Skip to main content

Full text of "Studies in the linguistic sciences"

See other formats



AT l*R8AUA-CHAi'^P.'.:.>i 


The person charging this material is responsible for 

us return to the library from which it was withdrawn 

on or before the Latest Date stamped below. 

Theft mutilation, and underlining of books are reasons for discipli- 

?o7en!J, c Tt T"): '""" '" *'"""*^^' '^°'" ••'^ University. "^ 
To renew call Telephone Center, 333-8400 


Modern |ang& Lilt Library 
k 425 33-0071 

LI61— O-I096 


Itudies in 

?he Linguistic Sciences 




ESTHER BENTUR Orthography arid the Formulation of 
Phonological Rules 

Headless Relative Clauses in Quechua 

AND S.N. SRIDHAR On the Acquisition 
of Subjecthood 

tY OJ. 





in Imbabura Quechua 



On Ergativity in Selected South Asian Languages 

JAMES LEVINE The Relative Pronoun and the Long 
Form Adjective in Russian 

DAVID ODDEN Aspects of Iraqi Arabic Verbal Phonology 

Government: The Case of the Passive 
Rule in Hindi 

IRMENGARD RAUCH Language-Likeness 


of the Class 5 Prefix in Shona 

S. N. SRIDHAR Linguistic Convergence: Indo- 
Aryanization of Dravidian Languages 

SUSAN U. STUCKY How a Noun Class System may 
be Lost: Evidence from Kituba (Lingua 
Franca Kikongo) 

department of Linguistics 
' University of Illinois 



EDITORS: Charles W. Kisseberth, Braj B. Kachru, Jerry L. Morgan 

REVIEW EDITORS: Robert N. Kantor and Ladislav Zgusta 

EDITORIAL BOARD: Eyamba G. Bokamba, Chin-chuan Cheng, Peter 
Cole, Georgia M. Green, Hans Henrich Hock, Yamuna Kachru, Henry 
Kahane, Michael J. Kenstowicz, Chin-W. Kim and Howard Maclay. 

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. Especially invited 
papers by scholars not associated with the University of Illinois will also be in- 

SPECIAL ISSUES: Since its inception SLS has been devoting one issue each 
year to restricted, specialized topics. A complete list of such special issues is 
given on the back cover. The following special issues are under preparation: 
Fall 1978: Linguistics in the Seventies: Directions and Prospects (Forum Lec- 
tures delivered at the 1978 Linguistic Institute of the Linguistic Society of 
America), edited by Braj B. Kachru; Fall 1979: Relational Grammar and 
Semantics, edited by Jerry L. Morgan; Fall 1980: Papers on Diachronic Syn- 
tax, edited by Hans H. Hock; Fall 1981: Studies in Language Variation: 
Nonwestern Cast Studies, edited by Braj B. Kachru; Fall 1982: South Asian 
Linguistics: Syntax and Semantics, edited by Yamuna Kachru. 

BOOKS FOR REVIEW: Review copies of books (in duplicate) may be sent to 
the Review Editors, Studies in the Linguistic Sciences, Department of 
Linguistics, University of Illinois, Urbana, Illinois, 61801. 

SUBSCRIPTION: There will be two issues during the academic year. Requests 
for subscriptions should be addressed to SLS Subscriptions, Department of 
Linguistics, 4088 Foreign Languages Building, University of Illinois, Urbana, 
Illinois, 61801. 

Price: $5.00 (per issue) 



Charles W. Kisseberth 
Braj B. Kachru, Jerry L. Morgan 


Robert N. Kantor and Ladislav Zgusta 


Eyamba G. Bokamba, Chin-chuan Cheng, Peter Cole, 

Georgia M. Green, Hans Henrick Hock, Yamuna Kachru, Henry Kahane, 

Michael J. Kenstowicz, Chin-W. Kim, and Howard Maclay. 

SPRING, 1978 



Esther Bentur: Orthography and the Formulation of Phonological 

Rules 1 

Peter Cole, Wayne Harbert and Gabriella Hermon: Headless Relative 

Clauses in Quechua 26 

Peter Cole, V/ayne Harbert, Gabriella Hermon and Shikaripur Sridhar: 

On the Acquisition of Subjecthood 42 

Peter Cole and Janice Jake: Accusative Subjects in Imbabura 

Quechua 72 

Janice Jake: l\'hy Dyirbal isn't Ergative at all 97 

Yamuna Kachru and Rajeshwari Pandharipande : On Ergativity in 

Selected South Asian Languages Ill 

James Levine: Tlie Relative Pronoun and the Long Form Adjective 

in Russian 127 

David Odden: Aspects of Iraqi Arabic Verbal Phonology 137 

Rajeshwari Pandharipande: Exceptions and Rule Government: 

The Case of the Passive Rule in Hindi 153 

Irmengard Rauch: Language-Likeness 174 

Pamela S. Silver and Scott R. Krause: A Reanalysis of the Class 5 

Prefix in Shona 181 

S.N. Sridhar: Linguistic Convergence: Indo-Aryanization of 

Dravidian Languages 197 

Susan U. Stucky: How a Noun Class System may be Lost: Evidence 

from Kituba (Lingua Franca Kikongo) 217 

studies in the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 



Esther Bentur 

Within the framework of generative phonology, a 'rule' stands for an 
hypothesis about a linguistic generalization perceived by speakers. In a 
linguistic model which aims at characterizing linguistic competence realis- 
tically, only valid generalizations--namely, only those which speakers may 
abstract given the input data--should be expressed. An adequate analysis 
should neither overstate nor underestimate the speaker's internalized knowl- 
edge. It is usually assumed that only oral data counts as the basis on 
which hypotheses are constructed by speakers (and consequently by linguists) . 
To demonstrate that this tenet cannot be maintained it will be shown that a 
generalization, which is recoverable synchronically only from the written 
data, is nevertheless perceived by speakers and becomes an integral part of 
their linguistic knowledge. 

The example will be constructed as follows: 

First, the relevant data and the historical background will be presented. 
An abstract analysis to account for those facts will be proposed. Several 
arguments against it will be provided, and consequently, it will be rejected. 
An alternative proposal will be offered and its validity will be tested in 
two productivity tests; one involving hypothetical lexical items and the other 
the elicitation of non-existing extensions of actual lexical items. The re- 
sults of the second test will point, however, to the inadequacy of the alter- 
native (i.e. non-abstract) proposal. A solution to the apparent paradox will 
be offered by incorporating an additional factor into the model, namely the 
influence of orthography. To substantiate the claim, that this is indeed 
the crucial factor, the results of a third productivity test, this tima a 
written one, will be mentioned. 

1 . DATA 

Consider the patterns of past and present forms of the following verbs: 

(1) kara 'he read' kore 'he reads' 
kafa 'he froze' kofe 'he freezes' 
maca 'he found' moce 'he finds' 


(2) kara 'he tore' korea 'he tears' 
mana 'he prevents' monea 'he prevents' 
?ama 'he heard' 2omea 'he hears' 

Whereas the past tense forms of both 'find' and 'hear' have the same 
configuration CaCa , the corresponding present tense forms exhibit different 
patterns CoCe (for 'finds', 'freezes' and 'reads') but CoCea (for 'tears'. 

'prevents' and 'hears'). Comparison with the pattern of regular verbs (i.e. 
where all the three radicals surface) such as: 

(3) lamad 'he studied' 
savar 'he broke' 

lomed 'he studies' 
sover 'he breaks' 

suggests, that the present tense singular form in this verb paradigm is 
C^oC eC . The CoCea forms, thus, manifest an additional final vowel a 
(beside the non-occurrence of the third radical) . 

A similar pattern shows up in the following verbs: 

(4) maSax 'he pulled' 
na§ax 'he bit' 

mosex 'he pulls' 
no^ex 'he bites' 

(5) barax 'he ran away' 
carax 'he screamed' 

boreax 'he runs away' 
coreax 'he screams' 

Here, the present tense forms are either of the type C-]^oC2ex (as in 'pulls' 
and 'bites'), or CioC2eax, with an additional [a] preceding the third radical 
[x] (as in 'runs away ' and 'screams'). 


These synchronic patterns of alternations are reflexes of an historical 
rule which inserted the vowel a when either /h/, /h/ or /'^Z occurred in 
final position following a non-low vowel (/i/, /e/, /o/ or In/). 

Expressed in distinctive features, the rule can be formulated as: 

a-Insertion Rule 






In MH, however, the following changes have taken place: 

(7) / / is always realized as [ ?] or 0. 

/h/ surfaces always as [x] and thus has merged completely 
with [x] which is an allophone of /k/ . 

/h/ is unstable in informal speech and is realized by most 
speakers as [ ?] or as 0. 

In spite of those mergers, a still surfaces in MH only when the fol- 
lowing consonant was historically /^/, /h/ and /h/. 


Consider first the analysis which recapitulates the historical events. 
According to it the a-Insertion Rule would still be a synchronic rule. Its 
application must precede the merging rules: 

(8) "" >> 7 

h ) X 

1^ > ? 

The ordering is necessary to ensure that [a] will surface only when 
it is followed by [ ?] or [x] , which are realizations of historical /^ / , /h/ 
and /h/. 

This analysis, which is crucially based on the postulation of underlying 
Z'^/, /h/ and /h/ as phonemes of MH, is maintained by most standard Hebrew 
grammar books, and was formulated within a generative framework by Oman 
(1972). It "explains" the fact that phonetically identical segments behave 
differently with respect to the application of the same phonological rule. 
As Oman mentions, given this analysis, the morphology of Hebrew has become 
considerably easier. 

The price, however, one should note, is a highly abstract phonemic 
system with underlying segments that never show up on the surface. In what 
follows, arguments against this abstract analysis will be presented. It will 
be demonstrated that it cannot possibly be maintained, given the existing 
internal evidence which clearly points to its inadequacy. 


The postulation of non-occurring surface segments as underlying elements 
in the synchronic system has been a controversial issue within generative 
phonology. 2 Even the proponents of abstract analyses admit that the burden of 
proof, in such cases, is on the one who advocates abstractness. For many 
linguists, however, analyses incorporating hypothetical segments are auto- 
matically excluded from the range of viable synchronic analyses (see Hooper 
1976) . No additional arguments are needed to convince these linguists of 
the inadequacy of the analysis proposed so far, since it involves the positing 
of the pharyngeals and voiceless glottal fricative as underlying phonemes in 
MH even though these elements are not manifested on the surface. 

Yet, I would like to discuss additional evidence to indicate that an 
abstract approach in this case of MH cannot be maintained even by those who 
allow for hypothetical segments under certain conditions (if, for example, 
several rules crucially refer to them; see Kiparsky 1971) . 

4.1 The selection of the feature F+low] 

It is widely accepted that the advocate of an abstract approach has to 
demonstrate clearly that the nature of the postulated segment is determined 
uniquely by the structure of the language. The feature assigned to it must 
be well-established within the internal system. 

In the Yawelmani case (Kisseberth 1969), for example, the abstract seg- 
ments were assigned the feature [+high] . The selection of this specific 
property was motivated by the behavior of the corresponding surface segments 
with respect to several phonological rules. Kisseberth notes that "the 
choice of u: and i^: as underlying source of some surface o: and all surface 
e_: is motivated since no other choice would perform the same function," 
Arguments along the same lines are provided by Brame (1972) who argues for 
an underlying voiced pharyngeal fricative in Maltese. 

The selection of the feature [+low] for the abstract segments in Hebrew 
cannot, however, be supported on internal grounds. There are no additional 
[+low] segments, which trigger the a-Insertion Rule, that could be cited as 
independent evidence that the presumed underlying feature involved is [+low] ; 
the only (+low) consonants, which historically conditioned the rule, are 
precisely those whose existence as elements in the synchronic system is being 
questioned. The assignment of the feature [+low] to those abstract segments 
is therefore totally arbitrary from a synchronic point of view and cannot be 
motivated given the internal organization and structure of the language; 
theoretically other phonemic features could have been posited as well. 

4.2 Current changes in MH 

The classical argument in favor of the postulation of abstract under- 
lying segment is based on the observation that this specific segment be- 
haves, with respect to several phonological rules, not in accordance with 
its surface properties, but rather similarly to other surface segments, 
which share a specific feature. In the Yawelmani case, for example, (Kisse- 
berth 1969), surface long non-high vowels were posited as high ones under- 
lyingly, since they behave like surface high vowels rather than non-high 
ones. If it can be shown, however, that, with regards to some of the rele- 
vant rules, the abstract segment stops functioning as predicted by its pre- 
sumed underlying features, and starts behaving according to its surface 
properties, this would cast doubt at the validity of an abstract analysis. 

An abstract analysis implies that speakers are able to maintain a 
phonemic system which includes non-surfacing segments. No subsequent changes 
in the direction of a more surface-oriented system should be anticipated. 
Ideally, all rules should continue referring to the underlying (abstract) 
rather than surface properties. In what follows several developments which 
have taken place in MH will be discussed; each of them has affected the ab- 
stract segments. What is common to all of these changes is the fact that 
the abstract segments are being treated according to their surface features. 
A proponent of the abstract analysis would probably be able to provide an 
explanation for the emergence of each individual development; on the whole, 
however, he would have to claim that the fact that all of these developments 
affect specifically the abstract segments is totally accidental. 

A non-abstract analysis (i.e. one which maintains that pharyngeals can- 
not be posited underlyingly in MH) predicts that those developments will in- 
evitably take place. Such changes are not regarded as random individual in- 
stances but rather as related developments within a general scheme. Only a 
non-abstract analysis can thus offer a unitary explanation for the emer- 
gence of those changes in MH ; the abstract one will have to treat them on a 
fragmentary accidental basis. 

Barkai (1972) mentions several rules which were conditioned histori- 
cally by /?/, 1^ I , /\\/ and /h/ (henceforth: guttural segments). He argues 
that, since historical /h/ still functions synchronically as a guttural 
with respect to these rules it must be posited underlyingly as such, even 
though it always shows up on the surface as a voiceless velar fricative 
[x] (i.e. as a segment which is not [+low]). Otherwise, he states, each 
of those rules would have to be complicated. 

However, for most of those rules mentioned by Barkai, it can be demon- 
strated that they are no longer obligatory for historical /h/. 

4.2.1 The Vowel Lowering Rule 

Consider, for example, the synchronic status of the historical Vowel 
Lowering Rule : 

Vowel Lowering 
(9) p-syllabd 
C hi 

liable 1 r -hls-^ / r-sylUblc-l V J 

igh J [20< low / +10W 

The reflexes of this rule still show up in MH. Compare verbs in the 
hiCCiC paradigm which do not have an historical guttural as a first radical, 

(10) hixmin 'invite' 
hisbir 'explain' 

with those having historical /?/ or / / in the same position. 

(11) he?ezin 'listen' 

he?enik 'grant' (historically: he enik) 

(The vowel following the guttural is an inserted echo vowel.) 

The rule no longer applies obligatorily when the first radical was an 
historical /h/; variants with a non-lowered prefix vowel have emerged, i.e. 

(12) hexvir % hixvir 'become pale' (historically: hehvir) 
hexzir 'v hixzir 'return (historically: hehzlr) 

Both alternants are found in free variation in the speech of the same 


Similarly one finds: 


(13) texanek "^ tixanek will suffocate' 

' she 

texatex '\' tixatex will be cut' 


Given an abstract analysis, these data can be accounted for by assuming that 
the Vowel Lowering Rule has become optional for some mysterious reason. 

Thus, if it does apply to underlying /h/ the output, following the merger 
h X will be [hexvir] . If, on the other hand, Vowel Lowering was skipped 
in the course of the derivation the output would be [hixvir] . The fact that 
the rule has become optional in MH is viewed, within this framework, as an 

A non-abstract analysis predicts that a variant such as [hixvir] will 
indeed emerge, since the element which triggered the Lowering rule histori- 
cally (i.e., /h/ is no longer a segment in the synchronic system. [hexvir], 
on the other hand, would be treated as an historical residue. 

Another related development, which has taken place in MH , is the emer- 
gence of the following variation: 

(14) hixtiv '^ hextiv 'dictate' (cf. katav 'write') 
hixlil "^ hex^il 'fail s.o' (cf. kisalon 'failure') 
hixbid 'V' hexbid 'make heavier' (cf. kaved 'heavy') 

Whereas in the case of hixzir and hixvir the velar fricative surfaces in all 
related forms (cf. xazar 'return', xazara 'rehearsal', xiver 'pale') there 
is a k 'V x alternation in the pairs of the type hixtiv - katav , hixsil - kilalon . 
To capture the relatedness between the latter pairs, an underlying /k/ must 
be postulated as the first radical; it surfaces as [x] due to the operation 
of spirantization. Since the segment which underlies the surface [x] in 
those verbs cannot possibly be /h/, the occurrence of the alternant with the 
lowered prefix vowel (i.e. hextiv ) cannot be attributed to the effect of a 
following [+low] consonant. The emergence of these variants may be viewed 
as a case of analogy. 

(15) hixvir - hexvir 

hixtiv - X = hextiv 

The Lowering Rule has thus been extended to all surface [x]'s regardless of 
their historical origin. Notice that as a consequence of the merger h x 
this rule has become a minor one; it applies only to some [x]'s (i.e. the 
reflexes of historical /h/'s) which had to be lexically marked. The over- 
generalization of the rule implies the elimination of the arbitrary marking, 
a development which results, of course, in grammar simplification. Given 
the emergence of the variations in (14) it becomes evident why the existence 
of alternants with the lowered vowel cannot be cited as an argument to sup- 
port the selection of the feature [+low] to the abstract segment. 

4.2.2 Historical constraints on surface strings #C l+low| , L+lowJ C 

Historically, Hebrew had restrictions on surface strings of the type 

#C [+low] and * [+low] C ; to break such impermissable sequences an epen- 
thetic vowel was inserted (for details see Barkai 1972, Bolozky 1972). The 
reflexes of these constraints are still manifested in MH: 

Compare : 

(16) tmixa 'support' 
svita 'strike' 

(17) te?ima 'tasty' (f. sing.) 
S£7iva ' pumping ' 


(18) mam^ix 'continue' (m. sing, pres.) 
matxil 'start' (m. sing, pres.) 

(19) ma?amin ( 'v mamin) 'believe' (m. sing, pres.) 
maTatik ('vmatik) 'copy' (m. sing, pres.) 

These constraints no longer hold for surface [x]'s which are reflexes of 
historical /h/'s. 

Thus we find in MH : 

(20) txina 'grinding' 

sxita 'slaughter' (compare (17)) 

taxsov 'you will think' 

maxzik 'hold' (compare (19)) 

These are clear cases where the presumed abstract segment no longer behaves 
according to its abstract underlying properties, but rather according to its 
surface features. 

Within an abstract analysis this development can be handled by ordering 
the relevant rules as: 

1) h ^x 

2) Vowel Epenethsis 

A proponent of this approach may even claim, perhaps, that the mere fact 
that an abstract segment stops functioning according to its underlying 
properties with respect to one (or even several rules) is an immaterial acci- 
dent and cannot be regarded as a strong argument against abstractness. He 
might go on to suggest that as long as for some other rules (or even one??) 
the segment in question continues to behave in accordance with its abstract 
properties this would be a sufficient indication for its existence. 

A non-abstract analysis, on the other hand, predicts the emergence of 
such changes, i.e. that rules and constraints will become sensitive to sur- 
face features rather than to abstract underlying ones. 

4.2.3 The convergence of the le?eCoC and la7aCoC patterns 

Another process which is presently taking place in MH supports the 
claim that the historical phonemic contrast 111 : 1^1 is being abolished. 

Consider the following data: 

The infinitive of the Hebrew verb paradigm C,aC aC has the configura- 
tion liC-C oC as in: 

(21) lilmod 'to study' (root L-M-D) 
liSbor 'to break' (root S-B-R) 

For verbs which had as a first radical /?/, the infinitive surfaced as 
le7eC2oCo; if, however, the first radical was 1^1 the infinitive was of the 
pattern Ia'^aC„oC-. Compare: 

(22) le?exol 'to eat' (7-K-L) 
le?esof 'to collect', 'to gather' (7-S-P) 


to work' ( -B-D) 
la^amod 'to stand' ('^-M-D) 

If the phonemic distinction / / : /?/ is indeed kept in MH one would 
have expected those patterns to be maintained; this however is not the case. 
We find the following forms in free variation in standard MH even in the 
speech of educated adults: 

(24) le?esof -v laTasof 'to collect' 

le?enox 'v. la?anos 'to rape' 

le?efot % la?afot 'to bake' 

leTeroz '\> la?aroz 'to pack' 

In all of the forms in (24) the first radical is historical 111 . Yet it 
seems that the pattern laTaC^^oC^ is more common in these cases (only two 
verbs: leTexol 'to eat' and levehov 'to love' do not show these variations). 
Within the framework of an abstract analysis which maintains the /?/ : 1^1 
contrast, this development must be treated as accidental. 

Viewed, however, in combination with other cases which have been cited 
above, this process is readily explained; the speakers' inability to keep 
the two patterns apart is another indication that the relevant phonemic 
contrast (in this case /?/ : /^/) is no longer maintained in their phonemic 

4.2.4 The behavior of morphemes containing several abstract segments 

Consider the case of lexical items which contain several occurrences 
of the same presumed underlying segment. 

The adjective taxuax 'crumbled, crushed (soil)' is derived from the 
historical root T-H-H. Within an abstract analysis this morpheme would be 
represented underlyingly as /tahuh/. The third radical behaves as predicted 
by this analysis; it triggers the a-Insertion Rule. The second radical, 
however, does not behave as a [+low] segment; the corresponding fem. sing, 
form in [txuxa] namely a vowel is not inserted to break the impermissible 

sequence #C [^+low| . 

An abstract analysis cannot provide a plausible explanation to the 
question: Why did this root receive such a mixed representation? Why was 
the same segment analyzed as [+low] only in one of its occurrences in the 
same lexical item? 

A non-abstract approach, on the other hand, would assume that both the 
second and the third radical are underlyingly /x/, but one of them (i.e. the 
third) is arbitrarily marked for the a-Insertion Rule. 

4.3 Children's behavior 

The linguistic behavior of children speaking MH can be cited as another 
piece of evidence for the claim that the abstract analysis is inadequate for 
this dialect of Hebrew. 

Verbs whose third radical is historical /h/ or 1^ I may surface in 
children's speech with or without the preceding [a]. On the other hand, there 
are many cases where overgeneralization of the rule takes place and it applies 
in verbs whose third radical is historical /?/ or /k/ (remember that histor- 
ical /k/ as a third radical always surfaces as [x] due to spirantization) . 
There does not seem to be a general principle governing the direction of the 
process. Bar Adon (1959) concludes that "roots having as their third radical 
historical /k/ and /h/ got so mixed up and it is rather difficult to distin- 
guish between them. Analogy operates in both directions." He cites the fol- 
lowing variations (radicals indicated in accordance with the historical pro- 
nunciation) . 

'he hides' (H-B-?) 
'he froze' (K-P-?) 
'he invented' (M-C-?) 

'he paints' (C-B- ) 
'he swallows' (B-L-*^) 

and similarly 

(26) linsox -v linsoax 'to bite' (N-S-K) 
lispoax -vlispox 'to spill' (S-P-K) 

Why is it that the variability occurs precisely with the presumed ab- 
stract segments? 

An abstract analysis can "explain" this by invoking a principle which 
states that the acquisition of abstract segments is inherently more diffi- 
cult and thus will be accomplished only at a later stage. Given this ap- 
proach, the specific developments which have taken place in MH are to be 
attributed to children's unfamiliarity with the abstract segments. Once 
those segments are acquired, however, one would have probably anticipated 
the elimination of variability and confusion. If this were indeed the case, 
major revision in adults' grammar should have been expected. Namely, the 
distinction between infinitives for verbs with first radical /?/ and /*^/ 


maxbi '^ maxbia 

ikpi % ikpia 

imci '\' imcia 

as well as 

cove ^ cove a 

bole -v bolea 


would reemerge; Vowel Lowering would become again an obligatory rule. How- 
ever, we do not find such revisions taking place. 

4.4 Conclusions 

Each of the separate developments cited above can be accounted for in 
some way within an abstract approach, and thus when treated individually 
they cannot be accepted as conclusive arguments against abstractness. Yet, 
when viewed in combination, these developments constitute strong circum- 
stantial evidence against the abstract analysis. 

Notice that the abstract analysis has been rejected not only on the 
basis of axiomatic beliefs such as "all phonological abstractness in the 
form of imaginary or abstract segments must be ruled out" (as a consequence 
of the True Generalization Condition, see Hooper 1976), or because we "take 
it as self-evident that an acquisition theory faces an extraordinary burden 
if it is required to account for the learning of the existence and locus of 
occurrences of phonomes that do not exist as perceptual realities anywhere 
in the language" (Brian 1974). Rather, it has been rejected since it was 
proven to be inadequate on internal grounds . 3.^ 

The issue of the behavior of historical gutturals in MH has to be 
viewed given the background of the revival of the Hebrew language. MH has 
been revived on the basis of Biblical and post-Biblical texts. The Biblical 
corpus reflects a certain phonemic system in which gutturals functioned as 
a distinct group and conditioned phonological rules which were phonetically 
perfectly motivated (Vowel Lowering, for example). The phonemic system of 
MH is not identical to that of Biblical Hebrew; mergers have taken place 
and, as a result, phonemic contrasts, on which the application of those 
phonological rules depended crucially, have been neutralized. Yet due to 
the relatively short period of time (MH was revived at the beginning of 
this century) as well as the normativistic pressure exerted by the school 
system, many surface systematicies, which are rooted in Biblical Hebrew, 
are still maintained. 

Some of the surface systematicies can no longer be treated as the out- 
come of phonological rules, which recapitulate the historical ones. Rather, 
some instances have been reanalyzed as morphologically conditioned, while 
others must be treated as involving 'exception features'. 

Given the above mentioned arguments one must conclude that the fact 
that adult speakers of MH still maintain the a only when the final radical 
was a historical 1^ I , /h/ or /h/ cannot possibly be accounted for by a 
synchronic rule, which reflects exactly the historical one, since such a 
rule must necessarily be sensitive to phonemic contrasts which cannot be 
posited underlyingly in the synchronic system. 

As an alternative one might consider an analysis according to which 
morphemes are marked in an arbitrary fashion by a diacritic which will 
trigger the a-Insertion Rule. The rule itself doesQnot recapitulate the 
historical rule; it no longer makes reference to [+low] as the conditioning 


environment. The question whether a certain morpheme would have an allo- 
morph in which the a surfaces is treated under this analysis as an idiosyn- 
cratic property of this morpheme, and is no longer entirely dependent on its 
phonological make-up. In other words: this fact has to be memorized by the 

To test the adequacy of this analysis two productivity tests were con- 
ducted. Before presenting the results, the relevance of such tests for the 
issue involved will be discussed. 


Kiparsky (1974) suggests that productivity should be viewed as a gra- 
dient phenomenon rather than a clear-cut dichotomy between productive and 
non-productive. On one side of the scale there are alternations which are 
fully productive in the sense that they extend to every new formation that 
meets their structural description. Such a productivity pattern would be 
taken as an indication for the extence of a major rule. On the other end 
one finds the traditional suppletive patterns of the type go - went , is- are ; 
these would never be productive. 

In between are alternations which are only partially productive: new 
words may but do not necessarily have to become subject to them. Berko 
(1958) found for example that 40% of the adults in her experiment when pre- 
sented with the word * heaf gave *heaves as the plural rather than * heaf s . 
She remarks: 

"l-rtien a small group of common words exist as a category by virtue of 
their great phonetic similarity and their morphological consistency, a new 
word having the same degree of phonetic similarity may be treated according 
to this special rule." 

These would be cases characterized in standard generative phonology by 
minor rules (or ad hoc subcategorization) . In the productivity pattern of 
those cases one does find variability among speakers, as well as inconsis- 
tent behavior within one speaker, who might apply the rule only in some 
cases but fail to do it in others, even though they meet the same phonologi- 
cal conditions. 

This behavior follows from the fact that the application of those rules 
depends not only on certain phonological configurations but is determined 
also by an arbitrary marking, whose existence is unpredictable. In cases 
involving minor rules and ad hoc subcategorization there will always be two 
possible outputs, each exhibited by some actual morphemes in the lexicon. 
When confronted with a new formation, which meets the structural description 
of the rule, the speaker has to decide whether to mark it as [+] or [-] the 
rule. This is an arbitrary decision, since nothing in the phonological shape 
of the new morpheme will dictate the 'correct' output. All the speaker can 
rely on is his previous acquaintance with lexical items which seem to struc- 
turally match the new one. Whether or not the rule will apply to the new 
formation depends on the pattern of alternation manifested by the actual 
lexical item selected as a model of reference. The factors determining the 
association of a new item with a specific model are intricate and may vary 


from speaker to speaker. All this predicts that the productivity of such 
pattern may be random rather than systematic (see Ohala 1973). 

If by psychological validity of a rule we mean 'perception of a certain 
recurring pattern', then rules can be psychologically real without necessar- 
ily being extended to all new formations which meet their structural descrip- 
tion. The variability factor, mentioned earlier with respect to rules in- 
volving ad hoc marking refers only to the question of applicability of a cer- 
tain rule; once it applies the output is totally predictable. This pattern 
of productivity clearly differs from that of the traditional suppletive 
cases: those involve a pattern of alternation which is manifested in one 
lexical item (is- are , go-went ) or by a small class of morphemes which lack 
a common phonetic denominator ( ox-oxen , child-children ) , and would hardly 
ever be productive. 

It will be assumed therefore that the degree of productivity of phono- 
logical rules should be taken not only as a verification of its psychologi- 
cal validity but also as an external manifestation of its most adequate 
characterization in the speaker's grammar. 

The subcategorization analysis suggested for the Hebrew data makes the 
claim that the application of the rule involves arbitrariness. This ap- 
proach predicts that in productivity tests variability would be expected. 
If on the other hand, a systematic behavior will be found, this would be un- 
explainable by the proposed analysis and, consequently, would point to its 


The subjects were 20 native speakers of Hebrew (10 male and 10 female). 
They were selected from among the Israeli students in Urbana-Champaign who 
agreed to participate in the experiment. All of them had high school educa- 
tion at least; 12 had academic education as well in either science or engi- 
neering, i.e. in areas that do not relate directly to linguistics in general 
or to Hebrew language in particular. 

The experiment involved the formation of present tense forms from the 
7 hypothetical past tense verbs. The three root radicals of each of those 
verbs do not occur in this combination in any actual lexical item. 

Each subject was tested individually. The items were presented to the 
subjects wholly auditorily, one at a time. The two sample items ( ^amarti - 
'I guarded' and katavti - 'I wrote') as well as the two practice items 
( gamarti - 'I finished' and zaxarti 'I remembered') were 'neutral' in the 
sense that they did not involve the application of the a-Insertion Rule, 
Thus the subjects were not deliberately biased in favor of any of the two 
possible outputs. The subjects were told that the purpose of the experiment 
was to gather information about the ability of Hebrew speakers to use new 
verbs. Two lexical items which do not meet the conditions for the a-Insertion 
Rule, were included to divert the subjects' attention from the real purpose 
of the experiment. 

Table 1 represents the pattern of application of the a-Insertion Rule 
for each of the relevant items. 


Table 1 
Responses in the Hypothetical Verbs Test 

a Insertion 

a Insertion 












Notice that even though there were 20 subjects, the total responses for each 
stimulus was 22; this fact will be explained momentarily. 

Table 2 represents the pattern of the rule application for each sub- 
ject (the letters refer back to the stimulus -verbs in Table 1). 

Table 2 
ittern of Rule Application for Each Subject 

No. of Subjects 

a Insertion 

- a Insertion 

a, d, e, f, 

a, f 

a, e, f 

a, d, e , f , 

a, d, e, f, 

a, d, f 

d, e, g 

d, g 

a, d, e, f, 

A high degree of variability was manifested in the responses of the 
group as a whole (Table 1). Furthermore, for half of the subjects there 
was what seems like a totally inconsistent behavior with regards to the 
application of the rule; they applied it only for a few items even though 
all the five hypothetical verbs met the structural description of the rule 
(Table 2) . 

A possible explanation for the variability is that the subjects were 
analogizing to lexical items they already had in their vocabulary. They 


were aware of the possible patterns but their choice of each was arbitrary. 
They might have referred to the pattern of alternation which is manifested 
by the actual item which was picked as a model. Reference to macati 'I 
found' or masaxti 'I pulled' which have present tense forms moce and mo sex 
respectively, would lead to an output such as gone and no lex (i.e. repre- 
sented as [-'a' insertion]). If on the other hand samati 'I heard' or 
baraxti 'I ran away' serve as a model, the outputs would be gonea and noleax 
respectively by analogy to the present tense forms Cornea 'he hears' and 
boreax 'he runs away' . Since any of these actual lexical items could serve 
as an adequate model, no wonder that two of the subjects have come up with 
two possible outputs in each case (that is how we got a total of 22 respon- 
ses for each stimulus rather than 20). 

The results so far seem to support the ad hoc subcategorization 


The subjects were the same as in the first experiment. 

The second experiment involved the creations of non-existing exten- 
sions to actual lexical item (such forms are referred to as 'lexical gaps'). 
One of the characteristics of Hebrew (and other Semitic languages) is the 
fact that nouns and verbs which are semantically related are usually derived 
from the same root, i.e. share the consonantal element. Whereas in English 
one 'draws a picture' and 'wears a sock' in Hebrew he mCaYeR CiYuR and GoReV 
GeReV respectively (a similar case in English is that of cognate objects, 
i.e. dream a dream). 

A list of 7 Hebrew nouns which do not have verbs derived from the same 
root as actual lexical items, was compiled and the subjects were asked to 
come up with related verbs (see Appendix A) . 

The stimulus-nouns were presented orally one at a time. To facilitate 
the subjects' task, examples of actual nouns and verbs which manifested the 
derivational relationships and were semantically close to the stimulus nouns 
were provided. For example: 

'To wear a sock' is ligrov gerev . 
'To wear a belt' is laxgor xagora . 

How would you say: 

He wears a hat? 'a hat' is kova 

He puts on a blanket? 'a blanket' is smixa 

On the basis of the actual lexical items, which manifest semantic and 
morphological relationships, and were provided as a model, the subjects were 
expected to come up with hypothetical verbs which would be related to the 
stimulus noun in the same fashion (i.e. be derived from the same root). 

In the case of masexa and zxuxit , for example, the final radical is 
realized as [x] ([a] and [it] are suffixes). Since in the infinitive, or 


present tense singular masculine, this radical occurs in final position, the 
structural conditions for the a to be inserted would be met. 

The results are indicated in Table 3. 

Table 3 
Responses in the Lexical Gaps Test 


a Insertion 

a Insertion 


kova hat 

smixa 'blanket' 

masexa 'mask' 

calaxat 'plate' 

garin 'seed' 

emca 'middle' 

zxuxit 'glass' 

20 kovea 



^-^ gorea-4 

20 mamcia 

20 somex 
moxe s - 8 



20 mezaxex-9 

All of the subjects behave identically with respect to the application 
of the rule (even though the actual outputs differed depending on the choice 
of the verb paradigm).' At first, it might seem that the results actually 
support the abstract analysis, which has been earlier rejected (cf. 4). 
Thus, one might argue that the subjects' systematic behavior was due to the 
fact that the identical surface segments differed underlyingly, some of them 
being pharyngeals and that this is why they conditioned a-Insertion. Since 
the stimulus-nouns were everyday words, speakers have probably had the oppor- 
tunity to establish these crucial underlying contrasts. 

Beside the theoretical arguments and empirical evidence against the va- 
lidity of an abstract analysis mentioned above, there is an additional reason 
why such an approach cannot possibly provide a satisfactory explanation in 
the case of the specific lexical items included in the second experiment. The 
recoverability of abstract segments crucially depends on the availability of 
surface clues, namely the presumed abstract segment must behave in a unique 
way which is not predicted by its surface properties. Its behavior must pro- 
vide clear indications that its underlying features are indeed distinct from 
its phonetic makeup. 

The stimulus -nouns in the second experiment, however, do not have deriva- 
tives in which the third radical of the root occurs in an appropriate 


environment to either undergo or condition any rule which could identify it 
as being underlyingly something different from its surface appearance. Given 
the non-existence of any clues which could have enabled speakers to differ- 
entiate between identical surface segments, it must be concluded, that an 
abstract approach, even if sometimes it can be motivated for handling data 
involving alternations, cannot possibly offer a plausible explanation for 
the subjects' systematic behavior in the second experiment. 

The other alternative, namely the arbitrary marking analysis, cannot 
account for the results either. It is evident that speakers do 'know' more 
about the pattern involved than what is predicted by such an analysis. If 
we hypothesize that the subjects were merely analogizing to existing lexical 
items, we would have expected in such a case to find a great degree of vari- 
ability, as was found in the first test. A stimulus-noun such as smixa 
'blanket', for example, could be analogized on the basis of phonetic simi- 
larity, either to cmixa 'growing', slixa 'pardon', 'forgiveness' (which have 
corresponding verbs come ax and soleax ) or to tmixa 'support' nsixa 'a bite' 
(whose related verbs are tomex and nofeex ) . If the hypothetical verb 'to put 
on a blanket' was indeed derived by analogy, one would have expected both 
somex and someax as possible outputs; only the first was, however, suggested 
by all the subjects. 

For other stimuli it was rather difficult to find an actual lexical 
item, which could serve as a model. For zxuxi t a completely matching mor- 
pheme does not exist in the whole lexicon of MH. Under such conditions one 
might have expected the subjects simply to guess the 'correct' output, and 
as a result to exhibit variability in their responses. Instead, a totally 
systematic pattern of productivity was found. 

A possible explanation might be that speakers were governed in their 
responses by the tendency to avoid homophony. Thus, since lehamci 'invent' 
is an already existing item, lehamcia was proposed for 'put in the middle'. 
This hypothesis, however, can be easily refuted: whereas kovea and somex 
are actual lexical items (meaning 'decide' and 'rely on' respectively) the 
other theoretically possible outputs, namely, kove and someax do not occur 
in the lexicon of Hebrew, yet all of the subjects suggested the former 
rather than latter. 

The only plausible way to account for this systematic linguistic be- 
havior is to conclude, that the subjects were not operating by analogy to 
subjective models, but rather all of them were applying the same rule: what 
conditions the application of this rule is, however, a mystery since it can- 
not be attributed to any surface or underlying difference, as has been demon- 
strated earlier. 

We seem to have arrived at a paradox; earlier it has been argued, that 
a speaker of MH could not have formulated a major rule to account for the 
forms where the a surfaces. On the basis of the second test we have come, 
however, to an opposite conclusion, namely, that speakers do operate with 
a very general rule. 



Both tests were conducted orally. The second one, however, involved 
stimuli which were every-day words whose spelling was obviously known to 
all the subjects. 

The outputs suggested in the second test reveal the following pattern: 

The a surfaces only when the third radical is spelled with the Hebrew 
letters 'xet'n and 'ayin' y but never where it is spelled with 'kaf J) 
and 'alef ^ . In other words, it was inserted only in words that were 
spelled with letters corresponding to the historical /^ / and /h/. Thus, 
the phonemic contrasts between /?/ : /^ / , /h/ : /x/, which have been abso- 
lutely neutralized in MH, but are still kept in the modern orthography, 
serve as clues for the modern speaker. 

The first test involved non-existing lexical items. Given that their 
orthographic representations were unknown and could not have been utilized 
for the purposes of the application of the rule, all the subjects could do 
was operate by analogy: that is what caused the variability. 

To test the validity of this hypothesis an additional test was con- 


The experiment was conducted with the same subjects as in the first 
and second tests. 

The task and material were identical to those of Experiment I, i.e. 
to form present tense forms of hypothetical past tense verbs. 

Each of the hypothetical verbs (cf. 7) was presented orthographically 
and each subject was asked to pronounce its present tense form. The third 
radical was indicated half of the time by ;? or r\ and half of the time as X 
or V. The experiment took place a month after the first one. 

The results clearly confirm the hjrpothesis about the use of ortho- 
graphic clues for the purposes of application of the rule; the [a] surfaced 
only when the third radical was spelled with n or V but never when it was 
represented as 3 or X . 

The results of the third test can be cited as additional argument 
against the validity of an abstract analysis. Such an analysis clearly 
fails to account for the differences in speakers' behavior in these three 

A proponent of the abstract approach might argue that the randomness 
in the first test has to be attributed to the difficulties in the recover- 
ability of the abstract segments. He would probably try to account for the 
systematic behavior in the second test by claiming that the lexical repre- 
sentations of these familiar items contain a pharyngeal. Why is it, one 
must ask, that the responses in the third test differ significantly from 
those of the first one, although both included the same hypothetical lexical 


items? If the systematic behavior in the third test was due to the recover- 
ability of the abstract segments what made this process suddenly possible? 
Obviously it was not any new phonetic information, since the test was a 
written one. The only difference between the first and third test was the 
availability of orthographic representations in the latter. Given this 
fact, even an advocate of an abstract analysis would have to admit that the 
orthographic element must be treated as a relevant factor affecting speakers' 
linguistic behavior. 


The orthography-based analysis makes the prediction that in the case 
of preliterate speakers, variability would be found in the second test (the 
one involving lexical gaps) as well. No significant differences should be 
expected in their performance in the first and second tests, since the only 
factor distinguishing between the stimuli in these two tests, namely the 
orthographic one, would be irrelevant in their case. 

To verify this hypothesis the second test was conducted with 15 first 
graders in an Israeli elementary school. The results are represented in 
Table 4 (for comparison, the adults' responses to the same stimuli are 
indicated as well) . 

Table 4 

Responses of Adults and First Graders 

in the Lexical Gaps Test 

First Graders 



+ a Insertion 

- a Insertion 

+ a Insertion 

- a Insertion 


7 (46.7%) 

8 (53.3%) 

20 (100%) 



6 (407o) 

9 (60%) 


20 (100%) 

ma sex 

8 (53.37,) 

7 (46.7%) 


20 (100%) 


9 (60%) 

6 (40%) 


20 (100%) 


10 (66%) 

5 (33.3%) 

20 (100%) 



9 (60%) 

6 (40%) 

20 (100%) 


Nine of the first graders manifested a 'systematic behavior' in the 
sense that they treated all of the items alike: 4 always inserted the vowel 
a, the other 5 never did. There were only 2 out of the 15 whose responses 
were identical to those of the adults. 

The uniformity in the adults' performance versus the variability in 


the first graders' responses are rather striking. They clearly confirm the 
hypothesis that the orthographic pattern serves as the crucial clue for the 
application of the rule. 


It seems that as consequence of familiarity with the spelling of 
lexical items, speakers might realize the correlation between certain ortho- 
graphic representations (i.e. letters) and a pattern of morphophonemic al- 
ternation. Based on the contrast which is still reflected in the orthog- 
raphy the modern speaker is able to formulate a generalization which is 
rather close to the historical a-Insertion Rule; as has been argued earlier, 
this rule could not have been recovered independently from the available 
surface data. 

The exposure to orthography might thus lead to modification of the 
speaker's grammar; patterns of alternations which synchronically involve 
arbitrariness (i.e. marking with a diacritic) may be reinterpreted as totally 
predictable when orthographic clues are available. For an educated speaker 
of Hebrew the contrast between kore 'he reads' and korea 'he tears' is no 
longer arbitrary; it is tied up with the orthographic distinction ( /\ versus 
y) which reflects the historic phonemic contrast. 

One might suggest that the spelling pattern serves in this case as a 
diacritic, i.e. as an arbitrary symbol which triggers the application of the 

It seems desirable to avoid using the term 'diacritic' in this context 
because of its connotations; when used within the current linguistic theory 
it implies an element which is external and totally ad hoc. Even though 
orthographic representations are arbitrary to a certain extent, they clearly 
have psychological reality for educated adults and cannot possibly be re- 
garded as of the same status as a mere diacritic of the type [+rule X] 
(+rule 235] . One might propose that since orthography is external to the 
phonological system it should still be referred to as a diacritic. Such a 
definition is valid only within the basic theoretical assumptions of modern 
linguistics. The orthographic factor can be treated as external only be- 
cause the domain of linguistics has been pre-defined as the spoken variety 
of language alone. Such a division seems to be unmotivated. Rather, the 
orthographic system should be incorporated and treated as relevant linguis- 
tic data, since it clearly has its effect on the formulation of the speaker's 

Within a model which incorporates orthography as legitimate linguistic 
data available to speakers and utilized by them, the a-Insertion Rule can 
no longer be viewed as triggered by a totally ad hoc marking. It can be 
claimed that for literate adults the rule has actually become a major one: 
it applies whenever its structural description (i.e. a specific grapheme) 
is met. 

The version of the synchronic a-Insertion Rule suggested here, however, 
is not a regular phonological rule. There is no empirical evidence to sup- 
port the hypothesis that spelling has indeed affected the speaker's grammar 


to such an extent as to introduce new phonemic segments (i.e. [+low] ones). 
Rather the application of the rule crucially depends on orthographic clues. 

The best that can be done is to formulate the synchronic rule by incor- 
porating non-phonological information into its structural description, i.e.: 


+ syll 
+ low 



Or, if one accepts the position that historical / / are postulated synchroni- 
cally as 111 whenever they serve as one of the root radicals (see Bolozky 
1972), the a-Insertion Rule can be formulated as follows: 


■+ syir 
+ low 




Such a formulation implies that for the purposes of the rule applica- 
tion, literate speakers of MH rely on orthographic distinctions rather than 
on phonemic contrasts. 

In spite of all these unconventional properties of the rule, it has 
been clearly demonstrated above that the new formulation reflects rather 
accurately the speaker's linguistic knowledge. The results of the second 
test indicate that the orthographically based major rule, rather than the 
version involving arbitrary marking, is the psychologically valid analysis. 
Ignoring the relevance of orthography might lead to a distorted picture of 
what the literate speaker's competence actually is. 

The influence of the orthographic system must be viewed as potential 
rather than automatic. Not every historical rule which is potentially re- 
coverable from the modern orthography will necessarily become part of 
speaker's grammar: whereas speakers of Hebrew do perceive the historical 
correlation between the spelling pattern and the application of the a- 
Insertion Rule, other rules which depended originally on the same phonemic 
contrasts, and are reflected in the modern orthography as well, have not 
been maintained with their historical environments but are memorized as 
arbitrary alternations.^ This fact does not, however, invalidate the claim 
made here, namely that historical information may sometimes reach syn- 
chrony due to its recoverability from the orthographic system. 

To sum up; a common criticism directed towards phonological analyses 
involving 'abstractness' is that they have lost psychological validity by 
overstating the speakers' ability to abstract generalizations from given 
surface data. This view is expressed, for example, by Braine (1974), who 
notes that the issue is not that such analyses "fail to capture generali- 
zations; rather, it is whether the native speakers make generalizations as 
the theory captures them." 


What I have suggested is that disregarding the relevance of ortho- 
graphic data in phonological analyses, results again, in a misrepresenta- 
tion of the speaker's knowledge; this time, however, by understating it; 
i.e. by failing to express certain valid generalizations. Analyses con- 
taining rules referring to contrasts which have been absolutely neutralized 
on the surface are generally considered as expensive if not automatically 
ruled out. The results mentioned above suggest, however, that historical 
rules whose synchronic formulation involves reference to segments which 
never surface may nevertheless be recapitulated and become an integral 
part of a speaker's system as long as the relevant historical distinctions 
are reflected in the orthographic system (assuming of course that we are 
dealing with a literate society). 


*The paper is a chapter from a dissertation entitled, "Some Effects 
of Orthography on the Linguistic Knowledge of Modern Hebrew Speakers." 
I would like to thank Charles Kisseberth, Hans Hock, and Michael Kenstowicz 
for the valuable comments and suggestions they provided. 

Historically, [x] was an allophone of /k/ derived via spirantization. 
Historical /k/'s which occur as the third radical, would always surface as 
[x]'s, because spirantization in Hebrew applied postvocalically, and there 
is never in the verb paradigm a configuration where a third radical occurs 
postconsonantally (a position where spirantization would not apply) . Thus 
these [x]'s never alternate with [k]'s. Whether this surface [x] has to be 
represented synchronically as a phoneme or still as an allophone derived 
from /k/ is irrelevant for the present purposes. All that matters is that 
there is no way to tell [x]'s which were historically /h/ from those which 
were allophones of /k/ in the third position: both are ' always realized on 
the surface as [x] . 

For the controversy about the abstractness of phonological analyses 
see Kiparsky (1968) , Kisseberth (1969) , and Crothers (1971) . 


Bolozky (1972) suggests that gutturals should be still postulated 
underlyingly primarily because of pattern correspondences. He argues "If 
one wants to account for awareness of the connection between root, pattern 
and semantic reading, and for semantic correspondences between realiza- 
tions of different verbs in the same pattern - then one is logically led 
to assume gutturals in the underlying representations. If we want yaavod 
'he will work', for instance, to be related to avad 'he worked' in the same 
way that yiksor 'he will tie' is related to ka^ar 'he tied' the most logi- 
cal way of doing it is to represent yaavod as /yi+'^vod/ and let 'a-insertion' 
for gutturals and assimilation across a prefix map it into ya ?avod (ulti- 
mately yaavod) ." 

Given, however, that those gutturals are hardly ever realized as such 
on the surface he suggests (p. 81) that "it is quite possible that instead 
of having 111 , 1^1 and /h/ it would be sufficient to fill the slot they 
used to occupy by /?/'s only, since it is almost only as [ ?] that they are 


optionally realized. ...Preserving /?/ in the ex-gutturals positions will 
not cause the loss of significant generalization, and would still satisfy 
the need for some additional consonants for pattern correspondences and 
for a plausible explanation of some phonological phenomena attributed to 
the gutturals. And since it is (almost) only /?/ that is optionally found 
in the phonetic output our underlying representation would not be too ab- 

The arguments in favor of the postulation of underlying /?/'s do not 
hold for /h/: in this case the problem of filling a slot does not arise, 
since there is always a surface segment in the place of the historical /h/, 
namely, [x] . Historical /h/ could thus be postulated as underlying /x/ and 
the pattern correspondences could still be maintained. 

Thus, the argument mentioned by Bolosky may motivate the postulation 
of an underlying /?/ as a 'slot filler': it cannot however be cited as evi- 
dence for postulating phonemic contrast between the members of the histor- 
ical guttural group (i.e. /?/ : / /) or for an underlying /h/. 

