign has grown considerably since the publication of the first 
volume of Illinois Studies in Korean Linguistics in Fall 1986 as a 
special issue of Studies in Linguistic Sciences, vol. 16, no. 2. The 
number of graduate students in Korean linguistics has steadily 
increased reaching a peak of twenty in 1991, a full one-fifth of the 
student body of the Department, eliciting a facetious proposal that 
the Department change its name to the Department of Korean and 
Linguistics. It may have been facetious but not entirely fallacious. In 
the six years since 1986, the Department has produced twelve 
doctoral dissertations on Korean, again one-fifth of 59 total 
dissertations written during that period, and added James H-S. Yoon 
to the faculty (the list of dissertations is found on the next page). 
More often than not, Illinois had the largest representation in 
meetings and conferences devoted to Korean linguistics in the U. S. 
(e.g., biennial meetings of the International Circle of Korean 
Linguistics, Harvard Workshops on Korean Linguistics.) In Korea, The 
University of Illinois at Urbana-Champaign is known as the 
institution that has produced the largest number of Korean Ph.D.'s in 
linguistics. The three editors of this volume can now adequately 
cover nearly all aspects of the Korean language from phonetics and 
phonology to syntax and pragmatics. Few linguistic programs outside 
Korea can so boast. 

Contributors to this volume were limited to those who graduated 
with Ph.D.'s since 1986 and to graduate students who were ABD's, i.e., 
post-Prelim students, in Spring 1991 when this volume was planned. 
Not everyone who was eligible contributed, however. Hectic academic 
life in Korea and miscommunications of one kind or another 
prevented a few from sending their papers. 

We regret that the publication of the volume was delayed one 
year, primarily due to the editors' indolence. We apologize for it. 
Before closing, we would like to thank Ms. Amy Cheatham for her 
immense help in editing the volume. She was more than a 
proofreader and a typesetter. Her remarkable editorial skills 
improved not only the texts in uncountably numerous places but also 
the design of the layout immeasurably. 

February 1992 Chin W. Kim 

Jerry L. Morgan 
James H-S. Yoon 
(Editors) 



vi Studies in the Linguistic Sciences 21:2 (Fall 1991) 

List of dissertations on Korean between 1987-92 at 
The University of Illinois at Urbana-Champaign 

1987 

Hyang-Sook Sohn: Underspecification in Korean phonology 

1988 

Euiyon Cho: Some interactions of grammar and pragmatics in Korean 

Yeon Hee Choi: Textual coherence in English and Korean: An analysis 
of argumentative writing 

1989 

Hye Suk Yoon: A restrictive theory of morphosyntactic interaction 
and its consequences 

1990 

Hyoung Youb Kim: Voicing and tensification in Korean 

1991 

Seok Ran Shim: Word structure in Korean 

Virginia K. McClanahan: Some interactions of grammar and 
pragmatics in negation in Korean 

1992 

Seikyung Cho: Universal Grammar and the Subset Principle in second 
language acquisition: The acquisition of the Governing Category 
Parameter by adult Korean learners of English 

Seok Keun Kang: A morale study of some phonolgoical phenomena in 
English and Korean 

Han-Gyu Lee: The pragmatics and syntax of pragmatic morphemes in 
Korean 

Yongsoon Kang: Phonology of consonant-vowel interaction: With 
special reference to Korean and Dependency Phonology 

Jang Song Lee: Semantics of Korean noun phrases in Discourse 
Representation Theory 



StVKJigS in thg Linguistic Scignggs 
Volume 2 1 , Number 2, Fall 1991 



VOWEL DELETION AND EPENTHESIS: THE VOWEL •* 

Sang-Cheol Ahn 

The main purpose of this paper is to deal with a two-fold 
issue: The deletion and epenthesis of the high back 
unrounded vowel /'. For this purpose I first discuss the 
deletion process and the representation of i. Relating to this 
issue, 1 also reexamine the so-called conjugational h- 
deletion. In proposing a more explanatory rule formulation 
and representation of deleting segments, I employ the 
theories of underspecification and feature geometry. For the 
epenthesis processes, on the other hand, I discuss the 
analogical epentheses as well as loanword epentheses. Then 
I specify phonological as well as morphological 
environments in rule formulation. 

1. Introduction 

According to the theories of underspecification and feature 
geometry, it is assumed that representation of segments (and 
autosegments) should be underspecified to capture better 
phonological generalization (Archangeli 1984, Archangeli & 
Pulleyblank 1986, etc.) and that feature representation should be 
organized geometrically (Clements 1985, 1989, Sagey 1986, 
McCarthy 1988, etc.). 

Employing these frameworks, this paper discusses two opposing 
issues. For the first part, it will be shown how the high back 
unrounded vowel i (known as the least marked vowel) is 
represented in Korean and why the so-called controversy of i- 
deletion and i-epenthesis is solved in favor of deletion.' Moreover, 
within the frameworks of underspecification and feature geometry. I 
will show that the representation of the vowel i is not specified 
below the root node, while the representation of any other vowel 
lacks this node. Then I will briefly reexamine an alternative proposal 
by pointing out its problems and attempt to reformulate the 
controversial /-deletion process. Relating to this issue of /-deletion, I 
will also discuss how the so-called conjugational /i-deletion can be 
treated. 

In addition to the discussion of /-deletion, the second part will 
show in what context an epenthetic vowel is allowed and in what 



2 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 

shape it appears. For this purpose, I will deal with analogical 
epenthesis and loanword epenthesis. For analogical epenthesis, I will 
show how this process is constrained by morphological as well as 
phonological environments. For loanword epenthesis, I will show 
various data triggering vowel epentheses. Then it will be shown how 
epenthetic vowel variation can be handled in terms of assimilation. 

2. i'-deletion 

2.1. Deletion versus epenthesis 

In Korean, many case markers and affixes following a noun or a 
predicate have the high back unrounded vowel i initially in their 
underlying representation. The underlying i also occurs stem-finally 
in many predicates. The affix-initial /' is retained when it is preceded 
by a consonant-final predicate, e.g. mak-'ila -4 [magira] 'in order to 
block". A stem-final /' of a predicate also remains intact when it is 
followed by a consonant-initial suffix e.g. camki-ta -> [camgida] 'to 
lock'. However, the underlying /' is often deleted when in contact with 
another vowel or when between certain sonorant consonants. 

(1) a. po + ila [pora] 'in order to see' 
b. camki + i [camgi] 'to be locked' 

This i-deletion is considered to be a vowel-hiatus breaking 
phenomenon (B.-G. Lee 1976, Kim-Renaud 1982). As in several 
studies, it could be argued that /' is not a part of the underlying 
representation but is inserted by i-epenthesis between obstruents 
(H.-P. Choi 1937, C.-W. Kim 1971, Y.-S. Kim 1989). This assumption 
seems plausible, particularly when i is an affix-initial segment. If we 
follow this assumption, however, then we cannot explain the 
following situation with the z-epenthesis rule. First, there are 
consonant-initial affixes which do not insert i when following a 
consonant-final stem, as illustrated in (2). Thus, if we assume that i- 
epenthesis applies, we have to mark these as exceptions. 

(2) ip-hi-ta [ip^'ida]. *[ibihida] 'to make someone dress' 
mak-ci [makc'i], *[magiji] 'eat (Aspectual)' 

sal-ki [salgi], *[sarigi] 'living' 
k'ak'-ni [k'aT)ni], *[k'ak'ini] 'Do you cut?' 

Second, in the following example (3), there is no way to 
differentiate the aspect affix from the interrogative ending affix if 
we assume the application of the i-epenthesis rule. 

(3) Aspect Ending (Interrogative) 
po-ni 'As we see' po-ni 'Do you see?' 
tat-ini 'As we shut' tat-ni 'Do you shut? 



Ahn: Vowel deletion and epenthesis: The vowel / 3 

Even if we consider the boundary distinction between the two 
affixes, both are expressed as '+' morpheme boundaries. Thus, unless 
we assume i as a part of the underlying representation of the aspect 
affix, it is impossible to distinguish between the two affixes unless 
the interrogative affix is marked [-i-epenthesis]. 

Third, / can be the stem-final segment of a predicate and it does 
not disappear unless it is followed by a vowel initial affix. By 
claiming that i is an epenthetic vowel between consonants, it follows 
that an underlying representation of a predicate stem can be formed 
by a single consonant in the following examples. As we know, 
however, this is not possible because the underlying representation 
of a stem cannot consist of a single consonant. 

(4) /s'V + a/ [s'a] 'Write (Imperative)' 
/k^i + ato/ [k^ado] 'Although it is big' 

Thus, any mono-syllabic i-final predicate stem can be a counter- 
example to i-epenthesis because the stem-final i cannot be inserted 
by j-epenthesis. 

Finally, as was argued by B.-G. Lee (1976:139), the so-called li- 
irregular predicates are strong evidence against the assumption of i- 
epenthesis. In //-irregular predicates, / is deleted but another / is 
added before a vowel, e.g. [[puli] a] -^ [puUa] 'call!'. (Kim-Renaud 
1973, 1982 posits two /'s in the underlying representation such as 
/pulliV, but this different underlying representation does not affect 
the current discussion.) As we can see in (5), it is impossible to 
derive the correct result in the second example. 

(5) Adv-forming Purpose 

[[pull] la] [[pull] la] 

i-epenthesis 

[pulla] [pulla] Intersonorant i-deletion 

[pulla] *[pulla] Phonetic representation 

([purira]) 

As will be discussed later, an intersonorant i-deletion rule is needed 
as a separate rule which is unrelated to i-epenthesis and i-deletion. 
Unless the second example is marked as an exception to 
intersonorant i-deletion, the correct result [purira] cannot be derived. 

Nevertheless. Y.-S. Kim (1989) recently revived the earlier 
argument for epenthesis. In order to support this claim, he compares 
those inflectional suffixes which take his epenthesized i with tho.se 
that do not. 



4 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(6) a. cap-i'ni 

cap-ina 

cap-imyan 

cap-ila 

cap-il 

cap- in 

cap-ima 

b. cap-ta 
cap-ko 
cap-se 
cap-ca 
cap-ci 

cap-ni (Interrogative) 
cap-na (Interrogative) 
cap-ne (Indicative) 

In this categorization, he observes that those suffixes taking /' begin 
with sonorants, while those not taking i begin with obstruents, except 
two interrogative and one indicative suffixes. Thus he proposes the 
following rule of i-epenthesis. 

(7) i-Epenthesis (Obligatory) 

^ V/C + __C].i„,,.,„d 

I I 

1 [+son] 

This rule implies that, save three exceptional affixes (i.e., two 
interrogatives and one indicative), an i is epenthesized between a 
consonant-final verb stem and a sonorant-initial suffix. And this sort 
of interpretation seems quite convincing at a glance. 

There are, however, several problems encountered besides those 
mentioned above. First, it should be explained why a vowel is 
inserted before a sonorant, rather than before an obstruent. If the 
function of the vowel i' is to euphonize the consonantal sequence, it 
should be inserted between two obstruents, rather than a consonant 
and a sonornant consonant which is closer to vowels in nature. 
Second, it is not a desirable rule if we have to allow an additional 
description for exceptions such as '-Int' or '-Ind'. Third, besides the 
exceptions shown in (6b), there are other exceptional cases such as 
-nya and -nin. 

(8) cap-nya 'Do you hold?' 
cap-nin 'holding' 

Especially, the latter, -n'in is not an interrogative or an indicative 
suffix. Rather, it is an aspectual one and it does not take i, unlike the 
other aspectual suffixes shown in (6a). Finally, there are also many 



Ahn: Vowel deletion and epenthesis: The vowel / 5 

obstruent-initial suffixes which should take /, such as the honorific 
suffix -si and the style suffix -p. 

(9) cap-isi-ta 'hold' (cf. po-si-ta 'see') 
cap-ip-ni-ta (cf. po-p-ni-ta) 

All the evidence shown here does not indicate that i is epenthesized 
between a consonant and a sonorant. Rather, it indicates that / exists 
in the underlying representation and it is deleted when adjacent to a 
vowel. Therefore, we assume /-deletion, rather than epenthesis, since 
it is hard to retain the proposal for epenthesis without encountering 
numerous problems. 

2.3. Rule formulation of i-deletion 

The issue of /-deletion was first formulated within the 

framework of generative phonology by B.-G. Lee (1976, 1979). He 

introduced a generalized /-deletion rule (10), by employing the 
mirror image convention. 

(10) i-» % [+voc] "]b<N>VA+ fa<n>1 a<b<[+seg] f+seg 1>[+low]> 

J isg J La<+cor> J 

When we remove the complicated symbols from this rule, (10) is 
regarded as the combination of the two rules in (11). 

( 1 1 ) a. 1^0/ [-t-voc] 1 VA + So _ 

(When preceded by zero or an indefinite number of 
segment 's', /' is deleted after the stem of a verb or an 
adjective.) 

b. 1^0/ VA + [+voc] 

(A verb or an adjective's initial /' is deleted before an affix 
initial vowel or an /.) 

This complex generalized /-deletion rule (10) is later revised and 
simplified as another mirror image rule by Lee to account for the 
deletion of affix-initial or stem-final / 's (B.-G. Lee 1979:4). 

(12) i'^ 0% [+voc] + 

This generalization looks very plausible and attractive in its 
simplicity. It is, however, subject to substantial revision due to many 
instances in which it does not apply, e.g. [[ki]^ eke] -> [kiege] 'to him', 
[iikko];^j^, -^ [iilck'o] "at last', [hilk] -> [hik] 'soil'. In other words, (12) 
does not apply to a noun or an underlying representation. Because of 
this, Kim-Renaud (1982) claims that /-deletion should be regarded as 
consisting of four separate rules. 



6 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

(13) a. Affixal i-Deletion 

i -^ / V + 

(The initial i of an affix is deleted when following a stem 
which ends in a vowel.) 

e.g. ki + imyan [kimyan] "if one crawls' 

b. Verbal Stem Final /-Deletion 

i -> / &V 

(The final i of a verb stem is deleted when followed by 
an affix beginning with a vowel.) 

e.g. k'i' & ato [k'ado] 'though (we) extinguish (it)' 

c. Casual i-Deletion (optional) 

X V 1 Y 

12 3 4 

-^1 2 4 
[+long] 

(/■ is truncated when meeting another vowel and the 
remaining vowel is lengthened.) 
e.g. /mail/ -^ [ma:l] 'village' 

d. Interconsonantal i-Deletion (optional) 
"1 ^ / 1 «fe [ml 

e.g. ul-imyan [ulmyan] 'if one cries' 

These rules in (13) can explain those counter-examples unexplained 
by B.-G. Lee's rule. In Ahn (1985a), however, it is argued that we 
need only three rules, instead of four as posited by Kim-Renaud. In 
this paper, however, as the modification made by Ahn (1985a, 1986, 
1991) for Casual and Interconsoantal i-deletion is not so crucial for 
discussing the nature of i, the discussion of this paper will be limited 
to the so-called general i-deletion process. ^ 

Now, as Kim-Renaud (1982:474-5) stated, in addition to Affixal 
i-Deletion, the Verbal Stem Final i-Deletion rule is used to prevent a 
noun stem ending in i' from undergoing i-deletion as in [[ki]eke] -^ 
[kiege] 'to him'. In other words, two separate rules were used to 
explain a single phenomenon, i-deletion in 'stem + affix'. Therefore, 
Kim-Renaud's Affixal i-Deletion and Verbal Stem Final i-Deletion can 
be combined into one rule by using a curly bracket as follows. 

(14) 1^0/ 



V 


[ 




- : -N 


[ V 


(-N = all grammatical 
categories except Noun) 



Ahn: Vowel deletion and epenthesis: The vowel / 7 

As a noun-final /' does not undergo i-deletion, we need to specify this 
grammatical information. Moreover, Kim-Renaud's grammatical 
boundaries can be replaced by a single bracket. This rule can now be 
simplified further by employing a mirror image convention as 
follows: 

(15) i->0% ].n[V 

Because i is deleted except when it is a noun-final segment, we do 
not need additional grammatical information in formulating this 
mirror image rule. Thus the following examples illustrate the 
application of (15) whether /' is preceded or followed by a vowel. 

(16) a. [[s'Ov ato] -^ [s'ado] 'although (we) write' 

[[khiJA ato] -^ [k^ado] 'although (it is) big' 

(cf.[[ki]N eke] -^ [kiege] (*[kege]'to him') 

b. [[ka]v ini] -> [kani] 'because (we) go' 

[[hili]\ ini] -^ [hilini] 'because (it is) fuzzy' 

(cf.[[ka]N ilo] -> [karo] 'to the edge') 

Although the new rule (15) replaces the earlier two rules by 
Kim-Renaud, there remains a crucial factor to be considered in 
relation to the so-called /i-irregular conjugation. 

Predicate final h is deleted in the so-called /i-irregular 
conjugation if followed by a vowel-initial suffix, regardless of the 
grammatical category of the stem. 

(17) [[noh] ato] -» [noado] 'although (we) locate (it)' 
[[nolah] ato] -> [norado] 'although (it is) yellow' 

If the suffix-initial vowel is /', however, /' is deleted obligatorily only 
after the adjective-final h, e.g. [[noh]vimyan] -^ [noimyan] 'if (we) 
locate (it)' vs. [[nolahl^imyan] -> [noramyan] (*[noraimyan]) 'if (it is) 
yellow'. In order to solve this problem, Ahn (1985a, 1986) appeals to 
the methods of CV Phonology. In other words, because this difference 
is related to the grammatical category, Ahn claims that the verb-final 
h is associated to the C slot of the CV tier, while the adjective-final h 
does not have its own C slot underlyingly. 

(18) a. verb C V C b. adjective C V C V 

III I I I I 

n a h nolah 

In other words, the final segment /i in a verb is associated to the 
timing C slot, while the final h in an adjective does not have its own 
timing slot underlyingly. Therefore, as there are two different 
underlying forms for the stem-final h deleting before a sonorant, the 
rule of /i-deletion was formulated as (19). 



8 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



(19) Affixational /i-deletion: (Ahn 1985a,b, 1986) 

<C> ^ / V ] <v>/A EX 

I I 

h [+son] (X = any timing slot) 

Now, related to the affixational /i -deletion, the non-linear rule 
formulation for i-deletion was also proposed by Ahn (1985a), as 
shown in (20). 

(20) i-deletion: mirror image 

<CT o 

I I 

V -^ % X ± .N <p/c> V 

I 



(Note: X = either C or V, p/c = passive/causative affix, 
-N = all grammatical categories except noun) 

As specified in the rule formulation, if an /-final predicate stem is 
followed by a vowel-initial passive/causative affix, the stem should 
be multi-syllabic. 

[kani] 'since (it) goes' 
[camgi] 'to be locked' 



(21) a. /ka-ini/ 
/camki-i/ 
(passive) 



b. /s'l-i/ -^ [s'ii] 'to be used' 

/th'i-i/ -^ [thii] 'to be opened' 

In other words, the nonlinear rule of i-deletion applies to (18b) by 
scanning the CV tier, not the segmental tier. As shown below, we can 
obtain the correct results with these two nonlinear deletion rules. 

(22) 



cvc" 


vc V ■ 




■ CVC V 

1 1 1 1 


VCV 

1 1 1 


inflection 


1 1 1 

n a h. 


1 1 1 
V 1 n i . 




1 1 1 1 
. n 1 a h . 


A 1 n i . 








CVC V 

1 1 1 1 
n 1 a h 


CV 

1 1 
n i 


i-deletion 








CV 

1 1 


VC V CVC V 

III 1 1 1 1 


CV 

1 1 


/i-deletion 


1 1 

n a 


1 1 1 
i n i 




n o 1 a 


n i 





[naini] [norani] 1 ^ r / V V 

Here, however, there are some problems in the formulation of 
two deletion rules. First, as for the affixational /i -deletion, the V slot 
preceding h is redundant at the timing tier for two reasons: i) Due to 
the boundary bracket in the rule formulation, it is possible to predict 
that the h is stem-final, so we can eliminate the morphological 



Ahn: Vowel deletion and cpcnthesis: The vowel / 9 

condition '<V>/A' from the rule description, ii) As syllable-final 
consonant clusters are not allowed in Korean, the stem-final h is 
truncated even after an underlying consonant. 

(23)/manh-a/ [mana] 'much' 

/k'inh-ini/ [k'lni'ni] 'since (we) cut' 
/talh-ato/ [tarado] 'though (it is) worn out' 

Moreover, at the melody tier, the representation of the segment h is 
specified along with the feature representation [-i-son] and this is 
apparently not a desirable rule description. 

In order to solve these problems in a consistent way without 
causing any redundancy, we can employ the theories of 
underspecification and feature geometry. Thus two kinds of 
underlying representation for /h/ are proposed as follows. 3 

(24) a. verb: X b. adjective: 

I 

R-[-f-cont] R-[-Kcont] 

I I 

L L 

I I 

[+s.g] [+s.g] 

(Note: X =timing slot, R = root node, 

L = laryngeal node, [s.g] = [spread glottis]) 

(24) shows that the stem-final h is specified only for the root feature 
[-»-cont] and the laryngeal feature [-Hspread glottis], while other 
predictable features are filled in by redundancy rules: e.g. [ ] -^ 
[-voice], [ ] -> [-son], [ ] -> [-constricted glottis], [ ] -^ [-labial], etc. Here 
the question arises as to whether a simpler underlying 
representation for /h/ can be postulated by eliminating the root 
feature [-^cont]. This possibility, however, should be discarded since, 
without the root feature, the representation of /h/ may not be 
distinguishable from the representation of /t^/ shown in (25). (See 
Iverson (1989) for the representation of Korean obstruents.) 

(25) /th/: X 

R-[+cont] 

I 

L 

(+s.g] 

Based on the arguments above and (24), the earlier version of 
affixational /i-deletion is now reformulated as in (26).'* 



1 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(26) Affixational /i-deletion (revised) 

(X) -^ / ] X 

I I 

R-[+cont] R-[+son] 

I 

L 

I 

[+sg] 

Turning to the earlier rule formulation of /-deletion, however, 
there also remain a couple of things to be reconsidered. First, the rule 
formulation in (20) is overly complex due to the condition on syllable 
counting of causative/passive forms. This problem can be avoided if 
we treat those several cases of monosyllabic stems of 
causative/passive forms as true exceptions being listed in the lexicon 
(Sohn 1987). Second, being neutral to vowel harmony and being the 
epenthetic vowel between consonants (Ahn 1985a; Sohn 1987), i 
should be regarded as the least marked vowel in Korean. Thus, 
within the framework of underspecification, the least marked vowel i 
is represented only as a nucleus root without any features. Based on 
this, the rule formulation for /-deletion is revised as the simple 
figure in (27). 

(27) (Revised) /-deletion: mirror image 



N 
1 








N 
1 


1 
X 

1 


-> 


0/_ 


— J-noun 


1 
[X 



(Note: N = nucleus 

R = root node) 

With the rule formulation (27), I refer to the least marked vowel /' 
with the nucleus node, the timing X-slot, and the root node, but 
without any features, while I underspecify the root node as well as 
the class node, only to simply represent a vowel, regardless of its 
quality. 

Sohn (1987) also proposed a similar rule of /-deletion called 
'empty nucleus deletion' where she deleted the condition '-noun' as 
well, saying that /kiV 'he' is the only non-verbal stem for which i- 
deletion needs the morphological constraint. 

(28) 



N 
1 


N 
1 


X ^ / 
1 


1 
X] 


1 





Ahn: Vowel deletion and epenthesis: The vowel / 1 1 

Here she claims that the epenthetic i" of a loanword noun should be 
treated separately since this process is a matter of Korean syllable 
structure not allowing CC clusters. However, it should be noted that /- 
epenthesis in loanwords is very productive and the epenthetic vowel 
never undergoes the obligatory /-deletion rule, as shown in (29a). 
Moreover, i-epenthesis can apply not only to loanwords but also to 
native words in children's speech as well as in some dialects, a 
process which is called '/-analogy' (Ahn 1985a), as shown in (29b). 

(29) a. /milkhi-e/ -^ [milkhie] 'to milk' 

/c^'ilisimasi-ini/ -^ [c*'risimasi(i)ni] 'since (it is) 

Christmas' 
b. /mak-ta/ -> [magida] 'eat' 

Another problem in Sohn's proposal is that she sets up two 
underlying representations for /', i.e., deleting / and epenthetic /. 

(30) a. deleting /' N b. epenthetic /' N 

I I 

X X 

I 

[] 
However, as both of them surface very freely without any difference 
in their usages, there is no reason to distinguish these two forms. 

3. i-epenthesis 

The obligatory /-deletion (27) applies in derivation or in 
inflection. Thus, a suffix initial / is deleted after a vowel. There is, 
however, an opposite phenomenon which inserts an / before a 
consonant initial suffix, in a consonant final predicate stem. And 
there are two cases which allow /-epenthesis. 

3.1. Analogy 

First, /-epenthesis often occurs in children's speech or in the 
imitation of children's speech: Children insert /'s in the environment 
mentioned above, their way of breaking a consonantal hiatus, in the 
process of language acquisition (P.-G. Lee 1979, Ahn 1985a). Thus, 
considering the morphological information, /-analogy is formalized as 
follows (Ahn 1985a). 

(31) /-Analogy 

-> 1 / CJv/A _ C 

/■-analogy is morphologically conditioned as it applies to predicate 
conjugations only, not to case-markings. 



1 2 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(32) a. Predicate inflection 

mdk- 'eat' -» mak-[i]ta, mak-[i]ci 

nopf^- 'high' -^ noph-[i]ta, noph-[i]ci 

an- 'hug' -^ an-[i]ta, an-[i]ci 

b. Case marking 

mdk 'ink-stick' -^ *mak-[i]to 'ink-stick too', 

*mak-[i]man 'ink-stick only' 
pap 'rice' -^ *pap-[i]to, *pap-[i]man 

In (32), the second morpheme in each example has no alternation in 
different phonological environments. In other words, the underlying 
representation does not include i in every case. By the morphological 
condition specified in (31), i' is inserted only after a predicate stem, 
not after a noun or a case marking. 

Moreover, as inflection is the domain of application, i-analogy 
does not apply to derivation either. Therefore, i-analogy is 
considered to be applied to inflection as in (33a), not to derivation in 
(33b), or to compounding in (33c). 

(33) a. Inflection 

[[kam] ki] -^ [kamigi] 'winding' 

b. Derivation (passive) 

[[kam] ki] -^ * [kamigi] 'to be wound' ([kamgi]) 

c. Compounding 

[puth][cap] -^ *[puthijap] 'to catch' ([putc'ap]) 
[kam][s'a] -> *[kamis'a] 'to protect' ([kams'a]) 
'to wind' 'to wrap' 

Finally, in relation to irregular conjugations, which will be 
discussed in the next chapter, /-analogy precedes an irregular 
conjugation process as follows, (p-irregular conjugation refers to 
stem-final 'p -^ w' change before a long-vowel initial suffix. ^- 
irregular conjugation refers to deletion of s before a vowel initial 
suffix.) As we can see here, /'-analogy feeds the application of an 
irregular conjugation process which changes a stem-final consonant 
before a vowel-initial suffix. As /-analogy precedes an irregular 
conjugation, the stem-final consonants never appear in phonetic 
representations which are derived after /-analogy. Therefore, in (34), 
*[kobida] 'be pretty' or *[cisida] 'make' never appear in Standard 
Korean. (But note that some southern dialects do not follow the 
irregular conjugational process and hence they would pronounce the 
stem-final consonants in their phonetic representations.) 



Ahn: Vowel deletion and epenlhesis: The vowel / 1 3 

(34) a. p-irregular 

/ko:p-ta/ "to be pretty' 

ko:p i' ta /-analogy 

ko:w 1 ta p ^ w 

ko w i ta V shortening 

[kowida] Voicing 

(or [kowuda]) Rounding of i' (*[kobyda]) 

b. 5-irregular 

/ci:s-ta/ 'to make' 

ci:s 1 ta /-analogy 

ci: 1 ta s -> 

ci i' ta V shortening 

[ciida] Voicing 
(*[cisida]) 

c. //-irregular 

/noh-ta/ 'to locate' 

noh 1 ta /-analogy 

no 1 ta h ^ 

[noida] Voicing 

3.2. Loanwords. 

Besides the analogical i-epenthesis process, there is another type 
of i-epenthesis which occurs when a foreign word is borrowed from 
an Indo-European language such as English. When a loanword has a 
complex consonant cluster which is not allowed in Korean syllable 
structure, /' is epenthesized between consonants in order to be 
compatible to Korean syllable structure which does not allow a 
consonant cluster in a syllable on the surface. In order to examine 
the characteristics of this process, we can observe the following facts. 
First, / (bold) is inserted to break up a word-initial consonant cluster. 
For the analogical epenthesis, I discussed the morphological 
environment for rule application 

(35) Christmas -^ [khirisimasi] 
trump -> [thiramphi] 
play -» [p*iirei] 

Second, / is inserted after a word-final consonant if it is a fricative or 
a stop which can be exploded. 

(36) 



bus -> 


[p'asi] 


light 


— » 


[lait"!] 


news -» 


[nyusi] 


seat 


— > 


[S'ithl] 


serve -* 


[s'abi] 


card 


— > 


(khadi] 


Gulf ^ 


[kalph.] 


mike 


— > 


[maikhi] 


pump -^ 


[pfiamp^i'i] 


smog 


— > 


[simogi] 


bunt — > 


[p'anthi] 









1 4 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

However, if the final consonant is an unexploded stop, i will not be 
inserted. Thus, in the following example, we do not get any 
epenthesized i word-finally as the final consonants are unexploded 
ones. 

(37) jeep ^ [c'ip], ??[c'ibi] 

Contac -^ [khonthaek], *[khonithaek], 

team -^ [t^'im], *[thimi] 

can -^ [k^aen], *[khaeni] 

wool -^ [ul], *[uri] 

pool -^ [p^ul], *[p*'uri] 

For the same reason, i is epenthesized between consonants word- 
internally only if the first consonant is exploded. Thus we have 
epenthesized (i.e., bold) i's between consonants only in (38a), not in 
(38b). 

(38) a. disk -> [disikhi] 

list -> [lisit'^i] 

whisky -^ [wisik^'i] 

b. bond -> [p'ondi], *[p'onidi] 

mint -> [mint^i], ♦[minit^'i] 

tent -> [thenthi], *[thenithi] 

trunk -> [thiraT)khi], *[thiraT)ikhi] 

milk -^ [milkhi], *[mirikhi] 

As we saw in the last two examples of (37), i is not epenthesized 
after the final / which is not released. If the final segment is an r 
which is realized on the surface, however, i' is usually inserted as 
shown in (39a) since [r] cannot occur syllable-finally in Korean. But if 
[r] is not realized, i-epenthesis does not apply as in (39b). Thus the 
application of i-epenthesis depends on the realization of [r] on the 
surface. And the realization of [r] is a matter of conventional 
representaion, rather than a morphological or phonological issue. 

(39) a. tar -^ [t^ari] 

cork -^ [khorikhf] 
b. car -> [k^a] 
par -^ [pha] 

Based on the observation made so far, we can formulate a rule 
of i-epenthesis applying to loanwords as follows. 

(40) i-Epenthesis in Loanwords 

[[-(-release]] 
i" ^ / [[Vcont] J ] a 

By this rule, i epenthesizes to form a new syllable if the preceding 
consonant is a released stop (such as [t^] or [d], not [f]) or a fricative 
(such as s). 



Ahn: Vowel deletion and epenthesis: The vowel / 1 5 

Before we close this section, there remains a fact to be 
considered: When a syllable-final segment of a loanword is palatal, 
the epenthetic segment surfaces as [i], rather than [i]. 

(41) orange -^ [orenji] 
George Bush -> [cojibusi] 
French -^ [phirenc>ii] 
Apache -> [ap^'ac'^i] 

This sort of variation could also be handled by establishing a 
separate rule of /-epenthesis, where i is deleted after a palatal 
segment syllable-finally. 5 Note, however, that we have already made 
the rule of epenthesis (40) and that there is a rule of assimilation 
which is required anyway in Korean phonology. (The non-existant 
/f/ is usually realized as [p^] in Korean.) 

(42) /phlench/ Trench" 
philenchi Rule (40) 
philenc'^i Place Assimilation 
p^irenc'ii 1 -> r 

Thus, by making use of the already existing rule (40) and the rule of 
place assimilation, we can obtain not only the correct results but also 
simplicity of linguistic description. 

4. Concluding summary 

So far I have discussed two opposite phenomena: i-deletion and 
/-epenthesis. In the first part of the discussion, I showed how the 
controversy over the status of i can be resolved. Moreover, relating 
to the discussion over the issue of /-deletion, I reexaminined the so- 
called /i -irregular conjugation as well. Furthermore, in order to 
provide more explanatory feature representation and rule 
formulation, I employed the theories of underspecification and 
feature geometry. Here I adopted generally accepted concepts, rather 
than any specific stipulations. Finally, I briefly reexamined an 
alternative proposal and its problems with respect to these issues. 

In the second part of this paper, I discussed the /-epenthesis 
processes for two separate environments: Epenthesis by analogy and 
epenthesis for loanwords. In discussing the analogical epenthesis, I 
examined the constraining factors from morphological as well as 
phonological points of view. For the discussion of loanword 
epenthesis, I specified the phonological environment of the 
epenthesis rule. Then I d'scussed how the two rules epenthesizing /' 
and / can be merged as one. 



1 6 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

NOTES 

*This work was supported by a research grant from Kyung Hee 
University, Seoul, Korea. 

1 Through this paper, the symbols [a] and [i] refer to the mid 
back unrounded vowel [s] and the high back unrounded vowel [i], 
respectively. 

2 The other two i-deletion processes are formulated as follows. 

Casual i-deletion: mirror image 

V V 

V i 

/■ is truncated when meeting another vowel and the remaining vowel 
is lengthened. 

Intersonorant i-deletion 

fml 

1 ^ / 1 V/A ^ 1 i 

3 When we adopt more recent proposals by McCarthy (1988) or 
Clements (1989), these representations may appear differently. 
However, I will adopt a more general way of representation since 
taking any specific model is not our major concern here. 

'^ The underspecified representation for the adjectival h is not 
the only one which lacks a timing slot. For example, there are nouns 
which have a final 'floating' h in their underlying representations 
(Ahn 1986). This h does not appear as an individual form or before a 
vowel, but it appears on the surface before an obstruent by 
aspirating the following consonant. 

an pak /anh -t- pak/ [anp^ak] 'in and out' 

su pal /suh + pal/ [sup'^al] 'a drone' 

malikalak /mali + kalak/ [marik^arak] 'hair' 

Here I also assume that the final h lacks the timing slot since the rule 
of i-deletion applies to these examples as it does to /i-final adjectives: 
e.g. /mali + ilo/ [mariro] 'with hair'. Moreover, if any of these nouns 
has a vowel immediately before h, it takes -ka, rather than -/, as its 
subject marker, just as other vowel final nouns do. 

5 We may formulate a rule roughly as follows. 

/-epenthesis: i -> / C ] a 

[-Hpalatal] 



Ahn: Vowel deletion and epenthesis: The vowel i 1 7 

REFERENCES 

AHN, Sang-Cheol. 1985a. The interplay of phonology and morphology 

in Korean. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 
1985b. i-deletion in Korean revisited. Studies in Korean 

Linguistics, ed. by S. Kuno et al., 237-246. Cambridge, MA: 

Department of Linguistics, Harvard University. 
1986. On the nature of h in Korean. Studies in the Linguistic 

Sciences 16:2.1-13. 
1990a. Say-lo-wun ca-cil-i-lon-uy ceng-lip-ul wi-ha-ye 

(Features in nonlinear phonology). Linguistic Research 9.1-44. 
1991a. Mi-phyo-si-i-lon-kwa mo-um-phyo-ki (Underspecifi- 

cation and the Korean vowel system). To appear in the Tong- 

pang-hak-ci (Journal of Eastern Studies) Institute of Korean 

Studies, Yonsei University, Seoul. 
1991b. Types of node-dependent underspecification. Linguistic 

Journal of Korea, 445-462. 
1991c. An introduction to Korean phonology. Prepublication 

draft. Kyung Hee University, Seoul, Korea. 
Archangeli, Diana. 1984. Underspecification in Yawelmani phonology 

and morphology. MIT Ph.D. dissertation in Linguistics. 
, & Douglas Pulleyblank. 1986. The content and structure of 

phonological representations. Prepublication draft. University of 

Arizona and University of Southern California. 
Choi, Hyun-Pay. 1937. Wu-li mal-pon (The Grammar of Our 

Language). Seoul: Ceng-um-sa. 
Chomsky, Noam, & Morris Halle. 1968. The sound pattern of English. 

New York: Harper & Row. 
Chung, Kook. 1980. Neutralization in Korean: A functional view. 

University of Texas at Austin Ph.D. dissertation in Linguistics. 
Clements, George, N. 1985. The geometry of phonological features. 

Phonology Yearbook 2.225-25. 
1989. A unified set of features for consonants and vowels. Ms. 

Cornell University. 
, & Samuel J. Keyser. 1983. CV phonology. Cambridge, MA: MIT 

press. 
Goldsmith, John. 1976. Autosegmental phonology. MIT Ph.D. 

dissertation in Linguistics. Distributed by the Indiana University 

Linguistics Club. 
IVERSON, Gregory K. 1989. On the category supralaryngeal. Phonology 

6:2.285-303. 
Kim, Chin-W. 1968. The vowel system of Korean. Language 44.516- 

527. 



1 8 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

Kim, Kee-Ho. 1987. The phonological representation of distinctive 

features: Korean Consonantal Phonology. University of Iowa 

Ph.D. dissertation. 
1991. Kye-chung-cek ca-cil-swu-hyeng-to-ey iss-e-se-uy pi- 

phyo-si-wa cam-cay-phyo-si (Underspecification and 

nonspecification in feature hierarchy). Linguistic Journal of 

Korea 15.153-193. 
KiM, Young-Seok. 1984. Kwuk-e-uy cang-mo-um-ey elk-hin myet- 

ka-ci mwun-cey-tul (Several problems related to long vowels). 

Festschrift to Professor Lee Hey-Sook, 41-61. Seoul: Hanshin 

Publishing Co. 
KiM-RENAUD, Young-Key. 1973. Irregular verbs in Korean. Language 

Research 9.206-225. 
1974. Korean consonantal phonology. University of Hawaii Ph.D. 

dissertation in Linguistics. 

1975. On /j-deletion. Kwuk-E-Hak 3.45-64. 

1982. i-deletion in Korean. Linguistics in the Morning Calm, ed. 

by I.-S. Yang, 478-488. Seoul: Hanshin Publishing Co. 
Lee, Byung-Gun. 1976. Hyen-tay han-kwuk-e-uy sayng-seng um- 

wun-lon (Generative phonology of modern Korean). Seoul: Il-ci- 

sa. 
1979. I-lun-pa ke-kkwu-lo mek-i-ki swun-se (The so-called 

counterfeeding order.) Linguistic Journal of Korea 4.1-24. 
Lee, Pyung-Keun. 1979. Um-wun-hyen-sang-ey iss-e-se-uy Cey-yak 

(Constraints in phonological phenomena). Seoul: Tower press. 
Levin, Juliette. 1985. A theory of syllabicity. MIT Ph.D. dissertation in 

Linguistics. 
McCarthy, John. 1988. Feature geometry and dependency: A review. 

Phonetica 43.84-108. 
SaGEY, Elizabeth. 1986. The representation of features and relations 

in non-linear phonology. MIT Ph.D. dissertation in Linguistics. 
SOHN, Hyang-Sook. 1987. Underspecification in Korean phonology. 

University of Illinois at Urbana-Champaign Ph.D. dissertation in 

Linguistics. 



Studies in the Linguistic Sciences 
Volume 21. Number 2, Fall 1991 



NOTES ON SOME TESTS FOR SUBJECTHOOD IN KOREAN 

Euiyon Cho 

In this paper, I examine some syntactic phenomena of 
Korean which make reference to subjecthood: honorific si, 
reflexive casin, and plural copying. It will be shown that 
what controls these linguistic phenomena is not just the 
grammatical function 'subject'; semantic and pragmatic 
factors such as animateness, in addition to the grammatical 
function of subject, govern them. Arguments for this 
conclusion will be presented with discussions about the 
subjecthood of nominals in the so-called "double/multiple 
subject constructions". 

1. Introduction 

Korean has some grammatical phenomena which are seemingly 
sensitive to the function 'subject'. Rules governing the distribution of 
honorific si, reflexive casin, and plural copying are representative. 
Such grammatical phenomena have been, therefore, cited in Korean 
linguistics to justify one's claim for the subjecthood of a nominal in 
certain constructions when inquiries concerning which nominal in a 
sentence bears the grammatical relation subject arise. • Especially 
queries concerning which nominal is a subject in the so-called 
"double/multiple subject constructions", exemplified below, have 
forced Korean linguists to utilize the above-mentioned subject- 
referring grammatical phenomena. 2 

(l)Inho-ka kho-ka khu-ta. 
NOM nose-NOM big-Dec 
'It is Inho whose nose is big.' 

(2) Inho-ka apeci-ka khu-si-ta. 
faiher-NOM big-HON-Dec 
'It is Inho whose father is big.' 

For instance, in his study of Korean multiple nominative 
constructions within the framework of Relational Grammar. Youn 
(1990) uses such subject-referring phenomena in order to support 
his arguments for the subjecthood (final 7 -hood) of the first 
nominative nominal in sentences like (1) and for the subjecthood of 
the second nominative nominal in sentences like (2). 



20 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

In this study, I shall show the following regarding the 
subjecthood of a nominal in sentences like (1): a non-term R-sign 
such as 'Focus' as well as a subject is able to serve as a target of 
honorific si and antecedent of reflexive casin if the grammatical 
subject fails to refer to an animate entity; what controls plural 
copying is not the plural marker tul on a subject nominal but the 
semantic plurality of the subject nominal. At the same time, contrary 
to Youn's claim for the subjecthood of the first nominative nominal in 
sentences like (1), it will be argued that in sentences like (1), which 
will be called "inalienable double subject constructions" (IDSC), the 
second nominative nominal is the (final) subject. This will pave the 
way of giving a unified account of the two types of double subject 
constructions. 

In section 2, Youn's work on IDSCs will be presented with 
discussions about honorific si, reflexive casin, and plural copying. 
This will be followed by my arguments for the subjecthood of the 
second nominative nominal in IDSCs. Section 3 concludes the paper 
discussing some consequences of this study. 

2. Searching for Subject in IDSCs 

2.1. Honorific si 

It is commonly accepted in Korean linguistics that the 
appearance of honorific affix si in the predicate of a sentence implies 
that the speaker shows respect to the subject referent. The following 
sentences show that the grammatical property of honorific si is 
linked to a subject nominal: the property of honorific si anchors to 
the subject nominals apeci 'father' and halapeci 'grandfather' in (3) 
and (4), respectively. 

(3) apeci-ka Kim sensayngnim-ul mana-si-ess-ta. 
father-NOM teacher-ACC meet-HON-Past-Dec 
'Father (Hon) met teacher Kim.' 

(4) halapeci-ka cap-hi-si-ess-ta. 
grand father-NOM calch-Pass-HON-Past-Dec 
'Grandfather (Hon) was arrested.' 

Thus honorific si has been called "subject referent honorific." 

On the basis of the examples such as (3) and (4), Youn (1990) 
proposes the following rule governing honorific si. 

(5) A final / controls S(ubject) H(onorification). 

With the above rule of honorific si, Youn attempts to show that the 
first nominative nominal is a (final) subject in IDSCs. In general, as 
he shows, the first nominative nominal in IDSCs appears to be a 
target for honorification by means of honorific si. For example, what 
is honorified by the linguistic element si in the following sentences 



E. Cho: Notes on some tests for subjecthood in Korean 2 1 

are the referents of the first nominative nominals enemin 'mother' 
and apenim 'father'. 

(6) emenim-i nwun-i khu-si-ta. 
mother-NOM eyes-NOM big-HON-Dec 
'It is mother whose eyes are big.' 

(7) apenim-i phal-i pwuleci-si-ess-ta. 
father-NOM arm-NOM become broken-HON-Past-Dec 
'It is father whose arm was broken.' 

Therefore, he concludes that the first nominative-marked nominal is 
the subject of IDSCs. 

In the following, we shall first see that it is not the case that 
only subject nominals control honorific si. After that it will be shown 
that sentences like (6) and (7) belong to a case where a non-term R- 
sign — nominal in terms of Relational Grammar — controls honorific 
si. That is, it will be argued that in the above sentences, the second 
nominative nominals are subjects although the referents of the 
nominals are not interpreted as being honorified by the honorific si.^ 

That a non-term R-sign controls honorific si is shown in the data 
(8) and (9) below, taken from Cho (1988) and Kim (1990). 
respectively. What is honorified by the presence of si in (8) and (9) 
are the referents of the nominals kyocang sensayngnim 'principal' 
and halmeni 'grandmother', which are not the subjects of the 
sentences; the nominal 'grandmother' is a part of the topic phrase 
halmeni-uy sangsi-eyse-nun 'in Grandmother's life' while the 
nominal 'principal' is part of the postpositional phrase cang 
sensayng-nim-ulopwute 'from the Principal'. 

(8) kyocang sensayngnim-ulopwute cisisahang-i iss-usi- 
Principal teacher-from message-NOM exist-HON- 
kkeyss-upni-ta. 

Future-HON-Dec 

'There will be a message from the Principal.' 