Another possible argument against abstractness involves rule ordering. 
If it can be established independently that the application of rule A which 
must refer to the surface properties of segment X precedes in the derivation 
rule B, which treats the same segment according to its presumed underlying 
features, this would obviously constitute strong evidence against postu- 
lating as an abstract segment. In many cases, however, abstract analyses 
recapitulate the historical developments: if the rule which has led to the 
neutralization of the historical underlying contrasts was the last to enter 
the language, and if rule reordering has not taken place, there will be no 
data which will enable us to construct this type of an argument. Further- 
more, such argument cannot be invoked if the relevant rules cannot be proven 
independently to be crucially ordered one with respect to the other. This 
is, for example, the case with the a-Insertion Rule, which is presumably 
triggered by /h/, and the Vowel Epenthesis Rule (i.e. the rule inserting a 
vowel to break'the impermissable sequences #c|>lowl, l+5owJ C , which no longer 
refers to historical /h/ as a [+low] consonant) . 

Notice, however, that an argument based on rule ordering can be used 
only negatively, namely to demonstrate that a certain abstract analysis can- 
not be maintained. The nonavailability of such argument does not, of course, 
constitute positive evidence in favor of an abstract analysis. 

Since the number of morphemes where the a surfaces is approximately 
equivalent to those which do not undergo the rule, it seems preferable to 
refer to the situation as 'ad hoc subcategorization' rather than a 'minor 
rule' (see Kenstowicz and Kisseberth 1977, p. 126) <, Essentially, both in- 
volve arbitrary marking of morphemes; the only difference is that in the 
case of minor rules one class can be identified as regular and unmarked « 

Instead of the subcategorization analysis the CaCa(x) - CoCea(x) 
alternation could be represented by suppletion; i.e. by listing both forms 
in the lexicon. In both analyses memorization is involved: one has either 
to memorize the fact that a certain morpheme is marked with a diacritic (a 
fact which is totally unpredictable) or two allomorphs of a certain morpheme 
have to be memorized (in a suppletive analysis) . 


The analysis involving subcategorization seems preferable for the fol- 
lowing reasons: 

a) Whereas a suppletive alternation (as defined traditionally) would 
hardly be expected to be productive (e.g. extend to new items) the alterna- 
tion of the type CaCa(x) - CoCea(x) is productive at least to a certain 
degree . 

b) Two aspects of predictability should be distinguished. One has to 
do with whether the application of the rule is predictable, whereas the 
second involves the pattern of alternation. In cases of ad hoc subcategori- 
zation what is unpredictable is whether a certain morpheme will manifest the 
alternation or not (i.e. whether the rule will apply). Once, however, the 
rule applies the pattern of alternation is totally predictable: if a mor- 
pheme of the type CaCa is marked we 'know' that the alternant will be of the 
configuration CoCea . From this second aspect it is as predictable as a 
major rule. For the traditional cases of suppletion (is- are , go-went), both 
their occurrence as well as their form (pattern of alternation) are totally 
unpredictable. The cases of the type CaCa(x) - CoCea(x) seem to be somewhere 
in between suppletive paradigms, on one hand, and totally predictable alterna- 
tions on the other. They share properties with both. Since they seem, how- 
ever, to differ essentially from clear cases of suppletion (in the sense that 
they are both productive and predictable at least to some extent) , I prefer 
the ad hoc subcategorization to the suppletive analysis. 

For the purposes of the present argument the question as to which of 
the two approaches is more adequate is not of crucial importance; both imply 
that for the native speaker of MH the alternations discussed involve arbi- 
trariness, as opposed to the abstract analysis, which claim that they are 
totally predictable. 

The various verb patterns of Hebrew (i.e. C^iC„eC , hiC^C iC , etc.) 
are generally associated with certain semantic properties such as causa- 
tivity, incohativeness, etc. The variability found in subjects' responses 
( licloax lecaleax, limsox lemasex) is, however, irrelevant for the purposes 
of the present issue; with respect to the application of the a-Insertion 
rule all subjects manifested a unanimous behavior. 


For example, the historical distinction betv?een infinitives of the 
type la^aCoC :le?eCoC , which depended on the contrast 111 : 1^1 is not 
maintained even though it is reflected in the orthography. 


BARaDON, a. (1959). Children's language in Israel. Unpublished Ph.D. 

thesis. University of Jerusalem (in Hebrew). 
BARKAI, Malachi (1972). Problems in the phonology of Israeli Hebrew. 

Unpublished Ph.D. thesis. University of Illinois, Urbana. 
BERKO, Joan (1958). The child's learning of English morphology. 

Word 14 150-177. 

BOLOZKY, Shmuel (1972). Categorical Limitations on Rules in the Phonology 

of Modern Hebrew. Unpublished Ph.D. thesis, University of Illinois, 

BRAINE , Martin D. S. (1974). On what might constitute a learnable phonology. 

Language 50 270-299. ^ 

BRAME, Michael (1972). On the abstractness of phonology: Maltese . In M. 

Brame (ed.). Contributions to generative phonology 22-61. University 

of Texas press. 
CROTHERS, J. (1971). On the abstractness controversy (POIA reports, 2nd 

series 12) Berkeley, Phonology Laboratory, Department of Linguistics. 
HOOPER, Joan (1976). An introduction to natural generative phonology. 

New York; Academic Press. 
KIPARSKY, Paul (1968). How abstract is phonology? Bloomington: Indiana 

University Linguistic Club. 
KIPARSKY, Paul (1971). Historical linguistics. In W. 0. Dingwall (ed.) . 

A survey of Linguistic Science 576-649. College Park: University of 

KIPARSKY, Paul (1975). What are phonological theories about? In David 

Cohen and Jessica Wirth (eds.). Testing Linguistic Hypotheses, 187- 

209. New York: Wiley. 
KISSEBERTH, C. W. (1969). On the abstractness of phonology: an evidence 

from Yawelmani, Papers in Linguistics, 1 248-282. 
OHALA, John (1973). On the design of phonological experiments. Monthly 

Internal Memorandvmi 25-62. Berkeley: Phonology Laboratory, Department 

of Linguistics. 
ORNAN, Uzzi (1972) . Ordered rules and the so called phonologization of 

ancient allophones in Israeli Hebrew. In Luigi Heilmann (ed.) 

Proceedings of the eleventh international congress of linguists. 

Bologna -Florence . 



a) To wear a sock is liGRoV GeReV. 
To wear a belt is 

(laXGoR XaGoRa) 

To put on a tie is 

(la?aNoV ?aNiVa) 

What would be similar constructions to convey 
the following ideas: 

He wears a hat. 

He puts on a blanket, 

b) To sculp a statue is leFaSeL PeSeL 
To draw a picture is 

How would you say: 
He makes a mask. 

He makes a plate. 

c) To take off the peel is leKaLeF KLiPa. 
How would you say: 

He removes the seeds. 


How would you say 

'to put in the middle 


How would you say 

'to make glass' . 

('a hat' is kova) 

('a blanket' is smixa) 

(leCaYeR CiYuR) 

('a mask' is masexa) 

('a plate' is calaxat) 

('a seed' is gar in) 
('middle' is emca) 
('glass' is zxuxit) 

Note: For readers who are not familiar with Hebrew, I have indicated 
in brackets the relevant Hebrew words. 

studies in the Linguistic Sciences 

Volimie 5", Number ± , i^pi'ing lyVO 

Peter Cole ard Wayne Harbert and Gabriella Hermon 

A wide variety of hypotheses regarding the underlying 
structure of relative clauses have been put forward in the 
generative literature. Among the more imaginative-and, at 
first glance, less plausible proposals-is that of Michael 
Brame, who suggested that restrictive relative clauses 
(with surface heads) are underlyingly headless (i.e., have 
dummy heads). Surface heads are formed by a rule which cop- 
ies the relativized NP from the embedded clause into the 
matrix clause. 

Brame 's proposal gains considerable plausibility in the 
light of recent work on relativization in Quechua. The p\ir- 
pose of this paper is to demonstrate the existence of surface 
headless relative clauses in three members of the Quechua 
language family: Ancash, Huanca and Imbabura. Evidence for 
headless structure is based on case marking, word order and 
the distribution of relative clause types. The results derive 
primarily from field work by the authors. The existence of 
surface headless relative clauses in Quechua has not been 
reported previously in the literature. 

The appearance in surface structure of just that struc- 
ture proposed by Brame as underlying restrictive relative 
clauses generally constitutes significant support for Brame' s 
proposal. It suggests that Brame 's analysis of English de- 
serves new and serious consideration. 

The purpose of this paper is to argue for the existence of 
"headless" relative clauses in three dialects of Quechua: Imbabura 
(a Quechua II dialect spoken in northern Ecuador), Ancash and Huanca 
(both Quechua I dialects spoken in central Peru).l The existence of 
such relative clauses in Quechua is of considerable theoretical import- 
ance because it lends support to a recent claim that relative clauses 
are headless underlyingly even in languages like English where heads 
invariably occur in surface structure. 

We shall begin by examining briefly the motivation for positing 
underlyingly headless relative clauses in English. We shall then turn 
to Quechua and show that there is compelling evidence for the existence 
of sxirface headless relative clauses in each of the above-mentioned 
dialects. 2 The clear existence of headless relative clauses in Quechua 
provides important support for the proposal that relative clauses are 
underlyingly headless cross-linguistically. ^ 

Headless Relative Clauses in English 

Recent studies of relativization in English suggest that English 
relative clauses are headless in underlying structure.^ According to 


this hypothesis the underlying structure of (l) would be roughly (2) 
(in which there is a dummy head5): 

the £^ we made headway 

A rule of promotion copies the object NP into head position, yielding (3): 

(3) m 

det NP 

the headway that we 


The evidence for such a derivation in English is quite convincing. 
Heads of relative clauses behave syntactically (with respect to a var- 
iety of rules and constraints ) as though they were constituents of the 
embedded clause (S.) rather than the matrix clause. Such behavior is 
expected if they originate in the embedded clause and are later promoted 
to constituency in the matrix clause. But it is unexpected and would 
constitute an irregularity in the grammar if heads were matrix clause 
constituents in (both surface and) underlying structiire.T 

Let us examine an example of a head behaving as though it were an 
original constituent of (S.). Consider relative clauses on the head 
noun headway , which in environments other than relative clause construc- 
tions can appear only as the object of make , as is illustrated in (k) 
and (5): 

(k) We made satisfactory headway 

(5) *The headway was satisfactory 

When headway is modified by a relative clause, however, the restric- 
tions on its distribution are waived. Compare the \mgrammatical (5) 
with the grammatical (6): 

(6) The headway Cwhich we madeD was satisfactory 

The grammaticality of (6) is not predicted if we assume that (l) derives 
from an underlying structure in which the head originates outside the 
embedded S. . But if (l) derives from a structure similar to (2), the 
grammaticality of (6) is predictable. Headway is gra mma tical as a head 


NP in (l) according to that analysis because it originates as the ob- 
ject of make . Schachter provides a variety of arguments of similar form 
shoving that relative heads with pronouns , picture noun reflexives and 
reciprocals all behave as though they originate vithin the relative 
clauses rather than as heads. 

If relative clauses with heads derive universally from headless 
structures (i.e., structures with dummy heads), it would seem likely 
that there would be languages in which relative clauses are headless 
in surface structure. 8 We shall now turn to evidence that various dia- 
lects of Quechua provide instances of such languages. 

We shall first consider relativization in Imbabura. Before pre- 
senting the argtunents for headless relative clauses in Imbabxira, it 
will be helpful to review some general facts about relativization in this 
dialect. In Imbabura, as in other varieties of the Quechua language 
family, relative clauses, and subordinate clauses generally, appear in 
norainalized form. The choice of nominalizer in Imbabura is determined 
by the temporal relationship of the relative clause and the matrix cla\ise. 
This is illustrated in (T)-(9): 

(T) Relative Clause with Present Nominalizer 

C0. punu-ju-jJ wawa. mana cai-pi-chu 

Csleep-prog-p^resent nomD child not this-in-neg 
'The child who is sleeping is not here.' 

(8) Relative Clause with Past Nominalizer 

C0. pxinu-shcaU wawa. mana cai-pi-chu 
C sleep-past nomD child not this-in-neg 
'The child who was sleeping is not here.' 

(9) Relative Clause with Future Nominalizer 

^0- pufiu-na^ wawa. mana cai-pi-chu 

C sleep-future nomH child not this-in-neg 
'The child who will sleep is not here.' 

In contrast to other dialects, when a head is present, the choice 
of nominalizer does not depend on the grammatical role (subject, direct 
object, etc.) of the relativized noun phrase. ■'■'-' For instance, - shea 
may be used in the relativization of transitive subjects and - j_ may be 
used in the relativization of direct objects. This is illustrated by 
(10) and (11): 

(10) Relativization of Transitive Subject with - shea 

110. warmi-ta Juya-shcaD runa. aicha-ta micu-Ju-n 
C woman-acc love-past nomD man meat-acc eat-prog-3 
'The man who loved the woman is eating meat.' 


(ll) Relativization of Direct Object vith -j_ 

Cruna 0. juya-j3 wa.niii. aicha-ta micu-ju-n 

Cman ■"■ love-present noin] voman meat-acc eat-prog-3 
'The woman whom the maxi loves is eating meat.' 

As would be expected in an SOV language, in relative clauses where 
a head is clearly present, it appears to the right of the embedded clause, 
this is represented schematically in (12): 

^-■■^^ NP^ S."- ^S. NP^ ^NP ^NP 
1 1 

It should be noted, furthermore, that word order in relative clauses, 
in contrast with matrix clauses, is strictly SOV. Compare (l3a) and 
(l3b) with (lUa) and (lUb): 

(l3) Word Order in Relative Clauses 

(a) SOV Order 

110. aicha-ta micu-jD warmi. 

C meat-acc eat-present nomH woman 

*(b) SVO Order 

C0. mdcu-j aicha-tal warml . 

L eat-present nom meat-acc H woman 

'A woman who eats meat' 

(lU) Word Order in Matrix Clauses 

(a) SOV Order 

warmi-ca aicha-ta micu-n 
woman-topic meat-acc eat-3 
'The woman eats meat.' 

(b) SVO Order 

warmi-ca mlcu-n aicha-ta 
woman-topic eat-3 meat-acc 
'The woman eats meat.' 

Case marking inside relative clauses follows the same principles 
that operate in main clauses. Subjects receive zero case marking, direct 
objects are marked with the accusative case marker - ta (or its allomorph 
-da after nasals). Case marking applies within the relative clause 
on the basis of grammatical relations within that clause. In addition, 
the relative clause as a whole receives case marking on the basis of 
its grajmnatical role in the matrix clause. These processes are illus- 
trated in (15): 



(15) parla-rca-ni C0. runa-ta ricu-shca3 warmi-wan 
speak-past-lsg L man-acc see-past noml woman-vith 
'I spoke with the woman who saw the man.' 

In (15) runa 'man' receives accusative case marking because it is 
the direct object within the embedded clause. The relative clause as a 
whole is marked with the comitative postposition - wan , the appearance 
of which is governed by the matrix verb parla 'speak'. Case marking within 
the relative clause differs from matrix case marking in only one way. 
The accusative case marker - ta is optional inside relative clauses though 
it is obligatory in natrix clauses, as shown by (l6a) , (l6b) and (l7): 

(16) .-ta Suppression in Relative Clauses 

(a) Unsuppressed - ta 

C0. wagra-ta michi-jD wawa. 
C cow-acc herd-pres nomH child 
' child who herds cows ' 

(b) Suppressed - ta 

C0. wagra michi-jD wawa. 
C cow herd-pres nomD child 
'child who herds cows' 

(it) Absence of - ta Suppression in Matrix Clauses 

*wagra . , . 
wawa-ca _^ michi-n 


1. •-, J ^ • *cow-0 , , _ 
child-topic herd-3 


'The child herds cows.' 

Ve shall now turn to headless relative clauses. The clearest and 
most straightforward examples of headless relative clauses in Imbabura 
are in sentences like those of (18): 

(18) Headless Relativization of Embedded Direct Objects 

(a) Cwambra wagra-ta randi-shcaD wanu-rca 
Cboy cow-acc buy-past nomD die-3 past 
'The cow which the boy bought died.' 

(b) Cruna alcu-ta jatu-shca3 ali-mi 

Cman dog-acc sell-past nomU good-validator 
'The dog which the man sold is good.' 

There are two clear cut arguments that these relative clauses are 
in fact headless. H Both arguments are based on a fundamental struc- 
tural difference between relative clauses vith heads and those without 
heads. Cross-linguistically, in those relative clauses with heads, 
the head is , in derived structure at least , a constituent of the matrix 


rather than the embedded clause. It appears either immediately to the 
left or immediately to the right of the embedded clause, the position 
typically correlating with "basic word order in the language in question 
(to the left of the embedded clause in VO languages and to the right 
in OV languages). In contrast, in headless relative clauses, unless rules 
changing its position apply, the relativized NF appears in its normal 
position within the relative clause. It is treated by case marking and 
other rules coding grammatical role as a constituent of the embedded 
rather than matrix clause. 

The first argument that the relative clause constructions in (l8) 
are headlessHhas" to do with word order. The \inderlined NPs in (l8) 
appear in the middle of the embedded clause rather than to the left or 
right, as would be expected if they were heads. Compare (l8) with (19): 

(19) Relativization of Embedded Direct Object (with Head) 

(a) Cwambra 0. randi-shcall wagra. wanu-rca 
Cboy buy -past nom II cow die-3 past 
'The cow the boy bought died.' 

(b) Cruna 0. jatu-shcaD alcu. ali-mi 

Cman ■"" sell-past nomD dog good- validator 
'The dog the man sold is good.' 

In (18) the relativized NP appears in the normal position for a direct 
object within a relative clause, that is to say, between the subject 
and verb. In contrast, in (19) the underlined NP appears to the right 
of the embedded verb. The differences in order between (18) and (19) 
are unexceptional if the sentences of (I8) involve relative clauses 
without heads and those of (19) with heads. 

The second argument is based on case marking. In (18) the underlined 
NP receives accusative case marking. This would be expected on the basis 
of the case marking process described previously if that noun phrase is 
a constituent of the embedded clause, but would be imexpected if it were 
a head, which is a main clause, constituent . 

Thus, we conclude that the sentences of (18) illustrate headless 
relative clauses. 

Having established the existence of headless relative clauses in 
Imbabura, we would now like to examine briefly the distribution of the 
construction. In the preceding examples we showed the headlessness of 
certain instances of relativization where the relativized NP is a direct 
object within the embedded clause and the past nominalizer - shea is 
employed. This type of relative clause is represented schematically in 

(20) , 

It should be noted that the sentences of (18) are unambigous with 
regard to whether the subject or the object has been relativized. 



These sentences cannot be interpreted as involving relativization of 
the subject I^. That is, (l8a) cannot mean 'The boy who bought the cov 
died.'. Nor can (l8b) mean, 'The man who sold the dog is good.'. Head- 
less relativization with the nominalizer - shea is restricted to the 
relativization of direct objects. -^2 

Thus, in headless relative clauses, unlike those with heads, the 
nominalizing suffix indicates both tense and the grammatical role of 
the relativized NP. Such a mixture of functions is not unusual in Quechua 
generally, but as we showed previously, it is not found in Imbabura in 
relative clauses with heads. We shall employ the difference in function 
of the nominalizing suffixes in relative clauses with and without heads 
to help us determine whether certain other relative clauses have or do 
not have heads. The difference in function of the nominalizer should 
be particiilarly helpful in examples where case marking and word order 
fail to distinguish whether a relative clause is or is not headless. 

Let us consider the examples in (2l): 

(21) Headless Relativization of Embedded Subjects 

(a) C wamhra wagra-ta randi-jD waSu-nga caya 

C boy cow-acc b\ay-pres nomU die-fut\ire 3 tomorrow 
'The boy who is buying the cow will die tomorrow.' 

(b) C runa alcu-ta jatu-j] ali-mi 

[man dog-acc sell-pres nomH good-validator 
'The man who is selling the dog is good.' 

The sentences of (2l) are similar to those of (l8) in that they must be 
interpreted as involving the relativization of a single grammatical role, 
subject, in this case. Thus, (21a) cannot constitute a statement that 
the cow will die tomorrow nor (21b) that the dog is good. The sentences 
of (21) differ from those of (l8) in that in (l8) the relativized NP is 
the embedded direct object while in (2l) it is the embedded subject. We 
shall argue that these sentences, like those of (l8), are instances of 
headless relative clauses (as indicated by the bracketing and the caption 
given in the examples). Alternatively, one might want to argue that these 
sentences should be interpreted not as instances of headless relative 
clauses, but rather as instances of right branching relative clauses with 
heads on the left. The two possible structures are illustrated in (22): 

(22) a. Headless Relative Clause 

j^C gC SUBJECT Object Verb- jig ^j^p 

b. Right Branching Relative Clause 

^:iffiAD:^p 3,CSTI|JECT 

We shall argue that the sentences of (2l) are not in fact right branching, 
but rather are headless. 


The two hypotheses in question make the sane predictions regarding 
case marking and word order. Both i)redict that the underlined noun 
phrases of (2l) would receive zero case marking and appear in initial 
position. Thus, word order and case marking cannot be used as tests for 
headless relativization in these cases. 

It is, however, possible to determine empirically whether the sen- 
tences of (21) are headless or right branching. It vail be remembered 
that in clear instances of relative clauses with (non-dumiror) heads, 
left branching relative clauses with heads on the right , relativization 
was possible regardless of the grammatical role of the relativized NP 
within S. and regardless of the nominalizing suffix employed. If 'the 
relative''"clau3es of (2l) are right branching, it would, therefore, be 
expected that sentences similar to (2l) but involving the relativization 
of the object of S. should be grammatical. But, as the sentences of 
(23) show, such relative clauses are not well-formed: 

(23) a. ?? wagra. Cwarabra 0. randi-jD wanu-rca 

cow Cboy buy-pres nomH die-past 3 
'The cow which the boy was buying dies.' 

b. ?? alcu. Cruna 0. jatu-jj ali-mi 

dog [man sell-pres nom] good- validator 
'The dog the man is selling is good.' 

« i 

Right branching relative clauses in which the object of the lower clause 

is relativized and the present nominalizer -j_ is em.ployed are ungrammatical . 

It might be proposed that the sentences of (23) do not constitute j 

conclusive evidence that the right branching hypothesis is false. Per- 1 

haps in right branching relative clauses the nominalizers encode both I 

tense and the gramniatical role of the relativized NP. Relative nom- I 

inalizers have both of these functions generally in most dialects of | 

Peruvian Quechua. They also have both functions in Imbabura headless j 

relatives, as demonstrated by the lack of ambiguity in the examples in j 

(18). Thus, it would not be surprising if the -j_ nominalizer were re- j 

stricted to relativized subjects and - shea to relativized objects in ^ 

right branching relative clauses. Hence, according to the revised right \ 

branching hypothesis, the ungrammaticality of (23) would be due to the j 

choice of -j_ as nominalizer. This proposal may be easily tested by sub- ■ 

stituting - shea and/or - na for -j_ in the examples in (23): i 

{2k) a. ?? wagra. Cwambra 0.. randi 7 _ 1^ \ ^ wanu-rca 

E boy buy (Ip^^t""";^^ ^ die-past 3 
•The cow which the ^oy |^^°f ^^^^^^^^ died.' 

-na 7* 

1 ali-mi 


dog Cman ^^1^ [ijast^'noml ^ good-validator 

'The dog which the manV^ ^/ is good.' 


The ungrammaticel examples of (2^1 ), in vhich - shea and -na have 
been so substituted show, we believe, that the ungrammaticality of (23) 
is, in fact, not due to the choice of nominalizer. If the right branching 
hypothesis were to be maintained in the face of such facts , it wo\ild 
be necessary to restrict right branching relative clauses to the rela- 
tivization of embedded subjects. Furthermore, headless relative clauses 
would have to be restricted to object relativization. Thus, the right 
branching analysis is frought with irregularities and ad hoc restrictions. 

But if the sentences of (2l) are analyzed as headless, such 
restrictions are \xnnecessary. In headless subject relatives, the nom- - 
inalizer -j_ is employed while in headless object relatives the nominalizer 
- shea is employed. The absence of sentences like those of (23) and 
(2U) is predicted if Imbabura does not allow right branching relative 
clauses at all. 

Thus , we conclude that in Imbabura there are two types of relative 
clauses, left branching and headless. Headless relative clauses may 
be formed on either embedded subjects or embedded objects. The choice 
of relativized NP is indicated by the nominalizer. A similar, though 
not identical pattern will be seen in Huanca, to which we shall now turn. 


The discussion of Huanca is based entirely on the excellent reference 
grammar for the dialect written by R. Cerron Palomino. ^3 Cerron-Palomino 
describes Huanca as having both left branching relative clauses (like 
(25)), and right branching relative clauses. 

(25) C0. lisqi-sha-ykil nuna-kak-man . kuti-y 

■'"i:know-nom-21 man-the-to return-imperative 
'Return to the person that you know.' 

Right branching is claimed to be predominant. He notes that if this 
description is correct, the situation in Huanca is anomalous since there 
does not appear to be any other dialect of Quechua in which a similar 
preference for right branching relativization is to be found. We shall 
propose in this section that relativization in Huanca is not anomalous, 
but rather that it is quite similar to that found in Imbabura. We sug- 
gest that all those instances of relative -clauses analyzed by Cerron- 
Palomino as right branching may in fact be headless. Thus, the relati- 
vization strategies in Huanca are reanalyzed as headless and left 
branching, the same two strageties found in Imbab\ira. The preference for 
putative "right branching" over left branching is treated in o\ir analysis 
as a preference for headlessness over left branching. Given the wide- 
spread occurrence of headless relative clauses in Quechua generally, 
the latter preference would not seem particularly surprising. 

We would like to begin by arguing that certain relative clauses 
analyzed by Cerron-Palomino as right branching must be headless. The 
argument is based on word order, as was the first and most straight- 
forward argument from Imbabura. Consider relativization of embedded 
direct objects when the -sha nominalizer (cognate to Imbabura - shea ) 
is employed: 


(26) Relativization of Embedded Direct O'bject with - sha 

luwis-pa kavallu lanti-sha-n-kak alfak-ta miku-yka-n 
Luis-genitive horse buy-nom-3-the alfalfa-acc eat-prog-3 
'The horse that Luis bought is eating alfalfa.' 

Cerron-Palomino treats sentences like (26) as right branching. 
But this clearly cannot be correct because the putative head, kavallu 
appears in normal object position within the embedded clause. ( Kavallu 
fails to receive accusative case marking in Huanca because the suppression 
of the accusative case marker is obligatory in relative clauses.) Thus, 
sentences like (26) must have a derived structure like (27) rather than 
one like (28): 

(28) NP^ NP^ \^P S^ ^S \P 

The fact that some putative right branching relative clauses must 
be analyzed as headless raises the question of vhether all putative 
right branching might in fact be headless. An examination of the range 
of data adduced by Cerron-Palomino provides no counterexamples to the 
headless hypothesis: 

(29) Relativization of Embedded Subject vith - sha 

walash libra lanti-sha-n-kak wanu-ku-n 

boy book buy-nom-3-the die-refl-3 
'The boy who bought the book died.' 

/ \ 1^ 

(30) Relativization of Embedded Subject vith -3^ 

allqu chuqllu suva-q-kak yana-m 
dog corn steal-nom-the black-validator 
'The dog that steals the corn is black.' 

In (29) and (30) the \mderlined NPs , valash and allqu , might, on 
the basis of the data available, be analyzed either as heads or as the 
subjects of the embedded clauses. Ultimately, this question must be 
resolved by finding dialect specific tests for the constituency of walash 
and allqu , as we in fact did in the previous section for Imbabura. We 
are not yet in a position to do this for Huanca, so any conclusions 
drawn at this time must be tentative. It would seem likely, however, 
that Huanca relativization is not as anomalous as Cerron-Palomino 
believed, and that headless relativization is central to relative clause 
formation in Huanca. ■'-5 


Relativization in Ancash is considerably more complex than in 
Imbabura and Huanca. As in other dialects, left branching relative 
clauses are fovuid, as in example (3l): 


(31) poncho-ta rura-q warmi sumaq-mi 
poncho-acc make-nom woman good-validator 
'The woman who makes ponchos is good. ' 

But, in addition, there is found in this dialect a range of case mark- 
ing facts which, so far, has eluded complete analysis on our part. We 
are unaware of any satisfactory general account of the complexities we 
have encountered. In particiilar, we are unable to specify which gramma- 
tical processes interact to derive apparent right branching relative 
clauses like (32) : 

(32) kuya-a-mi taqay warmi -ta qotsu-q-ta 
love-1-validator that woman-acc sing-nom-acc 

'I love the woman who is singing.' 

in which the apparent head and the nominalized verb both receive matrix 
case marking, and apparent right branching relative clauses like (33): 

(33) noqa kuya-a-mi warmi -ta qam dansa-sqa-iki-wan 

I love-1-validator woman-acc you dance-nom-2-obl 
'I love the woman you dance with.' 

in which the nominalized verb receives the case marking of the relati- 
vized NP in the embedded clause and the apparent head receives matrix 
case marking. In other examples, a noun phrase bearing embedded clause 
marking appears to the left of the embedded clause and the nominalized 
verb receives the same case marker as that noun phrase. This is illus- 
trated by example (3^): 

(3U) noqa urya-a nuna-ta qam qellay-ta qusqa-iki-ta 

I work-1 man-acc you money-acc give-nom-2-acc 
' I work with the man you give money to . ' 

In still further examples , the NP bears embedded case marking and the 
nominalized verb matrix case marking as in (35): 

(35) kuya-a-mi taqay siudad-caw noqa yaca-q-ta 
love-1-validator that city-loc I live-nom-acc 
'I love the city I live in.' 

Furthermore, the range of possibilities accepted often appears to vary 
iriiosyncratically from verb to verb. Until the regularity behind this 
complex of surface forms is revealed, it will be risky to make claims 
about the overall structure of the relative clause in Ancash."'-' 

There do, however, seem to be certain relative clauses that appear 
almost certainly to be headless. As in Imbabura and Huanca, the clear 
cases involve the relativization of direct objects: 

(36) a. noqa dansa-rqa-a mario warmi-ta kuya-sqa-n-wan 

I dance-past-1 Mario woman-acc love-nom-3-obl 
'I dance with the womsin Mario loves.' 


{3C) b. mario varmi-ta kuya-sqa-n wafiu-sqa 
Mario woman-acc love-nom-3 died-past 
'The woman Mario loves died.' 

The sentences of (36) appear to be completely analogous to instances 
of headless relativization of direct objects in Imbabura. As in both 
Imbabura and Huanca, the position of the underlined NPs is the one 
appropriate if those NPs are constituents of the embedded clause, but 
not if they are heads. Similarly, the accusative case marking of the 
italicized NPs is clearly due to their grammatical role within the 
embedded clause.-'-^ This case marking wo\ild be unexplained if these NPs 
were heads and, hence, matrix clause constituents. Additional examples, 
in which oblique NPs within the embedded clause are relativized appear 
in (37): 

(37) a. noqa kuya-a-mi qam warmi-wan dansa-sqa-iki-ta 

I love-1-validator you woman-obl dance-nom-2-acc 
'I love the woman you danced with.' 

b. mario way i -caw yaca-sqa-n inipa-sqa 
Mario house-loc live-nom burn-past 
'The house Mario lived in burnt down.' 

The argument that relative clauses like those in (36) and (37) 
are headless is further strengthened by the fact that when a direct 
object or oblique relativized noun phrase remains in its normal position 
between the subject and the verb it may not receive matrix clause case 
marking. This is shown in the examples in (38): 

(38) a. *noqa dansa-rqa-a mario warmi-wan kuya-sqa-n-wan 

I dance-past-1 Mario woman-obl love-nom-3-obl 
('I danced with the woman Mario loved.') 

b. *kuya-a-mi qam warmi-ta dansa-sqa-iki-ta 
love-1-validator you woman-acc dance-nom-2-acc 
('I love the woman you dance with.') 

The \ingrainmaticality of the sentences of (38) is particularly striking 
because matrix case marking is possible when the apparently analogous 
noun phrase appears to the left of the subject, as in (33) above. Thus, 
the underlined NPs in (36) do not behave as heads with respect to either 
word order or case marking. 

Although it is possible to establish that some relative claxises 
are headless in Ancash, it is not possible on the basis of what we know 
to determine the distribution of the construction. Are headless relative 
clauses limited to certain instances of the relativization of objects 
as in (36) and (37), or is the construction pervasive in this dialect, 
as it is to some extent in Imbabura and to perhaps an even greater ex- 
tent in Huanca? 



We have taken pains to demonstrate that three Quechua dialects have 
headless relative clauses. (Cur informant for a third dialect, Cocha- 
bamba, appeared to lack the construction. Headless relative clauses 
are apparently not found in southern Quechua.) It is of some importance 
that Imbabura on the one hand and the two Quechua I dialects on the other 
are about as distantly related genetically and geographically as any 
two dialects of Quechua. It would seem unlikely that either dialect 
acquired the construction by borrowing and, given the rareness of headless 
relative clauses among the languages of the world, it is improbable that 
the three dialects developed the construction independently. ^9 Thus, 
it would seem that headless relativization was likely to have been found 
in Common Quechua, as P. Muysken has suggested. Headless relativization 
appears to be a relativization strategy of some importance among Quechua 

We have observed that the existence of this strategy provides strong 
support for the Brame/Schachter hypothesis, according to which even rela- 
tive clauses with heads derive from underlyingly headless constructions. 
In addition to the theoretical significance of headless relatives for 
the study of the structure of relative clauses generally, they provide 
a testing ground for a variety of theoretical claims. Was Ross^^ 
correct in defining the 'Complex NP Constraint on structures of the form 
NP-S, or should the constraint prohibit extractions out of relative clauses 
as such? Headless relative clauses may provide the relevant test case. 

right in holding that the more a relative 
clause preserves of normal sentential form, the more it will be able 
to relativize "difficult" positions on the TIP Accessibility Hierarchy? 
Are Quechua headless relatives in fact used for such purposes? These 
are indicative of some of the questions that Quechua headless relatives 
may assist us in answering. 


The Quechua language (family) is spoken by six to eight million 
persons in western South America. There are tvro major dialect groups, 
termed Quechua I and II by Alfredo Torero ("Los dialectos quechuas". 
Anales Cientificos de la Universidad Agraria , Vol II, No.U. (Lima, 196k), 
pp. kh6~hTQ.) . For information on the classification of Quechua dialects, 
see Torero, op. cit. and Gary J. Parker ("La Clasificacion Genetica de 
los Dialectos Quechuas", Traducido por Gabriel Escobar. In: Revista 
del Museo Nacional . Tomo XXXII. (Lima, 19^3), pp. 2»+l-252). 

The present study is based on informant work carried out in Ecuador, 
Peru and at the University of Illinois (Urbana). An earlier version of 
this paper was presented at the Workshop on Andean Linguistics held at 
the University of Illinois in July, 1978. The authors gratefully acknow- 
ledge the support of the National Science Foxindation (grunt No. SOC 
75-0021+1+ and grant No. BNS 77-27159), the Research Board and the Center 
for International Comparative Studies of the University of Illinois for 
their support of this research. 


■^Paul Schachter ("Focus and relativlzatioo". In: Language , Vol.i+9, 
No.l (1973), pp. i9--6) cites data from Charles S. Bird (Relative clauses 
in Bamtara". In: Journal of West African Languages , Vol.5 (1968), 
pp. 35-1+7) which strongly suggest that Bambara relative clauses are headless 
in surface structure. Avery D. Andrews ("A typologcal survey of relative 
clauses". (1972) Unpublished Manuscript) also argues for the existence 
of headless relative clauses in several languages. Pieter C. M\iysken 
("Relative clause formation in Ecuadorian Quechua". (1976) Unpublished 
Manuscript) makes claims similar to ours for Quechua. 

This is not to suggest that evidence of \inderlying headlessness 
is not necessary for English and other languages. 

For example, Schachter, (op. cit.), and Brame, (Ms., "A new anal- 
ysis of the relative clause: evidence for an interpretative theory") 

As a matter of convenience, we adopt Schachter' s dummy head pro- 
posal. An alternative analysis is one in which there is no head at all 
underlyingly. A third possibility is that the head is deleted under 
coreference with the relativized noun phrase. We know of no empirical 
consequences for the hypothesis of a dimmy head rather than of under- 
lying headlessness or head deletion. Head deletion would not, however, 
be compatible with Brame 's data from English. 

These arguments are taken from Schachter (op. cit.), but are 
apparently due \iltimately to Brame. 

For an alternative explanation of the English data, see Pauline 
Jacobson ("Some aspects of movement and deletion". In: Berkeley 
Linguistic Society , Vol. Ill (1977). 

It is immaterial for our pruposes whether superficially "headless" 
relative clauses contain a dummy head like that posited in deep structure, 
or whether the head is deleted prior to surface structure. For the sake 
of simplicity in representation, we will assume that the dummy head is 
deleted. f 


'Data in this section is based on informant work conducted by Peter * 

Cole ajid Janice L. Jake. We would like to thank our principal informant, ^ 
Carmen Chuquin, for her generous assistance. The data presented here repre- 1 

sents the Quechua spoken in Mariano Acosta, in the eastern part of the : 

province of Imbabura. We do not know whether all of our claims are true | 

for Imbabura generally. In particular, the uses of the nominalizing * 

suffixes reported here differ from those discussed by Miaysken (op. cit.). * 

Possibly the informants Muysken consulted were speakers of a subdialect ^ 
of Imbabura with rules different from those in Mariano Acosta. 

In most Quechua dialects both grammatical role and tense are 
involved in the choice of nominalizers . See Muysken (op. cit.) for a 


survey of nominalizer choice in a variety of dialects. Typically, the 
nominalizer cocnate to Imbabura -j_ is used with subjects and - shea with 
direct objects. A similar pattern is found in Imbabura, but only in 
surface headless relative clauses. This is discussed further belov. 

That is to say, they have dummy heads or no heads. See Fn. 8. 


And, possibly, intransitive subjects. We have elicited such 

sentences as the following: 

(i) jari shamu-shca yarjachi-n 
man came-past nom hungry-3 
'The man who came is hungry-. ' 

Rudolf o Cerron-Palomino ( "Gramatica Quechua Junin-Huanca ", 
Ministerio de Educacion/Instituto de Estudios Peruanos (Peru, 1976)). 


The -2^ nominalizer is cognate to Imbabura -j_. 

- na relatives might be useful in testing our claims, but unfor- 
tionately, all of Cerr6n-PaJ.omino ' s examples involve pronominal subjects 
(so-called transitions). Thus, a word order test cannot be used: 

(i) ponchu tr\ila-ku -na -n -kaq -ta apamu-y 

poncho put on-refl -nom -3 -the -ace bring-imperative 
'Bring the poncho that he is going to put on.' 

Note that the sole reflex of the subject of trulaJm- is the third 
person possessive suffix -n_, which is appended to the nominalizer -na. 
Such examples do not tell us whether ponchu would precede or follow a 
subject NP. 

We woiild like to express our appreciation to Celestino Figueroa 
Bedon, originally from the town of Tarica, Department of Ancash for 
information on Ancash. 


According to Peter Landerman (personal communication) David 

Webber has foxmd a similar case marking pattern in relative clauses in 

other central Peruvian varieties of Quechua. These facts are to be 

discussed in Webber's forthcoming Masters Thesis. 


Though there appears to be some variation from one sub-dialect 

of Ancash to another, in the sub-dialect of our informant (Huaraz-region) 

the choice of the nominalizing suffixes, -3^ and - sqa , depends solely 

on the grammatical role of the relativized NP. Relativized subjects 

(and locatives) take the -3^ nominalizer, whereas direct objects, indirect 

objects and obliques taJce the - sqa nominalizer. 



We woiild like to show that eeoercrhically intervening dialects 
lack headless relatives , but we do not have sufficient infonuation to 
do so at this time. 

Thesis (1967)). MIT. 

Keenan, Edward L. and Bernard Comrie ("NP accessibility and universal 
gramnar". In: Linguistic Inquiry, Vol. 8 (l972), pp. 63-100.) 


Andrews, Avery D. 1972. A typological survey of relative clauses. 

Unpublished manuscript. 
Bird, Charles S. I968. Relative clauses in Bambara. Journal of West 

African Languages 5.35-^7. 
Brame, Michael. 197 • A new analysis of the relative clause: evidence 

for an interpretative theory. Unpublished manuscript. 
Cerron-Palomino , Rudolfo. 1976. Gramatica Quechua Junin-Huanca. Peru. 
Jacobson, Pauline. 1977- Some aspects of movement and deletion. 

Berkeley Linguistic Society 3. 
Keenan, Edward L. and Bernard Comrie. 1972. NP accessibility and 

universal grammar. Lingustic Inquiry 8.63-100. 
Muysken, Pieter C. 1976. Relative clause fonaation in ecuadorian 

Quechua. Unpublished manuscript . 
Parker, Gary J. 1963. La Clasificacion Genetica de los Dialectos 

Quechuas . Tracucido por Gabriel Escobor. Revista del Museo 

Nacional 32.2lH-252. 
Ross, John R. I967. Constraints on variables in syntax. Ph.D. 

Schachter, Paul. 1973. Focus and relativization. Language kQ.19-h6. 
Torero, Alfredo. 1961+. Los dialectos quechuas. Anales Cientificos de 

la Universidad Agraria 2.UU6-H78. 

studies in the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 

P, Cole W. Herbert G. Hermon S.N. Sridhar 

This paper constitutes an examination of the process 
by means of which subject properties are acquired by a noun 
phrase not previously having those properties. The evidence 
presented indicates that behavioral subject properties (like 
the ability of an NP to delete under Equi ) are acquired his- 
torically prior to subject coding properties (like nominative 
case marking and control of verb agreement). 

I. Introduction 

The purpose of this paper is to examine cross-linguistically the 
historical principles governing the acqvdsition of subject properties 
by a noun phrase.J Several investigators, including Keenan (19T6), 
have shown that these properties do not always converge on a single 
noun phrase. In this paper we are especially concerned with the ques- 
tion of whether certain kinds of properties associated with subjecthood 
are acquired prior to other kinds of properties. In particiilar, are 
morphosyntactic (i.e. coding) properties like nominative case and control 
of verb agreement acquired before or after transformational (i.e. be- 
havioral) properties like control of reflexivization and deletability 
by rules normally affecting only subjects? We shall argue that behavioral 
properties are consistently acquired prior to coding properties. 
Evidence in support of this claim will be presented from three groups 
of languages: Germanic, Polynesiaji, and Georgian. For each language 
(or group of languages), we will consider both behavioral and coding 
properties of subjecthood at various historical stages. Among the be- 
havioral properties considered will be the capacity to trigger reflex- 
ivization in languages in which only subjects control that process, 
the capacity to undergo and to trigger rules of deletion under ident- 
ity which are normally restricted to subjects, and the capacity to under- 
go subject raising rules. The coding properties considered include 
subject case marking and control of verb agreement. It should be noted 
that we do not consider initial position among the subject coding pro- 
perties. Rather, following Timberlake (19T6: 560), we regard initial 
position as primarily indicative of topicality rather than subjecthood. 
V.Tiile there is a frequent correlation between topicality and subject- 
hood, the two do not necessarily coincide. Thus, initial position will 
not figiire in the arguments that follow. 

For each language (family), we shall argue that at an early stage 
in the development of the construction (Stage A) , the noun phrase in 
question displays no subject properties, either behavioral or coding. 
At a later time (Stage B), the ansilogous noun phrase exhibits behavioral. 


but not coding properties. If coding properties are foimd at any stage, 
they are acquired subsequent to behavioral properties (Stage C). We 
propose that such a sequence is the unmarked (and perhaps universal) 
order of acquisition of subject properties cross-linguistically. The 
bulk of the paper vill be devoted to arguments that such a sequence is 
an accurate portrayal of the acquisition of subject properties in the 
languages discussed here. We know of no instances in which coding 
properties are acquired before behavioral properties. 

II. Germanic Languages 

We shall begin by tracing the development of two constructions 
in Germanic: dative experiencer constructions as exemplified in (l): 

(1) Dative experiencer in Germanic (example from German) 

Mir gefallen diese Damen 
me-dat please these ladies 
'I like these ladies.' 

and passive sentences with verbs taking non-accusative objects, as in (2): 

(2) Passivized objects of verbs taking dative objects in Germanic 
(example from German) 

Uns wird von der Polizei geholfen 
us-dat aux-3sg by the police help-past part 
'We are helped by the police.' 

We shall argue that at Stage A, no subject properties are associated 
with dative experiencers or passivized non-accusative objects. At 
Stage B, these noun phrases come to exhibit the behavioral accoutre- 
ments of subjecthood, but fail to display subject coding properties. 
Finally, at Stage C, these noun phrases, in some Germanic languages, 
come to possess both syntactic and morphosyntactic subject properties. 

The argioment from Germanic is complicated by the number of languages 
involved and by the fact that the parent language , Proto-Germanic , 
which corresponds to Stage A, is unattested. We must produce an arp;- 
ument, based on a comparison of the attested Germanic langixages, that 
at this unattested stage no subject properties were associated with 
dative experiencers and passive subjects. 

Germanic consists of three subfamilies. East Germanic, North 
Germanic and West Germanic. The sole extensively documented member of 
the East Germanic subgroup, Gothic, is represented by a relatively small 
corpus, the bulk of which is translated from Greek. Thus, only a very 
limited amount of evidence is available which bears on the status of the 
noun phrases in question in East Germanic. There is at least some 
evidence from Gothic, however, indicating that dative experiencers 
had no subject properties. Two rules, Equi and Conjunction Reduction, 
which normally applied only to subjects, treated the nominative NP 
in experiencer constructions, rather than the dative experiencer, as 


subject. This is illustrated in (3a) and (3b): 

(3) a. Nominative deletes under Equi (example from II Cor 5:9) 

inuJ)-J)is usdaudjam. . . vaila galeikan irama 
because of-this strive(lpl) well VCdat:(inf) him(dat) 
'Because of this wc strive to please him well.' 

b. Nominative undergoes conjxinction reduction (ex. from Thess U:l) 

hvaiwa skiilu^ gaggan jah galeikan guda 
how sho\ild(2pl) go(inf) and V:dat:(inf) god(dat) 
'how you should live and please God.' 

As illustrated by {h) , the dative experiencers in such constructions failed to control verb agreement: 

(k) Dative experiencers do not control verb agreement (ex. 
from I Thess 3:l) 

galeikaida uns ei bilii)anai weseima 
V:dat:(3sc) ucCdat) that left might be 
'It pleased us that we might be left.' 

Thus, we conclude that in Gothic, the nominative, rather than the dat- 
ive experiencer in experiencer constructions , had both the coding and 
the behavioral properties of subjects. The significance of this evid- 
ence for the development of experiencer constructions in Germanic will 
be seen below. 

In the earliest extensive North Germanic texts (Old Icelandic), 
behavioral subject properties appear to have been associated with the 
dative experiencers of only one verb, ^ykkia 'seem'. In an extensive 
study of reflexive pronouns in Old Icelandic, Rose (19T6) lists several 
instances in which dative experiencers of this verb serve as anteced- 
ents of reflexive pronouns (a function normally restricted to subjects). 
This is illustrated by example (5): 

(5) Dative experiencer of ^ykkia controls reflexivization (Njall. : 278) 

honum. i)6ttir ^u hafa haft vij) sik. fjtjrrajj 
him(dat) seem(3sg) nom aux (inf) part prep refl death-plot 
'He thought you to have had a death-plot against him.' 

There appear to be no attestations in which such reflexivization is 
controlled by passivized non-accusative objects, or by experiencer 
NPs other than those occ\irring with fykkia . ' »° Thus, the earliest Ice- 
landic texts appear to represent a transitional period, between Stage 
A and Stage B. At this stage, at least one behavioral subject property 
was associated with the dative experiencer of a single predicate. 

In contemporary Icelandic, a substantial change appears to have 
taken place. Andrews (1976) notes a variety of behavioral properties 
associated both with a wide range of experiencers and with passivized 
non-accusative objects in Modern Icelandic. Example (6) illustrates 


dative experiencers and (7) illustrates passivized non-accusative objects; 

(6) Dative experiencers 

a. trolli-nu svelgdist a stulku-nni (A 10c) 
trcll-the(dat) mis-svallowed on girl-theCdat) 
'The troll swallowed the girl wrong.' 

b. mer ifkar/lika J)eir bilar (A lOd) 
me(dat) like(3sg/pl) those cars(nom) 

'I like those cars.' 

(T) Passivized non-accusative objects 

a. mer var bjarga4 (A 13a) 
me(dat) aux VCdat^C supine) 

'I was saved, ' 

b. nun var be§id (A 13d) 
me (gen) aux VCgenHC supine) 

'I was waited for.' 

As in Old Icelandic, the trigger for reflexivization must be a 
subject. Example (8) shows that dative experiencers control reflex- 

(8) Dative experiencer controls reflexivization 

henni svelgdist a steikinni sinni (A 39^) 
she(dat) mis-swallowed on steak her(refl) 
'She swallowed her steak wrong.' 

Andrews does not provide examples parallel to (T) involving reflexiv- 
ization. He does give examples involving the passivization of indirect 
objects. Example (9a) illustrates an active sentence with both a dir- 
ect and an indirect object. In (9b) the indirect object has been pass- 
ivized and controls reflexivization. 

(9) a. Active bitransitive sentence 

teir seldi honum drengina (A UOa) 
him(dat) sold him(dat) boys-the(acc) 
'They sold him the boys.' 

b. Passivized indirect object controls reflexivization 

honum voru seldir drengirnir af fraendum 
him(dat) aux(inf )sold(part) boys-the(nom) by relatives 

sfnum/hans (A lt2b) 

'He. was sold the boys by his. /self's relatives.' 


Dative experiencers and passivized non-accusative otjects can also be 
raised "by subject-to-object raising, a process vhich normally applies 
only to subjects. This is shown in (lO)-(ll): 

(10) Dative experiencer undergoes SOR 

eg tel honum lika \)e±r bilar (A 19a) 
I believe him inf those cars(nom) 
'I believe him to like those cars.' 

(11) Passivized non-accusative object undergoes SOR 

eg tel i)eirra hafa veri-d be4i4 (A19d) 
I believe them(gen) aux(inf) aux(sup) VC gen: (supine) 
'I believe them to have been visited.' 

Furthermore, experiencers and passivized dative objects Eiay be the vic- 
tims of Equi : 

(12) Dative experiencer deletes under Equi 

trolli-d vonast til ad svelgjast ekki a 
troll the(nom) hope comp mis-svallowed not on 

stulkinni (A 26a) 
girl the 

'The troll hopes not to swallow the girl wrong.' 