(9) halmeny-uy sayngsi-eyse-nun samsiptay-ka kacang 
grandmothcr-Poss life-Loc-Top thirties-NOM most 
hayngpokha-si-ess-ta. 

happy-HON-Past-Dec 

'In Grandmother's life, (her) thirties was the happiest.' 

That is, in (8) and (9), the nominative case-marked subject nominals 
cisisahang 'message' and samsiptay 'thirties' do not serve as targets 
of honorific si. As the above examples show, nominals bearing a non- 
term R-sign can be the target for honorification by honorific si. What 
we need to note in the above examples is the semantic relation 
between the target of honorification and the subject nominal: 



2 2 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 

kyocangsensayngnim 'Principal' cisisahang 'message'; halmeni 
'grandmother' and samsiptay 'thirties'. It is a possessor-and- 
possessee relation (in the interpretation of the above sentences.) This 
relationship is the same as that between the first and second 
nominative nominals in IDSCs. I conjecture that the reason why 
graimnatical subjects fail to serve as the target of honorification in 
the above sentences is that the subject referents are not human. 

My analysis of honorific si honorification is as follows: si is 
grammatically linked only to a subject (final 1).'* But if the subject 
nominal denotes non-personage and the denotation of the subject 
nominal turns out to be a possessee of the person(s) referred to by a 
non-term bearing nominal, then in order for a speaker to show his 
deferential attitude to the possessor, the honorific si is used in order 
to honorify the possessor by linking the possessee nominal (subject) 
with the honorific si. Seen from the perspective of language change, 
it would be said that by showing his deferential attitude to the 
referent of subject nominal which is grammatically linked to 
honorific si, the speaker shows respect indirectly to the person who 
possesses it. However, what the speaker eventually wants to achieve 
by means of the linguistic element si is such a communicative context 
is to show respect to the possessor (a person), not to the possessee (a 
non-person). This kind of use of honorific si has been 
conventionalized. Thus, the appearance of honorific si in 
constructions like (8) and (9) as well as in IDSCs is quickly 
interpreted as being connected with a possessor denoted by a non- 
term R-sign bearing nominal. 

This explanation of honorific si phenomena applies equally to 
the interpretation of the honorific si in IDSCs. Let us consider the IDS 
sentence (6) repeated below: 

(6) emenim-i nwun-i khu-si-ta. 
mother-NOM eyes-NOM big-HON-Dec 
'It is mother whose eyes are big.' 

I claim that the second nominative nominal nwun 'eyes' is the 
subject of the sentence (6), which is predicated of by the predicate 
khu 'big'. Since the speaker knows that the honorific si is 
grammatically linked to the subject nominal nwun 'eyes' denoting a 
possession of a person, by using the honorific si, the speaker intends 
to show his deferential attitude to the possessor of eyes. Mother. 
Thus, the appearance of the honorific si in IDSCs like (6) conveys the 
implication that the speaker shows a deferential attitude to the 
possessor. That is, in IDSCs like (6), the referent of the first 
nominative nominal can be honorified by the speaker's use of 
honorific si not because it is the subject of IDSC but because its 
denotation is a human possessor of the subject referent (= the second 



E. Cho: Notes on some tests for subjecthood in Korean 2 3 

nominative nominal) which is grammatically linked to honorific si. In 
short, in Modern Korean, it is no longer true that what is honorified 
by honorific si is a subject. But a non-term R-sign as well as subject 
can be connected to the effect of honorific si only if what they denote 
is interpreted as a possessor of the referent (non-personage) of the 
subject nominal. 

Finally to further support my claim that the second nominative 
in IDSCs is a subject, let me cite a situation in which only the referent 
of second nominative nominal in IDSCs can be honorified by honorific 
si. Suppose that in a science fiction story computers are personified 
so that they think, speak, operate machines, and control the 
relationships between computers. In such a context, the use of 
honorific si as shown in the following utterance, has a communicative 
effect that the speaker of (10) shows deferential attitude to the RAM. 

(10) i kompwute-ka raym-i khu-si-ta. 
Ihis compuler-NOM ram-NOM big-HON-Dec 
'It is this computer whose ram is big.' 

To conclude, honorific si is grammatically linked to a subject 
nominal but the speaker's use of honorific si for the purpose of 
honorifying the person(s) referred to by a nominal does not always 
cover the cases in which the person is the subject referent: The 
referent(s) referred to by a non-term R-sign can be honorified by the 
speaker's use of honorific si if the nominal bearing the non-term R- 
sign is interpreted as a possessor of an inanimate subject referent. 

2.2. The Reflexive casin^ 

According to Youn (1990), the interpretation of reflexive casin 
'self is dependent on the nominal bearing the subject relation as the 
following examples show: 

(1 1 ) Chelswu-ka Swuni-lul casin-uy chayk-ul cwu-ess-ta. 
Chelswu-NOM Swuni-ACC self-Poss office-Loc meet-Past-Dec 
'Chelswuj met Sunij selfsj/^j office.' 

(12) Chelswu-ka Swuni-eykey casin-uy chayk-ul cwu-ess-ta. 
Chelswu-NOM Swuni-DAT self-Poss book-ACC give-Pasl-Dcc 
'Chelswuj gave Sunij selfsi/«j book.' 

Since Youn argues that the first nominative nominal in IDSCs is the 
subject (final /), it is predicted under his analysis that only the first 
nominative nominal can antecede the reflexive casin. The following 
IDSCs show that this prediction is borne out: 

(13) Chelswu-ka elkwul-i casin-uy kotnong-ulo ilkuleci- 
Chelswu-NOM facc-NOM self-Poss pain-wiih become iwisled- 
ess-ta. 

Past-Dec 
'It was Chelswuj whose face was twisted with selfs, pain." 



2 4 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

However, for a subject nominal to serve as antecedent of 
reflexive casin, it should have the semantic or pragmatic property of 
'being animate'. For example, if an IDSC has two inanimate nominals 
as in (14), no matter whether the first or second nominative nominal 
is the subject of (14), neither of them is eligible as antecedent of the 
reflexive casin. 

(14) i pangi-i pyekj-i casin*i/*j-uy nolyek-ulo 
this room-NOM wall-NOM self-Poss effort-with 
twukkeyp-ta. 

thick-Dec 

'It is this room whose walls are thick with selfs effort.' 

However, if we personify the referent of the second nominative 
nominal, then the reflexive casin can be coreferential with it. This 
example serves as evidence for the subjecthood of the second 
nominative nominal in IDSCs. Suppose that in a fairy tale, walls have 
ears to hear and mouths to speak. In addition, they have a magical 
power to make themselves thick or thin whereas rooms do not have 
such abilities although they are possessors of the walls. In this 
speech context, the second nominative nominal of the sentence (14) 
serves as the antecedent of casin while the first one fails to. This 
clearly shows that the second nominative nominal is the subject of 
IDSCs, and that for a subject nominal to be coreferential with the 
reflexive casin, it needs to be interpreted as animate. 

The question which naturally arises at this point is thus: in most 
IDSCs, what makes the first nominative nominal a possible 
antecedent of reflexive casinl Is it because the first nominative 
nominal in IDSCs bears the grammatical relation of subject like the 
second one? The answer to this question is no. This is what I shall 
show in what follows. 

We shall see an example showing that a nominal within a topic 
phrase in a non-gap topic construction, i.e., a non-subject nominal 
bearing a non-term R-sign serves as antecedent of reflexive casin. 
Let us look at the following sentence (15): 

(15) [[Inho-uy Seoul saynghwal-un] [ton-i casin-uy 
Inho-Poss living-Top money-NOM self-Poss 
checi-eyse kacang kun eleywem-i-ess-ta]] 
situation-Loc most big difficulty-be-Past-Dec 

'In Inhoi's life in Seoul, money was the greatest difficulty in 
seif'Si situation.' 

In (15), the nominal ton 'money' is undoubtedly the subject of the 
sentence. But it cannot serve as antecedent of the reflexive casin 
because it is inanimate. What is eligible as a possible antecedent of 
casin in (15) turns out to be the nominal Inho, a person, in a topic 



E. Cho: Notes on some tests for subjecthood in Korean 25 

phrase. What this phenomenon tells us is that reflexive casin finds its 
antecedent in an extra-sentential constituent if it fails to have the 
subject as antecedent because the subject is inanimate. But the 
nominal in an extra-sentential constituent needs to be interpreted as 
a possessor of the inanimate subject referent. What is happening in 
IDSCs like (13), repeated below, is the same as that of (15) as far as 
the phenomenon of reflexive casin is concerned. 

(13) Chelswu-ka elkwul-i casin-uy kotnong-ulo ilkuleci- 

Chelswu-NOM face-NOM self-Poss pain-with become twisted- 

ess-ta. 

Pasi-Dec 

'It was Chelswuj whose face was twisted with selfsj pain.' 

In (13), the second nominative nominal elkwul 'face' fails to serve as 
antecedent of reflexive casin because the nominal lacks the property 
'being animate' although it is the subject of the sentence (13). Thus, 
the first nominative nominal naturally becomes available as a 
possible antecedent of the reflexive casin because it does not only 
bear a non-term R-sign 'Focus' but also is interpreted as a human 
possessor of the subject referent 'face'. 

If we assume the structure of sentences like (13)-(15) as 
[[focus/topic] [s]] and treat the initial element as an extra-sentential 
constituent, then the above reflexive phenomena shown in (13)-(15) 
lead us to the conclusion that a nominal in a topic/focus constituent 
can serve as antecedent of reflexive casin if it is interpreted as an 
animate possessor of the subject referent when the subject nominal 
in [s] fails to serve as antecedent of it because it lacks the property of 
'being animate'. Thus, not every nominal which serves as antecedent 
of reflexive casin can be taken as a subject. 

2.3. Plural copying 

Korean has a plural marker tul. One of the interesting facts 
concerning the distribution of the plural marker is that it appears 
with categories other than the noun it pluralizes. The following 
examples show this: 

(16) haksayng-tul-i nuckey(tul) o-ess-ta. 
siudcnts-PL-NOM late (PL) come-Past-Dec 
'Students have come late.' 

(17) chayk-tul-i kapang sok-ey-tul iss-ess-ta. 
book-PL-NOM bag inside-LOC-PI. exisl-Past-Dcc 
'Books are in the bag.' 

Based on the above data as well as other ones, Youn proposes a 
condition on plural copying as in the following: 

(18) A final / can control Plural Copying. 



2 6 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

Since he assumes that the first nominative nominal is the subject 
(final 7) of the IDSCs, he predicts that plural marker tul attaches to 
categories other than the first nominative nominal when it pluralizes. 
The following examples are cited in support of the first nominative 
nominal is the subject of the IDSCs: 

(19) a. Chelswu-ka elkwu-i manhi-(*tul) yewi-ess-ta. 

Chelswu-NOM face-NOM much-(PL) become thin-Past-Dec 
'Chelswu's face became very thin.' 

(Youn's 51 in Ch. 2) 

b. ai-tul-i elkwul-i manhi-(tul) yewi-ess-ta. 

child-PL-NOM face-NOM much-(PL) become thin-Past-Dec 
'The children's faces became very thin.' 

(20) a. ku uyca-ka tungpati-ka manhi-(*tul) hyeeci-ess-ta. 

the chair-NOM back-NOM much-(PL) wear-Past-Dec 

'The back of the upholstered chair was very worn.' 

(Youn's 53) 

b. ku uyca-tul-i tungpati-ka manhi-(tul) hyeeci- 
the chair-PL-NOM back-NOM much-(PL) wear- 
ess-ta. 
Past-Dec 
'The backs of the upholstered chairs were very worn.' 

With the data like (19)-(20), Youn uses the plural copying 
phenomenon as a test for the subjecthood of a nominal in IDSCs. 

In the ensuing discussion, first of all, I would like to show that 
what really controls plural copying is not the plural marker tul 
attached to a subject nominal but the concept of plurality of a subject 
in the semantic interpretation of a sentence. Let us consider the 
following sentences (21) and (22), which have two nominative case- 
marked nominals and the first nominative nominal is undoubtedly 
interpreted as a possessor of the referent of the second one. 

(21) Inho-ka (yangccok) nwun-i manhi-tul pwu-ess-ta. 
Inho-NOM (both) eyes-NOM much-PL swell-Past-Dec 
'It is Inho whose eyes were very swollen.' 

(22) i ai-ka sin-i manhi-tul tahl-ess-ta. 
this child-NOM shoe-NOM much-PL wear-Past-Dec 
'This child's shoes were very worn down.' 

As the above sentences show, there is no plural marker attached to 
the two nominals: only the adverb man hi 'much' has the plural 
marker. But the sentences are grammatical, though. What is it that 
granmiatically licenses the plural marker tul attached to the adverb 
manhi in (21) and (22)? The appearance of the plural marker in the 
above sentences cannot be attributed to the first nominative 



E. Cho: Notes on some tests for subjecthood in Korean 27 

nominals Inho 'Inho' and / ai 'this child' since they are singular in 
number. The appearance of tul on an adverbial constituent in the 
above sentences is attributed to the plurality of the second 
nominative nominal derived from their denotation nwun '(both) 
eyes' and sin 'shoe(s)', even though no plural marker tul is attached 
to them. 

Thus, viewed from this perspective, the appearance of the plural 
marker tul on the adverb manhi 'much' in the IDSC (19b), repeated 
below, is attributed to the plurality of the things referred to by the 
second nominative nominal elkwul 'face', which is possessed by the 
referents of ai-tul 'children'. 

(19) b. ai-tul-i elkwul-i manhi-(tul) yewi-ess-ta. 

child-PL-NOM face-NOM much-(PL) become thin-Past-Dec 
'The children's faces became very thin.' 

In other words, even though there is no plural marker attached to 
the second nominative nominal elkwul 'face', since in the linguistic 
context of (19b) it denotes the faces (of the children), the second 
nominative nominal elkwul 'face' is interpreted as plural. Thus, the 
plural marker tul attached to the adverb manhi 'much' owes its 
existence to the plurality of the things referred to by the second 
nominative nominal elkwul 'face'. 

Thus we are led to draw the conclusion that what controls the 
appearance of plural marker tul on non-nominal syntactic categories 
in a sentence is the semantic plurality of a subject nominal. 

3. Conclusion 

I have attempted to show that what controls the phenomena of 
honorific si, reflexive casin, and plural copying is not just the 
grammatical relation 'subject'. It was shown that reflexive casin and 
honorific si are vulnerable to semantic or pragmatic factors such as 
animateness. On the other hand, it was argued that what controls 
plural copying is not the plural marker tul but the semantic plurality 
of a subject nominal. 

On the basis of these findings, 1 have argued that the second 
nominative nominal in IDSCs is a subject. This will make the 
structure of IDSCs no different from that of ADSCs since the second 
nominative nominal in ADSCs has been claimed to be a subject (see 
Youn 1990: §2.2). This is certainly a welcome consequence for the 
future study of double/multiple nominative constructions if we aim 
to provide a unified account of their syntactic structure. 



28 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

NOTES 

1 In this paper, I use the term "subject" neither in the sense of 
Relational Grammar in which the notion is taken as primitive of 
syntactic theory, nor in the sense of other syntactic theories such as 
Government and Binding in which it is taken as a derivative one. But, 
irrespective of one"s syntactic theory, what is found as a subject by 
heuristic tests in a language can be incorporated in the grammar of 
the language a linguist aims to design. 

2 These constructions have been given various names such as 
"double/multiple nominative constructions"'. Although I use the term 
"double subject constructions" I do not assume that there are two 
(final) subjects in a sentence like (1) and (2) in the main text. Those 
readers who are interested in the works on these constructions are 
referred to the works quoted in Youn (1990). 

3 This argument has been already made in works like Yoon 
(1987) and Cho (1988: Ch. 2). 

'♦ There is evidence for the claim that the honorific si is 
granmiatically linked to only a subject nominal even though there 
are data which show that the target of honorific si can be a non-term 
R-sign nominal. In Cho (1991), I presented data whose syntactic 
structure is the same as that of (9) as shown below: 

(i) halmeni-uy sangcon-si-nun halapeci-ka kacang 

grandmother-Poss life-time-Top grandfather-NOM most 

hayngpokha-si-ess-ta. 
happy-HON-Past-Dec 

'When Grandmother was alive. Grandfather was the 
happiest." 

If the topic phrase or focus phrase as well as the subject nominal in a 
non-gap topic construction can be grammatically linked to the 
honorific si, then it is predicted that the interpretation of the 
honorification of the referents of the two nominals halmeni 
■grandmother" in a topic phrase and halapeci 'grandfather'. However, 
in (i), only the subject nominal 'grandfather' is interpreted to be 
connected to the pragmatic effect of honorific si. 

5 Korean has another reflexive element caki whose usage is 
different from the reflexive casin. For the grammatical and pragmatic 
aspects of caki, the reader is referred to O'Grady (1987). 



E. Cho: Notes on some tests for subjecthood in Korean 29 

REFERENCES 

Cho, E. 1988. Some interactions of grammar and pragmatics in 

Korean. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 
1991. A reanalysis of some counter-examples against the 

existence of an AGR element in Korean (in Korean), MS, Dongguk 

University. 
KIM, Y-J. 1990. The syntax and semantics of Korean case: The 

interactions between lexical and syntactic levels of 

representation. Harvard University Ph.D. dissertation in 

Linguistics. 
O'Grady, W. 1987. The interpretation of Korean anaphora: The role 

and representation of grammatical relations. Language 63:2.251- 

277. 
PERLMUTTER, D. (ed.) 1983. Studies in Relational Grammar 1. Chicago: 

University of Chicago Press. 
YOON, J. 1987. Some queries concerning the syntax of multiple subject 

constructions in Korean. Harvard Studies In Korean Linguistics 

n, ed. by S. Kuno et al., 138-162. 
YOUN, C. 1990. A relational analysis of Korean multiple nominative 

constructions. University of New York at Buffalo Ph.D. 

dissertation in Linguistics. 



Studies in the Lipsuistic Scienggs 

Volume 21. Number 2. Fall 1991 



THE ACQUISITION OF ENGLISH REFLEXIVES BY KOREAN ESL 
LEARNERS 

Seikyung Cho 

This study investigates the acquisition of English 
reflexives by adult Korean learners of English. Wexler and 
Manzini (1987) propose the Governing Category Parameter 
for which five different values have been posited to account 
for the differences in languages with respect to the binding 
pattern of reflexives. According to this proposal, English 
represents the narrowest (unmarked) value and Korean 
represents the widest (marked) one. Thus, by inspecting 
how Korean learners acquire English reflexives, we can 
address the issue of the availability of the Subset Principle 
in second language acquisition. This investigation also 
answers the question of what type of value of the 
parameter second language learners choose: The transfer of 
the LI value, the acquisition of the correct L2 value, or the 
choice of an intermediate value which is distinct from both 
LI and L2. Two different types of tests were administered 
to Korean learners of English: A reading comprehension test, 
and a sentence-picture matching test. Results from these 
tests indicate that the Subset Principle is not available to 
adult second language learners. However, the successful 
acquisition of English reflexives by advanced subjects 
suggests that the resetting of a parameter is possible in 
second language acquisition even in the absence of relevant 
positive evidence. To account for this, I suggest that the 
application of the subtle solutions like 'indirect negative 
evidence' and 'indirect positive evidence' is available to 
second language learners. 

I. Universal Grammar and the Subset Principle 

During the past three decades, we have seen that such 
approaches as Contrastive Analysis and Creative Construction fail to 
address the traditional issues of second language acquisition 
research. Consequently, second language research focuses on 
developing a single theory which can integrate what was found in 
the Contrastive Analysis model and the Creative Construction model. 
Recently, one of most promising developments in this direction has 



3 2 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 

been the theory of Universal Grammar. This is the approach 
developed primarily in works by Chomsky as realized in the 
syntactic theory of Government and Binding (1981). According to this 
theory, children are born with a very specific set of cognitive 
structures, called Universal Grammar, which controls and guides the 
way children handle all the aspects of their language. Universal 
Grammar defines possible human languages and consists of a set of 
'principles' which constrain the types of hypotheses that children can 
entertain about their language. Along with these principles. Universal 
Grammar also makes available to children a set of 'parameters' which 
cover the variations among languages. Children set the parameters 
differently for different languages, thus creating the characteristic 
grammar for that language. 

Within this model, the children's task is straightforward. They 
only have to discover where Universal Grammar differs from the 
target language. In areas where Universal Grammar and the target 
language match, children simply apply UG principles to the language. 
Since such principles are innately built-in to children's minds, 
nothing has to be learned. When children are faced with a structure 
which varies from one language to another, they set the parameter to 
the appropriate value of the target language. 

This theory of language acquisition naturally leads us to the 
question of whether or not such a theory would account for second 
language acquisition. Despite the fact that Universal Grammar itself 
was not intended to account for second language acquisition, recently 
a lot of empirical evidence has been accumulated in support of the 
UG hypothesis in second language acquisition. For example, Flynn 
(1987) investigated the acquisition of the Head Parameter by adult 
second language learners and claimed that she found evidence 
indicating that second language learners actually have access to UG 
principles, and that parameter-resetting is possible in second 
language acquisition. In similar studies with adults learning a second 
language. White (1985a, 1987) investigated the accessability of UG 
principles in second language acquisition and reported similar 
results. 

Another important aspect of UG theory is related to the type of 
relevant evidence that language learners need to fix the parameter. 
UG theory rejects the use of negative evidence (correction, for 
example) as a reliable source of evidence to disconfirm inappropriate 
hypotheses. I Therefore, in this model, children must be able to learn 
their language solely on the basis of positive evidence. If this is true, 
how do language learners reject incorrect hypotheses which require 
direct negative evidence? From the UG point of view, language 
learners can reject inappropriate hypotheses by virtue of UG 
principles that constrain how and where the learners may apply 



S. Cho: The acquisition of English reflexives by Korean ESL learners 33 

their hypotheses. However, even with the operation of UG principles, 
there still remains a possibility that children arrive at incorrect 
hypotheses which must be disconfirmed by negative evidence. 

To overcome this problem, a proposal known as the Subset 
Principle (Wexler & Manzini 1987) has been proposed as a learning 
principle. The idea is that the order of hypotheses that children can 
entertain is constrained by a certain markedness hierarchy. 
According to this proposal, children start out with the unmarked 
value of a parameter. They never arrive at the marked value of a 
parameter unless positive evidence from the target language 
warrants it. In this way, this principle guarantees that children can 
proceed from the unmarked value of a parameter to the marked one 
without benefit of negative evidence. 

In order for the Subset Principle to apply, we need a parameter 
whose two values generate sentences in a subset/superset relation, a 
requirement referred to as the 'Subset Condition' (Wexler & Manzini 
1987). This is a situation where sentences generated by one value of 
a parameter (superset) are not only compatible with sentences 
generated by the other value of the parameter (subset), but with 
additional sentences as shown in the following figure. 




Figure 1. Subset/superset relation 

Here, the wider grammar that yields the Y sentences includes the X 
sentences generated by the narrower grammar. Then, Y sentences 
are compatible with two grammars; the grammar that yields X 
sentences and the grammar that yields Y sentences. In this case, X is 
a proper subset of Y and Y is a superset, and thus they meet the 
Subset Condition. 

2. The Governing Category Parameter 

One example of a parameter of this type is the Governing 
Category Parameter proposed by Wexler and Manzini (1987), which 
has five different values, each representing one type of language. 
Wexler and Manzini argue that languages can be divided into five 
types depending on how far away the reflexive can be from its 
antecedent and that these differences among languages can be 
described by a subset relationship as presented in the following 
figure. 



3 4 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 




Figure 2. Governing Category parameter 

English is a type (a) language, which is the most restrictive 
language in that it allows only the closest NP to the reflexive to be its 
antecedent regardless of whether the sentence is finite or non-finite. 
Therefore, in the following example, only 'Fred' can serve as 
antecedent for the reflexive 'himself in a type (a) language. 

( 1 ) Mike recalled that John wanted Fred to paint himself. 

By contrast, in a type (e) language like Japanese or Korean, the 
governing category is a root clause sentence (meaning the whole 
sentence). 

(2) Chulsoo-nun Kinam-ika caki-lul kkociputtako malhatta. 

Nom Nom self-Acc pinch Comp say-Past 

'Chulsoo said that Kinam pinched himself.' 
'Chulsoo said that Kinam pinched him.' 

In (2), both NPs can serve as an antecedent for the reflexive. Thus, 
we can see that English and Korean occupy opposite ends of the 
hierarchy of the GCP values in figure 2 which Wexler and Manzini 
postulate. That is, English is a type (a) language, which is the most 
restrictive in that it allows only the closest NP to the reflexive to be 
its antecedent. On the other hand, Korean is a type (e) language, 
which is the least restrictive language in that any NP in the sentence 
can serve as an antecedent for the reflexive. 

Thus, we are in a situation where English and Korean grammars 
happen to yield sentences which are in a superset/subset relation; 
English grammar which generates the same subset sentences, and 
Korean grammar which generates not only subset sentences of 
English type, but also additional ones by allowing non-local NPs to be 
antecedent for the reflexive. In other words, Korean grammar allows 



S. Cho: The acquisition of English reflexives by Korean ESL learners 3 5 

other non-local NPs as well as every NP that is allowed in English. 
Thus, the parametric value of the English reflexive and that of the 
Korean reflexive in this kind of relationship meet the Subset 
Condition. 

3. Analyses of previous L2 studies on the GCP 

3.1. Hirakawa's study (1990) 

Hirakawa investigated the effects of the GCP (Wexler & Manzini, 
1987) in the acquisition of English by four groups of Japanese high 
school students: Grade 10, grade 11, grade 12, and grade 13.2 
Japanese is like Korean in that the reflexive can be bound either 
locally or non-locally. Therefore, it is another widest-type of 
language with respect to the GCP. The test sentences included five 
types as can be seen in (3). 

(3) Type A: Two-clause sentence (finite) 

John said that Bill hit himself. 
Type B: Three-clause sentence (finite) 

Mary remembers that Jane said that Ellis blamed 

herself 
Type C: Two-clause sentence (infinite) 

Mary asked Ann to introduce herself. 
Type D: Three-clause sentence (infinite) 

Ann knows that Mary told Jane not to hate herself. 
Type E: One-clause sentence 

Bob talks to Paul about himself. 

A total of 25 sentences (5 each type) were presented. A multiple- 
choice grammaticality judgment task was conducted in which 
subjects were given a sentence with five alternatives. Subjects were 
asked to indicate who 'himself or 'herself referred to in each 
sentence by circling one of five choices as in the following example in 
(4). 

(4) John said that Bill hit himself 

a. John 

b. Bill 

c. either John or Bill 

d. someone else 

e. don't know 

If they could not find an antecedent in the alternatives, they were to 
circle 'someone' and to write down to whom it referred in the blank 
space as in (4d). When they did not understand the sentence, they 
were to circle 'don't know'. 

Hirakawa's results show a high incidence of non-local response 
for the finite clause. The subjects were transferring their LI value of 



36 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

the GCP. Hirakawa's interpretation of these results is that since the 
subjects were relatively less advanced than those of Broselow and 
Finer (1991), it may be that learners move from the widest value to 
a narrower value as they become more proficient in English. In other 
words, at a less advanced stage, L2 learners transfer their LI value, 
but as learning progresses, they gradually reset the GCP to the value 
of their target language. However, Hirakawa did not give any answer 
to the question of exactly how learners accomplish the resetting of 
the parameter. 

A total of 65 subjects were tested, and 5 types of sentences with 
an equal number of test items each were employed. Hirakawa 
conducted a screening test to investigate the subjects' metalinguistic 
knowledge to perform this kind of task. Background information of 
subjects was provided as to number of years of English study, 
amount of exposure to English, age when English study began, and 
method of English instruction they were given. She provided five 
choices for each test item to avoid the possibility of response bias of 
subjects. She compared the results of the study with Japanese and 
English native speakers' judgments. This is useful since this study is 
to observe the pattern of learners' judgments as they approach those 
of native speakers'. 

However, this study is methodologically not desirable in several 
aspects. First, the inclusion of 'someone else' option creates some 
confusion. She said that she included this option because in Japanese 
the reflexive 'zibun' can be interpreted as having the speaker as its 
antecedent. So she wanted to see if subjects might made this 
interpretation in English. However, as far as this grammaticality 
judgment task is concerned, only one sentence was given to subjects 
rather than a paragraph or conversation of some length where some 
context is given. Therefore, it is doubtful of whether subjects will 
really choose this option. Actually, Hirakawa's results show that 
almost none of her subjects chose this option. 

The second problem with this study is that it does not contain 
distractor items. One disadvantage of the inclusion of dummy items 
is that it often results in lengthy tasks which would tax the patience 
of subjects. On the other hand, however, the non-inclusion of dummy 
items often results in perturbation due to learning effect and fatigue. 
In addition, this kind of test is vulnerable to the recognition of 
grammatical focus by subjects. Therefore, as Birdsong (1989) points 
out, it may be desirable to prevent subjects from recognizing the 
grammatical focus of the test. The focus can be disguised by the 
inclusion of distractor items on the test. 

The third problem of this study is that only one type of test was 
provided, a multiple-choice grammaticality judgment task. The 



S. Cho: The acquisition of English reflexives by Korean ESL learners 3 7 

purpose of this study is to indicate the L2 learners' stage of 
development. However, as many researchers note, a grammaticality 
judgment task alone may not be a very sensitive indicator of 
learners' competence in their L2. Therefore, it may be that 
grammaticality judgment tasks serve as a valid measure of L2 
knowledge only when they are compared with data derived from 
other subtle and varied methods of empirical data-gathering. 
Judgment should be validated by the comparison with data from 
other measures on the same test items and subjects. 

3.2. Broselow and Finer's study (1991) 

Broselow and Finer conducted a study to investigate the 
acquisition of English reflexives by non-native speakers of English. ^ 
They tested 97 subjects: 30 native speakers of Korean, 37 native 
speakers of Japanese, and 30 native speakers of Hindi. The subjects 
were students at the State University of Stony Brook, or their friends 
and spouses. Their reported level was high intermediate. A picture 
identification method was employed. The subjects simultaneously 
heard and read 24 English sentences containing reflexives as can be 
seen in the following. 

expects 
(5) a. Mr. Fat Mr. Thin to paint himself, 

tells 

believes 

b. Mr. Fat that Mr. Thin will paint himself. 

thinks 

threatens 

c. Mr. Fat Mr. Thin to paint himself. 

promises 

Subjects were asked to choose one picture out of four which was the 
most appropriate for the sentence according to their judgment. Here 
are Broselow and Finer's results. 

Table 1 Mean percentages of responses 





Finite 
L NL L&NfL NL* L* 


L 


Non-finite 
NL L&NL NL* L* 


Korean 

Japanese 

Hindi 


97% 2% 1% 
88% 8% 2% 2% 
100% 


88% 
70% 
96% 


7% 1% 2% 3% 
20% 2% 4% 5% 
2% 1% 


L: Local 


NL: Non-local 







38 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

These results show that the Japanese and Korean subjects chose local 
antecedents a great deal more frequently than non-local antecedents 
in sentences like (5b). On the other hand, they took local antecedents 
comparatively more frequently in sentences like (5a) than they did 
in sentences like (5b). Therefore, it seems that the Korean and 
Japanese subjects chose the English setting of the GCP value in 
sentences like (5a), and their native language's setting in sentences 
like (5b). Browselow and Finer interpreted these results as showing 
that the subjects treated the English reflexive as if it occupied 
position (c) or (d) on the hierarchy of the GCP values in figure 2. To 
put it another way, Broselow and Finer argue that L2 learners start 
out with the parameter settings of their native languages and then 
move in stages through the intermediate settings in the direction of 
target language settings. 

However, an alternative interpretation for these results is 
possible. First, although Broselow and Finer claim that their Korean 
and Japanese subjects prefer non-local antecedents in non-finite 
sentences like (5a), as table 2 shows, the majority of the subjects 
chose local antecedents: Koreans 88%, Japanese 70%. Only 7% of 
Koreans and 20% of Japanese chose non-local antecedents in non- 
finite sentences. Thus, it is too powerful a claim to say that Korean 
and Japanese subjects chose non-local antecedents in non-finite 
clauses. 

Secondly, Broselow and Finer argue that the more frequent 
choice of non-local antecedents in non-finite sentences is not 
traceable to their LI, since the subjects made such a distinction even 
though their LI does not have a grammatical distinction between 
finite and infinite sentences. However, if we assume that the subjects 
have not mastered the distinction between finite and non-finite 
sentences in English, they would treat a non-finite clause as a simple 
sentence rather than a complex one. And in fact, in Korean, there is a 
strong tendency to prefer a non-local antecedent over a local one 
(Lee, 1984). Then, the choice of non-local antecedents in non-finite 
sentences made by Korean and Japanese subjects can be attributed to 
the effect of transfer from their native language. 

Methodologically, this study is not desirable in several ways. 
First, only one type of test (picture identification task) was used in 
this study. This type of study is suitable for illiterate or semi- 
illiterate subjects who do not have the metalinguistic skills necessary 
to perform other types of tests. However, Finer and Browselow's 
subjects seem to have enough metalinguistic knowledge to perform 
tests designed in other ways. Therefore, this study should have been 
done along with other measures on the same items and with the 
same subjects to increase the validity and informativeness of the L2 



S. Cho: The acquisition of English reflexives by Korean ESL learners 39 

study by providing comparisons with data derived from other 
measures. 

Secondly, in this study, no comparison with native judgments 
was given. Therefore, the judgment as to which picture represents 
the test sentence is the experimenter's alone. Comparison with native 
speakers' judgments is essential in this kind of study in which one's 
L2 acquisition theory makes explicit predictions about learners' 
intuitions for grammaticality relative to those of native speakers. 

4. Hypotheses and predictions 

In this section, I will present the hypotheses which derive from 
the analyses discussed previously and their predictions with respect 
to the acquisition of the GCP in English by adult Korean learners of 
English. Here is the first hypothesis. 

A. The Subset Principle operates in L2 acquisition exactly as it 
does in LI acquisition. That is, learners initially pick up the 
unmarked value of the GCP which generates the subset 
sentences. When learners encounter positive evidence that 
requires the GCP to be set the other way, they simply switch the 
setting to the value of the L2. 

This position is compatible with the 'pure UG hypothesis' (White 
1989) which assumes that UG operates identically in LI and L2 
acquisition, and that there is an acquisition sequence of unmarked 
before marked. Learners' LI does not play any role in L2 acquisition 
because UG reverts to its preset options regardless of the actual 
situation in target language. Thus, there is no transfer of marked 
parameter settings of LI to L2. 

Thus, this hypothesis predicts that the Subset Principle enables 
Korean learners of English to assume that the unmarked value of the 
GCP applies to English. Since there is no positive evidence requiring 
the marked setting in English, subjects make no misinterpretation 
about the binding pattern of English reflexives. Korean learners 
simply start out with the correct English value of the GCP and do not 
accept any non-local antecedents for the reflexive. Everything goes 
automatically and the subjects' native language does not play any 
role in this case. 

The second hypothesis is as follows: 

B. The subset Principle is not available to adult second language 
learners and transfer is the dominant factor in L2 acquisition. 
Thus, if the learners' native language has a different setting for 
the GCP from the target language, they simply transfer their LI 
value to the target language. 



40 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

This position is equivalent to the 'transfer hypothesis' of Contrastive 
Analysis (Fries 1945, Lado 1957) which claims that in areas where 
the structures of two languages differ, learners should encounter 
difficulties due to interference and negative transfer from the native 
language. Thus, where the LI has adopted a superset value, learners 
will assume that this superset value is also appropriate for the L2 
data. UG principles are assumed not to be involved in L2 acquisition 
in this hypothesis. 

Then, this hypothesis predicts that speakers of Korean whose 
native language employs the most marked setting of the GCP would 
start off with their superset LI setting and fail to choose the correct 
English value. They would make mistakes of choosing non-local NPs 
as the antecedent for the English reflexives. 

The third hypothesis is similar to the second hypothesis with 
one important difference. 

C. The Subset Principle is not available to adult L2 learners. 
Learners initially transfer their LI value of the GCP to the target 
language. However, as their learning proceeds, they eventually 
achieve the resetting of the parameter to the correct value of 
the target language even though no positive evidence is 
provided in the target language. 

This position claims that where the LI has adopted a parametric 
value generating superset sentences, learners transfer their LI value 
to the L2 under the assumption that this superset value also applies 
to the L2 data. Thus, they initially produce incorrect sentences, but 
as their learning proceeds, they gradually reset the parameter to the 
value of the L2. This position assumes that the resetting of a 
parameter is not necessarily done in a once-and-for-all fashion. As in 
LI acquisition, L2 learners move along a development continuum, 
starting off with simple structures and moving to more complex 
structures. Therefore, it seems to be that it takes a considerable 
length of time for L2 learners to reset some parameters to the values 
of the target language. 

This hypothesis is compatible with Zobl's (1988) study in which 
his advanced Japanese learners of English did not allow sentences 
violating the adjacency condition which is required by a 
configurational language like English between the verb and its 
complements, while less advanced learners transfer their LI 
superset value. Hirakawa (1990) has also reported similar results, 
that advanced Japanese learners of English eventually achieve the 
resetting of the GCP to the subset value of English from their own 
superset LI value. 



S. Cho: The acquisition of English reflexives by Korean ESL learners 4 1 

Then, this hypothesis predicts that Korean L2 learners first 
transfer their superset LI value to English, making mistakes such as 
picking up non-local NPs as the antecedent for the reflexive, which 
are not allowed by English grammar, but, as their learning proceeds, 
they eventually achieve the switching from their LI superset value 
to the correct English subset value. 

As we discussed before, some UG parameters are multi-valued, 
rather than binary. The following two hypotheses are relevant to 
multi-valued parameters. Here is the fourth hypothesis. 

D. The Subset Principle is not available to L2 learners, and the 
transfer of LI value is not in operation in L2 acquisition either. 
L2 learners pick up an intermediate value which is not 
predicted either by the Subset Principle or the transfer 
hypothesis. 

This position is possible with multi-valued parameters, especially 
with the GCP which has five different values. With the GCP, L2 
learners have three other possibilities than those of the native and 
target language. However, these three possibilities are consistent 
with the UG hypothesis because they are on the hierarchy of the GCP 
values (Wexler & Manzini 1987) all representing natural languages. 
In other words, these possibilities represent values permitted by UG, 
and therefore, possible in natural languages. 

This hypothesis is compatible with the results of the study by 
Browselow and Finer (1991), in which they report that their subjects 
converge on a parameter setting somewhere between the native and 
target language. Their subjects start with the parameter setting of 
the native language and then arrive at a setting that is midway 
between the LI and the L2. 

Then, this hypothesis predicts that Korean learners of English 
have three possibilities other than values from their LI and L2: (b) 
type (Italian), (c) type (Russian), and (d) type (Icelandic) value on 
the hierarchy of the GCP proposed by Wexler and Manzini (1987). 
For example, if Korean learners of English choose the Russian type of 
binding pattern of reflexives, they would pick up either local or non- 
local NPs as the antecedent for English reflexives in nonfinite clauses. 
On the other hand, in finite clauses, they would choose only local 
antecedents for English reflexives. 

The fifth hypothesis is similar to the fourth one with one 
important difference. 

E. None of the following is available to L2 learners: The Sub.set 
Principle, transfer from the LI. and the UG principles. L2 
learners would pick up an intermediate grammar not permitted 
in natural languages. 



42 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Although no one has ever reported the existence of this type of 
grammar in L2 learners' interlanguage, this is a logically possible 
alternative. For example, if L2 learners choose only non-local 
antecedents in finite clauses and only local antecedents in nonfinite 
clauses, this would be an unnatural grammar. Another example is a 
situation where L2 learners choose non-local antecedents regardless 
of whether the clause is finite or non-finite. 

If Korean learners of English assume that this possibility is 
appropriate for English data, this hypothesis predicts that they would 
pick up only non-local antecedents in finite English sentences and 
only local antecedents in non-finite English sentences. Or, they would 
choose only non-local antecedents regardless of the finite or non- 
finite status of the clause. 

5. The experiment 

5.1. Subjects 

In order to test the above hypotheses, an experimental study on 
the acquisition of English reflexives was conducted. Seven groups of 
subjects were involved in this experiment. Five experimental groups 
were composed of native speakers of Korean learning English as a 
foreign language. There were two control groups: A Korean control 
group composed of native Korean speakers and an American control 
group composed of native speakers of English. 

5.1.1. Motivation for selection of subjects 

The experimental groups were composed of a total of 104 
subjects and they were from five different levels: 23 grade 1 1 
subjects (age 16-17), 22 grade 12 subjects (age 17-18), 20 college 
freshmen subjects (age 18-19), 21 college senior students (age 21- 
22) and 18 graduate students (age 23 and over). The choice of these 
four groups of subjects was based upon two considerations. First, as 
many researchers (Ellis 1990, Sorace 1985) point out, grammaticality 
judgments may not be very sensitive indicators of a learner's stage 
of development, in particular for beginning learners. Beginning 
learners often perform very poorly in judgment tests. Therefore, 
researchers conclude that if tests are to be valid measures of L2 
knowledge, subjects must pass a certain threshold. Based upon this 
consideration, I chose three intermediate-level groups and two 
advanced-level groups. 

The other reason for such a choice of subjects is that one of 
purposes of this study is to investigate whether or not Korean L2 
learners show a significant difference between an early stage of their 
L2 development and that of a later stage. In other words, we want to 
see if Korean learners fail to reset the GCP to the value of the L2 at 
the early stage. And if that is the case, we also wanted to see 



S. Cho: The acquisition of English reflexives by Korean ESL learners 4 3 

whether or not they eventually accomplish the resetting of the 
parameter at a later stage. To satisfy such needs, we need both 
intermediate and advanced subjects. 

5.1.2. Control for background variables 

When we choose subjects for L2 experiments, there are some 
other things we must take into consideration. First, we need to 
control for the background variables. Things like the number of 
years of language study, length of residence in target-language 
environment, age when language study began, methods of language 
instruction, and scores of standardized language tests are important. 
Such data are essential to homogenize groups of subjects. By 
constraining the variables, we can limit the possible sources of 
variability of response data, thus enhancing the reliability of a test. 

In this respect, we were fortunate to choose Korean subjects. In 

Korea, students start learning English from grade 7 with almost no 

knowledge of English. Furthermore, they have almost no access to 

English outside the classroom. Therefore, Korean subjects seem to 
share a similar background in terms of the starting age of learning 

English, the number of years of language study, and the amount of 
exposure to English. 

The grade 11 and grade 12 subjects were students at a senior 
high school located in Seoul, Korea. In a Korean high school, students 
receive four hours of English lessons per week. None of the high 
school teachers were native speakers of English. They were all 
Koreans who had studied English language and literature at the 
university. The teaching methods used in Korean high schools are by 
and large traditional. Rules of grammar are intensively taught and 
practiced in classes, and usually instruction focuses on reading and 
writing rather than hearing and speaking. All explanations are given 
in Korean regardless of the level of students. However, teachers 
reported that no form-focused instruction or error-correction was 
given in class with respect to the behavior of English reflexives. 

The grade 13 subjects were first-year students of a college 
located in Seoul, Korea. They receive three hours of English lessons 
per week. Their instructions focus on reading with a belief that 
college students should be prepared to read materials of their own 
field written in English. Another group of college students was 
composed of juniors and seniors who took courses like English 
literature and English composition. Many of them had some 
experience with native instructors. 

Unlike high school students, among college students, the amount 
of exposure to English outside classroom varies slightly from person 
to person; some reported that they were taking private English 



44 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

conversation lessons given by native speakers; some subjects 
subscribed to English newspapers or magazines; others had attended 
special lectures given by the US Cultural Center in Seoul, Korea, etc. 

The graduate students were the most advanced among the five 
groups of subjects. All of them were students of the Graduate 
Institute of Peace Studies in Kyung-Hee University, Seoul, Korea. This 
institute was founded years ago with the support of the United 
Nations with the special aim of developing specialists in international 
peace studies. Therefore, in this school, they placed a heavy 
emphasis upon proficiency in English. Most of the lectures are given 
in English and many international exchange programs are provided 
in this institute. Students reported that they utilized English in their 
normal interactions with teachers. English magazines like Time and 
Newsweek were widely read among the students. 

From all five groups of subjects, those who had early exposure 
to English or any experience in an English-speaking country for more 
than three months were eliminated. Therefore, each subject seemed 
to share a common background with other subjects in the same 
group with respect to the starting age of learning English, the 
number of years of English education, and the amount of the 
exposure to English. The following table presents the subjects' 
background information such as their age range, mean age, and 
average number of years of English education. 

Table 2 Subjects background information 











Ave. length oi 


Group 


N 


Age range 


Ave. age 


Eng. learning 


Grade 11 


23 


16-18 


16.2 


4.3 


Grade 12 


22 


17-19 


17.4 


5.3 


Freshman 


20 


18-23 


19.8 


6.7 


Jr. & Sr. 


21 


20-26 


22.6 


9.8 


Graduate 


18 


23-33 


27.9 


11.4 



The two control groups were composed of fifteen subjects each. 
A Korean control group consisted of college students who attended 
the same school as the experimental groups. Fifteen American 
students at the University of Illinois at Urbana-Champaign served as 
the English control group. The purpose of testing control groups was 
to compare L2 learners' judgements with those of native speakers. 
This is really important if the purpose of the study is to observe 
learners' judgements as they approach those of native speakers, or if 
a linguistic theory makes explicit predictions about learners' 
intuitions for grammaticality relative to those of native speakers. 



S. Cho: The acquisition of English retlexives by Korean ESL learners 45 

5,1.3. Screening test 

In order to secure comparability among subjects within a group, 
a screening test was administered. When we investigate the 
acquisition of L2 syntactic properties, we need to test the subjects' 
basic grammatical skills. We conducted a preliminary test which 
examined the subjects' ability to distinguish between a pronoun and 
a reflexive and their proficiency in vocabulary items which were on 
the test. Six high school students who failed to meet the criteria of 
the test were eliminated. 

5.2. Materials 

The actual test was composed of two types of tests of different 
methods: A reading comprehension test and a sentence-picture 
matching test. Four types of sentences were employed for both of 
these tests. They are two-clause finite sentences, two-clause non- 
finite sentences, three-clause finite sentences, and three-clause non- 
finite sentences. Example (6) illustrates each type. 

(6) a. Mary said that Sue pinched herself. 

b. John asked Fred to paint himself. 

c. Jean said that Mary remembered that the waitress 
deceived 

d. Tom remembers that Mike asked Al to dress himself. 

5.2.1. Sentence-picture matching test 

The first test of the experiment involved a sentence-picture 
matching task, where subjects heard an English sentence containing a 
reflexive which was read two times in a row by a native speaker at 
normal speed. Then they were asked to look at three pictures and 
make a choice as to which picture they believe best correctly 
describes the sentence. An example is given in (7). 

(7) (Subjects will hear) 

John said that Tom shot himself. 






46 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

A total of 12 sets of pictures were presented, of which 8 items were 
relevant to the effect of the English value of the GCP. Since basically 
the same predictions are made for this test as in the above- 
mentioned three types of grammaticality judgment tests, I will not 
spell them out here. 

5.2.2. Reading comprehension test 

The reasoning behind the presentation of a reading 
comprehension test is a methodological one. Chomsky's grammar 
aims at describing and explaining the learner's 'competence'. 
However, L2 researchers within the UG framework, more often than 
not, gather data from an empirical examination of language learners' 
'performance'. This, as Ellis (1985) claims, poses a serious problem of 
what type of performance provides the best window for looking at 
competence. Most of L2 acquisition research is based upon results 
from standard grammaticality judgment tests. Therefore, in an effort 
to enhance the reliability and validity of grammaticality judgment 
tasks, it is important to examine whether or not there is a significant 
difference between grammaticality judgment tasks and different 
measures. If similar results could be obtained by different methods, 
the conclusions would be strengthened. 

In this test, subjects were asked to read an English passage 
which contains reflexives and answer the questions by selecting one 
option among three choices. A small sample from one of the reading 
passages is given in (8). 

(8) Last year, Chinese students staged a massive demonstration 
against the Communist government. They called on 
government leaders to make a series of political and 
economic reforms to save ailing China. The Chinese police 
immediately suppressed the students in a brutal way. 
However, the students believed that the police disgraced 
themselves. 

Q Who did the police disgrace in last year's Chinese uprising? 

A. police 

B. students 

C either police or students 

Subjects were given two passages with four questions, of which two 
are relevant to the interpretation of the English reflexive. The other 
two items are distractor items even though they are perfect 
questions in their own right. Each passage runs around 100 words. 

5.3. Procedures 

Before taking the tests, the subjects were given written 
instructions for the experiment in their native language. They were 



S. Cho: The acquisition of English refkxives by Korean ESL learners 47 

informed that the purpose of this study was to examine their 
intuitions about English sentences rather than their knowledge of the 
grammatical rules of English. Thus, we instructed subjects to focus on 
how they felt about English sentences in the test in an effort to 
insure that they would not regard the tasks as a grammar test which 
is determined by prescriptive rules. 

This was followed by an explicit description of the tasks which 
subjects were to perform, written in plain 'everyday' language to 
avoid the possibility of idiosyncratic interpretation of the judgement 
criterion. Attempts were made to use 'ever>'day' vocabulary in actual 
test items, too. Otherwise, there is a possibility that subjects would 
judge the correctness of a sentence upon the lexical interpretation 
rather than grammatical judgement. Extra efforts were also made to 
devise sentences that are semantically plausible to avoid the 
possibility that students would make judgments based on a semantic 
interpretation rather than a grammatical one. 

The subjects were tested group by group in the classroom of 
their schools by a native Korean who also had a good knowledge of 
English and experience in carr>'ing out this kind of experiment. Ten 
minutes were given for the sentence-picture matching test, and 
fifteen minutes for the reading comprehension test. A Samsung 
cassette recorder was used for the sentence-picture matching task in 
which subjects were to choose a picture after hearing an English 
sentence read by a native speaker of English. 

6. Results and discussion 

6.1. Results 

6.1.1. Results from the sentence-picture matching task 

In this task, subjects heard an English sentence read by a native 
speaker of English twice, and then they were asked to choose which 
of the three pictures they believed most correctly described the 
sentence. Results from this task are given in Table 3 in the form of 
percentages of responses by sentence types and levels of subjects, 
and in Table 4 which presents percentages of correct choice of local 
NPs and their means by sentence types and levels of subjects. 

The overall results show that approximately 69.69r of the 
subjects' responses were correct choices of a local antecedent for the 
reflexive. However, there were a lot of subjects who failed to choose 
the correct English value of the GCP. Therefore, these results seem to 
suggest that the Subset Principle does not operate in Korean learners' 
acquisition of English. If the subjects had the Subset Principle, they 
would have initially chosen the English subset value, independently 
of their LI. and not made the mistake of taking non-local NP<; as 
antecedents for English reflexives. 



48 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Table 3 The percentage of responses by type and grade in the 
sentence-picture matching task 

Korean English 
Gil G12 Fr Jr.&Sr. Grad. controls controls 



2 


clause finite 
















L 69.4 


84.2 


85.0 


88.1 


97.2 


33.3 


100 




NL 22.2 


10.5 


7.5 


4.8 


2.8 


56.7 







Either 8.4 


5.3 


7.5 


7.1 





10.0 





2 


clause non-finite 
















L 44.4 


50.0 


57.5 


83.3 


88.9 


26.7 


96.7 




NL 44.4 


50.0 


35.0 


14.3 


11.1 


63.3 


3.3 




Either 11.2 





7.5 


2.4 





10.0 





3 


clause finite 
















L 61.1 


68.4 


65.0 


76.2 


86.1 


33.3 


93.4 




NL 25.0 


26.3 


27.5 


23.8 


13.9 


53.3 


3.3 




Either 13.9 


5.3 


7.5 








13.4 


3.3 


3 


clause non-finite 
















L 47.2 


47.4 


55.0 


64.3 


72.2 


36.7 


100 




NT. 41.7 


44.7 


37.5 


31.0 


2.8 


46.7 







Either 11.1 


7.9 


7.5 


4.7 





16.6 






Table 4 The percentage of correct responses by type and group 



Gil 



G12 Fr. Jr.&Sr. Grad. mean 



2 clause finite 

2 clause non-fin. 

3 clause finite 

3 clause non-fin. 



69.4 


84.2 


85.0 


88.1 


97.2 


84.8 


44.4 


50.5 


57.5 


83.3 


88.9 


64.9 


61.1 


68.4 


65.0 


76.2 


86.1 


71.4 


47.2 


47.4 


55.0 


64.3 


72.2 


57.2 



55.5 62.6 65.6 78.0 86.1 



However, results from the advanced-level subjects do not 
support the transfer hypothesis either. In particular, graduate 
subjects made 86.7% of correct choices, a performance close to that of 
the native English control group, indicating that they were almost 
unaffected by transfer from their LI value. 



S. Cho: The acquisition of English teflexives by Korean ESL learners 49 

These results do not support the hypothesis advanced by 
Broselow and Finer either. While their hypothesis predicted that 
Korean subjects would choose an intermediate value which is distinct 
from either the LI or L2 (middle non-local NP in three-clause 
sentences, as discussed previously), they chose more left-most NPs 
than middle ones for English reflexives, suggesting that they chose 
them not as an intermediate value, but as a value of their LI. 

In general, Korean learners showed clear effects of UG in two 
ways. First, they showed an overall performance which is 
significantly better than chance even though less-advanced subjects 
showed some signs of transfer. Secondly, the choice of values they 
made were within the constraints of UG. That is, none of the values 
they chose represents an unnatural language. All of them represent 
natural languages definable by UG. 

The next question is whether or not there is a significant 
correlation between the subjects' performance and factors like age 
and length of learning. We can observe a clear developmental 
progress from grade 11 subjects to graduate subjects; while grade 11 
subjects made an average of 55.5% of correct choices, graduate 
subjects showed an average of 86.1% of accurate choices for the 
English value. Such a poor performance of less-advanced subjects is 
probably attributable to the fact that their knowledge of English was 
still insufficient, so that there was no appropriate structural basis for 
UG principles to operate on. On the other hand, the graduate subjects' 
near native-like performance, even in the absence of the Subset 
Principle, can be interpreted as indicating that they have had enough 
access to UG principles and have achieved the resetting of the GCP to 
the value of English. 

Now, let us look at whether or not there are differences between 
different sentence types. Subjects showed better performance in 
two-clause sentences (mean 74.9%) than in three-clause sentences 
(mean 64.3%). This result is consistent with our expectation that the 
more complex a structure is, the more difficulty subjects would have, 
because in Korean, there are more choices of NPs permissible as 
antecedents. The same pattern of response is found in finite vs. non- 
finite sentences. Subjects showed better performance in finite 
sentences (mean 78.1%) than in non-finite sentences (mean 61.1%). 
Thus, it seems that Korean subjects were affected by both complexity 
and non-finiteness of sentences, but slightly more affected by 
complexity than non-finiteness. 

As we expected, the English control group showed an 
overwhelming choice of local antecedents at a level ranging from 
93.4% to 100% over non-local ones regardless of finite or non-finite 
and two-clause or three-clause. On the other hand. Korean control 



50 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

group chose non-local antecedents more often than local ones. In 
two-clause sentences, where not only local antecedents but non-local 
ones are permissible in Korean, subjects chose non-local options 
(60.0%) more often than local options (30.0%). In three-clause 
sentences, where subjects have three possible choices permitted in 
Korean, the same pattern of response is found. These results, as a 
whole, lend support for the view that in Korean, there is a preference 
for the non-local antecedent over the local one even though both are 
grammatically legitimate. 

In summary, the overall performance of the experimental 
Korean subjects is clearly distinct from that of the English control 
group, suggesting that the Subset Principle is not in action in their 
acquisition of English. Less-advanced subjects failed to choose the 
correct English value suggesting that they transferred their LI value 
to English. However, they did not behave exactly as Korean controls 
did. And, we found that subjects, as their age increases, showed a 
developmental progress which became gradually close to the binding 
pattern of the English control group. In particular, the most- 
advanced graduate subjects behaved almost like the English control 
group, indicating that they already had achieved the resetting of the 
parameter to the value of English. 

6.1.2. Results from the reading comprehension task 

This task involved a reading comprehension test, where subjects 
were asked to read an English passage which contains reflexives and 
to answer questions by selecting one of three or four choices. The 
focus of the test was to see whether subjects make a correct choice in 
questions which are relevant to the interpretation of English 
reflexives. The results are given in Table 5 in the form of 
percentages of responses in relation to sentence types and levels of 
subjects, and in Table 6 which presents percentages of correct choice 
of local NPs and their means by each group and sentence type. 

The overall results from this task are similar to those of the 
previous test in that more-advanced learners performed better than 
less-advanced learners confirming the developmental progress we 
found in previous tests. Graduate subjects showed the highest 
performance of 81.9% of correct choices. Grade 11 and grade 12 
subjects showed worse performance than those of the previous test, 
27.8% and 44.8% of correct choices respectively. No significant 
difference was found between freshmen subjects (55.0%) and junior 
and senior subjects (57.2%). 



S. Cho: The acquisition of English reflexives by Korean ESL learners 5 1 

Table 5 The percentage of responses by level and type in the reading 
comprehension task 

















Korean 


English 






Gil 


G12 


Fr. 


Jr.&Sr. 


Grad. 


controls 


controls 


2 


clause finite 


















N 


16.7 


36.8 


35.0 


52.4 


83.3 


80.0 


86.6 




NL 


72.2 


57.9 


65.0 


47.6 


16.7 


13.3 


6.7 




Either 


11.1 


5.3 











6.7 


6.7 


2 


clause non- 


finite 
















L 


33.3 


57.9 


65.0 


42.9 


55.6 


73.3 


93.3 




NL 


38.9 


36.6 


35.0 


47.6 


44.4 


20.0 


6.7 




Either 


27.8 


10.5 





9.5 





6.7 





3 


clause finite 


















L 


22.2 


63.2 


60.0 


76.2 


100 


80.0 


86.6 




NL 1 


44.4 


15.8 


25.0 


9.5 





13.4 


13.4 




NL2 


27.8 


10.5 


15.0 


14.3 





6.7 







Either 


5.6 


10.5 

















3 


clause non- 


finite 
















L 


38.9 


21.4 


60.0 


57.1 


88.9 


73.3 


93.3 




NL 1 


16.7 


26.3 


20.0 


19.0 


5.5 


20.0 







NL2 


16.7 


21.1 


10.0 


19.0 


5.5 


6.7 


6.7 




Either 


27.8 


36.5 


10.0 


4.8 













Table 6 The percentage 


of correct choice by 


group and I 


[ype 








Gil G12 


Fr. Jr 


.&St. Grad. 


mean 




2 clause 


finite 


16.7 36.8 35.0 52.4 83.3 


44.8 




2 clause 


non-fin. 


33.3 57.9 65.0 42.9 55.6 


50.9 




3 clause 


finite 


22.2 63.2 60.0 76.2 100 


64.3 




3 clause 


non-fin. 


38.9 21 


.1 60.0 57.0 88.9 


53.2 




mean 




27.8 44.8 55.0 57.2 81.9 





However, if we look at the results closely, it is difficult to discern 
the consistent response pattern we found in the sentence-picture 
matching test. Subjects did well on some items, but poorly on others. 
For example, grade 12 subjects showed 63.2% accuracy in three- 
clause finite sentences, but 21.1% accuracy in three-clause non-finite 
sentences. Such an unexpected inconsistency is also found even in 



5 2 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

the performance of the most advanced graduate subjects; while they 
showed 100% accuracy in three-clause finite sentences, in two-clause 
non-finite sentences, they showed a surprisingly low 55.6% accuracy. 
However, in general, more-advanced learners showed the response 
pattern in relation to sentence types which we found in the 
sentence-picture matching test. Such a pattern is not found among 
less-advanced subjects. The subjects' performance in this test in 
relation to age share the same problem. That is, among more- 
advanced learners, the tendency toward more accurate 
interpretation of English reflexives with increasing age is found; 
graduate subjects outdid the junior and senior subjects in all 
categories of sentence types. However, among less-advanced 
subjects, it appears to be a disaster. For example, grade 12 subjects 
outdid freshmen subjects in two types of sentences; 36.8% vs. 35.0% 
in two-clause finite sentences, and 63.2% vs. 60.0% in three-clause 
non-finite sentences. However, at the same time, they showed 21.1% 
accuracy in three-clause non-finite sentences which is much worse 
than the grade 11 subjects' 38.9% accuracy. 

Such an inconsistent response pattern among less-advanced 
subjects is probably attributable to several factors. First, the level of 
the reading comprehension test is appropriate for advanced learners, 
not less-advanced learners. Even though we expected in advance that 
they could handle this kind of test, it turned out that their 
knowledge of English is not sufficient to carry out the task yet. They 
seem to be more affected by the misinterpretation of context of the 
passage than by the binding pattern of English reflexives. The other 
possibility is that the number of test items is too small to draw 
reliable results. Due to the constraint of time and length of the 
passage, one test item was assigned to each sentence type. 

As in the sentence-picture test, English controls showed an 
overwhelming choice of correct interpretation of English reflexives. 
However, Korean subjects showed a very different response pattern 
from that of the previous test. Rather they showed a performance 
which is close to that of English controls in that they chose local NPs 
for English reflexives. This is not too surprising given the fact that 
the Korean version was presented to them. They seemed to be more 
affected by context of the passage than the ambiguous binding 
pattern of Korean reflexives. Therefore, as far as this kind of 
particular test is concerned, it is not too meaningful to compare the 
results of the Korean control group with those of the experimental 
groups. 



S. Cho: The acquisition of English reflexives by Korean ESL learners 53 

6.2. Discussion 

6.2.1. Results in relation to hypotheses 

In considering whether or not the Subset Principle is available 
to L2 learners, two hypotheses were proposed. One is the subset 
hypothesis, which suggests that the Subset Principle operates in L2 
acquisition exactly as it does in LI acquisition. That is, Korean 
learners of English initially pick up the unmarked subset value of the 
GCP, and therefore, they make no misinterpretations of the binding 
pattern of English reflexives. The other possibility is the transfer 
hypothesis, which says that the Subset Principle is not available to 
L2 learners, but instead transfer is the dominating factor in L2 
acquisition. That is, Korean learners of English initially choose the 
unmarked superset value of the GCP and transfer this value to the 
target language making mistakes such as taking non-local NPs as the 
antecedents for English reflexives. 

The results from the two tests indicate that the Subset Principle 
does not operate in Korean learners' acquisition of English. This result 
is consistent with findings by other researchers (Zobl 1988, White 
1989a, Hirakawa 1990). Contrary to what the Subset Principle 
predicts, Korean subjects, especially less-advanced subjects, made 
lots of mistakes in choosing non-local NPs as antecedents for English 
reflexives. If the Subset Principle had been available to L2 learners, 
they should have chosen only local NPs for English reflexives. 

If the Subset principle is no longer available to L2 learners, are 
they influenced by their native language? The results from less- 
advanced learners, especially grade 11 subjects, seem to be 
compatible with the transfer hypothesis in that they set the value of 
the GCP wider than required by English. They may have assumed 
that the LI superset value is also appropriate for English, thereby 
allowing non-local NPs as antecedents for English reflexives. 
However, the results from more-advanced learners are not consistent 
with the transfer hypothesis. Even though some of them still chose 
the wrong value of the GCP, the majority of them showed a 
performance more like that of the native English control group than 
that of the native Korean control group. Especially, the performance 
of the graduate subjects is close to that of the English control group 
in not allowing non-local NPs for the reflexive. This suggests that 
transfer is no longer a factor in their L2 acquisition in spite of the 
fact that they cannot directly apply the Subset Principle to L2 data 
any more. 

In this context, it may worth noting that the inaccessibility of 
the Subset Principle to adult L2 learners does not necessarily mean 
that UG principles are not available to L2 learners. As we discussed 
earlier, the Subset Principle is a learning principle, which is 



5 4 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

independent of UG principles, to enable language learners to initially 
choose the unmarked subset value for a given parameter. The fact 
that Korean learners switch their LI value to some other values of 
the GCP suggests that UG is still available to them because 
parameters are part of UG. If UG is not available to adult L2 learners, 
they are not expected to choose a value which is one of GCP values. 
However, even if UG is available to adult L2 learners, there still 
remains a question as to exactly how more advanced Korean learners 
of English achieved the resetting of the GCP from the marked 
superset value to the unmarked subset value without the benefit of 
positive and negative evidence from their target language, a problem 
we will return to below. 

About the hypothesis of acquiring an intermediate value which 
is distinct from either LI or L2 (Broselow & Finer 1991), a possibility 
was raised that Korean learners of English would have three other 
possibilities than their LI value and L2 value: That is, (c) or (d) type 
value on the hierarchy of the GCP proposed by Wexler and Manzini 
(1987). Browselow and Finer base such a claim upon the observation 
that their Japanese and Korean subjects took 'Mr. Fat' as the 
antecedent in a non-finite sentence like (a), but 'Mr. Thin' in a finite 
sentence like (b). 

(9) a. Mr. Fat expects Mr. Thin to paint himself. 

b. Mr. Fat thinks that Mr. Thin will paint himself. 

It seems to be that the subjects followed an English type binding 
pattern in a sentence like (9a), but a Korean or Japanese type binding 
pattern in a sentence like (9b). They interpreted this result as 
indicating that the subjects were treating the English reflexive as 
though it occupied position (c) or (d) on the hierarchy of the GCP. 

However, as I mentioned before, this study raises some issues 
which require explanation. First, if the subjects had picked up an 
intermediate value on the basis of whether the sentence is finite or 
non-finite, the same pattern should be found in more complex 
sentences such as the following in example (10). 

(10) Tom says that Mr. Fat asked Mr. Thin paint himself. 

In a non-finite sentence like (10), subjects should choose 'Mr. Fat', 
not 'Tom', as the antecedent for the reflexive 'himself. However, the 
results from the tasks of our study show that many Korean subjects 
chose the first NP like 'Tom' as the antecedent for the reflexive in 
three-clause sentences like (10). It follows, then, that they were not 
choosing the intermediate value, but their own LI value of the GCP. 

Secondly, Korean exhibits no distinction between finite 
sentences and non-finite ones. Therefore, Korean subjects generally 
have more difficulty in choosing the correct English value of the GCP 



S. Cho: The acquisition of English reflexives by Korean ESL learners 5 5 

in non-finite sentences than in finite ones. Furthermore, as the 
results from the Korean control group indicates, in Korean there is a 
strong preference for a non-local antecedent over a local one even 
though both are permissible in the language. Thus, the Korean 
subjects' choice of non-local antecedents in non-finite sentences is 
probably attributable to the combined possibility of treating non- 
finite sentences as simple ones and being led by the preference for 
non-local antecedents over local ones. If this is the case, we can 
conclude that they were not choosing the intermediate value, but 
their own relatively wide Korean value. 

Now, let us move to the logical possibility of choosing an 
intermediate value which is not permitted in natural languages. Two 
possibilities were predicted by this hypothesis; either that Korean 
subjects would have picked up only non-local antecedents in finite 
sentences, and only local antecedents in non-finite sentences; or that 
they would have picked up only non-local antecedents regardless of 
finite or non-finite. These possibilities are totally disconfirmed by 
the results from the tasks in which not a single Korean subject 
showed such a binding pattern of reflexives. This lends additional 
support for the operation of UG in L2 acquisition. If L2 learners have 
no access to UG, and they rely upon general problem-solving 
hypotheses for the construction of a grammar of the target language; 
there is no reason that they should not adopt logical possibilities 
which do not represent UG parameters. In other words, parameters 
are part of UG, and therefore, they all represent natural human 
languages. Unless UG operates in L2 acquisition, language learners 
may adopt logically possible alternatives other than only the 
possibilities represented by UG parameters. 

6.2.2. Correlation between performance and factors 

The next question is whether or not there is a significant 
correlation between the subjects' performance and factors like their 
ages and length of English education. With the exception of the less- 
advanced subjects in the reading comprehension task, our results 
clearly show that there is such a correlation. Korean subjects showed 
a developmental progress from the transfer stage of grade 1 1 to the 
parameter-resetting stage of graduate level. 

This result is not compatible with Hirakawa's (1990) outcomes 
where she found no tendency towards more accurate interpretations 
of English reflexives with increasing grade. This is probably 
attributable to the fact that Hirakawa's choice of subjects was not 
desirable: Most of Hirakawa's subjects were low-intermediate or 
intermediate learners. If one of the purposes of the study is to 
investigate whether or not there is a significant difference between 
earlier stages and later stages of L2 development, we need to 



56 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

investigate subject groups which are more widely separated from 
each other by factors like age and length of English education. 

6.3. Possible solutions 

6.3.1. Possible L2 learning stages 

Generally speaking, less-advanced Korean subjects seem to fail 
to choose the correct English value for the GCP. They set a wider 
value than is required by English reflexives, and as a result, they 
take non-local NPs as the antecedents for English reflexives. This 
suggests that the Subset Principle is not available to adult L2 
learners, and that transfer still plays a considerable role in their 
acquisition of English reflexives. They appear to have more difficulty 
in choosing the correct English value in sentences of complex 
embedding. And the inability to make a distinction between finite 
and non-finite sentences often results in making more mistakes of 
choosing non-local NPs for English reflexives in non-finite sentences 
than in finite sentences. However, even though they fail to choose the 
correct English value, they do not seem to choose an intermediate 
value of the GCP, either. They appear to be affected by both UG 
principles and transfer at this stage. 

However, as their learning progresses, there is an increase in the 
number of learners who make more choices of correct English value 
for the GCP than wrong ones. They show a developmental progress as 
they approach later stages of development. This suggests that 
intermediate-level learners are less affected by transfer from their 
native language, and that they are in the process of resetting the GCP 
from a wider LI value to a narrower L2 value. However, they appear 
not to have made a clear distinction between finite and non-finite 
sentences yet. 

Advanced-level learners like graduate students seem to set the 
correct English value for English reflexives. They correctly choose 
local NPs as the antecedents whether the sentence is finite or non- 
finite, indicating that they make a distinction between finite and 
non-finite sentences, and that they already achieved the resetting of 
the GCP from the LI value to the L2 value. In principle, as White 
(1989b) points out, resetting may never take place at all, presumably 
either because of the absence of the relevant kind of input, or 
because L2 learners are unable to use it even if it occurs. However, 
some other previously cited studies as well as ours have reported 
that resetting does happen. In these studies, more-advanced learners 
do not inappropriately apply the incorrect LI superset value to the 
target language. 

However, there still remains a question of exactly how they 
achieve the resetting of a parameter. We have already seen that in 



S. Cho: The acquisition of English reflexives by Korean ESL learners 5 7 

L2 acquisition of the GCP, learners have no access to the Subset 
Principle and positive evidence from the L2 which is necessary to 
disconfirm the inappropriate LI superset value. This means that 
direct negative evidence is the only available alternative for the 
learners to achieve the resetting of the GCP. This is especially true for 
the Korean L2 learning situation where most of English learning 
takes place solely in a classroom setting. 

6.3.2. Indirect negative evidence 

Then, how can Korean learners of English proceed from the 
wider grammar of their own language to the narrower grammar of 
English without the benefit of direct negative evidence? One 
possibility is what is known as 'indirect negative evidence'. This idea, 
proposed by Chomsky (1981), is that when language learners fail to 
hear certain sentences, this will be taken as (indirect) evidence that 
such sentences are ungrammatical. Therefore, this kind of evidence 
can be obtained even without correction or adverse reactions, etc.'* 

Chomsky has suggested the Prodrop Parameter can be set by 
virtue of such indirect negative evidence. If children's target 
language is English, sentence like (11) is not allowed in English, while 
allowed in prodrop languages like Italian and Spanish. 

(11) *Left at night. 

Then, if children mistakenly hypothesize that the unmarked value is 
appropriate for the target language, the hypothesized grammar will 
generate a grammar properly containing English. 5 To solve this 
problematic situation, Chomsky's argument proceeds as follows: Since 
every language would prefer a null subject to an overt one, children 
expect null subjects to occur. Thus, when children notice the non- 
occurrence of null subjects, their absence can serve as relevant 
evidence that such sentences are ungrammatical in English. 

Although many researchers (e.g. White 1989b) call into question 
the validity of this idea, if we apply this insight from LI acquisition 
research to Korean learners' acquisition of English GCP value, Korean 
learners fail to hear English sentences with reflexives referring to 
non-local antecedents, and take this as evidence that such sentences 
are ungrammatical in English. In other words, Korean L2 learners 
start with an incorrect initial guess for English, but the absence of 
English sentences with reflexives bound to non-local NPs will 
indirectly indicate that the initial guess is wrong. Then, the incorrect 
superset setting will be replaced by the correct subset setting even 
in the absence of direct negative evidence. 

However, a question still remains as to under what 
circumstances the absence of a sentence will be taken by Korean 
learners as evidence for ungrammaticality. In the case of the Prodrop 



58 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Parameter, the preference of null subjects among languages to overt 
ones leads language learners to expect the occurrence of such 
sentences, and therefore, the absence of such sentences is taken as 
relevant evidence. However, in the case of Korean reflexives, which 
are not preferred to an unmarked English value, it is not clear what 
aspect of Korean reflexives eventually lead Korean learners to expect 
the occurrence of superset sentences in English. 

6.3.3. Indirect positive evidence 

Zobl (1988) presents another possibility, which one might call 
'indirect positive evidence', in his study of Japanese adult learners' 
acquisition of English VP construction. His idea is that language 
learners who initially hypothesized an incorrect wider grammar may 
find evidence to disconfirm the hypothesis from another area of the 
grammar. That is, a certain grammatical knowledge obtained from 
one area of the grammar may serve as a trigger for disconfirming the 
incorrect hypothesis of another area of the grammar. Zobl showed 
how this kind of evidence is used as the basis for rejection of an 
incorrect hypothesis by using the example of passive construction in 
English. 

The passive construction in English involves movement of the 
NP in direct object position to an empty subject position, leaving 
behind an empty category. Then, how can Japanese learners of 
English notice that there must be adjacency between the verb and 
the empty object left behind? Zobl's answer to this question is given 
by virtue of the notion of 'deterministic parsing'. Van Buren (1988) 
has presented a good explanation of this rather technical notion. 
According to his explanation, one of the basic tenets of parsing 
theory is 'deterministic', meaning that probability is not involved in 
the parsing process, but the process is totally governed by a 
corresponding grammar on a yes/no basis. 

Given the assumption of a determinacy of parsing process, NP 
movement in English passive construction entails that the empty 
category must be adjacent to the verb. Then, the Japanese learners of 
English only have to learn the rule of NP movement in English, which 
is a relatively easy task to do because enough positive evidence for 
the NP movement is provided in English. Then, there automatically 
follows the adjacency between the verb and its empty direct object. 
Once Japanese learners obtain the rule of English passive, they 
realize that their initial hypothesis is not appropriate for English VP 
structure, thus replacing the incorrect assumption with the correct 
one. 

Zobl argues that some properties from one area of a certain 
grammar, which only become available to advanced L2 learners from 
positive evidence, triggers reanalysis of the incorrect wider 



I 



S. Cho: The acquisition of English reflexives by Korean ESL learners 5 9 

hypothesis, and eventually lead the language learners to reset the 
parameter to the correct narrower grammar. 

In principle, there is no reason to rule out the possibility that 
language learners may use 'indirect negative evidence' and 'indirect 
positive evidence' as the source for the rejection of an incorrect 
initial hypothesis when input from the target language provides no 
clear disconfirming evidence for the hypothesis. In practice, 
however, these proposals may sound implausible.^ Still, they deserve 
consideration. For our present purpose, it is sufficient to outline the 
class of available solutions. The choice of a correct solution requires 
much more extensive research. 

NOTES 

1 This does not imply that UG theory denies the existence of 
negative evidence. Since negative evidence available to language 
learners is usually accidental, namely not uniformly available to all 
language learners, UG theory rejects this kind of evidence as a 
reliable source for disconfirming purposes. 

2 Hirakawa also examined the effects of the Proper Antecedent 
Parameter (Wexler & Manzini 1987). However, here I just 
concentrate on results for the GCP. 

3 Broselow and Finer also investigate the acquisition of 
phonology: The mastery of particular syllable onset clusters. 
However, I will not discuss the acquisition of phonology here. 

'♦ White (1989b) argues that the non-occurrence of certain 
sentences does not guarantee ungrammaticality of those sentences; 
there are many other sentences that language learners will not hear 
which are nevertheless grammatical. Furthermore, she points out, 
this presupposes that learners are able to notice non-occurrence, 
which is doubtful. 

5 This argument proceeds under the assumption that the 
prodrop value forms the unmarked value, and the non-prodrop, the 
marked. 

6 For example, van Buren (1988) argues that Zobl's proposal is 
unwarranted in the theoretical context in which it is drawn. 

REFERENCES 

BIRDSONG, D. 1989. Metalinguistic performance and interlanguage 

competence. New York: Springer. 
Broselow, E., & D. Finer. 1991. Parameter setting in second language 

phonology and syntax. Second Language Research 7.35-59. 



6 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: 

Foris. 
Ellis, R. 1985. Understanding second language acquisition. Oxford: 

Oxford University Press. 
. 1990. Grammaticality judgements and learner variability. 

Proceedings of the tenth meeting of the second language 

research forum, ed. by H. Burmeister & P. L. Rounds. 
Flynn, S. 1987. A parameter-setting model of L2 acquisition. 

Dordrecht: Reidel. 
Fries, C. 1945. Teaching and learning English as a foreign language. 

Ann Arbor, MI: University of Michigan Press. 
HIRAKAWA, M. 1990. A study of L2 acquisition of English reflexives. 

Second Language Research 6:1.60-85. 
Lado, R. 1957. Linguistics across cultures. Ann arbor, MI: University 

of Michigan Press. 
Van Buren, P. 1988. Some remarks on the subset principle in second 

language acquisition. Second Language Research 4.33-40. 
WEXLER, K., & R. Manzini. 1987. Parameters and learnability in 

binding theory. Parameter setting, ed. by T. Roeper & E. 

Williams. Dordrecht: Reidel. 
White, L. 1985. The acquisition of parameterized grammars: 

Subjacency in second language acquisition. Second Language 

Research 1.1-17. 
1987. Markedness and second language acquisition: The question 

of transfer. Studies in Second Language Acquisition 9.216-286. 
1989. Universal Grammar and second language acquisition. 

Philadelphia: John Benjamins Publishing Co. 
ZOBL, H. 1988. Configurationality and the subset principle: The 

acquisition of V by Japanese learners of English. Learnability 

and second languages: A book of readings, ed. by J. Pankhurst, 

M. Sharwood Smith, & P. Van Buren. Dordrecht: Foris. 

APPENDICES 

Appendix A: General Background Questions 

Age: years months 

Sex: M F 

1 . How old were you when you started learning English? 

2. How many years have you studied English? 

3. How many hours do you study English in class in a week? 



Cho: The acquisition of English reflexives by Korean ESL learners 6 1 

4. Do you study English outside of classroom? If yes, how? 
(examples: watching TV programs in English, reading English 
newspapers or magazines, attending lectures given in English, 
receiving English conversation lessons, etc.) 



5. Have you lived in an English-speaking country? 
If yes, how long? years months. 

Appendix B: Screening Test Questions 

I. Translate the following English words into Korean. 

1 . to humiliate 

2. to deceive 

3. a candidate 

4. a surgeon 

5. behavior 

6. a demonstration 

7. to stir 

8. intervention 

9. a critic 

10. generosity 

II. Translate the following English sentences into Korean. 
1 1 . Mary wanted her mother to give a present to her. 

12. Mike think that Tom hates himself. 



13. 1 think that president Roh considers Kim to be the best 
candidate. 



14. They shot the arrows at each other. 



15. The doctor told the patient not to go out by himself. 

III. Look at the following English sentences. If you think that 
the sentence is correct, mark 'C, or incorrect, mark T. 

16. I like Tom's picture of his father. 

17. Mr. Lee wants Al to assist him. 

18. Ann hopes that her father will help herself. 

19. They think that both will win the game. 

20. John thinks that Mary will kill himself. 



62 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



Appendix C: Test items of the experiment 

I. Listen carefully to the reading of the English sentence. Each 
sentence will be read twice at normal speed. And then look at 
pictures on the exam and choose one of them which you think best 
describes what you hear. 












Cho: The acquisition of English reflexives by Korean ESL learners 63 

4. A B C 




i^ 









(f 


) 

7^ 


rl 




.,-.i. 


/ 


„.l4. 








64 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

8. A B C 









ft 


- 




10. A 




Cho: The acquisition of English reflexives by Korean ESL learners 
12. A B C 



65 




H«y 






Test Sentences 

1 . He answered the phone himself, not the secretary. 

2. John said that Tom shot himself. 

3. John considers himself to be taller than Fred. 

4. Mr. Kim says that Tom thinks that John might kill himself. 

5. I heard that the doctor asked the patient to talk to himself 
about the dream. 

6. The soldiers shot the arrows at each other. 

7. They say that the president is not sure that the minister will 
vote for himself. 

8. My daughter said that her friend pinched herself. 

9. Fred said that John pulled the blanket around him. 

10. The fact that she is fatter than Mary annoys Sue. 

1 1 . My brother recalled that the teacher wanted Mike to hit 
himself. 

12. Mary asked her mother to dress herself. 

13. I remember that Bill asked John to shave himself. 

II. Read the following English passages and answer the English 
questions by circling one item which you think is the most 
appropriate to the questions. 



66 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Passage A 

Last year, Chinese students staged a massive demonstration 
against the communist government. They called on the 
government leaders to make a series of political and economic 
reforms to save ailing China. The Chinese police immediately 
suppressed the students in a brutal way. However, the students 
believed that the police disgraced themselves by refusing 
people's sincere desire for democracy and a free-market 
economy. The Chinese leaders claimed that evil-minded Chinese 
dissidents stirred the innocent students to destroy themselves. 
The one month-long struggle of the students was finally ended 
by the military intervention of the Chinese army. 

14. Which of the following did the students not ask for in last 
year's demonstration? 

(A) Democracy 

(B) Free market economy 

(C) Free travel to foreign countries 

15. Who did the police disgrace in last year's Chinese uprising? 

(A) police 

(B) the students 

(C) Either police or the students 

16. By whom was the demonstration finally ended? 

(A) By the students 

(B) By the police 

(C) By the army 

17. Who did the Chinese authorities say would be destroyed by 
the demonstration? 

(A) Chinese leaders 

(B) The students 

(C) The dissidents 

(D) Either the students, Chinese leaders or the dissidents 

Passage B 

The Italian tenor Enrico Caruso was probably the most famous 
singer of all time. He was born in Italy in 1873. After he came 
to the U.S. in 1904, he made a great success at the Metropolitan 
Opera. He had a voice of exceptional beauty with a superb 
technique. But, above all, people loved the soul of his singing. 
When he sang, the music and words were in his breath and in 
his blood. This is something God-given. Actually, he was so 
perfect that every critic considered Caruso to set the standard 
for himself. As a person, he left lots of wonderful stories about 
his generosity and his love for people of all kinds. To poor 
people, he liked to give money and even his coat. In Caruso's 



Cho: The acquisition of English reflexives by Korean ESL learners 6 7 

biography, his friend recalled that his manager once asked 
Caruso to save some money for himself. He suddenly died in 
1921 of lung illness at the age of 48. 

18. How old was Caruso when he came to the U.S.? 

(A) 27 

(B) 3 1 

(C) 4 4 

19. For whom did Caruso set the standard of singing? 

(A) Critics 

(B) Caruso 

(C) Either Caruso or critics 

20. Which of the following was not mentioned about Caruso's 
singing? 

(A) beautiful voice 

(B) superb technique 

(C) good stage presence 

21. For whom did the manager ask Caruso to save some 
money? 

(A) For the friend 

(B) For the manager 

(C) For Caruso 

(D) For Either the manager, the friend or Caruso 



Studies in the Linguistic Sciences 
Volume 21, Number 2. Fall 1991 



DISCOURSE REFERENCE IN WRITTEN KOREAN FOLK TALES 

Yeon Hee Choi 

This study investigates referential chains in written 
Korean narrative, exploring the notion of discourse 
structure. It examines three referential forms (zero 
anaphora, pronouns, and full NPs) in written Korean folk 
tales by analyzing text structure, using a simplified version 
of Fox's rhetorical structural analysis (1987b). The findings 
of the study reveal a dichotomy between zero anaphora and 
noun forms, on the one hand, and pronouns as referential 
devices in Korean, on the other. The distribution patterns in 
terms of referent-types and grammatical coding forms 
reflect the type of the texts analyzed: narrative writing. 
Futhermore, the results of the study show that referential 
choices were not affected by such factors as linear 
referential distance and ambiguity as much as in the 
distance-based approach. Rather, accessibility in terms of 
structural distance and continuity of the semantic role are 
crucial in determining referential devices. The study thus 
concludes that in order to better understand referential 
choices, they should be explored in the hierarchical 
structure of the text, not simply in linear distance. 

1. Introduction 

Reference to characters plays a crucial role in developing a 
narrative text and it also helps to comprehend the text. A clear 
referent and effective referential devices are thus major factors 
determining the textuality of narrative discourse. For these reasons, 
referential expressions have been one of the key issues in studies on 
narrative texts in various languages including Korean (Clancy 1980, 
Givon 1983, Hwang 1983, Lee 1989). Futhermore, contrastive studies 
of written texts have investigated the use of referential devices in 
different languages, because differences in their use are easily 
noticeable across languages (Y. Kachru, 1982; Pandharipande. 1982; 
Tsao, 1982). However, most of these studies have simply counted the 
frequency of different referential devices in order to investigate 
their distribution in or between languages, and their functions. 

Discourse anaphora has also been a main concern in the research 
on Korean text. Some Korean linguists have analyzed discourse 
anaphora in Korean narrative writing, following Givon's (1983) 



70 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

theory of topic continuity (Hwang, 1983). These studies have 
overemphasized, however, the linear nature of texts and have not 
incorporated a hierachically structural view of text. On the other 
hand, Lee (1989) has explored referential choice in Korean discourse 
by analyzing discourse structure. His data include casual 
conversation and personal letters which are different from the type 
of data analyzed in the previously mentioned studies (i.e., narrative 
texts) so that the findings from the two groups of studies, using 
linear and structural approachs respectively, are not really 
comparable. In this research, thus, referential forms in written 
Korean narrative discourse are examined, following Fox's (1987b) 
structure-based approach; it aims at testing the adequacy of the 
linear and structural analysis of discourse anaphora in written 
Korean narrative. 

By investigating referential chains, more specifically identity 
chains,' the present study examines factors affecting Korean writers' 
strategies of referential choices in narrative text which have 
antecedents in the preceding discourse. In the study, referential 
forms are categorized as zero-anaphora, pronouns, and full NPs 
including subtypes. The data analyzed consist of written Korean folk 
tales taken from a Korean textbook. In these texts, the three types of 
anaphora are examined in hierarchical text structure, using a 
simplified version of Fox's rhetorical structural analysis model (Aston 
1977, Tirkkonen-Condit 1985, Choi 1988). 

2 . Distance-based and structure-based analysis of discourse 
reference 

In terms of the way of looking at text, there are two types of 
analysis of reference in narrative discourse. First, a text is viewed as 
a linear sequence of sentences; thus, referential devices are 
examined in terms of linear referential distance and decay, and 
referential ambiguity, 2 in order to understand the linguistic coding of 
the concept topic (Givon, 1983). In the second approach, on the other 
hand, a text is seen as hierarchical relationships of sentences. 
Referential choices are, therefore, analyzed in text structure (Clancy 
1980, Fox 1987a,b). 

In exploring the concept topic, Givon has proposed the iconicity 
principle (the Continuity Hypothesis), which says "The more 
disruptive, surprising, discontinous or hard to process a topic is, the 
more coding material must be assigned to it." (p. 18) The factors 
which influence disruptive or discontinous topics include the distance 
to the referent's last mention, ambiguity from other referents, 
availability of semantic information, and availability of thematic 
information. The first two factors are the major concerns of the 
studies in Givon (1983). These studies present the anaphoric patterns 



Choi: Discourse reference in written Korean folk tales 7 1 

that pronouns are used when the distance to the last mention of the 

referent is small, without interfering referents, while full NPs are 

used when that distance is somewhat great and/or if there are 
interfering referents. 

Referential expressions have also been analyzed in Korean 
narratives, following Givon's approach. Hwang (1983) has 
investigated topic continuity of participant arguments in a written 
Korean narrative text (a short story) to explore Givon's continuum of 
topic continuity in connection with human discourse processing. She 
has found that zero-anaphora is used for the most continuous topic, 
the pronoun for the second most continuous, and finally, full NPs for 
the least continuous. Her findings of the correlation between the 
degree of grammatical encoding and topic continuity in Korean 
support Givon's hypothesis. In addition, her study suggests the 
important roles of other factors affecting referential choice: 
Humanness of antecedents, grammatical functions of referential 
expressions, socio-cultural constraints (e.g., politeness), and 
discourse-pragmatic factors (e.g., the speaker's intentions). 

The studies following Givon's cognitive approach overemphasize 
the linear nature of text and do not incorporate a hierarchically 
structural view of text. On the contrary, Clancy (1980), whose study 
was followed by Givon, had examined discourse anaphora in short 
spoken English and Japanese narratives, exploring the notion of 
discourse structure. The original purpose of her study was to analyze 
referential choice in terms of ambiguity and distance; however, 
besides these factors, she noted the impact of discourse units on the 
use of full noun phrases. She found that the main discourse 
structures influencing referential choice are episode boundaries, the 
beginning of a new line of action, and world shifts, in other words, 
"the speaker moves from one mode of talking to another (e.g., from 
digression to the plot line, or from film-viewer mode to story-teller 
mode). Both of these structure-types tend to be associated with use 
of full noun phrases" (Fox 1987a: 159). 

Likewise, Fox (1987a) has explored the anaphoric patterning in 
written English narratives by analyzing them in hierarchical 
structure. Her study has suggested that the patterns based on the 
structuring functions in narrative, such event-line, plans and actions, 
describe a very large proportion of the anaphors in the narrative 
texts she examined, including in the environment of interfering 
referents (p. 172). Criticizing Givon's linear approach to reference, 
she claims that to understand the use of various referential devices 
in discourse, they must be examined in the structure of the text. 

The structural approach has also been adopted in the research of 
Korean reference. Lee (1989) has analyzed referential choice in 



72 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Korean casual conversation and personal letters from a cognitive and 
social perspective, using both the linear and structural analysis. His 
study brings out the differing distributions of referential forms 
between the two types of discourse, as found between English 
conversation and expository text in Fox (1987b). The study also 
points out the important roles of not only cognitive factors (iconicity 
principle and plannedness)^ but also social factors (text structure and 
collaborative interaction,"* and stance representations). Based on his 
findings, Lee suggests the need of cognitive and social approaches to 
discourse reference including their analysis in discourse structure. 