(13) Passivized dative object deletes under Equi 

eg vonast til ad ver4a bjargad (A 26d) 
I hope comp avix(inf) VCdatD (supine) 
'I hope to be saved.' 

Thus, the data from Andrews indicate an extension in the history 
of Icelandic of the syntactic properties of subjecthood exhibited by 
experiencers and passivized non-accusative objects. It sho\ild be noted, 
however, that these noun phrases do not show any morphosyntactic subject 
properties. In experiencer constructions like (6b) (repeated): 

(6) b. mer likar/lfka ^eir bilar (A lOd) 
me(dat) like(3sg/pl) those cars(nom) 
'I like those cars.' 

^eir bilar , not mer, appears in nominative case. Similarly, verb 
agreement is triggered by the nominative no\in phrase, or fails to apply. 
This is illustrated by (6b) as well. Hence, Modern Icelandic appears 
to be an instance of Stage B. Dative experiencers and passivized non- 
accusative objects acquired behavioral properties typical of subjects, 
but not coding properties . 

We will now ttirn to evidence for the transition from Stage B to 
stage C in North Germanic. First, let us consider Faroese, which is 
closely related to Icelandic. In Faroese, accorking to Lockwood (1961*: 103), 


passivized non-accusative objects in some types of passive constructions 
may assume nominative case. Compare (lU)-(l7) with Andrews' examples 
from Modern Icelandic : 

(lit) Active dative object sentence 

teir fagnadu Depilsmonniun vael 
they VCdat: Depilsmen(dat) well 
'They welcomed the Depilsmen well.' 

(15) Passivized dative object becomes nominative obligatorily 

Depilsmenn voru va_el fagnaeLLr 
Depilsmen(nom) aiix(3pl) well VCdatnCpart) 
'The Depilsmen were well received.' 

(16) Passivized dative object becomes nominative optionally 

hon var bo4in i brudleyp 
she(nom) aux VC dat II ( part ) to wedding 
'She was invioed to a wedding.' 

(it) Passivized dative object remains dative 

maer var ikki bodid at koma 
meldat) aux neg VCdat!] (supine) comp inf 
' I was not invited to come . ' 

Examples (lU)-(l7) show that passive subjects of verbs taking dative 
objects in Faroes e have begvm to take on morphosyntactic subject pro- 
perties of Stage C. 

In North Germanic languages other than Icelandic and Faroese, dative 
and accusative case have merged. These languages therefore no longer 
have non-accusative objects. However, the acquisition of subject coding 
properties by historically non-nominative experiencers is well advanced 
in them. Examples will be limited to Swedish, but similar examples 
from Danish and Norwegian could be supplied as well. The examples in 
(18) and (19) and similar ones cited by Evans (1975) demonstrate that 
in Old Swedish, as in Icelandic, such experiencers did not have the 
coding properties of subjects, although they had at least some behav- 
ioral subject properties. 

(18) Dative experiencers do not control verb agreement 

thz tykker them allom wara thz besta 

that V:dat:(3sg) them(3pl dat) all be the best 
'That seemed to all of them to be the best.' (Rim-Kroniker I 

p. 391) 

(19) Dative experiencers control reflexivization 

honom thykte sik wara J enom lystelikom stadh 
him(dat) VCdatU refl be in a pleasurable city 
'He thought himself to be in a pleasurable city.' (Siaelinna Thrust, 

p. 280, ii. 17-18) 


In modern Swedish, as in Danish and Norwegian, these historically non- 
nominative experiencers have acquired the subject codinc properties 
of nominative case and control of verb agreement. A partial list 
of predicates which have come to occur with nominative subject-exper- 
iencers is given in (20): 

(20) Historically non-nominative experiencers acquire nominative 
case and control of verb agreement 

jag drommer 

•I dream' 

jag tycker 

'I think' 

jag kommer i hug 

'I recall' 

jag bor 

•I ought' 

jag ar varm 

'I am warm' 

jag samtycker 

'I agree' 

jag saknar 

'I lack' 

jag langtar 

•I miss' 

jag behover 

'I must' 

This list could be readily extented. Thus, Swedish and other contem- 
porary Scandinavian languages are illustrative of stage C. 

Although in North Germanic there is no direct evidence of a stage 
at which no_ subject properties are associated with dative experiencers 
and passive subjects, the attested development from behavioral proper- 
ties with a limited range of predicates to behavioral, and ultimately 
coding properties with a wide range of predicates strongly suggests 
that this was the case at an earlier stage in Germanic . Evidence from 
West Germanic, to which we now turn, supports that hypothesis. 

We shall discuss data from two West Germanic languages , German and 
"nglish. In German dative experiencers and passivized non-accusative 
objects display no subject properties, and we have found no evidence 
that the situation in Modern German differs from that at earlier stages 
in the development of the language. The absence of morphosyntactic 
subject properties is shown in (2l)-(22): 

(21) a. Active dative object sentence 

Die Polizel hilft uns 
the police help(sg) us(dat) 
'The police help us.' 

b. Passivized dative object remains dative sind fails to 
control verb agreement 

Uns wird von der Polizei geholfen 

us(dat) aux(3sg) by the police VCdat:(past part) 

'We are helped by the police.' 


(32) Passivized dative objects do not delete under relative clause 

*Der vom Lehrer geholfene Junge bekam eine gute Note 
the by the teacher help(part) boy got a good grade 
(•The boy helped by the teacher got a good grade.') 

(33) Dative experiencers do not delete under relative clause 

*Der das Buch gefallende Junge sitzt da in der Ecke 
the the book VCdatD(part) boy sits there in the corner 
('The boy liking the book is sitting there in the corner.') 

Thus, with respect to three behavioral properties of subject- 
hood, neither passivized non-accusative objects nor dative experiencers 
behave as subjects in German. For further discussion of the non-subject- 
hood of passivized dative objects, see Reis (1973), Ebert (1975) and 
Breckenridge (1975). We conclude that in German these noiin phrases 
show neither coding nor behavioral subject properties. The relevance 
of this fact for our argxjment will be discussed shortly. 

We will now turn to English. As was the case with Swedish, neu- 
tralization of the dative/accusative distinction in English causes 
us to limit our discussion to experiencers. In Old English, there are 
sporadic instances in which experiencer NPs behave as subjects, although 
they do not exhibit subject coding properties. In (3^+) for example, 
the accusative-experiencer verb langian 'to desire , long for' appears 
embedded under the subject-to-subject raising verb onginnen 'begin'. 
The raised experiencer retains its accusative case. 

(3H) Accusative experiencer undergoes SSR 

Jja ongan hine eft langian on his cyiipe 

then began him(acc) again VCaccD(inf) for his kin 

'Then he b'egan to long for his kin.' (Blick. Horn. 113:15, A.D. 971) 

This example substantially antedates the first appearance of langian 
with a nominative experiencer, for which Gaaf (l90ii) gives the date 

As was noted by Gaaf (l90l^)and Harris (1975) inter alia, the ex- 
periencer NPs occurring with a niomber of verbs , such as liken and dremen , 
had subject behavioral properties in early Middle English, but had not 
yet acquired subject coding properties. Note the accusative case of 
t)e 'thee' in (35a) and of him in (35^): 

(35) Accusative experiencers 

a. ..a foreward t)at ^e mai full well like 

an agreement that you(acc) may full will VCaccD 

'an agreement that may full well please you' (Dame Girith: 

255, A.D. 1275) 


(35) tJ. of ^at him drempte in prisun 

of that hin(acc) VCaccD in prison 
'He dreamed of that in prison.' (Genesis and Exodus, 

2116, A.D. 1250) 

Although the experiencers in (35a) and (35h) show non-subject 
coding properties , they undergo deletion by Equi , undergo and control 
reduction under identity with a nominative subject, and undergo indef- 
inite subject deletion, all properties generally restricted to subjects: 

(36) Accusative experiencers delete under Equi 
Him buri) to liken well his lif 

him befits to VCaccD will his life(nom) 
'It befits him to like his life well.' (Dame Sirith: 82, 

A.D. 1275) 

(37) Accixsative experiencers undergo conjunction reduction under 
identity with nominative subject 

a. Lewed men leued hym well and liked his wordes 
'Ignorant men loved him well and his words pleased Cthem] . ' 

(Langland, Piers Plowman , prol. 72, A.D. 1362) 

b. Arthur loked on the swerd and liked it passynge well 

(Mallory, Morte D'Arthur . A.D. 1U7O) 

(38) Accusative experiencers control deletion by conjunction 
reduction of a nominative subject. 

Us sholde neither lakken gold ne gere. But ben 
(ace) should neither VCacc] gold nor gear but be 

honovired while we dwelten there 
honoured while we dwelt there 

'We should not lack gold nor gear, but be honoured while we 
dwelt there.' (Chaucer, Canterbury Tales ) 

(39) Accusative experiencers undergo indefinite subject deletion 

Good is qiiaj) losef, to dremen of win 

'CltH is good, said Joseph, to dream of wine.' (Genesis and 

Exodus: 2067, A.D. 12 50) 

The above examples, and possibly others cited by Gaaf and Harris, show 
that experiencers in English exhibited behavioral subject properties 
before they exhibited coding properties . 

The West Germanic data, taken in isolation, might be analyzed in 
either of two ways , which we shall label Hypothesis One and Hypothesis 
Two. According to Hypothesis One, experiencers axid passivized dative 
objects displayed no subject properties in the parent language. In 
English, behavioral properties were acquired by some experiencer NPs 
by the thirteenth century, and coding properties later. But in German, 


neither behavioral nor coding properties were acquired. Thus, German 

is considered to have preserved the situation found in the proto-language . 

This is the position for vhich we shall argue. 

According to Hypothesis Two, in the proto-language, as a resvat 
of the loss of subject coding properties at an even earlier stage, 
behavioral but not coding properties were associated with experiencers 
and passivized dative objects. In German these behavioral properties 
were lost, but in English they were not. At a later stage in English, 
these noun phrases re-acquired subject coding properties. 

Although taken in isolation, the two hypotheses just outlined might 
appear to be equally plausible, when viewed in conjunction with the 
North Germanic and East Germanic data disc\issed earlier, only Hypothesis 
One is tenable. It will be remembered that in Gothic, the nominative 
NP in experiencer construction, rather than the dative experiencer, 
exhibited subject behavioral properties. Moreover, in Old Icelandic 
dative experiencers had only incipient subject behavioral properties, 
and passivized non-accusative objects appear to have had none. A clear 
trend toward the development of additional behavioral and, later, coding 
properties was seen in the case of Icelandic. This trend is consistent 
with the claim that the noun phrases in question showed no subject pro- 
perties in the proto-language , but is is inconsistent with the hypo- 
thesis that in the proto-language a f\ill panoply of subject behavioral 
properties were present. Hence, we conclude that the most plausible 
reconstruction is that in the proto-language experiencers and passivized 
non-accusative objects had no subject properties. Subsequently, first 
behavioral and then coding properties were developed in some of the 
daughter languages . Thus , the most plausible reconstruction for 
Germanic supports our hypothesis regarding the order of acquisition 
of behavioral and coding subject properties. 

III. Polynesian 

Further evidence in support of our hypothesis is provided by the 
existence of morphologically ergative languages in which transitive 
subjects are assigned historically oblique case ma.rkers. In a number 
of such languages, subjects marked in this way appear to have been 
historically oblique NPs (Stage A), which acquired the behavioral pro- 
perties of subjects while retaining their oblique case marking (Stage B), 
with subsequent reinterpretation of the oblique marking as the normal 
marking for transitive subjects. In this section, we will trace a dev- 
elopment of this type in Polynesian, as discussed in Chung (19T6). 
Additional possible instances of this sort of change are treated by 
Anderson (19TT). 

The Polynesian languages fall into two large groups, those with nomin- 
ative-accusative case marking and those with ergative-absolutive case 
marking. In the nominative-accusative languages, the subjects of ac- 
tive transitive verbs are unmarked. Upon passivization, those sub- 
jects are assigned the marker e_. Compare the following examples from 
Maori : 


{ko) Active sentence in Maori 

ka patu i te tuna te kotiro (Chung, ITc , p.l6) 
aor. hit ace the eel the girl 
'The girl killed the eel.' 

(Ul) Passive sentence in Maori 

ka tuku-a e Pacwa te waka (Chung, 19a, p. l8) 
aor. leave-pass agt Paova the canoe 
'The canoe was left by Paowa. ' 

In the ergative group, the subjects of active transitive verbs are roark- 
ed with e_. Observe the following example from Tongan, a language of the 
Tongic subgroup of Polynesian: 

{k2) Transitive sentence in Tongan 

(Chung, 22c, p. 20) 

na'e taa'i 'e 




past hit agt 




'Mary hit John' 

Thus, e_ is used for both passive agents in Maori and transitive subjects 
in Tongan. Chung argues convincingly that Proto-Polynesian was a nom- 
inative-accusative language, like Maori, and that the ergative case 
marker e_ in languages like Tongan developed from an original passive 
agent marker of the sort represented in the Maori example in {hi). 
Chung (19T6: 316) represents this development in the following way: 

(1+3) Passive: Verb- Cia e_ Agent (=underlying subject) Subject 
(=underlying DOl 

reinterpreted as 

Transitive: Verb-Cia e Sub j DO 

We will not repeat her extensive arguments for such a development in the 
present paper. Instead, we will accept the correctness of those argu- 
ments, and will demonstrate the way in which the development represented 
in (1+3) supports our hypothesis that behavioral properties are acquired 
prior to coding properties. 

In the nominative-accusative languages representative of Stage 
A, the e_ NP (the underlying subject) in the passive sentences does 
not exhibit ajiy of the behavioral properties of subjects. Maori, for 
example, has a rule of Equi which deletes those subjects which are also 
agents, as in (1+U): 

(1+1+) Equi deletes agentive subjects 

i hiahia au ki te patu i te poaka (Chung,l+l+a,p.l27) 
past want I comp hit ace the pig 
'I wanted to kill the pig.' 


The e NPs in passive sentences cannot be deleted even though they are 

(1*5) Equi does not delete passive agents 

*i hiahia au ki te patu-a 'te poaka (Chung,5Ta,p.l30) 
past want I comp hit-pass the pig 
('I vanted to kill the pig.') 

Maori also has a rule of subject -to-subject raising, governed by sen- 
tential negatives like kahore 'not', which can apply to subjects, but 
not to direct objects or other non-subject NPs. Compare (U6a) and (U6b): 

ik6) a. Subjects undergo SSR 

kahore a Hone i patu i te poaka (Chung, 158a, p. 179) 
not pers John past hit ace the pig 
'John didn't kill the pig.' 

b. Non-subjects do not undergo SSR 

*kahore te poaka i patu ai a Hone (Chung, 155a, p. 178) 
not the pig past hit pro pers John 
('John didn't kill the pig.') 

The rule can apply to the unmarked NP (the underlying DO) in passive 
sentences, but not to the e_ NP (the vinderlying subject). Compare (U7a) 
and ( U7b ) : 

(U7) a. Passive subjects imdergo SSR 

kahore raua kia kite-a mai e nga tangate 
not they sbj see-pass here agt the man 

o te whare (Chung, 159a, p. 179) 
of the house 

'They were not seen by the people in the house.' 

b. Passive agents do not undergo SSR 

*kahore (e) nga tangata kis kite-a (ai) raua 
not agt the-pl man sbj see-pass pro they-dual 
('They weren't seen by the people.') (Chung, 156a, p. 178) 

Thus, at Stage A, as represented by Maori, underlying subjects in passive 
sentences in Polynesian lacked both the coding properties and the be- 
havioral properties of subjects; passivized underlying direct objects, 
on the other hand, exhibited both subject behavioral properties and the 
zero case marking of subjects. 

In the innovating, morphologically ergative languages of Polynes- 
ian, the e_-marked NP (the historical agent phrase) in transitive sent- 
ences is treated as subject by such rules as Equi and Raising, while 


the historical passive subject is not. Tongan, for example, has a rule 
of Equi which can delete intransitive or middle subjects, but not non- 
subject NPs . Compare the examples in (U8): 

(U8) a. Equi deletes subjects 

pea na'e 'alu 'a e tangata 'o folau mama'o 
and past go abs the man comp sail far 
'Then the man went and sailed away.' (Chung, l8a, p.llU) 

b. Equi does not delete non-subjects 

*na'e 'alu 'a e tamaiki tangata 'o ngalo 
past go abs the children man comp forgotten 

'enau lesoni (Chung, 25a, p. 117) 
their lessons 

('The boys went and forgot their lessons.') (deletion of 
stative agent) 

In transitive sentences, the e_ NP can be deleted by Equi, but the his- 
torical passive subject cannot. Compare the examples in {h9) : 

{k9) a. Historical passive agents delete under Equi 

na'e 'alu 'a e tangata 'o taa'i 'a e kuli 
past go abs the man comp hit abs the dog 
'The man went and hit the dog.' (Ch\ing, 26a., p. 117) 

b. Historical passive subjects do not delete under Equi 

*na'e 'alu 'a e fefine 'o u'u 'e he kuli 
past go abs the woman comp bite erg the dog 
('The vroman went and dog bite 0.') (Chung, 23a, p.ll6) 

Thus, Equi in Tongan treats the historical oblique agent, not the his- 
torical passive subject, as subject. 

Tongan also has a rule of subject-to subject raising, governed by 
the verb lava 'be able', which can raise intransitive subjects, but 
not non-subject NPs. Compare (50a) with (50b): 

(50) a. Subjects undergo_SSR 

'e lava 'a e pepe 'o lea 
unm can abs the baby comp talk 
'The baby can talk' (Chung, 63a, p. 136) 

b. Non-subjects do not undergo SSR 

*tp 2iau lava pe 'o ngalo ia 
uim we can emp comp forgotten it 

('We might be able to forget about it.') (Chung, 86b, p. 11*5) 
(raising of stative agent) 


In transitive sentences in Tongan, subJect-to-subJect raising treats 
the e-marked NP, but not the historical passive subject, as subject. 
Compare (51a) and (51^); 

(51) a. Historical passive agents undergo SSR 

*e lava 'e ho famili 'o tauhi koe 
xmm can erg your family comp care you 
'Can your family take care of you?' (Chung, 87c, p.llt6) 

b. Historical passive subjects do not undergo SSR 

*na'e lava 'a e fefine 'o taa'i 'e Sione 
past can abs the woman comp hit erg John ^ 

('The woman could be hit by John.') (Chung, 8Ua, p. 11+5) 

In Tongan, therefore, historically oblique passive agents have 
taken on the behavioral properties of subjects while retaining oblique 
case marking. Correspondingly, original passive subjects have lost 
behavioral subject properties, but are still marked in the same way as 
intransitive subjects. These facts are consistent with the prediction 
of our hypothesis, which holds that subject coding properties will not 
be acquired by an NP prior to subject behavioral properties. 

A similar situation exists in Samoan, another morphologically erg- 
ative language, in which historically oblique passive agents with e_ 
have come to behave as surface subjects, while historically passive sub- 
jects have ceased to behave as subjects. Samoan has a rule of subject- 
to-subject raising, governed by a nximber of verbs, including mafai 
'be able', which can raise intransitive subjects, but not non-subject 
NPs. Compare (52a) and (52b): 

(52) a. Subjects undergo SSR 

e mafai e tatou 'ona nonofo 
unm can erg we comp stay 
♦We can stay.' (Chung, 9Ta, p.l5l) 

b. Non-subjects do not undergo SSR 

*e mafai (e) le la'au 'ona 'ou ta-ina 'oe 

unm can erg the stick comp I hit-trans you 

('I can hit you with the stick.') (Chung, llBb, p.l6o) 
(raising of instrument) 

As in Tongan, Samoan subject -to-subject raising can operate on his- 
torical oblique passive agents, but not on historical passive subjects 
Compare the examples in (53): 

(53) a. Historical passive subjects do not \indergo SSR 

*e mafai le tali 'ona maua e le tama 
unm can the answer comp catch erg the boy 
('The boy can come up with the answer,') (Chung, ll6a, p. 159) 


(53) b. Historical passive agents undergo SSR 

e mafai e le tama 'ona maua le tale 

linm can erg the boy comp catch the ansver 

'The boy can come up with the ansver.' (Chung, 119a, p.l60) 

In summary, at Stage A in Polynesian, vhich is still represented 
by Maori, there was a productive rule of passive, which made under- 
lying direct objects into subjects, and demoted underlying subjects 
to the role of oblique NP. At this stage the ^-marked underlying sub- 
ject in passive sentences exhibited neither the behavioral nor the coding 
properties of subjects, while the underlying direct object exhibited 
both coding and behavioral properties of subjects. At Stage B, repre- 
sented by Tongan and Samoan, the rule of passive lias become non-produc- 
tive, and the behavioral properties of subject-hood have been transferred 
from the historical passive subject to the historical agent phrase. 
Both NPs , however, preserved their original case marking, so that the 
historical passive subject, which is no longer a subject behaviorally , 
bears the same case as the subjects of intransitive sentences, while 
the historical oblique agent, which is a subject behaviorally, bears 
oblique case marking. 

Stage C, the acquisition of subject coding properties by his- 
torically oblique passive agents, does not appear to have been reached 
in any of the Polynesian languages. In other language families, however, 
in which morphological ergativity arose through a development similar 
to that found in Polynesian, there is evidence for Stage C. Comrie 
( 1977: p. 38) cites the case of Indo-Iranian, in which passive constructions 
became dominant in the perfect, to the exclusion of active constructions * 
with the consequent loss of a productive passive in the perfect. Sub- 
sequently, behavioral subject properties were transferred from the his- 
torical passive subject to the historical oblique agent phrase. The 
former, however, retained control of verb agreement, and the zero case 
marking characteristic of subjects in other tenses, as illustrated by 
the following example from Hindi: 

(5li) Absolutive object, not ergative subject, controls agreement 

is larke ne pustak parhi 
this boy erg book(f abs) read(f) 
'This boy read a book.' ( Comrie, 139, p. 39) 

Comrie observes, however, that in some Indo-Iranian languages, control 
of agreement is being assumed by the historical passive agent, and that 
in Modern Persian, the historical passive agent has acquired all sub- 
ject properties. Thus, the development and subsequent loss of ergativity 
may illustrate the three stages of the acquisition of subjecthood. 

IV. Georgian 

So far, we have examined the acquisition of subject properties in 
Polynesian and Germanic . In both cases there is good reason to believe 
that behavioral properties were acquired prior to coding properties. 
Although stage A is not directly attested in either case, there is 


strong motivation for reconstructing a stage in which the relevant noun 
phrases lack any subject properties. 

We shall now turn to Georgian. Our discussion is based on Harris 
(19T6), Tschenkeli (1958) and Braithwaite (1973). We shall consider the 
status of dative experiencers like gelas in (55):^"^ 

(55) Dative experiencers 

gelas uqvars nino 

Gela(dat) him-loves-she Nino(nom) 
'Gela loves Nino.'-'-° 

In both Old Georgian and Modern Georgian, the dative noun phrase 
in (55), gelas , exhibits a nunber of behavioral subject properties. 
A change has taJten place in Modern Georgian which shows that these noun 
phrases are taking on morphosyntactic subject properties. Hence, data 
from Georgian corroborate our claim that coding subject properties are 
acquired subsequent to behavioral subject properties. 

We shall turn first to the behavioral subject properties of dative 
experiencers. Examples are given from Modern Georgian and Old Georgian. 
There are two relevant tests of behavioral subject properties. The first 
test for subjecthood which we shall consider is tav- reflexivization. 

As was shown by Harris , tav- may be coreferential only to the sub- 
ject of its own clause. This is illustrated for sentences with nomina- 
tive subjects in (56)-(57): 

(56) Direct objects do not control reflexivization 
mxa-^vari daxatavs vanos tavistvis 
painter(nom) he-paints-him VeLno(dat) self- for 

'The painter, will paint Vano. for himself^.' (Harris, 1:2) 

(57) Indirect objects do not control reflexivization 

nino acvenebs patara givis tavis 

Nino(nom) she-shows -him-him little Givi(dat) self's 

tavs sarkesi (Harris 1:3a) 

self(dat) mirfor-in 

'Nino, is showing little Givi herself^ in the mirror.' 

Sentence (56) shows that tav- can refer only to subjects and not to 
direct objects. Similarly, in (57) tav- may only refer to the subject 
Nino, 6ind not to the indirect object Givi . 

Turning to verbs taking dative experiencers, we see that it is 
the dative which is the antecedent of tav- . This is shown in Modern 
Georgian in (58^-(6o), ajid in Old Georgian in (61): 


(58) Dative experiencers control reflexivization 

temurs uqvars tavisi tavi 
Temur(dat) him-loves-he self's self(nom) 
'Temur loves himself.' (Harris 8:28a) 

(59) Nominatives do not control reflexivization 

*tavis tavs uqvars temuri 
self's self(dat) him-loves-he Temur(nom) 
('Himself loves Temur.') (Harris 8:28b) 

(60) Initial nominatives do not control reflexivization 

*tem;iri uqvars tavis tavs 
Temur(nom) him-loves-he self's self(dat) (Harris ,8:28c) 

(61) Dative experiencers control reflexivization 

cudad mogimedgrehies tavi 

poorly you-give-up-it self(nom) (^_ ^^^^. 3^^^! 

'You have given up on yourself hadly.' communication) 

Sentence (58) shows that the nominative may not be the antecedent. 
It should be noted that these facts cannot be due to word order. In 
(60) the nominative noun phrase precedes the dative, but the former still 
cannot serve as the antecedent for tav- reflexivization. Thus, dative 
experiencers are treated as subjects by tav- reflexivization. 

We shall now consider a second rule that treats dative experiencers 
as subjects (rather than as indirect objects). This r\ile is causative 
clause \mion. Causative clause union is a nile, found in many languages, 
which maps biclausal structures like (62): 

onto simplex strucstures like (63): 

(63) s'-'^^l cause-verb NP^...] 
In Georgian, as in many other languages, though not all, when the comp- 
lement verb is intransitive, the subject of that verb is a direct 
object in the output of clause union. This is illustrated in (6i+)-(65); 

(6U) Active intransitive 

bavsvi VI inebs 

child(nom) he-sleeps 

'The child is sleeping.' (Harris 5:l'+a) 


(65) Complement subject of intransitive becomes direct object 
in causative clause union 

dedajn bavsvi daajina 

mother(erg) child(nom) she-caused-sleep-him 

'The mother got the child to sleep.* (Harris 5:l6a) 

(Direct objects receive nominative case with certain verbs and tenses. 
See Harris, Chapter 1, for details.) 

In contrast, the subjects of transitive verbs are indirect objects 
in the output of the rule. This is shown in (66)-(67): 

{66) Active transitive 

mziam cecxli daanta 
Mzia(erg) fire(nom) she-lit-it 
'Mzia lit the fire.' (Harris 5:17a) 

(67) Complement subject of transitive becomes indirect object 
in causative clause union 

mamam mzias daantebina cecxli 

father(erg) Mzia(dat) he-ca\ised-light-her-it fire(nom) 
'Father made Mzia light the fire.' (Harris 5:19a) 

Indirect objects of intransitive verbs remain indirect objects, as is 
seen in (68)-(69). (The effect of the riole on indirect objects of tran- 
sitives is not relevant to the present discussion. ) 

(68) Active intransitive with indirect object 

komisia vanos estumra 
commission(nom) Vano(dat) it-visited-him 
'The commission visited Vano.' (Harris 5:20b) 

(69) 10 remains 10 in causative clause union 

direktorma komisia vanos astumra 

director (erg) commlssion(nom) Vano(dat) he-caused-visit-him-it 

'The director had the commission visit Vano.' (Heirris 5:22b) 

We have seen that grammatical relations in the output of clause 
union are determined by the transitivity of the complement clause and by 
the grammatical relations of the noun phrases in the complement in the 
input to the rule. Thus, with intransitive sentences with dative exp- 
eriencers like (70), 

(70) Dative experiencer 

tusaijs sioda 

prisoner (dat) him-hungered 

'The prisoner was hungry.' (Harris 8:38a) 


if the dative nominELl is a subject in the input to clause vmion, it is 
predicted that it will be a direct object in the output of the rule. 
But if the dative nominal is an indirect object, it should remain an 
indirect object subsequent to clause vmion. (Compare (68)-(69).) As 
is seen in (Tl), dative nominals in sentences like (70) become direct 
objects : 

(Tl) Dative experiencer becomes direct object in causative 
clause union 

tusajfi moasives 

prisoner(nom) pro-cause-hunger-him 

'They starved the prisoner. '/'They let the prisoner go hvmgry.' 

Therefore, we conclude that dative experiencers are treated as subjects 
and not indirect objects by causative clause union. 

We have shown that dative experiencers behave in a manner typical 
of subjects rather than of indirect objects. This is true not only in 
Modern Georgian but in Old Georgian as well. We shall now turn to 
coding properties. In Old Georgian these nominals displayed no subject 
coding properties. Rather, they manifest coding properties typical of 
indirect objects. In Modern Georgian, however, a change appears to 
have taken place. Dative experiencers behave as subjects rather than as 
indirect objects with respect to certain coding properties. 

There are three relevant coding properties: case marking, person 
agreement, and n\amber agreement. With respect to case marking and 
person agreement, in sentences like (55) (repeated), 

(55) Dative experiencers 

gelas ugvars nino 
Gela(dat) him-loves-she Nino(nom) 
' Gela loves Nino . ' 

the nominative noun phrase, nino , manifests subject coding properties 
in both Old Georgian and in Modern Georgian. Note that nino is in 
nominative case and triggers subject person agreement on the verb. 
The situation with regard to number agreement, however, is more complex. 
Verbs agree in nvimber with any first or second person subject, direct 
object, indirect object. But they agree in number with third person 
noun phrases only when they are subjects. Third person niamber agreement 
is illustrated in (T2)-(T3): 

(72) Number agreement (Third person subject) 

a. darca 

'He remained.' (Harris 9:iia) 

b. darces 

'They remained.' (Harris 9:5a) 


(73) Number agreement (Third person non-subject) 

a. icnobs cems studentebs 
he-knows-it my students (dat) 
'He knows my students.' (Harris 9:6a) 

b. acukebs tavis cigns cems studentebs 
he-gives -him-it self's book(dat) my students (dat) 
'He is giving his book to my students.' (Harris 9:6b) 

Thus, number agreement distinguishes between third person subjects and 
third person non-subjects. Only the former govern number agreement. 

We shall now consider the status of the two noun phrases in sen- 
tences like (55). Does the verb agree with the nominative noun phrase 
or with the dative experiencer? In Old Georgian (Tschenkeli: 1958, 1+53^U) 
the verb always agreed in number with the nominative noun phrase. 20 
This is shown in ilk). Note the pl\iral agreement on the verb. 

(7*+) Nominative governs nimiber agreement exclusively 

me miqvaran 

me (dat) me-love- they 

'I love them.' (Tschenkeli: n.l+^lt) ' 


In Standard Modern Georgian a change has taken place. V/hen both 
the dative experiencer and the nominative noun phrase are third person, ' 
the verb agrees with the dative rather than the nominative. Contrast 
(75) » in which the first person nominative controls number agreement 
and not the third person dative, and (76), in which the third person dat- 
ive , and not the third person nominative , controls verb agreement : 

(75) First person nominative controls niomber agreement 

a. mas vuqvarvar me 
him(dat) him-love-I l(nom) 
' He loves me . ' 

b. mas vuqvarvart_ even 
him(dat) him-love-we we(nom) 
' He loves us . ' 

c. mat vuqvarvar me i 
them (dat) him-loves-I l(nom) 1 
'They love me.' | 

d. mat vuqvarvart_ even J 
them (dat) him-loves-we we(nom) ^ 
•They love us.' (Tschenkeli: p.U59) t 


(76) Third person dative and not third person nominative controls 
number agreement 

a. mas uqvars is 
him(dat) him-lcves-he he(nom) 
'He loves him.' 

b. mas uqvars isini 
him(dat) him-loves-they they(nom) 
'He loves them. ' 

c. mat uqvart_ is 
them(dat) them- love-he he(nom) 
'They love him. ' 

d. mat uqvart isini 
them(dat) them-love-they they(nom) 

'They love them.' (Tschenkeli : P.U59) 

Hence, in Standard Modern Georgian, the dative experiencer has acquired 
partial control of number agreement, a coding property. 

In colloquial Modern Georgian, the acquisition of control of number 
agreement by the dative experiencer has, according to Tschenkeli (1958), 
been generalized. While in Standard Modern Georgian control of agreement 
by third person datives is limited to environments in vhich the nomina- 
tive no\in phrase is also third person, in colloquial Modern Georgian, 
this has been generalized for some speakers to environments in vhich the 
nominative noun phrase is first or second person. Thus, in (77) (from 
Harris (1976)), the dative experiencer triggers plural agreement despite 
the presence of a first person pronoun in the clause. 

(77) Dative governs number agreement regardless of person of 

msoblebs vuqvarvart_ (me) 

parents (dat) them- love-I l(nom) 

^Ity parents love me.' (Harris p. 283 fn.5) 

This example illustrates the extension of the control of number agree- 
ment by dative experiencers in environments not possible in Standard 
Modern Georgian. 

We have shown that in Old Georgian behavioral subject properties 
are controlled by the dative experiencer and coding properties by the 
nominative noun phrase. In Modern Georgian, especially in the colloquial 
language , one coding property , number agreement , has come to be controlled 
by the dative experiencer. Hence, Georgian corroborates our claim that 
nominals acquiring subject properties first acquire behaviroal properties 


V. Conclusions 

We have examined the historical process "by means of vhich, in a niam- 
ber of languages, subject properties were acquired by a noun phrase not 
previously having those properties. Among the languages examined vere 
Polynesian (Maori, Tongan, and Samoan), Germanic (Gothic, Icelandic, 
Faroese, Swedish, German, and English), and Georgian. 

From the languages surveyed, a common pattern emerges. There are 
three stages in the acquisition of subject properties. In the first 
stage, the noun phrase in question shows no subject properties. In the 
second stage, the noun phrase begins to take on transformational pro- 
perties associated in the language with subjecthood. During the third 
stage, subject coding properties are acquired. Among the coding pro- 
perties are control of such morphosyntactic processes as case marking 
and verb agreement. What distinguishes the third stage from the second 
is that the syntactic properties acquired during the third stage are 
reflected in the morphology of the language. Today's morphology, as 
Givon (19T1: ^13) and others have noted, is the ossification of yester- 
day's syntax. Hence, it is not surprising that morphosyntactic processes 
are acquired later than purely transformational processes. It is just 
such an order that we have observed. 


*The idea for this paper was suggested by the work of Yael Ziv 
(1976), who insightfully recognized the diachronic implications of 
Keenan's (1976) claims regarding the synchronic acquisition of subject 
properties in a derivation. A number of linguists offered us valuable 
comments and supporting data. These include Alice Harris, who supplied 
crucial data on Georgian (as well as helpful discussion of the data), 
and Anthony Britti, Sandy Chung, Bernard Comrie, Dee Holi sky, Hans H. 
Hock, Janice L. Jake, Jerry Morgan, Alan Timberlake, Rochelle Wright, and 
Ladislav Zgusta, who provided comments on the data and/or the analysis. 

We should also like to thank the National Science Foundation 
(grants SOC 75-002UU and BNS 77-27159) and the Research Board of the 
University of Illinois for supporting this research. Earlier versions 
of this paper were presented as a Forum Lecture at the 1977 Linguistic 
Institute in Hawaii, as well as at the 1977 Simmier Meeting of the LSA, 
the Third International Congress on Historical Linguistics (Hamburg), 
the linguistics colloquia of Cambridge University and Tel Aviv University, 
and at the Seventh All-India Conference of Dravidiaji Linguists, 
Bangalore, June 1977- 

"TThere are two possible ways of conceptualizing the changes dis- 
cussed in this paper. We are claiming that certain originally non- 
subject NPs are becoming subjects historically, and that they are 
therefore coming to be treated as such by rules which are sensitive to 
subjecthood. Alternatively, it might be claimed that the rules themselves 


are changing, starting to apply not only to subjects, but to certain 
non-subject NPs (e.g. experiencers ) as well. We reject the latter view 
primarily because of its iinnecessary complexity. It is generally the 
case that if one rule of a given type (e.g. a behavioral rule ) treats 
such NPs as it treats subjects, others of the same type will treat them 
similarly. The second of the above positions would require the positing 
of several changes in individual rules in order to account for this fact. 
The position that it is the NPs which are changing in grammatical role 
accounts for these parallel extensions of the applicability of indivi- 
dual rules as the consequence of a single change. Moreover, the rule- 
change position unnecessarily requires the weakening of the otherwise 
valid claim that certain processes, e.g. deletion by Equi , are cross- 
linguistically restricted to subjects (of. Keenan (197^) inter alia). 

For a fuller list of behavioral and coding properties , see 

Keenaji (19T6). The distinction between behavioral and coding properties 

is based on Keenan 's paper. Our paper provides additional support 

for this distinction. 

p.32U) lists sentence-initial position as a specific 
coding property of subjects. Even if this claim is assumed to be 
correct, and if initial position can be shown to be acquired prior to 
behavioral subject properties, our major contention, that there is a 
language-independent order of acquisition of the form: (l) behavioral 
subject properties, and (2) all coding properties aside from word order, 
remain \maffected. 

h „ 

Although we refer to these nominals as passivized non-accusative 

objects" for all the Germanic languages, we should like to emphasize 
that for languages like German in which these NPs have acquired no_ 
subject properties, this terminology is somewhat misleading. In the 
case of Stage A, the term "passivized non-accusative objects" should 
only be taken as a means of identifying those nominals that at a later 
stage come to demonstrate derived subject properties in the output of 

The question of the internal grouping of the Germanic languages 
has not been settled with certainty. Various investigators have ad- 
vocated a tripartite division into North, West, and East Germanic, a 
bipartite division into West and Northeast Germanic, and a division 
into East and Northwest Germanic. The arg\iments for these divergent 
positions have been reviewed in Kufner (19T2). This issue does 
not affect our argviment. 

The evidence regarding the acquisition of behavioral and coding 
properties by passivized non-accusative objects in Gothic is conflicting. 
In most instances, such objects seem to have acquired properties of both 
types. That is, they seem to have become true subjects. Compare the 
following examples : 


(i) ei Ceisl in filuwaurdein seinai andhatisjaindau 
that they(nom) in verbosity refl. VCdatD pass 3pl 
'that they will be heard in their verbosity' (Mat 6:7) 

(ii) gaveisodai wavuijun daurawaurdos 
VCgen] part a\ix 3pl porters (nom) 
'Porters were appointed.' (Neh 7:l) 

Note that in (i), the passivized dative object not only appears as a 
nominative and controls verb agreement, but serves as the controller 
of reflexivization as well. In all such cases, however, the Gothic 
construction translates a normal Greek passive construction with a 
nominative subject. This suggests the possibility that the Gothic con- 
struction has been influenced by the Greek, especially in view of the 
fact that in the few instances in which the Gothic NPs did not acquire 
coding properties, the Gothic constiniction was either independent of, 
or in opposition to the Greek, as in (iii) and (iv): 

(iii) J>airh t)atei is briikjaidau 
through that it VCgen 3 pass 
'through that it is used' /'through its use' (Col 2:22) 

(iv) jah bajoJ)um gabairgada 

and both (dpi) VCdatll pass sg 

'And both will be saved.' (Mat 9:1T) 

There are instances in which the reflexive possessive adjective 
refers to dative experiencers with verbs other than bykkia . Rose 
(19T6: 128) cites one example in particular involving the verb syna 
'seem'. She observes, however, that the reflexive possessive adjective, 
unlike reflexive pronouns, is used to refer to a variety of non-subject 
NPs, as in the following example: 

(i) hann skyldi fa henni eign sina 
he should get her(dat) possessions refl. 
'He should get her her possessions.' (Heimskringla 1:1+22) 

Thus, the ability of an NP to serve as the antecedent for a reflexive 
possessive adjective in Old Icelandic does not constitute a reliable 
subject property. 


There are a number of attestations in Old Icelandic in which 
experiencer predicates appear embedded lander subJect-to-obJect raising 
verbs , as in the following examples : 

(i) Grettir kvaj) ser J)at betr i)ykkia 

Grettir said her(dat) that better seem (inf) 

'Grettir said that to seem better to her.' (Grettis Saga, 39) 

(ii) Grettir kvaj) ser i)at allvel l£ka 

Grettir said her(dat) that well like (inf) 

•Grettir said that to please her well.' (Grettis Saga, T8) 


(iii) Hrolfr kua-zk ekke sua J)ykkia 

Hrolf(nom) said-refl not so seem (inf) 

'Hrolf said CitU not to seem so to him.' (cited in Heusler, 

1967: 39) 

In (iii), the \inderlying experiencer pronoun has cliticized to the main 
clause verb. Unfortunately, it does not appear possible to detennine 
whether in cases like the above the complement experiencer or the comp- 
lement nominative has been raised by SOR into matrix object position. 
It will be seen below that in Modem Icelandic , when a dative exper- 
iencer undergoes SOR, it retains its dative case. 

^See Andrews (1976: 13ff) for details. 

Passivized genitive objects, however, may not: 

(i) *eg vonast til ad verda bedid 

I hope comp aux inf VCgenI] supine 
( ' I hope to be waited for . ' ) 

"TThe information provided by Lockwood on experiencers in Faroese 
is insufficient for determining to what extent these have undergone 
the B to C shift. 


The experiencers in these languages display behavioral as well as 

coding properties. In the following Swedish examples, nominative exper- 
iencer subjects which developed from historically non-nominative exper- 
iencers control reflexivization, delete through Equi , and are raised by 
subject-to-subject raising; 

(i) han tyckte sig vara i en trevlig stad 
he(nom) thought refl be in a pleasurable city 
'He thought himself to be in a pleasvirable city.' 

(ii) Jag hoppas dromma 

I hope dream (inf) 
'I hope to dream, ' 

(iii) han tycks minnas 

he seems remember (inf) 
'He seems to remember.' 

The examples in (37) are cited by Harris as examples (lb) and 
(id) in the appendix to her paper. Our example (38) was cited by Butler 
(1976). We have \ised examples (3^*) ajid (36) rather than similar ones 
cited by Gaaf (190U: 32) and Harris (1975: 20) for chronological reasons. 
The examples which they cite axe of vmcertain significance, since they 
postdate the first occiirrence of the verbs involved with nominative 
experiencers, and since nominative experiencers appear with these verbs 
elsewhere in the works of the authors cited. Our examples (3*+) and 
(36) present no such problem, since they substantially antedate the first 
occurrences of these verbs with subject-coded experiencers. 


Some speakers of Austrian German allov passivized non-accusative 
objects to be deleted by Equi . For these speakers, sentences like (23) 
are grammatical. Thus, such NPs seem to be acquiring some behavioral 
properties in some dialects of German. 

C_ represents a set of consonants whose occurrence in the passive 
suffix is lexically determined. 

Also undeletable are the underlying DOs of passive sentences, 
since these do not meet the agency condition of the application of 
Equi , as in the following example : 

(i) *i hishia au ki te patu-a e Rewi 
past want I comp hit-pass agt Rewi 
('I wanted to be hit by Rewi.) (Chiing, 5^a, p. 150) 


Georgian also has a special mode described by Harris as "eviden- 
tial" (p. 226). This is the mode used where a speaker reports an event 
which he did not directly witness, but of which he saw the results of, 
or some evidence that it occurred. Using this mode also relieves the 
speaker of direct responsibility for the accuracy of the statement and 
Harris uses turme 'apparently' to gloss this mode. This mode usually 
appears in a perfective tense. Thus, certain verbs show up in a nominative- 
subJect/dative-DO and 10 pattern in the present, but take the dative 
case for the subject ajid nominative for DO (and a postpositional phrase 
for the 10) in the evidential mode. Thus, in (i) and (ii) the same rela- 
tions hold between the berb and the NPs. The only difference is in mode 
ajid in case marking: 

(i) glexi tesavs teslebs 

peasant (nom) he-sows -it seeds (dat) 
'The peasant is sowing seeds.' 

(ii) turme glexs dautesavs teslebi ' 

apparently peasant(dat) him-sowed-it(nom) seeds (nom) { 

'The peasant has apparently sown seeds.' | 

Though we will primarily discuss the status of the experiencer dative ; 
in this paper, all the facts hold also for evidential mode datives, and | 
we shall occasionally use examples from this construction. I 

18 S 

Harris glosses this sentence as: j 

(i) gelas uqvars nino I 

Gela(dat) he-loves-her Nino(nom) j 

(he=Gela, her=Nino) '■, 

However, in our gloss we have, for the sake of clarity, decided to use 
nominative pronouns for those markers which agree with nominative nouns, 
and dative pronouns for those agreeing with dative-nomdnals . This is 
closer in spirit to what actually happens in Georgian: the markers re- 
ferring to the dative are object markers and those referring to the nom- 
inative are subject markers . We do not give the markers and the verb root 


in the same order in vhich they appear in Georgian (where the ohject marker 
usually precedes the root and the subject marker follows the root) due 
to the fact that the markers are sometimes discontinuous (i.e., circiom- 
fixes) and sometimes morphemes. 


The verb can also mean 'weaken' . In the verb form, -£- is the 

second person (singular) object marker: - es is the third person sin- 
gular subject marker, agreeing with tavi . The verb is in the perfect 
or first evidential, which is part of Series III. The initial subject 
is 'you' , which undergoes inversion and becomes the final indirect 
object, triggering indirect object agreement and occiirring in the dative 
case. See fn. IT for a brief discussion on inversion (in the evidential 
mode ) • 

The same facts are reported by Braithwaite (1973: 110 ) (our 
glossing) : 

Old Georgian (i) mas mohuklavs_ igi 

him(dat) him-killed-he he(nom) 

'Apparently, he has killed him.' (-s_ marks 3sg) 

(ii) mas mohuklavan igini 

him(dat) him-killed-they they(nom) 
'Apparently, he has killed them.' (-an marks 

3pl subject) 

Here we see that in Old Georgian the derived subject (Nom) triggers 
third person number agreement. In Modern Georgian, the Nom does not 
have control over third person number agreement anymore and as a result, 
even though we have a plural Nom NP, the verb is marked for third person 
singular subject in agreement with the dative NP, as observed in (76b). 


Georgian differs from Germanic in that we have no evidence for a 

stage in which dative experiencers show no_ syntactic subject properties. 

We hypothesize, however, that there was such a stage. 


Anderson, Steven R. 1977. On the mechanisms by which languages become 

ergative. In C. Li, ed. : Mechanisms of syntactic change. Austin, 

Texas: Texas Press. 
Andrews, Avery. 1976. The VP complement in Modem Icelandic. In 

Montreal working papers in linguistics, vol. 6. 
Bennett, J.A.W. and G.U. Smithers (eds.). 1966. Early Middle English 

verse and prose. Oxford: Clarendon Press. 
Braithwaite, Kim. 1973. Case shift and verb concord in Georgian. 

Dissertation, University of Texas at Austin. 
Breckenridge, Janet. 1975. Rules which nothing undergoes. Bachelor's 

thesis. Harvard University: Cambridge, Mass. 


Butler, M.L. 1976. The reanalysis of Middle English impersonal 

constructions and the characterization of 'suhject of a sentence.' 

In Carlota S. Smith and Susan F. Schmerling, (eds.): Texas Lingustic 

Forum, vol. k. Austin. 
Chung, S.L. 1976. Case marking and grammatical relations in Polynesian. 

Dissertation, Harvard University: Cambridge, Mass. 
Comrie, B. 1977. Ergativity. Mimeo. 
Ebert, R. 1975. Subject raising, the clause squish, and German scheinen 

constructions. CLS 11. 
Evans, E.E. 1975- Psychological process experiencer verb impersonals : 

a case grammar approach. Dissertation, University of Texas. 
Falk, Hjalmar and Alf Torp. I9OO. Dansk-Norskens syntax. Kristiania 

H. Aschehoug. 
Gaaf , W. van der. 190i|. The transition from the impersonal to the 

personal construction in Middle English. Heidelberg. 
Givon, Talmy. 1971. Historical syntax and synchronic morphology: an 

archaelogist's field trip. CLS 7. 
Harris, Alice. 1975. On the mechanism of syntactic rule loss: the 

evidence from English and Udi. Unpublished paper. 
1976. Grammatical relations in Modern Georgian. 

Dissertation, Harvard University: Cambridge, Mass. 
Heusler, Andreas. 1967. Altislandisches Elementarbuch. 7th edition. 

Heidelberg: Carl Winter. 
Keenan, E. 1976. Towards a vmiversal definition of 'subject'. In 

C. Li, (ed.): Subject and Topic. Academic Press. 
Kufner, Herbert L. 1972. The grouping and separation of the Germanic 

languages. In Frans van Coetsom and Herbert L. Kufner, (eds.): 

Toward a grammar of Proto-Germanic. Tubingen: Max Niemeyer. 
Lockwood, W.B. 196h. An introduction to Modem Faroese. Copenhagen: 

Nygaard, M. 1905- Norr^n syntax. Kristiania: H. Aschehoug. 
Rose, M.L. 1976. The syntax of the reflexive pronoun in Gothic and 

Old Norse. Ph.D. Thesis, University of Wisconsin. 
Streitberg, Wilhelm (ed.) 1971. Die gotische Bibel. 6th edition. 

Darmstadt: Wissenschaftliche Buchgesellschaft. 
Tschenkeli, Kita. I958. Einfiihrung in die Georgische Sprache. Zurich: 

Ami rani Verlag. 
Ziv, Yael. 1976. On the reanalysis of grammatical terms in Hebrew 

possessive constructions. In P. Cole, (ed.): Studies in Modern 

Hebrew and Semantics, pp. 129-152. North-Holland Pub. Co. 

studies in the Linguistic Sciences 
Volume 8, Number 1, Spring I976 


Peter Cole and Janice L. JaJce 

This paper examines tvo impersonal constructions in Imbabura 
Quechua in which a noun phrase in accusative case displays a 
variety of syntactic subject properties. One impersonal constuc- 
tion is governed by lexical, or non-derived, experiencer verbs. 
In the other construction experiencers are formed by adding the 
desiderative suffix - naya- to a verbal stem. Although in both 
constructions the accusative experiencer nominal possesses certain, 
though not all, properties of subjects, the two constructions 
differ with regard to which subject properties are displayed in 
each, indicating that the nominals in - naya- desideratives have 
reached a more advanced stage in reanalysis from non-subject to 
subject than have those with lexical verbs of experience. 

1. Introduction 

In Imbabura Quechua (IQ) , and in other varieties of Quechua as 
well, subjects typically appear in nominative case and control verb 
agreement. There are, however, two 'impersonal constructions in IQ 
in which a noun phrase in accusative case displays a variety of syn- 
tactic subject properties. The two constructions are illustrated in 
(1) and (2). 

(1) Experiencer verbs'^ 

a. can-da rupa-n 
2sg-acc burn-3 
'You are hot. ' 

b. nuca-ta nana-n 
Isg-acc hurt-3 
'I hurt.' 

(2) - naya- desideratives 

a. can-da micu-naya-n 
2sg-acc eat-desid.-2 
'You would like to eat.' 

b. nuca-ta punu-naya-n 
Isg-acc sleep-desid.-3 
'I would like to sleep.' 

The sentences of (l) are instances of what we shall refer to as the 
lexical, or non-derived, experiencer construction. In addition to 
rupa- 'to be hot' and nana- 'to hurt' (intransitive), chiri- 'to be 
cold', yar j a- 'to be hungry' and muna- 'to wsint' may be used in this 


construction. The sentences of (2) are - naya- desideratives . These, 
are formed by adding the desiderative suffix - naya- to a verbal stem. 

The purpose of this paper is to sketch the morphological and syn- 
tactic properties of the nominals in each of the above constructions. 
We shall argue that in both constructions the 'experiencer' nominal 
(can-da and fiuca-ta in (l) and (2)) possesses certain, though not all, 
properties of subjects in IQ. Furthermore, the two constructions 
differ with regard to which subject properties are displayed in each. 
We shall show that the nominals in - naya- desideratives have reached 
a more advanced stage in reanalysis from non-subject to subject than 
have those with lexical verbs of experience. 

Our discussion is organized as follows: In each section, we deter- 
mine the extent to which the nominals in question are treated as sub- 
jects or as objects by grammatical processes found in IQ. Different 
groups of rules are considered in each section. In section 2 we examine 
the morphosyntactic properties of the experiencer nominals in the con- 
structions with regard to the subject coding rules^ of Case Marking (CM) 
and Verb Agreement (VA) and the object coding rule of Pronominal Object 
Cliticization (POC). Although experiencer nominals are typically coded 
as non-subjects by these riiles , those in the desiderative construction 
may, and vmder certain conditions must, display partial subject coding 
properties. TViis is not the case for nominals with non-derived exper- 
iencer verbs like those of (l). 

In section 3 we compare the nominals in the two constructions with 
regard to three rules: Equi NP Deletion (Equi), Subject Raising (SR) 
and Passivization. Nominals in both construction undergo the rules 
typically restricted to subjects, Equi, SR, and demotion by Passivization, 
and fail to undergo promotion to subjecthood by Passivization, this 
being restricted to objects. Thus, the experiencer nominals in both 
constructions behave imambivalently as subjects with regard to the rules 
discussed in this section. 

Two constructions in which subjects and non-subjects are treated 
differently are discussed in section h. These are switch reference 
ptirpose and adverbial clauses. In these constructions, nominals with 
- naya- verbs behave unambivalently as subjects. Nominals with lex- 
ical experiencer verbs, however, may be treated either as subjects or 
as non-subjects. 

In section 5, we consider the implications of the above facts for 
the diachronic and synchronic analysis of the two constructions. We 
argue that experiencer nominals with - naya- verbs are in the final 
stages of reanalysis as subjects. Nominals with lexical verbs of exper- 
ience are at an earlier stage of reanalysis. This position is supported 
not only by data from IQ, but also by cross-linguistic evidence. 

C. Morphosyntactic coding rules 

In this section we shall consider the behavior of experiencer 
nominals in the two constructions with regard to three morphosyntactic 
coding rules: Case Marking, Verb Agreement and Pronominal Object 


Cliticization. Experiencer nominals in both constructions behave as 
non-subjects with regard to VA. This is not true, however, with respect 
to CM or POC. 

With the exception of the two constructions under consideration, 
subjects appear in nominative case (0 case marking), control VA and 
fail to under POC. Cirect objects receive accusative case (-ta) , 
fail to control VA and may, iinder appropriate conditions, optionally 
undergo POC.' These facts are illustrated in (3)-(T): 

(3) Subject in nominative case and controls VA 

can-0 nuca-ta ricu-ngui 
2sg-nom Isg-acc see-2sg 
'you see me. ' 

(U) Object cliticized 

can-0 (nuca-ta) ricu-wa-ng\ii 
2sg-nom Isg-acc see-l-2sg 
'you see me. ' 

(5) Object may not control VA 

*can-0 nuca-ta ricu-ni 
2sg-nom Isg-acc see-lsg 
(You see me. ) 

(6) Subject may not appear in accusative case 

*can-ta nuca-ta ricu-ngui 
2sg-acc Isg-acc see-2sg 
(You see me. ) 

(T) Subject may not undergo cliticization 

*iiuca-0 can-da ricu-wa-ni 
2sg-nom 2cg-acc see-1-lsg 
(I see you. ) 

To return to experiencer constructions, with both desideratives 
and lexical verbs of experience the experiencer nominal is frequently 
coded as a superficial direct object. It receives accusative case, 
fails to trigger VA and, in the case of lexical experiencers , may under- 
go POC. This is shown in (8)-(ll): 

(8) - naya- experiencer in accusative and fails to constrol VA 

nuca-ta-ca can-da ricu-naya- •< ^ .\ 

Isg-acc-topic 2sg-acc see-desid- 
'I wo\ild like to see you.' 

i*lsg I 


(9) Lexical experiencer In accusative and fails to control VA 

'nuca-ta) nana-<. .C 

Isg-acc hurt-/,^g\ 
' I hurt . ' 

(10) Lexical experiencer undergoes POC 

a • ( f!uc a-t a ) nana-va-n 
Isg-acc hurt-1-3 
'I hurt.' 

b. (nuca-ta) chiri-va-n 
Isg-acc cold-1-3 
'I am cold. ' 

Desiderative experiencers differ from lexical experiencers in 
that the desiderative experiencer nominal may optionally appear in 
nominative case and can never undergo POC. The option of nominative 
case marking does not exist for lexical experiencers .8 Furthermore, 
desiderative experiencers differ from lexical experiencers in that the 
former fail to undergo the object coding rule of POC. This is illustrated 
in (11) (cf. 10): 

(11) - naya- experiencer fails to undergo POC 

a. *(nuca-ta-ca) can-da ricu-naya-wa-n 

Isg-acc-topic 2sg-acc see-desid-1-3 
(l would like to see you.) 

b. *micu-naya-wa-n 


(I vould like to eat.) 

c . 'punu-naya-wa-n ,; 

sleep-desid-1-3 J 

(I would like to sleep.) i 

Despite their appearance in noiinative case, desiderative ex- ' 

periencers fail to control VA: ! 

(12) - naya- experiencer in nominative case but fails to control VA 

fiuca-0-ca can-da ricu-naya-J^ .C 

Isg-nom-topic 2sg-acc see-desid' 
'I would ].ike to see you.' 



(l3) Lexical experiencer may not appear in nominative 


'nuca-0-ca nana- .5 .> 

lsg-ncF,-topic hurt- ^ :.' ( 

(I hurt.) 

We have shown that experiencers with - naya- verbs and ex- 
periencers with lexicel verts of experience are treated in different 
ways by morphosyntactic coding rules, - naya- experiencers may option- 
ally be coded as subjects by CM and are obligatorily coded as subjects 
by POC (if POC applies). They fail, however, to trigger VA. In con- 
trast, lexical experiencers are invariably treated as objects. As will 
be seen below, the two constructions differ in a similar way for certain, 
though not all, syntactic rules as well. 

3. Some syntactic indications of subjecthood 

In this section, we shall examine the behavior of experiencer 
nominals in both constructions with respect to three syntactic rules : 
Equi , SR, and Passivization. The ability of an KP to undergo these 
rales has been shown cross-linguistically to be indicative of the gram- 
matical role of the nominal in question. Equi and SR take subjects 
as their input. The former rule deletes complement subjects under 
identity with a matrix noun phrase, while the latter promotes a comp- 
lement subject to matrix direct object. In contrast, Passivization 
promotes iinderlying object noun phrases to subjecthood in a manner to 
be illustrated below. 

For each rule, we shall first show that in IQ it is normally 
restricted to noun phrases with a certain grammatical role (subject 
for Equi and SR, object for promotion by Passivization, and subject 
for demotion by Passivization). We shall then show that experiencer 
nominals with both desiderative and lexical experiencer verbs behave 
like subjects rather than objects: They undergo deletion by Equi and 
promotion by SR; they undergo demotion by Passivization, but they fail 
to undergo promotion by Passivization. Thus, we claim, accusative 
experiencer nominals in both constructions exhibit significant subject 
behavioral properties. 

Sentences (ll+)-(l5) illustrate the application of Equi to sen- 
tences with nominative subjects: 

[ik) a. Input structxire to Equi 


i:pai-0. trabaJa-D gushta-n 

3sg-nom 3sg-nom work like-3 

Output of Equi 

pai-0. 0. trabaja-na-ta gushta-n 
3sg-nom work-infin-acc like-3 
'She likes to work.' 

(l^) a. Input structure to Equi 

:pai-0. mishqui-ta micu-H gushta-n 


X _i. uaj.— V. . 


3sg-nom Ssg-nom candy-acc eat like-3 

b . Output of Equi 

pai-0. 0. mishqui-ta micu-na-ta gushta-n 
Ssg-nom candy-acc eat-infin-acc like-3 
'She likes to eat candy,' 

Sentences (i6)-(1T) shov Equi does not apply to objects: 

(16) Complement object coreferntial to matrix subject 

rmca-0. ELuna-ni pCpai-0 nuca-ta. ricu-chunD 
Isg-nom want-lsg 3sg-nom Isg-acc see-subjvmctive 
'I want her to see me.' 

(17) Equi fails to apply 

*nuca-0. muna-ni pai-0 0. ricu-na-ta 
Isg-nom want-lsg 3sg-nom see-infin-acc 
(I want her to see me.) 

Equi can, however, delete the experiencer nominal in both the 
desiderative and the non-derived experiencer construction: 

(18) a. Input structure to Equi with lexical experiencer 

warmi-0. mana gushta-n-llu Cwarmi-ta. nana-ju-3 
woman-nom not like-3-neg woman-acc hurt-prog 

b. Output of Equi 


woman-nom not like-3-neg hurt-prog-infin-acc 

'A woman doesn't like to hurt.' 

(19) a. Input structure to Equi with - naya- desiderative 

wira warmi-0. mana gushta-n-llu gCwira warmi-ta. micu-nayal 
fat woman-nom not gushta-n-llu fat woman-acc eat-desid 


(19) b. Output of Equi 

wira wanni-0 mana gushta-n-llu 0. micu-naya-na-ta 
fat woman-nom not like-3-neg eat-desid-infin-acc 
'The fat woiri£.n doesn't like wanting to eat.' 

Examples (l8)-(l9) show that unlike object nominals, experiencers in 
both of the constructions in question may be deleted by Equi. Thus, 
Equi treats the experiencer nominals in these constructions like sub- 
jects rather than objects. 

The rule of SR also distinguishes between accusative experiencers 
on the one hand and objects of the other. 9 SR applies to nominative 
subjects and to - naya- and lexical experiencers. The examples of (20) 
show the effect of SR on nominative subjects: 

(20) a. Input to SR 

pai-0 „Cnuca-0 micu-j-ta3 cri-rca 
3sg-nom Isg-nom eat-pres . comp-acc believe-3past 

Output of SR 

pai-0 nuca-ta cri-rca micu-j-ta 

3sg-nom Isg-acc believe-Spast eat-pres. comp-acc 

'He believed me to eat.' 

*pai-0 nuca-0 cri-rca micu-j-ta 
3sg-nom Isg-nom believe-Spast eat-pres .comp-acc 
(He believed me to eat.' 

d. Application of POC to (20b) 

pai-0 cri-wa-rca micu-j-ta 

3sg-nom believe-l-3past eat-pres. comp-acc 

'He believed me to eat.' 

In general, the application of SR can only be detected on the basis 
of word order (which is relatively free, and hence a poor test), and 
of the application of subsequent rules: e.g., case marking, matrix 
clause POC and matrix clause Passivization. Note that in (20b) the 
underlying complement subject is in accusative rather than nominative 
case. In (20d) this noun phrase has been cliticized to the matrix 
verb. These facts indicate that the underlying complement subject has 
become the derived matrix object. 

Further indication that the vinderlying complement subject has 
become a derived matrix object is provided by Passivization The ex- 
amples of (21) show that these nominals may be promoted to matrix 
subject by Passivization. 


(21) a. Input to Passivization 

Maria-0 nuca-ta cri-n ^.rnilcu-shca-taD 
Maria-nom Isg-acc believc-3 eat-past part-acc 
'Maria believes me to have eaten, 

b. Output of Passivization 

nuca-0-ca Maria-0 cri-shca ca-ni 

Isg-nom- topic Maria-nom believe-past part be-lsg 

eat-past part-acc 

'I am believed by Maria to have eaten.' 

The sentences of (22) show that SR and subsequent POC and Passivi- 
zation are restricted to underlying complement subjects. Note that 
underlying complement objects may not undergo SR. This is demonstrated 
by the failure of these NPs to cliticize to the matrix verb or to be 
promoted to matrix subject by Passivization. (Accusative case marking 
is inapplicable as a test for SR because the complement object is marked 
accusative as a result of its grammatical role in the complement clause.) 

(22) a. Transitive object complement clause 

Jose-0 cri-n _CMaria-0 nuca-ta ricu-shca-taH 
Jose-nom believe-3 Maria-nom Isg-acc see-past part-acc 
'Jose believes that Maria sav me.' 

b. Complement object fails to cliticize to the matrix verb 

*Jose-0 cri-va-n Maria-0 ricu-shca-ta 
Jose-nom believe-1-3 Maria-nom see-past part-acc 
(Jose believes that Maria saw me.) 

c. Complement object fails to become matrix passive subject 

*nuca-0-ca Jose-0 cri-shca ca-ni 

Isg-nom-topic Jose-nom believe-past part be-lsg 

Maria-0 ricu-shca-ta 
Maria-nom see-past part-acc 

(I am believed by Jose for Maria to have seen me.) 

Vfe have shown in the preceding paragraphs that SR distinguishes 
between complement subjects and complement objects. The former may, 
subsequent to SR, be cliticized to the matrix, or become the matrix 
passive subject. The latter may not. Thus, SR, in conjunction with 
POC and Passivization, provides a syntactic test for the status of 
- naya- and lexical experiencer nominals . If these nominals vmdergo 


matrix cliticization and Passivization to matrix sutject, we are 
Justified in claiming that they are treated as sutjects rather than 
objects by SR. We 2hall now show that this is, in fact, the case. 

The senteiices of (23) show that accusative - naya- experiencers 
may be cliticized to the matrix verb. This is shown for lexical ex- 
periencers in (2ii). 

(23) a. Input to POC 

Jari-0 fiuca-ta cri-n micu-naya-shca-ta 
man-nom Isg-acc believe-3 eat-des id-past part-acc 
'The man believes me to have wanted to eat.' 

b. Output of POC 

jari-0 cri-wa-n micu-naya-shca-ta 
man-nom believe-1-3 eat-desid-past part-acc 
'The man believed me to have wanted to eat.' 

(2l+) a. Input to POC 

jari-0 nuca-ta cri-n chiri-shca-ta 
man-nom Isg-acc believe-3 cold-past part-acc 
'The man believes me to have been cold.' 

b. Output of POC 

jari-0 cri-wa-n chiri-shca-ta 
man-nom believe-1-3 cold-past part-acc 
'The man believes me to have been cold.' 

Similarly, experiencers in both constructions may undergo Passiviza- 
tion and become matrix subjects, as is seen for desideratives in (25) 
and lexical experiencers in (26). 

(25) a. Input to Passivization 

jari-0 nuca-ta cri-rca micu-naya-shca-ta 
man-nom Isg-acc believe-3 past eat-desid-past part-acc 
'The man believes me to have wanted to eat.' 

b. Output of Passivization 

fiuca-ca jari-0 cri-shca ca-ni 

Isg-topic man-nom believe-past part be-lsg 

eat-desid-past part-acc 

'I was believed by the man to have wanted to eat.' 

(26) a. Input to Passivization 

jari-0 can-da cri-n chiri-shca-ta 
mnn-nom 2sg-acc believe-3 cold-past part-aco 
'The mam believes you to have been cold.' 


(26) t. Output of Passivization 

can-ga jari-0 cri-shca ca-ngui 

2sg-topic man-nom believe-past part be-2sg 

cold-past part-acc 

•You were believed by the man to have been cold.' 

Thus, examples (23)-(26) indicate that accusative experiencers in 
both constructions are treated as subjects rather than as objects 
by SR. 

The final rule to be discussed in this section is Passivization. 
Sentences involving this rule were given above in conjunction with 
ovir discussion of SR. What is relevant to the present argument is 
that Passivization promotes the non-subject to derived subjecthood 
and demotes the underlying subject to non-sub.lect nominative. For 
example , in ( 27 ) , 

(27) nuca-ca wawa-ta micu-chi-rca-ni 
Isg-topic child-acc eat-caus-past-lsg 
•I fed the child.' 

the accusative noun phrase, wawa 'child', is the direct object. The 
verb agrees with fiuca 'I', the subject. In the output of Passivization 
as shown in ( 28 ) , 

(28) wawa-ca nuca-0 micu-chi-shca ca-rca 
child-topic Isg-nom eat-caus-past part be-3 past 
'The child was fed by me.' 

vava 'child' has been promoted to subject, as is shown by third per- 
son verb agreement. As was noted previously, the promotion aspect 
of Passivization fails to apply to nominative subjects while the de- 
motion aspect of the rule fails to app]y to objects. Hence, an NP 
that is promoted by Passivization may be assumed not to be a subject 
prior to the application of the rule and an NP that is demoted by the 
rule may be assumed to be a subject in the input structure. 

To turn to the two accusative experiencer constructions , the 
experiencer nominals fail to undergo promotion by Passivization, as 
is indicated for - naya- verbs in (29) and lexical experiencers in (30). 

(29) a. Input to Passivization 

nuca-ta puRu-naya-rca 
Isg-acc sleep-desid-3 past 
'I would like to sleep.' 


(29) b. Ungramniatical output of Passivizr.ti'jn 

*nuca-ca punu-naya-shca ca-rca-ni 

Isg-topic slcep-desid-past part be-past-lsg 
( I vas slept . ) 

(30) a. Input to Passivization 

fiuca-ta chiri-rca 
Isg-acc cold-3 past 
'I was cold. ' 

b. Ungrammatical output of PaEsivization 

*nuca-ca chiri-shca ca-rca-ni 
Isg-topic cold-past part be-past-lsg 
(I was colded. ) 

Thus , the experiencer nominals in both constructions behave like nom- 
inative subjects rather than like accusative objects with regard to 
promotion by Passivization. 

The experiencer nominals also behave as subjects with regard to 
the demotion aspect of Passivization. This is shown for - nay a- desi- 
deratives in (31) and lexical experiencers in (32). 

(31) a. Input to Passivization 

nuca-ta-ca mishqui-ta micu-naya-rca 
Isg-acc-topic candy-acc eat-desid-3 past 
'I would like to eat candy.' 

b. Grammatical output of Passivization of patient 

mishqui-ca nuca-0 micu-naya-shca ca-rca 
candy-topic Isg-nom eat-des id-past part be-3 past 
'Candy was wanted to eat by me.' 

c. Ungrammatical output of Passivization of experiencer 
*nuca-ca mishqui- ) > micu-naya-shca ca-rca-(ni) 

Isg-topic candy- |'^0'°? eat-desid-past part be-(3) past-(lsg) 
(I was eaten by candy / I was eaten candy.) 

(32) a. Input to Passivization 

nuca-ta-ca mishqui-ta m\ma-rca 
Isg-acc-topic candy-acc want-3 past 
'I wanted candy.' 


(32) t. Grammatical output of Passivization of patient 

mishqui-ca fiuca-0 muna-shca ca-rca 
candy-topic Isg-nom want-past part be-3 past 
'Candy was wanted by me.' 

c. Ungrammatical output of Passivization of experiencer 

*nuca-ca mishqui- f'f V- muna-shca ca-rca-(ni) 

Isg-topic candy- J^°'^c want-past part be-(3)past-(lsg) 

(I was wanted by candy / I was wanted candy.) 

The sentences of (3l)-(32) show that the experiencer nominal under- 
goes demotion by Passivization as would be expected for subjects. 
This is in contrast with the patient, which is promoted to derived sub- 
ject. Note that the patient may not undergo demotion nor the experiencer 

These facts are predictable from the hypothesis that in the input 
to Passivization the experiencers are parsed as subjects and the pat- 
ients as objects. But the counter-hypothesis that both accusative 
nominals are non-subjects would provide no explanation for the differ- 
ent behavior of the two groups of accusative nominals. Hence, we 
conclude that both aspects of Passivization provide evidence for the 
subjecthood of the experiencers in both constructs oas. 

In this section we have examined the effect of three grammatical 
rules, Equi, SR, and Passivization, on the two experiencer constructions. 
With respect to each of these rules, the experiencers in both construc- 
tions behaved syntactically like subjects rather than like non-subjects. 
The rules did not distinguish between the constuctions in any way. 
In the section which follows, we shall see that this is not the case 
for all sjTitactic rules. For certain rules, the experiencer nominals 
in the desiderative construction behave more like nominative subjects 
than do experiencer nominals with lexical verbs of experience. 

h. Switch reference constructions 

In this section we shall consider the behavior of experiencer 
nominals in two types of switch reference constructions : adverbial 
clauses and purpose clauses. In these constructions, the suffixes used 
when the embedded subject is coreferential with that of the main clause 
differ from those used when the embedded and matrix subjects are not 
coreferential. This is illustrated for nominative subjects in the 
examples which follow. 


(33) a. Adverbial clause: matrix and embedded subject coreferential I 
(nuca-0.) wasi-man chaya [^"^^P^] ^^^^_^_ can-da ricu-nl 

-coref adv 1 

Isg-nom house-dat arrive) \ -, o t„~ 

(,*-non~cor adv) Isg-nom 2sg-acc see-lsg 

'When I arrive home, I see you. 



(33) b. Adverbial clause: Matrix and crabeddcd subjects non-coref. 

nuca-0. vasi-man chayaA~"^P { can-0. fluca-ta. ricu-ngui 

1 (*snpa> _ ■ J 1 ° 

Isg-nom house-dat arrive}" '^'^'^"^"^^ \^\ you-nom Isg-acc see-2sg 
( -coref adverb ) "' ^ ^ 

'When I arrive home, you see ne.' 

c. Purpose clause: matrix and embedded subjects coref erential 

nuca-0. wasi-man ri-ju-ni (fiuca-0.) ^ can-da, ricu-j . ~^f^^^"^/ 

1 1 J C*-chun b 

Isg-nom house-dat go-prog-lsg Isg-nom 2sg-acc seeV~^°^^''' P^^P i 

C -non-cor piirpj 

'I am going home to see you.' 

d. Purpose clause: matrix and embedded subjects non-coref. 

nuca-0. wasi-man ri-ju-ni can-0, nuca-ta. ricu). .> 

1 '^ '^j 1 (*-ngapaj^ 

, , , . ^ „ T (-non-cor. purp"? 

Isg-nom house-dat go-prog-lsg 2sg-nom Isg-acc seei ^ j:- ^ v 

^ -^ '^ "^ ^ ^ ^-coref. purp J 

'I am going home for you to see me.' 

As is shovn in the examples of (33), the suffix - shpa is used in 
adverbial clauses in which the embedded and matrix subjects are core- 
ferential. When these NPg are not coref erential, the suffix - jpi is 
found. In purpose clauses - ngapaj indicates coref erence between matrix 
and embedded subjects. The suffix - chun shows non-coref erence. It should 
be noted that when there is merely coref erence between a non- subject 
nominal in the matrix clause and the embedded subject, - jpi and - chun 
are used, as was seen in (33b) and (33d). Similarly, - jpi and - chun 
are found when the matrix subject and an embedded non-subject nominal 
are coref erential: 

(3^) Coref erence between matrix subject and embedded non-subject 

a. Adverbial clause 

nuca-ta. ricuV , > fiuca-0. pai-ta, maca-s 

1 c*-shpa^ 1 ^ j 



^-non-cor. advJ . _ , -j. n ^ j. 

jsg-nom Isg-acc seeTL ^ ^ > Isg-nom 3sg-acc hit-lsg fut 

(^-coref. adv. J 

'When he sees me, I will hit him.' 

b. Purpose clause 

fiuca-0. pai-ta. maca-rca-ni pai-0 ^^^^Y*_ncar)a1 S 

, -J. J. -, -, (-non-coref purp/ 

Isg-nom 3sg-acc hit-past-lsg 3sg-nom cry < „ r- i- t 

'I hit him so he would cry.' 


We shall now employ svitch reference to determine whether ac- 
cusative experiencers are treated as subjects or as non-subjects. 
It will be shown that the accusative experiencer nominals in - naya- 
desideratives behave syntactically like nominative subjects with re- 
spect to the assignment of adverbial and purposive suffixes. In con- 
trast, accusative experiencers with lexical verbs of experience may be 
treated either as subjects or as non-subjects for piu-poses of adverbial 
and purposive suffix assignment. 

We shall first consider - naya- experiencers. The sentences of 
(35) show that accusative nominals with - naya- experiencers are viewed 
as subjects for pupposesi- of adverbial and purposive suffix assignment. 
This is shown by the use of - shpa and - ngapaj , which, as was seen above, 
are restricted to sentences in which the matrix and embedded subjects 
are coreferential. Note that the sentences of (35) would be xingramma- 
tical if - jpi and - chun , the suffixes used when matrix and embedded 
subjects are non-coreferential, were used. It will be seen below that 
this is not the case for lexical experiencer nominals. 

(35) Desiderative experiencers treated as subjects in switch 
reference clauses 

a. Adverbial clause: desiderative matrix clause 

(nuca-0.) chagra-pi trabajaf'^_. . L nuca-ta. 

^. n •, . , r -coref . adv. / -, 
Isg-nom field-m work ^^ ^ ^ V Isg-acc 
^ \J*-non-coref . adv^ 


'When I work in the field, I want to sleep.' 

b. Adverbial clause: desiderative embedded clause 

(nuca-ta.) punu-naya ^ ^"^ ?^ ^ nuca-0- ca. punu-ni 

1 -1 J .,r -coref. adv. ") 

Isg-acc sleep-desid^_^^^_^^^^^_ adv^ Isg-nom-topic sleep-lsg 

'When I want to sleep, I sleep.' 

c. PUcpOBe clause: desiderative matrix clause 

cunan tuta nuca-ta. punu-naya-n caya maimi 0. 
now night Isg-acc""" sleep-desid-3 tomorrow a lot 

C*-non-coref . purp._^ 
'I'd like to sleep tonight in order to work a lot tomorrow.' 


(35) (!• Fnrpose clause: desiderative embedded clause 

nuca-0-ca. ali micuna-ta rura-ni 0. 
Isg-nom-topic good food-acc make-lsg 

r -ngapaj? 
^*-ch\in J 

-desid f/c°^ef- P^- "^ 
t -non-corer. purpj 


'I make good food so that I vill want to eat.' 

Lexical experiencers exhibit a pattern different from that seen 
in (35). As is seen in (36), subject to certain variation from speaker 
to speaker to be discussed below, both sets of suffixes may be used: 
those indicative that the matrix and embedded subject are coreferential 
( - shpa and - ngapaj ) and those indicative of non-coreferentiality of 
subjects (- jpi and - chun ) : 

(36) Adverbial clauses with lexical experiencers treated as either 
subject or non-subject 

a. Lexical experiencer in matrix clause 

0. urcu-pi trabaja ) ~^ ?^ > jari-ta-ca. 

J. . . , \ -coref . adv. > 
mountain-m work f ^ j \ man-acc-topic 
(^-non-coref . adv.^ 

yacu-ta muna-rca 
water-acc waiit-3-past 


'While working on the mountain, the man wanted water.' 

b. Lexical experiencer in subordinate clause 

5uca-0. jambi-dur-man ri-rca-ni 0. nana 7 » • < 

•,. . . -, ^ ^ -. ^ ^ C-coref. adv. < 
Isg-nom medicine-agt-dat go-past-lsg hurt ^_non-coref . advA 

'I went to the doctor hurting.' 

Note that when the subject NPs are non-coreferential, as in (37), 
only - Jpi is used: 

(37) Adverbial clauses with lexical experiencer: non-corefer- 
ential subjects 

a. Lexical experiencer in matrix clause 

jari-0. urcu-pi trabaja ^^if^JaS 

C -non-core f. adv.^ 
man-nom mountain-in work Qk_coref. adv. J 


(37) a. (continued) 

wawa-ta-ca. yacu-ta muna-rca 

child-acc-topic water-acc want-past 

'While the man was working on the mountain, the child 
wanted water. ' 

b. Lexical experiencer in subordinate clause 

nuca-0. jambi-dur-man ri-rca-ni wawa-ta 
Isg-nom medicine-agt-dat go-past-lsg child-acc 


, , r -non-coref. adv.? 
^^^*l?-coref. adv. S 

'I went to the doctor when the child was hurting.' 

With regard to purpose clauses, some variation was found among the 
speakers consulted. In the examples of (38), the symbol (#) indicates 
that the sentence in question was accepted by one infonnant and rejected 
by another. 

(38) Purpose clauses 

a. Lexical experiencer in matrix clause 

fiuca-ta. trabaja-na-ta muna-n 0. culqui-ta 
Isg-acc work-infin-acc want-3 money-acc 

take ( -c°^e^- P^^' ? 
C# -non-coref. purp.J 

' I want to work in order to make money . ' 

b. Lexical experiencer in subordinate clause 

pai. larca-pi wambu-rca ama 0. ^^P^/#_(,h\in J 
3sg brook-in swim-3 past neg burn 
'He swam in the water in order not to be hot 

(#-chun J 

C*-coref. purp. '^ 
(,#-non-coref . purp.r 

The examples of (38) show two patterns. For one speaJier, switch 
reference in purpose clauses with lexical experiencers show the same 
pattern as in adverbial clauses : the experiencer NP may be analyzed 
either as a subject or as a non-subject. For the other speaker con- 
sulted, the lexical experiencer in a purpose clause could be treated 
only as a subject, a pattern like that found for - naya- experiencers. 
Parallel to what was found with adverbial clauses , when the lexical ex- 
periencer was not coreferential with the subject of the other clause, 

-Chun was invariably the only form accepted. 

(39) Purpose clauses vith lexical experiencers : non-coreferential 

a. Lexical experiencer in matrix clause 

nuca-ta-ca. taita-man. cara-na-ta muna-n 
Isg-acc-topic father-dat servi-infin-acc want-3 

sinlli 0. caC-^^^ .7 
J L -ngapaj ^ 

strong te£-^°"-f^^^- P^^P'^ 
C.*-coref. purp. J 

'I want to serve it to father so he vill be strong.' 

b. Lexical experiencer in subordinate clause 

nuca-ca. vava-ta. wambu-chi-rca-ni ama 0. 
Isg-topic child-acc swim-caus .-past-Is g neg 


{-Chun ^ 
*-ngapaj 3 

, C* -non-coref . adv.~? 
\*-coref. adv. j 

'I caused the child to svim so he wouldn't be hot.' 

We have shown in this section that - naya- desiderative exper- 
iencers and lexical experiencers differ in behavior with regard to 
switch reference constructions. Desiderative experiencers are invar- 
iably analyzed as subjects. Lexical experiencers, however, with some 
variation from speaker to speaker, may be analyzed either as subject 
or as non-subject with respect to switch reference. The synchronic 
and diachronic implications of these facts will be discussed in the 
following section. It is sufficient at present to note that a pattern 
similar to that found for W and POC (section 2) is found with switch 
reference: desiderative experiencers are (in this case, obligatorily) 
treated as subjects while lexical experiencers are (in this case, 
optionally) treated as non-subjects. 

5. Conclusions 

In the preceding sections we examined the degree to which two 
types of accusative nominals , - naya- desiderative experiencers and 
lexical experiencers, exhibit morphological and syntactic properties 
otherwise restricted in IQ to nominative subjects. The properties as- 
sociated with each type of experiencer are reviewed schematically 
in (UO). 



Subject Properties Experiencers 









Switch Reference Purpose 



Svitch Reference Adverbial 












yes=behaves like nominative subject 
no=does not behave like nominative subject 
#=optional or individual variation 

In (Uo) subject properties are divided into behavioral and coding 
properties. (Cf. footnote 5) Lexical experiencers exhibit some , though 
not all behavioral properties , while - naya- desideratives exhibit all 
behavioral properties, and, in addition some coding properties assoc- 
iated with IQ subjects. The question which naturally arises is whether 
there is an explanation for these facts. Does the pattern of subject 
properties seen in (Uo) reveal any significant diachronic or synchronic 
regularities in IQ? 

The distribution of subject properties seen in (Uo) appears to 
have a principled explanation. It has been shown in previous work 
(Cole et al., 1978) that when a historically non-subject nominal comes 
to be a subject, properties associated with subjecthood are often ac- 
quired gradually rather than as a single historical change. Further- 
more, in the process of historical change, behavioral subject properties 
are acquired prior to coding properties . The cross-linguistic patterns 
observed in Cole et al. (1978) suggest an analysis for the data described 
in this paper. The pattern seen in (i+O) is just what woiild be predicted 
on the basis of Cole et al. if both experiencer constructions are in- 
stances of grammatical change in progress. Viewed from this perspective, 
the distribution of subject properties would be taken as reflecting 
the degree of historical change of the experiencer from non-subject 
(presumably, object) to subject. But, although the accusative nominals 
in the two constructions are undergoing a similar change, the change 
would appear to be preceding indei)endently in each construction. In 
the case of - naya- experiencers , the change would appear to be nearly 
complete. This is shown by the fact that both behavioral and coding 
properties are associated with - naya- experiencer nominals. In contrast, 
lexical experiencer nominals appear to be at an earlier stage in the 
same process of change. 


The interpretation of (Uo) suggested by Cole et al., which consti- 
tutes , in effect , a partial internal reconstruction of the constructions 
in question, appears to be supported by the somewhat scant comparative 
data available. To the best of our knowledge, no detailed study has 
been made of the properties of experiencers in analogous constructions 
in other dialects of Quechua. Thus, we are not able to provide convinc- 
ing arguments that experiencers lacked all behavioral subject properties 
in common Quechua. But it does seem fairly clear that the subject 
coding properties associated with - naya- experiencers are an innovation. 
In nearly all Quechua dialects - naya- experiencers are found in acc- 
usative case. The possibility of nominative is quite exceptional. 
Thus, in the absence of convincing counter-evidence, such as relevant 
archaisiiis , the most likely reconstruction wouid be the one in which the 
property fovind in most of the d.aughter languages, accusative case, is 
hypothesized to be a property of the proto-language . It would seem, 
then, that comparative evidence suggests that - naya- experiencers are 
innovative vis-a-vis lexical experiencers, the same conclusion drawn 
above on the basis of dialect internal evidence. 

The synchronic analysis of the two constructions lends support 
to the iiypothesis that the desiderative construction represents a 
more advanced stage in historical change than does the lexical exper- 
iencer construction. Harris (to appear .,a) and Ferlmutter (to appear) 
inter alia have proposed that experiencer constructions like those 
examined here involve a rule, of Inversion. This rule demotes the un- 
derlying subject to object. Such an analysis is well-motivated syn- 
chronically, because it provides as explicit explanation for the fact 
that the experiencer NP has both subject and non-subject properties. 
If an Inversion analysis is accepted, there are certain interesting 
consequences. The synchronic analysis of the two experiencer con- 
structions, in conjunction with historical data presented by Harris 
on the development of experiencer constructions in other languages, 
provides additional evidence that the two experiencers are in transition 
from non-subject to subject, and that the - naya- experiencers have 
reached a more advanced stage than have the lexical experiencers. 

An example of how Inversion allows the association of super- 
ficial object properties with the underlying subject is given in 
(i+2) which is roughly the derivation of (ijl): 

(Ul) nuca-ta punu-naya-n 

Isg-acc sleep-desid-3 

(i+2) a. Underlying representation 

nuca (subject) pufiu-naya- 
Isg sleep-desid 

b. Output of Inversion 

nuca (object) punu-naya- 
Isg sleep-desid 


(U2) c. Output of Case Marking 

nuca-ta punu-naya- 
Isg-acc sleep-desid 

d. Output of Verb Agreement 

nuca-ta puiiu-naya-n 
Isg-acc sleep-desid-3 

Once an Inversion analysis is assiomed, certain rule orderings are 
necessary in order to generate the sentences found in the language. 
According to this analysis, desiderative and lexical experiencer con- 
structions differ only in terms of how Inversion is ordered with respect 
to other rules . The fact that desiderative experiencers are more sub- 
ject-like than lexical experiencers is explained by ordering Inversion 
later in the derivation for - naya- than for lexical experiencers. 
The ordering of rules needed for each construction is presented in 
(U3). We would like to emphasize that the rule ordering posited in 
(U3) is motivated on pixrely synchronic grounds. The necessity for these 
orderings is justified briefly in footnote 15- 

(Us) Ordering of Inversion 

a. Lexical experiencers 




#Switch Reference Purpose 
#Switch Reference Adverbial 




b. Desiderative experiencers 




Switch Reference Adverbial 

Switch Reference Prupose 



#indicates variability in ordering 

The rule ordering seen in (1+3) correctly predicts that - naya- 
experiencers will exhibit a wider range of subject properties than lex- 
ical experiencers. In general, the later Inversion is ordered vis-a-vis 
other rules, the more subject properties will be associated with the 


experiencer nominal. Thus, the Inversion hypothesis, in conjunction 
with the rule orderings proposed in (^3), accounts for the differences 
between the two constructions. 

We shall turn now to the diachronic implications of the difference 
in rule ordering in (H3a) and (U3b). Harris (to appear, b) examines 
grammatical changes leading ultimately to the loss of non-nominative 
experiencer constructions in two unrelated languages, English and 
Udi . Employing an Inversion analysis, Harris considers the synchronic 
state of the languages at various stages in their development. She 
shows that in both cases Inversion was ordered prior to morphosyn- 
tactic coding rules (case marking and verb agreement) at an early 
stage in the development of the language, but that at later stages 
the r\ile was successivelj' reordered so as to follow the coding rules. 
The effect of the reordering was to dininish, and eventually, to 
eliminate non-subect coding properties. Upon being reordered after 
all other rules, inversion ceased to have any surface effects, and 
the rule, presiimably, was lost. 

The changes reported for English and Udi provide cross-linguistic 
corroboration for the analysis oflQ proposed above. We suggested 
that - naya- experiencers are at a later stage in the acquisition of 
subjecthood than are lexical experiencers. We shall assi^me that IQ 
would be expected to undergo a process of change similar to that 
found by Harris in English and Udi. On the basis of this assumption, 
o\ir diachronic analysis predicts that Inversion should be ordered later 
in the derivation with - naya- experiencers than with lexical exper- 
iencers. This is, in fact, the rule ordering found in (US). 

To summarize our conclusions, - naya- desiderative experiencers 
manifest a considerably wider range of subject properties than do 
lexical experiencers. The differences between the two experiencer 
constructions suggest that, while experiencers in both constructions 
are in the midst of a change from objecthood to subjecthood, the change 
is more advanced in the desiderative experiencer construction than in 
the lexical experiencer construction. 


"The dialect of Quechua described in this paper is spoken in 
northern Ecuador. There are between iiO,000 and 60,000 speaker of 
IQ and between 6,000,000 and 8,000,000 speakers of Quechua as a whole. 
For a general description of IQ, see Stark (1973). 

This paper is based on data from speakers born and raised in 
the town of Mariano Acosta, in the eastern part of the Province of 
Imbabura. We would like to express our great gratitude to Carmen 
Chuquin, our primary informant, now a graduate student in linguistics, 
with whom we worked extensively at the University of Illinois, and 
to Mariana and Emilia Chuquin, with whom we worked during a visit to 
Otavalo (Imbabura) in January 1978. Some aspects of the data reported 
on in this paper may be restricted to the subdialect spoken around 
Mariano Acosta. 


v;e vculd like to thank Gaty Hermon, Wayne Harbert , and Chuck 
Kisseberth for their suggestions regarding the analysis of the data 
presented here and for comnents on earlier versions of this paper. 
We would also like to thank Jean Hannah for finding a variety of incon- 
sistencies and problems with the original manuscript. Her assistance 
in preparing the paper for publication was invaluable. 

This work was supported in part by the National Science Foundation 
(50c T5002Ui+), the Research Board of the University of Illinois and 
the Center for International-Comparative Studies of the University 
of Illinois. The assistance of these institutions is gratefully aclaicv- 

Due to a phonological voicing rule, the accusative suffix is 
pronounced - da following nasals and - ta elsewhere. 

The verb m\ina- 'want 'may also be used in 'personal' con- 
structions with a nominative experiencer, in which case it agrees with 
the nominative NP in person and number. 

In addition, - naya- may be added to a limited number of nouns, 

e.g. yacu 'water', to form a desiderative verb, yacu-naya- 'desire 

to drink, be thirsty'. 

Both lexical and desiderative verbs of experience may be con- 
verted into 'personal' verbs, tailing nominative experiencer. This is 
done by the addition of the morpheme - chi : nuca-ta punu-naya-n 
(experiencer in accusative, third person agreeement) but nuca punu- 
naya-chi-ni (experiencer in nominative, first person singular agree- 
ment ) . Both sentences are glossed identically as 'I want to sleep' 
by speakers. The suffix - chi is normally a causative morpheme, but 
that does not appear to be its function here. We shall have no fur- 
ther discussion of this construction. 

It should be noted that in - naya- constructions the experiencer 
is understood to be identical to the subject of the verb stem. That 
is, the construction may be used to express the notion 'I would like 
to eat ' , but not ' I would like Harry to eat ' . 

Following Keenan (1976), we distinguish between subject coding 
properties and subject behavioral properties. The former includes 
rules involving the morphosyntactic coding of gr amma tical relations, 
such as case marking and verb agreement. The latter includes non- 
morphological syntactic rules, such as deletion under identity and 
subject raising. 

For a justification of the distinction, see Keenan (19T6) and 
Cole et. al. (19T8). The distinction between coding and behavioral 
properties plays a major role in section 5. 

Passivization in Imbabura promotes an object NP to subject, de- 
motes transitive subjects to non-subject nominatives and makes approp- 
riate changes in verbal morphology. The two aspects of the rule, pro- 
motion and demotion, constitute tests for objecthood and subjecthood: 


Objects, both direct ana incirect , though i.ox obliques, nay be pronoted 
to sutject, but subjects may not. Subjects are demoted to non-subject 
noirdnatives by Fassivizaticn. 

'rcC in IQ, and in Ecuadorian dialects generally, is restricted 
to first person objects. In Peruvian dialects, second person objects 
may be cliticized as well. The rule is optional in IC. A full NP 
object may co-occur vith the clitic for eiri'hasis . 

the nond-ncitive, 

they are invariably marked as topics (-ca_). Topic marking is not re- 
stricted tc any single type of NP such as subjects. It is frequently 
used to disambiguate subject from non-subject, as in (i): 

fiuca-ta •• ^ 






'I want to see h 



The potential aonbiguity in the ve:^-sion of (i) without -ca arises 
from the fact that HPs in Quechua are frequently conjoined by juxta- 
position. Thus, the version of (i) without -ca is , normally, inter- 
preted as a conjoined object construction (unspecified subject). The 
use of - ca shows that nuca-ta and pai-t a are not a single constituent. 
Tlie topic marker has a similar furiction in the passive, as is shown below. 

SR is governed by approximately the same class of verbs that 

[;,cverns subject to object raising in English: verbs cf jei-ception 

(e.g., ricu- 'see', xiya- 'hear') and verbs of propositional attitude 

(e.g., cri- 'believe', yacha- 'know'). 

"'"Complement clauses and, in fact, nearly all subordinate clauses 
in IQ are non-finite. The verb of the complement clause receives a 
nominalising suffix which indicates relative time reference vis-a-vis 
the matrix clause: The suffix - shea indicates past time reference, 
-j_ and -^ indicate present and - na indicates future. The same suffix- 
es (with the exception of -^) are found in relative clauses. In addition, 
- shea has a role analogous to that of past participle in the formation 
of the perfect and of the passive. 

roc is not always a clear test for SR. With certain matrix 
clause verbs (e.g., muna- 'want') clitic climbing tal:es place: A comp- 
lement object cliticizes to the matrix verb, as in (i): 

(i) pai-0 muna-wa-n ayuda-na-ta 

3-sg-nom want-1-3 help-infin-acc 
'He wants to help me.' 

The verb cri- 'believe' is employed in (22) because clitic climbing 

is not poFsible with cri- , as is shown by the ungrammaticality of (22b). 

For further discussion of clitic climbing, see Kviysken (1977). 

"The retention of the optional noun phrase is, in this example. 


axid in several others in this section, grammatical but stylistically 


Ve were unable to obtain clear informant judgements regarding 

whether the assignment of the suffixes in question is strictly con- 
trolled by derived grammatical relations . Our informant failed to 
have consistent intuitions with respect to the suffixes needed in adverb- 
ial and purpose clauses subordinated to passive matrix clauses. 


According to Perlmutter, and Harris, the rule universally de- 
motes a subject to indirect object. The appearance of the underly- 
ing subject as a direct object in IQ (as evidenced by accusative case) 
would presiomably be acco\inted for by the subsequent promotion of the 
indirect object to direct object. We shall take no position on whether 
such a derivation is in fact justified in IQ. 

The examples presented previously provide the basis for the 
rule ordering statements which follow. It is assumed that rvtle order- 
ing is transitive: If A precedes B and B precedes C, then A precedes C. 

SR feeds Passivization, and thus, must precede it. Equi must 
precede Inversion or Inversion would bleed Equi. With lexical exper- 
iencers. Inversion may precede or follow both switch reference rules 
for some speakers. For others it must follow Switch Reference Purpose 
and precede Switch Reference Adverbial. Thus, Switch Reference Pur- 
pose precedes Switch Reference Adverbial. (see section U for details.) 
In some derivations , Inversion follows the switch reference rules but 
precedes POC. Thus, the switch reference rules must precede POC. 

With desiderative experiencers , Inversion may come between POC 
and CM. Hence, POC precedes CM. It may also follow CM but precede 
Verb Agreement. Therefore, Verb Agreement must follow CM. 

The historical acquisition of subject properties does not al- 
ways involve the reordering of a rule in the derivation. See Cole 
et al. for details. 


Cole, Peter, Wayne Harbert, Gabriella Hermon, and S.N. Sridhar. 19T8. 

On the acquisition of subjecthood. Studies in the Linguistic 

Sciences 7.2 
Harris, Alice. To appear, a. Inversion as a r\ile in universal grammar: 

Georgian evidence. Studies in Relational Grammar, ed. by David 

To appear, b. On the mechanism of syntactic rule loss: 

evidence from English and Udi. Studies in Relational Grammar, 

ed. by David Perlmutter. 
Keenan, Edward. 1976. Towards a imiversal definition of 'subject'. 

Subject and Topic, ed. by Charles Li. Academic Press. 
M\;iysken. 1977. Clitics as Moving Features. Manuscript. 
Perlmutter, David. To appear. Evidence for inversion in Russian, 

Japanese, and Kannada. Studies in Relational Grammar, ed. by 

David Perlmutter. 


Stark, Louisa and Lawrence Carpenter. 1973. El Quichua de Imbabura: 
una gramatica pedagogica. Institute Interandino de Desarrollo. 
Otavalo, Ecuador. 

studies in the Linguirtic Science: 
Voliur.e a, Nunber 1, l-pring 197 » 

Janice Jake 

The recent interest in ergativity has underi-cored the singular 
position of Dyirbal as a syntactically ergative language. Even 
amon^j the morphologically ergative aborigine languaceo of Australia, 
Dyirbal represents a unique situation. In most morphologically 
ergative languages it can be easily demonstrated that the morpho- 
logy' does not dii^ctlj'- represent the grammatical relations reflected 
by the sj-ntactic rules of the language, Dixon (1972), hofwever, has 
claimed that absolutive and ergative, rather than subject and object, 
are the relevant grammatical relations in Dyirbal, This paper exam- 
ines the evidence supporting Dixon's claim that Dyirbal is syntac- 
tically ergative and shows how this evidence is equally' supportive 
of a nominative-accusative anal^^sis in which subject and object are 
the grammatical relations referred to by the syntactic rules of 
Dj'irbai, In addition to the data provided by Subject-Subject-Equi- 
NP-Deletion, Conjunction Reduction, Helativization and Reflexiviza- 
tion discussed in Dixon (1972), Imperative Deletion and Object-Sub- 
ject-Equi-NP-Deletion offer two sets of data which can adequately 
be accounted for by a subject-object analysis, but not by an abso- 
lutive -ergative analysis. 

Prior to an examination of Dyirbal, a few examples from Tongan and Avar 
are presented to show the typical situation in morphologically ergative lang- 
uages. Then Dixon's analysis of Dyirbal as a syntactically ergative language 
will be coiT?3ar6d to an alternative subject-object analysis. The two analyses 
are then evaluated in terms of cross -linguistic evidence and the predicticxis 
they make for the other parts of the grammar of Dyirbal, Here not only do 
cross -linguistic precedence arguments favor the subject-object analysis, but 
the language internal evidence, conprlsed of two sets of data ^rtiich the subject- 
object analysis can adequately account for, but which the absolutive -ergative 
analysis cannot, includes a counterexample to Dixon's analysis. 

Ergativity has traditionally referred to languages in which intransi- 
tive subjects and direct objects ate marked by the same morphological form 
and transitive subjects are marked by another morphological form. In some 
languages morphological ergativity is indicated by case marking, as shown by 
the Tongan examples in (l): 

(1) Tongan (Chung, 1976a) 

a, Ergative Case Marking: 

Verb or »a Subject (intrans) 

e Subject or 'a Direct Object (trans) 

b. (intrans) »oku fu'f hela'ia 'a 'Alani 'i he ngaue 

prog very tired abs Alan caus the work 
'Alan is very tired because of work.' 


c, 'oku fa'u 'e hoku tolcuoa 'a e tepile 
prog make erg my sibling abs the table 
•>i3' brother is making a table.' 