Referential forms in languages including Korean have been 
analyzed from both distance-based and structure-based approach. 
However, the results of these studies are not really comparable in 
order to determine which approach provides a more accurate picture 
of referential choice since the data are of different types. 
Nevertheless, the structure-based research seems to suggest that the 
writer's choice of one referential device over others can be better 
understood when it is examined in the structure of the text, not only 
in terms of linear distance. 

3. Research design 

3.1 Data 

Five written Korean folk tales were analyzed in quantitative 
terms. Taken from a Korean textbook for foreign students, the folk 
tales were chosen for the study because they are one of the main 
types of narrative discourse analyzed across languages. Although 
they are shorter than other types of narrative texts such as short 
stories and novels, they still clearly have discourse units (e.g., climax) 
within the text; thus, it was expected to see the effect of discourse 
structure and units on referential choice, as found in Clancy (1980) 
and Fox (1987a). 

The five folk tales contained 23 paragraphs, 115 sentences, and 
164 finite clauses excluding nominal and relative clauses. Since folk 
tales are usually told in the third person (the observer), there was no 
first person reference in the data. On the other hand, the texts 
contained four types of reference: human, animal, inanimate, and 
text reference. The title of each text is as follows: 

Senpi-wa toduk ('Scholar and Thief) 

Han Sek-pong-kwa emeni ('Han Sek-bong and His Mother') 

Sicip sali ('Married Life in the Parents-in-law's House") 

Horangi kkori ('The Tail of a Tiger') 

Cheng kaykuli ('The Green Frogs') 



Choi: Discourse reference in written Korean folk tales 73 

3.2 Data analysis 

The five Korean folk tales were analyzed to identify three types 
of referential forms: Zero anaphora, pronoun, and full NP. The 
pronoun form included personal and demonstrative pronouns. The 
noun forms were classified into several subtypes: repeated or 
partially repeated noun forms including names and kinship terms 
{kanan-ha-n senpi 'a poor scholar* — senpi 'the scholar'); lexical 
replacement {'cheng kaykuli hyengcey-tul 'the green frog brothers' — 
ahi-tul 'the children'); possessive-plus-NP {ku cung 'that monk'); 
demonstrative adjective-plus-NP (iren myenuli 'such a daughter-in- 
law'); and NP-plus-relative clause (/ mal-ul tul-un si-apeci 'the 
father-in-law who heard this'). 

The three types of referential forms were examined in the 
hierarchical structure of the text. The structure was analyzed by the 
simplified form of Fox's rhetorical structure analysis (1987b) which 
was modified by adapting the interactive text analysis model (Aston, 
1977; Tirkkonen-Condit, 1985; Choi, 1988). The basic unit of the 
analysis was finite clauses, excluding relative clauses and nominal 
clauses such as complementary clauses of the verb in which 
anaphors are syntactically controlled. 

In the analysis, each text was broken down into finite clauses. 
Any quoted material was not analyzed, although each quote was 
counted as one clause if they appeared as an independent unit, but 
not as part of the main text, such as a quote within a sentence (see 
(1)). After breaking down each text into clauses, the hierarchical 
relationships of clauses were determined by their functional roles 
and generality in topic. No label was given to relationships between 
clauses (e.g., elaboration). 

A sample of the analysis is found in example (1). The text in the 
example is the first paragraph of the folk tale Cheng Kaykuli (Green 
Frogs). It has three rhetorical units narrating temporally situated 
actions; eight clauses (Clauses 1-5 and 9-11) are in the same 
hierarchical level. Clauses 6-8 provide background information for 
the action described in clause 9. In the diagram following the English 
translation, the equivalent English expressions of the tokens of the 
referential forms are provided under each clause. 

{\) Cheng kaykuli ('Green Frogs') 

1) Yeysnal enu cheng kaykuli kacok -i sal -ko iss- 

once upon a lime a green frog family SM live PROG 

ess -upnita. 2) Kurentey cheng kaykuli hyengcy- tul -un 
PAST DEC but green frog broiher PL TOP 



7 4 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

emma -uymal -ul cal tut -ci anh-ass -upnita. 
mother POS speech AOC well listen not PAST rSC 

3) Haru -nun emma cheng kaykuli -ka ahi -tul -eykey 
one day TO* mother green frog SM child PL DAT 

'onul -un param-i pul -ko nalssi -ka nappu-nikka, cip 
today TOP wind SM blow and weather SM bad since house 

an -eyse nol -aya ha-n -ta.' -ko malha -ass -upnita. 
inside LOC play must PRES ffiC IDS say PAST DBC 

4) Kurena cheng kaykuli hyengcey-tul-un 'yay, uri pakk 
but green frog brother PL TCP hey we outside 

-ey naka -se nol -ca.' ha -ko -nun 5) motu pakk -uro 
LOC go out and play PRO say and TCP all outside UOC 

naka- ss -upnita. 6) Ttoharu -nun pi -ka manhi 
go out PAST DBC Another day TOP rain SM a lot 

wa -SS- upnita. 7) Cangma -ka ci -ese. 8)yeki 
come PAST DEC rainy season SM set in because here 

ceki -se yatan- i na -ss -upnita. 9) Emma 

there LCC trouble SM break out PAST CBC mother 

cheng kaykuli -nun ahi -tul- eykey 'mul -i manh-un 
green frog TCP child PL DAT water SM plenty 

kot -ey -nun tomapaym-i iss-unikka, ka-myen khu-n 
place LOC TCP lizard SM be since go if big 

il -i -ta.' ha-ko malha-ass -upnita. 10) I mal-uh 
event be raC IDS say PAST DBC this speech 

tul -un cheng kaykuli hyengcey-tul-un ipen- ey-to 
hear PAST green frog brother PL TCff* this time again 

'yay, uri tomapaym kukyeng-ka-ca.' ha -ko -nun 
hey we lizard see go PRO say and TCP 

11) motu pakk -uro naka -ss -upnita. 
all outside LOC go out PAST LBZ 

1) Once upon a time, there was a green frog family. 2) The 
frog brothers did not listen to their mother. 3) One day the 
mother green frog said to her children, 'Since it is very windy 
today and the weather is not so good, don't go out to play but 



Choi: Discourse reference in written Korean folk tales 75 

stay at home." 4) But, the green frog brothers said, 'Let's go 
outside to play.' and 5) all of them went out. 6) Another day, it 
poured rain. 7) Because the rainy season set in, 8) people were 
in trouble here and there. 9) The mother green frog said to her 
children, 'Since there are lizards at the wet place, don't go there. 
It is very dangerous.' 10) The green frog brothers, who heard 
this, said this time again, 'Let's go to see the lizards' and 11) 
then they all ran out. 

/\ 
/\ / \ 

/\ / \ / \ 

/ \ / / \ / /\ 

/ \ / / \ / / \ 

1) 2) 3) 4) 5) I 9) 10) 11) 

the green frog the mother the (the I the mother this/ (the 

brothers/ green frog/ green green frog l_ green the green 

their mother her children frog brothers) 6) \ frog/her green frog 

brothers I _ children frog brothers 

7) \ brothers 
I 
8) 

After identifying the text structure and referential forms, tokens 
of the forms which were classified into four categories in terms of 
the types of their referents (human, animal, inanimate, and text)^ 
were quantitatively analyzed in terms of the following aspects: 1) 
frequency distribution with respect to the grammatical functions 
(e.g., topic or subject)'' and the type of clause (initial and non- 
initial);8 2) the number of intervening clauses and sentences 
separating the two mentions; 3) the presence of intervening 
referents of same and different genders; and 4) text-structural 
factors (e.g., starting a new unit). 

4. Results and discussion 

4.1 Frequency of referential forms and types of referent 

Tokens of the three types of referential forms were counted in 
terms of the type of referent. As presented in Table 1, the most 
common referential form was noun forms (66%); the frequency of 
zero anaphora (31%) was relatively high. On the other hand, 
pronouns were least frequently used (3%). This distribution pattern 
has also been noted in another type of written Korean narrative 
(Hwang's analysis of a Korean short story (1983)), while it does not 
hold true in Lee's (1989) study of Korean conversation and personal 
letters where zero anaphora occurred most frequently. 



76 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Table 1 

Frequency of Referential Forms and Types of Reference 

Human Animal Inanimate Text Total 



Zero 



56 (38%) 3 (27%) 2 (8%) (0%) 61 (31%) 



Anaphora 

Pronoun (0%) (0%) 3 (13%) 3 (27%) 6 (3%) 



Full NP 93 (62%) 8 (73%) 19 (79%) 8 (73%) 128 (66%) 



Total 



149 



24 



195 



The high occurrence of full NPs and the rare occurrrence of 
pronouns contrasts with the dominance of pronominal reference in 
various types of English discourse (Brown, 1983; Fox, 1987b; Choi, 
1992). Such a skewed distribution pattern can be accounted for by 
the formal features of pronouns and zero anaphora, and the socio- 
cultural constraints on their use in Korean. Korean is a pro-drop 
language in which not all constituent elements of a sentence (e.g., 
subject or object) have to be on the surface, in other words, their 
presence is not obligatory. Zero anaphora thus can be used where 
pronouns would be most acceptable in English, without jeopardizing 
the grammaticality of the sentence. In addition, Korean pronouns are 
coded in diverse forms, some of which cannot be easily distinguished 
from nouns (e.g., ku saram 'the person'); the third-person singular 
personal pronoun ku is used for both male and female referent so 
that it can cause referential ambiguity. Korean pronouns are, 
moreover, similar to English stressed pronouns, which are expected 
to carry some contrastive meaning or emphasis. As a result, 
pronouns tend to be avoided as much as possible in Korean. Zero 
anaphora or full NPs are instead chosen over them, depending on the 
explicitness of the referent and the writer's strategy of presenting 
his/her message. 

The overall distribution pattern of referential devices exhibits 
some common points and variations across the four types of 
reference. A huge difference was found between the total tokens of 
human and non-human reference, which may be due to the feature 
fo the narrative texts analyzed that humans were the topics of 
narration. On the other hand, the distribution of the three referential 
devices was similar across the four types of referent, except for the 



Choi: Discourse reference in written Korean folk tales 7 7 

high occurrence of pronouns for inanimate and text referents. The 
high frequency of full NPs in all types of referent does not support 
Lee's statement that "human referents are very topic continuous and, 
therefore, are encoded by zero-anaphora without regard to the 
specific type of speech activity." (Lee 1989:103) 

Another variation was noted between animate (human and 
animal) and inanimate (inanimate and text). Pronouns were never 
employed for animate referents in the texts analyzed for this study, 
while they were relatively frequent for inanimate referents. This 
seems to suggest a contrast between the role of zero anaphora and 
nominal forms, and of pronouns as referential devices in Korean. 

For text reference, zero anaphora was never used, while the 
frequency of pronouns was relatively high. The distinctive 
referential pattern for text reference suggests the need of more 
specified classification of reference than just a dichotomy such as 
human/non-human or animate/inanimate in investigating the 
writer's referential strategies. 

4.2 Clause types and case roles 

The frequency of the three types of referential devices were 
counted in terms of the type of clause and their grammatical coding 
forms. The frequency distribution varied with these two factors. 
Nominal forms were most frequent at the initial clause across the 
four types of referent, while zero anaphora occurred most frequently 
in the non-initial clause (NI), except for inanimate and text reference. 
The latter finding is not very surprising because in Korean ellipsis is 
common when the antecedent is mentioned in the previous clause, 
especially within a sentence. 

Another dichotomy was present between topic and non-topic. 
Most of the tokens of human and animal reference were in topic 
positions, whereas inanimate and text reference were more frequent 
in non-topic positions, specifically in object positions. This indicates 
that animate referents, humans and animals, tend to be agents and 
thus appear in the topic position; on the other hand, inanimate 
referents including text are typically the recipient rather than the 
actor, appearing in the object position, the differences between 
animate and inanimate reference have also been noted in Hwang 
(1983), which found the high occurrence of human reference in 
subject positions contrasted with that of non-human reference in 
accusative and oblique positions. 



78 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



Table 2 Frequency of Referential Forms, Clause 
Types, and Case Roles 

Topic Subject Accusative Dative Genitive Oblique 
Init. NI Init. NI Init. NI Init. NI Init. NI Init. NI 



Human 


























Zero Anaphora 


9 


36 


2 


1 





4 


2 


2 














Pronoun 






































Full NP 


50 


5 


9 


1 


8 


1 


6 


1 


9 


1 


2 





Animal 


























Zero Anaphora 





2 


1 





























Pronoun 






































Full NP 


3 








1 


1 











3 











Inanimate 


























Zero Anaphora 





1 


























1 





Pronoun 


2 














1 




















Full NP 


1 


1 


1 





1 1 











1 





4 





Text 


























Zero Anaphora 






































Pronoun 














2 




















1 


Full NP 














6 











1 





1 





Init.: Initial 


NI: 


Non 


initial 





















4.3 Referential distance and text structure 

The distance to the most recent mention of a given referent for 
each referential form was measured by counting the number of 
clauses from a token to the nearest clause where the referent 
appeared. The average distance for pronouns was shortest, and that 
for noun forms longest, as shown in Table 3. All referents of 
pronominal forms (inanimate and text) were present in the 
immediately preceding clause. This finding was surprising because 
the shorter referential distance had been expected for zero anaphora, 
as found in other studies on Korean discourse (Hwang 1983, Lee 
1989). This may be due to the absence of pronouns for animate 
referents in the data. 



The average distance for noun forms was greater than for the 
two other forms. However, it was much shorter than that found in 
Hwang (1983:71) in which the average distance for relative clause + 
NP was 16.98 for non-human referents; that for names, 5.6 for 
human. Such a discrepancy can be explained by the differences in 



Choi: Discourse reference in written Korean folk tales 



79 



the type of data: The present study analyzed short folk tales, while 
Hwang's data was a short story containing about 4100 clauses where 
many characters and inanimate objects are introduced and then 
reintroduced through the changes of scenes. 

A variation across the referent types was also noted in terms of 
referential distance. The referential distance was the longest in the 
case of zero anaphora and full NPs for inanimate referents. On the 
other hand, that of nominal forms was shortest for human referents. 
These results seem to support Givon's claim about the higher 
topicality of human/animate/agents. It cannot explain, however, the 
short distance for text referents (1.00 for pronouns and 2.25 for full 
NPs), which are also inanimate. The findings of the present study 
thus point out that referential distance itself cannot account for topic 
continuity as accurately as expected in distance-based studies. 

Table 3 Referential Distance of Referential Forms 

Human Animal Inanimate Text Average 



Zero 



Anaphora 


1.16 


1.00 


2.50 


0.00 


1.55 


Pronoun 


0.00 


0.00 


1.00 


1.00 


1.00 



Full NP 



2.08 



4.00 



5.18 



2.25 



Average 1.62 



2.50 



2.89 



1.63 



1.9! 



In contrast to Givon's iconicity principle and Lee's (1989) 
findings that zero anaphora is most frequently used to encode 
referents in the previous clause in Korean personal letters and 
conversation, the present study also found, as shown in Table 4, that 
full NPs were frequent when the referent was mentioned in the 
immediately preceding clause. For human and animal reference with 
the antecedent in the preceding clause, zero anaphora and noun 
forms were used in the exactly same frequency (50% both); for 
inanimate and text reference, the use of full NPs was much greater 
(60% and 67%, respectively) than that of other referential forms. 



80 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



Table 4 Referential Forms and the Scope of Referents 





Referent in 


Referent not 


Referent in 


Referent not 




Preceding 


in 


Preceding 


Preceding 


in preceding 




Clause 


CI 


ause 


Sentence 


sentence 


Human 












Zero Anaphora 


52 (50%) 




4 (9%) 


11 (17%) 


3 (9%) 


Pronoun 


(0%) 




(0%) 


(0%) 


(0%) 


Full NP 


51 (50%) 




43 (91%) 


55 (83%) 


33 (92%) 


Animal 












Zero Anaphora 


3 (50%) 




(0%) 


(0%) 


(0%) 


Pronoun 


(0%) 




(0%) 


(0%) 


(0%) 


Full NP 


3 (50%) 




5 (100%) 


3 (100%) 


4 (100%) 


Inanimate 












Zero Anaphora 


1 (10%) 




1 (7%) 


(0%) 


1 (8%) 


Pronoun 


3 (30%) 




(0%) 


2 (22%) 


(0%) 


Full NP 


6 (60%) 




13 (93%) 


7 (78%) 


11 (92%) 


Text 












Zero Anaphora 


(0%) 




(0%) 


(0%) 


(0%) 


Pronoun 


3 (33%) 




(0%) 


2 (25%) 


(0%) 


Full NP 


6 (67%) 




2 (100%) 


6 (75%) 


2 (100%) 



When the distribution of referential forms was examined in terms of 
the location of the referent in the sentence level, the frequency of 
noun forms increased across all types of reference. Such referential 
forms were chosen, however, more frequently when the referent was 
in the immediately preceding sentence than when there was more 
than one intervening sentence. This result indicates that referential 
devices perform more than just connecting a reference item with its 
antecedent, as claimed by Fox (1987b: 144). 

... full NPs in the written texts are doing much more than just 
the standard referent-tracking work attributed to them: that is, 
if they occur even when their referent is plainly retrievable 
from the preceding clause, then they are not simply performing 
an anaphoric duty. Rather, they are helping to block the text 
into its structure units. 

In order to see the impact of text structure on the choice of 
referential devices, the tokens of reference were counted in five 
types of text-structure features: referent in the same rhetorical unit, 
return pop,^ referent in the same unit that is unaccessible, referent 
in the closed unit, and starting a new rhetorical unit. When the 
referent was in the same unit, variations were noted across the four 



Choi: Discourse reference in written Korean folk tales 



81 



types of reference. For human reference, zero anaphora was 
relatively more preferred than noun forms, while such preference 
was not noted when the referent was animal. On the other hand, 
noun forms were much more frequently used for inanimate 
referents. Pronouns were of the most favorable for text reference. 

Table 5 Referential Forms and Position in Rhetorical Unit 



Same 
Unit 



Return 
Pop 



Same 
Unit-Closed 



New 
Unit 



Closed 
Unit 



Human 

Zero Anaphora 48 (60%) 5 (19%) 1 (20%) 

Pronoun (0%) (0%) (0%) 

Full NP 32 (40%) 22 (81%) 4 (80%) 



1 (6%) 1 (5%) 

(0%) (0%) 

16 (94%) 18 (95%) 



Animal 

Zero Anaphora 3 (50%) (0%) (0%) (0%) (0%) 

Pronoun (0%) (0%) (0%) (0%) (0%) 

Full NP 3 (50%) 1 (100%) (0%) (0%) 4 (100%) 



Inanimate 

Zero Anaphora 1 (13%) (0%) (0%) 

Pronoun 2 (25%) (0%) (0%) 

Full NP 5 (63%) 4 (100%) (0%) 



(0%) 1 (11%) 

(0%) (0%) 

2 (100%) 8 (89%) 



Text 

Zero Anaphora (0%) (0%) (0%) 

Pronoun 3 (60%) (0%) (0%) 

Full NP 2 (40%) (0%) (0%) 



(0%) (0%) 

(0%) (0%) 

2 (100%) 4 (100%) 



Except for the first type of text-structural position, however, full 
NPs were of the most frequent across all types of reference. These 
results show that in the folk tales analyzed, noun forms were 
selected over the other types when the referent was not accessible in 
the same rhetorical unit. In the studies on English discourse (Fox 
1987b, Choi 1992), pronouns have been found to be the most 
favorable referential form for return pops. In this study, human, 
animal, and inanimate return-popped reference tended to be coded 
in noun forms. Likewise, full NPs were preferred when the popped- 
over material was structurally complex, or did not contain mentions 
of the referent; when the writer tended to create the effect of 
beginning a new unit; or when it was in other rhetorical units and 
thus unaccessible. These findings suggest that referential choices are 
more sensitive to structural distance than to referential distance. 



82 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



4.4 Referential ambiguity 

Since the presence of interfering referents often determine the 
referential choice, the tokens of reference were counted in terms of 
the presence of other third-person animate referents of the same or 
different genders. This quantitative analysis was done for human, 
animal, and animate reference. When there were no interfering 
referents between the referential form and the antecedent (NIR), full 
NPs were the most common across the three types of reference, 
while for human reference the percentage of zero anaphora was as 
high as noun forms (45% and 55%, respectively). Noun forms were 
also chosen when there were interfering different-gender (DG) as 
well as same-gender (SG) referents. The high frequency of full NPs 
when containing no intervening referents has not been noted in 
other studies on Korean discourse (Lee, 1989; Hwang, 1983) as well 
as on English discourse (Fox, 1987b; Choi, 1992). In such cases, 
rather, the most frequent type was zero anaphora in the former and 
pronouns in the latter. 

Table 6 Referential Forms in the Environment of 
Interfering Referents 



NIR 



DG 



SG 



Role Role 

Continuity Change 



Human 

Zero Anaphora 40 (45%) 7 (23%) 9 (30%) 

Pronoun (0%) (0%) (0%) 

Full NP 49 (55%) 23 (77%) 21 (70%) 



47 (70%) 9 (11%) 

(0%) (0%) 

20 (30%) 73 (89%) 



Animal 

Zero Anaphora 3 (30%) (0%) (0%) 

Pronoun (0%) (0%) (0%) 

Full NP 7 (70%) (0%) 1 (100%) 



3 (75%) (0%) 

(0%) (0%) 

1 (25%) 7 (100%) 



Inanimate 

Zero Anaphora 1 (7%) (0%) 1 (11%) 

Pronoun 3 (20%) (0%) (0%) 

Full NP 11 (73%) (0%) 8 (89%) 



2 (17%) (0%) 
1 (8%) 2 (17%) 

9 (75%) 10 (83%) 



In Choi's (1992) study on written English narrative, grammatical 
or semantical roles of reference have been noted as the crucial factor 
affecting anaphora choices in interfering same- and different-gender 
referents. In this study, thus, the distribution patterns of referential 
forms were analyzed in terms of role continuity or change, in other 
words, whether the reference had the same grammatical or 



Choi: Discourse reference in written Korean folk tales 83 

semantical role as its referent. Regardless of the type of reference, 
zero anaphora was most frequently chosen when the semantic role 
was constant, whereas noun forms were preferred when the role 
changed. This suggests that in Korean narrative text the 
continuity/discontinuity of the semantic role has a significant impact 
on referential choices. 

5. Conclusion 

The distribution of three types of referential forms has been 
explored in written Korean folk tales with respect to reference-type, 
clause type and case roles, linear referential distance, the position in 
the rhetorical unit, and the environment of interfering referents. 
First, zero-anaphora and full NPs were much more frequent 
referential forms than pronouns. The low frequency of pronouns 
shows that pronominal forms are not important referential forms in 
Korean, compared to zero-anaphora and full NPs, which is due to the 
formal features of pronouns and the socio-cultural constraints on 
their use in Korean. In addition, another contrast between zero 
anaphora and full NPs, on the one hand, and pronouns, on the other, 
was noted in terms of referent-types. The former referred to 
inanimate. The largest tokens for human reference as well as this 
difference reveal the strong impact of animacy or the ability of an 
entity that performs an action on referential choice in Korean 
narrative writing and also in narrative text in general (Y. Kachru 
1982, Choi 1992). 

Next, the analysis of the three types of referential devices in 
terms of clause type and their grammatical coding forms has also 
revealed a dichotomy between animate and inanimate reference. 
This dichotomy reveals a characteristic of narrative text: Animate 
entities are usually actors, the main characters in the story, so that 
they appear in the topic or subject position. 

Third, in contrast to the expectation of the shortest referential 
distance for zero anaphora, pronouns had the shortest distance, and 
the distance for noun forms was longest. A variation across the 
referent types was also noted in terms of referential distance. The 
distance for inanimate referents was greatest; that for human 
referents was relatively short. However, text reference also exhibited 
short distance. Furthermore, full NPs were frequent when the 
referent was mentioned in the immediately preceding clause. When 
the anaphoric patterns were analyzed in text structure, it was found 
that noun forms were selected over the other types when the 
referent was not accessible in the same rhetorical unit. These 
findings thus suggest that referential choices are more sensitive to 
structural distance than to referential distance. Referential distance 
itself cannot account for topic continuity as accurately as expected in 



8 4 Studies in the linguistic Sciences 21:2 (Fall 1991) 

distance-based studies, because referential devices perform more 
than an anaphoric duty. 

Finally, the presence of more than one referent of same or 
different gender triggered changes in referential choice; however, not 
just their presence but the continuity/discontinuity of the 
grammatical and semantic role of the reference was crucial in 
determining which reference took zero anaphora. 

This study has shown that referential forms were not chosen as 
simply as Givon states in his Topic Continuity Hypothesis. Unlike 
Hwang (1983), no significant correlation was found between the 
degree of grammatical coding and referential distance. Referential 
choices were affected by various factors including reference-type, 
structural distance, and continuity of semantic roles. Structure-based 
analysis can account for the use of referential devices in written 
Korean folk tales more accurately than distance-based analysis, even 
though linear referential distance cannot be completely ignored. In 
order to have an accurate picture of the writer's referential 
strategies, thus, referential choices should be analyzed not only 
linearly but also structurally. 

The results of the study have exhibited some common points 
with Hwang's study on a Korean short story, because both studies 
have analyzed narrative text. However, the similarities are 
overridden by the differences in their findings. Thus, future studies 
on various types of narrative prose are needed to have a better 
generalization of anaphoric patterns exhibitied by all types of Korean 
narrative writing. Such studies may find variations within one text- 
type. In addition, if other types of Korean discourse are explored in 
text structure, insight can be provided into referential forms used in 
Korean in general. It can also help Korean students, who are not 
usually taught how to compose in Korean, to write a better text in 
their native language. 

NOTES 

1 Identity chains are one of the two types of cohesive chain 
proposed by Halliday and Hasan (1985). A cohesive chain is "a set of 
items each of which is related to the others by the semantic relation 
of co-reference, co-classification, and/or co-extension" (p. 84). The 
items related to each other by co-reference form an identity chain: 
"every member of the chain refers to the same thing, event, or 
whatever" (p. 84). 

2 Referential distance is the distance to the most recent mention 
of a given referent. Referential decay is the measurement of distance 
forward: the number of following clauses where the given referent is 



Choi: Discourse leference in written Korean folk tales S5 

mentioned. Referential ambiguity refers to the presence of possible 
referents besides the actual referent (Givon 1983: 13-15). 

3 Ochs (1979) defines planned discourse as "the discourse that 
has been thought out and organized (designed) prior to its 
expression" (p. 55). 

4 The term collaborative interaction in Lee (1989) is based on 
Grice's Cooperative Principle. In other words, the speaker's 
referencing process is a collaborative process not to break down 
communication but to continue it. 

5 Stance representation is a concept from Grimes (1975). In Lee 
(1989) it is defined as "verbal behavior representing the speaker's 
(or writer's) affective and epistemic altitudes toward a referent, or 
towards his/her relationship with hearers through the act of 
referencing in the context of interaction" (p. 271). 

6 When animals were the main characters with no human 
character, their reference was counted as human in the study (e.g.. 
Cheng kaykuli 'Green Frogs'). 

7 The grammatical functions of each referential form were 
determined by the case marker following it. 

8 In the analysis of English data, clauses are classified into main 
and non-main including subordinate, participial, relative clauses 
(Givon 1983, Brown 1983); this set of categories has also been used 
in Hwang's (1983) study of written Korean narratives. In Korean, 
main clauses always appear at the end of the sentence unless 
inversion occurs for special effects. Consequently, a great deal of zero 
anaphora can occur in the clauses including the main clause after the 
first. In order to obtain better information about referential choices 
when a sentence consists of more than one clause, the occurrence of 
reference at the initial clause and in the following clauses was 
differentiated in the present study. 

9 Return pop is a sentence which ties back to an earlier, a 
superordinate sentence other than the immediately preceding one. 

REFERENCES 

ASTON, Guy. 1977. Comprehending value: Aspects of the structure of 

argumentative discourse. Studi Italiani di Linguistica Teorica ed 

Applicata 6.465-509. 
Brown. Cheryl. 1983. Topic continuity in written English narrative. 

Topic continuity in discourse, ed. by Talmy Givon, 313-341. 

Amsterdam: Benjamins. 
Brown, Gillian, & George Yule. 1983. Discourse analysis. Cambridge: 

Cambridge University Press. 



86 Studies in the Linguistic Sciences 21:2 (FaU 1991) 

Chafe, W. 1976. Giveness, contrastiveness, definiteness, subjects, 

topics and point of view. Subject and topic, ed. by C. Li, 25-56. 

New York: Academic Press. 
Chang, Suk Jin. 1982. Linguistics and written discourse in particular 

languages: English and Korean. Annual Review of Applied 

Linguistics 3.85-98. 
Choi, Hyenbay. 1971. Uri malbon (A grammar of our language), 4th 

ed. Seoul: Chungumsa. 
Choi, Yeon Hee. 1988. Textual coherence in English and American: An 

analysis of argumentative writing by American and Korean 

students. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 
1991. Discourse reference in Korean students' writing in English. 

Paper presented at the 25th Annual Convention of TESOL, New 

York. 
1992. Co-referential forms in English narrative text. Eungyong 

Enehak (Journal of the Applied Linguistics Association of Korea) 

5.63-88. 
Clancy, Patricia. 1980. Referential choice in English and Japanese 

narrative discourse. The pear stories: Cognitive, cultural, and 

linguistic aspects of narrative production, ed. by W. Chafe, 127- 

201. Norwood, NJ: Ablex. 
CONNOR, Ulla. 1987. Research frontiers in writing analysis. TESOL 

Quarterly 21:4.677-696. 
Fox, Barbara. 1987a. Anaphora in popular written English narrative. 

Coherence and grounding in discourse, ed. by Russell S. Tomlin, 

157-174. Amsterdam: John Benjamins Publishing. 

1987b. Discourse structure and anaphora: Written and 

conversational English. Cambridge Studies in Linguistics 48. 

Cambridge: Cambridge University Press. 
GIVON, Talmy. 1983a. Topic continuity in discourse: An introduction. 

Topic continuity in discourse, ed. by Talmy Givon, 1-41. 

Amsterdam: Benjamins. 
(ed.) 1983b. Topic continuity in discourse. Amsterdam: 

Benjamins. 
Grimes, J. 1975. The thread of discourse. The Hague: Mouton. 
Halliday, M. a. K., & Ruqaiya Hasan. 1976. Cohesion in English. 

London: Longman. 
1985. Language, context, and text: Aspects of language in a 

social-semiotic perspective. Oxford: Oxford University Press. 
Hankuke Hakdang of Yonsei University (The Korean Language Center 

of Yonsei University), ed. 1989. Hankuke Tokpon: Chokup 

(Beginning Korean). Seoul: Yonsei University Press. 
Hinds, John. 1975. Korean discourse types. The Korean language and 

its structure, ed. by Ho-min Sohn, 81-90. Hawaii: The Center for 

Korean Studies, University of Hawaii. 



Choi: Discourse reference in written Korean folk tales 87 

1983. Topic continuity in Japanese [1]. Topic continuity in 

discourse, ed. by Talmy Givon, 43-93. Amsterdam: Benjamins. 
Huh, Wong. 1984. Kukehak: Uri mal-ui onul-kwa oje (Korean 

linguistics: The modern Korean and the history of the Korean 

language). Seoul: Sam Munhasa. 
Hwang, Myongok. 1983. Topic continuity in Korean narrative. Journal 

of Korean Linguistics 3.47-79. 
KaCHRU, Yamuna. 1982. Linguistics and written discourse in English 

and Hindi: A contrastive study. Annual Review of Applied 

Linguistics 3.56-77. 
Lee, Won-phyo. 1989. Referential choice in Korean discourse: 

Cognitive and social perspective. University of Southern 

California Ph.D. dissertation in Linguistics. 
Ll, Charles N., & Sandra Thompson. 1976. Subject and topic: A new 

typology of language. Subject and topic, ed. by Charles N. Li & 

Sandra Thompson, 457-489. New York: Academic Press. 
OCHS, Elinor. 1979. Planned and unplanned discourse. Discourse and 

syntax, ed. by Talmy Givon, 51-80. New York: Academic Press. 
Pandharipande, Rajeshwari. 1982. Linguistics and written discourse 

in particular languages: Contrastive studies: English and Marathi. 

Annual Review of Applied Linguistics 3.118-136. 
Park, Sa-yeol. 1985. Topic pronoun and grammatical agreement. 

Yenge-yengmunhak-yenku (The Journal of English Language 

and Literature: Chungcheng Branch) 28.331-352. 
1986. Lexical-replacements in co-referential chains: Focused on 

Korean. Yenge-yengmunhak-yenku (The Journal of English 

Language and Literature: Chungcheng Branch) 28.237-264. 
TIRKKONEN-CONDIT, S. 1985. Argumentative Text Structure and 

Translation. University of Jyvaskyla, Finland, Ph.D. dissertation 

in Linguistics. 
TSAO, F. F. 1982. Linguistics and written discourse in English and 

Mandarin: A contrastive study. Annual Review of Applied 

Linguistics 3.99-117. 



iJ 



Studies in the Linguistic Sciences 
Volume 21. Number 2, Fall 1991 



MORAIC REPRESENTATION OF AMBISYLLABICITY: 
EVIDENCE FROM KOREAN* 

Seok Keun Kang 

The purpose of this paper is to consider the formal 
representation of 'ambisyllabicity' in terms of moraic 
phonology. Presenting evidence from Korean, first, I claim 
that ambisyllabicity DOES exist; i.e., the notion of 
ambisyllabicity capturing the shared feature of a consonant 
has a real intuitive appeal. Second, I also claim that 
ambisyllabicity and gemination are not properties which are 
in complementary distribution among languages, and that 
they should be given different representations from each 
other. Finally, 1 show that the moraic representation of 
ambisyllabicity is preferred over the CV representation. 

1. Introduction 

There has been much discussion in the literature on the formal 
representation of ambisyllabic consonants since a multi-tiered 
phonological representation was introduced in the field of phonology. 
Kahn (1976) argues for the representation of ambisyllabicity in (la), 
explaining the English flapping which changes, for instance, /t/ into 
[D]. Claiming that all phonetic properties characteristic of 
ambisyllabic segments are derived from their coda status as in (lb), 
Selkirk (1982:355) believes that ambisyllabicity does not exist. 
Borowsky et al. (1984:34), on the other hand, assert that the formal 
representation of ambisyllabicity is identical to that of gemination as 
in (Ic). Furthermore, they claim that ambisyllabicity and true 
gemination are properties which are in complementary distribution 
and that phonetic rules interpret the representation with reference 
to the phonology of the particular language. 

(1) a. $ $ b. $ $ c. \ / 

\ / \ ;»f C C 

t C \/ 

I t 

t 

Kahn (1976) Selkirk (1982) Borowsky et al. (1984) 

In what follows, I will claim that unlike Selkirk's assertion, 
ambisyllabicity DOES exist. I will present some evidence from Korean 
which shows that phonetic properties characteristic of ambisyllabic 



C€> 



90 Studies in the Linguistic Sciences 2 1 :2 (FaU 199 1) 

segments are not derived from their coda status nor from their onset 
status. In addition, I will also show that unlike Borowsky et al.'s 
assertion, ambisyllabic and geminate consonants should be 
represented differently from each other, for both configurations 
occur in language, e.g. Korean. 

2. Ambisyllabicity 

Selkirk (1982) argues that all phonetic properties typical of 
ambisyllabic segments are derived from their coda status. Thus she 
formulates the English flapping rule as resyllabification in (lb) 
above. As Borowsky et al. (1984) point out, however, the 
resyllabified structure in (lb) violates a universally observed 
preference for onsets. In addition, phonetic properties of 
ambisyllabic segments are not derived from their coda status, as will 
be discussed below. 

In Korean, there is evidence that certain consonants function as 
neither the onset of a syllable nor as the coda of the preceding 
syllable. A syllable-initial lateral /I/, for instance, becomes [n], as 
shown in (2). 

(2) /kamlam/ -^ [kamnam] 'olive' 

/kuklip/ -^ [kuTjnip] 'government established' 
/lakwan/ -> [nakwan] 'a paradise' 

/loin/ -> [noin] 'an old man' 

The /l/-Nasalization rule can be formulated as follows. 

(3) /l/-Nasalization 

/ \ 
/ 1 / ^ [n] / V 

Rule (3) reads that the lateral /I/ is nasalized into [n] in the syllable- 
initial position. 

But it is interesting to note that an intervocalic lateral /I/ does 
not undergo /l/-Nasalization but rather changes into [r], which is 
illustrated in (4). 



(4) /malu/ 


-^ 


[maru] 


'a wooden floor' 


/tali/ 


— > 


[tari] 


•leg- 


/suli/ 


-^ 


[suri] 


'repair' 


/polu/ 


— > 


[poru] 


'a fort' 



Of interest here is that /I/ is not subject to rule (3) even if it appears 
to satisfy the structural description of the rule. Syllabification of 
/tali/ 'leg; bridge', for example, will proceed as follows. 



S. K. Kang: Morale representation of ambisyllabicity 9 1 

(5) $ $ 

I I 

II /\ /\ 

t a 1 i (UR) tali (Syllabification) 

In (5), /I/ is syllabified as the onset of the second syllable, which is 
predicted by the Onset First Principle. 

(6) The Onset First Principle 

a. Syllable-initial consonants are maximized to the extent 
consistent with the syllable structure conditions of the 
language in question. 

b. Subsequently, syllable-final consonants are maximized 
to the extent consistent with the syllable structure 
conditions of the language in question. 

(Clements & Keyser 1982:37) 

Despite the fact that (5) meets the structural description of l\l- 
Nasalization after syllabification takes place, it does not undergo the 
rule. The l\l in the example is phonetically realized as [r], not as [n] 
or [1]. Selkirk's (1982) resyllabification analysis given in (lb), which 
would resyllabify l\l in (5) into the coda of the preceding syllable, 
cannot hold here, for an allophone [r] of IM is not allowed to occur in 
the coda position. Put differently, the /I/ in (4) is not the onset of the 
second syllable nor the coda of the first syllable, but rather it should 
be interpreted as an ambisyllabic segment, which is produced by the 
ambisyllabification rule given in (7). 

(7) Ambisyllabification 

V 1 V 

The structural description of the /IZ-Nasalization rule is no longer 
met in (5) after the Ambisyllabification takes place. This is 
predictable by the so-called Linking Constraint. 

(8) Linking Constraint 

Association lines in structural descriptions are 
interpreted as exhaustive. 

(Hayes 1986) 

As can be seen in (3), the structural description of the l\l- 
Nasalization rule requires a unique association line between the 
segment l\l and the mora V'- But the ambisyllabification associates 
the segment l\l with two moras. Thus the ambisyllabified segment 



92 StmiesiBAeLBgiBSOc Sciences 21:2 (Fan 1991) 

1' no longer meets the stmctural description of the /lANasalization 
rule. Rather it is subject to the following rule, by which the 
ambisyllabic segment /!/ is changed into [r]. 

(9) H \i 
/ \ / \ 

/I/ -^ [r] / V V 

It is clear from the above that unlike Selkirk's (1982) assertion, the 
notion of ambisyllabic it>' capturing the shared feature of a consonant 
has a real intuitive appeal. 

Now. let's turn to how ambisyllabic and geminate consonants are 
represented. Do they have the same representation or different 
representations? According to Borowsky ei al. (1984), the formal 
representation of ambisyllabicitv is identical to that of gemination. 
They assert that ambisyllabicity and gemination are properties 
which are in complementary distribution and that any phonetic 
differences between them are predicted by phonetic rules, which 
interpret the representation with reference to the phonology of a 
particular language. In what follows, 1 will show that this is not the 
case. That is. ambisyllabicity and gemination should be represented 
in different ways, and that they are not in complementary 
distribution. 

There is evidence that in Korean, there exist both ambisyllabic 
and geminate consonants, which are phonetically realized differently 
from each other. Consider the following examples. 

(10) a. /msli/ -^ [mari] head' 
[malli] far' 
[in] 'here' 
[illi] some reason' 
[irida] 'to arrive' 
[illida] 'to inform' 

(lOa-c) are examples that show ambisyllabicity. The syllabification of 
/mali/ 'head', for instance, proceeds as follows. 

(11) $ S 

I I 

i I /\ /\ 

mali (UR) -^ mali (Syllable Association 

& Onset Rule) 

The oatpat in (11) first undergoes the Ambisyllabificatioo role in (7). 
and then the rule in (9) which changes /U into [r], as shown below. 



0>a. 


/mali/ 


a'. 


/malli/ 


b. 


/ili/ 


b'. 


/illi/ 


c 


/ilita/ 


c'. 


/illita/ 



C12) 



S 

I 

u 



s s 



93 



m 9 1 





S S 

1 1 


ri] 


/\\ /\ 
mar i 



As Ao wm m ( 12). 
man vwlexljn^y I 
^jflafoificatkM. la < 
aifjgrd a waan is tke 




(/I/-^[rl) 



/I/ IS Mt 



1991aV la 

below. 



/I/ m (lOa'-c-) is 
(HafB 19». Kane 

coatra$t to ^b^H/. ^aaffi/ 'aCv^ is syilaiitf»d as skon 



(13) 





S S 

1 1 


M P »i 
1 1 1 

mall aiii 


i 1 

- = 3 I i 


s s 




N 1 
H Pit 

/I in 

-^ aia 1 i 


(CbdaRafe) 



(S>ilabie AsKici^oa 
& Omsa Rafe» 



la (13K the ymimm rtmmmmt A/ is fin 
Ike secoMi sjllaMe by *e Ouet Rale md *ea its aon is 
k witk tke precedBS sylUUe by Ae Coda iMe. The fia^ o 
' does Boc Mxt the stractaral coadhioa of the fVSi 

aor thai of the /I/ -» [i] nric- k ■< dev froa (12) aad (13> thM 
aarinsyflabicity aad gcauaatioa shoald aot have the ideatical 
repmeatatioa. This iwcscals c w i d tat*, asaiast Borawsky et aL's 
(19&4) claiai that the fofaal wpwwa i a i i oa of iMJiijIlibiiirj is 
ideatical to that of geauaatioa. Tkat is, their assettioa that 
ua^isyllabicity aad trae geaaatioa are propeities that are ia 

itractmres. 



9 4 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Some evidence for ambisyllabicity is also found in Middle 
Korean. The bilabial voiced fricative /p/ of a Humble morpheme 
/sap/, for example, occurs neither syllable-initially nor syllable- 
finally (Lee 1975, Back 1985). In (14), therefore, the intervocallic 
/p/ is first syllabified as the onset of a syllable by the Onset First 
Principle in (6), and then it is also associated with the preceding 
syllable by Ambisyllabification (Back 1985:10). 

(14) $ $ $ 

/l\ /l\ / \ 

tit + s a P + a 
'hear' HUMBLE COMPLETIVE 

In the coda position, on the other hand, /p/ gets devoiced, and its 
preceding vowel is lengthened, as shown below. 

(15) 



$ 




$ 


$ 


$ 


$ $ 


/l\ 




/l\ 


/\ 


/l\ 


/l\ /\ 


t i t 


+ 


sap + 


k -^ 


t i t + 


sap + k 


'hear' 




HUMBLE 


'and' 




(Baek 1985:10) 



In connection with ambisyllabicity, Baek examines the tonal behavior 
of the morpheme /sap/. The underlying tonal melody of /sap/ is LH, 
of which H is a floating tone. The high tone is associated with the 
following syllable if the syllable begins with a vowel, as in (16a). 
Otherwise it is associated with the vowel of /sap/ itself, which results 
in a contour tone LH, as in (16b) (Baek 1985). 

(16) a. $ $ $ b. $ $ $ 

/l\ /\ /\ /l\ /l\ /\ 

t i t s a P a tits a:p k o 

I I /I I /\ I 

L LHH LLHH 

Baek asserts that phonetic realization of the underlying tonal melody 
LH in Middle Korean is related with syllable structures. In order to 
account for the interaction of the tonal behavior with the syllable 
structures, he posits an underlying empty V slot on the CV tier. 

(17) CVCV 
/sap/ 

LH 

(18a) and (18b) show how [tit sa Pa] and [tit sa:p ko] are derived, 
respectively. 



il 



S. K. Kang: Morale representation of ambisyllabicity 9 5 

(18) a. $ $ $ $ $ $ 

/l\ /\ //\ /l\ /l\ /l\ 

cvccvcvv cvccvcvv 

I I I I I I I I I I I I I \/ 



t i t 

1 


sap 

1 


a 

1 




t i t 

1 


s a p a 

1 /\ 


L 


LH 


H 


— > 


L 


L HH 


$ 


$ 


$ 




$ 


$ $ 


/l\ 


/l\ 


/\ 




/l\ 


/ /\ \ /\ 


cv c 


cvc V cv 




CVC 


CVVC C V 


1 1 1 


1 1 1 


1 1 




1 1 1 


1 \/ 1 II 


t i t 

1 


sap 

1 


k 

1 




t i t 

1 


s a p k 

/\ 1 


L 


LH 


L 


— > 


L 


L H L 



Back (1985:12) says, 

"...the empty V before a vowel-initial syllable [in (18a)] is 
associated with the neighboring vowel, and that the floating 
tone H is associated with the same vowel, resulting in two 
contiguous high tones. ...the two tones are collapsed into each 
other. On the other hand, the empty V before a consonant- 
initial syllable [in (18b)] is associated with the vowel of the 
preceding syllable, producing a long vowel. In this case, the 
floating tone H is associated with the same vowel, forming a 
contour tone LH." 

His analysis, however, has several shortcomings. To begin with, the 
empty V slot posited in the underlying representation of the 
morpheme /sap/ is not independently motivated. And it is not clear 
how his syllabification works. In the first half of the derivation in 
(18a), the empty V slot is already syllabified with both the preceding 
consonant and the following vowel, whereas in the case of (18b), it is 
not syllabified yet. This kind of syllabification is ad hoc in that in 
some cases the empty V slot is syllabified and in other cases it is not. 
Finally and more crucially, his derivation in (18b) encounters a 
serious problem; i.e., how can the empty V element be syllabified 
into the preceding syllable? He gives no account for this problem. 
Suppose that his analysis is correct. Then there seems to be two 
possible ways of syllabifying the empty V slot into the preceding 
syllable in (18b). One possibility is given in (19). 



96 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(19) 



$ $ $ 
/l\ /l\ /\ 

CVC CVCVC V 

1 1 1 1 1 1 II 

t i t s a p k 
1 1 1 




$ $ $ 

/l\ /l\ /\ 

CVC CVCVC V 

1 1 II L+' 1 1 

t i t s a p k o 

1 1 1 


L LH L (Syll.) 


—> 


1 1 1 

L LH L 

(Vowel Lengthening) 


$ $ $ 

/l\ /A\ /\ 

CVC CV CV C V 

1! 1 1 1.^ 1 1 

t i t s a p k 

1 /\ 1 

^ L L H L 


(Association of the 
floating H tone) 



Vowel lengthening in (19) would raise eyebrows, since we end up 
with crossing association lines. An alternative mechanism is some 
kind of metathesis, which will exchange the empty V slot with the 
preceding C slot, as shown below. 

(20) 



$ $ $ 

/l\ /l\ /\ 

CVC CVCV C V 

1 1 1 1 1 1 II 

t i t s a p k o 
1 1 1 




$ $ $ 

/l\ // \ /\ 

CVC CVVC C V 

1 1 1 1 1 III 

t i t s a p k 
1 1 1 


L LH L (Syll.) 


-^ 


1 1 1 
L LH L 
(Metathesis 


$ $ $ 

/l\ /A\ /\ 

CVC CVVCC V 

1 1 1 1 IxM 1 1 

t i t s a p k 

1 /"s 1 

-> L L H L 


) 

(Vowel lengthening. 
Association of H & /p/^[p 



But the metathesis is not supported by any empirical evidence. 

Now let's turn to an account in terms of morale phonology. As 
discussed above, when /p/ occurs syllable-finally, it becomes a 
voiceless bilabial stop [p] and the preceding vowel is lengthened. 
Assuming that /p/ is moraic but /p/ is not, we can give a natural 
account for the devoicing of /p/ with compensatory lengthening of 



S. K. Kang: Morale representation of ambisyllabicity 



97 



the preceding vowel. In the process of syllabification, that is, /p/ in 
the coda position is assigned a mora by the Weight by Position in 
(21), which is the revised version of Hayes (1989), as shown in (22). 

(21) Weight by Position 



(22) 



$ 
1 




$ 
l\ 












^ 


— > 


1^ M 












1 




1 1 


Where $ dominates the 


only n and 


ap 




aP 


P is 


moraic. 








$ 

1 


$ 

1 


$ 

1 




$ 
1 


$ 
l\ 


$ 
1 




[i 


\i- 


\i 




^i 


^ U 


^i 




/W 


/I 


/\ 




/i\ 


/I 1 


/\ 




t i t 

1 


s a p 

1 


k 

1 (Onset 




t i t 

1 


s a p 

1 


k 

1 


(Coda 


L 


LH 


H Rule) 


— > 


L 


LH 


H 


Rule) 



Then the devocing rule which changes /p/ into [p] delinks /p/ from 
its mora. As a result, the mora is stranded so that it can be 
reassociated with the preceding vowel. Example (23) shows how this 
takes place. 



(23) 



$ 

1 


$ 
l\ 


$ 
1 


\^ 


[i- U 


^i 


/l\ 


/I 1 


/\ 


t i t 

1 


sap 
1 


k 
1 


1 
L 


1 
LH 


1 
H 




$ 
1 


$ $ 
l\ 1 




^ 


^ M U 




/w 


/U' /\ 




t i t 

1 


s a p k 
l\ 1 


-^ 


L 


LH H 




$ 
1 


$ $ 
l\ 1 




M 


^ M M 




/l\ 


/l/l /\ 




t i t 
1 


s a p k o 
l\ 1 


-^ 


L 


LH H 



>t> J> 4> 

I l\ I 

/l\ /I /\ 

tit s a p k o 
I I I 

L LH 

(Devoicing) 



H 



(Reassociation of /p/) 



98 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

i 



The floating high tone in (23) is associated with the long vowel /a:/, 
producing a contour tone LH. On the other hand, no contour tone is 
derived when the bilabial voiced fricative /p/ occurs as the onset of 
a syllable, which is shown below. 



(24) 



$ $ $ 

1 1 1 




$ $ $ 
1 1 1 


1 1 1 
/W /I /I 

tit s a p a 
1 1 1 




1 1 1 

/l\ /l\/l 

tit s a p a 

1 1 1 


1 1 1 
L LHH 


(Syll.) ^ 


1 1 1 

L L H 

(Ambi. & SEC) 



Suppose that in Middle Korean, short vowels cannot bear a contour 
tone. Then, the unassociated high tone above is deleted by the Stray 
Erasure Convention given in (25). 

(25) Stray Erasure Convention 

Erase segments and skeleton slots unless attached to 
higher level of structure. (Steriade 1982) 

It is shown above that unlike Back's (1985) assertion, the devoicing 
of /p/, vowel lengthening and the association of the floating high 
tone are phenomena related with one another. That is, the devoicing 
of /p/ triggers compensatory lengthening of the preceding vowel, 
which in turn feeds the association of the floating high tone, 
producing a contour tone LH. Here, we do not need to posit with Back 
(1985) an empty V slot underlyingly which has no independent 
motivation. Once a morale structure is assumed, then those 
phenomena discussed above can be given a natural account. 

3. Conclusion 

I have discussed so far how ambisyllabicity should be 
represented. It has been shown that unlike Selkirk's (1982) 
assertion, ambisyllabicity DOES exist. Arguing against Borowsky et al. 
(1984), I have also shown that ambisyllabicity and gemination have 
different structures from each other. A geminate consonant is 
assigned a mora in the underlying representation, and in the process 
of syllabification, it is syllabified both as the onset of a syllable and 
as the coda of the preceding syllable. In contrast, an ambisyllabic 
consonant does not bear any mora underlyingly. It is syllabified as 
the onset of a syllable in the process of syllabification, and then it is 
also linked with the preceding mora by the Ambisyllabification. 
Ambisyllabicity and gemination are not in complementary 
distribution as Borowsky et al. (1984) assert. Rather they can exist in 



S. K. Kang: Morale representation of ambisyllabicity 99 

a single language, and the distinction between them even contributes 
to differences in the meanings of words, as discussed above. 

Finally, the moraic representation of ambisyllabicity in (26a) is 
preferred over Clements and Keyser's (1983) in (26b). 

(26) a. n n b. $ $ 

/\/\ \/ 

V t V c 

I 

t 

As Borowsky et al. (1984:35) point out, first, questions with respect 
to descriptive power may arise. Observe (27). Can the V-slot be 
shared by different syllables in a way analogous to the ambisyllabic 
C-slot in (26b)? 

(27) $ $ 

\/ 
V 
I 
a 

Unlike CV theory, however, moraic theory correctly predicts that no 
ambisyllabic vowel occurs. Under moraic theory, that is, ambisyllabic 
segments are distinguished from vowels underlyingly in that the 
former do not bear any mora, while the latter bear one or two moras. 
Only unmorified segments can become ambisyllabic segments. This 
leads to the correct prediction that vowels can never be ambisyllabic. 

Second, the loss of an ambisyllabic segment does not induce 
compensatory lengthening (CL). Consider a case of this sort from 
English given below. 

(28) $ $ $ $ 

II II 

^1^1 -^ \i \i (/h/ -^ 0) 

l\/l I I 

ni h i lism n i ilism 

The fact that in (28), no CL is triggered by the deletion of the 
ambisyllabic segment /h/ comes from the moraic representation of 
ambisyllabicity. That is, the deletion of /h/ has no mora stranded, so 
that no CL occurs. Under CV theory, however, this prediction cannot 
be made. CV theory would incorrectly predict that the loss of the 
ambisyllabic segment /h/ triggers CL, as exemplified in (29). 

(29) $ $ $ $ $ $ 
l\/ 1 l\/ 1 l\/ 1 

VCV ^ VCV (/h/ -^ 0) ^ VCV (CL) 

III II 1/ I 

nih ilism n i ilism n i ilism 



100 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

NOTES 

*I am grateful to Chin W. Kim, Charles W. Kisseberth, Jennifer 
Cole, Jose I. Hualde and members of the Illinois Club in Korean 
Linguistics for discussion and comments on the earlier version of this 
paper. 

REFERENCES 

Baek, Eung-Jin. 1985. Syllables and tones in Middle Korean. Chicago 

Linguistic Society 21.1-14. 
BOROWSKY, T., Junko Ito, & Ralf-Armin Mester. 1984. The formal 

representation of ambisyllabicity: Evidence from Danish. North 

Eastern Linguistics Society 14. 
Clements, G. N., & S. J. Keyser. 1983. CV phonology: A generative 

theory of the syllable. Cambridge: MIT Press. 
Hayes, B. 1986. Inalterability in CV phonology. Language 62.321-351. 
. 1989. Compensatory lengthening in moraic phonology. Linguistic 

Inquiry 20.253-306. 
Kahn, D. 1976. Syllable-based generalization in English phonology. 

MIT Ph.D. dissertation in Linguistics. 
Kang, Seok Keun. 1991a. Moraic phonology and /l/-irregular 

predicates in Korean. Studies in the Linguistic Sciences 21:1. 
1991b. Compensatory lengthening in Korean revisited. Eastern 

States Conference On Linguistics '91. 
, & Chin W. Kim. 1991. A moraic account of consonant cluster 

simplification in Korean. Formal Linguistics Society of 

MidAmerica II. 
MS. Compensatory lengthening in Farsi: A moraic account. 

University of Illinois. 
Kenstowicz, M. 1982. Gemination and spirantization in Tigrinya. 

Studies in the Linguistic Sciences 12:1.103-133. 
SCHEIN, B., & Donca Steriade. 1986. On geminates. Linguistic Inquiry 

17:4.669-744. 
Selkirk, E. 1982. The syllable. The structure of phonological 

representations (Part II), ed. by Harry van der Hulst & Noval 

Smith. Foris publications. 
. 1988. A two root theory of length. North Eastern Linguistics 

Society 19. 
Zec, D. 1989. Sonority constraints on prosodic structure. Stanford 

University Ph.D. dissertation in Linguistics. 



Studies in the Linguistic Sciences 

Volume 21. Number 2. Fall 1991 



THE LOCALITY CONDITION OF TONAL SYSTEMS: 

WITH SPECIAL REFERENCE TO NORTH KYUNGSANG DIALECT 

IN KOREAN 

Yongsoon Kang 

This paper is an attempt to reduce the power of 
phonological rules. To be specific, this paper deals with the 
tonal system of the North Kyungsang dialect in Korean and 
shows that a couple of suggested tonal rules violate the 
Locality Condition (LC), which prohibits rule application 
between nonadjacent elements. This paper proposes that 
these rules be removed and that following the idea of LC 
counting, the counting capability of phonological rules be 
limited to two operations, i.e. trigger itself and trigger the 
adjacent target. 

1. Introduction 

It was the tonal system which introduced the concept of non- 
linearity into phonology. Recently, Korean tonal systems also have 
been reanalyzed under the non-linear framework and many new 
rules have been suggested. Since these rules are part of phonological 
rules, they are expected to obey other phonological conditions and 
constraints. 

The purpose of this paper is to show that a couple of tonal rules 
suggested in the previous works (Kim, G-R. 1988, Chung, Y-H. 1989, 
Han, S-H. 1989) on the tonal system of North Kyungsang (Taegu) 
dialect are too strong in that they violate the Locality Condition (LC) 
(Hewitt & Prince 1989), and thus should be removed. 

In the next section, I will introduce the notion of the LC and 
discuss its theoretical importance. In section 2, I suggest that several 
previously formulated tonal rules violate the LC and that, therefore, 
they can be dispensed with. 

2. Locality Condition 

In this section, I will introduce the LC suggested by Hewitt & 
Prince (1989) and discuss its theoretical significance. According to 
the LC of Hewitt & Prince (1989:3). phonological rules may affect a 
single element (tier) adjacent to the rule trigger. Under the non- 
linear framework, they define the notion of structural adjacency as 
follows. 



102 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(1) a. Adjacent 

XXX X 

III I 

AB AB AB AB 

b. Nonadjacent 

XXX 

I I 

A B 

In (1), X signifies the tone bearing unit (TBU) and capital letters (A, 
B), the tonal tier. Tones which are not linked to the TBU are floating 
and are assumed to be adjacent to any linked tones as (la) shows. 
The two floating tones are also assumed to be adjacent. However, 
linked tones are adjacent only if their connected TBUs are adjacent. 
Phonological rules may affect a single tone which is adjacent to the 
rule trigger. Thus we can say that A and B can affect each other in 
(la) but not in (lb). 

This kind of constraint is of theoretical importance in that it 
restricts the power of phonological rules. Assuming that the ultimate 
objective of studying language is to explain language acquisition of 
children, we do not want our grammar to be so powerful as to 
generate everything. On the contrary, we want it to be as simple as 
possible so that we can explain how a child learns his or her own 
language with such ease and quickness. 

By the LC, we can restrict the power of phonological rules to 
apply only between the adjacent trigger and target. Moreover, 
following the idea of the LC, I suggest that the counting capacity of 
phonological rules be limited to 'two', i.e. the trigger itself and the 
adjacent target. Thus we can prevent the following kind of powerful 
phonological rule from taking place. 

(2) Xq Xi X2 ... Xn 

I / 

1/ 

A (A spreads to the nth segement) 

In this paper, following the underspecification theory of 
Pulleyblank (1986), I assume that only high tones are represented 
on the underlying structure and all low tones will be realized on the 
surface by the following default low tone insertion rule. 

(3) Default Low Insertion rule 

X ^ X 
I 
L 



Y. Kang: The Locality Condition of a tonal system 103 

3. Previous analyses 

In this section, I will present three tonal rules suggested in 
previous analyses (Kim, G-R. 1988, Chung, Y-H. 1989, Han, S-H. 1989) 
which violate the LC, and I will then suggest an alternative solution 
in §4. 

As an adjustment rule, several rhythm rules were introduced in 
the previous studies. However, the LC was never considered. Example 
(4) illustrates some of these rules. 

(4) a. Pre-linked H-deletion (Kim. G-R. 1988:58) 

H -^ / # (x) X + (x) X (x) # 

I I 

H 

b. Second H-delinking rule (Han, S-H. 1989:141) 

X Xj] X 

I I 

I = (Xj is one or more sequence of 

I I empty x's) 

H H 

These rules apparently violate the LC because one or more TBUs 
are between two high tones. In both rules of example (4), H- 
delinking is motivated by the neighboring (but not adjacent) H. The 
situation is worse, especially in (4b), since the number of TBUs 
between two H-bearing TBUs is not limited. In the next section, I will 
show that the same data can be explained without these rules. 

Another powerful tonal rule found in all of the above previous 
works is the Third H-delinking rule. 

(5) Third H-delinking rule 

H ^ / X X x^ 
I I 

I I 

H 

This rule says that all the high tones are delinked except the 
first two. This rule was motivated to explain the data which show 
[HHLL...] pattern. First, they assume the universal free tone 
association which forces the maximal association of a high tone to the 
right, and then rule (5) applies since high tones are not realized from 
the third on. 

Rule (5), however, is too powerful in the sense that it does not 
restrict the power of the counting capability of phonological rules. If 
we allow that it can count three, then we can ask why not four, five, 
six and so on. In fact, the rule is a by-product of the free tone 



104 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

association rule, which unnecessarily spreads to the maximum. Thus 
its sole motivation is to derive the correct surface form. Therefore, 
the rule makes the grammar too powerful by giving it unlimited 
counting capability, which is a violation of the LC. The original LC 
proposed by Hewitt and Prince (1989) limits the power of counting 
to the adjacent unit. 

Given the above reasons to reject this rule, I will adopt the Tone 
Mapping Rule instead which was suggested by S-H. Han (1989:136) 
to explain the same data. 

(6) Tone Mapping Rule (TMR) 

a. Map a floating H tone onto the first TBU. 

b. Mapped H tone spreads only to the next TBU to the 
right. 

4. Alternative analysis 

In this section, I will show that the same data discussed in 
earlier works can be explained only with the rules which do not 
violate the LC. To begin with, I will introduce the data and some 
tonal rules which will be adopted in this paper. Then I will present 
derivations of the data. 

4.1. Data 

The following are some examples which I will discuss in this 
paper. First, there are some words in which the location of the tone is 
not affected by the attachment of postpositions. The underlying 
forms of these words are given in (7d) (Chung, Y-H. 1989). i 

(7) a. myenuli [HLL] 'daughter-in-law' 

myenuli-»-ka [HLLL] 'daughter-in-law (Subj.) 

myenuli+chelem [HLLLL] 'like daughter-in-law' 
myenuli+pota [HLLLL] 'than daughter-in-law' 

b. kkamaki [LHL] 'raven' 
kkamaki-i-ka [LHLL] 'raven (Subj.)' 
kkamaki+chelem [LHLLL] 'like raven' 
kkamaki-Hpota [LHLLL] 'than raven' 

c. hanul [HL] 'sky' 
hanul+i [HLL] "sky (Subj.)' 
hanul+chelem [HLLL] 'like sky' 
hanul-t-pota [HLLL] 'than sky' 

d. myenuli kkamaki hanul 

I I I 

H H H 

Second, in the following, the first two tones are always high, 
regardless the number of TBUs. 



Y. Kang: The Locality Condition of a tonal system 



105 



(8) a. mwucikay [HHL] 

mwucikay+ka [HHLL] 
mwucikay+chelem [HHLLL] 
mwucikay+pota [HHLLL] 


'rainbow' 
'rainbow (Subj.) 
'like rainbow' 
'than rainbow' 


b. kwulum [HH] 
kwulum+i [HHL] 
kwulum+chelem [HHLL] 
kwulum+pota [HHLL] 


'cloud' 

'cloud (Subj.)' 
'like cloud' 
'than cloud' 


c. so[H] 
so+ka [HH] 
so+chelem [HHL] 
so+pota [HHL] 


'cow' 

'cow (Subj.)' 
'like cow' 
'than cow" 


d. mwucikay kwulum 


so 


H H 


H 


Third, the location of the high tone 


is shifted with the addition of 


postpositions. 




(9) a. satali [LLH] 'ladder' 

satali+ka [LLHL] 'ladder (Subj.)' 
satali+chelem [LLLHL]~[LLHHL] 'like ladder' 
satali+pota [LLLLH] 'than ladder' 


b. poli [LH] 
poli+ka [LHL] 

poli+chelem [LLHL]~[LHHL] 
poli+pota [LLLH] 


'barley' 
'barley (Subj.)' 
'like barley' 
'than barley' 


c. cha [H] 
cha+ka [HL] 

cha+chelem [LHL]~[HHL] 
cha+pota [LLH] 


'car' 

'car (Subj.)' 
'like car' 
'than car' 


Following Y-H. Chung (1989). I assume 
for words in (9) and that the surface 
insertion rule as in the following. 


that there is no underlying H 
forms are realized by a H- 


(10) H-insertion 




X] -> X] 




1 

H (Condition: 


no H in the domain) 


4.2. Derivations 





In order to explain the above data. I introduce three more rules 
which are given in the following examples. First, there is a H-shifting 
rule as in example (11). 



106 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

(11) H-shifting 

X X (Condition: a. prelinked H doesn't shift. 
I ./ b. it doesn't shift to the last 

= ./ TBU of the domain.) 

1/ 
H 

Second, we need a rule which docks the floating H of a 
postposition to the last TBU of the lexical item, which is given in (12). 

(12) Floating H-docking Rule (Y-H. Chung 1989:122) 

X + X 
\ 

\ 
H 

This rule precedes the Tone Mapping Rule and we don't have to 
specify this rule ordering in the grammar, for this is determined 
naturally by the Elsewhere Condition which makes a rule A apply 
before a rule B if and only if the structural description of A properly 
includes that of B. 

Third, we need Meeussen's rule in which the application of the 
rule takes place only between adjacent TBUs. 

(13) Meeussen's Rule 

H-> 0/H_ 

To summarize, the rules I have proposed are as follows in example 
(14). 

(14) Floating H-Docking Rule (FHDR) 
Tone Mapping Rule (TMR) 
H-insertion 

H-shifting 
Meeussen's rule 

Before examining the rule ordering relationship of (14), I would 
like to say a little more on postpositions in Korean. Given the fact 
that all postpositions in Korean are bound forms and that they cannot 
occur alone, I suggest here that tonal variation of postpostions is not 
realized until they are attached to the independent lexical items in 
the sense of Lexical Phonology (S-C. Ahn 1985), even though they 
have the underlying tonal structure as in example (15). 

(15) -ka (-i) -chelem -pota 

H H 



Y. Kang: The Locality Condition of a tonal system 107 

With these rules we need to specify the rule ordering relations 
between the rules. To begin with, Meeussen's rule should precede the 
FHDR since the former bleeds the latter as shown below. 

(16) a. myenuli+chelem b. myenuli+chelem 

I I 

H H H H 

Meeussen x x x + x x 

I \.... FHDR 

FHDR H H 

Meeussen 

[HLLLL] *[HLHLL] 

Second, FHDR precedes the H-shifting rule since the former feeds 
the latter as in (17). 

(17) a. satali+chelem b. satali+chelem 

H H 

FHE^ XXX + XX H-shifting 

H X X X + X X FHDR 

H-shifting \.. 

X X X + X X H 

I ./ 

H 
[LLLHL] *[LLHLL] 

The ordering relation between the five rules is summarized in 
(18). 

(18) Meeussen's rule — 

I 

-FHDR 

I 

I TMR 

I I 

---H-shifting rule 

H-insertion rule 

With these rules and their ordering relation, we can derive the 
data as seen in (19). 



108 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



(19) 


a. kkamaki-chelem 




H H 


Meeussen 

FHDR 

TMR 

H-shifting 

H-insertion 

[LHLLL] 




b. kkamaki-pota 


all 


1 
H 


rules 


[LHLLL] 


insertion ru 


lie doesn't apply to 



meli-chelem 
I 
H H 



[HLLL] 

meli-pota 
I 
H 



[HLLL] 

H-insertion rule doesn't apply to -pota in (19b), since there is 
another H in the domain. Low tone is realized by the default low- 
insertion rule (2) after all other rules have applied. 

TMR plays a crucial role to explain following data. 



10) a. 


mwucikay+ka 


kwulum+i 




so-)-ka 




H H 


H H 




H H 


Meusseun 













FHDR 












TMR 


XXX + \ 


X X + X 




\ + \ 




1/ 


1/ 




1.../ 




H 


H 




H 


H-shifting 













H-insertion 















[HHLL] 


[HHL] 




[HH] 


b. 


mwucikay-i-chelem 


kwulum+chelem 


so-J-chelem 




H H 


H 


H 


H H 


Meeussen 













FHDR 











TMR 


X X X + X X 


X X + X 


X 


X +X X 




1/ 


1/ 




1../ 




H 


H 




H 


H-shifting 













H-insertion 














[HHLLL] 



[HHLL] 



[HHL] 



Y. Kang: The Locality Condition of a tonal system 
c. mwucikay+pota kwulum+pota 



109 



Meeussen 

FHDR 

TMR 

H H H 

H-shifting 
H-insertion 



H 



X X X + X X 

1/ 



[HHLLL] 



H 



X X + X X 



[HHLL] 



so+pota 
H 



X + X X 

1..../ 



[HHL] 



In (20), FHDC does not apply because this rule applies to only the 
floating H of postpositions. 

The H-insertion rule is not affected by rule ordering, and if 
there is no H in the given word then it applies, making the last TBU 
high-toned. (21) gives some examples. 



1) a. satali 




poli 




cha 


Meeussen 












FHDR 












TMR 












H-shifting 












H-insertion 










X X x] 




XX] 




X] 


H 




H 




1 
H 


[LLH] 




[LH] 




[H] 


b. satali-(-ka 


poli+ka 


cha-fka 




H 




H 


H 


Meeussen 




... 







FHDR X X x + x 


X x + 


X 


x + x 


\.. 








\.. 




H 




H 


H 



TMR 

H-shifting 
H-insertion 



[LLHLl 



[LHL] 



[HL] 



110 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



c. 


satali+chelem 


poli+chelem 


cha+chelem 




H 


H 


H 


Meeussen 











FHDR 


X X X + X X 


X X + X x 


X + X X 




\... 


\... 


\... 




H 


H 


H 


TMR 











H- 


X X X + X X 


X X + X X 


X + X X 


shifting 


1 ./ 


1 ./ 


1./ 




=/ 


=/ 


=/ 




H 


H 


H 


H-insertior 












[LLLHL] 


[LLHL] 


[LHL] 


d. 


satali+pota 


poli+pota 


cha+pota 


Meeussen 











FHDR 











TMR 











H-shifting 











H- 


X X X + X x] 


X X + X x] 


X + X x] 


insertion 


1 


1 


1 




H 


H 


H 




[LLLLH] 


[LLLH] 


[LLH] 



In (21c), if we change the H-shifting to H-spreading which spreads 
the H only to the next TBU without delinking the trigger, then we can 
explain the alternative tonal form. 

(22) a. H-spreading 

X X 

I ../ 

1/ 

H 

b. satali-f-chelem 

H 

Meeussen 

FHDR X X X + X X 

\.. 
H 

TMR 

H- X X X + X X 

spreading I ./ 

1/ 
H 

H-insertion 

[LLHHL] 



Y. Kang: The Locality Condition of a tonal system 111 

5. Conclusion 

The Locality Condition is theoretically significant in that it 
restricts the power of phonological rules. For instance, the application 
of Meeussen's rule or H-delinking would be undesirable if we assume 
that they can apply between the nonadjacent segments. Moreover, if 
we allow this, there is no way to prevent them from applying to 
cases such as in (23), which apparently look wrong. 

(23) XX X ... X X X 
I I 

H H 

In this paper, I have suggested the removal of some rules which 
violate the LC because they make a grammar too powerful. I have 
proposed that, following the idea of the LC, the counting ability of 
phonological rules be limited to 'two'. I adopted TMR which was 
suggested by S-H. Han (1989) and restricted the application of 
Meeussen's rule to apply strictly to adjacent environments. Finally, I 
have showed that, with certain modifications, these rules can derive 
correct surface forms. 

NOTES 

1 For the transcription of the Korean data, I use the Yale 
Romanization system. 

REFERENCES 

AHN, S-C. 1985. The interplay of phonology and morphology in 

Korean. University of Illinois Ph.D. dissertation in Linguistics. 
Chung, K. 1980. Neutralization in Korean: A functional view. 

University of Texas at Austin Ph.D. dissertation in Linguistics. 
Chung, Y-H. 1989. Tone system of the North Kyungsang dialect of 

Korean. Papers from the Sixth International Conferences on 

Korean Linguistics, ed. by Eung-Jin Back. University of Toronto. 
Goldsmith, J. 1976. Autosegmental phonology. MIT Ph. D. dissertation 

in Linguistics. Distributed by the Indiana University Linguistic 

Club. 
Han, S-H. 1989. The accentual system of the Korean Kyungsang 

dialect: A lexical analysis. Papers from the Sixth International 

Conference on Korean Linguistics, ed. by Eung-Jin Back. 

University of Toronto. 
Hewitt, M., & A. Prince. 1989. OCP, locality, and linking: The N. 

Karanga verb. Department of Linguistics. Brandeis University, 

MS. 



112 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Kim, G-R. 1988. The pitch accent system of the Taegu dialect of 

Korean with emphasis on the tone sandhi at the phrasal level. m 

University of Hawaii at Manoa Ph.D. dissertation in Linguistics. ^ 

KiPARSKY, P. 1982. Lexical morphology and phonology. Linguistics in 
the morning calm, ed. by I-S. Yang, 3-91. Seoul: Hanshin 
Publishig Co. 

PULLEYBLANK, D. 1986. Tone in lexical phonology. Dordrecht: Reidel 
Publishing Co. 



Studies in (he Ling»i?;tic Sciences 

Volume 21, Number 2. Fall 1991 



THE CHARACTER OF KOREAN GLIDES 

Chin W. Kim 
(University of Illinois at Urbana-Champaign) 

and 

Hyoung-Youb Kim 
(Korea University at Chochiwon, Korea) 

The maximum surface syllable structure of Korean is CGVC. 
The exact status of G within the syllable hierarchy has been 
in dispute in recent years: It has been viewed either as (a) 
a full segment in flat structure, (b) a part of a complex 
consonant, (c) a part of the onset cluster, or as (d) a part of 
the nucleus. In the first part of the paper, we present 
various morpho-phonological phenomena in Korean, i.e., 
morpheme structure conditions, consonant cluster 
reduction, initial consonant constraint, reduplication in 
ideophones, a language game, etc., as evidence for the 
nuclear nature of glides. In the second part, we show how 
the proposed view may simplify phonological descriptions 
of rules involving glides beyond the Korean data, e.g., glide 
formation, metathesis, compensatory lengthening, etc. In 
conclusion, we show how correctly and ingeniously, Han'gul, 
the five and a half century old native script, represents the 
nuclear character of glides in the Korean syllable. 

1. Introduction 

Our task today is a simple one; we would like to argue that in 
the syllable structure of Korean, the glides should be regarded as a 
part of the nucleus, rather than a part of the syllable onset. The 
paper is divided into two parts. In the first part, we will present 
evidence for the nuclear position of glides, and in the second part, we 
will show how this view simplifies a phonological description beyond 
the Korean case. 

Korean phonologists have not been in agreement with the 
canonical syllable structure. With respect to the position of glides, the 
syllable structure of Korean may be divided into the following 
several types: 



114 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(1) a. Flat structure 



(C)(G)V(C) 

b. Glide as onset 

i. Right-branching Rime 




ii. Left-branching Core 



Kim & Shibatani (1976) 
Y-S. Kim (1984) 



B-G. Lee (1982) 
S-C. Ahn (1985) 



(Onset) Nucl. 

C (G) 

c. Glide as nucleus 
a 



(Coda) 




C-G. Gim (1987) 
S-C. Ahn (1988) 



Kim-Renaud (1978) 
J-M. Kim (1986) 
H-S. Sohn (1987) 



Our evidence for a glide as a part of the syllable nucleus in 
Korean will take the form of presenting and examining several 
phonological phenomena that involve interaction with glides. Much of 
such evidence was first given by H-S. Sohn (1987) where she 
advances excellent arguments for representing complex Korean 
vowels including diphthongs and glides as a sequence of vowels 
dominated by syllable nucleus. The exact mechanism by which two 
vowels coalesce to produce surface complex vowels is admirably 
worked out by Sohn, I will not go into details here. Below we give a 
brief summary of her arguments. Let's first begin with what Sohn 
(1987) calls an ex silentio argument for the nuclear nature of glides. 



C. W. Kim & H-Y. Kim: The character of Korean glides 1 1 5 

2. Evidence of a nuclear glide 

In many languages where there are consonant clusters, two 
things are very common: One, a severe constraint in the sequencing 
of the segments, what used to be called "Morpheme Structure 
Conditions", and two, a mirror-image relation between the onset 
cluster and the coda cluster. In English, for example, only non-nasal 
sonorants may follow stops, which in turn can follow only s, e.g., pr-, 
pi-, py-, tr-, tw-, kr-, kl-, kw-, ky-, spr-, spl-, str-, skr-, etc. It is also 
true that for all the initial clusters listed above, there are also final 
clusters in reverse sequence, i.e., -rp, -Ip, -lb, -rt, -rk, -Ik, -rps, -Ips, - 
rts, -rks, etc. Note that there are no final clusters *-wp,*-wb, just as 
there are no initial clusters *pw-, *bw-. Returning to Korean, if we 
regard the glides w and y to be parts of the onset cluster, then we 
might expect that there may exist a certain constraint in the 
sequencing of glides with respect to other consonants. But the fact is 
that there is none. Any consonant may precede a glide. A possible 
exception might be the nonexistence of a sequence of a palato- 
alveolar consonant and the palatal glide y. But it is a case of 
nondistinction in pronunciation between cya and ca, rather than a 
case of nonexistence of the underlying sequence cya. Also, if we 
grant that Cy- and Cw- sequences are onset clusters, then we might 
also expect the mirror-image sequences to occur as coda clusters, i.e., 
-yC and -wC. But such a cluster is simply nonexistent in Korean. 

Curiously enough, on the other hand, there exist co-occurrence 
constraints both in the permissible coda consonant clusters and in 
the complex nuclei involving glides. In Korean, there are morphemes 
and words that have two-consonant clusters at the coda position. 
However, the composition and the order of these clusters are 
severely constrained, so that only -ps, -ks, -Ip, -Ip'^, -Ik, -It^, -Ik, -Ih, 
-Im, -nh, and -nc are permissible coda clusters. In complex nuclei, 
there is a negative condition such that *yi, *yu, *yi, *wu, *wo, *wo, 
*wi, and *wu do not occur, and the only off-glide diphthong occurring 
in Korean is iy. This is not a place to explicate a rule and a reason for 
non-occurrence of the complex nuclei listed above, nor does this 
phenomenon by itself constitute evidence for assigning glides to the 
nucleus rather than to the coda, but on typological grounds it makes 
one bet on the former as a more probable and likely case. 

Since we are on the subject of consonant clusters, let us examine 
the phenomenon of consonant cluster reduction in Korean and its 
relevance to the position of glides. As was mentioned in the 
preceding paragraph, some words in Korean may end in one of the 
small set of two-consonant clusters. When such a word is followed by 
a vowel, both consonants survive, as the second consonant becomes 
the onset of the following syllable, thus keeping the canonical 
syllable shape. But when the next morpheme/word begins with a 



116 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

consonant, as in example (2), it creates an impermissible three- 
consonant cluster medially, which prompts deletion of one of the 
cluster consonants. 

(2)kaps 'price', kaps-i (Nom), feur kap-man 'price only' 

naks 'spirit', nsks-i (Nom), nak-to 'soul also' 

ilk- 'to read', ilk- ra (Imp), ik-ca 'let's read' 

calm- 'young', calm-in (Adj), cam-ta 'is young' 

anc-a 'to sit', anc-atta (Past), an-kara 'sit' (Imp) 

Suppose now that a glide is a part of the onset. Then a word- 
initial consonant followed by a glide should be regarded as a cluster. 
If such a word is preceded by a word ending in a consonant, we 
would again have a medial three-consonant cluster, and we should 
expect the cluster reduction rule to apply to delete one consonant. 
But this does not happen. Examine example (3). 

(3) ol-pya 'this year's crop' 

sil-kwa 'fruit' 

sok-pyang 'internal illness' 

cal-myo 'exquisiteness' 

kak-phyo 'each vote/ballot' 

san-nya 'angel, fairy' 

One can of course argue that this is not a genuine case of a 
three-consonant cluster by appealing to the intervening CV tier 
independent of the segmental tier, as in example (4). 

(4) Syllable tier: a a 

CV tier: C V C C V 

III /\ I 

Segment tier: s i 1 k w a 'fruit' 

k a k p*^ y 'each vote' 

In this representation, there is no violation of the canonical 
syllable structure, and naturally the cluster reduction rule does not 
apply. But note that this gain is made only at the expense of 
postulating a large set of complex consonant phonemes in Korean. 
While there are complex consonants such as affricates and 
prenasalized consonants in the phonetic inventory of the world's 
languages, the set is very limited, and a large number of complex 
consonants allegedly involving glides in Korean is highly suspect on 
universal phonetic grounds. 

Even if one grants that Cy and Cw are acceptable complex 
consonants, one runs into a difficulty when it comes to pre-y 
sonorant deletion in the word-initial position in Korean. In Korean, 



C. W. Kim & H-Y. Kim: The character of Korean glides 1 1 7 

word-initial / and n are deleted although they i survive in the medial 
position, as in example (5). 



(5) Base 


form 


Medial 


Initial 


iyuk 


'ground' 


tae-lyuk 'continent' 


yuk-ci 'land' 


nya 


'female' 


su-nys 'nun' 


ya-ca 'woman' 


nyo 


'urine' 


pang-nyo 'urination' 


yo-to 'urethra' 


lyang 


■pair" 


yal-lyang 'ten pairs' 


yang-ka 'both 

families' 



If ly and ny are regarded as complex consonants linked to a 
single C in the CV tier, then deleting the initial segment would be like 
deleting the first half from affricates or prenasalized stops, i.e., the 
stop component from ts,dz,pf, etc., or the nasal component from mh, 
nd, ng, etc. It may not be an impossible thing to do or an impossible 
to rule to write, but certainly such a rule would be a complicatd one 
reflecting a rather rare phenomenon. 

Ideophones in Korean also give evidence that a glide is a part of 
the nucleus. Ideophones normally take the form of reduplication, as 
in example (6). 

(6) k'opul-k'opul 'zigzagging, winding' 
taekul-taekul 'rolling, rumbling down' 
pintung-pintung 'loafing, idling' 
c^'ullang-chullang 'lapping, slopping' 

There is, however, a set of ideophones with partial reduplication 
in that the initial consonant of the first isotope is missing, e.g., 

(7) aki-caki 'sweet, intriguing' 
osun-tosun 'friendly, chummily' 
ulkit-pulkit 'colorful' 
ampsng-tampang 'sloppily, carelessly' 

Now, examine the following examples in this context: 

(8) yam-nyam 'tasty' 
yak-lyak 'vivid' 
yong-nyong 'teasing' 

These forms suggest that the glide y is a part of the nucleus, not 
a part of the onset, for, if the latter, a partial reduplication without 
the onset consonant would have given us non-occurring *am-ny(im, 
*k-ly k, *onf>-nyong, etc. 

Sohn (1987:108) cites a language game to support her argument 
of glides as parts of nuclei. This game involves copying a CV from 
every syllable in which C is prelinked to p and V copies the vowel of 
the preceding syllable, as in example (9). 



118 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 

(9) hgkong 'empty sky' ha-ps ko-po-ng 
p^ato 'waves' p*^a-pa to-po 
camsil 'a place name' ca-pa-m si-pi-1 

What is noteworthy is that where the syllable contains a glide, 
this glide is also copied as a part of the nucleus. Examine the 
following: 

(10) kwanse 'power' kws-pws-n se-pe 
yaku 'baseball' ya-pya ku-pu 
cwasak 'seat' cwa-pwa sa-ps-k 

If the glide was a part of the onsent consonant, then we would 
not expect the glide to be copied along with the vowel. As Sohn 
(1987) notes, the fact that it is, provides substantial evidence that 
glides constitute a prosodic unit with their following vowel rather 
than with their preceding onset consonant. 

There are two pieces of phonetic evidence showing that a glide 
is a part of the syllable nucleus. The first is palatalization of /s/ 
before /wi/ in such words as swin 'fifty', swi-ta 'rests', swip-ta 'is 
easy', etc., where the initial consonant [s] is palatalized to [s]. If [w] is 
a part of the onset cluster /sw/, it is difficult to explain this 
palatalization phenomenon. But if [w] is a part of the vowel complex 
[wi] such that the complex nucleus is a round front vowel [u], then it 
is natural for the preceding consonant to be palatalized. ^ 

The most transparent phonetic evidence showing that glides are 
a part of the syllable nucleus not the onset cluster is found in the 
pronunciation of the liquid phoneme IM before a glide in Korean. One 
of the most salient phonetic rules in Korean is the change of [1] to [r] 
in the prevocalic (= syllable-initial) position, as in example (11). 

(ll)kil 'street', kil- to 'st. also', cf. kil-e [kire] (Loc) 

tal 'moon', tal-pic^ 'moonlight', cf. tal-i [tari] (Nom) 
pul 'fire', pul-k'och 'flame', cf. pul-il [puril] (Ace) 

Now, when a glide-initial morpheme follows a morpheme ending in /, 
this / uniformly becomes r, as in example (12). 

(12) il-yo-il [iryoil], *[ilyoil], 'Sunday' 

sal-yok [saryok], *[s8lyok], 'vindication' 

kil-w 1 [kirw 1], *[kilwal], 'writing' 

mil-w 1 [mirw 1], *[milwal], 'honeymoon' 

If the glides are onset consonants, then the preceding liquid 
ought to remain as the coda consonant of the preceding syllable 
rendering the lateral pronunciation [1]. But the fact that it changes to 
[r] suggests that glides y and w are a part of the syllable nucleus, 
allowing the preceding / to fill the empty onset position during 
resyllabification, as in example (13). 



C. W. Kim & H-Y. Kim: The character of Korean glides 1 1 9 

(13) a o a ceo 

AAA resyll. I A A 
il-yo-il -> i-lyo-il [iryoil] 

/X /X resyll A yC^ 

mil-wal -> mi-lwsl [mirwal] 

There is in fact a southern dialect (Kyongsang Province) in which 
the above words are pronounced with the lateral [1], e.g., [ilyoil], 
[milwal], etc. This suggests that in this dialect the glides function as 
the onset consonant forcing the liquid to remain as the coda 
consonant. Significantly, it is just in this dialect that we find no 
nucleus involving a glide after another consonant, i.e., there is no 
such complex nuclei as wa, w a , yu, ys , etc. post-consonantally. 
Compare the following sets of words: 

(14) Standard dialect Kyongsang dialect Gloss 

kwaca kaca 'cookie' 

pwara para 'see' (Imp) 

kyaul keul 'winter' 

pyai pel 'star' 

This fact is significant, for it shows that there is a non-accidental 
relation between the fact that the glides function as onset consonants 
in the Kyongsang dialect and the fact that glides are not found in the 
nucleus position in the same dialect, for, obviously, glides cannot 
function both as onset consonants and as nuclei. The fact that the 
liquid is pronounced as a lateral [1] in this dialect indicates that it is 
in the syllable-final position with the following glide acting as the 
onset consonant for the following syllable. On the other hand, the fact 
that the pre-glide liquid is pronounced as [r] in the standard dialect 
indicates that here the glide is a part of the syllable nucleus. 

Interestingly, not only do we not find prevocalic glides in the 
Kyongsang dialect, but also there is no post-vocalic glide. It may be 
recalled that there is one diphthong in Korean involving an off-glide, 
namely, iy. As a native suffix, it is a genitive marker, and as a Sino- 
Korean morpheme, it has a few homophonous meanings, e.g., 'justice', 
'medical', 'clothing', 'will', etc. In Kyongsang dialect, all these 
homophonous morphemes are uniformly pronounced as [i]. 
3. Theoretical implications 

We have now come the second part of the paper. We will show 
in the following that, cast against the current format of non-linear 
(CV or X-tier) phonology and a theory of underspecification, the 



120 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

nucleus glide achieves a descriptive simiplicity that the onset glide 
does not. 

Take first the common phenomenon of devocalization in Korean, 
as in example (15). 

(15) no- a [nv^'a] 'to put down' ci-e [eye] 'lose (game, war)' 
tu-e [tue] 'leave behind' phi-e [phye] 'bloom' 

cu-e [ewe] 'give' 
po-a [pwa] 'see' 

One can describe this phenomenon as in (16): 

(16) a o 
/\ I 
C V V 

i I I 

n o a 

i 
w 

This description entails three problems: 

(17) a. V(o) gets linked to node C, 

b. compensatory lengthening obligatorily results, and 

c. it creates a complex consonant. 

In order to avoid these problems, one may posit a rule changing 
the skeletal position V into C as in example (18). 

(18) a a 
/\ I 
C V V 

i 

c 

If we do this, all of the above problems will disappear, but we 
have now created an onset cluster; a solution equally untenable as 
the previous one. 

These descriptive glitches will disappear in a "nuclear fusion", so 
to speak. In this process, two nucleus vowels coalesce or fuse as 
shown in example (19). 

(19) 



[nwa] 





a 
1 




^ 


a 


R 


R 







R 


1 +,' 


''1 




1 


^ 


C V 

1 1 


V 
1 


— > 


c 

1 


V V 
1 1 


1 1 
n 


1 

a 




1 
n 


1 1 
a 



C. W. Kim & H-Y. Kim: The character of Korean glides 1 2 1 

In this configuration, two vowels with identical features will 
generate a long vowel under the OCP, but when the two vowels have 
non-identical feature specifications (in underspecified features), the 
first vowel will be realized as a glide. The exact mechanism of this 
process has been worked out by Sohn (1987), and we won't go into 
further details here. What is to be noted is that this "nuclear fusion" 
gives a "clean" power (We're on an extended metaphorical trip here). 
We said "clean", because it does not generate any of the undesirable 
by-products mentioned above in (17). 

C-W. Kim (1977) pointed out that devocalization does not take 
place in Korean when the first of the two vowels is long, as in 
example (20). 

(20) coo(h)-a -^ *cwaa 'good' cf. no-a -^ nwa 'put down' 

too(w)-a -^ *twaa 'help' cf. tu-e -> twe 'leave behind' 

Since a long stem vowel is in general shortened before a vowel- 
initial affix, as in example (21), 

(21)t'wii-e -> t'wi-e 'to run' 
k'uu-e -^ k'w-e 'borrow' 

it creates a problem of indeterminacy at the point of application of 
the devocalization rule, i.e., how to prevent to-a (from too-a) from 
becoming *twa. But a simple condition on the nuclear fusion, i.e., the 
nuclear fusion cannot involve three or more vowels, can take care of 
this. It's as if a certain chemical reaction cannot take place when 
there is an extra chemical element. 