Verb agreement can also be ergative, as it is in Avar: 

(2) Avar (Comrie, forthcoming) 

a. vas V- ekerule 
boy -abs sg/masc/abs-run 

• The boy runs , • 

b. jas J- ekerule 
girl-abs sg/fem/abs-run 
•The girl runs,' 

c. vas -as: jas j- ec:ula 
boy -erg girl-abs sg/fem/abs -praise 
"The boy praises the girl.' 

In morphologically ergative languages the NPs which are in nominative 
case, as in (l), and the NPs which control verb agreement, as in (2), have 
been referred to as absolutives. Transitive subjects are referred to as 

Since nominative case marking and control of verb agreement have been 
generally regarded as properties of subjects, the applicability of the notion 
subject has been questioned for ergative languages, A more detailed examin- 
ation of subjects shows that in addition to morphological properties like 
nominative case marking and control of verb agreement being cross -linguisti- 
cally characteristic of subjects, there are also certain syntactic properties 
associated with subjects. Some of the properties are the ability to delete 
under Equi-NP-Deletion and Conjunction Reduction, the ability to control but 
not undergo clause -bounded Ref lexivization, and the ability to vindergo Sub- 
ject -to-Subject Raising, In morphologically ergative languages in a transi- 
tive sentence, it is not the nominative NP which syntactically patterns with 
the intransitive subject, but the ergative NP. The examples in (3) show that 
Equi-NP-Deletion deletes subjects rather than absolutives in Tongan, 

(3) a. pea na'e "alu 'a e tangata'' 'o folau mama'o 

and past go abs the man coap sail far 

'Then the man went andsailed away.' _ 

b, na'e 'alu 'a e tangata'' 'o taa'i 'a e kuli 
past go abs the men comp hit abs the dog 
'The man went and hit the dog.' 

c, -x-na'e 'alu 'a e fefine'''o u'u 'e he kulI 

past go abs the woman comp bite erg the dog 
(The woman j^went the dog bit 0^)) 

In (li) both transitive and intransitive subjects, but not objects, undergo 
Subject-to-Subject-Raising in Tongan: 

(U) a. 'e lava 'a e pipe 'o lea 
\mm can abs the baby comp talk 
•The baby can talk,' 


b, 'oku ou lava 'o ongo'i 'a e tame 'a mei he fale ko e 
prog I can comp hear abs the music from the house pred that 
'I can hear the music (coming) from that house,' 

c, *na'e lava 'a e fefine 'o taa'i 'e Sione 

past can abs the woman comp hit erg ^ione 
(The woman^ can John hit 0.,) 

As exemplified by Tongan, morphological ergativity does not correspond 
to syntactic ergativity. The grammatical relation of subject is applicable 
to morphologically ergative languages. The term absolutive only reflects 
the common morphology of the intransitive subject and the direct object. 
Ergative is not a graranatical relation but a case marking used for transitive 

In order for a langiaage to be syntactically ergative the syntactic rules 
of the language must treat absolutive and ergative as grammatical relations. 
Johns en il91h) attempted to characterize a hierchary of grammatical illations 
which would be applicable to syntactically absolutive -ergative languages as 
well as subject-object languages, ■^ Grammatical relations were ranked, the 
Specific relation of each rank being determined by the particular grammatical 
relations a language possessed. Thus absolutive represents the highest gram- 
matical relation in Dj^irbal, parallel to the relation of subject in languages 
like English, The ergative is the equivalent of the second highest relation 
of direct object. In Dyirbal the dative case marks an oblique or chStaer simi- 
lar to the passive agent by-phrase in English, In the following discussion 
of Dyirbal this terminology will be adapted to^ facilitate a conqDarison of the 
analyses, Absolutive, ergative, oblique, subject, and direct object will 
refer to grammatical relations; nominative, instrumental, and dative will be 
reserved for case marking, ^ 

The subject of an intransitive sentence in Dyirbal is in nominative, 
as shown in (5), The sentences in (6) sho;^ the three possibilities for 
expressing an (at least underlj'ingly) transitive sentence. All three senten- 
ces fundamentally express the same thing: in (6a) the relationship between 
patient, agent, and verb is viewed as one event; in (6b) the dative NP is 
affected by the action of the agent^ in (6c) the degree of success of the 
action on the patient is not clear? 

(5) bayi yafa bani-pu 

cm n man n coin9<rnon-rut 
•man came' 

(6) a, balan ^ugumbil bangul yafangu balgan 

cm n woman n cm i man i hit-non-fut 

b, bayi ya^a bagun ^ugumbilgu balgal-i^a-^u 
cm n man n cm d woman d hit-nay-non-fut 

c, bayi ya^i bar^gun ^ugumbijni balgal-ija-pu 
cm n man n cm i woman i hit-nay-non-fut 
•man hit woman' 

In an analysis treating Dyirbal as an abs olutive -ergative language, 
(6a) would be ccnsidered to be closest to the underlying structure of a 
transitive sentence. The transitive subject is in inatrumental case, which 
corresponds to the ergative relation and the object is in nominative case, 
as is the intransitive subject in (5). The other two sentence types could 


be transformationally derived from the structure \inderlying (6a), In (7) 
Antipassive promotes the underlying ergative to absolutive and the underly- 
ing absolutive is demoted to an oblique NP. ° 

(7) a. balan (Jugumbil bai^gul ya^ar^gu balan 

cm n woman n cm i man i hit-non-fut s 

(absolutive) (ergative) 
b, bayi yafa bagun cjugumbilgu balgal-J)a-/iu 
cm n man n cm d woman d hit-nay-non-fut 
(absolutive) (oblique) 

After Antipassive has applied to promote the underlying ergative and demote 
the underlying absolutive, a rule of Dative Promotion could apply to promote 
the oblique NP to an ergative relation as shown in (8), 7 

(8) a* bayi yafa bagun ^ugumbilgu balgal-^a-_pu 

cm n man n cm d woman d hit-nay-non-fut -_^ y 
(absolutive) (oblique) 
b« bayi ya^a bar^gun ^uguii4)ipi balgal-na^^u 
cm n man n cm' i woman i hit-ijay-non-fut 
(absolutive) (ergative) 

This absolutive -ergative analysis of Dyirbal accovmts for the fact 
that only absolutives can be deleted by Equi-NP-Deletion and Ccxi junction 
Reduction and that only absolutives can undergo Re lativization and trigger 
but not undergo Ref lexivization, as shown in the following examples. 

Equi-NP-Deletion: only nominative NPs delete 

(9) balan cjugumbil barjgul yaj:ar)gu manda-n 
cm n woman n cm i meui i take -non -fut 
•man took woman* 

(10) a. balam miran baijgun (^ugumbipi babi-n 

cm n beans n cm i woman i scrape -non -fut 
b« balan ^ugumbil bagum mira^gu babil-ija-pu 

cm n woman n cm d beans d scrape -na-non-fut 
'woman scrape beans ' 

(11) a. *balan cjugumbil bangul ya^angu raunda-n " balam miran babi-n 

cm n woman n cm i man 1 take-non-fut cm n beans-n scrape-non-f 
b. balan ^ugumbil baifgul ya^angu munda-n bagum mir^gu babll-i^a-pu 

-" '-■ - take-ncai-fut cm d beans d scrape-nay-no 

si ' fll 

cm n woman n cm i man i 
man took woman to scrape beans' ' fu 

The ungrammaticality of (lla) shows that the instrumental NP in (lO) cannot be 
deleted under identity with the nominative one in (9), although the nominative 
NP in (10b) can be deleted, as shown in (lib). 

Conjunction Reduction: only nominative NPs delete 

(12) bayi yaja bani-^u 

cm n man n come -non -fut 
'man came' 

(13) a, balan ^ugumbil baijgul yaraijgu balga-n 

cm n woman n cm i man i hit-non-fut 
b, bayi ya^a bagun ^ugumbilgu balgal-na«^u . 
cm n man n cm d woman d hit-nay-non-fut 

'man hit woman' 

(lii) a. *bayi ya^a bani-pu balan ^ugumbil balga-n 

cm n man n cQme-non-fut cm n woman n hit-non-fut 
b, bayi yaj:a bani-pu bagun dugnimbilgu -balgal-i^a-^iu 

cm n man n come-non-fut cm d woman d hit-i^ay-non-fut 
'man came and hit woman' 

The nominative IIP in (12) can delete the nominative NP in (l3b) zander identity 
but not the instrumental NP in (l3a), 

Relativization: only nominative NP can be relativized 

(1$) a. bayi yajra bagal-i)a-nu bagul yuj'igu banagsynu 

cm n man n ^spear-i^ay-rel cm d kangaroo -d7 retum-non-fut 
'man who speared kangaroo is returning' (bayi yara \0) 

b, -Ji-bayi yapra /baga-^u bayi yu^iZ banaganu 

cm n man n spear-rel cm n kangaroo n retum-non-fut 

(man who speared kangaroo is returning) (bai^gul yaj'angu .^ 0) 

c, -x-bayi yafa ^alan 4iigumbil huTa-na-pn/ wayndin 

cm n man n cm n woman n watch-nay-rel go uphill-non-fut 
(man watched by woman is going upnill) (bagul yaragu > 0) 

Ref lexivization : only nominative NP can control clausebounded Reflexivization 
(l6) a, bayi ya^a buyba-yiri-;?.u 

cm n man n hide-raf lex-non-fut 

b, ---bai^gul yararjgu buyba-yiri-^u 

cm i man i hide-reflex-non-fut 

c, ^-bagul yaragu buyba-yiri-;iu 

cm d man d hide-reflex-non-fut 
'man hid himself' 

This absolutive-ergative analysis will be compared with the following 
subject-object analysis. In this analysis (6a) i& claimed to be closest to 
the underlying structure of transitive sentences. The subject is marked 
with nominative case and the object ; dative case. The - nay suffix marks 
the verb as being active transitive. In the derivation of (oa) Passive 
applies, promoting the object to subject and demoting the subject to an 
oblique NP marked by instrumantal case. The verb is intransitivized by the 
deletion of - gay . Object Demotion would apply in the derivation of senten- 
ces like (6c), The object is demoted to an oblique NP and marked by the 
instrunental case. 

The iiubject-object analysis also account'? for the facts about Equi- 
NP-Deletion, Conjunction Reduction, Relativization, and Reflexivization 
in (9)-(l6), The nominative NPs which were absolutives in the absolutive- 
ergative analysis are subjects in the subject-object analysis. Both analy- 
ses v/ork equally' well in accounting for these data because they are mirror 
images of each other. Each analysis chooses one sentence whose structure 
is closest to the underlying form of a transitive sentence and derives the 
other two sentence types by appropriate transformations. What is underly- 
ing in one analysis is derived in the other and vice versa. Because the 
nominative case reflects the highest grammatival relation in both analyses, 
transfonrjations referring to stirface subjects or surface absolutives will 
not distinguish between the two analyses. One way to evaluate the compet- 
ing analyses is to find other areas of the grammar of Dyirbal for which the 
two anal;/ses make different predictions and see which one is correct. Rules 


referring to the jecond highe-st grammatical relation, yuch as Objcct-oub- 
ject-Equi, ought to, as well as rulej referring to underlying subjects. 9 
Two sets of data dealing with t3uch areas of the gratninar of i3yirbal are 
discussed below. 

Lower subjects can oe deleted only under identity with terms. Thus, 
in English a direct object can trigger deletion, out a passive -agent (chd- 
raer) cannot. 

(17) a. i'-lary told Bob to go. 

b. -K-Mary was told by Bob to go. 
(with the reading 'Bob go' ) 

In Dyirbal a dative NP can trigger deletion of a nominative NP, but an 
instrumental NP cannot, 

(18) a. bayi yaj:a bagun ^ugumbilgu giga-i)a-jiu bagum mirsyigu 

cm n man n cm d woman d tell-nay -non -fut cm d beans d 


scrape -nay -non -fut 

'man told woman to scrape the beans' 

b. -M-bayi yara bai^gun dugumbipi giga-i)a-/iu bagum mira^pgu 

cm n man n cm i woman i tell-^ay-non-fut cm d beans d 
scrape -nay -nu 
(man told woman to scrape beans) 

c. -w-bayi ya^^a bangun ^ugumbiru giga-n bagum mirajigu 

cm n man n cm i woman i tell-non-fut cm d beans d 
scrape -r^ay-non-fut 
(womanitold man (for) 0^ to scrap? beans) 

A subject-object analysis correctly predicts that objects ( in dative case) 
can trigger deletion, but that oblique (instmunental) NPs cannot. An abso- 
lutive-ergative analysis incorrectly predicts that ergatives (instrumental 
NPs) should be able to trigger deletion and that dative case NPf should not 
be aole to. Thus, the subject-object analysis correctly predicts that dat- 
ive NPs represent the direct object grammatical relation and that instrumen- 
tal NPs a e not terms. 

Another sire a of the grammar which provides evidence distinguishing 
between the two analyses is Imperative jDeletion. lOXhe following three sen- 
tences demonstrate how Imperative Deletion applies in I^^irbal. 

(19) a. (ninda) bani 

(2sg-n) come 
'come J ' 

b. (ninda) bagul yafagu balgal-ga 
(2sg-n) cm d man d hit-ijay 
'hit the mani ' 

c, (ninda) bayi ya^ balga 
(2sg-n) cm n man n hit 
'hit the man! ' 


(let 1 
j„^jthe man be hit!) 


A subject-object analysis predicts that subjects of imperatives will 
be deleted and an absolutive-ergative analysis predicts that aosolutivea 
will oe deleted. An exajnination of the examples in (19) shows that only 
in the subject-object analysis is it possible to capture the relevant gen- 
eralization about Imperative Deletion. In (l9a) the subject of an active 
intransitive sentence is deleted in the subject-object analysis. In the 
absolutive-ergative analysis the absolutive is deleted. In (l9b) the tran- 
sitive subject of an active transitive sentence is deleted in the subject- 
object analysis. In the absolutive-ergative analj'sis a derived absolutive 
(underlying ergative) is deleted from a derived intransitive sentence. In 
(I9c) the passive agent (or underlying subject) of a derived intransitive 
sentence is deleted. In the absolutive-ergative analysis the ergative ic 
deleted from an active transitive sentence, 

/or the subject-object analysis the generalization about Imperative 
Deletion is that underlying subjects delete, .'"or the absolutive-ergative 
analysis there is no clear generalization referring to only one grammatical 
relation. It appears that in intransitive (derived and underlying) senten- 
ces, the absolutive is deleted and that in transitive sentences, the erga- 
tive is deleted. Evaluation of the two generalizations shows the subject- 
object analysis to be preferred. Not OTily does it capture the generaliza- 
tion that the underlying subject deletes, but is does so without resorting 
to unprecedented grammatical relations. 

Perhaps one objection to the subject-object analysis of Imperative 
Deletion in Dyirbal is that it allows passive imperatives. This objection 
is easily answered by the fact that other languages, like haori, Javanese, 
and Bahasa Indonesian, also allow passive imperatives. In Maori the dis- 
tribution of the active sentence is so marked that active imperatives are 
ungramraabical, as shown in (20), Imperative Deletion applies to delete the 
passive agent, that is, the underlying subject, 

(20) Kaori (Chung, 1976a)_ 

a. tua-ina te rakau 
fell-pass the tree 
•fell the tree I' 

b. -Ktua-i te rakau 
fell-act the tree 
(Fell the tree') 

c. puhi-a 
shoot -pass 
'Shoot (it)!' 

d. *patu te pooka 
hit-act the pig 
(Kill the pigl) 

In passive imperatives in Javanese the underlying subject is also 

deleted, as shown in (21), 

(21) Javanese (Home, 1961) 

a, KOntSte di-adepke" ron6. 
car pass -head over that way 
'Head the car over that way.' 

b, Kertase' di-tulisi djenengmu! 
paper pass -write name -your 
•Write yoiQT name cm the paper! • 



Thus L^irbal is like other languages in which the passive ir; relatively 
unmarked. In these languages the generalization':- aoout Imperative :)eletion 
seems to be to delete the underlying subject. The subject-object analysis 
can correctly account for the facts about Imperative Deletion in iJyirbal, 
since Dyiroal behaves like other subject-object languages in which the dis- 
tribution of the passive construction is less restricted than in English. 
This cross -linguistic data argues against Dixon's (1977) position that pas- 
sives are universally the more morphologically complex and restricted in 
distribution than the active. There are languages in which the distribution ' 
passive is less restricted than that of the The situation in Kaori 
(Chung, 1976a) parallels that fo'jnd in Dyirbal. The passive occurs in main 
clauses and the active usually occurs in reduced or subordinate clauses. 
While all main clauses in haori are passive, some main clause in Dyirbal do 
occur in the active, as in the topic chain in (22). 

(22) balagara ^urmangu bu^-alga/iu 

two people-n track-d see-ijay-non-fut 

'the two (women) saw the (man's) track (i.e. the path worn by his con- 
stant travel to and from the scrub).' (Text XV, 12) 

In languages like Javanese and Bahasa Indonesian the passive also has 
a relatively unrestricted distribution. Thus, when one is faced with decid- 
ing between a subject-object analysis with an unmarked passive similar to 
passive in other languages or an absolutive-ergative analysis which proposes 
grammatical relations which have not been attested in any other langiiage, 
the preferred analysis is the subject-object analysis. Although the syntac- 
tic evidence from Imperative Deletion and Object-ouoject-Equi indicates the 
active transitive sentence is the more morphologically complex -i^ construc- 
tion, Dixon maintains that this construction is derived and that Dyirbal is 
syntactically absolutive-ergative. This position entails the necessity of 
such a language being able to refer to more than one set of primary grammati- 
cal relations. This complication of the grammar of Dyirbal is avoided in 
the subject-object analysis,, ^ 

There are two possible objections which might be raised against the 
subject-object analysis of Dyirbal based on morphology: first, the active 
transitive form of the verb is morphologically more complex than the intran- 
sitive form, and, second, while the - i^ay suffix is absent in the Passive 
derived sentence, it is present in the Object -Demotion derived sentence. 

In answer to the objection that the active verb morphology is more 

coraplex, it can be pointed out that in the Malaysian languages in which there j 

is more than one kind of passive, the non-intentional passive is less com- i 

plex than the active transitive, j 

(23) Bahasa Indonesian (R, MacDonald, 1976) 

a, Ali men-yangka saya orang Palembang 
Ali trans -thought I man Palembang 
'Ali thought that I was a Palembang man,' 

b, Saya Ali sangka orang Palembang 
I Ali thought man Palembang 
'Ali thought that I was a Palembang man' (passive) 

c, Ali meng-anggap soal itu beres 
Ali trans -considered problem the settled 
'Ali considered the problem settled,' 


(23) (con't) 

d. ;/oal itu Ali anggap be res 

problem the Ali considered settled 
'Ali considered the pboblem settled,' (passive) 

Thus it is not the case that crosslinguistically the pasisive form of the 
verb is more complex than the active. 

^v'ith regards to the objection that the -qiay suffix is not deleted in 
Object Demotion derived sentences, two replies are available, j^irst, it 
is not the case that morphology always corresponds directly to the syntax 
of a lang-uage, For example, in Walbiri, a morphologically' ergative language, 
the subject of an Antipassive derived sentence remains ergative ly marked, 

(2U) Walbiri (Comrie, forthcoming) 

a, Nat uluj.u ka-^a wawiri Juwaqi 
Isg-erg tense -Isg kangaroo-abs shoot 
•I £im shooting the kangaroo,* 

b, Nat^ululu ka-na-la-t inta wawiri-ki luwani 
Isg-erg* tenSe-lsg-3sg-dat-intran kangaroo-dat shoot 
•I am shooting at the kangaroo,' 

Comrie(forthcoiiiing) points out that in Basque the subject remains ergative 
and the auxiliary verb is marked tjith a third person singular patient after 
Object Deletion applies, as shown in (25). Catford (1975) discueses ergative 
marking on subjects in Chechen and Dargi after Object Incorporation or 
Object Deletion, rules which are similar to an Object Demotion Antipassive, 

(25) Basque (Bakaikoa, personal communication) 

a, guk zuei artikulua irakurri d-i-zue-gu 

Ipl 2pl-dat article-def read 3sg-aux-2pl-lpl 
'We read the article to you,' 

b, guk zuei irakurri d-i-zue-gu 

Ipl 2pl-dat read 3sg-aux-2pl-lpl 

'We read to you,' 

Perhaps a second reason why the -way suffix is not deleted in the Object 
Deraoticai construction in Dyirbal is the ambiguity which would result from the 
merger of the passive with the Object Demotion derived structure. Both have 
nominative and instnimental NPs, The free word order in Dyirbal would not 
distinguish whether the nominative "HP was an underlying subject and Object 
Demotion had applied or whether it was a derived subject and Passive had 
applied. The presence or absence of -nay distinguishes between these two 
constructions involving a nominative NP and an instrumental NP, 

One problematic area for both analyses of Dyirbal is the prcaioun sys- 
tem. The chart in (2?), taken from Dixon (1972), has been annotated to make 
explicit the correspondences of case marking to the grammatical relations of 
each analysis. In the absolutive -ergative analysis absolutive intransitive 
subjects and absolutive objects have different case marking. Transitive sub- 
jects are marked with the sane form as intransitive subjects (except for sing- 
ular pronouns in the Giramay dialect). Dative case represents absolutive 
objects. In the subject-object analysis tha dative case represents the direct 
object relation, Dixon's (O) forms are subjects derived from Passive, The 
(S) and (a) forms represent underlying subjects; passive agents have the same 


pronoun form as do other underlying subjects, that is, the transitive sub- 
jects in the -nay construction and underlying intransitive subjects. A 
passive subjectTs distinguished form suJ'face subjects which were underly- 
ing by the suffixation of -na cf -ga to the pronoun form referring to under- 
lying subjects. 

This use of -na/-:Pa is also found optionally v;ith proper and common 
names, usually referring to humans and other animates. Thus object NPs (in 
dative case) have two forms; so do underlying objects which have been pas- 
sivized to derived subjects. 

(26) underlying subject: 
luiderlying object: 
derived passive subject: 

burbula (only) ^ 

burbulagu ,buTbalyiangu 
burbula, burbal^a 

'burbula' (name) 

This optional process seems to have been made obligatory for derived pronom- 
inal subjects. The problem of mapping the case morphology of the pronoun 
system onto grammatical relations has a simple solution in the subject- 
object analysis: underlying subjects have one form; derived passive sub- 
jects have an obligatory suffixj surface objects have another form. In the 
absolutive-ergative analysis there does not seem to be a clear generaliza- 
tion about the relationship of case marking in the pronoun system to the 
grammatical relations of absolutive and ergativej one form is used for all 
derived subject absolutives, all ergatives, amd some underlj'ing subject 
absolutives. These data have led Dixon (197?) to conclude that some parts 
of languages can and/or must refer to subjects, -"-^ 

(2 7) Dyirbal pronoun system (Dixon 1972 :$0) 

Abs -Erg Sintrans/Atrans abi oblique NP 

Analysis (GtSccly) (G;Aonly) object da tive 

Sub -Ob j underlying Sub derived object NP 
Analysis (G:Sintrans) (G:Strans) passive sub 



G 5ayba 
DM r^a^a 




G r^ali 
D ijali^i 
K ijali 






G ijana 
D ijan^i 
M ^ana 






G ijinba 
DM jiinda 





G Tiiibila^i 
D ;iubala^i 
M j;mbala 




G pura 
D ^ura^i 
M jiuray 






In sunmarj'', it seems that all of the evidence presented in this paper 
leads to the conclusion that Dyirbal is not syntactically ergative, A com- 
parison of the two competing analyses has shown the subject-object analysis 
not only to have crosslinguistic precedence, but also to make correct pre- 
dictionc about specific facts about Dyirbal for other parts of the grammar. 
It was first pointed out that while the subject-object analysis rnade the 
implicit claim that the passive was unmarked in Dyirbal, this situation was 
found to occur in other Janguages, Maori representing an extrene case, in 
which the active has a very marked distributicai. The fact that such lang- 
uages may not be as common as those in which the active is unmarked doss 
not justify positing the elsewhere unattested grammatical relations of abso- 
lutive and ergative. 

The subject-object analysis was shown to make correct predictions 
about the grammatical relation of dative NPs« The were shown to be able 
to trigger Object-Subject-Equi-NP-Deletion, an argument for their status 
as the second highest grammatical relation. In Imperative Deletion the 
subject-object analysis was able to capture the generalization that under- 
lying subjects were deleted. It was shown that in other languages allowing 
passive imperatives, the passive agent, that is, the underlying subject, 
is also deleted. Maori provided an example of a language in vrtiich impera- 
tives could only be passive. 

Two possible objections about the verbal morphology in the subject- 
object analysis were shown to have parallels in other languages. The rela- 
tive complexity of the transitive sentence is not unprecedented. The reten- 
tion of the - nay suffix in sentences with demoted objects was shown to have 
parallels in the retention of ergative case marking in formally intransitive 
sentences in other languages, again demonstrating that surface morphology 
is not always a valid indication of the the syntax of a language. A disam- 
biguating function was also mentioned. The pronoun system of Dyirbal was 
presented and shown to have a relationship with the grammatical relations 
of subject and object, but not absolutive and ergative. 

While not conclusive, I hope that this initial evaltution of these 
two competing analyses has shown that not only is the subject-object ana- 
lysis to be preferred because it accounts for the core data without recourse 
to unprecedented grammatical relations, but because crosslinguistic evidence 
and language internal facts show that of the two analyses, only it can 
adequately account for Dyirbal syntax. 


""Research for this paper has been supported by UoiP grant SOC75-002UU 
and the Jtesearch Board, University of Illinois. I would like to thank 
C.C, Cheng, Peter Cole, Wayne Harbert, Gabi Hermon, Kim Hodges, Chuck 
Kisseberth, David Odden, and Anne Roalef for their help and comments. Ear- 
lier versions of this paper were presented the University of Wisconsin 
(i'.adison) and the University of Texas (Austin), Comments received at these 
times have been helpful and are gratefully acknowledged, A less complete 
version has appeared in CLS lU, Of course, all errors remain my own. 

In "Kappoteur's introduction and suiranarj' for the section on "Are 
Australian Languages Syntactically Nominative -Ergative or Nominative -Accu- 
satice?" in Dixon (1976), Blake has the following remarks about the distri- 


Dution of ergativity in Australian language^: : 'The evidence availaole is 
not overwhelmingly strong, but the indications are that the typical Aus- 
tralian language is syntactically nominative -accusative. Cn the other hand 
Dyirbal is syntactically nominative -ergative, and it is the only Australian 
language that has been clearly documented as being ergative in syntax." 
(Dixon, 1976:ii88) 

^^or a more extensive listing of subject properties, see Keenan (1976), 

•^Jol-jnson (197U) proposes the following Generalized Accessibility 
Hierarchy: Primary > Secondary ^ 10 ^ Oblique NP > .c. _^ chwner 

^Dixon (1972) distinguishes an ergative case and an instrumental case 
although they are morphologically identical. Rather than maintain this dis- 
tinction, the two kinds of NPs will be treated as being marked with the same 
case out as representing two grammatical relations. In Dyirbal instrumental 
(true instruments) and coramitative NPs can be promoted to surface subject 
in the -m(b)al construction, but agent NPs in instrumental case cannot. This 
distinction is similar to the one between the ability of instruments, loca- 
tives, and other obliq\ie3, but not passive agents, to undergo passive under 
certain consitions. The fact that instruments but not agents can undergo 
-m(b)al promotion can be accoiinted for by both analj'ses. In an absolutive- 
ergative analysis the inability of the ergative to be promoted indicates that 
it is a distinct grammatical relation. However, since this implies that erg- 
ative NPs are very low on the Generalized Accessability Hierarchy, it is per- 
haps more problematical for the absolutive -ergative analysis than for the 
nominative -accusative analysis in which these non-promotable NP are treated 
as passive agents. 

The distinction between sentences like (6b) and (6c) is difficult to 
make explicit. In order not to misrepresent the situation, the following 
comment by Dixon (1972:66) is offered: "The difference between (examples 
parallel to (6c)) and (examples parallel to (6b)) is felt by speakers of 
Dyirbal to be a crutial and important onej it cannot be brought out through 
English glosses, but is explained in terms of the 'deep syntax', «• Roughly, 
in (examples parallel to (6c)) the actor, goal, and action make up an event; 
(examples parallel to (6b)) imply something more -- that the actor is posi- 
tively implicating the goal in the event. The difference is essentially one 
of topic.'* 

Use of the term Antipassive to characterize this kind of rule is per- 
haps misleading, especially for languages in which ergativity is a morpholo- 
gical phenomenon. In terms of grammatical relations, the rule is best refer- 
red to as Object Demotion, The effect of this demotion is to in trans itivize 
the verb, which, in a morphologically ergative language, like West Circassian, 
results in absolutive (nominative) case marking on the subject and oblique 
(often identical with ergative) case marking on the demoted object. Note thatj 
when the transitive subject is marked with the same case as obliques, the sur- 
face morphology appears to have switched the cases of the absolutive NP, the 
direct object, and the ergative NP, transitive subject. Actually the subject 
is unaffected by the rule; the only change is the demotion of the diject. 

Wept Circassian (Anderson, 1976) 

(i)p:sasa-m c'9y-aj» ya-d-a 

girl-^rg cherkessa-abs 3sg-(3sg)-sew-pres/trans | 

'The girl is sewing the cherkessa. • X 



(ii) p:sasa-r c 'ay-em ya-d-a 

girl-abs cherkessa- obi 3sg-(3sg)-sew-pras/intrans 
'The girl is sewing on the cherkessa. ' 


This is not the only possible relationally based absolutive — erga- 
tive analysis. An alternative would be to posit two Antipassives, one 
de.-noting absolutives to obliques and another demoting absolutive s to the 
second grammatical relation, ergative. Viihat all relationally based treat- 
ments would have in common is that the ergative NP would represent the 
second highest relation and the dative case would not. Johnson (197^1) 
tentatively proposes such a hierarchy of grammatical relations for an abso- 
ergative analysis, (See note 3.) 


George (1975) proposed an alternative analysis of Dyirbal in which 
he posits structures underlying sentences like (6b) as closest to the under- 
lying representation of transitive sentences. This is all that his analy- 
sis has in common with the one which follows, 

^Because the data for idiom argument? which could be used to establish 
which structure is underlying and which is derived is not available for Dyir- 
bal, distinguishing between underlying and derived subjects will have to 
rely cai other tests, such as Imperative Deletion. (See discussion of (19).) 

^'^Imperative Deletion involves the pronoun system of I)yirbal,and the 
relaticyiship of the surface case morphology to grammatical relations. (See 
discussion of (27).) 

^•'•In Dixon (1977) i't is suggested that some transformations universally 
result in the sane syntactic effect in every language in which they occur. 
^ypotheses regarding the possible kinds of 'split' syntactic ergativity fall 
essentially' into two categories! those in which the syntax is split according 
to levels; either deep, shallow, surface, or cyclic, post and/or last cyclic, 
and those in which certain kinds of rules are sensitive to one of the possi- 
ble sets of grammatical relations. The first positicxi was basically that 
apopted by George (1975) in an earlier relational grammar treatment of Dyir- 
balj cyclic rules applied to a nominative -accusative syntax, then post cyclic 
rules applied to an absolutive -ergative syntax. However, in this analysis 
rules like Equi-NP-Deletion are treated as post-cyclic. The second position 
can be found in Woodbury's (1977) treatment of Greenlandic Eskimo where he 
posits a neutral NP hierarchy and general principals for determining the pos- 
sible ranges of absolutive -ergative and nominative -accusative rules. Gener- 
ally, absolutive -ergative relations are posited for morphological rules, 
rather than relation changing rules. The few exceptions to this include 
rules like Causative Clause Union and Antipassive (Object Demotion, see note 
6), Close examination of these rules shows that not only is a neutral NP 
hierarchy unnecessary, but the notion of absolutive -ergative grammatical 
relaticais is unnecessary as wel]^ since the only evidence for such relations 
is morphological J case marking and verb agreement rulae can apply to the out- 
put of relation changing rules, as has been shown by Chung (1976a) and Harris 
(1976), Note that in a language like Georgian where the system of case mark- 
ing is determined by tense, aspect/and verb classes, the grammatical rela- 
tions of subject, object, indirect object, etc, only make sense in view of 
their sjnitactic behaviorial properties, not morphological coding properties. 


12 ' 

Note that in Dyirbal passives Lhe pronounc referring to the passive , 

subject and the passive agent are both in nominative case, but t,he derived I 

passive subject is suffixed with -naZ-ga. This parallels the .orphology I 

of lassives in Inibab'^a Quechua in whfch the passive subject is obligatorily | 

marked as a topic with -ca. This distinguishes it from the passive agent^ ' 

a nominative case chdiner. Note that in Imbabura Quechua the verb agrees [ 

with the derived subject in number and person. ' 

(i) nuca-ca pai-lla maca-shca ka-ni 

lS£-top 3sg-just hit-pass be-isg •! was hit by him.' 

(ii) kan-guna fiuca-chi yach-chi-shca ka-ngui-chi 

2-pl 1-pl know-caus-pass be-2-pl 'You are taught by us.' 

because passive agents are not marked by instrumental case, there are two mor- 
phological forms of passive in D irbal. As showi above, in the pronoun system | 
it is parallel to Imbabura Quechua. In the noun system the passive agent's , 

demotion is indicated by instrumental case and the subject is optionally suf- ' 
fixed with -aa/ -jia « j 

(i) ijayguna ijinda balga-n i 

1 sg-na 2 sg hit-non-fut 'I was hit by you' ' 

(ii) bayi yaj-a baggun (^ugumbipi balga-n 

cm n man n cm i woman i hit-.ion-fut 'the man was hit by the woman 
/or a language to have two forms of passive is not unprecedented. In Bahasa 
Indonesian (see Chung, 1976b) the distribution of the two passives seems to 
be similar to that of the two in Dyirbal} in general the non -canonical pas- 
sive is used with first and second imderlying subjects. Appartently this ten- 
dency may reflect an earlier stage in which the tendency was a restriction on 
third person underlying subjects; the canonical passive was supposed to have j 

been used in these cases, 


Anderson, S.R, 1977. On mechanisms by which languages become ergative. In Li,' 

ed,: Mechanisms of syntactic change, Austin? U Texas Press. ! 

Anderson, S.R. 1976, On the notion of subject in ergative languages. In Li, 

ed,: Subject and topic, NY: Academic Press. 
Blake, B.J. 1976, Rapporteur's introduction and summary. In Dixon, ed,; Gram- 
matical categories in Australian languages, Canberra: Australian Institute 

of Aborigine Studies, 
Catford, J.G. 1976, Ergativity in Caucasian languages, NELS 6. 
Chung, S. 1976a, Case marking and syntactic structure in Polynesian, Harvard 

University dissertation, 
Chung, S, 1976b, The subject of two passives in Bahasa Indonesian, In Ii, ed,| 

Subject and topic, NY: Academic Press. 
Comrie, B, forthcoming. Ergativity. ' 

Dixon, R.M.W. 1972, The Dyirbal language of North Queensland, Cambridge Uni- , 

versity Press, : 

bixon, R.M.W. 1977. The syntactic development of Australian languages. In Li,-; 

ed,: techanisms of syntactic change, Austin: U Texas Press, j 

George, L, 197^. Ergativity and relational grammar, NELS V, j 

Harris, A, 1976, Ci*ammatical relations in Modern Georgian, Harvard University] 

dissertation, \ 

Home, E. 196I. Beginning Javanese, Wew Haven: Yale University Press. i 

Johnson, D. 197U, Towards a theory relationally based grammar. University of 1 

Illinois dissertation, '' 

Keenan, E.L. 1976, Towards a universal definition of 'subject'. In Li, ed,: i 

Subject and topic, NY: Academic Press, | 

MacDonald, R. 1976, Indonesian reference grammar, Washington, D.C. George- < 

town Press, 
V.'oodbury, A. 1977. Greenlandic Eskimo, ergativity, and relational grammar. SS8. ' 

Studies in the Linguistic Sciences 
Volume 8, Number 1, Fall 1978 


Yajiiuna Kachru 
Rajeshwari Pandharipande 

In a recent paper, it has been suggested that morpho- 
logical ergativity may be "a consequence of syntactic erga- 
tivity and/or various types of semantic markings" (Dixon 
1977, section 3.0). In this study, we examine data from 
a number of South Asian languages to determine (a) if these 
languages are syntactically and/or morphologically ergative 
and (b) in case they are morphologically ergative, if it 
is a consequence of syntactic ergativity or if it is condi- 
tioned by the semantic nature of NP's, verbs or Aspect-tense. 
We conclude that in the languages we have looked at, there 
is no evidence for syntactic ergativity. They may, however, 
be said to be morphologically ergative. But their morpholo- 
gical ergativity is not conditioned by any semantic marking 
and seems to be simply a consequence of the historical de- 

1.0 In a recent paper [Pandharipande and Kachru 1977), we dis- 
cussed the phenomenon of ergativity in Hindi-Urdu in the framework 
of Relational Grajimiar (Postal and Perlmutter 1976, Johnson 1974). 
In this paper, we attempt to look at ergativity in a wider context, 
namely, in the context of following South Asian languages: Bengali, 
Hindi-Urdu, Kashmiri, Marathi and Nepali. These languages represent 
both the inner (Hindi-Urdu) and outer group (Bengali and Marathi) of 
New Indo-Aryan (Chatter jee 1926), an Indo-Aryan language outside India 
(Nepali) and a Dardic language (Kaslimiri). These languages provide 
us with a spread which is interesting both from theoretical as well 
as typological points of view. In particular, we examine whether erga- 
tivity in these South Asian languages is determined by the semantic 
nature of Noun Phrases (NP's, hereafter). Verbs (V's, hereafter), 
or Aspects-Tenses. 

This study follows the format used in our earlier paper (Pandhari- 
pande and Kachru 1977) and proceeds as follows. First, we discuss 
morphological ergativity in the languages under consideration in terms 
of case (or postposition) marking and verbal agreement (section 2.0). 
Next, we determine if the morphological ergativity in these languages 
is "a consequence of syntactic ergativity and/or various types of 
semantic marking" (Dixon 1977, section 3.0). The final section states 
our conclusions. 

2.0 A language is said to be ergative if the intransitive sub- 
ject (SI, hereafter) is treated in the same manner as the transitive 
object (DO, hereafter), and differently from the transitive subject 
(ST, hereafter). The clearest evidence of such different treatment 
is in the case (or postposition/preposition) narking: in ergative 
languages, the ST is marked differently from both SI and DO. In the 
following sub-section, the marking of SI, ST and DO in each of the 


languages under consideration is discussed separately. Any effect 
such marking has on verbal agreement is also pointed out. 

2.1 In Bengali, neither the SI nor the ST or DO is marked with 
special inflection or postposition. Only an animate definite object 
is followed by the accusative/dative postposition ke^. Consider the 
following sentences: 

1. ami barT jacchT 
I home go 1st p. 
I am going home. 

2. turn! barT jaccho 
you home go 2nd p. 
You are going home. 

3. se barT jacche 
he/she home go 3rd p. 
He/she is going home. 

4. arnra barT jacchT 
we home go 1st p. 
We are going home. 

5. chelera barT jacche 
boys home go 3rd p. 
The boys are going home. 

6. amT boT por I am 
I book read 1st p. 
I read a book. 

7. se boT porlo 

he book read 2nd p. 

He read a book. 4 


8. ma cheleke daklen , J 
mother boy ace. called hon. ^j 
Mother called the boy. 'J I 

9. se tomake dakchilo M 
he you ace. call was 2nd p. M 
He was calling you. 1 

Notice that in each of the above examples, the SI as well as ST ) 

is unmarked and only the DO in 8-9 is marked with the accusative post- j 

position k£. The verb shows agreement in person (1st, 2nd, 3rd), j 

and honorific (sentence 8). We need not discuss the case of Bengali ] 

any further, as Bengali does not show any overt marking for ergativity. i 

We shall, however, come back to Bengali later in our discussion (sec- !; 

tion 5.0). ! 

2.2 Unlike Bengali, in Hindi-Urdu, the transitive verb in the 
perfective requires the agent to be marked with a special agentive 


postposition ne , e.g., compare the following non-perfective and per- 
fective sentences with intransitive and transitive verbs: 

10. larka sota he / soya 

boy sleeps / slept niasc. sg. 
The boy sleeps/ slept. 

11. lerki sotT he / soT 

girl sleeps / slept fern. sg. 
The girl sleeps/sleot . 

12. me sota hu 

I sleep 1st p. masc. sg. 
I sleep. 

13. jiarke/ ne kitab parhT 

(boy / ag. book read 

The boy/girl read the book. 

Jiarkef ne axbar oarha 

jboy > ag. newspaper read 
(girl j 
The boy/girl read the newspaper. 

15. ne ne kitab parhT 
I ag. book read 

I read the book. 

16. me ne axbar parha 
I ag. newspaper read 

I read the newspaper. 

Notice that neither the SI in 10-12, nor the DO in 13-16 are marked 
with any postposition, whereas the ST in 13-16 is followed by the 
post-position ne. Notice further that in 10-12, the verb agrees with 
the SI, whereas in 13-16, the verb agrees with the DO, i.e., with 
kitab 'book' (fern, sg.) in 13 and 15, and axbar 'newspaper' (masc. sg.) 
in 14 and 16. On the basis of data such as the above, Hindi-Urdu has 
been described as an ergative language (Allen 1950). This account 
of NP-marking and verbal agreement in Hindi-Urdu, however, is incom- 
plete. Since the details of NP-marking and verbal agreement in Hindi- 
Urdu have been discussed elsewhere (Pandharipande and Kachru 1977), 
we will simply summarize them here. The following facts about Hindi- 
Urdu have to be taken into" account before characterizing Hindi-Urdu 
as syntactically ergative. First, some transitive verbs in the perfec- 
tive do not allow the agent marked with ne, (e.g., lana 'bring'), 
whereas others allow for optional marking with n£, (e.g., s ana J hna 
'understand'). Second, some intransitive verbs allow the SI to be 
marked with ne, (e.g., nahana 'bathe oneself, chTkna 'sneeze', etc.). 
Third, the verb agreement is not sensitive to ne , it is sensitive to 
any postposition. As a consequence, the verb either agrees with any 
unmarked NP in the sentence, or, if all NP's are marked, shows neutral 
agreement. Consider the following examples of such neutral agreement: 


114 i 

17. ilerke/ ne kitabo ko utha I i ya 

jboy 1 ag. books ace. picked took 
The boy/girl picked up the books. 

18. mujhko becco ko khilana he 
me dat. children ace. to feed is 
I have to feed the children. 

19. usse bscco ko sembhala nahT jaega 
him by children ace. care not go will 

He will not be able to care for the children. 

Note that since all NP's in 17-19 are followed by postposition, 
the verb is in the unmarked, i.e., masc. sg. form. Note also that 
18 and 19 are not perfective, rather, they are modal and passive, 
respectively, (see Pandharipande 1977 for a discussion of Passive in 

Since the ne-marker is not determined either by transitivity or 
by perfective aspect, it is worth considering if the semantic nature 
of NP or V conditions the occurrence of ne in Hindi-Urdu (as suggested 
in Dixon 1977). Two possibilities are worth considering in this res- 
pect: ne^ always marks the agent as opposed to the patient or ne_ al- 
ways marks the volitional agent as opposed to the non-volitional agent. 
Unfortunately, neither one of the above is borne out by the data. 
Compare the following: 

20. ran (ne) chTka 
Ram (ag.) sneezed 
Ram sneezed. 

21. ran ko chTk aT 
Ram dat. sneeze came 
Ram sneezed. 

22. ne (ne) khasa 

I (ag.) coughed 
I coughed. 

23. mujhko khasT aT 
me dat . cough came 
I coughed. 

It appears as though the dative subject construction characterizes 
the patient subject (in 21 and 23) whereas the agentive SI optionally 
takes ne (in 20 and 22). But, consider further data: 

24. bacca gira 
child fell 

The child fell down (accidentally or deliberately). 

25. *b9cce ne gira 

child ag. fell 


26. ram ne tasvTr dekhT 
Ram ag. picture saw 

Ram saw/looked at the picture. 

27. *ran tesvTr dekha 

Ram picture saw. 

28. ran kitab laya 
Ram book brought 

Ram brought the book. 

29. *ram ne kitab I aT 

Rapi ag. book brought. 
Ram brought the book. 

30. bar is ne sara maza kirkira kar diya 
rain ag. all enjoyment ruined 

The rain ruined all fun. 

24-30 illustrate the fact that ne does not consistently mark the 
(volitional) agent in Hindi-Urdu. In dative subject construction, 
the dative postposition ko does mark the experiencer (Fillmore 1968) 
consistently (Kachru, Kachru, Bhatia 1976). 

2.3 Nepali is similar to Hindi-Urdu in that the ST is marked 
with the instrumental postposition l_e. It is different from Hindi- 
Urdu in that the marking is not restricted to perfective only. Also, 
the verb agreement is not affected by le^, the verb still agrees with 
the ST: 

31 . ma hase 

I laughed. 

32. keto haso 
boy laughed 

The boy laughed. 

33. mai le kitap parhe 
I instr. book read 
I read the book. 

34. ketale kitap parho 
boy instr. book read 
The boy read the book. 

The postposition le^ occurs with the third person ST in the future, 
and with all subjects (i.e., both SI and ST, in all persons) in the 
tenses formed with the infinitive (as in 37-40 below) : 

35. usie bhat khala 

he instr. rice eat will 
He will eat rice. 

36. *usle ghara Jala 

he instr. home go will 
He will go home. 


37. usie janu pa re ha 
he instr. go inf. should 
He should go. 

38. usIe kitap parhnu parcha 
he instr. book read inf. should 
He should read the book. 

39. maile kitap parhnu parcha 

1 instr. book read inf. should 
I should read the book. 

40. usle / uslai janu cha 

he instr. /he dat. go imp. is 
He has to go. 

Notice that in case of modal constructions (37-40) the verb agree- 
ment is neutral, the verb does not agree either with SI or ST. In 
sentence 40, SI or ST may take either the instrumental le^ or the dative 
lal . A generic ST of an imperfective (habitual present) obligatorily 
takes le , as in the following: 

41. rughale dukha dincha 

• cold instr. suffering gives 
Cold makes one suffer. 

42. baghle bakhra khanchan 
tigers ag. goats eat 
Tigers eat goats. 

43. *baghle jangalma bascha 

tiger ag. forests in live 
Tigers live in forests. 

To complicate matters further, le may occur with the ST in the 
perfective of verbs such as hacchiu garnu 'sneeze', khoknu 'cough', 
etc. : 


usle hacchiu garo 
he ag. sneeze did 
He sneezed. 


mai le khoke 
I ag. coughed 
I coughed. 

To summarize, the distribution of le in Nepali is as follows; 

46. a. Simple past and perfective tenses: all ST, some SI 

b. Future: 3rd person ST only 

c. Present imperfect (habitual): generic ST only 

d. Vnu par-: all SI and ST (i.e., in all persons) 

e. Vnu ch-: all SI and ST (i.e., in all persons) 
[In e, le~lai alternate] 


Again, it is difficult to detect any consistent semantic motivation 
for this distribution. 

2.4 As far as NP-marking is concerned, in Marathi , only the third 
person ST is marked with the instrumental postposition ne (variant 
nT) in the perfective, e.g.: 

47. nT kame kelT 

I jobs did 3rd pi. neuter 
I did the jobs. 

48. anhT kame kelT 

we jobs did 3rd pi. neuter 
We did the jobs. 

49. tU kame kelT(_s) 

you jobs did 3rd pi. neuter (2nd d.) 
You did the jobs. 

50. tunhT kame kelT(t) 

you pi. jobs did 3rd pi. neuter (pi.) 
You did the jobs. 

51. tyane /tine kame kelT 

he instr./she instr. jobs did 3rd pi. neuter 
He/she did the jobs. 

52. tyannT kame kelT 

they instr. jobs did 3rd pi. neuter 
They did the jobs. 

Contrast the above sentences in the perfective with the following 
in the imperfective: 

53. ram ambe khato 

Ram mango eats (masc. sg.) 
Ram eats mangos. 

54. sudha ambe khate 

Sudha mangos eats (fem. sg.) 
Sudha eats mangos. 

Except for the optional second person ending in 49 and 50, the 
verb agreement is similar to Hindi-Urdu. The NP-marking is different 
in that in Marathi, only the third person ST is marked with ne (ni) . 
This is, however, an incompletepicture. In Marathi, the third person 
SI and ST are marked with ne ( ni ) in the desiderative and modal 'should' 
construction also, e.g.: 

55. ram ne ata ghsrT zave 
Ram instr. now home may go 
Ram may go home now. 


56. tyane panT mhantiT pah I jet 

he instr. songs sing should pi. 
He should sing songs. 

Notice that the verb agrees with the DO in 56 and is in neutral 
agreement in 55, cf: 

57. nT ata zave 

I now may go 
I may go now. 

53. tyane / tumhT kam kerave 
he instr. / you job may do 
He/You may do the job. 

As in the perfective, in the desiderative (55 and 57) and modal 
(56 and 58), too, the second person subject marking suffix may occur 
with the verb, e.g.: 

59. tu zavas / zavTs 
You may go. 

60. tu gharT ala puhijes 

you sg. home come masc. should 
You should come home. 

Note that in the modal construction, all subjects may optionally 
be marked with the dative -la., in which case, in the third person, we 
may get either - ne or -la, e.g.: 


lari ala pahije 

J ma I a / 

\ tula Y gh£ 


P ] 

) you > 
Che J 

dat. home come should 

I /You/He should come home. 

The NP-marking and verb agreement pattern in Marathi seems to be 
as follows: 

62. a. SI unmarked : V agrees with SI 

b. ST unmarked in imyierfective : V agrees with ST 

c. ST marked (0 with lst/2nd, ne (ni) with 3rd) in 
perfective : V agrees with unmarFed DO 

d. ST marked (0 with lst/2nd, ne (nl) with 3rd) in 
desiderative/ modal : V agrees with DO if transitive, 
otherwise neutral 

e. ST marked, DO marked : Neutral agreement 

An example of e is as follows: 

63. tyane mulTla pahila 

he instr. girl ace. saw neut. 
He saw the girl. 


Note that Marathi is similar to Nepali as far as NP-marking is 
concerned: in both languages, subjects of perfective as well as cer- 
tain modal constructions are marked. We will have more to say about 
this in 4.0. 