Another phonological phenomenon involving glides in Korean is 
METATHESIS which takes place under certain conditions, as exmplified 
in (22). 



(22) pyel 


— > peyl — > pel 'star' 


o-si-e-yo 


-> o-sy-e-yo -^ o-sey-yo -> o-se-yo 'come!' 




(Hon) 


hyeng-nim 


-^ heyng-nim -^ heng-nim 'elder brother' 


koyang'i 


-^ koayng'i -> kwayng'i ^ [k wae T)i ]'cat' 


syem 


-> seym -> se:m 'island' 



A description in terms of the glide as a part of an onset cluster, 
as shown in (23), involves the switching of two segments, one 
belonging to Onset and the other belonging Rhyme. While this may 
not be an implausible process (cf. English brid — bird, hrose — horse, 
etc.), it is more complicated and less plausible than a switch of two 
elements dominated by the same node, as shown in (24). 



122 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



(23) 



(24) 




< 



[-back] 

The example in (22) is not that of metathesis but compensatory 
lengthening. 2 If it is a case of CL and if glide is a part of onset, then 
we are describing here a cross-linguistically rare case of an onset 
consonant inducing CL. Ingria (1980) and others have shown that CL 
normally occurs within the Rhyme consitituent, as in example (25). 

(25) c o 

/\ 

R 

1 I 
C V 

I I 

d o -> s i i d o 'I sit' 

Regarding a glide as a part of the nucleus makes CL of the 
adjacent V within the rhyme a typologically more plausible and a 
descriptively simpler phenomenon. 

4. Concluding remarks 

In conclusion, we have argued that glides in Korean are best 
regarded as a part of the nucleus rather than as a part of the onset 
on both internal and external grounds. C-W. Kim (1968) argued 
earlier in an article on the system of Korean vowels that glides are 




i 



C. W. Kim & H-Y. Kim: The character of Korean glides 123 

distinctive features of vowels to be realized as superimposed 
(simultaneous) components in some cases (e.g., wi -^ [u]), but to be 
sequentialized as falling diphthongs in other cases (e.g., wa, we, etc.). 
Nearly a quarter of a century later, after a reexamination of the 
behavior of Korean glides, we find ourselves holding a modified but 
basically unchanged view. 

The title of this paper is "The character of Korean glides". As you 
know, the word character has two meanings: One meaning 'a 
distinguising quality,' and the other meaning "a graphic symbol.' As 
the final tidbit, we will examine how the glides are represented in 
the native script han'gul, which we think you will find very 
revealing. 

Five centuries earlier. King Sejong described the Korean vowel 
system as in (26). 

(26) a. Simple vowels 

• [o] depicts the (round) heaven; the tongue is retracted, 
and its voice is deep. ( -^i^iSjif-i^ ) 

— [i] depicts the (flat) earth; the tongue is slightly 
retracted, and its voice is neither deep nor shallow. 

(-g-'his^^Tvi^T-;^ ) 

I [i] depicts a (standing) man; the tongue is not 
retracted, and its voice is shallow. ( -^^^^^'.k) 

— [o] is the same as • , but the mouth is contracted/rounded. 

(^#»IS1-S7D^) 

!• [a] is the same as • , but the mouth is stretched/spread. 
( |. )^ . 15] ^ o jt ) 

T [u] is the same as — , but the mouth is contracted/rounded. 

•I [a] is the same as — , but the mouth is stretched/spread. 

b. Complex vowels 

-ii- [yo] is the same as -i- , but rises from I . (•f.^iis) r^^i^ \ ) 

\l [ya] is the same as !• , but rises from I . 

~^ [yu] is the same as ~ , but rises from I . 

:| [ya] is the same as •! , but rises from I . 

Note especially the description of the so-called "complex vowels" 
(26b). The sage monarch knew exactly what he was doing. He could 
have devised separate characters for [y] and [w], but he knew better. 



124 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Chong Lin-ji, one of the courtiers, wrote in the postscript to the 
royal edict proclaiming the creation of a native script: 

"Although there have been thousand kings in the East, none 
was wiser than his majesty." 

With admiration, we can only agree. 

NOTES 

*An earlier version of the paper was presented at the first 
Formal Linguistic Society of Midamerica meeting, May 1990, 
University of Wisconsin, Madison, WI, and at Linguistics Seminar, 
October 1990, University of Illinois, Urbana, IL. The authors 
benefited from the discussions that followed the presentation at both 
places. Any flawed arguments, however, are solely ours. 

1 The reason why l\J is not palatalized to [c] in the same context 
in such words as twi 'back' is probably because the /t/ - /c/ distinc- 
tion is phonemic in Korean, while the [s] - [s] distinction is not. 

2 We are not sure how general this historical process was in 
Korean, so we may be walking on thin ice here. 

REFERENCES 

Ahn, S-C. 1986. Syllabification process in Korean. Papers from the 

1985 Mid-America Linguistics Conference, 1-12. 
1988. A revised theory of syllable phonology. Linguistic Journal 

of Korea 13.333-362. 
GiM, C-G. 1987. A study on the syllable structure and some processes 

in its nucleus in Korean [in Korean]. Journal of the Language 

Institute [Maal], Yonsei University, Seoul, Korea, 12.25-69. 
INGRIA, R. 1980. Compensatory lengthening as a metrical 

phenomenon. Linguistic Inquiry 11.465-495. 
Kim, C. 1968. The vowel system of Korean. Language 44.516-527. 
. 1977. Vowel length in Korean. Studies in the Linguistic Sciences 

7:2.184-190. 
, & H. Sohn. 1986. A phonetic model for reading: Evidence from 

Korean. Studies in the Linguistic Sciences 16:2.95-105. 
, & H-Y. Kim. 1990. The plight of Korean glides. The First Annual 

Meeting of the Formal Linguistics Society of MidAmerica. 
Kim, J-M. 1986. Phonology and syntax of Korean morphology. 

University of Southern California Ph.D. dissertation in 

Linguistics. 
Kim, K-0. & M. Shibatani. 1976. Syllabification phenomena in Korean. 

Language Research 12:1.91-98. 



C. W. Kim & H-Y. Kim: The character of Korean glides 1 25 

KIM-RenAUD. Y-K. 1978. The syllable in Korean phonology. Papers in 

Korean Linguistics, ed. by C. Kim, 85-98. Columbia, SC: Hornbeam 

Press. 
Lee, B-G. 1982. A well-formedness condition on syllable structure. 

Linguistics in the morning calm 11, 489-506. Seoul, Korea: 

Hanshin Publishers. 
SOHN, H-S. 1987. Underspecification in Korean phonology. University 

of Illinois at Urbana-Champaign Ph.D. dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21. Number 2. Fall 1991 



PROSODIC PHONOLOGY OF KOREAN 

Hyoung Youb Kim 

Traditional accounts of Korean phonology have been 
syntax-insensitive in the sense that they have been done 
exclusively within a phonological framework without 
reference to any syntactic information. In recent years, 
however, evidence has accumulated to suggest that there is 
an area of phonology-syntax interface which could be 
fruitfully explored by way of accounting for certain 
phonological phenomena. Recently, Cho-Yu (1987, 1990) 
attempted to describe certain phonological phenomena in 
Korean within this new framework, now known as phrasal 
phonology. In this paper, I point out three problems in Cho- 
Yu (1987, 1990) that she does not address, and show how 
they can be resolved with the model proposed in Nespor & 
Vogel (1986). I will also argue that such morpho-syntactic 
phenomena such as compound words, multiple subjects, 
topicalization, and rightward movement can be more 
readily explained within a new model of prosodic domains 
that I postulate for Korean phonology. 

1. Introduction 

In Korean all phonological phenomena have been regarded as 
being syntactically insensitive. They have been accounted for 
exclusively within a phonological framework without reference to 
any syntactic information. However, Y-M. Cho-Yu (1987, 1990) 
illustrates some inconsistently applied phonological rules even 
though they are generally accepted as post-lexical rules which are 
applied without exceptions. She attempts to account for them by 
assigning different prosodic domains which are based on different 
syntactic structures. The prosodic domains are syllable and foot, 
phonological word, phonological phrase, and intonational phrase 
(Nespor & Vogel 1982. 1986). Each domain has different phonological 
rules which are limited to it. Unless they satisfy the conditions of 
each domain, the rules are not allowed to apply. 

Hayes (1984), Nespor & Vogel (1982. 1986). Selkirk (1984). and 
Kaisse (1985) have suggested several conditions for the domains. 
Cho-Yu (1987, 1990) suggests three domains: The phonological word, 
phonological phrase, and intonational phrase. However, the 



128 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

conditions for each domain have three serious problems. First, the 
definition of a phonological word does not show how compound 
words can be incorporated into the domain. Moreover, it cannot 
distinguish the difference between a subcompound and a co- 
compound. In the latter, the Post-Obstruent Tensing Rule does not 
apply even though in the former it does. Second, Cho-Yu (1987, 
1990) categorizes relative clauses, with a head and a specifier, into 
the same domain as the phonological word. The application of the 
Post-Obstruent Tensing Rule is supporting evidence for putting 
relative clauses in the domain of the phonological word. In order to 
account for them within the domain, Cho-Yu puts the clauses into the 
lexicon. However, the analysis of the clauses shows that she does not 
understand the structure of the clause. The relative clause is actually 
generated by a syntactic operation outside of the lexicon. Third, 
Cho-Yu claims that the phonological phrase formation is cyclically 
applied from the innermost maximal projection. The established 
phonological phrase is not sought to be reanalyzed even within the 
higher maximal projection. However, the cyclic application of the 
formation can produce an incorrect prosodic domain which blocks 
certain phonological phenomena even though they are expected at 
that position. 

Below I will elaborate these problems and show how they can 
be accounted for more appropriately within the framework of Nespor 
& Vogel (1982, 1986). 

2. Phrasal phonology 

In this section I will give a summary of Cho-Yu (1987, 1990) 
where we can see how a theory of phrasal phonology may be at work 
in Korean. In order to show why a theory of phrasal phonology is 
required she illustrates two sentences in (1) where the Voicing Rule 
is applied only in (la). Although in (lb), /k/ is located between the 
same vowels /i/ and /a/ as in (la), it does not become [g]. In order to 
account for the difference between (la) and (lb) she claims that the 
two /k/'s belong to two separate domains. In (la), /k/ is in the same 
domain with two flanking vowels, but in (lb), /k/ and the preceding 
vowel are in different domains. For the same reason, /p/ of /pang/ 
in (2) is voiceless, while /p/ of /kapang/ in (b) is voiced. 

(1) a. apaci-ka pang-e tilaka-si-n-ta. 

father-nom room-loc enter-honor-pr.- VE 
[absjiga p a r^e dilagasinda] 
'Father is entering the room.' 

b. apaci kapang-e tilaka-si-n-ta. 
father bag-loc 

[abaji kapar^e diragasinda] 
'Father is entering the room.' 



H. Y. Kim: Prosodic Phonology of Korean 1 29 

In her papers. Cho-Yu (1987, 1990) claims that there should be 
three domains in Korean within a theory of phrasal phonology: The 
phonological word (W), phonological phrase (P), and intonational 
phrase (I).i The domains are assumed to have hierarchical structures 
like that of syntax. Thus, every terminal constituent can be assigned 
to 'W, and the matrix sentence to T. 

The phonological word is defined as a domain which contains 
lexical items (N, V, A, etc.) and endings like particles and inflectional 
suffixes. All terminal constituents of a sentence can be phonological 
words. 

(2) The phonological word consists of a lexical item (noun, 
verb, adjective, determiner, and adverb) and any particle 
or any inflectional ending that follows it. 

(Cho-Yu 1987:330) 

The main phonological rules working in the domain are Palatalization 
and Post-Obstruent Tensing. In (3) below are Cho-Yu's examples of 
the two rules. 2 

(3) Palatalization 

a. tot-i -> toji 'rising'(derivation) 
rise-Nominalizer 

b. pat*i-i -> pac'^i 'the field'(inflection) 

Post-Obstruent Tensing^ 

a. hal kil -^ hal k'il 'way to do' 

b. titil panga -^ titil p'anga 'tread mill' 

The domain of the phonological phrase is defined on the basis of 
the concept of X'-theory where a head is the main part of a certain 
maximal projection: N of NP and V of VP, etc. A domain is constituted 
with a head and complements. 

(4) Korean P-Phrase Formation 

a. In [...Y" X] where X is the head of X" and Y" is an 
adjacent complement, the sequence Y" X forms a P- 
Phrase. 

b. All P-Words unaffected by (1) form P-Phrase. 

(Cho-Yu 1987:335) 

A prime example of a phonological rule working in the P domain is 
the Voicing Rule. For example, (5) shows how /k/ in (la) and (lb) 
are assigned to separate domains. 



130 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(5) a. S 




apaci-ka pang-e tila-ka-sin-ta 

[abajiga] 




tila-ka-sin-ta 



3. Problems 

As I have already mentioned in section 1, the conditions which 
decide each domain are not enough to account for all phonological 
phenomena related to each domain. In this section, I will examine 
some of these problems in more detail. 

3.1 Compound words 

In Cho-Yu (1987:333) a compound is a word, and it is 
constructed in the lexicon. She refers to the leveling strata 
organization of the lexicon to show that compounding is a part of the 
lexicon. Any compound word can be a phonological word, and the 
Post-Obstruent Tensing Rule is applied in the domain. Indeed, a great 
deal of compound words in Korean go through the rule in the domain. 
However, rule (2) is not enough to assign the domain to a compound 
word. It actually has two problems which make (2) account 
inadequately for compound words. First, according to (2), a 
compound word cannot be a phonological word because all compound 
words are composed of more than two lexical items (Selkirk 
1982:13). Second, in the examples in (6) we see that some compound 



H. Y. Kim: Prosodic Phonology of Korean 1 3 1 

words are not influenced by the Post-Obstruent Tensing Rule even 
though they are phonological words and have the same phonological 
environment as those influenced by the rule. In addition, they have a 
different morphological structure from those undergoing the rule. For 
instance, we can insert kwa 'and' between the internal elements 
without changing their semantic contents. On the other hand, the 
compound words influenced by the Tensing rule have the 'adjunct + 
head' structure. Thus, different morphological structures appear to 
determine their prosodic structures. The former are Co-compounds 
and the latter are Subcompounds (Y-S. Kim 1984:152-157). 

(6) a. Co-compounds 

kyecip casik 'wife and children' 

kusok kusok 'every nook and cranny' 

kos kos 'place and place; everywhere' 

b. Subcompounds 

pom p'i 'spring rain' 

ip p'alis 'manner of speech' 

sol p'angul 'a pine cone' 

3.2 Relative clauses 

According to Cho-Yu's analysis, the head and the preceding verb 
of a relative clause should be grouped together into a phonological 
word because they make a tight phonological unit in which the initial 
consonant of the head becomes tensed when the preceding verb ends 
with '1', which comes from the inflectional suffix of the future tense: 
'-(i)!'.'' But, as mentioned in fn.3, '1' is not an obstruent. In order to 
account for this case, she argues that relative clauses must be 
constructed in the lexicon by the compounding process. Namely, a 'C 
(an underspecified segment) is inserted between the verb and its 
head, which satisfies the environment of the Post-Obstruent Tensing 
Rule. 5 After tensing the initial consonant of the head, 'C is deleted by 
the Stray Erasure Convention. <^ 

(7) /hal C kil/ 'way to do' 

hal C k'il Post-Obstruent Tensing 

h a 1 C k' i 1 Syllabification 

s s 

h a 1 k' i 1 Stray Erasure 

^^ ^^ 

s s 

[hal k'il] 



132 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

However, derivation of [hal k'il] in example (7) has two serious 
problems. First, the ending /-I/ of /hal/ requires the looping 
between the subcompound level and the inflection level, since /-I/ is 
an inflectional suffix and /hal C kil/ is at best a subcompound. This 
weakens the theory of strata of the lexicon. ^ Second, the derivation 
in (7) is not a morphological, but a syntactic process. The relative 
clause formation is a syntactic construction which is generated by 
syntax. Song (1978) shows how the phrase structure rule (PS rule) 
produces the relative structure of Korean: NP ^ S N. 

3.3 Cyclicity of P-phrase formation 

In the case of the complex sentences with embedding clauses, 
rule (4) exhaustively assigns phonological phrases to the sentences 
from the innermost maximal projection of embedding clauses. Once a 
domain is set, it is not reanalyzed because of the higher maximal 
projection. Example (8) illustrates how the phonological phrases are 
assigned to complex sentences. In (8b), the initial consonant of 
/poassta/ 'saw' is voiced, but in (8a), the same consonant is not 
voiced even though it appears in the same phonological environment. 
According to Cho-Yu (1987) the voicing of /p/ depends on whether it 
is located in the same domain with the preceding and following 
elements. 

(8) a. Syntactic Bracketing 

[[[s9nsaeT)nim-k'ess][cusi-n]] kilim-il] po-ass-ta] 
teacher-Nom give-Mod picture-Acc see-Past-VE 

'(I) saw the picture that the teacher gave." 

Phrasal Bracketing 
[ssnsaerjnimk'esa] [cusin girimil] [poatt'a] 

b. Syntactic Bracketing 

[[[s8nsaeT)nim-k'es8 [[haksasT)-eke] cusi-n]] kilim-il] 

student-Dat 
po-assta] 
'(I) saw the picture that the teacher gave the student.' 

Phrasal Bracketing 
[s9naeT]nimk'esa] [haksaer^ege jusin] [kirimil boatt'a] 

However, the cyclic application of rule (4) causes two problems. 
First, it actually cannot produce the two different prosodic structures 
found in (8a') and (8b'). In (8a'), /cusin/ 'give' is under the VP of the 
embedding sentence, and the VP can be an independent phonological 
phrase. When a new domain is established in an upper hierarchy like 
NP, the verb is excluded from the new domain. According to this 
analysis, (8a) and (8b) have the same prosodic structures. 



i 



H. Y. Kim: Prosodic Phonology of Korean 
(8)a'. S 

NP VP 



133 




NP 

I 
N N 
I I 

(I) sasaeT)nim-k'es9 cusi-n kilim-il po-ass-t; 




NP V 

I I 

(I) sasstinim-k'csa haksaT)-eke cusi-n kilim-il po-ass-ta 

P P P 

Second, it is not appropriate that the phonological phenomenon is 
blocked due to an added constituent within the innermost maximal 
projection. For example, in (8b) the initial consonant of /poassta/ 
'saw' is not voiced because of the noun phrase /haksajr^eke/ 'to a 
student'. Unlike syntax where a question pronoun (who, which, what, 
etc.) is coindexed with the original position no matter how far the 
position is from the pronoun in a sentence, phonology is concerned 
only with specific environments within a limited range.** 

4. A new Tramework 

As I mentioned in section 1, I will show that the problems in 
section 3 can be solved by the new domain conditions of Nespor & 



134 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Vogel (1986). In section 4.1, I will introduce the conditions in detail, 
and in section 4.2, I will show how they can account for the problems 
that I have pointed out in the previous section. 

4.1 Prosodic domains 

In this section I will show how a phonological word and a 
phonological phrase may be defined. According to Nespor & Vogel 
(1986), a phonological word is constructed with a stem, and the 
preceding and following elements of phonology and morphology. 

(9) Phonological Word [W]9 

a. a stem 

b. any element identified by specific phonological 
and/or morphological criteria 

c. any unattached elements form a W on their own 

The definition of a phonological phrase consists of two parts such as 
Phonological Phrase Formation and Domain Restructuring (optional). 
The domain of a phonological phrase is constructed with a lexical 
head (X) and complements on the nonrecursive side. Because Korean 
is left recursive, the domain must consist of a head and complements 
on the right side of the head.io The domain restructuring is to 
eliminate as many non-branching phonological phrases as possible. 

(10) a. Phonological Phrase Formation (P) 

The domain of phonological phrase consists of a C 
which contains a lexical head (X) and all the Cs on its 
nonrecursive side up to the C that contains another 
head outside of the maximal projection of X. 

b. Phonological Phrase Restructuring (optional) (P') 

A nonbranching phonological phrase which is the first 
complement of X on its recursive side is joined into the 
phonological phrase that contains X. 

4.2 Solutions to the problems 

In section 3, 1 have pointed out three problems caused by the 
definitions of domains by Cho-Yu (1987). In this part, I will show 
how rules (9) and (10) can overcome these problems. 

First, it must be mentioned that rule (2) cannot put a compound 
word into a phonological word. If we observe the morphological 
structure of compound words, we will find that it can be included 
into the domain of a phonological phrase instead of a phonological 
word. In order to be a phonological phrase, a compound word must 
have a head. Selkirk (1982:20) shows how a compound word has a 
head in the case of English. Y.-S. Kim (1984:154) says that in Korean 
(endocentric) compounds, the head is invariably on the right-hand 
side of the compound. After the head is decided, rule (10b) can join 
the complement on the left side which is recursive in Korean. 



H. Y. Kim: Prosodic Phonology of Korean 



35 



Example (11) shows how (10a) and (10b) assign the phonological 
phrase to compound words. 

[ipp'alit] 'manner of speech' 



(11) ip pslis 
H 
I 
P 



ip palis 
I I 
I P 
1/ 
P' 



The words in (6), which are illustrated as counterexamples of the 
Post-Obstruent Tensing Rule, can also be accounted for if we consider 
each of them as having double heads (Y-S. Kim 1984:157). Thus, they 
have two separate phonological phrases, which will block the rule as 
shown in (12). 

(12) kyecip casik 
H H 



Second, rule (10) can account for phonological phenomena 
occurring between the preceding verb and the head of the relative 
structure without returning it to the lexicon.' • Rule (10) also can 
explain why the cyclic formation of phonological phrases from the 
innermost maximal projection is not appropriate to account for the 
phonological phenomena occurring in the matrix sentence. For 
example, I will show that (8a) and (8b) actually have the same 
phrasal structure. In (8a) and (8b), /cusin/ and /kilimil/ are always 
in the same phonological phrase even though the inner structure is 
changed in (8b). According to (10), the phrasal structures of (8a) and 
(8b) are as follows: 



(13) 




(10b) 



136 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 

VP 




N V 

I I I I I 

sansaeinimk'ess haksasT^eke cusin kilimil poassta 

I I III 

w w w w w 

I I \ I I 

P P Y P P (10a) 

P* (10b) 

4.3 Other morpho-syntactic phenomena 

In section 4.2 we saw how the new framwork solves the 
problems caused by that of Cho-Yu (1987). In this part, I will show 
that the range of its application can be expanded to account for such 
morpho-syntactic cases as complex compound words, multiple 
subjects, topicalization, and rightward movement. 

4.3.1 Complex compound word 

I assume that there are two kinds of complex compound words 
in Korean. One type is a compound word composed of more than 
three lexcal items (N+N+N+...). The other is a compound word 
containing affixes (V+V+...+Affix) (Y-S. Kim 1985:157). In a compound 
word with more than three composite lexical items, the initial stop 
consonant of a certain lexical item is tensified according to the 
function of the preceding item: Adjunct or non-adjunct. Some 
examples of this are shown in (14). In (14a), /kuksu/ 'noodle' is 
modified by /hankuk/ 'Korea'; and, in (14b) /cip/ 'restaurant' is 
modified by /kuksu/. Thus, (14a) means 'a restaurateur of Korean 
noodle' while (14b) means 'a Korean in a noodle restaurant' . 

[ciplN [salamJNlN 



(14) a. [[hankuk kuksu]^ 

Korean noodle 

[hanguk k'uks'u 

b. [[hankuklN [kuksu 

Korea noodle restaurant 

[hanguk kuks'u c'ip 



restaurant person 
c'ip s'aram] 



ciplN 



[salam]N]N 

person 

s'aram] 



H. Y. Kim: Prosodic Phonology of Korean 



37 



According to the derivation illustrated in (11), tensification occurs 
within the endocentric compound with an adjunct and a head, which 
also constructs a phonological phrase. The analyses of the phrases in 
(14) are as follows: 

(15) a. N b. N 



salam 





salam 



hankuk #kuksu =cip 
P 



hankuk = kuksu 
I 
P 



[hanguk k'uksu c'ips'aram] [hanguk kuks'uc'ips'aram] 

(Chen 1987:129) 

The examples in (16) are compound words which contain an 
affix. In (16a) the affix is attached only to the right element and 
changes the verb into a derived noun. In (16b) the affix is attached 
to a compound verb and changes it into a derived noun (Y-S. Kim 
1985:154). 

(16) a. [[hae]N # [tot + 1]^]^ 'sun rising' 

sun rise + Nominalizer 

[hae toji] 

b. [[[mil] # [tat]]v + i]N 'sliding door" 

push shut Nominalizer 
[mi taji] 

Example (9) can explain how the compound words with different 
morphological structures show the same phonological result with 
respect to rules like palatalization. Even though (17a) and (17b) have 
different morphological structures, the definition in (9) can assign 
the phonological word to a verb-affix constituent in which 
palatalization occurs. 

(17)a. 





(Ne.spor & Vogel 1986:117) 



138 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



4.3.2 Multiple subjects 

In (18), I show two sentences structured with the same words 
which, however, undergo different phonological phenomenon, e.g., 
voicing. In (18a), the initial consonant of the second word is not 
voiced, but in (18b), the same consonant is voiced. In order to 
account for the difference, I will claim that (18a) and (18b) have 
different syntactic structures. (18a) will be regarded as a double 
subject sentence which is possible in Korean (Y-J. Yim (1985) and YS. 
Yoon-James (1987)). If we assume that the subject markers (/-i/ and 
/-ka/) are optional, then we can generate (18a) without any 
problem. (18b) will be regarded as including a compound word 
subject. 12 

(18) a. [[khok'ili] [kwi] khita] 
elephant ear big 
[khok'iri kwi khita] 

b. [[khok'ili kwi] khita] 
elephant ear big 
[khok'iri gwi khita] 

In order to account for the difference we can assign the phonological 
phrase differently where voicing occurs. The diagrams in (19) show 
how the domain is assigned. 

(19)a. S 



'elephant has big ears' 



'elephant's ears are big' 



NP 

I 
khok'ili 



NP 



VP 



b. 



khok'ili 
I 
P 



kwi 
kwi 



khita 
khita 



NP 




VP 


PP 


^? 


khiti 


NP (P) 

khok'ili (-iy) 
khok'ili 


kwi 
1 
P 


khitj 



H. Y. Kim: Prosodic Phonology of Korean 1 39 

4.3.3 Topicalization (Yang 1973) 

In (20) I illustrate a pair of sentences that behave differentially 
with respect to the tensification rule even though they are composed 
of the same words. The initial part of (20a) does not have any 
tensified consonant, but that of (20b) does.' 3 



(20) a. [[cip] [puca-ka] ciasta] 
house richtnan-SUB built 



'a richman built a house" 



[cip pujaga cidtt'a] 

b. [[cip puca]-ka ciasta] 'a man of many 

[cip p'ujaga ciatt'a] houses built (a bldg.)' 

In order to account for the different phonological output, I claim 
that (20a) and (20b) have different syntactic structures. The first one 
is structured with topicalization which moves a word located in the 
middle of the sentence to the initial position of the sentence. The 
second one has a compound word in the subject position. 

(21) a. 




4.3.4 Rightward movement (Choe 1987) 

The sentence in (22) is one which does not undergo voicing even 
though it includes the appropriate environment for the phonological 
rule. 14 



140 



Studies in the Linguistic Sciences 21:2 (Fall 1991) 



(22) cangnim-i khitako mitninta ki cip-i. 

blind-SUB big believe that house-SUB 

[cangnimi khitago minninta ki cibi] 

Although this appears to be an exceptional case, I claim that it is not 
because the sentence is derived by moving /ki cip/ to the final 
position of the sentence. (23) shows how the sentence is constructed 
and how the phonological phrase is assigned to the sentence. 



(23) 




cangnim-i 



5. Conclusion 

In this paper I have attempted to show that the prosodic 
conditions of Nespor & Vogel (1982, 1987) are more appropriate 
than those of Cho-Yu (1987, 1990). In order to support this position, 
I have illustrated three problems not addressed in Cho-Yu (1987, 
1990), and have examined them in detail. In the last part of this 
paper, I also have showed how the prosodic conditions of Nespor & 
Vogel (1986) can readily account for other morpho-syntactic 
structures such as complex compound words, multiple subjects, 
topicalization, and rightward movement. 



NOTES 

1 In this paper, I will deal only with the phonological word and 
phonological phrase. 

2 The definition in (2) does not allow a derivational suffix in the 
phonological word. Thus, (3a) is not an appropriate example to 
illustrate (2). 

3 The examples are not relevant to the rule because /I/ is not an 
obstruent. 



H. Y. Kim: Prosodic Phonology of Korean 141 

^ When the preceding verb ends with the inflectional suffix of 
the present tense Anin/, the initial consonant of the following head 
does not become tensed. 

5 In order to account for tensing in compounding, several people 
have proposed a rule of insertion of a segment such as r (C-W. Kim 
1970b), 7 (C-B. Kim 1974), and X (H-S. Sohn 1987). 

6 It is to erase a segment which is not linked to any syllable. It 
also works in consonant cluster simplification. 

^ According to S-C. Ahn (1975), the lexicon of Korean is stratified 
as follows: 

(1) Subcompound 
Cocompound 
Derivation 
Inflection 

8 In modern linguistics, syntactic rules are context-free and 
phonological rules are context-sensitive. 

9 For the purpose of this paper I have paraphrased the 
definition of Nespor & Vogel (1987:141). 

10 In Korean, it does not seem to be possible to have some 
complements on the nonrecursive side, unlike English and French 
which are right recursive and have some examples where an 
adjective follows a noun. 

(2) something special 

robe elegante (Fr.) 'elegant dress' 

" In Korean, the Tensing Rule should be applied in both the 
lexicon and post-lexicon, when the preceding element ends with /. In 
the case of the post-lexicon, / is originated from the inflectional suffix 
of the future tense. Cho-Yu (1987:329-331) states that tensification 
occurs in both parts and also shows how the rule looks. 

12 The full representations of each sentence are as follows: 

(3) a. khok'ili-ka kwi-ka khita. 
b. khok'ili-ly kwi-ka khita. 

•3 The full representations of (20) are as follows: 

(4) a. cip-il pucaka cissta {<r- puca-ka cip-il ciasta) 
b. cip pucaka ciasta 

i"* The structure before rightward movement is as follows: 

(5) cangnim-i ki cip-i khitako mitninta. 
blind-SUB thai housc-SUB big believe 

'a blind believes that that hou-se is big' 



142 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

REFERENCES 

Chen, M. Y. 1987. The syntax of Xiamen tone sandhi. Phonology 

Yearbook 4.109-149. 
ChO-Yu, Y-M. 1987. Phrasal phonology of Korean. Harvard Studies in 

Korean Linguistics 2, ed. by S. Kuno et. al. Cambridge: Harvard 

University, 328-340. 
1990. Syntax and phrasing in Korean. The phonology-syntax 

connection, ed. by S. Inkelas & D. Zee. University of Chicago. 
Choe, Hyun-Sook. 1987. Successive-cyclic rightward movement in 

Korean. Harvard Studies in Korean Linguistics 2, ed. by S. Kuno 

et al. Cambridge: Harvard University, 40-56. 
Hayes, B. 1984. The phonology of rhythm in English. Lingusitic 

Inquiry 15.33-74. 
KaisSE, E. M. 1985. Connected Speech: The interaction of syntax and 

phonology. New York: Academic Press. 
Kim, Y-S. 1984. Aspects of Korean morphology. University of Texas at 

Austin Ph.D. dissertation in Linguistics. 
Kim, H-Y. 1990. Voicing and tensification in Korean: A multi-face 

approach. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 
Nespor, M., & I. Vogel. 1982. Prosodic domains of external sandhi 

rules. The structure of phonological representations Part I, ed. 

by H. van der Hulst & N. Smith et. al. Dordrecht: Foris, 225-255. 

1986. Prosodic phonology. Dordrecht: Foris. 

Selkirk, Elisabeth O. 1982. The syntax of words. Cambridge: MIT 

Press. 
1986. On derived domains in sentence phonology. Phonology 

Yearbook 4.371-405. 
SOHN, H-S. 1987. Derivational morphology and tensification in Korean 

Compounds. Harvard Studies in Korean Linguistics 2, ed. by S. 

Kuno et. al. Cambridge: Harvard University, 443-455. 
Song, Zeno. 1978. Noun complementation in Korean. Korean 

Linguistics, 108-127. The International Circle of Korean 

Linguistics. 
Yang, Dong-Whee. 1973. Topicalization and relativization in Korean. 

Indiana University Ph.D. dissertation in Linguistics. 
YIM, Young-Jae. 1985. Multiple subject constructions. Harvard Studies 

in Korean Linguistics 1, ed. by S. Kuno et. al. Cambridge: Harvard 

University, 101-109. 
YOON, James Hye-Suk. 1987. Some queries concerning the syntax of 

multiple subject constructions in Korean. Harvard Studies in 

Korean Linguistics 2, ed. by S. Kuno et. al. Cambridge: Harvard 

University, 138-162. 



Studies in the Linguistic Sciences 
Volume 21, Number 2. Fall 1991 



THE PRAGMATICS OF THE PRAGMATIC MORPHEME COM 'A 
LITTLE' IN KOREAN 

Han-gyu Lee 

Pragmatic morphemes in Korean are used to represent 
different presuppositions or attitudes of the speaker in a 
discourse. They play an important role in a discourse to 
help the hearer to understand the speaker's intention or 
goals. This paper is a study of how a speaker uses the 
pragmatic morpheme com 'a little' and how the hearer 
interprets it in the discourse. 

This paper intends to describe how a speaker uses the pragmatic 
morpheme com 'a little' and how the hearer interprets it in the 
discourse. The term 'Pragmatic Morphemes' (PM) will be used to 
indicate bound morphemes which do not make a contribution to the 
truth-conditional meaning of a sentence containing them, but which 
affect pragmatic interpretations of the sentence, and which are 
attached to any syntactic category in a sentence such as NP, VP, PP, 
AdvP, and S. For instance, each parenthesis in (1) indicates a possible 
position for pragmatic morphemes. 

(l)Suni-ka-() alumtawun kongwen-eyse-() Yonghi-lul-() 
NM beautiful park-in AC 

cacwu-() mannasseyo-(). 
often met 

'Suni often met Yonghi in the beautiful park.' 

There are dozens of PMs in Korean. They are used to represent 
different presuppositions or attitudes of the speaker in a discourse 
without changing the truth-conditional meaning of a sentence 
containing them. The purpose of this paper is to describe the uses of 
the PM com according to the Gricean Cooperative Principle. 

The PM com has various uses in a discourse. The speaker can 
use it to show politeness by minimizing a threat to the hearer, or to 
insult her by belittling her ability, and the like.' For instance, the 
speaker can say (2a) or (2b) to achieve his goal (getting the salt) by 
making the hearer pass him the salt at a dinner table. The only 
difference between (2a) and (2b) is whether the PM com is used or 
not. As a request, however, utterance (2a) containing the PM com 
sounds polite, while (2b) without it sounds rude. 



144 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(2) a. sokum-com cwuseyyo. 

salt-PM give 

'Please, pass me the salt." 

b. sokum cwuseyyo. 
salt give 

'Pass me the salt.' 

When the speaker says (2b) without the PM com, he gives the 
impression that he has a right to order the hearer to give him the 
salt and that she is responsible for doing what he orders. This is why 
(2b) sounds rude as a request. On the other hand, utterance (2a) as a 
request is not rude in this way because of the politeness function of 
the PM com. In a discourse, the speaker can use the PM com to show 
politeness toward the hearer by minimizing a face-threatening act 
such as a request or an order. 

Depending on context, an utterance containing the PM com may 
sound ruder than the corresponding one without it. For instance, 
when the speaker scolds the hearer, a student, who does not study, 
but just goofs off, he can say (3a) or (3b) to urge her to study. 
However, (3a) sounds ruder than (3b), because the PM com implies 
that the speaker threatens the hearer's face by minimizing the 
hearer's ability of studying. 

(3) a. kongpu-com hayla. 

study do 

'Please, do study.' 

b. kongpu hayla. 
study do 
'Study.' 

Furthermore, different locations of the PM com in a sentence 
represent the speaker's different goals in a discourse, as seen in (4). 

(4) A: mues-ul manhi cwulkka? 

what-AC a lot give-shall 
'What shall I give you a lot of?' 

B: a. ppang-ul-com manhi cwuseyyo. 
bread-AC a lot give 

'Please, give me a lot of BREAD.' 

b. #ppang-ul manhi-com cwuseyyo. 
bread-AC a lot give 

'Please, give me A LOT OF bread.' 

As an answer to A's question in (4), only (4a) is appropriate. The 
only difference of (4a) and (4b) is in the location of the PM com; after 
ppang 'bread' in (4a) and after manhi 'a lot' in (4b). Different 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 145 

locations of the PM com results in the (in-)appropriateness of (4a) 

and (4b) in situation (4). The speaker draws the hearer's attention to 

a phrase whose information is significant for his goal, by placing the 

PM com after that phrase. Because of this focus use of the PM com, 

(4b), not (4a), will be good as an answer to the question (4'). 

(4') ppang-ul elmana cwulkka? 
bread-AC how much give-shall 
'How much bread shall I give you?' 

I will demonstrate in this paper that various pragmatic uses of 
the PM com, such as those illustrated in (2) and (3), are inferred 
from the sense of the PM com in accordance with the Gricean 
Cooperative Principle (CP), and that the location of the PM com in a 
sentence interacts with its sense to produce an enriched pragmatic 
interpretation of the sentence as seen in (4). Following Lee (1992), I 
am using the term 'sense' to refer to how a form functions to 
contribute to an interpretation of a sentence's truth-conditionally or 
non-truth-conditionally. I claim that the sense of the PM com is that 
the significance of what a com -suffixed phrase represents is little (or 
minimized). What kind of significance will be minimized depends on 
the speaker's goals in a discourse. When the speaker uses the PM 
com in a request sentence like (2a), the amount of salt he needs is 
minimized so that he expects the hearer to understand that passing a 
little amount of salt will not be a burden to her. Using the PM com, 
he can hedge his request by diminishing the burden or pressure it 
puts on the hearer. This is why (2a) sounds polite as a request. The 
speaker can also use the PM com to insult the hearer by belittling 
her ability to do what a com -suffixed phrase represents, as seen in 
(3a). By saying (3a), the speaker implies that a little studying is 
enough from the hearer even though more than a little is expected. 
In this way, the speaker can belittle the hearer's ability. 

This paper is organized in the following way. In section 1, 1 
describe how a speaker and a hearer engage in a conversation 
according to the Gricean cooperative principle. In section 2, I will 
distinguish the PM use of the form com 'a little' from its adverb use, 
demonstrating their differences. In 3, various pragmatic uses of the 
PM com are discussed as inferences from its basic sense. Employing 
the concept 'face' (Brown & Levinson 1978, 1987), I will describe the 
pragmatic uses of the PM com. In section 4, I show how the location 
of the PM com is relevant to the speaker's goals, that is, how the 
sense of the PM com interacts with its location in a sentence to affect 
the total pragmatic interpretation of the sentence. I claim that the 
focus use of the PM com is a side effect of its pragmatic uses. 



146 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

1. The Gricean Cooperative Principle and the PM com 

Grice (1975) takes discourse to be a purposive behavior: Talk 
exchanges are cooperative efforts in that the speaker converses 
rationally to achieve his goals and the hearer tries to recognize what 
they are. Grice (1975:45) states this as the Cooperative Principle. 

The Gricean Cooperative Principle (CP) 

Make your conversational contribution such as is required, at 
the stage at which it occurs, by the accepted purpose or 
direction of the talk exchange in which you are engaged. 

According to the CP, when a speaker utters what he intends to 
say, he has goals, and believes that his hearer assumes his utterance 
is rational for his goals and that she is rational enough to be able to 
work them out from what is said. 7. 8 When hearing what is said, the 
hearer believes that the speaker uttered it to achieve his goals, and 
she rationally infers what they must be. So even when what the 
speaker said appears not to be directly related to his goals in a 
discourse, the hearer assumes that his utterance is goal-oriented, and 
she will infer how it is relevant to the goals which she assumes he 
has. The Gricean inferences account for the gap between what is 
literally said and what is intended to be understood, and keep talk 
exchanges rational and reasonable. 

The talk exchange in (5) is one of the examples which Grice 
presents to show how what is intended to be understood is inferred 
from what is said according to the CP. In example (4), A and B are 
talking about a mutual friend, C, who is working in a bank. 

(5) A: How is C getting on in his job? 

B: Oh quite well, I think; he likes his colleagues, and he 
hasn't been to prison yet. (Grice 1975:43) 

In B's answer, the remark that C hasn't been to prison yet is 
apparently irrelevant as an answer to A's question, and B appears to 
make no conversational contribution (violates the CP). However, what 
B intended in the remark is that C is potentially dishonest and his 
colleagues are also treacherous. He believes that A is able to infer his 
intended interpretation from the literal meaning of his utterance, if 
she believes that he is abiding by the CP. When hearing B's saying, 
which is apparently irrelevant, A assumes that B's remark was 
uttered as a reasonable answer to her question, and she will have to 
figure out how B's remark is relevant to his goal in the discourse in 
order to make his remark consistent with the assumption that B is 
conforming to the CP. That is, A thinks that B's remark will be 
rational if A supposes that B implies that C is dishonest because 
dishonest persons are in jail. So A understands that B intended to 
imply that C is potentially dishonest. 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 147 

In a way similar to the Gricean conversational inferences just 
discussed, how the speaker uses the PM com and how the hearer 
understands its uses can be explained according to the CP. That is, 
the speaker's using the PM com in his utterance is relevant to his 
goal at the moment in the way indicated by the sense of the PM com. 
When the speaker uses the PM, he expects the hearer to be able to 
recognize why he used it. The hearer assumes that the speaker used 
the PM to support his goal, and she will infer from its sense how it is 
relevant to his goal. In section 3, I will describe how the speaker 
uses and the hearer interprets the PM com in conformity with the CP. 

2. PM uses and adverb uses 

The form com is a contraction of the adverb cokum 'a little'. It 
has been taken for granted by Korean linguists that the form com is a 
free morpheme since cokum is a free morpheme. My argument here 
is that the form has two uses, PM use and adverb use, and the PM 
use (called PM com here) is a bound morpheme while the adverb use 
(called Adverb com here) is a free morpheme. 2 I will show this by 
comparing the linguistic (syntactic, semantic, and phonological) 
behaviors of the PM use and the adverb use of the form com. 

There are three phonological reasons to distinguish the PM com 
from the adverb com. First, although the adverb com and the PM com 
have the same form, a native Korean speaker distinguishes between 
them when uttering sentences containing them. As illustrated in 
(6a)-(6b), the PM com is pronounced in a sequence with its 
immediately preceding word as if it were a part of the word, so that 
they form a phonological word; that is, the speaker does not put a 
slight pause between the PM com and its preceding word. On the 
other hand, in (6c), a slight pause comes between the adverb com 
and its preceding word. 

(6) a. ton-ul com cwul-kka, manhi cwul-kka? 
money-AC a little give-Q much give-Q 

'Shall I give you a small amount of money or large 
amount of money?' 

b. *ton-ul-com cwul-kka, manhi cwul-kka? 
money-AC-PM give-Q much give-Q 

'Shall I give you a small amount of money or large 
amount of money?' 

c. ton-ul-com cwul-kka, sathang-ul-com cwul-kka? 
money-AC-PM givc-Q candy-AC-PM give-Q 
'Shall I give you money or candy?' 

In sentences (6a) and (6b), the two quantitative words, com 'a little' 
and manhi 'much' are in contrast. To convey the intended meaning. 



148 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

com in (6a) and (6b) is pronounced as an independent word by 
placing a pause before the word com. When com is pronounced as a 
part of its preceding word ton-ul 'money' as in (6b), the sentence 
cannot represent such a contrast as in (6a). When Koreans hear (6b), 
they expect some noun in contrast with ton 'money' in the first 
clause to appear in the second clause, because they realize that com 
in (6b) has no literal meaning and it is a PM. However, the second 
clause in (6b) does not have such a noun. This is why (6a) is good 
and (6b) is bad for conveying the intended meaning. On the other 
hand, in (6c), two things, ton 'money' and sathang 'candy', are in 
contrast and the PM com is pronounced as a part of both words. 

Second, the PM com undergoes tensification which occurs right 
after a stop within a phonological word level, but not between words. 
The first sound Id of the PM com is tensified where the word 
preceding the PM com ends with a stop, so that it is pronounced /ts/. 
On the other hand, the first sound Id of the adverb com is not 
tensified in the same environment. So the PM com in (7a) is 
pronounced /tsom/, and the adverb com in (7b) as /com/. 

(7) a. ttek-com manhi cwuseyyo. 

cake-PM much give 
'Please, give me a lot of cake.' 

b. ttek com manhi cwuseyyo. 
cake a little much give 
'Give me a little more cake.' 

Third, the PM com is not stressed in a sentence, while the 
adverb com can be. When the adverb com is stressed, it tends to 
undergo tensification to be pronounced /tsom/, irrespective of its 
circumstance. 

Semantically, the PM com does not affect the truth-conditional 
meaning of a sentence containing it, while the adverb com does. For 
instance, in (8a) where the PM com is used, if it is interpreted as 'a 
little', then the sentence would imply that the speaker is asking the 
hearer to read a lot. However, this interpretation is quite the 
opposite of the intended meaning given in (8a). Instead, the PM com 
is interpreted as 'please' because it indicates that the speaker's 
politeness comes from softening his request or order. On the other 
hand, the adverb com in (8b) has the meaning 'a little'. 3 

(8) a. chayk-com po-cimaseyyo. 

book-PM see-don't 

'Don't read the book, please.' 
b. kwuk-i com ccayo. 
soup-NM a little is-salty 
'The soup is a little salty.' 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 149 

Syntactically, there are four reasons to distinguish the PM com 
from the adverb com. First, they occur in different syntactic positions 
in phrases. The PM com is placed right after a phrase in a sentence 
(9), while the adverb com comes before the phrase it modifies (10a), 
but not after the phrase (10b). 

(9) nay-ka Suni-lul cacwu-com mannassta. 
I-NM AC often-PM met 

'I met Suni often.' 

(10) a. nay-ka Suni-lul com cacwu mannassta. 

I-NM AC a little often met 

'I met Suni somewhat often.' 

b. *nay-ka Suni-lul cacwu com mannassta. 
l-NM AC often a little met 

'I met Suni somewhat often.' 

Second, the PM com can be placed after any phrase of a sentence 
as shown in (11), while the adverb com cannot, as in (10b). 

(11) a. nay-ka-com Suni-lul cacwu mannassta. 

I-NM-PM AC often met 

'I met Suni often." 

b. nay-ka Suni-lul-com cacwu mannassta. 
I-NM AC-PM often met 

'I met Suni often.' 

c. nay-ka Suni-lul cacwu-com mannassta. 
I-NM AC often-PM met 

'I met Suni often.' 

d. nay-ka Suni-lul cacwu mannassta-com. 
I-NM AC often mct-PM 

'I met Suni often.' 

Third, the PM com can occur more than once in a sentence; it can 
appear in every possible position simultaneously, as illustrated in 
(12a). However, the adverb com lacks this property (12b). 

(12) a. chayk-ul-com manhi-com ilke-com poseyyo-com. 

book-AC-PM many-PM rcad-PM try-PM 

'Please, please, read many books.' 

b. *cahyk-ul com manhi com ilke com poseyyo com. 
book-AC a little many read try 

*Read a little a little more a little books a little." 

Fourth, the PM com can occur adjacent to the adverb com or its 
original form cokum 'a little' in a sentence, while the adverb com 



150 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

cannot modify or be modified by cokum or itself, as seen in (13)- 
(15). 

(13) a. ton-ul-com cokum cwuseyyo. 

money-AC-PM a little give 
'Please give me some money.' 

b. *ton-ul com cokum cwuseyyo. 
money-AC a little a little give 
'?Give me some, some money.' 

(14) a. ton-ul-com com cwuseyyo. 

money-AC-PM a little give 
'Please give me some money.' 

b. *ton-ul com com cwuseyyo. 
money-AC a little a little give 
'?Give me some, some money.' 

(15) a. ton-ul cokum-com cwuseyyo. 

money-AC a little-PM give 

'Please give me some money.' 

b. *ton-ul cokum com cwuseyyo. 
money-AC a little a little give 
'?Give me some, some money.' 

There is no superficial difference in standard orthography between 
each pair of (a) and (b) in (13)-(15). However, when uttered, they 
are clearly distinguished by a pause, as we have already observed. 
The PM com is pronounced in a sequence with its preceding word, 
and a slight pause follows. On the other hand, a pause is placed 
before the adverb com. 

3. Pragmatic uses 

I claimed that the sense of the PM com is that the significance of 
what the com -suffixed phrase represents is minimized. In this 
section, I demonstrate how the pragmatic uses of the PM com are 
conversationally inferred from its sense, according to the Gricean CP. 
I also describe how a speaker exploits strategies for using the PM 
com depending upon various situations. 

I will describe various uses of the PM com within the 
framework of the theory of 'face' as proposed by Brown & Levinson 
(1978, 1987). They explain social relations between the speaker and 
the hearer, employing the concept 'face'. 'Face' refers to the want that 
one's actions be unimpeded by others (negative face) or that his 
wants be desirable to others (positive face). According to Brown & 
Levinson (1978, 1987), each one's face is vulnerable in interaction 
with others and people cooperate in maintaining or keeping their 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 151 

own face. However, in certain situations, the speaker purposely 
damages his hearer's face or his own face, depending on the social 
relations with her. 

If the speaker threatens the hearer's face, he knows that she 
will have negative feelings toward him and that she is not likely to 
cooperate with him in getting his goal accomplished. So the rational 
speaker tries to maintain the hearer's face in order to fulfill his goal. 
The rational hearer perceives the speaker's cooperative effort. In this 
sense, I believe, 'Be polite' is a corollary to the Gricean CP. The 
speaker's use of the PM com in his utterance is one of the strategies 
Koreans exploit to be polite toward the hearer.'* 

When the speaker makes a request or an order to the hearer, he 
knows that his request or order impedes her freedom to act by 
putting some pressure on her to do (or refraining from doing) a 
certain act that he wants her to do (or refrain from doing): that is, 
the speaker damages the hearer's face. The rational speaker tries to 
minimize the threat to the hearer's face in a normal situation. This is 
what the PM com does pragmatically. When the speaker uses the PM 
com after a phrase in a request sentence to show politeness to the 
hearer, he believes she knows its sense, and he expects her to reason 
from the sense in the following way: the sense of the PM com is that 
the significance of what a com -suffixed phrase represents is 
minimized; since a com -suffixed phrase is a part of his request, 
minimizing the information of that phrase implies that what he 
asked her to do will be little burden or pressure to her; thus he 
intended to redress a threat to her face by minimizing the burden or 
pressure she will perceive from his request. In this way, the speaker 
can save the hearer's face from being damaged by his request and 
show politeness to her. 

For instance, the speaker at a dinner table can utter either (2a) 
or (2b) repeated below, when he is asking the hearer to pass him the 
salt. 

(2) a. sokum-com cwuseyyo. 
salt-PM give 

'Please, pass me the salt.' 

b. sokum cwuseyyo. 
salt give 

'Pass me the salt.' 

The speaker of (2a) (believes the hearer) knows that he impedes her 
freedom of action by asking her, and not the others, to pass the salt. 
His intention of using the PM com after the phrase sokum 'salt' in 
(2a) is to diminish the face-threatening effect of his request by 
minimizing its burden on the hearer, which is implied from 



152 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

minimizing the amount of salt requested. When hearing (2a) 
containing the PM com, the hearer infers from its sense that the 
amount of salt the speaker needs is minimized so that he believes 
that his asking her to pass a small amount of salt will be only a small 
burden to her. Then she understands that he intended to alleviate a 
threat to her face to achieve his goal (getting the salt) by diminishing 
the burden she got from his request, and that his using the PM com 
is an effort to maintain her face. This is why (2a) sounds polite. On 
the other hand, utterance (2b) without the PM com sounds blunt 
because nothing mitigates the speaker's face-threatening act. 

When a face-threatening act such as a request is implicated 
according to the CP, the speaker may use the PM com to maintain the 
hearer's face. Take for instance the situation in (16) below, where A 
boarded a plane and found that his seat was already taken by B. 

(16) cali-ka pakkwuyessnundeyyo. 
seal-NM was-changed. 
'You took the wrong seat.' 

When B hears (16), she tries to figure out why a total stranger. A, 
uttered it to her, and how the utterance in the situation is relevant in 
the situation which A and B are engaged in. With the belief that A's 
utterance is goal-oriented, B interprets his speech act this way: 'He 
said that I took the wrong seat; 1 have no reason to suppose that he 
is not observing the CP; if I think that I am taking his seat, his 
utterance is consistent with my supposition; so he implies that he 
wants me to leave the seat for him.' In this way, a request is 
implicated from (16). This indirect speech act makes (16) polite. 

In this situation, using the PM com after the phrase cali-ka 'seat' 
as in (16'), the speaker intends to maintain B's face by minimizing 
the significance of what that phrase conveys, that is, B's 
transgression of taking the wrong seat. This is what makes (16') 
sound more polite than (16). 

(16') cali-ka-com pakkwuyessnundeyyo. 
seat-NM-PM was-changed. 
'You took the wrong seat.' 

I demonstrated that the use of PM com is intended to show the 
speaker's politeness when his utterance or his intention (goal) 
threatens the hearer's face. So the speaker may not use the PM com 
under the circumstances where he barely damages the hearer's face, 
for example, where he provides her with offers which are purely in 
her interest (17a), or where face-saving is not a concern any more to 
him in the situations of robbing, coercing or such (17b). 5 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 153 

(17) a. phyenhi ancuseyyo. 
comfortably sit-down 
'Make yourself at home.' 

b. ton naynoa. 
money take-out 
'Give me your money.' 

Saying (17a), the speaker is making an offer for the hearer to sit 
down. The offer is purely for the benefit of the hearer, and the threat 
to her face is very small. She understands that he is not damaging 
her face because the offer is clearly in her interest. Therefore, even 
though the speaker does not use the PM com in (17a), the utterance 
does not sound rude. When a mugger intends to take money away 
from a person by threatening him, (17b) can be used. In this 
situation, the mugger does not care about the hearer's face, and using 
the PM com is unnatural. 

Even if the speaker's offer is obviously in the hearer's interest, 
he also can use the PM com as seen in (18). 

(1 8) phyenhi-com ancuseyyo. 
comfortably-PM sit-down 
'Make yourself at home, please.' 

The utterances (17a) and (18), however, are used in different 
situations. When somebody visits my home, I can ask her to sit 
down, uttering (17a) but not (18), which is awkward in the situation. 
However, when the visitor has already had a seat but looks 
uncomfortable, then I can utter (18). Although my offer is for the 
hearer's benefit, I am threatening her face by inviting her to do what 
I want her to do. So I can use the PM com after the phrase phyenhi 
'comfortably' in (18) which represents the way in which I want the 
hearer to sit. Hearing (18), the hearer understands from its sense 
that I intended to save her face by implying that a little effort will 
be enough for her to sit in the way I want her to sit. 

By placing the PM com after more than one phrase in his 
request sentence, the speaker intends to multiply the minimization 
of the threat to the hearer's face because the burden or pressure of 
his request is multiply redressed through the minimization of what 
each com-suffixed phrase represents. Since excessive politeness 
means that the speaker humbles himself excessively, he is damaging 
his face by using more than one com with the politeness use. In a 
normal situation, however, the speaker tries to save his own face in 
interaction. So when the speaker uses the PM com more than once, 
the hearer assumes that multiple occurrences of com are relevant to 
his goal, and recognizes from its sense and the context that his goal is 



154 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

SO urgent that he may damage his own face himself. Take the 
examples in (19) for instance. 

(19) a. cepal-com sallye-com cwuseyyo-com. 

please-PM save-life-PM give-PM 
'Please, please, don't kill me, please.' 

b. i kes-com kkok-com hay cwuseyyo. 
this thing-PM surely-PM do give 
'Please, please, do this for me without fail.' 

The speaker can utter (19a) when a gangster tries to kill him. In this 
situation, the speaker's urgent goal is to save his life, and his desire 
to keep his face is not stronger than his goal. By using more than one 
com in his utterance, he expects a gangster to understand how 
urgently he wants to save his life. And a speaker can utter (19b) 
when his goal (getting the work done) needs to be urgently 
accomplished, and he believes that it can be done only by the hearer, 
or when he knows that she doesn't want to or won't accept his 
request at the moment of uttering. In situations like (19), the hearer 
guesses why the speaker uses more than one com, and understands 
that his desire or longing to get his goal fulfilled by the hearer is 
greater than his want to keep his face in the interaction. 

Being humble is another way of being polite. People have an 
interest in having their desire recognized by others. So self-conceit or 
self-praise of the speaker's doings, ability, talents and so on can 
damage the hearer's face since he gives an impression that her 
ability, talent or such is poorer than his. In order not to threaten the 
hearer's face, the speaker can use the PM com and expects her to 
understand from its sense that he is minimizing his ability, talents or 
such to show humility as seen in (20). 

(20) A: wuntong-ul cohahaseyyo? 

sports-AC like 

'Do you like sports?' 

B: ney, tennis-com chyeyo. 
yes tennis-PM play 
'Yes, I play tennis.' 

Even though B plays tennis well, he does not tell the truth because he 
knows that such self-praise can make him look presumptuous and 
that it can make the hearer feel that he believes that she can't play 
tennis as well as he. Instead, by using the PM com, he expects the 
hearer to infer that he is minimizing his ability in (playing) tennis. 

Likewise, if the speaker uses the PM com when he is talking 
about the hearer's ability, doings, talents and so on, it implies that he 
is minimizing her desire that her wants be recognized by others so 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 155 

that her face is threatened. Therefore, in such a context the speaker 
does not use the PM com as in (21). 

(21 ) tennis-com chiseyyo? 
tennis-PM play 
'Do you play tennis?' 

Using the PM com in (21) has the effect of depreciating the hearer's 
ability in (playing) tennis. That is, the speaker damages the hearer's 
face because he disregards her desire that her ability be recognized 
by others. Therefore, in a situation of asking whether the hearer can 
play tennis or not, the utterance (21') without the PM com sounds 
more polite than the corresponding one (21) with it. 

(21') tennis chiseyyo? 
tennis play 
'Do you play tennis?' 

In a way similar to the case of example (20), when he refuses 
the hearer's request indirectly, the speaker can use the PM com to 
redress a threat to her by minimizing the importance of the reason 
why he refuses her request. A person has a desire for his wants to be 
desirable to others. When such wants are considered undesirable by 
others, his face is damaged. So when the speaker refuses the hearer's 
request, her face is damaged. Furthermore, if he makes light of her 
request but thinks much of the reason why he refuses it, her face 
will be much more threatened. However, he can mitigate the threat 
to her face if he gives the impression that he does not regard the 
importance of her request. The PM com used in an indirect speech 
act of refusing the hearer's request has this function. It allows the 
hearer to recognize from its sense that the speaker intends to imply 
that he does not disregard her request because he is minimizing the 
importance of why he refuses it. For instance, Lee's friend A gave a 
call to ask him to give her a ride to Chicago tomorrow afternoon, but 
he refused the request with the utterance in example (22). 

(22) nayil ohwu-ey yaksok-i-com issnundeyyo. 

tomorrow afternoon-in appointment-NM-PM have 
'I have an appointment tomorrow afternoon.' 

By saying (22), Lee intends for B to figure out that he cannot give her 
a ride since he has something else to do tomorrow. The rom-suffixed 
phrase 'appointment' is in contrast with giving B a ride, so that it is 
significant for Lee's goal of refusing her request. Since refusing the 
request threatens B's face, Lee intends to redress the threat to her 
face by implying that he is minimizing the importance of his own 
appointment. Accordingly he can imply that he does not disregard B's 
request. B will recognize from the literal meaning of (22) that Lee is 



156 Studies in the Linguistic Sciences 2 1 :2 (Fall 1991) 

refusing her request, and will understand the politeness function of 
the PM com by inference from its sense. 

The PM com can be used in a declarative sentence like (23) 
when the speaker believes that his utterance is not as informative as 
is required for what he assumes his hearer wants to know. Using the 
PM com, he expects his hearer to understand that he is restricting all 
the information he believes she wants to know to merely what he is 
saying through the utterance containing it, and that he is mitigating 
the threat to the hearer's face which results from the fact that her 
desire to know all the information she wants to know has not been 
satisfied. 

(23) A: ecey eti kasesseyo? 

yesterday somewhere went 

yelepen cenhwahay-to an patusiteyyo. 

several times phone-although not answer 

'Did you go somewhere yesterday? I called you several 

times, but you didn't answer.' 

B: ney. Seoul-ey-com kasseyo. 
Yes to-PM went 

'Yes. I went to Seoul.' 

In the situation (23), A is asking why B didn't answer the phone 
yesterday and B answers that he went to Seoul. If B believes from 
the context that what A wants to know is not just that he went to 
Seoul, but the reason why he went to Seoul, and believes that he 
does not have to explain all of this to A at the moment of uttering, 
then he may threaten A's face with (23). Using the PM com, B 
intended to save A's face, which he believes is threatened by not 
giving her all the information he believes she wants. 

However, when the speaker believes that his utterance 
represents all the information he believes the hearer wants to know, 
or that she can figure out all the information she wants from his 
utterance, her knowledge of the world and the discourse up to then, 
using the PM com is stupid as illustrated in (24). In the situation, A 
visited his friend Insik's house to see him and learned from B, Insik's 
mother, that he is serving in the army now. 

(24) A: Insik-i cip-ey isseyo? 

NM house-at stay 

'Is Insik at home?' 

B: a. kwunday-ey kassta. 
army-to went 

b. #kwunday-ey-com kassta. 
army-to-PM went 

'He joined the army.' 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 157 

In Korea, all young men serve in the army for 30 months by law. All 
Koreans know this. So when Insik's mother answers A, she believes 
that he can infer why Insik joined the army from her utterance and 
his knowledge of the world, and that she does not have more to say 
to him at the moment. So Insik's mother does not need to use the PM 
com in the situation. On the contrary, using the PM com in this 
situation is inappropriate as seen in B's answer (b) in (24). 

Even when the subject of a speaker's utterance (sentence) is in 
the third person, the PM com can be used if the speaker believes that 
what he says can be a threat to the hearer's face directly or 
indirectly. Take for instance the situation in (25) below, where Inse's 
father, coming back home from work, is looking for his son who does 
not study hard, and his wife says that Inse went to his friend's. 

(25) Father: Inse-ka an poicyana. 
NM not is-seen 

'I can't see Inse. (= Where is Inse?)' 

Mother: chinkwu cip-ey-com nolle kasseyo. 
friend house-to-PM play went 
'He went to his friend's to play.' 

As should be clear by now, the PM com is used in the direct 
interaction between the speaker and the hearer in order to express 
politeness. However, in (25), Mother is informing Father of the fact 
that their son went to his friend's. It is not the speaker, but the third 
person Inse that performed the action of going to a friend's. So 
Mother appears to have no reason to show her politeness to the 
hearer by using the PM com. However, if Father told Inse not to go to 
his friend's but to study, and Mother knows this, Inse's having gone 
to his friend's threatens his father's face since he disobeyed his 
father. Mother believes that conveying the fact is a threat to Father's 
face, and by using the PM com she tries to minimize the face- 
threatening effect of her utterance and save his face. In addition, 
since Mother knows what Father told Inse, she feels a sense of 
responsibility for not stopping Inse from going to his friend's, and 
she believes that her allowing Inse to go to his friend's against 
Father's command has threatened Father's face. So she uses the PM 
com to mitigate the threat to his face. When Father hears the PM com 
in (25), he understands from its sense that Mother uses it to redress 
his threatened face which she believes he can figure out from the 
conventional meaning of her utterance and his knowledge of the 
world. In the situation in (25), Mother can use the PM com even 
when Father did not enjoin Inse from going to his friend's. Because 
she assumes that parents are responsible for their children so that 
they have to discuss together anything concerning their children, she 
believes that she threatened Father's face by disregarding his 



158 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

opinion on the matter of whether or not to allow Inse to go to his 
friend's. So she can use the PM com to show politeness in minimizing 
the threat to Father's face. He understands her intention of using the 
PM com as required, since he knows its sense and what she assumes. 

The speaker can use the PM com to insult the hearer by 
belittling her ability on purpose, in cases where she has been 
negligent in doing her responsibility, where she has behaved badly, 
or where she did not do what he had asked her to do. In these 
situations he does not have to maintain her face, since she herself 
threatened her own face by not doing what she should do. Using the 
PM com in these situations, the speaker expects the hearer to 
understand from its sense that he is threatening her face by 
minimizing her ability to do what a com -suffixed phrase represents 
with respect to an event he is describing. Using the PM com under 
these situations is an insult to the hearer, and makes the speaker's 
utterance sound ruder and more coercive than when it is not used. 

Take for instance example (26). The speaker can utter it when 
the hearer whom he lent money broke her promise several times to 
pay it back on the due dates. 

(26) ton-corn cwuseyyo. 
money-PM give 

'Pay back my money.' 

In this situation, the hearer threatened her own face by breaking a 
social convention (returning borrowed money on the due date). 
Because she did not keep her promise to return his money, the 
speaker has negative feelings toward her. Using the PM com after the 
phrase ton 'money' in this situation, the speaker implies that the 
amount of money the hearer can pay will be just a little even though 
she is expected to be able to pay more than a little. In this way, he is 
belittling her ability to pay the money and insulting her. He expects 
her to figure out that he is urging her to pay back the money soon 
because he is insulting toward her as a result of her failure to do it.^ 

Similarly, Korean parents and teachers often say (27) to the 
hearer, a student, who does not study but only goofs off. 

(27) kongpu-com hayla. 
study-PM do 
'Please, do study.' 

Studying is what a student has to do. The hearer, however, damages 
her own face by just goofing off. By using the PM com after kongpu 
'study' in (25), the speaker expects the hearer to understand from its 
sense that he intends that a little studying will be enough from her 
even though more than a little is expected. 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 159 

Utterances (28)-(29) can be used in the same situations as 
utterances (26)-(27) are. The only difference is that (28)-(29) do not 
contain the PM com. Because of the exploited use of the PM com 
discussed just before, (26)-(27) sound ruder and more coercive than 
(28)-(29). 

(28) ton cwuseyyo. 
money give 

'Pay back my money.' 

(29) kongpu hayla. 
study do 
'Study.' 

4. Interaction of the sense of the PM com and its location 

In this section, I will describe according to the CP how different 
locations of the PM com affects pragmatic interpretations of a 
sentence. In a discourse the speaker places the PM com after a 
phrase whose information is relevant to his goal at the moment in a 
way indicated by its sense, and the hearer's attention is drawn to the 
phrase. In 4.1, I will demonstrate how the location of the PM com is 
relevant to the speaker's goals, that is, how its sense interacts with 
its location to affect total pragmatic interpretations of a sentence. I 
claim in 4.2 that the 'focus' use of the PM com is a side effect of its 
pragmatic uses. 

4.1. Interaction of senses and locations of PMs 

The speaker can use the PM com to minimize the face 
threatening act or to belittle the hearer's ability. As described in 
section 3, those uses are inferred from the sense of the PM com 
which I claimed is that the significance of what a corn-suffixed 
phrase represented is minimized. I will show how the speaker places 
the PM com after different phrases in a sentence, depending on the 
situation. In this section, I will show that the speaker uses the PM 
com after a phrase whose information is relevant to his goals in a 
way indicated by its sense and that its sense interacts with its 
location to produce an enriched pragmatic interpretation of a 
sentence. 

For instance, in different contexts, Mrs. Lee started a discourse, 
uttering (30) and (31) to her husband who just came back home. 
Mrs. Lee and her husband were hesitating in sending their three- 
year-old son to a nursery school because of the monthly fee which 
would be a burden for them. Her husband suggested to her to run a 
baby-pool. 



160 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(30) Daewoong-i-com ywuawon-ey ponayyo. 
nursery school-to send 

talun yaytul-to ta ywuawon-ey taninteyyo. 
other kids-also all nursery school-to go-heard 

'Please, send DAEWOONG to a nursery school. All the other 
kids also go to a nursery school, I heard.' 

(31) Daewoong-i ywuawon-ey-com ponayyo. 

nursery school-to send 
phwulhal yaytul-i epseyo. na-to icey cichyesseyo. 
do-pool kids-NM are-not I-also now is tired 
'Please, send Daewoong TO A NURSERY SCHOOL. I couldn't 
find kids for a pool. I am tired of it.' 

Mrs. Lee's request sentences containing the PM com in (30) and (31) 
have the same literal meaning, but have the PM com in different 
places. By using the PM com in both situations, Mrs. Lee intended to 
imply that her face-threatening act will not be much of a burden. 

In the context of (30), informing her husband that all the kids 
she knows go to a nursery school, Mrs. Lee expects him to infer that 
she thinks that they could afford to send their son to a nursery 
school, too, which she wants. Her goal is to make her husband allow 
their son to go to a nursery school like other kids. So, placing the PM 
com after the phrase Daewoong, she is minimizing (or restricting) the 
persons to send to a nursery school just to Daewoong. Thus she 
intended that it is just Daewoong who they need to send to a nursery 
school, so that sending just Daewoong cannot be much of a burden. 

In the context of (31), Mrs. Lee intends to convince her husband 
that there is no choice but to send their son to a nursery school 
because she didn't find kids for a pool. So, placing the PM com after 
ywuawon-ey 'to a nursery school', she is restricting the places where 
to send Daewoong only to a nursery school, and expects her husband 
to understand that sending Daewoong only to a nursery school will 
not be a big burden to him. 

Other examples are (32) and (33) which the speaker can utter to 
a mechanic at an auto-repair shop. 

(32) i ke-comonul kochye cwuseyyo. 
this thing today repair give 
cenyek-ey Seoul-ey ka-ya hayyo. 
evening-in to go-have to 
'Please, fix THIS (CAR) today. 

I have to go to Seoul this evening.' 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 161 

(33) i ke onul-com kochye cwuseyyo. 
this thing today repair give 

cenyek-ey Seoul-ey ka-ya hayyo. 
evening-in to go-have to 

'Please, fix this (car) TODAY. 
I have to go to Seoul this evening.' 

The speaker can utter (32) after hearing the mechanic say that she 
has several cars to fix today. His goal in this situation is making the 
mechanic understand that what should be repaired today is his car 
(which is necessary for accomplishing his ultimate goal which is 
driving it to Seoul this evening). So placing the PM com after / ke 
'this (car)', the speaker intended to minimize the significance of what 
the mechanic should repair today, and he expects her to infer that 
fixing his car, not others, will not be a burden. The mechanic 
understands that what the speaker wants her to fix today is his car. 
On the other hand, the speaker can utter (33) when the mechanic 
tells him to pick up the car tomorrow. In this situation, his goal is 
making the mechanic repair his car today, not tomorrow (his 
ultimate goal is driving his car to Seoul this evening). So placing the 
PM com after onul 'today', the speaker intends the mechanic to figure 
out why he placed it there, from its sense and location, and the 
context. The mechanic understands that the time the speaker wants 
her to fix his car is today, not tomorrow or the other day, and that he 
intended to alleviate his face-threatening request by minimizing the 
significance of the time when she has to fix the car. 

4.2. Focus use 

In the previous section, I showed that the information of a PM 
com -suffixed phrase is relevant to the speaker's goal in a way 
indicated by the sense of the PM com; that is, a total pragmatic 
interpretation of a sentence is a function of the sense of the PM com 
and its location (and the literal meaning of the sentence, a context 
and so on). In this section, I demonstrate how the hearer's attention 
is drawn to a PM com -suffixed phrase, not any other phrases in a 
sentence, and that the focus use of the PM com is a side effect of its 
pragmatic uses. 

If different locations of the PM com do not affect pragmatic 
interpretations, sentences containing it in different positions would 
have the same pragmatic interpretation and they could be uttered in 
the same discourse context. For instance, in the wh-question and 
answer context in (34), the two utterances (34a) and (34b) 
containing the PM com could be used appropriately as an answer to 
A's question 'What shall I give to Yonghi?'. 



162 Studies in the Linguistic Sciences 21:2 (Fall 1 99 1 ) 

(34) A: Yonghi-hantey mues-ul cwulkka? 
to what-AC give 

'What shall I give to Yonghi?' 

R a. Yonghi-hantey ppang-com cwuseyyo. 
to bread give 

'Please, give Yonghi BREAD.' 

b. #Yonghi-hantey-com ppang-ul cwuseyyo. 
to bread-AC give 

'Please, give YONGHI bread.' 

Hov^'ever, only (34a) is appropriate as an answer in this situation. 
The only difference of (34a) and (34b) is the locations of the PM com; 
after ppang 'bread' in (34a) and after Yonghi-hantey 'to Yonghi' in 
(34b). Different locations of the PM com in (34a) and (34b) make a 
difference in their appropriateness. 

The focus use of the PM com as seen in example (34) is not a 
main use of the PM, but a side effect of its pragmatic uses which are 
inferred from its sense. For instance, when the speaker is making a 
request, he can use the PM com to minimize his face-threatening act. 
Using the PM com after a phrase allows the hearer to figure out from 
its sense that the speaker intended to imply that what that phrase 
represents is insignificant. However, the PM com can be placed after 
any constituent in a sentence. So when the speaker uses it after a 
specific phrase, the hearer guesses why he placed it after that 
phrase, not any other phrase in his utterance. She assumes that he 
did not choose the phrase for it without purpose and that minimizing 
the significance of what that phrase represents is relevant to his goal 
at the moment, because she believes that his speech act is goal- 
oriented. Then her attention is drawn to that phrase. The speaker 
exploits the hearer's belief about his conforming to the CP, and can 
use the PM com after any phrase whose information fits in with his 
goal at the moment in the way indicated by its sense. In this way the 
speaker draws attention to a corn-suffixed phrase. 

For example, the PM com in both (34a) and (34b) is used to 
minimize the burden implied by his request. However, it is placed 
after different phrases; after ppang 'bread' in (34a) and after Yonghi- 
hantey 'to Yonghi' (34b). In the context of (34), B's immediate goal is 
to give A the correct information which is 'bread'. So the rational 
speaker places the PM com after the phrase 'bread' and focuses 
attention on that phrase by seeking to minimize the amount of bread 
that is needed. When A hears (34a), she recognizes that using the PM 
after the phrase 'bread', not any other phrase, is consistent with the 
goal she assumes he has at the moment. This is why (34a) sounds 
appropriate in the context. However, (34b) is inappropriate as an 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 163 

answer to A's question, because the sense of the PM com applied to 
Yonghi is not consistent with B's goal. Utterance (34b) will be 
appropriate as an answer to the question 'To whom shall I give 
bread?" 

5. Conclusion 

I described according to the CP how the speaker uses the PM 
com and how the hearer interprets it. When the speaker uses the PM 
com in his utterance, the hearer believes that his use of it supports 
his goals she assumes he has, and she rationally reasons its uses 
which she believes are relevant to the goals in a way indicated by its 
sense. The speaker exploits the hearer's belief like this to achieve his 
goals by using the PM com. In section 3, I have shown that various 
pragmatic uses of the PM com are inferred according to the Gricean 
CP from its sense, which I claimed is that the significance of what a 
com -suffixed phrase represents is minimized. For the description of 
pragmatic uses of the PM com, I employed Brown & Levinson (1987, 
1987)'s concept 'face'. The speaker uses the PM com to show 
politeness to the hearer by minimizing the face-threatening act 
which his utterance constitutes, or to insult her by minimizing her 
ability. The hearer interprets the PM com in the way that the 
speaker intended because she knows its sense and assumes that he 
uses it in conformity with the CP. In every use of the PM com, I 
demonstrated how using the PM com contributes to getting the 
speaker's goal at the moment accomplished. 

I also showed that different locations of the PM com produce 
different pragmatic interpretations which are used in different 
situations; that is, the speaker places it after a phrase whose 
information is significant to his goal in the way indicated by its 
sense, and draws attention to the phrase. I claimed that the focus use 
of a PM is a side effect of its pragmatic uses which are inferred from 
its sense according to the CP. 

This study of the PM com in Korean will contribute to the field 
of teaching Korean as a foreign language. Korean has a number of 
pragmatic morphemes and sentence-final particles, which play 
important roles in a discourse to help the hearer to understand the 
speaker's intention (or goals). Koreans use and interpret them 
intuitively and subconsciously, but they cannot explain consistently 
to foreign language learners how they use and interpret them in 
different ways in different situations. Nevertheless, there has been 
no systematic and theoretic study about them up to now. As a Korean 
teacher, I have experienced difficulties in teaching students how to 
use and understand PMs in a discourse, and I found foreigners have 
a hard time learning them. 



164 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

I claimed that all uses of the PM com are conversationally 
derived from its sense according to the CP, and that the location of 
the PM com is relevant to the speaker's goal in a discourse. Believing 
that Korean learners are rational enough to make cooperative efforts 
to understand the speaker's goals in a discourse, a Korean instructor 
just let his students know the sense of the PM com and point out how 
its sense and location interact to contribute to the speaker's goals in 
different situations. Then they will be able to use and interpret the 
PM com appropriately in a discourse. 1 believe that the claim made 
in the study of the PM com can apply to the study of other PMs in 
Korean, which will also contribute to the field of teaching and 
learning Korean. 

NOTES 

1 Strictly, 'hearer' is different from 'addressee', in that even by- 
passers who happen to hear what a speaker is saying can be hearers, 
but not addressees. Nevertheless, 1 use the familiar term 'hearer' for 
'addressee' here. 

For the sake of expository convenience, third person, male, 
singular pronoun is used to refer to a speaker, while third person, 
female, singular pronoun refers to a hearer. 

2 My claim that the PM com is treated as a bound morpheme is 
different from the general view of the form com held by Korean 
linguists. Korean linguists consider the form com as a free morpheme 
irrespective of its uses, PM use or Adverb use, in light of the fact that 
it is spelled as an independent word in written materials, even 
though they intuitively know the distinctions of the PM com and the 
adverb com discussed in §2. 

3 The speaker can utter (8b) in the situation where the hearer 
cooked him the soup, which is too salty for him. In (8b), the adverb 
com also can show the speaker's politeness toward the hearer. 
However, its politeness comes from a lie because it is not truth- 
conditionally true that the soup is a little salty. 

'^ The social convention 'Be polite' is even more important in the 
Korean culture where deferring to superiors in age or social position, 
expressing humbleness, not imposing on the hearer and so on have 
been considered highly-valued virtues. It is conventionalized in the 
Korean language and every aspect of Korean society. The complicated 
honorific system in Korean is an example. 

5 An offer can threaten the hearer's face if it implies weakness 
on her part. Then the speaker can use the PM com. 



H. Lee: The pragmatics of the Korean pragmatic morpheme com 165 

6 Likewise, example (2a) can also be used in a situation where 
the speaker is upset and impatient because the hearer did not pass 
the salt he asked her to pass. In this case, by using the PM com, he 
intended to minimize or doubt her ability in passing the salt and to 
threaten her face. He expects her to infer that his intentional threat 
to her face resulted from her disregarding his desires so that he is 
urging her firmly to do what he wants her to do. 

REFERENCES 

Brown, p. & Levinson, S. 1978. Universals in language usage: 

Politeness phenomena. Questions and politeness: Strategies in 

social interaction, ed. by E. Goody, 56-311. Cambridge, England: 

Cambridge University Press. 
1987. Politeness. Cambridge, England: Cambridge University 

Press. 
Chang, S. I. 1982. Modern conversational Korean 1. Evanston, IL: 

Amerstra Corporation. 
CHO, C. H. 1982. A study of Korean pragmatics: Deixis and politeness. 

University of Hawaii Ph.D. dissertation in Linguistics. 
Dunn, A. 1990. The pragmatics of selected discourse markers in 

Swahili. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 
Green, Georgia M. 1982. Linguistics and the pragmatics of language 

use. Poetics 11.45-76. 
1989. Pragmatics and natural language understanding. Hillsdale, 

NJ: Lawrence Erlbaum Association. 
1990. The universality of Gricean interpretation. Proceedings of 

the 16th Annual Meeting of the Berkeley Linguistics Society, ed. 

by K. Hall, J. P. Koenig, M. Meacham, S. Reinman, & L. A. Sutton, 

411-428. Berkeley, CA: Berkeley Linguistics Society. 
GriCE, H. p. 1975. Logic and conversation. Syntax and Semantics 3: 

Speech Acts, ed. by P. Cole and J. Morgan, 41-58. New York: 

Academic Press. 
Lakoff, G. 1972. Hedges: A study in meaning criteria and the logic of 

fuzzy concepts. Papers from the 8th Regional Meeting of the 

Chicago Linguistic Society, ed. by P. M. Peranteau, J. N. Levi, & G. 

C. Phares, 183-228. Chicago: Chicago Linguistic Society. 
Lakoff. R. 1972. Language in context. Language 48.907-927. 
1973. The logic of politeness: or Minding your P's and Q's. Papers 

from the 9th Regional Meeting of the Chicago Linguistic Society, 

ed. by C. Corum, T. C. Smith-Stark. & A. Weiser. 292-305. 

Chicago: Chicago Linguistic Society. 
Lee, H. G. 1992. The pragmatics and syntax of Pragmatic Morphemes 

in Korean. University of Illinois at Urbana-Champaign Ph.D. 

dissertation in Linguistics. 



166 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

Morgan, J. L. 1978. Two types of convention in indirect speech acts. 

Syntax and Semantics 9: Pragmatics, ed. by P. Cole, 261-280. 

New York: Academic Press. 
PARK, Y. T. 1991. Spoken Korean Book 1. NJ: Hollym International 

Corporation. 
Prince, E. 1981. Toward a taxonomy of given-new information. 

Radical Pragmatics, ed. by P. Cole, 223-256. New York: Academic 

Press. 
SCHOURUP, L. 1985. Common discourse particles in English 

conversation: like, well, y'know. New York: Garland. 
SCHIFFRIN, D. 1987. Discourse markers. Cambridge, England: 

Cambridge University Press. 
UYENO, T. Y. 1971. A study of Japanese modality: A performative 

analysis of sentence particles. University of Michigan Ph.D. 

dissertation in Linguistics. 



Studies in the Linguistic Sciences 
Volume 21. Number 2, Fall 1991 

THE PRAGMATICS OF NEGATION IN KOREAN 

Virginia K. McClanahan 

This paper argues that a principle of least effort governs 
the choice between the short and long form of negation in 
Korean. It is also suggested that while the short form of 
negation may be unambiguously descriptive, the long form 
is not the preferred device for expressing metalinguistic 
negation. 

1. Introduction 

Previous studies of the two types of negative declarative 
sentences in Korean have been primarily concerned with syntactic 
and semantic analyses of two constructions: Short NEG in which the 
negative morpheme an(i) is placed before the verb and Long NEG in 
which the verb is nominalized using ci, followed by the negative 
morpheme an(i) and the verb hata "to do' (usually shortened to the 
negative auxiliary anh-ta). For the affirmative sentence in (1), both 
(2) with Short NEG and (3) with Long NEG are possible negations. 

(1) Mary- ka ca- ess- ta 

NOM sleep PAST DECL 
'Mary slept.' 

(2) Mary- ka an ca- ess- ta 

NEG 
'Mary did not sleep.' 

(3) Mary- ka ca- ci an ha- ess- ta 

NMZ do 

'Mary did not sleep.' 

In the first part of this paper I argue that choice of Short NEG or 
Long NEG is governed in part by the "least effort" principle 
(McCawley 1978).' In the second part of the paper I discuss Horn's 
(1985, 1989) claim that negation is pragmatically ambiguous 
between cases of descriptive negation and metalinguistic negation 
and support his suggestion that Short NEG in Korean might be viewed 
as unambiguously descriptive but argue against his suggestion that 
Long NEG is used metalinguistically. 



168 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

2. The principle of "least effort" 

We conversationally implicate not only by what we say but also 
by what we do not say. McCawley (1978) proposed that when we 
choose one utterance rather than others that could have been 
produced, a conversational implicature results by virtue of the 
choice. In other words, when a speaker uses an utterance which 
takes more effort than other utterances that might be used in a given 
situation, he is conversationally implicating something by choosing 
the more complex utterance. In support of this least effort analysis, 
McCawley cites Householder's (1971) observation that while the 
adjective pale can readily be combined with a great many color 
words (pale green, pale blue), it does not readily combine with others 
(Ipale red). Householder suggests that pale red sounds strange 
because there is a common English word for 'pale red', namely pink. 
Pale blue and pale green do not sound odd because English does not 
have specific lexical items to designate these shades. McCawley notes 
that pale red requires more effort than pink due both to its greater 
complexity in surface structure and to its containing more 
phonological material. ^ Therefore, unless there is some particular 
reason for the extra effort involved in saying pale red, pink is 
preferred. When a speaker uses pale red instead of pink, he 
conversationally implicates something, for example, that the color in 
question is not pink but rather is some other shade of red, a shade 
which is paler than what might normally be considered red but not 
so pale as to be considered pink. 

This least effort analysis can also offer insights into the 
distribution of the Short and Long negative forms in Korean. Short 
NEG is less complex in surface structure and contains less 
phonological material than Long NEG. According to McCawley's least 
effort concept then, we can expect that the shorter, less complex 
Short NEG will be used by speakers unless there is some reason not to 
use that form. Furthermore, if the longer, more complex Long NEG is 
used rather than Short NEG, we should expect that the speaker is 
conversationally implicating something by virtue of the fact that he 
is expending the extra effort to produce the more complex utterance. 

If this is in fact the case, one obvious question is what might be 
conversationally implicated by the use of Long NEG rather than Short 
NEG. I argue that the fact that Short NEG is the negative construction 
of least effort is the reason it is viewed appropriate in casual, 
informal conversational situations, and that the relatively greater 
length and structural complexity of Long NEG, a construction 
demanding extra effort on the part of the speaker, is the reason it is 
considered appropriate in more formal conversational situations as 



McClanahan: The pragmatics of negation in Korean 169 

well as in written forms. In other words, Long NEG is used to imply 
formality. 

Since Short NEG takes less effort than Long NEG, the principle 
predicts that Long NEG would not be used unless there were some 
reason for expending the extra effort. Long NEG is generally 
considered to be a more formal, more deferential form, and I contend 
that this is the case due to the fact that it is the form requiring extra 
effort. In other words, use of the longer, more complex Long Form in 
more formal, more deferential, more polite speech comes about as a 
result of speakers operating on the principle of least effort. When the 
use of Long NEG is not necessitated by the grammar, what is most 
often being conversationally implicated when a speaker uses Long 
NEG is formality and/or deference. 3 

McCawley's least effort principle offers a straightforward 
explanation for a speaker's preference for Short NEG in casual 
conversational settings and for the fact that Long NEG is used almost 
exclusively in written forms and formal speeches. According to the 
principle of least effort, unless there is some reason to use a more 
complex linguistic form when there is a simpler form which will 
express what the speaker wants to express, the speaker will choose 
to use the simpler form. Since the Short NEG construction is shorter in 
length and syntactically simpler than the Long NEG construction. 
Short NEG involves less effort than the longer, more complex Long NEG 
construction. Therefore, unless there is some special reason for using 
the longer, more complex structure, a speaker would choose to use 
the form involving the least effort. I claim, therefore, that adherence 
to the principle of least effort is the reason that Short NEG is a more 
casual form than Long NEG and used by speakers in informal speech 
situations. 

3. Metalinguistic negation and pragmatic ambiguity 

Horn (1985, 1989) takes the position that negation "...must be 
taken as pragmatically ambiguous, with marked negation as an 
extended metalinguistic use of a basically truth-functional operator" 
(1985:121). Rejecting theories which view negation as semantically 
ambiguous or invariably truth functional, Horn argues that external, 
presupposition-canceling negation is part of a wider phenomenon of 
negation used to convey a speaker's unwillingness to assert a given 
proposition in a given way, i.e., a speaker's use of negation to object 
to the content or form of an utterance rather than to the proposition 
contained in the utterance. A speaker might object on any number of 
grounds including phonetic, morphological, syntactic, semantic or 
pragmatic, as well as stylistic or implicational. 

Before going any further, I will exemplify the difference 
between the two types of negation discussed by Horn, descriptive 



170 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

and metalinguistic. Internal negation, called descriptive by Horn, is 
used by a speaker to deny a proposition, in other words, it reverses 
the truth value of an affirmative proposition, as illustrated in (4a) 
with its negative counterpart in (4b).'* 

(4) a. Jane is a student, 
b. Jane is not a student. 

If (4a) is true, then (4b) must be false. Put simply, descriptive or 
internal negation is everyday, garden-variety proposition-denying 
negation. 

Speakers, however, do not limit their usage of negative 
constructions to simply deny propositions, as illustrated in the now 
classic example in (5). 

(5) The King of France is not bald (because there is no King 
of France). 

If we take the above example without the parenthetical continuation 
as internal or descriptive negation, we must judge it to be either 
false or to lack truth value. Given, however, that France is a republic 
and does not have a king, the example in (5) is true in a sense since 
that which does not exist cannot logically be bald. In (5) it is not only 
the proposition The King of France is bald that is canceled but also 
the presupposition that such a person as the King of France exists. 
This type of presupposition-denying or external negation is 
considered by Horn to be an instance of metalinguistic negation. 

Other examples of metalinguistic negation include instances of 
"more than" negation of scalar predicates in which the upper- 
bounding implicature is removed, as exemplified in (6b). 

(6) a. Steve is smart. 

b. Steve is not smart, he's brilliant. 

Examples of negation used to object to previous assertions on lexical, 
phonetic, and stylistic grounds are given in (7), (8), and (9) 
respectively. 

(7) (The baby bited the dentist's finger.) 

No, the baby didn't bited the dentist's finger, she bit the 
dentist's finger. 

(8) (She was sexually harrassed.) 

She was not sexually harrassed; she was sexually harrassed. 

(9) (The old coot kicked the oxygen habit.) 

The old coot did not kick the oxygen habit; your 
grandfather passed away. 



McClanahan: The pragmatics of negation in Korean 171 

Since these non-proposition-denying uses of negation do not alter 
truth conditions, Horn argues persuasively that they cannot be 
semantically analyzed in terms of truth but rather should be 
analyzed in terms of a speaker's unwillingness to assert something in 
a given way. 

Horn (1985) speculates that Korean is a language in which there 
is an unambiguous distinction between descriptive and 
metalinguistic negation in the sense that Short NEC can be used only 
descriptively which Long NEG may be used either descriptively or 
metalinguistically. However, such a distinction does not adequately 
describe the facts in Korean. In the following discussion of data, it is 
shown that neither Short NEG nor Long NEG is normally used for 
metalinguistic negation. Instead a periphrastic construction is used to 
express metalinguistic negation. In addition to Short and Long 
negation, there is a cleft construction in Korean which is used to 
negate declarative sentences. Examples are given in (10) and (11). 