2.5 In Kashmiri, the ST in the perfective is marked with speci- 
fic case forms as opposed to the following which are in the nominative: 
the ST in the non-perfective, the SI in both perfective and non-per- 
fective, and the DO in both perfective and non-perfective (Kachru 1969) 
Consider the following: 

64. bJ chus kh'ava:n 
I be masc. sg. eat 
I ajii eating. 

65. t im chi kh 'ava: n 
they be masc. pi. eat 
They are eating. 

66. ne kh'av bat J 

I dat. ate masc. sg. food masc. sg. 
I ate food. 

67. timav kheyi tsot 

they ag, ate fern. sg. bread fem. sg. 
They ate bread. 

The case forms utilized for marking the ST in the perfective are 
as follows (Kachru 1969): 

68. First and second person : dative case 
Third person: agentive case 

The following sentences exemplify this: 

69. raman dits me kitab 

Ram ag. gave fem. sg. me book fem. sg. 
Ram gave me a book. 

70. me per kitab 

I dat. read fem. sg. book fem. sg. 
I read the book. 

71. tse dits tamis tsot 

you dat. gave fem. sg. him dat. bread fem. sg. 
You gave him bread. 

72. tam' par kitab 

he/she ag. read fem. sg. book fem. sg. 
He/she read the book. 

The dative (with first and second person) and agentive (with 
third person) is used with the pazi -imperative, too (Kachru 1969:268): 

73. asi pazi kh on 
we dat. ought eat 
We ought to eat. 


Kashmiri has several nominal declension classes and very few 
nouns/pronouns have distinct forms for each case, hence, it is not 
unreasonable to label the so-called dative as dative/agentive in the 
context of first and second person pronouns. Note that mardav may 
be either the ablative or the agentive form of marJd 'men' . Now, 
we can say that in Kashmiri, the ST in the perfective is marked with 
the agentive case-marker. The verb agreement pattern is similar to 
Hindi in that the verb agrees with the SI and also ST in the non-per- 
fective tenses, while it agrees with the DO in the perfective tenses. 
The agreement pattern, however, is not as consistent as in Hindi. 
This is because, in Kashmiri, certain transitive verbs govern objects 
in oblique cases and others do not. For instance, the verbs 'see' 
and 'call' take objects in nominative case and therefore, the verb 
agrees with the DO; the verb 'hit', on the other hand, takes an oblique 
object, therefore, the verb is in its neutral form, i.e., it does not 
agree either with ST or DO.-'- Consider the following: 

74. a. me bulov su ladkl 

I dat. called that boy noin. 
I called that boy. 

b. me bul8:v so kUr 

I dat. called that girl nom. 
I called that girl. 

75. me loy tamis ladkl 

tas korl 
I hit that boy dat. 

that girl 
I hit that boy/that girl. 

In 74 a and b, the verb bulov versus bula:v shows agreement with 
the DO; in 75 the verb form is in third person masculine singular 

3.0 The above discussion suggests that except for Bengali, all 
the South Asian languages under study may be at least morphologically 
'split ergative' (Dixon 1977). In this section, we will determine if 
any of the above languages are syntactically ergative, too. 

According to Dixon 1977, there are three universal syntactico- 
semantic primitives - A and occurring in transitive sentences and S 
occurring in intransitive sentences. 'Subject' is a universal deep 
structure category, involving functions A and S. The operation of 
optional singulary transformations (passive, antipassive, reflexive, 
etc.) on deep structures yields shallow structures. It is at this 
level that generalized transformations operate, forming coordinate and 
subordinate constructions. These rules may treat (derived) S and A 
in the same way (i.e., S/A pivot) or they may treat (derived) S and 
in the same way (i.e., S/0 pivot). If a language utilizes the S/0 
pivot, it can be said to have an 'ergative' syntax. 

3.1 We have already shown elsewhere that neither Hindi-Urdu, nor 
Kashmiri work in terms of a S/0 pivot (Kachru, Kachru and Bhatia 1976), 
hence there is no evidence for syntactic ergativity in these languages. 



Sharma 1977 presents evidence to show that Nepali uses a S/A pivot in 
major subordinate and coordinate structures. Pandharipande 1978 pre- 
sents similar evidence for Marathi, The following is a summary of 
the evidence presented in the works mentioned above, 

3.2 First, in terms of the accessibility hierarchy (Keenan, 
Comrie 1972) , the marked ST is as accessible to relativization as 
SI or DO or any oblique NP in the South Asian languages under study 
here. The following exemplify this (H-U, K, M, N preceding the numbers 
stand respectively for Hindi-Urdu, Kashmiri, Marathi and Nepali): 

H-U 76. jis larke ne kitab xarTdT thT vah rezqarT lena bhUl jBya 

wh- boy ag. book bought had he change take forget went 
The boy who bought the book forgot to take the change. 

K 77. su Isdki yen ra't h bath g'ov chu yet i rozan 
that boy who yesterday song sang is here living 
The boy who sang the song lives here. 

M 78. jya mulane kal gane mhatle to ithec rahto 
which boy yesterday song sang he here lives 
The boy who sang the song yesterday lives here. 

N 79. jun ketale hi jo git gayo tyo yahT bascha 

which boy yesterday song sang that (boy) here lives. 
The boy who sang the song yesterday lives here. 

3.3 The conjunction reduction rule which reduces the first conjunct 
of a coordinate structure by deleting the identical subject and trans- 
forming the finite verb into an absolutive form, utilizes S/A pivot 

in the languages under consideration here.^ Note, however, that whereas 
in Hindi-Urdu, Kashmiri and Marathi, DO does not either control or un- 
dergo deletion; in Nepali, DO undergoes deletion even though it does 
not control deletion. For instance: 

H-U 80. me ne khana khaya or me so naya 
I ag. food ate and 1 sleep went 
I ate (some) food and went to sleep. 

H-U 81. khana kha kar me so gaya 

food eat having I sleep went 
Having eaten, I went to sleep. 

H-U 82. me ne bacce ko mara or bacca roya 
I ag. child ace. hit and child cried 
I hit the child and the child cried. 

H-U 83. *(me) mar kar bacca roya 
I hit having child cried 
I having hit (him), the child cried. 

K 84. ne kh'av bati ta Songus 
I food ate and slept 
I ate and went to sleep. 


K 85. bl Sogus bati kh'ath 

I slept food having eaten 

bat! kh'ath sogus bi 
food having eaten slept 1 
Having eaten, I went to sleep. 

K 86. me lo:v ladkas tl tan' h ot radun 
I hit child and he cried 
I hit the child and he cried. 

K 87. *ne narrith h ot ladkan radun 
I having hit child cried 
I having hit, the child cried. 

M 88. mT jevio anT mT zopio 
I ate and I slept 
I ate and I slept. 

M 89. jevITn mT zopIo 
having eaten I slept 
Having eaten, 1 went to sleep. 

M 90. nT mulala marl a anT mu I ga radla 

I boy obj . masc. hit and boy cried. 
I hit the boy and the boy cried. 

M 91. *mT mulala narUn mu I ga radla 

I boy obj. masc. having hit boy cried. 
1 having hit (him) the boy cried. 

N 92. ma i I e gulivorko yatra parhe ani ma base 

I Gulliver's travels read and I laughed 
I read Gulliver's Travels and I laughed. 

N 93. gulivarko yatra parhera ma hase 

Gulliver's travels having read 1 laughed 
Having read Gulliver's Travels, I laughed. 

N 94. usie chorilai kuto ra chori roi 
he instr. girl ace. hit and girl cried 
He hit the girl and the girl wept. 

N 95. usie kutera chori roi 

he instr. hit having girl wept 

He having hit (her), the girl wept. 

N 96. chori roi ra usie chorilai kuto 

girl wept and he instr. girl ace. hit 
The girl wept and he hit the girl. 

N 97. a. *roera usie chorilai kuto 

having cried he instr. girl ace. hit 
(The girl) having cried, he hit the girl, 


b. *chori roera usie kuto 

girl having cried he instr. hit 
The girl having cried he hit (her). 

3.4 The rule that yields past participle modifiers of NP's in 
Hindi-Urdu seems to be sensitive to ergative/absolutive distinction, 
for example: 

H-U 98. [Isrkiya bethT] larkiya 
girls sat girls 

H-U 98a. bethT huT lerkiya 
The seated girls. 

H-U 99. [larkiyo ne kitab parhT] lerkiya 
girls ag. book read girls 

H-U 99a. *kitab parhT huT larkiya 
The book-read girls. 

H-U 100. [larkiyo ne kitab parhT] kitab 
girls ag. book read book 

H-U 100a. larkiyo kT parhT huT kitab 
The read-by-the-girls book. 

Notice that 98a and 100a are grammatical whereas 99a is ungramma- 
tical: in 98a, the past participial modifies an NP identical to the 
S-NP, in 100a, it modifies an NP identical to DO of the participial 
verb; in 99a, were it grammatical, the participial would modify an NP 
identical to ST. It has, however, been shown (Pandharipande and Kachru 
1977) that the facts about participial izat ion in Hindi-Urdu are to 
be explained on the basis of the semantic nature of NP's and verbs 
rather than ergative/absolutive distinction. 

In Nepali, the rules that yield past participial and infinitival 
participial seems to utilize the S/0 pivot: 

N 101. [ketiharule kam gare] kam 
girls instr. work did work 

N 101a. ketiharule gareko kam 

The done-by-the-girls work 

N 102. [ketiharule kam gare] ket i 
girls instr. work did girls 

N 102a. *kam gareka ketiharule 
The work-done girls. 

N 103. [manche naro] manche 
man died man 

N 103a. mareko manche 
The dead man. 


N 104. [ketale kitap parh] kitap 
boy instr. book read book 

N 104a. ketale parhne kitap 

boy instr. to read book 

The book to be read by the boy. 

N 105, [ketale kitap parh] keta 
boy instr. book read boy 

N 105a. *kitap parhne keta 
book to read boy 

N 106. [mane he hase] manche 
man laugh man 

N 106a. hasne nancha 

The man to laugh 

Not much information is available on participialization in Marathi 
and Bengali. As far as Bengali is concerned, participialization seems 
to be as marginal a process as it is in English (Verma 1968). Marathi 
seems to be more like the Dravidian languages where a wide range of 
participial modifiers are used. There is no evidence at present to 
suggest that either one of these languages utilizes a S/0 pivot for 
deriving participial modifiers. 

3.5 The discussion so far makes it clear that except for Nepali, 
which uses S/0 pivot for deriving participial modifiers, there is no 
evidence for syntactic ergativity in any other South Asian language 
under discussion here. There is no rule such as antipassive in any of 
these languages, and the passive in South Asian languages does not neces- 
sarily yield a derived subject (Kachru, Kachru and Bhatia 1976). As 
far as participles are considered, the case of Hindi-Urdu (Pandharipande 
and Kachru 1977) suggests that a deeper examination of language-internal 
facts is necessary before arriving at any conclusions. What appears 
to depend upon S/0 pivot in Nepali may turn out to have an alternate 
explanation in Nepali, too. 

4.0 In conclusion, it is necessary to evaluate all the evidence 
we have presented so far. In the absence of any evidence for syntactic 
ergativity, the question naturally arises as to whether South Asian 
languages exhibit morphological ergativity. On the surface, the languages 
under discussion here, except Bengali, seem to be morphologically erga- 
tive. Hindi-Urdu has a special postposition ne^ to mark A as opposed 
to S and 0, Kashmiri has a special agentive case, and Marathi and 
Nepali utilize the instrumental postposition to mark the A. The marking 
of third person A only in Marathi and partially in Nepali also follow 
the principles suggested by Dixon 1977. That is, on the 'potentiality 
of agency scale', third person forms rank lower as compared to first 
and second person forms. The marking of A in perfective also follows 
the principles suggested in Dixon 1977. Note, however, that Marathi 
and Nepali (and marginally, Kashmiri) go beyond the past - non-past 
distinction suggested in Dixon 1977. The principle utilized by these 
languages suggests the following type of split: 


107a. Nepali: Now versus non-now (timeless and potential) 

b. Marathi: Past versus non-past, and actual versus potential 

Kashmiri is closer to Marathi in this respect. Even this, however, 
does not explain the marking exhaustively. If 107a and b were to be 
consistently applicable, imperative, and subjunctive forms should 
have fallen together with the modal constructions in these languages. 
That does not seem to have happened. 

In the absence of any synchronic explanation, it seems best to 
rely upon a diachronic explanation. The perfective in New Indo-Aryan 
is said to have developed from the past passive participle of old Indie 
(Chatter jee 1926). That explains the instrumental marking of the A 
in Marathi and Nepali. The agentive postposition and case in Hindi- 
Urdu and Kashmiri respectively seem to be parallel developments, too. 
It, however, raises a very interesting question with regard to Bengali. 
Notice that Bengali went through identical historical stages. Never- 
theless, instead of an NP-marking system (for A), what developed in 
Bengali was a system of verbal affixation which copied on to the verb 
the personal pronouns. Thus, both S and A are unmarked and control the 
verbal agreement. Several dialects of the so-called Hindi -area went 
through identical development and the tendency in Eastern as well as 
Dakhini varieties is toward the same kind of development. Western 
Hindi, however, has the stable ne-marker, and Marathi seems to be sta- 
bilizing the 0-V agreement in the transitive sentences by losing the 
optional second person suffix of the verb in sentences such as 49, 
repeated here for convenience: 

49. tu kame keli (s) 

you jobs did 3rd pi. neut. (2nd p.) 
You did the jobs. 

The verbal agreement pattern does not show a neat split, however, in 
that all the languages under consideration mark the in specific ling- 
uistic contexts. 

Dixon 1977 is no doubt a valiant attempt to bring some system to 
the chaos. It, however, seems doubtful, on the basis of data from 
South Asian languages, that it is always possible to explain morpholo- 
gical ergativity on the basis of semantic features of NP, V, or As- 
pect-Tense. The languages we have discussed here, except for Bengali, 
seem to be part accusative, part ergative and part neither and these 
parts seem to be determined by historical development rather than 
consistent semantic principles. This suggests that the inventory of 
types of split in Dixon 1977 should include a fourth category: split 
conditioned by historical syntactic/morphological development. 


*Part of the research reported herein was carried out with the help of 
research grant from the Research Board of the University of Illinois. The 
work on Marathi was done primarily by Rajeshwari Pandharipande and the 
Nepali data was supplied by Basudev Sharma. 

as in Hindi-Urdu, the verb agrees with the noun in the 
direct or nominative case. In the perfective, the ST is in the 


oblique case, (i.e., a case other than the nominative) and the DO is 
in the nominative, therfore, the verb agrees with the DO. In case a 
particular verb governs a DO in an oblique case, in the perfective, 
both the ST and the DO are in non-nominative cases, hence the verb 
agreement is neutral. This has been discussed in detail in Kachru, 
Kachru, and Bhatia 1976. 

The conjunction reduction rule discussed here works as follows. 
Given (i) and (ii) , the rule yields (iii): 

H-U (i) ram ghar gaya 
Ram home went 
Ram went home. 

H-U (ii) ram ne khana khaya 
Ram ag. meal ate 
Ram ate (his) meal. 

H-U (iii) ghar ja kar rarn ne khana khaya 
home go abs. Ram ag. meal ate 
Having gone home. Ram ate (his) meal. 

In Hindi-Urdu, there are only two constructions with absolutive forms 
which do not require the identity and coreferentiality of the S/A in 
the two clauses. These are as in (iv) and (v) : 

H-U (iv) car baj kar das minat hue he 

four strike abs. ten minutes happened are 
It is ten past four, 

H-U (v) bhaT ko dekh kar uska man sant hua 

brother DO see abs. his mind calm became 
He was reassured to see his brother. 

Sentences (iv) and (v) involve the sets given in (vi) and (vii) 

H-U (vi) a. car baje he 

four struck is 
It is four. 

b. das mi nat hue he 
ten minutes happened are 
Ten minutes have passed. 

H-U (vii) a. usne bhaT ko dekha 
he ag. brother D saw 
He saw (his) brother. 

b. uska man Sant hua 
his mind calm became 
He was reassured. 

Note that apparently, in (vi) , the two S's are distinct, and in (vii) 
the two A's are not identical: in (viia), the A is He , in (viib). 


it is His mind . The A in (viia), however, is coreferential with 
the possessive NP of A in (viib) . It is interesting to note that in 
Hindi-Urdu, other processes sensitive to coreferential relations be- 
have similar to the conjunction reduction as in (viib). 


Allen, IV. S. 1950, A study in the analysis of Hindi sentence struc- 
ture. Acta Linguistica Hafniensin 6. pp. 68-86. 

Chatterjee, S.K. 1926. The origin and development of the Bengali 
language. London: Allen and Unwin. 

Comrie, B. 1976. (rev. 1977). Ergativity. (mimeographed). 

Dixon, R.M.W. 1977. Ergativity. (mimeographed). 

Fillmore, C.J. 1968. The case for case. In E. Bach and R.T. Harms, 
eds.: Universals in linguistic theory, pp. 1-90. Holt. 

Kachru, B.B. 1969. A reference grammar of Kashmiri. Department of 
Linguistics, University of Illinois. 

Kachru, Y., B. Kachru and Tej Bhatia. 1976. The notion 'subject': 

a note on Hindi-Urdu, Kashmiri and Punjabi. In M.K. Verma, ed.: 
The notion 'subject' in South Asian languages, pp. 79-108. 
Madison, University of Wisconsin. 

Keenan, E. and B. Comrie. 1972. Noun phrase accessibility and uni- 
versal grammar. Linguistic 'Inquiry 8:1, 1977. 63-100. 

Pandharipande, R. and Y. Kachru. 1977. Relational grammar, ergativity 
and Hindi-Urdu. Lingua 41. pp. 217-38. 

Pandharipande, Rajeshwari. 1977. Exceptions and rule government; 
the case of the passive rule in Hindi. Paper presented at the 
187th meeting of the American Oriental Society held at Ithaca. 

. 1978. Participles in Marathi. (unpublished). 

Sharma, Basudev. 1977. The syntax and semantics of participles in 
Nepali. (unpublished). 

Verma, M.K. 1971. The structure of the noun phrase in English and 
Hindi. Delhi. Motilal Banarasidas. 

studies in the Linguistic Science; 

Volume 8, Kuinbor 1, Sj^rin^'j 1978 


James Lcvinc 
University of i-;aryland 

Babby's (1973, 1973) analysis of Russian attri- 
butive adjectives is critically discussed. A re- 
analysis is proposed on the basis of the assumption 
that J and not KOTOR is the relative i^ronoun in 

The syntactic derivation of adjectives is probably one 
of the less disputed issues within the theory of generative 
grammar. There is wide ajjreement in the assumption that 
attributive adjectives are related to underlying predicates 
(Harris 1952; Chomsky 1957; Lees 1960a) , and that they are 
transformationally derived from deeper relative clauses, i,e,, 
from embedded sentences (Smith 196l, 196^; Lees 1960b; 
Katz and Postal 1964; Lalcoff 1970; Ljung 1970, etc), iNThile 
there are difficulties in the analysis deriving attributive 
adjectives from predicate adjectives in relative clauses 
(liolinger 1967; Motsch 1967; Sussex 1974), a more promising 
explanation has yet to be found for the obvious relationship 
between adjectives in attributive and predicative positions. 

The relative clause source of attributive adjectives is 
given additional support in recent studies of Russian adjec- 
tives by Dabby (l973» 1975). At the center of Babby's 
investigations is the syntactic relation between the attri- 
butive long form (LF) and predicative short form (SF) of 
adjectives in Contemporary Russian. In particular, he 
argues that the LF and SF are surface structure categories 
which derive from a single lexical category, Verbal (v). 
Under this analysis, what is traditionally called "adjective" 
is V with the feature (+adj] , and what is traditionally 
called "verb" is V with the feature C-adj^ (cf. Lalcoff 1970 : 
Appendix a). Neither the SF nor the LF are assumed to be 
generated by the phrase structure rules. Rather, it is 
argued that the deep structure category V emerges from the 
transformational component as a SF if it acquires the features 
of gender, number and person by the Subject-Verb Agreement 
transformation. The category V emerges from the transforma- 
tional component as a LF, if in addition to acquiring these 
agreement features, it receives a case feature, Babby 
states his main hypothesis as follows (l975:9): "the LF is a 
SF that has acquired a case feature by virtue of its trans- 
formational introduction into the constituency of a NP (noun 
phrase)." This analysis claims that it is the derived phrase 
marker configuration [. , , V « . J NP which is the defining 
criterion of the LF, Fifrtncrmore , it assumes, in accordance 
with the "lexicalist" theory, that inflectional morphology is 


handled by a set of postsyntact ic morphophonemic rules which 
"spell out" features acquired in derivation as inflectional endings, 
For example, within this framework,^ the ending -AJA of the LF 
KRASIV-AJA in the NP KRASIVAJA DEVUSKA "the pretty girl" is as- 
sumed to be a "spelling" of the features assigned by rules of 
agreement (fern, sing, 3rd) and case marking (nom) . It is the 
opinion of the present writer that assuming the operation of such 
vague morphophonemic spelling rules does not adequately account 
for the morphological details of inflection in the LF adjective. 
The aim of this paper, therefore, is to present an alternative 
analysis in which some concrete, albeit tentative proposals are 
made as to how the syntax might account in a more explicit fashion 
for the inflectional endings of the LF. 

Descriptions of Contemporary Russian by linguists operating 
within both traditional and generative frameworks assume the 
relative pronoun to be the morpheme KOTOR- , plus inflectional 
endings. For example, in the generative account of Russian adjec- 
tives in Babby 1975 » the structure 

•girl*. 'girl' 'is pretty' 

which is posited as the underlying representation of the NP 
KRASIVAJA DEVUSKA 'pretty girl' undergoes Relative-clause For- 
mation to produce the intermediate configuration 

'girl' 'who' 'is pretty' 

In the present analysis I would like to propose that what is 
traditionally considered a morphologically whole inflectional 
ending in LF adjectives, actually contains the relative pronoun 
morpheme J, and that KOTORYJ is a surface configuration, trans- 
formationally derived by the prefixation of KOTOR- to this same 
morpheme J, In arguing that the relative pronoun is J, I am in 
effect suggesting that one of the rules underlying the formation 
of the LF adjective in Contemporary Russian is the same rule as- ^ 
suraed to have been operative in Old Russian. As is well known, 
in Old Russian the relative was the^orpheme J (j-b nom masc) < 
followed by the suffixed particle -ZE. The traditional analysis j 
of the formation of the LF adjective in Old Russiaji is that of j 
Vaillant 19^2, who argues that the relative pronoun eventually ^ 
became affixed to a regular (i.e. short) adjective to create the 
LF. In assuming the concatenation of short adjectives and J, 
Vaillant provides a plausible explanation for the synonymous 
readings of Adj+N phrases and corresponding relative ^lauses. 
Thus, in Vaillant 's view, the NP BLAGY ^=BLAG'Td+J^) CLOVEKT) [ 
'the good man' and the relative clause CLOVEKHd IZE (=*Jb-ZE) i 
BLAG Id 'the man who is good' are identical constructions (p.5), ] 
If one were to incorporate this insight in a generative-trajisf or- , 
mational treatment of the LF in Old Russian, then Adj(LF)+N 
phrases would be analyzed as deriving from the deep structure 


underlying corresponding relative clauses. Under such an analysis, 
a transformational rule which I shall call J-Af fixation , would 
detach the relative pronoun morpheme from the K-node and affix 
it to the rijht of the SF, as in (l). 


CLOvm: I3 

A late rule, known in the literature as Adjective or Modifier 
Shift, will then move the derived LF to the left of its head- 
noun to give the surface configuration BLAGY (BLAG'HD+Jb) 
Clovek Hd . 

The view that J is still the relative pronoun morpheme in 
Contemporary Russian and that the proposed rule of J-Affixation 
is still operative, finds support from the following morphological 
and syntactic considerations. Zaliznjak (1967:2329) argues that 
the inflectional endings of attributive adjectives in Russian 
contain a morpheme which functions as the marker of long formedness, 
His analysis is based on an examination of LF adjectives and their 
corresponding SF • s : 









Comparing the long and short forms above, Zaliznjak notes that 
the LF adjective ending consists of a vowel identical to the 
vowel in the corresponding SF ending plus a 'special morphological 
element' J, followed by a vowel which is a copy of the one im- 
mediately preceding this element. He calls the morpheme J the 
'marker of attributive function'. If one accepts Zaliznjalc's 
view that the inflectional endings of LF adjectives in Contem- 
porary Russian are complex items of the structure V +J+V , where 
J is the long formedness morpheme, then, in my view, it is 
plausible to assume that this morpheme J is the same relative 
pronoun J which Vaillant claims was affixed to the SF to create 
the LF in Old Russian. 

I Adopting the position that the relative pronoun in Contemporary 
; Russian is J makes it possible to explain certain syntactic facts 
about attributive adjectives and corresponding relative clauses 


with predicative adjectives. Consider, for example, sentences i 
(a)-(c) belov, taken frora Babby 1973: : 

•The boy who is ill with quinsy has to stay in bed i 


all week, ' ! 


'The boy ill with quinsy has to stay in bed all week, ' 

'The boy, who ill with quinsy has to stay in bed ,,,' 

Sentence (a) consists of a relative clause introduced by KOTORYJ 
'who', and contains the predicate adjective (SF) BOLEN 'ill'. 
Sentence (b) is a reduced version of sentence (a), containing 
BOL'NOJ 'ill' (LF) but not the relative pronoun. Sentence (c), 
marked deviant, contains both KOTORYJ and BOL'NOJ (LJ). Babby 
proposes an analysis of sentences (a) and (b), but does not 
address the question of why sentence (c) is ungrammatical. I 
shall repeat his analysis here and then propose an alternative 
analysis which will not only explicate the relationship between 
sentences (a) and (b) , but will also account for the deviancy 
of sentence (c). 

The deep structure proposed by Babby for the NP ' s in (a') 
and (b') is (2), 

'the boy who is ill with quinsy' 

'the boy ill with quinsy' 

fP, (nom) 

The derivation of (a') from (z) according to Babby, involves the j 
following transformations: (i) Subject-verb Agreement copies the j 
features of [mAL'CIK]]nP„ (gender, nvunber and person) onto the main ] 
verb BOL#N- 'ill', giving BOL#N- (masc sing 3rd), (ii) Relative- j 
clause Formation converts [mAL'^IKJ NP„ to KOTORYJ masc sing nom i 
'who, which'. The morphophonemic rules are assumed to subsequently. 


spell out BOl^fN- (masc sing 3rd) as BOLIiN (SF). The derivation 
of (b') from (2) involves both rules (i) and (ii) plus the op- 
tional transformation of Relative-clause Reduction, which results 
in the intermediate phrase marker of (3). 


P (nom) 




sing J 

(3) is mapped into (4), the final derived phrase marker, by the 
rule of S-node Deletion which deletes the node S and results in 
the introduction of BOI^/N- (masc sing 3rd) into the constituent 
of NP where it receives its case feature (nominative) by the 
rule of Case Marking, Again it is assumed that the morphopho- 
nemic rules will later spell out BOLi^N- plus the features 
acquired by Subject-verb Agreement and Case Marking, as BOL'NOJ 
(LF) masc sing nom 'ill'. 


Babby's transformational treatment of the SF and LF adjective 
has served both as a model and catalyst for the analysis I am 
about to propose. In fact the two analyses differ essentially in 
one small, but very significant aspect, viz., the assumption that 
J and not KOTOR- is the relative pronoun morpheme in Contemporary 

The proposed base structure for the NP ' s in (a*) and (b') 
in the present analysis is (2), the structure proposed by Babby, 
As in Babby's analysis, the derivation of (a') involves both 
Subject-verb Agreement and Relative-clause Formation, except 
that in the present analysis, the latter rule relativizes 


[maL'CIkInP to J, giving (5) 

^ Z+mascI 

+ 3TdJ 

As in Old Russian the relative pronoun morpheme is too 
slight morphologically to stand independently on the surface. 
Its syntactic status is that of an enclitic which is evidenced 
by the fact that it must either be an affix in its own position 
or be affixed to some other lexical item. Thus, (5) represents 
the stage in the derivation at which one of two mles may be 
invoked: either the rule of J-Aff ixation that attaches J to the 
lexical item at the V-node (#BOLbN#) to allow for the creation 
of the LF (#B0LbN7fj#), or the rule of KOTOR-Support is obliga- 
torily invoked, inserting the formative KOTOR- to the left of J, 
giving 7rK0T0R# J# . The phrase markers resulting from the appli- 
cation of the rules of J-Aff ixation and KOTOR-Support are 
represented in (6) and (7) respectively. 











The structure in (6) will subsequently undergo S-node 
Deletion and Case Marking, the latter rule supplying the nomina- 
tive case feature to #BOLbN#J# which, together with the features 
of gender (masc) ajid number (sing) constitute the necessary 
information for the lexical insertion of the appropriate in- 
flectional endings, i.e., ^BOLbN+ti^J-rtjjv'^. This form will emerge 
from the phonological component as the LF surface adjective 


( noni ) 




+3rd j 

The structure in (?) undergoes no additional syntactic { 
transformations. The case feature needed for the selection of 
the appropriate inflectional endings by the NP fr-K0T0R#J;7- , is 
supplied by the subject position of this NP in the embedded i 
clause. Thus, the features nom, masc, sing will provide the 
specification needed for the lexical insertion of inflectional 
endings, giving fr'K^^OR-f^ffJ+'bjf, The rules of phonology will 
convert this form to the surface relative pronoun, KOTORYJ, 

A comparison of this analysis with Babby's suggests that 
both analyses account equally well for the formation of the 
LF adjective. However, in my view, the alternative analysis 
presented in this paper has the further advantage of offering 
a more explicit explanation for why sentences like (c) *MAL'CIK, 
KOTORYJ BOL'NOJ ANGINOJ ,,, containing both the surface relative 
pronoun and a LF adjective are ungrajnmatical , Simply stated, if 
J is the same morpheme that occurs in both KOTORY(j) and BOL'NO(j), 
then the two forms cannot co-occur in the same clause. In syn- 
tactic terms, the rule of J-Aff ixation is complementary with the 
rule of KOTOR-Support . 

It has been argued above that the LF adjective is not merely 
a SF that has acquired the features of gender, number and case 
which later get spelled out as inflectional endings. Rather, it 
has been shown that a LF is a SF which, in addition to these 
agreement features, has undergone concatenation with the relative i 
pronoun morpheme by the rule of J-Aff ixation. Therefore, what 
has traditionally been considered a morphologically whole inflec- 
tional ending has been shown in this paper to be a complex item 
consisting of both derivational and inflectional formatives, 


Babby's treatment of participles in these studies will 
not be considered in this paper, 

For an earlier treatment of this question, see Levine 

(1977: Ch, h). 


According to Zaliznjalc, there is no anomaly in the 

masculine form, since in this instance the preceding; the J 

is merely a "vocalic infix" which brealcs up the inadmissable 

consonant cluster "consonant + J," Phonemic notation is 

employed here, 

Zaliznjak's discussion is limited to adjectives in the 

nominative case. In fact all of the nominative and accusative 

forms as veil as all the feminine forms in the oblique cases of 

LF adjectives show a J in their surfaces phonetic representation. 

However, this analysis derives less support from those oblique 

case endings where there is no evidence of a J, i,e,, in the 

masculine, neuter and plural forms. It should be pointed out 

that this is the same morphophonemic complication which obtains 

in Old Russian and which has resisted satisfactory historical 

explanation. One possible way to account for the absence of J 

would be to assume that certain oblique case forms arc subject 

to a transformational rule which deletes this morpheme. For a 

discussion of this problem and a proposed deletion rule of this 

type, see Coats' description of the inflectional morphology of 

LF adjectives in Old Russian (Coats 1973). 

I have altered Babby's representation of the vowel/zero 
alternation in the SF adjective, replacing ;/ with b, so as not 
to confuse this use of 77^ with its use as a boundary symbol, 

A few words should be said about the notational conventions 
employed in this description. In accordance with Chomsky and 
Halle (1968:364-369), I shall assiime that each formative in the 
lexicon is flanlced by a formative boundary symbolized by the 
plus sign +, e,g,, +X+, +Y+, +Z+, I shall further assume, 
following Chomsky and Halle, that the formative boundaries of 
lexical items inserted at the terminal nodes of phrase markers 
are deleted when adjacent to the word boundary symbol #, which, 
by convention, is automatically introduced at the beginning and 
end of every string dominated by a major lexical category, e,g,, 
#+X+f/-^ #Xf/, The concatenation of two formatives resulting from 
a CTAle of affixation will produce a string with an initial and 
final #, but with only one internal +, For example, the conca- 
tenation of j^Xf^ and ffYff will give j^X+Y,/, However, the concatena- 
tion of the relative pronoun enclitic J with a lexical item will 
be represented slightly differently. The syntactic status of J 
will be characterized by the presence of a single internal word 
botindary symbol #, between the lexical item and the enclitic J, 
Thus, when J is affixed to the complex item #X+Y#, the resulting 
new lexical item will be represented as #X+Y^J#, Finally, I will 
assume that inflectional endings are inserted to the left of the 
right word boundary of a formative, and in the case of a forma- 
tive to which J has been affixed, to the left of both right word 
boundary symbols, e,g,, tJ^X+Y+E^J+Ej^' , where E stands for inflec- 
tional ending. 



BABBY, Leonard H, 1973. The deep structure of adjectives and 

participles in Russian. Language ^9: 349-360, 
. 1975. A transformational grjumnar of Russian adjectives. 

The Hague: Mouton, 
BOLINGER, Dwight L. I967. Adjectives in English: attribution 

and predication. Lingua 18: 1-34, 
CHAPIN, Paul G, I967, On the syntax of word derivation in English, 

MIT Ph,D. Dissertation, 
CHOMSKY, Noam, 1957. Syntactic structures. The Hague: Mouton, 
and Morris Halle, I968. The sound pattern of English. 

New York: Harper and Row, 
CHOMSICY, Noam, 1970. Remarks on nominalizat ion. In Roderick A. 

Jacobs and Peter S. Rosenbaum (eds.), Readings in English 

transformational grammar. Waltham: Ginn, 
COATS, Herbert S, 1973. Old Russian declension: a synchronic 

analysis. In American contributions to the seventh inter- 
national congress of slavists. Vol, 1, linguistics and 

poetics, ed, Ladislav Matejka, pp, 67-99. 
HALLE, Morris, 1973. The accentuation of Russian words. 

Language 49: 312-48. 
HARRIS, Zellig S. 1952. Discourse analysis. Language 28: 18-23. 
JACKENDOFF, Ray. 1975. Morphological and semantic regularities 

in the lexicon. Language 5I: 639-7I. 
ICATZ, Jerrold and Paul M. Postal, 1964. An integrated theory of 

linguistic descriptions. Cambridge, Mass.: MIT Press, 
LAKOFF, George, 1970. Irregularity in syntax. New York; Holt, 

Rinehart & Winston, 
LEES, Robert B, 1960a, The grammar of English nominalizations, 

Blooraington: Indiana University, 
, 1960b, A multiply ambiguous adjectival constrxiction in 

English, Language 36: 207-221, 
LEVINE, James S, 1977. Towards a transformational account of 

derived adjectives in Russian, University of Illinois Ph.D. 

LJUNG, Magnus. 1970, English denominal adjectives, Gothenburg, 

Sweden: University of GcJteborg (Gothenburg Studies in 

English, 21,) 
MOTSCH, W. 1967. KOnnen attributive Adjektive durch Trans- 

formationen erklart werden? Folia Linguistica 1: 23-48, 
POSTAL, Paul M, 1964, Limitations of phrase stmcture grammars. 

In Jerry Fodor and Jerrold Katz (eds,), The stmcture of 

language, Englewood Cliffs, New Jersey: Prentice-Hall, 
Smith, Carlota S, I96I, A class of complex modifiers in English. 

Language 37: 11-54. 

, 1964, Determiners and relative clauses in a generative 

grajnmar of English, Language 40: 37-52. 
SUSSEX, Roland. 1974. The deep stmcture of adjectives in noun 

phrases. Journal of Linguistics. 10: III-I31. 
VAILLANT, Andre, 1942. L* Article en vievix slave. Revue des 

etudes slaves 20: 5-12. 
ZALIZNJAK, A. A. I967 . pokazateljax mnozestvonnogo cisla v 

russkom sklonenii. In To honor Roman Jakobson. Vol. 3. 

The Hague: Mouton. pp. 2328-2333. 

studies in the Linguistic bciences 
Volume b. Number 1, Spring 1978 

David Odden 

In this paper, I shall investigate certain phonological processes 
of Iraqi Arabic, based on data from Erwin (1963), Woodhead (196?) 
and supplemented by data from a native speaker of Iraqi Arabic, 
Since the majority of productive and regular morphology is found in 
the verbal system, the majority of forms for this study will 
be verbal. The primary descriptive goal of this study is to account 
for the stem and affix alternations in the various verbal paradigms, 
I shall argue that in^jerfective verb stems with the superficial 
shape CCVC are to be derived from an trndarlylng CVCVC pattern via 
a right-to-left iterative Syncope rule. It is noted that a similar 
rule exists in Tookawa which must apply in a left-to-right manner j 
this fact refutes a claim made by Jensen and Stong -Jensen (1973) 
that the direction of iteration for a rule can be determined by 
universal principles. An arguement for the cyclic application of 
the Stress Assignment rule is advanced, and a significant difference 
in the interaction between Stress Assignment and Epenthesis in 
Iraqi and Palestinian Arabic is noted. 

The Iraqi verb appears in a perfective, inqjerative and imj>erfective 
conjugatioii. An optional object pronoun suffix is avedlable in all coo- 
jugatictis. The perfective, inqjerfective, object pronoun and imperative 
affixes are provided below, 

(1) Perf. Subject Imperf. Subj. Pro. Obj. Imperative 

-ak ?i- 

-ic ?i i 

-kum ?i-.««-u 




Morphemes preceded by a dash follow the verb root and those followed by a 
dash precede the root. The subject prcaaioun sets associated with perfective, 
inperfective and inperative verb stems when realized as suffixes precede 
the object suffixes, 

lo Stress and Epenthesis 

The Stress Assignment rule of this language is essentially idBntical 
to the rule found in Classical Arabic and other Eastern Arabic dialects, 
A final heavy syllable (i.e, VCC or VC) is stressed, otherwise a penultimate 
heavy syllable is stressed, and the antepenultimate syllable is stressed 
if the two preceding conditicms are not met. The following exan?5les 
illustrate this pattern. 

Perf, Subject 

Imperf, Subj 












ti iin 













(2) 9alee 'on him' 

yihinarr 'he blushes' 
marii^ 'sick* 


naadi 'club' 
9iraa£ii 'Iraqi' 
sallatha 'her basket' 

majirasa 'school' 
larika 'conpany' 

mumaOOlla *ac tress ' 

As this pattern represents the stress pattern of virtually every word in 
the language, I assume the following Stress Assignment rule. 

(3) V * r+stress] / ^Co((VC)V(C))] 

A rule which interacts with Stress Assignment is found in the grammar 
which lengthens a vowel before the object pronouns and the dative preposition 
-il when it is cliticized to the verb. 



'we wrote' 


'we wrote them' 


'you f, studied' 


•you f, studied us' 


'he forgot' 


•he forgot her' 


'you p. sent' 


'you p, sent to him' 

The following rule is proposed here to account for this lengthening pheno- 
mencn, although the precise formulation of the rule is not critical for 

this study, 

(5) V * C+long] / + -(Dative Clitic, Object Pronounf 

I shall now turn to a consideraticn of some phonological alternations 
at work in the perfective paradigm. The first set of forms suggests the 
existence of a Syncope rule deleting an unstressed vowel in a doubly open 
syllable (i,e, when in the enviromnant VC ^CV), 










'he wrote' 
'she wrote' 
' they wrote ' 
'you m, wrote' 
'you f« wrote' 
•you p, wrote' 
'I wrote' 
'we wrote' 









'he phoned' 
•she phoned' 
'they phoned' 
'you m, phoned' 
•you f, phoned' 
'you p, phoned' 
•I phoned' 
•we phoned' 

Assuming the underlying stems to be kitab and xaabar, the following 
Syncope rule accounts for the stem alternations kitab • >. kitb and 
xaabar ~ xaabr. Examples such as kitabit and xaabarit motivate the C-stress] 
condition in the focal vowel and also indicate the necessity of restricting 
(7) from applying after word initial consonants. Examples will be provided 
in a later portion of this section showing the need to restrict this rule 
from applying after consonant clusters. 


V ♦ 

/ VC 


Additional exaii?)les of this pervasive nile are provided in (8), with stems 
followed by -a marking feminine, -een marking dual, -a marking one of a 
collective and -i? marking 'your f ,', ~ 



























'female mule' 
'female goat' 
'two lessons' 
•two graves' 
'two clerks' 
'two seas' 
'two benches' 
'a date' 
'a hair' 
■ fish' 

'your f. promise' 
'your f. books' 

Returning to the paradigms of (6), it will oe ooserved that the 1st 
and 2nd singular forms kitloit and xaab^rit have an anomalous ^phonetic 
shape in terms of stress" placement, since one would expect *kitbit and 
*xaabrit cai the analogy of 3 fern, kitbat, xaabrat, if the relevant 
verbal endings were taken to be underlying -it. The anomaly has am explanation, 
however, if we assume the underlying form of the ending to be -t and that 
the following E^senthesis rule is involved, ordered crucially after Stress 
Assignment (3). 

(9) -^ i / C 


This rule is supported by the fact that Iraqi has no final consonant 
clusters or internal clusters longer than two consonants. By this analysis, 
kitibit has the underlying representation kitabt, otress falls on the final ! 
vowel according to p), thus kit^ot, and subsequent application of Epenthesis (9J 
derives correct kitabit. If, on the other hand, we assume the relevant 
terminations to be -it, there is no natural explanation possible. 

There is additional evidence supporting the assumpticHi that the vowel 
found in kitabit is epenthetic, beyond the explanation this assumption ' 
provides for Stress Assignment, A rule (or set of rules) inserts ee between 
stem and consonant initial suffix if the stem ends in a vowel or geminated 
consonant, and optionally if the stem contains a medial glide (which is 
deleted) and a derivational prefix (such as in nlaam 'be blamed', from ! 
nliwam) . i 

(10) nfsa 

'he forgot 


'he sent' 




'she ' 




•they • 


'you m. 


'you m. ' 


'you f. 


'you f, • 


'you pi. 


•you pi.' 




'I ' 




'we ' 


•he was blamed' 


•she ' 


'they • 


'you m, • 


•you f. • 


'you pi. • 


'I • 


•we ' 

These three types of stems have in common the fact that they are in 
a sense "shorter" that the common CVCVC stem, lacking on the surface the 
final vowel or medial or final conscaiant. They thus may be characterized 


Whether this is the correct designation 
for such stems or not is not cinicial in this discussion; the important 
point is that the application of the ee-Epenthesis rule is conditioned to 
the right by a ccxisonant initial suffix. 

(11) ^ee /CC^VC^(V) +C 

The fact that niseet exhibits epenthetic ee_ whereas nisat does not is 
explainable by the hypothesis that underlying n-isa-»-at does not satisfy the 
conditions of the epenthesis nile, whereas underlying nisa+t for 'I forgot' 
doeso This contrast is brou^t out in the derivations below (elisi<ii of 
a will be considered later), 

(12) nisa+at nisa+t 

Epenthesis nisaeet 

Elision nijiat niseet 

Stress nisat niseet 

Even though -t sometimes has the vowel i, -t nevertheless behaves like a 
ccHisonant initial suffix in inducing ee-Epenthesis, 

2o Some Elision Processes 

Verb stems having a madial consonant cluster do not undergo Syncope 
when the stem precedes the endings -at and - aw, as suggested in the formulation 
of Syncope (?)• Instead, unstressed a is reduced to i. 

(13) staxdam 'he hired' staxdimat 'she hired' 
?abclae 'he made" ?aiidi9at 'she made' 
sta^mal 'he used' sti!'9milat 'she used' 
?anjaz 'he did' ?anjizat 'she did' 

The full perfective paradigm of staxdam is provided below: each verb of 
the type found in (13) conjugates in a similar fashion. 

(lU) staxdam 'he used' staxdapiit 'I used* 

staxdiijiat 'she • staxdamna 'we ' 

staxd^nit 'you m, ' sta^damtu 'you pi' 

staxdamti 'you f . ' staxdimaw 'they ' 

The following rule accounts for these alternations, 

JL$) a 4> i / [+seg]C ^CV +verb 


The patterns of reduction are limited to verbs, sipce nouns, for 
example madras a 'school', mioTnara 'pies of marble', sfarjala 'quince 'j 
exhibit no reduction of a to i. The -stress condition is required to prevent 

reduction of a in staxdaiiiit . The lefthand context C+segjC is required 

to prevent reduction of the initial vowel <xi ?ab^ddil 'I change'. 

It will be shown below that reduction also applies in the context V:C ^CV, 


It might oe assumed that ^eduction applies to underlying ki tabat 
yielding kitibat, and a revised version of byncope (7) would apply which - 

deletes only the high vowels i and u in the relevant environment. However, j 

Reduction doe^ not apply in nouns, yet unstressed a deletes in sax la , 'female , 
goat', from §axal a. Therefore, Syncope is correctly formulated to delete all 
unstressed vowels. 

We may novj return to additional examples of ^.yncope in order to show 
restrictions which must be placed on that rule, byncope also applies in 
the imperfective and imperative, assuming that the imperfective stem has the 
shape CVCVC ( I shall argue for this position below). The data in (l6) ; 

show that syncope must apply from right to left; the underlying stems i 

(containing the vocalism of the imperfective) are libas 'wear', tila9 'go out', 
kitib 'write' and suqut 'fail*. * i 

(16) yii.Das 'he wears' Tili'bsi ^ 'wear (f)l' 'j 
yltla9 'he goes out' yitil9uun 'they go out' 

t^tio 'you write' tikitba 'you write it' ^ 

yusqut 'he fails' yusuqtumi 'they fail' \ 

A full paradigm of libas is provided below. \ 

(17) yilbas 'he wears' ?aj.bas • I wear' j 
tilbas 'she wears' nilbas ^ 'we wear' j 
tiloas 'you m wear' tilibsuun 'you pi. wear' ^ 
tilibsiin 'you f wear' yilibsuun 'they wear' j 

Within the imperfective, verbs of the shape CVCi^C (such as libas ) have 
the form CCVC when followed by a consonant initial suffix or no suffix, and 
have the form CVCC when followed by a vowel initial suffix. Thus, yilibsuiin 
derives from yilibasuun via application of byncope (7) to the rightmost 
vowel in a doubly open syllable, a. The consonant cluster in the output of 
this rule created by application of Apocope prevents reapplication of ^ , 

Syncope to i. If Syncope were to apply from the left, incorrect -M- yilbasuun j 

would result. 

Jensen and Stong-Jensen (1973) have argued that the direction of ! 

application of rules can be predicted by xiniversal principles. One i 

principle is that "accent" rules such as Syncope apply in the direction that i 

minimizes application. Phelps (1975) has shown that their principles do • 

not predict the direction of a inile such as Syncope, where both directions J 
of application are equally minimizing. Phelps shows that Tonkawa has the 
following Syncope rule, 

(18) V t / VC((?)V) G V 

I^+stem^ [-suffix:] 

This rule must apply from left to right in Tonkawa. However, I have shown , 

that Syncope in Iraqi must apply from right to left. Since the rule in i 

Tonkawa and the rule in Iraqi are fundsimentally the same rule and differ in i 

directicxi of their application, the claim that directionality can be predicted j 
does not appear to be wholely tenable. i 


One might wish to argue for a different analysis of the derivation of 
yilbas ,as deriving from an underlying stem Ibas employed in the imperfective 
and derive yilibsuun from yilbas uun v ia some rule of metathesis which also 
changes vowel quality. One might then propose the following Metathesis 
rule with Raising. 

(19) V+CCVC+V -^ V+C V CC+V 


This rule directly converts underlying yilbasuun to yilibsuun . The placement 
of morpheme boundaries is crucial in this rule. For example, if the righthand 
+ is not specified, the rule incorrectly applies to the stem qtirih , which 
contains four radical consonants, in yx-^qtirifa 'he suggests', generating 
incorrect -M- yiqitrih . If the lefthand boundary is not specified, underlying 
staxdamat incorrectly becomes ^s- staxidmat rather than staxdimat 'she used', 

liider the Syncope analysis, the independently motivated Syncope 
rule is employed to account for the stem alternations within the imperfect. 
Under the Jfetathesis -Raising analysis, an additional Metathesis rule is. 
required. Furthermore, a special canonical shape of the stem, CCVC, is 
required in the iirq^erfective, since it is assumed that the surface shape of 
the imperfective libs found in yilibsuun 'they wear' derives directly from 
the vmderlying stem Ibas , The grammar must contain some statement describing 
the occurrence of CCVC stems in the imperfective, since not all stems have 
the shape CCVC on the surface. Note the following verbs which do not have 
the shape CCVC in the imperfect: in each case, the ingierfect and perfect 
are formed from the same stem, 

(20) ?a+qtiri^ 'I suggest' 
?a+st^dam 'I employ' 
?a+bdxidil 'I change* 
?a+ha^jaj 'I argue' 
?a+axrr 'I insist' 
?a+Iuuf 'I see' 
?ii+ntirid 'I am fired' 


Thus, it is only stems or the shape CVCVC in the perfect which have the shape 
CCVC in the imperfect. One can capture this relationship by the following 

(21) V ^0 / V+c ^CV Imperfective 

This rule does not claim that a phonological vowel deletion process converts 
the perfective stem into the in^serfective, merely that a relation exists between 
these two stems. 

The statement capturing the relation between imperfect and perfect stems 
under the Metathesis -Raising analysis is nearly identical to the independently 
needed Syncope rule. The Ifetathe sis -Raising analysis is therefore to be rejected 
because it contains an xinneeded metathesis rule to handle the data in (l6). 
Syncope to handle the data in (6) and (8), and a stem formation irule for the 
imperfectives of CVCVC perfects which is virtually identical to Syncope, 
The analysis advanced here assumes a single stem to underlie the perfect and 
imperfect and a alngle Syncope rule. 


One condition which must be placed on oyncope is that a vowel is 
elided if the lefthand context contains a geminated consonant. To that I 

extent, geminate consonants do not behave like clusters. 