(10) Na- nun kyohoy- ey ka- n kes- i ani- la 

I TOP church to go MOD COMP NOM NEG but 

hakkyo- ey ka- ass- ta 

school to go PAST DBOL 

'I didn't go to church, but I went to school.' 

(11) kongpu- hal ttay ttetul- nun key ani- ta 
study do lime make noise MOD COMP NEG DBC 
'You shouldn't make noise when studying.' 

(key is the contraction of kes and the nominative marker /.) 

Horn cites the sentence in (12) as an example of a metalinguistic 
scalar predicate negation in which the upper bounding implicature is 
removed. 

(12) Max doesn't have three children — he has four. 

In this example the proposition that Max has three children is not 
negated, but rather the implicature that he has ONLY three children is 
negated. 

Song (1982) presents a similar example in which Short NEG may 
be used for this type of metalinguistic negation. 

(13) John-i sakwa-lul twu kay ani mek-ess-ta 

-NOM apple-ACC two piece not ate-PAST-DECL 
'John didn't eat two apples.' 

Song suggests that the sentence in (13) could be used to assert 
simply that John did not eat two apples, i.e., he ate less than two 
apples, in which case it would be descriptive negation. However, Song 
also points out that it can be used to assert that John ate more than 
two apples, as is shown in the expanded sentence in (14). 



172 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(14) John-i sakwa-lul twu kay an mek-ko yele kay mek- 

eat-and several 
ess-ta 
'John didn't eat two apples, but ate several.' 

The reading in (14) is an example of the same type of upper- 
bounding-implicature-removing metalinguistic negation as that in 
(12) and Short NEG is used. These data indicate that Short NEG in 
Korean is not unambiguously descriptive as Horn suggests. However, 
Song's example when considered in conjunction with additional 
Korean examples of the type of negation which Horn considers to be 
metalinguistic suggests that while Short NEG in Korean may not be 
ruled out for metalinguistic usage in an absolute sense, it is in fact 
extremely awkward when used metalinguistically. While no native 
speaker of Korean judged (14) ungrammatical, many thought it to be 
awkward. In order to try to get as accurate a reading as possible of 
how metalinguistic negation is actually expressed in Korean, I asked 
native speakers to indicate the most natural way they would express 
certain external negations and then, after getting their responses, 
asked about the appropriateness of alternative constructions. When I 
elicited scalar predicate data, the results were consistent: (1) neither 
the Short or Long negative construction was the first choice, but 
rather the key ani-ta cleft construction was chosen first, (2) the Long 
NEG construction was preferred over the Short NEG construction, and 
(3) the Short NEG construction was judged awkward. The following 
examples illustrate these findings. 

(15) a. Mary-nun yeyppun-key ani- ko alumtap-ta 

TOP pretty CMP NEG but beautiful 

'Mary isn't pretty, she's beautiful.' 

b. # Mary-nun yeyppu- ci anh- ko alumtap-ta 

TOP pretty NOM NEG but beautiful 
'Mary isn't pretty, she's beautiful.' 

c. ## Mary-nun an yeyppu- ko alumtap-ta 

TOP NEG pretty but beautiful 
'Mary isn't pretty, she's beautiful.' 

(16) a. onul nalssi- nun ttattus-han- key ani- ko tep-ta 

today weather TCF warm CMP NEG but hot 

'It (the weather) isn't warm today, it's hot.' 

b. # onul nalssi- nun ttattus-ha- ci anh- ko tep-ta 

today weather TCP warm NMZ NEG but hot 

'It isn't warm today, it's hot.' 

c. ## onul nalssi- nun an ttattus-ha ko tep-ta 

today weather TOP NEG warm but hot 

'It isn't warm today, it's hot.' 



McClanahan: The pragmatics of negation in Korean 173 

In (15) and (16) an embedded sentence construction followed 
by the negative linking verb ani-ta was preferred over the Long NEC 
construction, and the Short NEG construction was judged very 
awkward. 

Since both (15) and (16) contain description verbs and at least 
some speakers prefer to negate description verbs with Long NEG 
rather than Short NEG, it is possible that judgements could be affected 
by this factor. To eliminate such possible interference, the verb in 
examples (17) and (18) is a monosyllabic action verb. Song's earlier 
example is repeated and expanded in (18). The judgements remained 
consistent with those in (15) and (16). 

,(17) a. ne- nun kwaca- lul myech kay- man mek- 

you IQP cookie ACC a few counter only cat- 
un key ani- ko ta mek-ess-ta 
ADJ COMP NEG but all ate 

'You didn't eat some of the cookies, you ate all of 
them.' 

b. # ne- nun kwaca- lul myech kay- man mek- 

ci anh- ko ta mek-ess-ta 

NMZ NEG but all ate 

'You didn't eat some of the cookies, you ate all of 

them.' 

c. ## ne- nun kwaca- lul myech kay- man an 

NEG 
mek-ko ta mek-ess-ta 
'You didn't eat some of the cookies, you ate all of 
them.' 

(18) a. John-i sakwa-lul twu kay mek- un key 

apple two pieces eat ADJ COMI' 

ani- ko yele kay mek-ess-ta 

several pieces ate 
'John didn't eat two apples but several.' 

b. # John-i sakwa-lul twu kay mek- ci anh- ko 

NMZ NHG but 
yele kay mek-ess-ta 
'John didn't eat two apples but several.' 

c. ## John-i sakwa-lul twu kay ani mek-ko yele kay 

NHG 
mek-ess-ta 
'John didn't eat two apples but several." 

Now, comparing these above scalar predicate negations with "more 
than" readings to their corresponding "less than" readings, the results 
are quite different. For the less than readings, an embedded sentence 



174 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

with key ani-ta was not used at all, and while Long NEC was the first 
response given, all speakers indicated that Short NEG was just as 
acceptable. 

(19) a. Mary- nun yeyppu- ci anh-ta 

TOP pretty NMZ 

'Mary is not pretty.' 
b. Mary- nun an yeyppu-ta 
NEG 
'Mary is not pretty.' 

(20) a. onul- un tep-ci anh-ta 

today TCP hot NOM NEG 
'It isn't hot today.' 

b. onul- u n an tep-ta 
NEG 
'It isn't hot today.' 

(21) a. ne- nun kwaca- lul mek- ci anh-ass-ta. 

you TOP cookies AOC eat NMZ NEG 
'You didn't eat the cookies.' 
b. ne- nun kwaca- lul an mek-ess-ta. 
NEG 
'You didn't eat the cookies.' 

Looking now at some nonscalar metalinguistic negations, the 
following examples contrast plain forms of verbs with honorific 
forms. All verbs and some other lexical items in Korean have both 
plain and honorific forms. In the following examples the subject 
nouns kyoswunim 'professor' and moksanim 'minister' refer to 
persons of high social status. To use the plain form of the verbs, 
cwuk-ta 'die' and mek-ta 'eat', with such subject nouns would be 
extremely insulting to the subject referents. If a child or an 
unknowing foreigner used the inappropriate verbs, she might be 
corrected as in (22) and (23). 

(22) a. kyoswunim-un cwuk- un key ani- ko 

professor die (plain) TOP COMP NEG but 

tolaka-syessta 

passed away (HON) 

'The professor did not die, he passed away. '5 

b. # kyoswunim-un cwuk- ci anh- ko tolaka-syessta 

NMZ NEG 
'The professor did not die, he passed away.' 

c. ## kyoswunim-un an cwuk-ko tolaka-syessta 

NEG 
'The professor did not die, he passed away.' 



McClanahan: The pragmatics of negation in Korean 175 

(23) a. moksanim-un cenyek-ul mek- un key 

minister dinner eat (plain) ADJ COMP 

ani-ko capswu-syessta 

NEC ate (HON) 

'The minister didn't eat dinner, he ate dinner.'^ 

b. # moksanim-un cenyek-un mek- ci anh-ko 

capswu-syessta 

NMZ NEC 
"The minister didn't eat dinner, he ate dinner.' 

c. ## moksanim-un cenyek-un an mek-ko capswu- 

syessta 

NBG 
'The minister didn't eat dinner, he ate dinner.' 

Again, native speakers responded first with the (a) utterances using 
the key ani-ta construction. The Short NEG in the (c) sentences was 
judged extremely awkward and the Long NEG, shown in the (b) 
sentences, was considered preferable to the Short NEG construction 
but not as appropriate as the key ani-ta construction. Based on this 
data, Korean appears to have a periphrastic negative form especially 
suited for metalinguistic negation. This is not to say, however, that 
the periphrastic construction is limited to metalinguistic usage; it 
may be used either descriptively or metalinguistically. 

The existence of a form which if especially appropriate for 
expressing metalinguistic negation is not unique to Korean. Horn 
(1989) includes Japanese data from McGloin (1982) which indicate 
that a periphrastic form must be used in order to negate the upper 
bounding implicature associated with scalar predicates, as shown in 
(24). 

(24) a. Atsui dokoroka nietagit-te i-ru yo. 

hot far from boiling be-PRES 

'It's far from being hot; it's boiling.' 

b. Atsui nante yuu monja na-i. Nietagit-te i-ru yo 
say 
'It's not something you can call hot. It's boiling.' 

Horn further notes that the Japanese construction wake de wa 
nai is described by Kato (1985) as being specialized for 
metalinguistic negation. In (25) Kato demonstrates that an adverb 
which cannot appear in negative sentences (either in or out of the 
scope of negation) can occur within the scope of wake de wa nai. 

(25) a. Kuruma ga totsuzen {tomat-ta/ *tomar-anakat-ta}. 

car suddenly stop-PAST stop-NEG-PAST 

'Our car suddenly {stopped/*didn't stop}.' 



176 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

b. Kuruma ga totsuzen tomat-ta wake de wa na-i. 
'It's not that our car stopped suddenly.' [It stopped 
gently, etc.] 

4. Conclusions 

To sum up, Korean and Japanese exhibit a stricter division of 
labor in the use of negative constructions than English does. While in 
English we sometimes employ it is not the case that... or it is not true 
that... to express a metalinguistic type of negation, both of these 
constructions have a rather formal, stilted sound. English speakers 
are, therefore, inclined to simply use what is normally a descriptive 
negation to express a metalinguistic negation. For instance, (26a) 
sounds more natural than (26b). 

(26) a. He's not (just) smart. He's brilliant. 

b. It is not the case that he's just smart. He's brilliant. 

In Korean, and in Japanese as well, however, periphrastic 
constructions which involve greater effort on the part of the speaker 
are used for metalinguistic negations, and the comparatively lesser- 
effort constructions are reserved for descriptive negation. The 
periphrastic constructions, however, are not limited to metalinguistic 
usage but may be used either descriptively or metalinguistically. 
Thus, the Korean data reinforces the general pattern discerned by 
Horn (1989) that "...no language contains two negative operators 
corresponding exactly to descriptive and marked [metalinguistic] 
negation" (442). 

While the Long NEG ci anh-ta construction is at least somewhat 
more acceptable for metalinguistic negation than the Short NEG an 
construction, neither of these is the negative construction preferred 
by speakers for metalinguistic negation. Instead, the periphrastic key 
ani-ta is the negative construction of choice. 

In the first part of this paper, the use of Short NEG in informal 
speech situations was analyzed as an instance of McCawley's least 
effort principle, and I now suggest that the preference for a more 
syntactically complex construction for metalinguistic negation than 
that used for descriptive negation is an example of the other side of 
the least effort coin. Of the three negative constructions discussed in 
this paper. Short NEG requires the least effort and the periphrastic 
construction the most effort, both phonologically and syntactically. 
Since natural language negation is overwhelmingly descriptive, it is 
logical that the lesser effort negative forms are appropriate for 
descriptive negation and that the negative construction requiring 
greater effort is more appropriate for the more pragmatically 
complex metalinguistic negation. The fact the Long NEG is more 
acceptable than Short NEG for metalinguistic negation but less 



McClanahan: The pragmatics of negation in Korean 177 

acceptable than the periphrastic construction is also predictable from 
the principle of least effort since the Long NHO form occupies the 
middle slot as far as the amount of effort required. 

NOTES 

1 Phonological and morphological restrictions which necessitate 
the use of Long NEC with certain verbs are not discussed in this 
paper. 

2 Syntactic and phonological complexity are not the only factors 
which McCawley takes into account, but they are the only ones 
relevant to negation in Korean. 

3 Though the notions of formality, deference, and politeness are 
interrelated, they are not interchangeable. I consider deference to 
imply formality but not necessarily vice versa. If a speaker is 
deferential to an addressee, the deference will be reflected in more 
formal speech. Formality in speech, therefore, may be a reflection of 
deference. On the other hand, formality in speech may be used by a 
speaker to distance himself from the listener(s). For example, if an 
employer uses a formal style of speech to an employee, it would 
more likely be a reflection of the difference in status between them 
than deference. Politeness seems to be a feature of both deference 
and formality. 

'^ This is not to imply that all negative propositions are uttered 
in response to or in the context of a corresponding affirmative 
proposition. 

5 In this context cwukta might be more appropriately translated 
with an idiom such as kicked the bucket' or bought the farm'. 

^ In this context mekta might be more appropriately translated 
'gobble' or 'pig out' or 'chow down'. 

REFERENCES 

Horn, L. R. 1985. Metalinguistic negation and pragmatic ambiguity. 

Language 61:1.121-174. 
1989. A natural history of negation. Chicago: The University of 

Chicago Press. 
HOUSFHOI.DKR, F. W. 1971. Linguistic speculations. London and New 

York: Cambridge University Press. 
Kato, Y. 1985. Negative sentences in Japanese. Sophia Linguistica 19. 

Tokyo: Institute for International Communication. Sophia 

University. 



178 Studies in the Linguistic Sciences 2 1 :2 (Fall 199 1) 

MCCawley, J. D. 1978. Conversational implicature and the lexicon. 

Syntax and Semantics 9: Pragmatics, ed. by P. Cole. New York: 

Academic Press. 
McGloin, N. H. 1982. Negation in a verb-final language. Unpublished 

Ms. University of Wisconsin-Madison. 



Studies in the Linguistic Sciences 

Volume 21. Number 2. Fall 1991 



INFLECTIONAL STRUCTURES IN KOREAN AND HEADEDNESS* 

James Hye Suk Yoon 

In this paper, I critically review the arguments presented 
in Sells (1991) that inflectional structures in Korean are 
lexically formed and left-headed. I show that the 
arguments for left-headedness lose force upon scrutiny and 
conclude the paper by rehearsing well-known arguments 
for the syntactic independence of i-affixes. 

1. Introduction 

Sells (1991) presents a number of morphosyntactic arguments 
for a LEXICAL analysis of Korean I(NFLECTIONAL)-STRUCTURESi against 
the backdrop of much current research that takes i-structures in 
these languages to be formed in the syntax (cf. Ahn & Yoon 1989, 
Yoon & Yoon 1990, Whitman 1989, etc.) Sells' attempt is worthy of 
note, since most syntactic analyses seem driven by an implicit desire 
to make the clausal structure of Korean parallel to those posited for 
English and French by Chomsky (1988) and Pollock (1988). 
Characteristically absent from the majority of syntactic analyses are 
robuts morphosyntactic ARGUMENTS that i-affixes must be heads in 
Korean syntax. 2 

The major component of Sells' analysis centers on the claim that 
i-structures are LEFT- HEADED. While the claim of L-headedness is 
logically independent of the issue of where i-structures are formed, ^ 
if i-structures are L-headed AND formed in the syntax, they would 
systematically violate the generalization that Korean syntax'* is 
otherwise head-final, weakening the syntactic position considerably. 

It is my goal in this paper to examine two contrasting claims 
about the LOCUS and HEADEDNESS of i-structures — one that claims that 
i-structures are LEFT-HEADED and formed in the LEXICON, and the other 
which claims that i-structures are formed in the SYNTAX and are 
RIGHT-HEADED. We argue that the weight of the evidence points in 
favor of the latter. 

2. Two analyses of Korean I-structures 

2.1. A lexical analysis 

According to Sells, the proper analysis of an inflected verbal 
form such as mek-ess-ta (eat-Past-Decl) is as in example (1). In this 



180 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

structure, only VO, and not any proper subpart of it, is visible to rules 
of syntax. Thus, in the words of DiSciullo & Williams (1987), it is a 
syntactic atom. In addition. Sells makes the claim that an inflected V 
is consistently left-headed, since categorial features are determined 
by the left-hand member at each level of structure. 

(1) ... VP5 

/ \ 
VO 

/ \ 
V-i af 
/ \ I 
V-2 af -ta 
I I 

V-3 -ess 
I 
mek- 

2.2. A syntactic analysis: I-affixes as phrasal affixes 

In this paper, I defend a syntactic analysis which specifically 
claims that a form like (1) consists of not one, but three distinct 
syntactic atoms, the Verb Stem (mek-). Tense {-ess) and Mood (-ta) 
morphemes, each heading their own projections (VP, TP, and MP, 
respectively) in syntax. The Tense and Mood affixes are therefore 
"phrasal affixes", since they are affixes subcategorizing for a phrase 
syntactically, even though phonologically they appear as affixes on 
verb stems. In the proposed structure, TP dominates VP and MP 
dominates TP. This is motivated both by logical 'scope' considerations 
and generalizations concerning affix ordering.^ 

(2) MP 
/ \ 

M' 

/ \ 

TP M 

/ \ I 

T' -ta 
/ \ 
VP T 
/ \ I 
V -ess 
/ \ 
V 
I 
mek- 



Yoon: Inflectional structures in Korean and headedness 181 

This type of analysis requires a mechanism to derive the actual 
inflected form of the verb. This could be achieved in either of two 
ways. In a derivational alternative, head movement raises V to T and 
this complex in turn raises to M. In a non-derivational, Autolexical- 
like alternative, (2) is the syntactic representation associated with 
the string mek-ess-ta. To this a morphological representation is 
associated in accordance with general principles. The Tense and 
Mood formatives, being clitic-like, are attached to the periphery of 
constituents they combine with syntactically. In either approach, the 
morphological representation would be as follows.^ 

(3) M 

/ \ 
T M 
/ \ I 
V T -ta 
I I 
mek- -ess 

In sum, in the syntactic analysis, inflectional structures are 
HEAD-FINAL, both syntactically (cf. 2) and morphologically (cf. 3), and 
the morphological and syntactic analyses of a string like mek-ess-ta 
are different and required independently. 

We turn now to arguments Sells offers for his analysis and show 
that they are either untenable or else turn out to be neutral with 
respect to the choice between the lexical and syntactic analyses. 

3. Arguments for a lexical account of inflection 

3.1. Phonological arguments 

The first class of arguments that Sells gives for the lexical view 
are phonological. The gist of these arguments is that phonologically, 
i-affixes behave clearly as AFFIXKS, rather than as Cl.ITICS or 
INDEPHNDKNT WORDS - i.e., there is a closer phonological cohesion 
between i-affixes and stems/roots than between clitics and their 
hosts or between two independent phonological words. They are 
lexically attached in this sense. 

In presenting this type of argument. Sells is relying on the "null 
hypothesis" - i.e, it is better to have an analysis in which the 
characterizations of "word" is uniform phonologically, 
morphologically, and syntactically. However, the independence of 
what is phonologically a word (P-word) and what is morphologically 
a word (M-word) is well-documented. Likewise, what is niorpho- 
phonologically determined to be lexical (i.e., M/P-word) need not 
necessarily be a syntactic word (syntactic atom). Besides, even if the 



182 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

null hypothesis turned out to be ultimately correct, this is exactly 
what Sells is trying to defend, and therefore arguments presupposing 
its correctness cannot be used to argue FOR it. Therefore, we do not 
find the phonological arguments terribly bothersome. We turn now 
to more robust morphosyntactic arguments offered by Sells. 

3.2. Non-local selection 

Auxiliary verbs in Korean select for the form ("COMP", a la Sells) 
of main verbs. For example, the negative auxiliary anhta constrains 
the main V it combines with to be in the -ci COMP form, as shown by 
the contrast between (4a) and (4b) below. However, certain kinds of 
suffixes can intervene between the -ci suffix and the selecting 
auxiliary, breaking the LOCALITY of selection (cf. 4c). 

(4) a. cek-ci anhta 

small-COMP not 

b. *cek-key anhta 
small-COMP not 

c. cek-ci-man-un anhta 
small-COMP-only-TOP-not 

Sells reasons that such a state of affairs will prove problematic for 
the syntactic view which holds that i-affixes are heads, since 
selection would have to "see" through several layers of affixes, in 
violation of selectional adjacency which is common in morphology 
and syntax. 

(5) VP 

/ \ 
Af3 V 
/ \ I 
Af2 afanh- 
/ \ I 
Afl af -un 
/ \ I 
V af -man 
I I 
cek- -ci 

However, if V is taken to be the head, selection can remain local, as 
shown in (6), which is adapted from Sells (1991). The requirement 
that the sister of the negative auxiliary be in [FORM:CI] is inherited 
through the sequence of head nodes to VO, as desired. 



Yoon: Inflectional structures in Korean and headedness 183 

(6) V as Head: VP 

/ \ 
/ \ 

/ \ 

/ \ 

V0[TYP:V-MOD, FORM:CI] V[FORM:CI] 
/ \ I 

V-i[TYP:V-MOD,FORM:CI] Af anhta 

/ \ I 

V-i[TYP:V-MOD,FORM:CI] Af -un[TYP:V-MOD] 

/ \ I 

V-2[TYP:N0] Af -man[TYP:V-MOD] 

I I 

V-3[TYP:N0] -ci[TYP:V-MOD, FORM:CI] 
I 
cek- 

It is not difficult to come up with an alternative syntactic 
analysis that does not countenance problems of non-local selection. 
One simply has to make the assumption (Grimshaw 1991) that the 
affixes that intervene (the so-called Delimiters) have no effect on 
selection because they are CATEGORY-NEUTRAL AFFIXES/HEADS, allowing 
categorial and selectional information on the non-head member to 
percolate through (cf. 7). 

(7) Affixes as Heads: VP8 

/ \ 

/ \ 

/ \ 

VP[+V, -N. CI] V[SUBCAT:<-i-V, -N, FORM:CI>] 
/ \ I 

/ \ anh- 

VP[+V, -N, CI] DEL[OV, ON, FORM:0] 
/ \ I 

/ \ -un 

VP[+V, -N, CI] DEL[OV, ON. FORM:0] 

/ \ I 

/ \ -man 

VP COMP[+V, -N, CI] 

/ \ I 

...cek -ci 

There is good evidence for taking the delimiters to be 
category/selection-neutral. As is well-known. they suffix 
indiscriminately to a variety of categories. 



184 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

(8) a. mek-e-nun 'eat-Comp-Top' (affixation to V-e) 

b. mek-key-nun 'eat-Comp-Top' (affixation to V-key) 

c. John-un 'John-Top' (affixation to N) 

d. John-eykey-nun 'John-Dat-Top' (affixation to N-eykey) 

e. ppalli-nun 'quickly-Top' (affixation to Adv) 

We have just demonstrated that a syntactic right-headed 
analysis can cope with apparent problems of non-local selection. We 
now like to point out some problems with Sells' claim that i- 
structures are uniformly L-headed in Korean. A problem for this 
claim comes from i-suffixes OTHER THAN THE NEUTRAL ONES, as these 
crucially determine the COMBINATORIC (distributional) potential of a 
given root/stem. 

To see this, we need some background in the feature system 
that Sells uses. Sells uses two different kinds of features in his 
account. The feature TYP, whose value ranges over {NO, V-MOD, N- 
MOD} is a COMBINATORIC feature, while the binary Boolean features V 
and N are CATEGORIAL features. We need not discuss the latter since 
its usage is familiar. Let us briefly discuss the function of the TYP 
feature. 

The TYP feature determines the syntactic distribution of a given 
item, in the following way. Constituents marked TYP:V-MOD combine 
with verbal categories, while TYP:N-MOD constituents combine with 
nominal categories. Stems/Roots, since they are by themselves 
incapable of occurring as sisters to N or V constituents, are 
categorized as TYP:NO. 

Now, it turns out to be the case that it is the i-affixes which 
crucially determine the COMBINATORIC POSSIBILITY of a given form. 
This is shown by -ci in (6) above, which turns V-2 of TYP:NO into 
TYP:V-MOD. A similar role is played by the CASE-MARKERS in the N 
system. The affixation of Case-markers turns N-2, which would be 
TYP:NO, into TYP:V-MOD/N-MOD. Needless to say, the prerogative of 
heads is to determine the distribution of constituents that they head. 
To the extent that i-affixes are responsible for this, one would want 
to view them as heads in the relevant sense. If so, we have a 
situation where i-affixes, the right-hand elements are heads. 

Of course, the traditional CATEGORIAL FEATURES V and N are 
always determined by the left-hand elements in Sells' analysis. 
However, the significance of this fact in Sells' system is not at all 
obvious, because these features do not play a crucial role in syntax 
and inflectional morphology in his system, for the following reasons. 
Normally, these features are taken to be responsible for constraining 
the syntactic distribution of constituents, that is, in the terminology 
just introduced, they are categorial AS WELL AS combinatoric features. 
But in Sells' system, since the latter function is taken over by the TYP 



Yoon: Inflectional structures in Korean and headedness 185 

feature, it is not clear what the role of V and N are in 
syntax/inflectional morphology. They obviously have a role to play 
in category-changing derivational morphology, which Sells does not 
deal with in the paper, but appear to have no direct relevance in 
syntax. 

Therefore, even if an analysis such as (6) proved to be correct, 
one must maintain the conclusion that V (left-hand element) is the 
head only in so far as the CATEGORIAL (what appear to be syntactically 
irrelevant) features are concerned, but (certain) i-affixes are heads 
regarding the COMBINATORIC (syntactically relevant) features, despite 
the fact that they are right-hand elements. 

3.3. Paradox between movement and selection 

Sells considers an alternative syntactic analysis, different from 
the one we proposed in the previous section. He shows that while 
this alternative may avoid problems of non-local selection, a 
paradoxical situation is reached, thus making it unworkable. We shall 
see that there is a simple solution to this "paradox" under the 
syntactic approach. 

Sells presents a strawman syntactic analysis which can avoid 
problems of non-locality noted earlier because this analysis takes 
selection for form to be determined AFTER head-movement. Taking 
(9) to be the S-structure formed after movement, we see that the 
constituent that is sister to anh- contains the -ci affix. 

(9) VP 

/ \ 
Af3 V 
/ \ I 
Af2 af anh- 
/ \ I 
Afl af -un 
/ \ I 
V af -man 
I I 

cek- -ci 

Sells reasons that this alternative predicts that morphological 
selection will be insensitive to outside affixes, since all that is 
required is that head movement put the subcategorized constituent 
in a position sister to the subcategorizing constituent by S-structure, 
regardless of how many other affixes may be added. 

However, he shows that contrasts such as those in (10a,b) falsify 
this. What (10) shows is that the requirement that the Copula select 
an N-' (N-stem) is satisfied in both structures after head movement 
in the strawman analysis we are considering, since the constituent 



186 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

containing N-i is sister to the Copula in both structures. However, the 
presence of -un, which turns an N' into an N^, makes (10b) ill- 
formed. Therefore, one cannot conclude that selection is satisfied 
AFTER movement, since the relevant contrast can only be explained if 
we assume that selection must be satisfied in the pre-movement 
structure. Only in (10a) is this requirement met. 

(10) a. John-eykey-man-ita vs. b. *John-eykey-man-un-ita 

[[[[N-3] N-2] N-1] ] [[[[N-3] N-2] N-i] NO] ] 

John-DAT-only-COP John-DAT-only-TOP-COP 

-ita selects N-i 

Let us see how this contrast may be dealt with in a syntactic 
analysis. As we have seen, in the syntactic account, there are two 
levels at which well-formedness can be stated — syntax and 
morphology. Since phrasal affixes are syntactic atoms, they select 
syntactically. However, being affixes, they must satisfy particular 
morphological requirements as well. Thus, phrasal affixes have a 
dual subcategorization, both of which must be independently 
satisfied (Sadock 1990). Taking the Copula to be such an affix, we 
could state its twin requirements as follows. 

(11) -ita: +V,-N, TYP:N-MOD, [XP ] (Syntax) 

+V,-N, TYP:N-MOD, [X-i ] (Morphology) 

(11) states that syntactically, the Copula combines with any maximal 
projection, but that morphologically, it has to be able to affix to a 
stem. 9 Given this, the Copula is free to combine syntactically with the 
relevant PP/DPs in (10a) and (10b). However, in (10b) the 
morphological requirement that Copula be a suffix on N-i (N-stem) is 
violated, since morphologically, John-eykey-man-un is an N^. 

3.4. Double marking 

Sells presents the following as another potential problem for the 
syntactic approach. Irregular honorific V-stems prohibit affixal 
honorific marking, as seen below. 

(12) *capsusi-si-ta 'eat (HON)-HON-Decr 
*tusi-si-ta 'eat (HON)-HON-Decr 

Sells argues that this is not predicted if HON is a head in the syntax 
(constituting AgrS, in the system of Chomsky 1992). Sells reasons 
that in the lexical view, MORPHOLOGICAL BLOCKING can account for (12), 
since suppletive forms already have honorific information marked, 
thus blocking redundant marking. 



Yoon: Inflectional structures in Korean and headedness 187 

One could respond to this in a variety of ways. J-M Yoon (1990) 
argues on the basis of constituency tests (cf. §4 below) that -si does 
not form an independent projection in syntax. If so, nothing need be 
added to Sells' lexical account. Since the attachment is lexical, it will 
be subject to whatever principles constrain lexical forms. 

Another response may be along the following lines. The 
irregular honorific V-stems always incorporate -si as part of its 
stem/root. Given this, further affixation of an identical affix would 
run afoul of the principle restricting duplicate affixation of the same 
affix — No Vacuous Affixation Principle (Marantz 1984). 

Still another way would be to assume that -si is Agr in syntax, 
but that selects a VP whose head cannot be marked [+HON] 
inherently. This is the syntactic analogue of "blocking" and is no less 
ad hoc than any lexical account of blocking, as far as I can tell. 

Sells also argues that if Case-markers head their own projections 
(DPs or KPs) in syntax, the redundant double marking of Nominative 
in example (13) remains a problem, since there would have to be two 
DPs headed by two Nominative affixes. 

(13) apeci-kkeyse-man-i 'father-HON NOM-only-reg. NOM' 

In responding to this argument, we wish to draw attention to 
the fact that Korean quite regularly allows double morphological 
combinations of Case-markers, as long as they are Inherent and 
Structural Case-markers, occurring in that order (J-M Yoon 1991). 
This is seen in example (14). 

(14) a. John-eykey-ka (J-DAT-NOM) 

Inh-Str 

b. *John-ka-lul (J-NOM-ACC) 
*Str-Str 

c. John-eykey-man-i (J-DAT-only-NOM) 
Inh-...-Str 

d. apeci-kkeyse-man-i (father-hon. NOM-only-pl. NOM) 
Inh-...-Str 

e. *John-ka-eykey (J-NOM-DAT) 
*Str-Inh 

Such Case-doubling is allowed because the two types of Case- 
markers differ morphologically, the Inherent Case-markers being N^ 
affixes, and Structural Case-markers being N' affixes. Now, if we 
make the reasonable assumption that honorific NOM is Inherent Case 
and regular NOM is a Structural Case, their combination would be 
allowed without stipulation. 'O 



188 Studies in the Linguistic Sciences 21:2 (FaU 1991) 

I have thus far either rebutted or presented alternative 
syntactic accounts of some of the facts that Sells considered 
problematic for a syntactic account. In the next section, I quickly 
rehearse some previously presented arguments FOR a syntactic 
analysis of Korean i-structures. 

4. Arguments for a syntactic account of Korean inflection^ i 

4.1. Constituency and coordination 

As noted in Yoon (1989), Yoon & Yoon (1990), J-M Yoon (1990), 
coordinate structures in Korean allow certain i-affixes to be missing 
in all but the final conjunct. In such cases, the information borne by 
the i-affixes on the final conjunct takes a distributive scope over the 
unmarked non-final conjuncts. This is illustrated below in (15). 

(15) a. John-i pap-ul mek-ess-ta, kuliko Mary-to pap 
J-NOM meal-ACC eat-Pst-Decl and M-also meal 
-ul mek-ess-ta 
-ACC eat-Pst-Decl 
'John ate the meal and Mary (also) ate the meal' 

b. John-i pap-ul mek-ess-(*ta)-ko Mary-to pap 
J-NOM meal-ACC eat-Past-Conj M-also meal 
-ul mek-ess-ta 

-ACC eat-Past-Decl 

'John ate the meal and Mary (also) ate the meal' 

c. John-i pap-ul mek-ko Mary-to pap-ul mek 
J-NOM meal-ACC eat-Conj M-also meal-ACC eat 
-ess-ta 

-Pst-Decl 

'John ate the meal and Mary (also) ate the meal' 

In non-affixal coordination, (15a), verbs of both conjuncts are 
independently inflected for Tense and Mood. However, in affixal 
coordination, (15b,c), Mood marking is prohibited, (15b), while Tense 
marking may be absent, (15c), in non-final conjuncts. Despite the 
absence of such marking, the verb of the initial conjunct is 
interpreted as Past and Declarative. 

In the works cited above, this peculiar behavior of coordination 
is taken to be evidence that TP and VP are syntactic CONSTITUENTS. 
Under the assumption that only constituents can be coordinated, it is 
proposed that (15b) instantiates coordinated TPs and (15c) 
coordinated VPs. This, together with the hypothesis that subjects are 



Yoon: Inflectional structures in Korean and headedness 1 89 

VP-internal (and may remain there), accounts for word order and 
the obligatory distributive scope of final conjunct i-affixes.'^ 

(16) 



(=15b) 


/ 
TP 


MP 

\ 
M 


b. 


(=15c) MP 
/ \ 
TP M 


/ 
/ 
TP 




\ 1 
\ -ta 
TP 




/ \ 1 
VP T -tc 

/ \ 1 


/ \ 
/ \ 




/ \ 
/ \ 




/ \ -ess 
VP VP 


c-ess-ko 


i mek-ess- 




/ \ / \ 
mek-ko ..mek- 



4.2. Negation and coordination 

The interaction between postverbal negation and coordination 
provides another piece of evidence for the posited constituent 
structures. Postverbal negation on the final conjunct can have 
distributive scope over non-final conjuncts. However, it cannot do so 
when the verb of non-final conjunct is inflected for Tense (cf. 17a vs. 
17b). 

(17)13 a. John-i pap-ul mek-ko Mary-ka swul-ul 

J-NOM meal-ACC eat-Conj M-NOM beer-ACC 
masi-ci anh-ass-ta 
drink-COMP not-Pst-Decl 

■John did not eat the meal and Mary did not drink 
beer' 
b. John-i pap-ul mek-ess-ko Mary-ka swul-ul masi-ci 
anh-ass-ta 

John ate the meal and Mary did not drink beer' 
'John did not eat the meal and Mary did not drink 
beer' 

J-M Yoon (1990) accounts for this contrast in the following way. 
Postverbal negation requires an uninfected V but precedes Tense. 
Therefore, it is natural to posit Neg as a head between VP and TP. 
Given this. If TP is selected without Neg (cf. 17b), such a clause will 
be necessarily interpreted as affirmative, while if VP is selected (cf. 
17a), Neg can have distributive scope over it. 

4.2. One tense-V per clause restriction 

Cho & Morgan (1987) note that two adjacent INFLECTED V's are 
always interpreted as clausal conjunction, while the sequence of 
uninflected and inflected Vs may be interpreted clausal or sub- 
clausal conjunction. This is curious, since if Korean patterns like 
English, where subclausal coordination demands identically inflected 
V's (cf. 19), the sequence of two V's should be interpreted as 



190 Studies in the Linguistic Sciences 21:2 (Fall 1991) 

subclausal or clausal conjunction. !'♦ However, when the first V is 
inflected, the structure can only be interpreted as S-conjunction. 

(18) a. John-i mek-ko (pro) ca-ss-ta 

= 'John ate and slept' (VP conjunction) 
= 'John ate and he/someone slept' (S conjunction) 
b. John-i mek-ess-ko pro ca-ss-ta 
= 'John ate and he/someone slept' (S conjunction) 
:^ 'John ate and slept' (VP conjunction) 

(19) a. John ate the meal and drank beer (VP conjunction) 
b. John bought and drank Kirin beer (V conjunction) 

This contrast between English and Korean is expected under the 
structures we have been assuming. The presence of inflected V 
implies the presence of Tense, which in turn implies TP, a clausal 
node. 

NOTES 

*An earlier version of this paper was presented at the Seoul 
International Conference on Linguistics (SICOL) in August 1992. 

1 The term i-structures designates any inflected (as opposed to 
derived, compounded) form of word, or a part of it. 

2 This is not surprising in view of the fact that neither Pollock 
nor Chomsky offered any morphosyntactic arguments for the 
dissociation of INFL into Tense and Agr. The strategy used (by 
Pollock, at least) to arrive at the intended conclusion seems to be as 
follows. There are at least three positions that verbs can occur in a 
clause in English/French. Supposing one is the base position inside 
VP, the other two must be Tense and Agr, respectively. This is hardly 
a morphosyntactic consideration, as should be obvious. 

3 The independence of these two issues is driven home in 
Lapointe (1991), who, working in Autolexical framework, recognizes 
the morphosyntactic independence of i-affixes in Korean but treats 
them as modifiers rather than heads. 

'^ And non-inflectional morphology, such as derivation and 
compounding. 

5 Sells (1991) and Cho & Sells (1990) posit three levels of 
structure below the WORD (xO) for Korean morphology, where x-3 
level is the ROOT, x-2 the SUB-STEM, and x-i the stem. This division is 
motivated by the loosely templatic character of Korean inflection. 
However, the distinction is sometimes obliterated in surface forms 
since direct inheritance into higher bar-levels without affixation is 
also possible. 



Yoon: Inflectional sinictiires in Korean and headedness 191 

While different kinds of i-affixes can be attached to a V-root, 
two kinds of affixes are obligatory. Tense and Mood affixes. These 
are the ones which "shift" the bar-levels to a higher level in Sells' 
system. In the syntactic alternative, this translates into the 
requirement that the projections of T and M are obligatory, while 
those of others will be optional. 

6 That is, while Tense and Mood are verbal affixes, their logical 
scope is over the entire proposition (sentence). See J-M Yoon (1990) 
for detailed justification. The significance of affix-ordering 
generalizations for syntax is due to Baker (1985). 

7 The mapping between (2) and (3) is sanctioned in ALS by the 
Incorporation Principle. Let us be a bit more specific. As phrasal 
affixes, M and T possess dual subcategorizations, thus, M 
syntactically selects TP, and morphologically selects T (a sub-stem). T 
s-selects VP and m-selects V (stem/root). 

8 I take -ci to be a "Comp" on VP, and the delimiters to attach to 
VP, which in turn is selected by the Neg auxiliary. 

9 The copula in Korean affixes to any category, and not just to 
nominals. We have adjusted the subcategorizations to reflect this 
fact. 

10 There remains the question of how two NOM's may be 
asssigned to a single NP syntactically. The analysis is similar to that 
proposed for DAT-NOM combinations (Yoon & Yoon 1991). Assuming 
subjects to originate within the VP, honorific NOM would be assigned 
there, while regular NOM assigned when the subject raises to Spl. 
The optional raising of Case-marked NPs to Spl is well-attested in 
Korean, even though such movement is not required by the Case 
Filter. Cf. Yoon & Yoon (1991) for more discussion. 

1 • The arguments sketched here are presented in much more 
detail in Yoon (1993) and interested readers are referred to that 
work. 

12 Yoon (1989) provides further evidence for TP and MP from 
various kinds of nominalizations. For reasons of space. I do not go 
into details. 

•3 (17a) marginally allows a reading where negative scope is 
confined to the final conjunct, meaning. "John eats the meal, but/and 
Mary did not drink beer." This possibility arises because the Present 
tense morpheme may be null in Korean. The oddity of this reading is 
due to an unnatural sequence of tenses. 

••* This possibility exists because Korean is a pro-drop language. 



192 Studies in the Linguistic Sciences 21:2 (FaU 1991) 

REFERENCES 

Ahn, H-D. & H-J. Yoon. 1989. Functional categories in Korean. Harvard 

Studies in Korean Linguistics III, 79-88. Seoul: Hanshin. 
Baker, M. 1985. Mirror principle and morphosyntactic explanation. 

Linguistic Inquiry 16.373-416. 
Cho, J-0 & J. Morgan. 1987. The interaction of syntax and morphology 

and Korean VP coordination. Harvard Studies in Korean 

Linguistics II, 27-39. Seoul: Hanshin. 
Cho, Y-M Yu, & P, Sells. 1991. A lexical account of phrasal suffixes in 

Korean. Ms. Stanford University. 
Chomsky, N. 1988. Some notes on the economy of derivations and 

representations. MIT Working Papers in Linguistics 10.43-74. 
. 1992. A minimalist program for linguistic theory. MIT Occasional 

Papers in Linguistics 1. 
Grimshaw, J. 1991. Extended projection. Ms. Brandeis University. 
LaPOINTE, S. 1991. Korean verb-markers and Autolexical theory. 

Proceedings of Formal Linguistics Society of Midamerica 2.144- 

162. 
LlEBER, R. 1992. Deconstructing morphology. Chicago: University of 

Chicago Press. 
Marantz, a. 1984. On the nature of grammatical relations, MIT Press: 

Cambridge. 
POLLOCK, J-Y. 1988. Verb movement, UG, and the structure of IP. 

Linguistic Inquiry 20.365-424. 
Sadock, J. 1990. Autolexical syntax. Chicago: University of Chicago 

Press. 
Sells, P. 1991. Korean and Japanese morphology from a lexical 

perspective. Ms. Stanford University. 
Whitman, J. 1989. Topic, modality and IP structure in Korean. 

Harvard Studies in Korean Linguistics III, 341-356. 
Yoon, J. 1989. A restrictive theory of morphosyntactic interaction and 

its consequences. University of Illinois at Urbana-Champaign 

Ph.D. dissertation in Linguistics. 
. 1991. Korean nominalizations, lexicalism, and morphosyntactic 

interface. Ms. University of Illinois at Urbana-Champaign. 
1993. Tense, coordination, and clausal structure of English and 

Korean. Paper read at the 5th Harvard International Symposium 

on Korean Linguistics. Cambridge, MA. 
, & J-M Yoon. 1990. Morphosyntactic mismatches and the 

function-content distinction. Chicago Linguistic Society 26. 

1991. Chain condition, ambiguity of government, and 

derivational grammars. Northeastern Linguistic Society 21. 
YOON, J-M. 1991. The syntax of A-chains. Cornell University Ph.D. 

dissertation in Linguistics. 



) 



^ 



i 



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 7V4%, IN 5%. MI 4%. MN 6%. OH 5%, WI 5%. 

While quantities last. Volumes 7:2 through 16.1 are being offered at half-price. 



Papers in General Linguistics $2.50 S^rOO 

Papers in General Linguistics $2.50 $^:06 

Papers in General Linguistics $2.50 $5.00 

Papers in General Linguistics $2.50 $&^ 

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 

(Editor: Yamuna Kachru) $2.50 $&m 

Papers in General Linguistics $2.50 $5.00 

Papers on Diachronic Syntax: $2.50 S^tOO 

Six Case Studies 

(Editor: Hans Henrich Hock) 

Papers in General Linguistics $2.50 $5.00 

Studies in Language Variation: $2.50 $5.00 

Non western Case Studies 
(Editor: Braj B. Kachru) 

Papers in General Linguistics $2.50 $5:00 

Language in African Culnire and Society $2.50 $5.00 

(Editor: Eyamba G. Bokamba) 

Papers in General Linguistics $2.50 &&rOQ 

Linguistic Studies in Memory of $2.50 $5:00 

Theodore M. Lighlner 
(Editor: Michael J. Kenstowicz) 

Papers in General Linguistics $3.00 $^:00 

Illinois Studies in Korean Linguistics $3.(X) $^:00 

(Editor Chin-W. Kim) 



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 


Vol. 18:1 


Spring 1988 


Vol. 18:2 


Fall 1988 


Vol. 19:1 


Spring 1989 


Vol. 19:2 


Fall 1989 



STUDIES IN THE LINGUISTIC SCIENCES 

The following issues are available: 



Papers from the 1986 South Asian $6.00 

Languages Analysis Roundtable 
(Editor Hans Henrich Hock) 

Papers in General Linguistics $6.00 

Papers in General Linguistics $6.00 

P£q)ers in General Linguistics $6.00 

Papers in General Linguistics $7.50 

The Contribution of African Linguistics $7 . 50 

to Linguistic Theory, Vol. 1 
(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 Aimiversary 
(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 

Vol. 21:1 Spring 1991 Pq)ers in General Linguistics $7.50 

Vol. 22:1 Spring 1992 Pqjers in General Linguistics $7.50 

IN PREPARATION: 

Vol. 20: 1 Spring 1990 The Contribution of African Linguistics $7.50 

to Linguistic Theory, Vol. 2 
(Editor Eyamba G. Bokamba) 

Vol. 22:2 Fall 1992 Twenty-Five Years of Linguistic Research $7.50 

at the University of Illinois at Urbana-Champaign: 
Post-graduate research by Doctoral and Master's 
degree students in the Department of Linguistics 
(Editors: Braj B. Kachru and Frances Vavrus) 

Vol. 23:1 Spring 1993 Papers in General Linguistics $7.50 

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, Dlinois 61801 



I