(22) baddal 'he changed' baddlat 'she changed' ^ 
yb^ddil 'he changes' ybiaddluun 'they change' i 
?ajjal 'he delayed' ?ajjlat^ 'she delayed' ., 
y?^jjil 'he delays' y?ajjluun 'they delay' i 

• ( 

The complete perfective and imperfective paradigm of baddal 'change' is ^ 

provided below. \ 

(23) ?abaddil Is. imperf. baddalit Is, perf. j 
nbaddil Ip. " baddaina Ip. " i 
tbaddil 2m. " badd^it 2m. " 

tbaddliiji 2f. " baddalti 2fo " . 

tbaddluun 2p. " baddaitu 2p, " 

ybaddil 3ra. " baddal 3m. " 

tbaddil 3f. " baddlat 3f. " ' 

ybaddluiSh 3p. " baddlaw 3p. " j 

On the other hand, if the righthand context contains a geminate j 

consonant, an unstressed vowel is not elided. | 

(2ii) ?axallj/ha »I let her' 1 

?abaddalkum 'I change you p,' 

kitabilkumi^aa 'he wrote it to you p. • 

yistimirmiun 'they continue' 

yi¥ma?izzuun 'they are disgusted' 

V/hile deletion of vowels after geminate consonants in (22) seems to indicate 

that geminate consonants behave like single consonants and should therefore 

be represented as single consonants with the feature +long , the retention I 

of vowels before geminate consonants in (2U) suggests that geminates must 

be treated as clusters of consonants. Geminate consonants in Iraqi are thus 

an example of the dual behavior of geminates in language in generalj in • 

this example, the very same segments behave in different ways, depending on 

which side of the focus they fall on. The following revision of Syncope reflects 

this fact, admittedly in a rather unsatisfactory manner. 

(25) V ^ ^0/V(C.)C. ^CV 

[.-stressj "■ "■ 

The second restriction on Syncope is that it cannot bring two identical 
consonants together. The unstressed vowel between identical conscaiants remains j 
unelided and is reduced to i by Reduction (l5)o 

'she argued' | 

'she furnished' 

'she moved' ] 

'she wasted' 

'she specialized' 

(26) haajaj 

'he argued' 



•he furnished' 



'he moved' 



'he wasted' 



'he specialized' 



This restriction is contained in (27). 

(27) V » / V(C^)C^ C.V i / j 


Again, this is a rather unsatisfactory representation of the restriction, 
although I believe it correctly represents a restriction on Syncope rather than 
auxiliary epenthesis rules. There is no evidence that the stem vowel ever 
elides in the relevant environments in (26), and if Syncope were to have applied 
to baa.ia.l+at, yielding haajj+at , it would merge that form in shape with forms 
such as dazzat 'she sent' and saadda 'his having closed', so no epenthesis rule 
can rescue the derivation of j^adjijat . 

A derivation of representative forms is provided below. Note that 
reduction of a in l;;iaajijat provides an example of reduction after a long vowel. 

(28) baddad+at baddal+at 

Stress baddad<-at b^ddal+at 

Syncope NA b^ddlat 

Reduction b^ddidat 

staxdam+at feaajaj+at 

st^dam+at )jiaajaj+at 


st^xdimat haijijat 

The next rule I propose deletes i from the imperfective prefixes when 
in an open syllable and followed by a closed syllable (i.e. VCC of WC), 
This rule thus deletes the vowels of the imperfective prefixes ti-, ^-, and 

but not in ?a- 








•he changes' 
'he journeys' 
'he closes' 
'he sleeps* 
' he turns ' 
'he proves' 







'she changes' 
'she journeys' 
'she closes' 
•she sleeps' 
•she turns • 
•she proves • 







•we change' 

'we journey' 

'we close » 

'we sleep' 

'we turn' 

•we prove ' 

Elision of i in these prefixes is to be contrasted with retention of a in the 
first person singular forms of (30) « 

(30) ?ab4<3dil 'I change' 
?asidj3 'I close' 

?adeevmr 'I turn' 


»I journey' 


•I sleep' 


'I prove' 

The following Prefix Elision rule accounts for the reduction of prefixes in (29). 

(31) i t / #C 


The disjunction {,V,C^ represents a strong syllable; this notation is unsatisfacto 
for expressing the notion 'strong syllable', but no more enlightening notation 
is available. The condition that i must itself be in an open syllable is 
shown by nonelision in the followii7g examples where i is followed by two consonant 

(32) yitxarraj 

•he graduates • 





•he bets' 

•he specializes' 

•he suggests' 

•he is broken' 

•he employs' 


The presence of the morpheme boundary in the inile is mo1>ivated by the fact I 

that kit^ti 'you f, wrote' retains initial i where no + boundary is present. I 

'sleep* derive from the structure 

CVGVC via a Glide Deletion rule. Since the vovrel of the iraperfective prefix 
Is elided in ynaam 'he sleeps' by Prefix Elision which requires the context 

^CWG, Glide Elision must precede Prefix Elision so that the requisite W 

cluster is created. Prefix Elision must precede Syncope, however, since a stem 
of the shape ^CVCG derived from CVGVC via Syncope does not condition (31) J 
thus yikitbuun 'they write' does not show Prefix Elision, 

3o Vowel Sequences 

This section will be concerned with the question of how sequences of 
vowels are modified vdien created by deletion of intervocalic glides or by 
morphological juxtaposition. Verbs having initial glides that appear phoneticall; 
in the perfective when ijiitial exhibit deletion of that glide intervocalically 
after the inperfective prefixes. The prefix yowel and following glide 
appear to combine as follows: i+w ^ oo , i^ ♦ ee , an[ ^ 22. s ilZ 1^ 22. • 
In addition to these alternations which hold true for all glide initial verbs, \ 
the two verbs ?axad 'take' and ?akal 'eat' show apparent combination of the ] 

glottal stop and prefix vowel such that a^l f aa , il? ^ aa . Glides appear i 

phonetically in the perfective forms yi?as 'he despaired', yibas 'he dried', ^ 

w^gaf 'he stopped', wusal 'he arrived •! i 

(33) yee?as 'he despairs' ye^as 'he dries' 

?e€(?as^ 'I despair' ?eebas 'I dry' ' 

teeTsuun 'you p. despair' teebsuun 'you p. dry' 

yo^af 'he stops' yoosal 'he arrives' ] 

?oogaf / 'I stop' ?oosal^ 'I arrive' ; 

toogfuxm 'you p. stop' too|luun 'you p. arrive' 

yaaxud 'he takes' yaajcul 'he eats' 

?aaxud^ 'I take' ?a£iail^ 'I eat' 

taaxduun 'you p. take' taakluun 'you p, eat' 

The stems ?akal and ?axad are peculiar in two ways. The first is that 
a glottal stop is deleted intervocalically in these forms, whereas all other 
verbs with glottal stop retaJJi that glottal stop; thus, ?ilaf 'he was tame', 
yj!?laf 'he is tame', yi?ilfuto 'they are tame'. Secondly, only these two 
verbs have the vocalic pattern CaCaC in the perfect (c.f, (6)), whereas all j 

other CVCVC and CVCV verbs have CiCaC or CuCaC in the perfect. These two j 

verbs also appear to have CaCaC as their vocalism in the imperfect, judging ' 

from the forms ?aaxu4, ya^d, whereas all other CVCVC verbs select either i 
or u as their initial stem vowel. If we assume these two stems to be axad 
and' "akal rather than ?axad and ?akal , with initial vowels functioning in the 
place of initial ccnson^ats in terms of cancaiical patterning, the divergence 
of these two steins from all other stems will be provided with a unified . 

We must now explain the remaining contraction problems. Two analyses 
appear to be at first glance plausible. In the first analysis, the sequences 
of glides and vowels are contracted by a single rule into a long vowel having 
features of the glide. In the second analysis, the analysis to be argued 
for here, features of the glide are present on the stem vowel, glides are deleted 
and the resulting vowel sequences will be appropriately contracted. The first 


analysis requires the following rule. 

(3U) V 

+ r-3yl 7 V ^0r+syl-l 
-cons 1 -hi 

[_+5on J |_+longJ 

This analysis however fails to account for yaaxud which, deriving from 
underlying yi+sixud , does not contain the requisite glide for application of 
(3it). Additionally, the glide -contraction analysis does not account for deletion 
of stem medial glides, as will be discussed oelow. 

The -vowel contraction solution we adopt here assumes that the high vowel 
after the glide in underlying yi-*vugaf and yi-^yi?a s must be identical to the 
glide, and employs glide deletion and vowel contraction rules. Thus, yoogaf 
derives from yi-hijugaf via the stage yi-higaf . An identity constraint on high 
vowels after glides can oe argued to be independently neces.'^ary for the 
imperfective. Glide Elision is governed by the following rule. 

(35) G ^ / V V 


In addition to Glide Elision, a Vowel Contraction rule is necessary to complete 
the derivation. It will oe observed that the resultant vowel is in all 
instances a lengthened and lowered stem vowel. This generalization is stated 
in the following rule. 

(36) sf 


bhi J 

The following derivations illustrate application of these rules 


Glide Elision 




Verbs which contain a glide as the second radical consonant retain 
that glide phonetically if the prededing vowel is long or the glide itself 
is long. A stem alternation exists in case the glide is deleted so that 
CaaC appears in the perfect before vowel initial endings or when the root 
is followed by and CiC or CuC appears in the perfective when followed 
by a C initial ending. Compare the perfective conjugations of ?aaf 'see' 

and baag 'steal' 



'he saw' 


•she " • 


•they" • 


'you m. ' 


'you f .' 


•you p. ' 


'I • 
'we ' 

baag 'he stole 
baagat'she " 
ba^aw'they " 
bugit 'you m. 
bu^ti 'you f . 
biigtu 'you p. 
bi^it 'I 
bugna 'we 


The choice of vowels in the shortened (preconscxiantal) forms might be 
expected to depend on the nature of the underlying glide, based an the facts 
of Classical Arabic, However, this is not the case, as can be. seen below, 
since the glide w can correspond to either u or i in pre consonantal forms, 
K the imperfectxve stems contains a high vowel rather than aa (as in xaaf 
'fear';^ that vowel is invariably dependent on the underlying glide. 




•I saw' 


•I sold' 



'I went' 


•I visited" 



'I brought' 



'I got up' 


'I stole' 



'I feared' 



•I owed' 


'make see' 
•make sell' 
•make go' 
'make get up' 
'steal hab,» 
'lend to' 


'he sees' 

'he sels' 

'he goes' 

'he visits' 

'he brings' 

'he rises' 

'he steals' 

'he fears' 

'he owes' 

We notice that there are two forms which indicate the identity of the 
medial glide in the above stems. The glide itself appears when preceded by 
a long vowel or vriien itself long, and secondly, a cognate high vowel appears 
in the imperfective. In general, the final stem vowel of the imperfective 
can be either a,i or u, A remarkable fact about glide medial stems is that 
no verb has ^ as~its medial glide and uu as its imperfective vocalism or w 
as its medial glide and ii as its imperfective vocalism. That is, we never 
find verbs like -«- sayyaf ^ » y suiif or ■w- mawwat^ '^^^ ymiit . This fact suggests 
some constraint on the relation between glides and imperfective vowels. One 

night ' ) show that a appears as a stem vowel freely. Thus at least some nonidenti 
vowels appear after glides in the imperfect. The correct constraint is 
therefore on high vowels, that any high vowel after a glide in the imperfect 
is identical to the preceding glide. 


V #r«<backl 
r+hi] [_« round] 

/ G 

pt back"! 
L «roundJ 


This constraint explains why a can appear freely in glide medial stems in 
the inperfective but i and u are restricted. At the same time, this constraint 
explains why all stems with initial glides have identical vowels, on the basis 
of the more general fact that the initial vowel for all CVCVC stems must be 
•*-high in Iraqi Arabic (with the exception of akal and axad). 

A further fact which supports the assumption of a constraint on high 
vovrels after glides in the imperfective is that as far as I have determined, 
all stems with medial glides which are retained require a folloj^ing high vowel 
to be identical to the glide j thus, y9ayyin 'he appoints', ydaayin 'he lends' 
yfawwur 'he boils', ydaawum 'he continues', yde^wur 'he turns ' , yista jwub 
' 'he questions ', I conclude that the highvowel conatraint expressed in 
(UO) is correct. 


Turning to the mechanism for selection of CaaC versus CiC ^ shovm in 
(38), I assume that the stems underlying these alternations are siwaf and 
buwagj which become liaf and buag by application of Glide Elision, These 
stems are them shortened by deletion of the final vowel before consonant 
initial suffixes by the following rule, 

(Ul) V ^0 / V ^CC 


Notice that this rule has applied to sifit 'I saw", although the 
Inquired consonant cluster is not found on the surface. If we assume that 
the 1st singular and 2nd m, suffixes are underlying -t as we have assumed 
earlier for other reasons, and assume a rule of Epenthesis such as (9), this 
anomaly is automatically explained. These forms have the intermediate represehtati 
jfiaft, which becomes ¥ift by application of (Ul) and ¥£f it by application of 
Epenthesis, In assuming underlying -t rather than -it for the 1st singular 
and 2nd m, perfective suffixes, we have explained three otherwise unexplainable 
and unrelated facts, vis. Stress Assignment, ee-epenthesis and Shortening, 

Having accounted for the mechanism which shortens glide medial stems, 
we turn now to the mechanism for generating the allomorphs with long aa, 
which will be derived from ua or ia. It will be noticed that the long vowel 
resulting from the vocalic cluster is essentially the righthand vowel, as 
was seen to be the case with glide initial and the two vcwel initial verb 
stems in the imperfective, A rule assimilating ua and ia to aa already 
exists in our analysis. Contraction (36), Unfortunately, that rule as formulated 
is not applicable to the relevant vcwel sequences in glide medial verbs, since 
(36) requires a morpheme boundary to intervene between the vowels of the cluster. 
If this boundary were removiBd from the rule, the assimilation would correctly 
apply to siaf , yielding saaf . The revised rule would then lower all sequences 
of two vowels within a stem, and should entail that no stem can have two high 
vowels consecutively, but would predict long midvowels whenever they should 
arise. This is clearly false, cf, ysuuf 'he sees', yjifb 'he brings'. 
These forms and the forms in (39) unambiguously show that no vowel lowering 
is active within glide medial stems. 

It is therefore necessary to separate the lengthening and lowering 
features of (36) into two rules with different domains of application so 
that these data can be properly accounted for. Since Lowering only applies 
between a prefix and a stem initial vowel. Lowering is restricted to that 
environment. The specification that the focal vowel is a stem vowel is necessary 
30 that suffixes such as the feminine imperative -i do not undergo Lowering 
after vowel final verb stems, since ?l->bna-*-i becomes lihnx 'build(f)l', not 
*?ibne (this will become clear immediately following), 

(U2) V ^[-highJ/V* 


The lengthening rule on the other hand has a wider domain of application, 
applying within glide medial stems as well as glide initial stems. It is not 
a general condition of the language that whenever two short vowels come together* 
a long vowel results. Therefore, lengthening must be stated as a specific 
rule and is not a metacondition on rules in Iraqi, 


(a3) V ♦ [nong]/ V 


To review this analysis, I have assvmed that vowel sequences are 
created by Glide Deletion and by prefixing vowel final affixes to the two 
vowel initial stems, A constraint on high vowels after glides disallows 
■x- yu and -M-wi in the imperfective stem. The observed relations between deleted 
glide and resultant long vowel are analyzed as the result of this identity 
constraint on high vowels and specific rules to lengthen the second vowel of 
a W sequence, plus a rule lowering the stem initial high vowels, A derivation 
of ¥a^ and yee?as is provided below, 

ikh) siwaf yi+yi?as 

Glide Elision 

Lengthening aa ii 

Lowering NA ^ ee 

others ¥a^ yee?as 

The final rule needed to complete the derivation of aa from ia is a 
vowel truncation rule. This rule is necessary to explain facts other than 
those discussed immediately above. Whenever two vowels come together by 
suffixatioi, one of the vowels is deleted. This process has been alluded to 
in the discussion of ee_-epenthesis; there, we noted that when a vowel 
final verb stem such as bina 'build' is followed by a vowel initial suffix 
or ee is inserted by ee-~ Epehthesis (11), the stem vowel is deleted. In 
the following examples, the lefthand example shews the stem vowel followed 
by no suffix, while the righthand example consists of stem plus affix. 

(15) bina 

•he built' 


•she built 


'he builds' 


•they build' 





'he grows' 


'you (f) grow 

These data suggest the following rule. 
(a6) V ^ / +V 

When we bring into consideration the effect of object pronouns added 
to verbs, we require an extension of this rule. Recall that oefore object 
pronoun suffixes, a vowel is lengthened by Lengthening (5), as in kitabnaahum 
•we wrote them^, from kitabna-^hum . If a vowel initial object suffix is 
added to a vowel final stem, the stem vowel is lengthened by this rule and 
the suffix vowel is delated. Thus, the vowel of -iS 'you (f)', -ak •you (m) 
and -a 'him' is deleted in the examples below, 

(U7) dirabna •we hit' dirabnaa^ •we hit you(f ) • 

nisa •he forgot' nisaaK 'he forgot you (f ) 

t^imu 'he grows' tinraujf 'he grows it^ 

qira 'he reads' qiraak •he reads you(m) 

We therefore see that a short vowel deletes before or after any long vowel, 
and before another short vowel. Assuming that mirror image rules abbreviate 

rules in the order i/ X , /X "l , ttese deletions can be accounted 

for by modifying (I46) as follows. 


(U8) V ^ ^ V 


lie stress and the Cycle 

I shall now turn to a consideration of the dative clitic -il and 
the pronominal link - yya which occurs under lyingly between the dative clitic 
and the direct object pronoun on a verb. There are two logical possibilities 
for the underlying form of each of these morphemes, given the existence of 
Syncope and Epenthesisj these are -il, -1, -ivy a and - yya . I shall argue 
that the choice of -il and - yy.a as the unH'erlymg forms is the only selection 
which accounts for critical data. The vowel i appears in -il when that suffix 
stands before a consonant initial suffix, and is not found before vowel initial 

(h9) kitab+;Ll+na 'he wrote to us' kitab+1+i? 'he wrote to you(f)' 

dazz+xl+na 'he sent to us' dazz+l+i5? 'he sent to you(f)' 

Under the analysis with underlying -il, the vowel is deleted by Syncope 
(which, it will be recalled, can delete an unstressed vowel after a geminated 
consonant, as indicated by (27)) . Thus, kit^bli^ derives from kit^ili^ 
via Syncope. 

The suffix - yya selects the vowel i after a consonant, and selects no 
vowel after a vowel, as indicated below, 

(50) kitabilnayyaa 'he wrote it for us' ki-tabilkumiyyaa 'he wrote it for yc 
dazzeetilnayyaa' 'you sent it to us' dazzeetilhumiyyaa 'I sent it o them' 

It will be noticed that the vowel of the indirect object suffix na 
in kitabilnayyaa does not become lengthened by Lengthening (5). This fact 
was anticipated in the formulation of that rule, since the pronoun linking 
morpheme - yya was not included in the set of suffixes which condition 
Lengthening. The fact that the preceding a vowel of na is not lengthened 
and yet not elided in favor of i clearly shows that kitabilnayyaa does not 
have the intermediate structure kitabilna^-iyyaa^ , since according to (U8), 
the lefthand vowel of a short vowel cluster elides (as in ?i'qri •read(f)l', 
from Tigra+i). Assuming the underlying form - iyya for this suffix results 
in incorrect a- kitabi Iniyyaa . Only the -underlying form - yya explains the 
retention of the final object suffix vowels, by deriving the vowel of iyya 
through Epenthesis, whenever that vowel appears phonetically. 

Having extablished the underlying form of - yya, we can now turn to 
forms which show that the underlying form of the dative clitic must contain 
the vowel i, and which also show the ordering relation between Epenthesis 
and Syncope (specifically, Epenthesis precedes Syncope), One argument for 
assigning the dative clitic the underlying representation -il is the fact 
th^t that vowel appears everywhere when not cliticized to the verb; thus 
?ilha 'to her', ?xla 'to him'. When -il is followed by one of the vowel initial 
object suffixes -jg or - ak which are in turn followed by iyyaa, the vowel of 
those object suffixes is elided, 

(51) kitabilkiyyaa 'he wrote it to you (m)' 
kitabiiyiyya/ ^ 'he wrote it to you (f)' 
dazzeetilkiyyaa 'I sent it to you (m) • 


Assuming the indirect object clitic to be -il and assuming that Epenthesis I 
feeds Syncope, these forms are derived in the following manner, 

(52) kitabilakyya/ kitabilijyyaa ^ 
Epenthesis kitabilaklyya/ kitabiliciyyaa 

Syncope kitabilkiyyaa" kitabilJfiyyaa" i 

On the other hand, if we assume the dative clitic to have the underlying ' 

representation -1, or if we assume that Syncope precedes Epenthesis, incorrect ' 
results follow, ~ ' 


(53) kitabilikyyaa' kitablikyyaa 1 
Sync. NA kitabUkiyyaa Epenth. ^ 
Epenth, kLtabilikiyyaa'^ NA . Sync. ' 

*kitabilikiyyaa *kitablikiyyaa * 

The only analysis consistent with elision of i and a in -iS and -ak is that 
Epenthesis precedes Syncope and that the dativie clitic has the underlying ' 

form -il, ' 

— I 

At this point, we have seen the relevant rules of Iraqi Arabic, so we 
may now turn to a set of facts arguing that stress must be assigned cyclically. 
It will be recalled that I have shewn that Stress Assignment applies before I 

Epenthesis, since Stress Assipnmant applies to the abstract representation J 

kjtabt rather than derived kitabit. Thus, an epenthetic vowel is not present j 
when stress is assigned, "Now consider the following examples in terms of * 

the ordering of Epenthesis and Stress Assignment, and also in terms of 
application of Syncope. These examples contain the suffix(es) t which receives 
its i through application of Epenthesis, ~ 

(5U) dirasit 'I studied' dirasftha 'I studied her' J 

9irafit 'you knew' 9irafitha 'you knew her' ^ 

We notice two unusual facts about Verb + Object forms on the righthand side, ' 

Epenthetic i is stressed in a syllable closed by the person suffix and object j 

suffix -ha.~ Fortbermore, the stem final unstressed vowel in a doubly ojjen ] 

syllable is not elided by Syncope. Since the epenthetic vowel is stressed, ' 

it must be present when stress is assigned, which means that Epenthesis must ' 

precede Stress Assignment, But we have already shown that Stress Assignment 
precedes Epenthesis and that the epenthetic vowel cannot be present when 
stress is assigned. Thus, an ordering paradox results. 

The ordering paradox dissolves if Stress Assignment is applied cyclically. 
Bracketing dirasitha a ppropriately (ie, f Cdirastfha^), cyclic application of 
Stress Assignment allows that rule to both preced and follow Epenthesis. In 
addition, if stress is assigned to the final stem vowel in the course of the 
derivation as predicted by the cyclic hypothesis and later reassigned to the 
epenthetic vowel, presence of reduced stress on the final stem vowel prevents 
that vowels elision. The cyclic derivation of dirasitha is provided below. 


r [dirast] hal 









The cyclic application of Stress Assignment in Iraqi thus allows two anomalies 
to be related and explained without resorting to adhoc conditions on rules. 

A different argument for cyclic application of Stress Assignment in 
Palestinian Arabic based on essentially different phenomena was provided by 
Brame (197U). A crucial difference between these two dialects exists in 
terms of their treatment of Stress and Epenthesis, Whereas in Iraqi an 
epenthetic vowel is stressable on a second cycle, as in dirasftha, in Palestinian 
Arabic, the epenthetic i is never stressable. Thus, Palestinian has dar^sit 
'I studied', dar^sitha "'"l studied her'. Nevertheless, these two dialects have 
the same Stress Assignment rule and Epenthesis rule. A language specific 
statement of Iraqi grammar specifies that Epenthesis may apply between the 
two cycles of Stress Assignment, whereas in Palestinian Arabic, Epenthesis 
does not intervene. Thus, the point should be made ( and has not been made elsewhere^ 
that part of the condition that a rule applies cyclically is that the set of 
rules which intervene between one cycle of a rule and the next cycle of a 
rule must be specified in the grammar. It is a question for future research 
how three cyclic applications of a rule are to be handled and what the 
interactions between two cyclic rules might be, 


"""l;y thanks to Mike Kenstowicz and Chuck Kisseberth for having read and 
commented on an earlier draft of this paper, Ky thanks also to Saad Ahmad, 
a native of Baghdad, for providing crucial material for this study, 


BiiAI^,M, 197U. The cycle in phonologj'': Stress in Palestiniajn, Maltese and 

Spanish, Linguistic Inquiry' 5o39-60 
ERWIN,W, 1963. A short reference grammar of Iraqi Arabic, Georgetown Arabic 

Series ^^J^: Washington, 
PHELPS, E 1975 Iteration and disjunctive domain in phonology. Linguistic 

Anallysis 1,137-172 
JENSEN, J and M, Stong-Jensen 1973. Ordering and directionality of Iterative 

rules. Papers in Linguistics 6,66-90, 
iA(OODHEAD,D, 1967 A dictionary of Iraqi Arabic: Iraqi-English. Georgetown 

Arabic Series #10: Washington, 

Studies In the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 


Rajeshwari Pandharipande 

The goal of this paper is to examine two major hypotheses 
(Lakoff 1965 and Green 1974, 1976) about exceptions to syntactic 
rules and their role in defining the concept of rule-government. 
In this context the case of the exceptions to the passive rule 
in Hindi is discussed. It is argued that the exceptions to the 
passive rule in Hindi do not present true irregularity and thereby 
fail to support Lakoff 's hypothesis which treats all exceptions as 
true irregularity. The case of the Hindi passive supports the 
hypothesis presented in Green (1974, 1976) in that the exceptions 
are regular and systematic. The passive rule in Hindi applies to a 
semantically specifiable class of verbs, i.e., verbs expressing a 
volitional act. Thus, the verbs which do not express a volitional 
act fail to undergo the rule and present apparent irregularity. 
Passive is a governed rule in Hindi in the sense that it admits a 
semantic class in its structural description. It is pointed out 
with illustrations that the above hypothesis is independently 
motivated and is needed in the language. 

1.0 Introduction 

A number of recent studies in linguistics (Lakoff 1965, Lakoff, 1968, 
Green 1974, Green 1976, and Green and Morgan 1976) have discussed the phenom- 
enom of exceptions to syntactic rules and their role in defining the concept 
rule-government. Two major questions were discussed in this context: (i) 
does every case of exception to the rule present true irregularity? and (ii) 
should a rule be labelled as governed on the basis of the fact that it admits 
exceptions? The following hypotheses have been proposed: hypothesis I 
(Lakoff 1965) treats all cases of failure to undergo a rule whose struc- 
tural condition is satisfied as lexical exceptions (true irregularities). 
No distinction is made between lexical exceptions and other kinds of items 
which for some general and independently motivated reasons fail to undergo 
the rule. All rules which have exceptions are claimed to be governed rules. 
In contrast to this, hypothesis II (Green 1974, Green 1976, and Green and 
Morgan 1976) makes a distinction between lexical exceptions and apparent 
irregularity. It is claimed that rules can apply to the semantically 
specifiable class of lexical items. In such cases, lexical items which do 
not belong to the appropriate semantic class fail to undergo the rule and 
produce apparent irregularity. Governed rules, according to this hypothesis, 
are rules whose structural description mentions semantic classes. 

The purpose of this paper is to discuss the passive rule in Hindi 
(Passive hereafter) and investigate the nature of the verbs which fail to 
undergo the rule (exception-verbs hereafter). The major points of focus are 
the following: (i) Neither the traditional grammars nor hypothesis I pro- 
vide any explanation for the "exceptional" behavior of the exception-verbs. 
Hypothesis I forces us to treat these exception-verbs as lexical exceptions, 
(ii) There is independently motivated evidence to show that these verbs are 
not true exceptions. They are systematic, i.e., they fall into a coherent 


semantic class, namely, verbs expressing a non-volitional act. (iii) In 
order for the hypothesis to be descriptively adequate, it is necessary to 
treat Passive as a governed rule, in the sense proposed in hypothesis II. 
If it is claimed that Passive in Hindi applies to the verbs expressing a 
volitional act, it would explain in a straightforward fashion why certain 
verbs fail to undergo the rule. Also, it would correctly predict whether 
or not a borrowed verb in Hindi would undergo the rule. In the light of the 
above-mentioned facts, some observations about the semantics of passive 
sentences in Hindi are made. It is pointed out that passive sentences in 
Hindi express more than one meaning. However, they all share one thing in 
common, i.e., they all express a volitional act. Subject-less passive sen- 
tences in the present tense and habitual aspect express a convention and 
prescribes a particular mode of behavior, e.g., baro se istarah zorse bola 
nahi jata 'it is not customary to talk loudly like this with elders.' This 
sentence thus expresses a convention, namely, one does not talk loudly with 
the elders, and thereby it is suggested that one should not talk loudly with 
the elders. It is observed that the same type of construction is found in 
the Brahmanical texts in Sanskrit literature. The purpose of these texts is 
to illustrate the ritualistic conventions and prescribe a particular mode of 

This paper also includes discussion about the theoretical implications 
of this proposal (cf. 6.1). 

1.1 Passive in Hindi 

In this section, I will describe the passive rule in Hindi and present 
evidence to support the claim that Hindi has a passive rule. Passive in Hindi 
applies to both transitive and intransitive verbs. When a verb is passivized, 
certain changes take place in the sentence. These changes can be illustrated 
as follows: (i) the main verb of the active sentence is in its perfective 
form in the corresponding passive sentence; (ii) an auxiliary verb jana 'to 
go' is attached to the main verb; (iii) the subject of the active sentence 
(ex-subject_hereaf ter) is followed by one of the following postpositions; 
se , (ke) dwara , (ke) zariye , and (iv) when the auxiliary verb bona 'to be' 
is present in the active sentence, it follows the auxiliary verb jana 'to go' 
in the passive sentence. The surface structure of a passive sentence can be 
illustrated as follows: (Kachru 1966:181) 

ya + ja Aspect + Tense 

Compare the following non-passive sentences (1) and (2) with the passive 
sentences (l.a) and (2. a). 

Intransitive verb calna 'to leave' 

(1) ram cala 
Ram left. 


(l.a) ram se cala gaya 
ram by left went 

It was left by Ram. (Ram left.) 

Transitive verb gana 'to sing' 

(2) ram gane gata he 
ram songs sings aux 
Ram sings songs. 

(2. a) ram se gane gaye jate he 
ram by songs sang go aux 

The songs are sung by Ram. 

Notice that the verbs calna 'to leave' (1) and gana 'to sing' (2) are in their 
perfective form in (l.a) and (2. a) respectively. The subject of (l.a) and 
(2. a) is_followed by the postposition se 'by.' Also, notice that the auxil- 
iary bona 'to be' is present in (2) and is retained in (2. a), where it 
follows the auxiliary jana 'to go.' 

2.0 Verb-agreement 

The verb-agreement is not the same in active and passive sentences. 
While the verb agrees with the subject in an active sentence, in a passive 
sentence it agrees with the object of the active sentence (ex-object here- 
after). Consider the following examples: 

(3) larka citthiya parhta he 
boy letters reads aux Sp.plu. 
mas. fem, mas. 

The boy reads the letters. 

(4) larke se citthiya parhi jati he 
boy by letters read go aux 3p.plu. 3p.plu. 3p.plu 3p.plu. 
mas. fem. fem. fem. 

The letters are read by the boy. 

Notice that while the verb and aux. parhta he 'reads' agrees with the subject 
larka 'boy' in (3), the verb parhi 'read' and both the auxiliaries agree with 
the ex-object citthiya 'letters' in (3. a). 

An intransitive verb agrees with the subject in an active sentence. 
However, it remains in its unmarked form in a passive sentence. Consider the 
following examples: 

(4) larke nahi soe 

boys not slept 
3p.plu Bp.plu. 
mas. mas. 

The boys did not sleep. 


(5) larko se soya nahi gaya 
boys by slept not went 
mas. mas. mas. 

It was not slept by the boys. 
(The boys could not sleep.) 

Notice that the verb soe 'slept' agrees with the subject layke 'boys' in (4) 
but it remains in its unmarked form in (4. a). The example (3. a) shows that 
when a transitive verb is passivized, if the ex-object is not followed by a 
postposition, the verb agrees with the ex-object in number and gender in all 
aspects and tenses. Notice that the subject in (3. a) is followed by the 
postposition ^ while the ex-object is not followed by any postposition. Now 
consider the following example of a passive sentence where both the ex-subject 
and the ex-object are followed by a postposition. In this case the verb 
does not agree with either of them. 

(5) larkiyo se zakhmi admiyo ko nahi dekha gaya 
girls by injured people not saw went 
3p.plu. 3p.plu.obj. 

fem. mas. mas. mas. 

The injured people were not seen by the girls. 
(The girls could not bear to see the injured people.) 

Notice that the ex-subject larkiya 'girls' is followed by the postposition 
se . Also, the ex-object adml 'people' is followed by the postposition ko. 
The verb (in its perfective form) and the auxiliary dekha gaya 'was seen' 
do not agree with either the ex-subject or ex-object. 

The agreement rules for passive sentences can be stated as follows 
(for further discussion see Kachru 1966:182) 

(i) \^en a transitive verb is passivized, if the ex-object is not 
followed by ko, the verb agrees with the ex-object in number and gender 
[recall (3. a)]. 

(ii) However, if the ex-object is also followed by a postposition ko, 
the verb remains in its unmarked form. 

(iii) If 
marked form. 

an intransitive verb is passivized, it remains in its un- 

The verb agreement in a passive sentence in Hindi abides by the axiom 
of the verb agreement in Hindi, namely, a verb does not agree with a noun 
which is followed by a postposition. Thus the case of the verb-agreement in 
the passive sentences is not different from the case of active sentences 
where the subject is followed by a postposition. In both cases the verb 
does not agree with the subject. Notice the following example of non-passive 

(6) larko ne kam kiya 
boys ag. job did 
mas. mas. mas. 

The boys did the job. 


(7) bacco ko ma yad ati he 
children mother memory comes 
3p.plu. obj.m. 
mas. fem. fem. 

The children remember the mother. 

Notice that in both (6) and (7), the subject larke 'boys' in (6) and bacce 
'children' in (7) is followed by a postposition' (ne and ]co respectively) and 
the verb is in its unmarked form in (6) and agrees with ma 'mother' in (7). 

2.1 Evidence for Passive: Ref lexivization 

In this section, I will present evidence to show that Hindi has a 
passive rule. The Hindi reflexive pronoun apna 'self refers to the sub- 
ject of the Hindi sentence. Consider the following sentence: 

(8) ram ne syam ko apni jagah par bheja 
ram ag. ^yam obj .M. self's place on sent 
Ram sent syam to his place. 

Ram' s 

V _ 

Notice that the reflexive apni 'self refers to Ram (subject) and not Syam 
(object). However, the following example shows that the reflexive in the 
corresponding passive sentence refers to the subject as well as the object. 

(9) ram se syam ko apni jagah par nahi bheja gaya 
ram by syam obj.m. self's place on not sent went 
Syam was not sent by Ram to his place. 

S^am ' s 
Ram' s 

The most straightforward way to account for the facts of ref lexivization 
in Hindi is by analyzing (8) and (9) as being transformationally related. 
For, consider the alternative. If the above assumption is not made, sen- 
tences like (9) may lead one to believe that the reflexive in Hindi refers 
to subject as well as object. This is patently false, as sentence (8) 
makes clear. The fact is that the reflexive in Hindi refers to both sub- 
ject and object if and only if the sentence is passive. In active sentences 
it refers only to the subject. This fact is easily accounted for if we 
treat (9) as being derived from (8) by the rule (Passive) which promotes the 
object to the subject position, then we can explain the behavior of apni 
'self's' in a straightforward fashion, i.e., examples (8) and (9) do not 
contradict the generalization namely, the reflexive apni 'self's' always 
refers to the subject and not the object. It is because Syam [object in 
(8)] is promoted to subject-position by the rule_(Passive) , that it behaves 
like the subject and therefore the reflexive apni 'self's' can refer to 
^yam [ex-object in (9)]. This evidence indicates the following: 

(i) Hindi has a passive rule. 

(ii) Passive is a relation-changing rule in Hindi 

and (iii) Although (9) provides evidence for object-promotion, it does not 
provide any evidence for subject-demotion [recall (9) where the reflexive 


continues to refer to ram (the ex-subject)] in a passive sentence. (For 
more discussion on the subject of passive sentences in Hindi, see Kachru 
et al 1976). 

2.2 Evidence for Passive: Subject-deletion 

The ex-subject is generally deleted in a passive sentence in Hindi. 
Thus Hindi has the following kind of passive sentences: 

(11) i kam kiya gaya 

work did went 
The work was done. 

If we do not consider this deletion as a consequence of the Passive, and 
thereby sentences like (11) as derived sentences, then we will have to treat 
these sentences as being identical to the other subject-less sentences 
like (12). 

(12) kam hua 
work happened 

The work got done. 

However, for the native speakers of Hindi, (11) and (12) are not semantically 
identical. For them the agent (subject) is implied in (11) while it is not 
in (12). This intuition of native speakers is correctly supported by the 
fact that when the agent and its agency is explicitly denied in (11) and 

(12) respectively, (11. a) presents contradiction while (12. a) does not. 

(11. a) * kisike na karne par bhi kam kiya gaya 
anybody's not doing inspite of work did went 
The work was done without anybody's doing it. 

(12. a) kisike na karne par bhi kam ho gaya 
anybody's not doing inspite of work happened 
The work got done without anybody's doing it. 

(11. a) indicates that the agent is implied in it while (12. a) indicates that 
it is not. The following assumptions explain the situation in a simple 

(i) The underlying structure of (11) is (13), in which the agent is 
present. (11) is derived from (13) by a rule which generally deletes the 
ex-subject after the application of passive rule. 

(13) X ne kam kiya 
X ag.M. work did 
X did the work. 

(ii) The underlying structure of (12) is (14) which is the same as its 
surface structure. 

(14) kam hua 
work happened 

The work got done. 


Notice that the subject is not present in the underlying structure of (12) 
whereas it is in the underlying structure of (11). Thus in order to differ- 
entiate the sentences like (11) from (12), it is necessary to assign to them 
two different underlying structures and derive (11) from (13) by the passive 

3.0 Exceptions 

A number of verbs in Hindi fail to undergo the passive rule. The list 
of the exception-verbs consists_of the verbs (intransitive) like dikhna 'to 
seem,' khilna 'to bloom,' murjhana 'to whither,' galti bona 'to make a mis- 
take,' bikni 'to sell,' tutna 'to break' and the verbs (transitive/ 
intransitive) which require their subject to be in the dative/oblique case 
i.e., lagna 'to feel,' pasand bona 'to like' (literally, liking happen), 
malum bona 'to know,' yad ana 'to remember' (literally, memory come) and 
sirdard honi 'to have a headache' (literally, headache happen) etc. 

The traditional grammars (Guru 1920, Vajpeyi 1958, Kellogg 1955), do 
not discuss these exceptions. Within the framework of hypothesis I (Lakoff 
1968), these verbs will have to be treated as lexical exceptions (true 
irregularity) and will have to be marked (-Passive) in their lexical entry. 
Also, since the passive rule in Hindi allows these exceptions, it will be 
marked as a governed rule. 

3.1 The "capabilitative" meaning 

Before I present a semantically oriented explanation for the "excep- 
tional" behavior of these verbs, it will not be inappropriate to see what 
a passive sentence means in Hindi. A passive sentence in Hindi, with the 
ex-subject present on the surface gnerally expresses the capacity of the 
subject to carryout the act expressed by the verb (Vajpeyi 1958, Kachru 
1966). Consider the following sentences: 

(15) larke se citthiya nahi likhi jati 
boy by letters not wrote go 

The letters are not written by the boy. 
(The boy is not capable of writing letters.) 

However, whether or not a verb expresses capability of the subject, does not 
help us define the exceptions for the following reasons: (i) verbs which 
express capability of the subject, do not necessarily undergo Passive and 
(ii) not all passive sentences in Hindi express the capabilitative meaning 
(Kachru 1966:183-184). Passive sentences in Hindi cover a wide range of 
meaning. A descriptively adequate analysis of the "exceptional" behavior 
of the verbs should take into account all the meanings expressed by those 

In what follows, I will present evidence to show that the capacity of 
the subject either to undergo or to carryout the act expressed by the verb 
is not crucial to decide whether or not a verb will undergo Passive. Consider 
the following sentences: 

(16) phul khilta he 
flower blooms aux 
The flower blooms. 


(17) mujhe accha lag raha he 
to me good feel prog, aux 
I am feeling fine. 

Notice that the subject phul 'flower' in (16) is capable of undergoing the 
act of blooming. Similarly, the subject jc^ 'I' in (17) is capable of feel- 
ing (fine). However, when the verbs khilni 'to bloom' and lagna 'to feel' 
are passivized, (16. a) and (17. a) are ungrammatical. 

(16. a) * phul se khila jata he 

flower by bloomed goes aux. 
It is bloomed by the flower. 

(17. a) * mujh se accha laga ja raha he 

me by good felt go prog. aux. 
It is felt good by me. 

This evidence shows that the subject's capacity to carry out or to undergo 
the act expressed by the verb is not crucial to decide whether or not a 
verb will undergo Passive. 

3.2 In this section I shall show that a passive sentence expresses the 
capabilitative meaning only when the ex-subject is present. When the ex- 
subject is deleted, passive sentences lose the capabilitative meaning 
(kachru 1966). Consider the following sentences: 

(18) larke se citthiya nahi likhi jati 
boy by letters not wrote go 

(i) The letters are not written by the boy. 
(ii) The boy is not capable of writing letters. 

(19) i citthiya nahi likhi jati 

letters not wrote go 

The letters are not written. 

Notice that when the ex-subject is present, (18) expresses the capabilitative 
meaning. However, when the ex-subject larke se 'by the boy' is deleted, (19) 
loses the capabilitative meaning. 

3.3 In (3.2) we have seen that subject-less passive sentences do not 
express the capabilitative meaning. Now I will present various meanings 
expressed by the subject-less passive sentences under different conditions. 
First, I will present subject-less passive sentences which express a con- 
vention, a custom or a tradition in the present tense and the habitual 
aspect. These sentences are used to prescribe a particular mode of behavior. 
Consider the following sentences: 

(20) baro se istarah zor se nahi bola jata 
elders with like this loudly not talked goes 

(One does not talk) loudly like this with the elderly people. 

(21) diwali par ghar ghar me dip jalae jate he 
diwali on _house house in lamps lit go aux. 
On the Diwali-day lamps are lit in every house. 


(22) am chilkar nahi khae jate 
mangoes having peeled not ate go 
Mangoes are not eaten having been peeled. 

Notice that the ex-subject is not present in (20), (21) and (22). (20) 
expresses a convention, namely, it is not customary to talk loudly with the 
elders and thereby suggests that it is not proper to talk loudly with the 
elders. (21) expresses the custom of lighting lamps of the Diwali-day 
(festival of lights, celebrated in India); and thereby suggests that one 
should light lamps on the Diwali-day. (22) expresses the convention that 
the mangoes are not peeled. (They are either cut into small pieces or 
their juice is squeezed out.) This sentence suggests that one should not 
peel mangoes. All the three sentences above [(20), (21) and (22)], carry 
an habitual aspect and present tense. These sentences lose their "pres- 
criptive" meaning when their aspect or tense is changed. Consider the 
following examples: 

(23) diwali par dip jalae gae 
diwali on lamps lit went 

(perfective) ^past) 
The lamps were lit on the Diwali-day. 

(24) am chilkar nahi khae ja rahe (he) 
mangoes having peeled not ate go prog, (aux) 

The mangoes are not being eaten having been peeled. 

Notice that (23) is in the perfective aspect and past tense, while (24) 
is in the progressive aspect and present tense. In both cases the "pres- 
criptive" meaning, expressed by (21) and (22) is blocked. Neither (23) 
not (24) express any custom, tradition or convention, instead, they refer 
to a particular incident at a particular point in time. Thus in order for 
passive sentences to convey the "conventional" meaning, the role of the 
aspect and the tense seems to be crucial. The "capabilitative" meaning 
is also blocked in both (23) and (24). 

This "prescriptive" usage of passive sentences in present tense and 
habitual aspect has precedent in the history of Indo-Aryan. The Brahmanical 
texts in Sanskrit make extensive use of subject-less passive sentences. 
The purpose of these texts is to interpret the Vedic texts in the context 
of the sacrificial rituals. These texts describe the sacrificial rituals 
and illustrate the duties of the priests and Yajamana (the host for whom 
the sacrifice is performed), etc. (For more discussion see Keith 1928 and 
Gonda 1975). A close examination of the data shows that subject-less pas- 
sive sentences here are utilized to express conventions and thereby precribe 
particular modes of behavior. The following examples illustrate the 

(25) sama giyate 
sacrificial prayer is sung 

gai + ya (pass.) + present 
The sama is sung. (l^atapatha Brahmana 5.2.46) 


(26) ahutir huyate 
oblation is offered 

hu + ya (pass.) + present 
The oblation is offered, ( ^atapatha Brahmana 3.1.23,28) 

(27) sa svahetyevajuhot tasmat u 

he svaha thus saying offered therefore indeed 
svahetyeva huyate 
svaha thus saying is offered 

hu + ya (pass.) + present 
He offered (the oblation) with (the_words) "svaha!". 
Therefore (it is) offered with svaha. ( Satapatha Brahmana 

Notice that the subject is not present in (25) — (27). (25) expresses a con- 
vention, namely, the sama is sung. It also expresses a "prescriptive" 
meaning namely, the sama should be sung. (26) expresses a convention, 
i.e., the oblation is offered and thereby suggests that tha oblation should 
be offered. In (27), the process of offering the oblation is described. 
The first half of the sentence describes an incident i.e., he offered the 
oblation, reciting the word "svaha!" (an exclamation used in offering 
oblations to gods). The second half of the sentence describes the conven- 
tion, 'therefore it is offered with the recitation of the word "svaha!".' 
This sentence conveys a "prescriptive" meaning, namely, the oblation should 
be offered with the recitation of the word "svaha! "^ This "prescriptive" 
meaning conveyed by the passive sentences in the Brahmanas , is restricted 
to the present tense and the habitual aspect. The following sentences in 
the past tense do not convey this "prescriptive" meaning: 

(28) tasmac saro nama yad asiryata 
therefore arrow named because was broken 

a(past M) + sri + past 3sg. 
Therefore the designation arrow, because it was broken off. 
( Satapatha Brahmana 2.4.1) 

(29) sapta rsayah asrjyanta 
seven sages' were created 

a(past) +srj + ya(passive) + past 3sg. 
The seven sages were created. 
( Satapatha Brahmana 8.4.3 — 6) 

Note that (28) and (29) do not express a prescriptive meaning like (26) and 
(27). While (25)~(27) are in the present tense, (28) and (29) are in the 
past tense. In this section I have shown that: 

(i) When the subject is deleted, the passive sentences in Hindi lose 
their "capabilitative" meaning. 

(ii) In the present tense and the habitual aspect, the passive sen- 
tences express a convention and prescribe a particular mode of behavior; 

(iii) The passive sentences in the Brahmanas behave similarly [For more 
discussion on the function of the passive sentences in the Satapatha 
Brahmana , refer to Pahdharipande : 1976 (unpublished) . ] 


This similarity between the function of the passive sentences in the 
Brahmanas and in Modern Hindi leads to a speculation namely, Hindi might 
have inherited this usage of passive sentences. However, in order to arrive 
at this conclusion, it is necessary to investigate the function of passive 
sentences in the literature_written between the period of the Brahmanical 
texts and Modern Hindi. Prakrt, Apabhram^a and Old Hindi mark three major 
stages of Indo-Aryan between the period of ^atapatha Brahmana and Modern 
Hindi. If the investigation of the texts shows that the "prescriptive" 
function of passive sentences is retained through those stages in the 
development of Indo-Aryan, then it can be claimed that Modern Hindi inher- 
ited the "prescriptive" function of passive sentences. At this point this 
is an open question and needs further investigation. 

3.4 In this section, I will show that in some "regi=;ters" (or 'function- 
ally determined language-types' Halliday :1964) of Hindi (i.e., the news 
paper register and the scientific register), the subject-less passive 
sentences express an "impersonal" meaning. In this case, a passive sentence 
is generally followed by a sentential object. These sentences are generally 
comparable to English sentences such as, "it is believed that smoking is 
bad for health" or "it is said that apples are good for health." Consider 
the following sentences: 

(30) i kahajatahe ki nehru bare dayalu admi the 

said goes aux that nehru very kind man was 
It is said that Nehru was a very kind man. 

- _ _ _ V '^^ 

(31) ajkal yeh suna jata he ki har des me 

now-a-days this heard goes aux that every country in 

kapro ke dam barh rahe h"^ 

clothes of cost increase prog, aux 

Now-a-days it is heard that in every country the prices of clothes 

are increasing. 

Notice that in (30) and (31) the ex-subject is not present and they convey 
an "impersonal" meaning. The "capabilitative" meaning is not conveyed by 
these sentences. B. Kachru (1977) suggests that this function of passive 
sentences (i.e., to express an "impersonal" meaning) has been borrowed from 
English into Hindi as a result of the language-contact between English and 
Hindi over a long period of time. 

A.O Volitional act and the Hindi Passive 

The discussion in (3.0) — (3.4) shows that passive sentences in Hindi 
express a variety of meanings. However, whether or not a verb will undergo 
Passive cannot be judged on the basis of these meanings because none of 
these meanings is shared in common by all passive sentences in Hindi. 
Recall that 

(i) The "capabilitative" meaning is lost in the subject-less passive 
sentences (cf. [3.2]). 

(ii) The "prescriptive" as well as the "impersonal" meaning is governed 
by the structure (it has to be a subject-less construction) and tense and 
aspect (present tense and habitual aspect for the "prescriptive" meaning). 


Therefore, neither the "prescriptive" nor the "impersonal" meaning can be 
said to be the characteristic of the verbs which undergo Passive. 

A close examination of passive sentences in Hindi shows that they 
express a volitional act, in the sense that the subject's volition controls 
the act expressed by the verb, as has been observed by Kachru (1972) and 
Hasan (1971). In the following discussion it will be shown that every 
passive sentence (irrespective of what kind of meaning it conveys) expresses 
a volitional act. The examples in focus are (3. a), (21) and (30). Notice 
that (3. a) expresses the "capabilitative" meaning, (21) expresses the 
"prescriptive" meaning and (30) expresses the "impersonal" meaning. The 
hypothesis that (3. a), (21) and (30) express a volitional act is supported 
by the following examples (32) — (34). Examples (32), (33) and (34) are 
the counterparts of (3. a), (21) and (30) respectively. In (32) — (34) it 
is stated that the act expressed by the verb took place although the subject 
did not do it intentionally. In other words, the subject's volition is 
explicitly negated in those sentences. If our hypothesis is correct, i.e., 
if the passive sentences express a volitional act, then such negation of 
the subject's volition must present contradiction. Notice that (32) — (34) 
do in fact present contradiction. 

(32) "larke se citthiya parhi jati he; janbujhkar nahi 

boy by letters read go aux intentionally not 

The letters are read by the boy (but) not intentionally. 

(33) *diwali par ghar ghar mi dip jalae jate he; janbujhkar nahi 

diwali on every house in lamps lit go aux intentionally not 
The lamps are lit on the Diwali-day, (but) not intentionally. 

(34) *Kaha jata he — janbujhkar nahi — ki nehru bare dayalu admi the 

said go aux intentionally not that nehru very kind man was 
It is said (but) — not intentionally — that Mr. Nehru was a very kind 

Notice that in the first clause in (32) — (34), the subject's volition is 
negated. Also notice that (32) — (34) are ungrammatical. This ungrammati- 
cality can be explained in a straightforward fashion if we assume that in 
(3. a), (21) and (30), subject's volition is presupposed; therefore, its 
negation in (32), (33) and (34) respectively, presents contradiction. The 
examples (32) — (34) show that the volitional act is characteristic of all 
passive sentences in Hindi. 

4.1 Volitionality and the exception-verbs 

The discussion in the following section is focused on two major points: 
first, whether or not a verb will undergo Passive crucially depends on 
whether or not it expresses a volitional act. Second, the classification 
of the verbs into two classes, i.e., verbs expressing a volitional act and 
verbs expressing a non-volitional act, is independently motivated. 

4.1. a. Inanimate subject and the exception-verbs 

The verbs whose subject is generally inanimate (and thereby does not 
have any volition) typically fail to undergo Passive. Notice that verbs 


such as khilna 'to bloom', mur j hana 'to wither', ubalna 'to boil' (intran- 
sitive) and bikna 'to sell', generally take inanimate subjects such as 
phul 'flower' ( khilna 'to bloom' and mur j hana 'to wither'), pan! 'water' 
(ubalna 'to boil') and kitab 'book' ( bikna 'to sell'). Those verbs do not 
undergo Passive. Consider the following examples: 

(35) *phul se khila jata he 

flower by bloomed go aux 
It is bloomed by the flower. 

(36) *Kaliyo se murjhaya jata he 

buds by withered go aux 
It is withered by the buds. 

(37) *pani se ubla jata he 

water by boiled goes aux 
It is boiled by the water. 

(38) *ghar se bika jata he 

house by sold goes aux 
It is sold by the house. 

Now consider the following verbs which take animate or inanimate subjects. 
The examples (39) and (40) show that those verbs are passivized only when 
their subject is animate. 

(39) larki andar ati he 
girl in comes aux 
The girl comes in. 

(39. a) (larki se) andar aya jata he 
(girl by) in came goes aux 
It is come in by the girl. 

(40) hava andar ati he 
air in comes aux 
The air comes in. 

(40.a)*hava se andar aya jata he 
air by in came goes aux 
It is come in by the air. 

Notice that the verb ana 'to come' can take either animate (39) or an inani- 
mate subject (40). However, when Passive is applied to (39) and (40) 
respectively, (39. a) is grammatical, while (40. a) is not. (40) differs 
from (39) in that the subject of_(40) is inanimate ( hava 'air'), while the 
subject of (39) is animate ( larki 'girl'). This discussion shows that in 
order for the verb to undergo Passive, the subject's volition must control 
the act expressed by the verb or, in other words, the verb must express a 
volitional act. 


4.2 The negation test 

The purpose of this test is to determine, independently of the passive 
construction, whether or not the subject's volition controls the act 
expressed by the verb. If the subject's volition controls the act in a 
sentence, its negation in the conjoined sentence should result in contradic- 
tion. Now contrast the_verb torn! 'to break' (which undergoes Passive) with 
the exception-verb girna 'to fall'. When it is claimed that the action 
took place against the subject's volition, the use of the regular verb 
torna ' to_break' produces contradiction, while the use of the exception- 
verb girna 'to fall' does not. 

torna 'to break' (regular verb — undergoes Passive) 

(41) ??? m£ ne phuldan tora halaki me torna nahi cahti thi 

I ag.M. vase broke although I to 'break not wanted 
I broke the vase although I did not want to break it. 

girna 'to fall' (exception-verb) 

(42) us bhir me vah larki gir pari halaki vah girna nahi 
that_crowd in that girl fell down although she to fall not 
cahti thi 


That girl fell down in that crowd although she did not want to. 

Notice that when the volition of the subject me 'I' in breaking the vase is 
negated in the conjoined sentence, (41) produces contradiction, while (42) 
is fine when the volition of the subject vah larki 'that girl' is negated in 
the conjoined sentence. This result of the negation test shows that the 
verbs which undergo Passive (i.e., torna 'to break') express a volitional 
act while the verbs which fail to undergo Passive (i.e., girna 'to fall) 
express a non-volitional act. 

5.0 Evidence from constructions involving volitionality 

In this section I will present evidence to show that the verbs which 
fail to undergo Passive, typically fail to participate in the constructions 
which require the verb to express a volitional act. The regular verbs are 
contrasted with the exception-verbs and it is shown that the regular verbs 
can participate in those constructions. However, then the exception-verbs 
are used, the result is ungrammatical . 

5.1 The Imperative Construction 

The first one of this type of construction is the imperative construc- 
tion. It is generally assumed that the subject's volition controls the act 
expressed by the verb in the imperative construction. Thus in English, in 
a sentence like 'carry it!', it is assumed that the volition of the implied 
subject 'you' controls the action of carrying. Also, notice that if a verb 
does not express a volitional act, it fails to participate in this 
construction, i.e., 

(43) *Like it! 


(44) *Resemble it! 

Notice that the subject's volition does not control the act of liking (43) 
and resembling (44). The following examples show that this construction is 
blocked for the exception-verbs. 

(45) *yad ao ! 'be remembered!' 

(46) Alago! 'feel!' 

(47) "khoo! 'lose yourself!' 

Now consider the following examples: 

(48) yad karo ! 'remember!' 

(49) toro! 'break it! ' 

(50) likho! 'write! ' 

Notice that the verb yad ana 'to be remembered' (45), lagna 'to feel' (46) 
and khona 'to lose oneself (47), are exceptions to Passive, while the 
verbs yad karna 'to remember' (48), torna 'to break' (49) and likhna 'to 
write' (50) are not exceptions to Passive. This discussion shows that 
subject's volition does not control the act expressed by the exception- 
verbs ((45) — (47)), while the subject's volition controls the act expressed 
by the verbs which undergo Passive ((48) — (50)). 

5.2 Kosis karna 'try to' construction in Hindi 

Another construction which requires the verb to express a volitional 
act is the one in the English sentences such as "Bill tried to help John". 
Notice that the verbs which participate in this construction are verbs like 
"to please" (i.e., Jim tried to please his boss), "to kill" (i.e., Sam 
tried to kill Bill) and not verbs such as "to resemble" (??? John tried to 
resemble Bill). Here again, the subject's volition is involved in the acts 
expressed by the verbs of pleasing, killing, and helping while it is not 
involved in the act expressed by the verbs such as "resemble". 

Now let us consider the Hindi verbs of this type. The following 
examples show that while this construction is_blocked for the exception- 
verbs, such as khona 'to lose one self, yad ana 'to be remembered', it is 
not for the verbs which undergo Passive. 

— — V— V — 

(51) sudha ne sab janne ki kosis ki 

sudha ag.M. all to learn tried 
Sudha tried to learn everything. 

~, — v— V — 

(52) mc ne soneki kosis ki 

I ag. M. to sleep tried 
I tried to sleep. 


(53) *dilip ne khoneki kosis ki 

dilip ag. M. to lose himself tried 
Dilip tried to lose himself. 

(54) *seema ne yad aneki kosis ki 

seema ag. M. to be remembered tried 
Seema tried to be remembered. 

Notice that the use of verbs such as janna 'to learn' (50) and sona 'to 
sleep' (52) in this construction, does not produce ungrammaticality . In 
contrast to this, when used in this construction, the verbs khona 'to 
lose one self (53) and yad ana 'to be remembered' (54), result in 
ungrammaticality. The verbs j anna 'to learn' (51) and sona ' to sleep' (52) 
undergo Passive while khona 'to lose oneself (53) and yad ana 'to be 
remembered' (54) do not. On the basis of this discussion, we can infer 
that the subject's volition does not control the act expressed by the 
exception-verbs while it does in the case of the verbs which undergo 

5.3 The dative-subject-verbs and non-volitionality 

The verbs which require their subject to be in the dative case are 
exceptions to Passive in Hindi. A native speaker of Hindi intuitively 
feels that those verbs do not express a volitional act. Thus for him the 
subject's volition does not control the act of remembering expressed by 
the dative-subject-verb yad ana 'to be remembered'. The following evidence 
supports this intuition. Notice that in (55) and (56), the dative-subject- 
verb yad ana 'to be remembered' (literally, memory come) is contrasted with 
the verb yad karna 'to remember' (literally, to do memory). In (55) and 
(56) it is stated that the act of remembering expressed by the verbs yad ana 
'to be remembered' (to come to memory) (55) and (56) respectively took 
place against the subject's volition. While (56) presents contradiction, 

(55) does not. 

(55) (mere) na cahnepar bhi mera dost mujhe bar bar 

(my) _ not wanting in spite of my friend to me again & again 

yad ata he 

remember aux 

Although I do not want to, I remember my friend over and over again. 

(56) ??? (mere) na cahnepar bhi me apne dost ko 

(my) not wanting in spite of I my friend Obj , M. 
bar bar yad karta hQ 
again & again remember aux 
Although I do not want to, I remember my friend over and over again. 

The examples (55) and (56) indicate that the subject's volition does not 
control the act expressed by the dative-subject-verb yid ana 'to be remem- 
bered' (55). Therefore, even when the act is said to have taken place 
against the subject's volition, no contradiction results. In contrast to 
this, the subject^s volition controls the act of remembering expressed by 
the verb yad karna 'to remember' (56). Therefore, when the act is said to 
have taken place against subject's volition, (56) presents contradiction. 
Notice that the verb yad ana 'to be remembered' is an exception-verb while 

yad karna 'to remember' is not. The hypothesis, i.e., dative-subject- 
verbs do not express a volitional act, is further supported by the cross- 
linguistic evidence presented in Krishnamurti (1975), McAlpin (1976), and 
Sridhar (1976). 

5.4 In this section I will present more evidence for the hypothesis that 
the_exception-verbs express a non-volitional act. The Hindi adverbial 
dhyan se 'carefully' is comparable to the English adverb 'carefully'. 
Notice that the adverb 'carefully' (in English) can accompany the verbs 
like "drive" (i.e., He drives carefully) and "listen" (i.e.. Listen care- 
fully!), where the subject's volition controls the act of driving and 
listening respectively. However, notice that "carefully" does not occur 
with the verbs such as "get hurt" (*John got hurt carefully) or "like" 
(*I carefully liked the movie), where the subject's volition does not 
control the act expressed by the verbs "to get hurt" and "to like" 
respectively. Thus in English, the adverb "carefully" does not accompany 
a verb which does not express a volitional act. The Hindi data shows that 
dhyan se 'carefully' accompanies only those verbs which undergo Passive 
and typically does not accompany the exception-verbs. Consider the follow- 
ing examples: 

(57) jan dhyan se parta he 
John carefully studies aux 
John studies carefully. 

(58) *jan dhyan se kho gaya 

John carefully lost himself 
John carefully lost himself. 

Notice that the adverb dhyan se 'carefully' can occur with the verb parhna 
'to study' but not with the verb khona 'to lose oneself. While p arhna "to 
study" undergoes Passive, khona 'to lose oneself does not. This shows 
that the exception-verbs do not express a volitional act whereas the verbs 
which undergo Passive express a volitional act. 

6.0 Conclusion 

The major points in the preceding discussion can be summed up as 
follows: First, that Hindi has a passive rule (section (1.0)). Second, 
that the list of exceptions to Passive in Hindi consists of both transitive 
and intransitive verbs. Within the framework of hypothesis I (Lakoff 1965), 
these verbs will have to be treated as lexical exceptions and be marked as 
(-Passive) in their lexical entry. Third, passive sentences in Hindi 
express a variety of meanings under various conditions (sections (3.2) — 
3.4)). However, all of them share one thing in common, i.e., they all 
express a volitional act (4.0). Fourth, the discussion in (4.1) — (5.4) 
reveals that the verbs which undergo Passive express a volitional act while 
the exception-verbs do not. 

On the basis of these observations, the following conclusions can be 
drawn: The exceptions to Passive in Hindi do not present true irregularity. 
Passive in Hindi applies to verbs which express a volitional act. Since 
the exception-verbs typically express a non-volitional act, they fail to 
undergo the rule and present apparent irregularity. Thus Passive in Hindi 


should be labelled as a governed rule in the sense discussed in Green 
(1974), Green (1976), and Green and Morgan (1976), i.e., it applies to a 
semantically specifiable class of verbs. 

The classification of verbs into two classes (i.e., the verbs express- 
ing a volitional act and the verbs expressing a non-volitional act) is 
independently motivated (recall the discussion in sections (5.0) — (5.4). 
The verbs belonging to these classes behave differently from each other 
with regard to other syntactic constructions in Hindi. Thus this classi- 
fication enables us to explain the "exceptional" behavior of the exception- 
verbs and captures the generalization, namely, that the exception-verbs 
are exceptions to the constructions which require the verb to express a 
volitional act. The hypothesis which marks these exceptions as true 
irregularity, misses this generalization. 

The hypothesis proposed in this paper explains why some borrowed 
verbs (i.e., verbs borrowed from English) undergo Passive while some others 
do not. Notice, Hindi has borrowed several nouns from English, e.g., 
ekting (acting), ^ak (shock), etc. (For further discussion see B. Kachru: 
1977). Some "hybrid" compound-verbs are formed with these nouns. The 
process of deriving those compound-verbs is as follows: An operator (a 
native Hindi verb) is attached to the word borrowed from English; these two 
(the borrowed noun and the operator) function as a unit — a compound-verb. 
Thus the following verbs are formed: ekting karna 'to act' or 'to pretend'. 
Notice that in this compound -verb, ekting 'acting' is the borrowed noun and 
karna 'to do' is the Hindi verb which is used as an operator. Now consider 
the other compound-verb ^ak lagna 'to get a shock' or 'to be shocked', in 
which ^ak 'shock' is the borrowed noun and lagna 'to feel^ to be attached' 
is the Hindi verb used as an operator. While ekting karna 'to act' is 
passivizable, sak lagna 'to be shocked' is an exception to Passive. Under 
hypothesis I, the grammar of Hindi will have to mark ^ak lagna 'to be 
shocked' as an exception to Passive. However, it will not explain why some 
of the verbs formed with the borrowed nouns undergo Passive while some do 
not. In the framework of our hypothesis, this situation can be_explained 
in a straightforward fashion. Notice that the exception-verb ^ak lagna 'to 
be shocked' is a dative-subject-verb. The discussion in (5.3) shows that 
the dative-subject-verbs express a non-volitional act and therefore it is 
only expected that it will not undergo Passive. Also, the behavior of 
these verbs with regard to the constructions discussed in (5.1), 
(5.2) and (5.4) shows that while ekting karna 'to act' expresses a 
volitional act, ^ak lagna 'to be shocked' does not.^ 

6.1 The discussion in this paper raises some theoretical questions: 

It has been observed (Green and Morgan 1976) that all rules that change 
grammatical relations are governed. However, it is not clear as to what 
degree of relation-changing capacity is necessary for a rule to be governed. 
The Hindi Passive changes the grammatical relations only partially. Notice 
that the ex-object in the Passive sentence (9) controls ref lexivization 
and thereby provides evidence for object-promotion. However, (9) does not 
provide evidence for subject-demotion (i.e., the ex-subject continues to 
control ref lexivization in (9)). Kachru et al. (1976) have shown that the 
subject of the passive sentence in Hindi is less subject-like. On the 
whole, the Hindi Passive only partially changes the grammatical relations. 

The discussion in this paper shows that Passive is a governed rule in 
Hindi. Can we then say that the rules which only partially affect the 
grammatical relations are also governed? This assumption will give rise 
to the following questions: 

(a) Are grammatical relations always discrete? 

(b) What degree of relation-changing capacity of a rule is necessary 
in order to establish a correlation between relation-changing rules and 

The "prescriptive" meaning and the "impersonal" meaning conveyed by 
some passive sentences in Hindi provide evidence for the role of language- 
use in determining the meaning of a sentence. This raises the question, 
how did Hindi acquire the "capabilitative" meaning? Is it a function of 
language-use or of the linguistic structure? 

The Hindi Passive shows that the syntactic behavior of the verbs is 
determined by their role in the semantic structure of the language. One 
might therefore ask, to what extent semantic structure controls the 
syntactic structures of a language? At present these are open questions 
and further investigation is needed. 


I am greatly indebted to Professors Yamuna Kachru, Braj Kachru and 
Jerry Morgan for their invaluable suggestions, criticisms and guidance but 
for which this paper would be much worse than it is. I claim the respon- 
sibility for the shortcomings that remain. 

The influence of English is seen at various linguistic levels (i.e., 
lexicon, morphology and syntax) in the newspaper register of Hindi. "A 
number of syntactic patterns which are transferred from English have a 
high frequency in the newspaper register such as in sports news and in 
headings and captions" (B. Kachru 1977). Kachru points out two major 
reasons for the direct influence on this particular register of Hindi 
"one reason being that until recently, English was used as the main source 
for translating news items into Hindi, since the facilities for direct 
communication of news in Hindi (say, teleprinters in Hindi) were not 
available. Second, a large number of successful English newspapers started 
sister publication in Hindi (i.e. , The Hindustan Times and Hindustan 
(New Delhi) etc,)." 

In Hindi, traditionally, active forms of verbs are used to convey the 
"impersonal" meaning. However, the following paragraphs from newspapers 
illustrate the use of passive to convey the "impersonal" meaning in the 
newspaper register: 

(a) yah bhi mag ki gai ki ukt sarasthan me dhan ki 
this also demand^did went that said province in finances of 
herapheri ki jac ki jay 

mishandling of investigation did go (optat.) 

It has been demanded that there be an investigation of the mishandling 
of finances in the said province. ( Navbharat Times , New Delhi, 
October 19, 1977). 


(b) yaha tak kaha gaya ki lina ke thik bote hi use aspatal 
here upto said went that line of recovering_soon her hospital 
se sidhe kahi esi jagah bhej diya gaya . . . 
from straight some such place sent went 

It was even said that soon after her recovery Lina was sent from the 
hospital to some place . . . ( Saptahik Hindustan , May 22, 1977). 

^There is evidence to show that the verb ekting karna 'to act' 
expresses a volitional act; in contrast to this, ' ^ak lagna 'to get_a 
shock' does not express a volitional act. Notice that ekting karna 'to 
act' participates in the constructions which require the verb to express 
a volitional act, while ^ak lagna 'to get a shock' fails to participate in 
those constructions. Consider the following examples: 

(a) The Imperative construction (cf. 5.1) 

1. ekting karo 'act!, do the acting!' 

2. *^ak lago 'get a shock!' 

(b) kosis karna 'try to' (cf. 5.2) 

■ = •> ^ _ M — M — 

3. ram ne ekting karne ki kosis ki 
Ram tried' to do the acting^ 

4. --ram ne ^ak lagne ki ko^i^ ki 
ram ag.M. shock get tried 
Ram tried to get a shock. 

(c) dhyan se 'carefully' (cf. 5.4) 

5. vah dhyan se ekting karta he 
he carefully acting does aux 
He does the acting carefully. 
6.* vah dhyan se ^ak lagta he 
he/him gets a shock carefully. 

Notice that constructions such as Imperative, 'Try to' and 'carefully' are 
blocked for the verb ^ak lagna 'to get a shock' (recall (2, 4, and 6)). On 
the other hand, those constructions not blocked for ekting karna 'to act' 
(1, 3, and 5). On the basis of this, we can conclude_that while ekting 
karna 'to act' expresses a volitional act, the_verb ^ak lagna 'to get a 
shock' does not. Also, the fact that ^ak lagna 'to get a shock' is a 
dative-subject-verb indicates that it expresses a non-volitional act. Our 
hypothesis_correctly predicts that while ekting karna 'to act' will undergo 
Passive, ^ak lagna 'to get a shock' will not. 


GONDA, J. 1975. Vedic literature. Otto Harrassowitz. Wiesbaden. 
GREEN, G. 1974. Semantics and syntactic regularity. Indiana University 

• 1976. The notion of rule government. Problems in linguistic 

matatheory. Department of Linguistics. Michigan State University. 
and MORGAN, J. 1976. Notes toward an understanding of Rule 

Government. Studies in Linguistic Sciences. Urbana; University of 

Illinois. Volume 6, Number 1. 
GURU, K. P. 1920, 1952. Hindi Vyakran. Kashi Nagri Pracarini Sabha. 



HALLIDAY, M. A. K. , MCINTOSH, A., and STREVENS, P. 1964. The Linguistic 
sciences and language teaching. Indiana University Press. Blooming- 

HASAN, R. 1971. Syntax and semantics: Biological and Social factors in 
Psycho-linguistics. Ed. John Morton, London. 

KACHRU, B. 1977. Englishization of Hindi. Linguistic Method. Papers in 
honor of Herbert Penzel. Ed. I. Rauch et al. De. Ridder Press. 
The Hague. 

KACHRU, Y. 1966. An introduction to Hindi syntax. Urbana; University of 
Illinois, mimeographed. 

. 1972. Notes on grammatical categories and participant roles 

in Hindi-Urdu (unpublished). 

, KACHRU, B., and BHATIA, T. 1976. The notion "subject": 

discussion of three North Indian languages. The notion of Subject in 

South Asian Languages. Ed. M. K. Verma. University of Wisconsin 

Publication No. 2. 
KELLOGG, S. H. 1955. A grammar of the Hindi language. Rouledge and 

Kegan Ltd. London. 
KEITH, A. B. 1928. A history of Sanskrit literature. Oxford. 
KRISHNAMURTI, Bh. 1975. Verbs of cognition in Telugu. Osmania Papers in 

Linguistics, 1:1. Hyderabad. Osmania University. 
LAKOFF, G. 1965. On the nature of syntactic irregularity. Indiana 

University dissertation. Published 1970 as Irregularity in Syntax. 

LAKOFF, R. 1968. Abstract syntax and Latin complementation. M.I.T. Press. 
MCALPIN, D. 1975. Dative subjects in Malayalam. The notion of subject 

in South Asian languages. Ed. M. K. Verma. University of Wisconsin. 

Madison publication series, publication 2. 
MCGREGOR, R. S. 1972. Outline of Hindi Grammar. Clarendon Press. London. 
PANDHARIPANDE, R. 1976. On the syntax and semantics of the ya-passlve 

(present) construction in Satapatha Brahmana (unpublished) . 
POSTAL, P. and PERLMUTTER, D. 1974. Relational grammar (unpublished). 
Shatapatha Brahmana. 1967. Volumes I, II and III. Published by The 

Research Institute of Ancient Scientific studies. New Delhi. 
SRIDHAR, S. N. 1976. Dative subject, rule government, and relational 

grammar. Studies in Linguistic Sciences, Vol. 6, No. 1. 
VAJPEYI, K. D. 1966. Hindi Shabdanuahasan. Kashi Nagri Pracarini Sabha. 


Studies in the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 

Irmengard Rauch 

Semiotics, the science of signs, derives its recent lin- 
guistic heritage from standard bearers such as Saussure, Sapir, 
Bloomfield, Jakobson and, indeed, Chomsky. Bloomfield's (1939) 
claim that 'Linguistics is the chief contributor to semiotic' 
and Jakobson's (1971) hypothesis that '...any human communication 
of non-verbal messages presupposes a circuit of verbal messages' 
indicate a modus operandi , namely the function in which all of 
semiotics displays an inlay of language, in which non-verbal 
communication is LANGUAGE -LIKE. It is proposed that the iso- 
lation and identification of discrete elements shared by idio- 
morphic semiotic systems and language semiotic systems provide 
the cardinal data. The universal primitive of LENGTH is 
extracted in both a set of language examples and a set of 
music/architecture examples. The LANGUAGE- LIKENESS of the 
music and architecture semioses is found to reside in the proc- 
essing of the primitives via the inferential function of ab- 
duction -- a function peculiar to language. 

Linguistic theory, which is currently in the midst of a new wave of 
iconoclasm, appears on the one hand to be leaning toward the laboratory 
sciences for insights into the composition and linguistic functions of 
the brain, while on the other hand the probings of linguists resemble ever 
more the speculations of philosophy. The duality is understandable, since 
language is both physical and mental. Several schools of philosophy claim 
language as their principal subject matter, thus providing a ready tap for 
linguists whose sole object is, in fact, language. More immediately, 
within our own recent linguistic heritage we can uncover an appropriate 
philosophical framework suggested by none other than Saussure, continued 
by such standard bearers as Sapir, Bloomfield, and Jakobson, and even 
endorsed by Chomsky. This, of course, is the discipline of semiotics, 
the science of signs. 

Saussure proposed in his Course in general linguistics (1966:16) 'A 
science that studies the life of signs within society. . . ' He says 

'I shall call it semiology. Semiology would show what consti- 
tutes signs, what laws govern them. Linguistics is only a part 
of the general science of semiology; the laws discovered by 
semiology will be applicable to linguistics, and the latter 
will circumscribe a well-defined area within the mass of 
anthropological facts. ' 

Sapir (1958:104) saw that '...every cultural pattern and every single act 
of social behavior involves human communication in an explicit or implicit 
sense.' The anthropological theorist Claude Levi-Strauss formulated very 


clearly a concept of society with an integrated theory of communication con- 
sisting of three types of exchange: exchange of messages (linguistics), ex- 
change of goods (economics), and exchange of mates (kinship). Thus semiotics 
is, in effect, a science studying communication within society. We will as- 
sume, for the moment, that this is society within the species homo sapiens 
and, for the major part of this paper delay reference to semiotics outside 
this species, i.e. zoosemiotics. Nevertheless, within human semiosis, we 
must account for all three types of Levi-Strauss ' communication, which, in 
turn, have many subtypes. 

Let us return to our linguistic heritage. Bloomfield tells us in his 
Linguistic aspects of science (1939:55) that 'Linguistics is the chief con- 
tributor to semiotic' How is this possible if we are to understand, ac- 
cording to Sebeok (1975:95), that 'Semiosis, independent of form or sub- 
stance, is... seen as a universal, criterial property of animate existence'? 
In other words, what do we do with that communication or semiosis which is 
not the chief contributor to semiotics? Jakobson (1971:662) writes '...the 
subject matter of semiotic is the communication of any message whatever, 
whereas the field of linguistics is confined to the communication of verbal 
messages.' Temporarily he answers our question: Linguistics '...has a 
narrower scope, yet, on the other hand, any human communication of non-verbal 
messages presupposes a circuit of verbal messages, without a reverse impli- 
cation.' Thus Levi-Strauss (1963:83) claims for his three types of communi- 
cation or exchange that '...language comes into play at all levels', i.e. it 
preexists each exchange, it accompanies each, and it serves as a means of 
translation for each non-verbalized exchange. This last function serves as 
the focal point of the present paper: We want to know how language serves 
as a means of translation for non-verbalized exchange, how non-verbal mes- 
sages presuppose a circuit of verbal messages, i.e. how all of semiotics has 
an INLAY of language, how non-verbal communication is LANGUAGE -LIKE. 

The centrality of linguistics prompted Roland Barthes (1970:11) to re- 
verse Saussure's statement ('Linguistics is only a part of the general sci- 
ence of semiology' cf. above) by the statement '...linguistics is not a 
part of the general science of signs, even a privileged part; it is semiol- 
ogy which is a part of linguistics....' However, upon close inspection we 
see that Barthes overstated his case, and that, in fact, he distinguished 
language which is the object of linguistics from what he termed 'second- 
order language,' which is the object of semiotics. 

Clearly semiotics and linguistics are independent sciences, but just 
as clearly they are intimately related, since language provides an inlay 
for all modalities of semiotics. When we observe human language in action, 
we observe an organism (a human being) producing an expression (a sound, 
orthographic sequence, gesture) in order to refer by it to something (an 
object) (Carnap 1961:8-9). Thus the language act is comprised of (1) a 
speaker, (2) an expression, and (3) a designatum, which correlate directly 
with the tripartite division of semiotics, viz. pragmatics, syntactics, and 
semantics, respectively. For some linguists the semiosis of verbal language 
appears to be a challenge easily commensurate with life-time devotion, so 
that they opt not to treat data from speech surrogates such as cryptography, 
whistle speech, drumbeat communication, and so-called paralinguistic struc- 
tures such as laughing, crying, whispering, and noises of the tongue and 


lips associated with interjections. Still more linguists exclude from their 
domain idiomorphic semiotic data, i.e. data from semiotic systems which are 
not variform direct superstructures on languages. Samples of idiomorphic 
systems can be readily named by reference to the senses, thus kinesics and 
proxemics relate to the senses of touch and sight, music to the auditory 
sense, architecture to sight, odor to the olfactory sense, culinary semiotics 
to the sense of taste as well, and so forth (cf. further Rauch forthcoming 

Let us consider the likeness between e.g. music and language, or ar- 
chitecture and language? Saussure (1966:17) simply says 

'...the characteristic that distinguishes semiological systems 
from all other institutions shows up clearly only in language 
where it manifests itself in the things which are studied 
least, . . . ' 

That was around 1915, and apparently we are still groping to identify those 
least studied things, to which Saussure ascribes the semiotic mechanism, 
but never mentions as such. Somewhat parallel is the proposal of Jakobson 
(1971:703) that 

'The cardinal functions of language--referential, emotive, 
conative, phatic, poetic, metalingual--and their different 
hierarchy in the diverse types of messages have been out- 
lined and repeatedly discussed. This pragmatic approach 
to language must lead mutatis mutandis to an analogous 
study of the other semiotic systems....' 

But we ask, what is this mutatis mutandis ? It is precisely the crux of the 
problem. One is then not surprised to read as recently as 1975 in an arti- 
cle entitled 'Spoken and unspoken meanings' by Cohen (24), that 'It remains 
quite an open question whether this clearly recognizable domain of study 
[the semiotic domain] will lend itself to the construction of non-trivial, 
empirically testable theories at some appropriately deep level of explana- 
tory generality' (cf. further Rauch 1977). 

It seems that the comparison among diverse modalities of massive or 
large scale relationships, as represented perhaps in the cardinal functions 
of language (above) is somewhat premature. The linguist may have more suc- 
cess in relying on what is recognized as a principal time-proven tool of 
all modern linguistic methodologies, viz. their ability to isolate and 
identify discrete elements. Of the several theories of semantic develop- 
ment, the universal primitives hypothesis, recently proposed especially by 
Postal (1966:fn. 10) and extended by Bierwisch (1967:1-4, 34-6 and 1970: 
181-2), appears in particular to link with semiotic theory. In effect. 
Postal and Bierwisch hold that the real world, or input into the human or- 
ganism, is interpreted in the form of universal semantic components or prim- 
itives in a species specific manner. 

Let us accordingly choose a semantic primitive and identify it in 
several modalities, in order to understand how it is interpreted into lan- 
guage in non-linguistic semiotic systems. The primitive which we choose is 


LENGTH or DURATION, pinpointed by Miller and Johnson-Laird (1976:27J as a 
universal. To be sure it is readily identifiable in linguistic semiotics, 
but even within linguistic data, LENGTH is more transparent in certain 
structures than in others. Thus, in the semantic component, e.g. in the 
sentence 'A mile is longer than a kilometer,' the perception of LENGTH is 
more direct than on the syntactic level, which may be exemplified by gapping 
in e.g. 'John reads Dickens, Joe Hemingway, and Jill Kant' as opposed to 
'John reads Dickens, Joe reads Hemingway, and Jill reads Kant.' A morpho- 
logical example of LENGTH often cited in semiotic literature for its icon- 
icity is the non-suppletive comparison of adjectives in Indo-European, where 
the positive, comparative, and superlative suffixes mirror a stratified 
phoneme increment, thus e.g. old , older , oldest . On the phonological level 
LENGTH is, of course, perceptible where it alone distinguishes one word from 
another; thus in Kamba, a Bantu language, three degrees of vowel LENGTH are 
contrastive, e.g. kwelela 'measuring', kwele- la 'moving backwards and for- 
wards', koele: la 'aiming at.' Consonantal LENGTH is observed in Italian 
fato 'fate' versus fat:o 'done.' We are, of course, familiar with express- 
ive or affective LENGTH as in the English expression too bad. This leads 
us technically into what have been considered borderline linguistic levels, 
specifically what Trager (1958) has called 'paralanguage. ' For example, his 
so-called voice qualifier of extent, which would oppose 'drawl' to 'clip- 
ping', represents the feature LENGTH on the paralinguistic level. 

Leaving language and paralanguage but still dealing with acoustic 
LENGTH, let us cross over to the semiotic system of music. An example is 
the choral sequence of the first twelve notes of equal LENGTH in the Ode to 
Joy of Beethoven's Ninth Symphony. Excepting performance modifications, any 
one of the twelve notes which might be played shorter or longer relative to 
its adjacent notes would be perceived as incorrect. Not surprisingly, 
Gerald Warfield, who in his Layer analysis (1976) explicates the highly 
structured principles underlying tonal music, stresses that (xi) ' 
can be (and j^, most of the time) understood intuitively. . . . ' 

Before considering how musical LENGTH is language- like, let us apply 
the LENGTH primitive to a very concrete semiotic system, viz. architecture, 
and specifically to the basic box, which is man's most common form of con- 
struction, be it a house, a skyscraper, or whatever. Miller and Johnson- 
Laird (1976) distinguish between two systems in conceptualizing an object in 
space, 1) a deictic system which is relative to the speaker's ego location 
and the coordinate axes of an object, and 2) an intrinsic system which is 
relative to the coordinate axes of the object itself. In the case of the 
box, if the horizontal dimension, labeled A, is its maximal dimension, it 
is intrinsically the 'length' or 'longer than' the secondary horizontal 
dimension, labeled B, which is its 'width' or 'shorter than' A. Interest- 
ingly, Miller and Johnson-Laird wrote (404) : 'Recognizing the intrinsic 
nature of the parts of some objects seems to be something that users of 
English manage to do without self-consciously conceptualizing it.' Once 
more we superimpose the question: How then, is the LENGTH feature in ar- 
chitecture language- like? 

It is possible that structure or component analogues between linguistic 
semiotic systems and non-linguistic semiotic systems may prove insightful, 
e.g. Uspenskij's (1973) level of articulation of an icon as an analogue to 


the phonological level in natural languages. This is a level of figuration 
which corresponds to the most general painting processes, which are in them- 
selves non-significant, but which allow the painter to represent meaningful 

However, beyond such analogues the primitive LENGTH, whether in the 
visual arts, music, paralanguage, or language, is ultimately interpreted as 
a perceptual judgment which derives solely from that semiotic system alone 
capable of propositions, i.e. language. Postal (1966:fn. 10) argued that 
'...the relation between the semantic primitives and their combinations 
which are part of the combinatorial structure of language and the world is 
not learned but innate.' Bierwisch (1970:181) hypothesized that '...all 
semantic structures might finally be reduced to components representing the 
basic dispositions of the cognitive and perceptual structure of the human 
organism. ' 

Now this innate ability of the mind to process semantic primitives can 
find an explanation in the theory of inference of Charles Sanders Peirce, 
the co-founder with Saussure of modern semiotic. According to Peirce we 
experience external objects as percepts and interpret them through a cogni- 
tive operation. All cognition involves inference and the type of inference 
which Peirce holds to be the most common kind of reasoning is abduction, 
whereby the minor premise is derived from the major premise and the con- 
clusion. Our first premises are perceptual judgments, viz. the formation 
of a mental proposition concerned with our sense experience when making the 
judgment. Peirce defines this innate manner of interpreting an experience 
or a perceived object as '...a judgment asserting in prepositional form what 
a character of a percept, directly present to the mind, is' (1960 V:par. 54). 
Meaning thus originates in the perceptual judgment, which itself can be re- 
garded as an extreme case of abductive inference. The term 'educated guess' 
closely approximates the abductive process, which Peirce claims is instinc- 
tive. The interpretation of our chosen primitive LENGTH then, as displayed 
in the various semiotic modalities, derives from one and the same innate 
ability of judgment and hypothesis formation, an ability which belongs only 
to the semiotic modality of language. 

Thus the LENGTH cognition of the non- linguistic modalities presented in 
this paper is language-like in that it proceeds from reasoning of the fol- 
lowing sort: 

I. for the musical semiosis: 

Each of the twelve notes equals one beat. 
E is one of the twelve notes. 
E equals one beat. 

II. for the architectural semiosis: 

Maximal dimension equals the length. 
A is the maximal dimension. 
A equals the length. 

Observe that the middle statements are abductively inferred. 


Before concluding let us turn briefly to zoosemiotics, purposely ex- 
cluded thus far. Peirce (1960 VI:par. 491) compared man's instinctive power 
of so-called guessing with the instincts of animals for feeding and breeding. 
Sebeok (1972:46-7) describes what we may discern as differences of LENGTH of 
the bee dance among six races of bees. The hundred meter distance between 
the hive and the food source is calculated at six different LENGTHS in revo- 
lutions per fifteen seconds: Italian 7.95, Austrian 8.4, German 9, Punic 
9.05, Caucasian 9.8, Egyptian 9.25. Man's observation of the LENGTH differ- 
ential is, of course, not the issue, and Sebeok rightly reports that an en- 
counter between e.g. an Austrian bee and an Italian bee would precipitate 
confusion of signals. 

Purposely excluded also has been any species designation for the word 
language , i.e. we have refrained from writing human language . Accordingly, 
the questions which arise directly from the present paper: Can one even 
contemplate a language-inlay in non-verbal semiosis? and if so. What is 
the language- likeness in the animal code? join those provocative questions 
presently stimulating the necessary and inevitable reappraisal of the con- 
cept of language itself (cf . Rauch forthcoming b) . 

To conclude, it appears appropriate to turn to Chomsky's little known 
endorsement of the Peircean theory of inference referred to at the begin- 
ning of this paper. In Language and mind (1972:93) Chomsky writes: 

"Speculating about the future, I think it is not unlikely that 
the dogmatic character of the general empiricist framework and 
its inadequacy to human and animal intelligence will gradually 
become more evident as specific realizations, such as taxonomic 
linguistics, behaviorist learning theory, and the perception 
models, heuristic methods, and 'general problem solvers' of the 
early enthusiasts of 'artificial intelligence' are successively 
rejected on empirical grounds, when they are made precise and 
on grounds of vacuity when they are left vague. And--assuming 
this projection to be accurate--it will then be possible to 
undertake a general study of the limits and capacities of 
human intelligence, to develop a Peircean logic of abduction." 


BARTHES, Roland. 1970. Elements of semiology. Translated by A. Lavers 

and C. Smith. Boston: Beacon. 
BIERWISCH, Manfred. 1967. Some semantic universals of German adjectivals. 

Foundations of Language 3.1-36. 
. 1970. Semantics. In J. Lyons, ed.: New Horizons in linguistics, 

pp. 166-84. Harmondsworth, Middlesex: Penguin Books. 
BLOOMFIELD, Leonard. 1939. Linguistic aspects of science. (International 

encyclopedia of unified science, vol. 1, no. 4.) Chicago: University 

of Chicago Press. 
CARNAP, Rudolf. 1961. Introduction to semantics. Cambridge, MA: Harvard 

University Press. 
CHOMSKY, Noam. 1972. Language and mind. New York: Harcourt, Brace, 



COHEN, J. Jonathan. 1975. Spoken and unspoken meanings. In T. A. Sebeok, 

ed. : The tell-tale sign: a survey of semiotics, pp. 19-25. Lisse: 

Peter de Ridder Press. 
JAKOBSON, Roman. 1971. Roman Jakobson: selected writings II. The Hague: 

LEVI -STRAUSS, Claude. 1963. Structural anthropology. Translated by 

C. Jacobson and B. G. Schoepf. New York: Basic Books. 
MILLER, George A. and JOHNSON- LAIRD, Philip N. 1976. Language and per- 
ception. Cambridge, MA: Belknap Press of Harvard University Press. 
PEIRCE, Charles S. 1960. Collected papers of Charles Sanders Peirce. 

vol. V and VI. Cambridge, MA: Belknap Press of Harvard University 

POSTAL, Paul M. 1966. Review article of Elements of general linguistics, 

by Andre Martinet. Foundations of Language 2.151-86. 
RAUCH, Irmengard. 1977. Review of The tell-tale sign: a survey of 

semiotics, ed. by T. A. Sebeok. Lingua 41.183-89. 
. forthcoming a. Semiotics and language. Language Quarterly 

(University of South Florida) . 
. forthcoming b. The language-inlay in semiotic modalities. 

SAPIR, Edward. 1958, Selected writings of Edward Sapir in language, 

culture and personality, ed. by D. G. Mandelbaum. Berkeley: Univer- 
sity of California Press. 
SAUSSURE, Ferdinand de. 1966. Course in general linguistics, ed. by 

C. Bally et al . Translated by W. Baskin. New York: McGraw-Hill. 
SEBEOK, Thomas A. 1972. Perspectives in zoosemiotics . (Janua linguarum, 

series minor, 122.) The Hague: Mouton. 
. 1975. Zoosemiotics: at the intersection of nature and culture. 

In T. A. Sebeok, ed. : The tell-tale sign: a survey of semiotics, 

pp. 85-95. Lisse: Peter de Ridder Press. 
TRACER, George L. 1958. Paralanguage: a first approximation. Studies 

in Linguistics 13.1-12. 
USPENSKIJ, Boris A. 1973. Per I'analisa semiotica delle antiche icone 

russe. Ricerche semiotiche: nuove tendenze della scienze umane 

nell' URSS, ed. by J. Lotman and B. A. Uspenskij , Italian edition 

ed. by C. S. Janovic. Translated by C. S. Janovic et al. Turin: 

WARFIELD, Gerald. 1976. Layer analysis: a primer of elementary tonal 

structures. New York: David McKay. 

^jtudies in the Linguistic Sciences 
Volume b. Number 1, Spring 1978 


Pamela S. Silver 
Scott R. Krause 

This paper is intended to refute the traditional ways of 
accounting for the word-initial voicing alternations in the class 
5/6 noun-adjective series of Shona. We will show that none of 
the traditional analyses is correct and that the voiced /voice less 
phenomenon under discussion should be looked upon as the result 
of a morphophonemic mile. In other words, we are proposing that 
there is neither a segmental nor a prefix in class 5, and that 
our morphophonemic analysis provides a more concrete solution. 

When examining any corpus of Shona data, one is immediately struck by 
the occurrence of a voice less /voiced alternation between pairs of consonants 
in certain paradigms, e.g.. 

(1) class 5 

class 6 







mater) ga 








bveni doko 

mapfeni madoko 

'small baboon' 

^ira 3ena 

macira ma'^ena 

'white blanket' 

daka gukutu 
gaoo ^ma 

madaka makukutu 

•hard mud' 

makiK^o matema 

•black baboon • 

J»are dzvuku 

matare matsvuku 

'red ball' 

Traditionally, this voicing alternation has been looked upon as the 
result of either: 

(2) a) a prefix being lost in the NOUN/ADJECTIVE series of class 5; 

b) a morpheme ( 2^ , i" , etc.) which has no overt phonological shape 
in class 5 but which shows itself in the voicing of certain 
consonants (Fortune, 173; Fivaz and Ratzlaff 72, etc.); 

c) the "prefix" for the NOUN/ADJECTIVE series of class $ actually 
being voicing (Heny, 211), or scMne similar claim. 

We hope to show in the following paper that none of these analyses 
provides a correct synchronic explanation and that the voiced/voiceless 
phenoBtenon under discussion be looked upon as the result of the following 
morphonemic rule: 







Our proposal, then, is that there is no prefix for the class 5 noun/ 
adjective series. 

We shall proceed as follows: First we will examine the dlatinctive 
features and environment of the proposed rule and show why each one is 
necessary. Second, in the course of the discussion, we will examine a few 
features which might be proposed for our rule and show why these are un- 
necessary or erroneous in nature. 

As we add more and more forms to our list of data, we note that the 
observed voicing alternation does not occur with every possible voiced/ 
voiceless pair of consonants that exist in Shona. That is to say, the voic- 
ing alternation only occurs with phonemes which are [♦stop]] : 



class 5 



6iM (^/c in 

our notation) 

class 6 



ma tare 



'wooden mortar' 







But such an alternation does not occur with fricatives: 
(5) v/f z/s i/^ 

class 5 

class 6 



•possessing spirit 

















ma zai 






ma fuza 





Thus it appears that since non-stops do not show the voicing alternat- 
ion, and since all stops or stop-initial phonemes do, the most natural way 
to account for the facts at this point in our discussion is to set up a 
voicing alternation between stops. 


_ c 


L-vc J 

c , 

+vc J 

Crucial evidence is lacking as to whether affricates should be analyzed" 
as a single (+stopJ phoneme (Chomsky and Halle, 317), or as consonant 
clusters consisting of two separate phonemes. If the second alternative 


were chosen, a rule of voicing assimilation would be necessary, i.e., 
(7) ^ C ^ [^vcJ^C 

Since the evidence needed to determine the status of affricates is lacking, 
we will adopt Chomsky and Halle's framework, and will refer to affricates as 

r+stop ~] . 

[+delayed releaaej 

If l+voicingj is the morphological marker for class 5, as proposed by 
Heny, Fivaz and Ratzlaff, etc., why is it that only stops are affected? In 
all other classes, all words, regardless of their initial consonants (or 
indeed vowel), receive the prefix associated with that class. For example, 
all class 6 nouns and ad.jectives are prefixed by #ma-. Because of this we 
would expect all initial consonants in class 5 to beconts voiced, if in fact 
the class 5 prefix were C*voicing] or some similar abstraction. That this 
is not the case, as has been seen from the preceding data, argues against 
such proposals. 

So far we have given evidence that there is a voicing alternation between 
stops in classes 5 and 6. The next logical question is in which direction 
this change takes place, i.e. (+vc] j' [-vcj , or L-vcJ :> t+vcj . The evidence seems 
to point toward [-vc]^l+vc] for several reasons. There are cases of 

(8) a) class 5 


^ class 6 


alternations, as in the 


class 5 

class 6 


'strong wind' 

as well as 

b) class 5 


^ class 6 


altematioiis, as in the 


class 5 
(;adza ■ • 

class 6 

•granary • 

Outside of the affricates and borrowed words, both of which we will discuss 
below, there are no cases of 

c) class S 


class 6 


alternation, and there are no cases of 
d) class $ /^ class 6 

lass 5 '^ class 

[^vc] Oc] 

alternations in the language whatsoever. 

As such, the voicing distinction between stops has been virtually 
neutralized in root-initial position in class 5« To take the tack 
L+vc];7[.vcJ in class 6 would necessitate an explanation not only of why 
there are no voiceless stop-initial class 5 words, but also an explan- 
ation of exactly in what environment a voiced stop devoices, which is 
impossible to state even when taking tonology into consideration (i.e., 
would require marking each item which undergoes this devoicing in the 

Some linguists may claim, however, that admission into class 5 is 
dependent upon a root-initial voiced consonant. This assumption is 
inherently claiming that there is a devoicing rule for class 6 (exactly 
the opposite of our proposal - a voicing rule for class 5). In view of 
the fact that a devoicing environment for class 6 is impossible to state, 
a linguist making such a claim would have to say that all root-inital 
stops that do not devoice in class 6 are exceptions. However, in view 
of the fact that an initial voiced consonant is not a valid requirement for 
inclusion, into class 5, since root-initial voiceless fricatives occur in 
abundance^, and in view of the fact that our analysis (whereby a voiceless 
stop voices in class $) is exceptionless in that all stops voice in the 
following environment, 

(9) /# 

the [-vcj-^l+vc] directionality does indeed seem preferable. (This voicing 
alternation only occurs in root-initial position, as we have seen in the 
data thus far and will see in greater detail later.) Thus, in the view of 
simplicity and the lack of convincing evidence to the contrary, we 
tentatively have 

(10) C ^[:*vcJ/# 

[-VC J 

We have implicitly assumed up to this point in the discussion that the 
voiced/voiceless alternation only occurs in classes $ and 6. This is not 
entirely true, but as we shall see below, all those voicing alternations 
outside of classes 5 and 6 may be accounted for by a separate rule, 5 
and 6 being the only place where our proposed non-phonemic alternation 
takes place. 

The next question is how many mornhological features we need to include 
in our rule, i.e., I+5j» K+N T] , C-Vj , etc. Since this particular 
. . altamaioo only takes place in class $ nouns and adjectives, 

it seems that in the strictest senae both the feature \*^~\ and (+N 



might not be needed. In the 21 noun classes of Shona, all 21 agreement 
series have overt prefixes, and in the noun/adjective series, all classes 
except 5 and 9 have overt prefixes, although it appears that some sort of 
synchronic nasal may be posited for class 9.^ Aa such, the only place 
in which we have word-initial non-prefix consonants of either series of 
noun classes is the class 5 series of nouns and adjectives. Therefore, 
to state both [+$] and fC+N ^~] would be redundant, but to state one or the 

other is necessary, because outside of the noun class series, this rule is 
not productive. The following data will show that the root-initial stop 
of verbsj^ even when in word-initial position (this only happens in the 
imperative), never voices. 

(11) infinitive imperative 

kutenga ter)ga 'buy' 

kutarisa tarisa 'look' 

kupa ipa ' give ' 

kukura kura ' grow ' 

ku'J^e^geta ^engeta 'save' 

kupfugama pfugama 'kneel' 

The voicing altemati<»i does seem to serve a function of sorts in 
class 5» although it is not as pervasive as one might hope if it were a 
purely morphological function, as other scholars would have it. So we 
now have 

(12) C > [>vc]/# 

-vc / 

^5 J 

Thus far, our voicing rule has been formulated so as to affect only 
word-inital voiceless stops in class 5 nouns and adjectives, since we have 
already seen that verbs are not affected by our rule. It must now be 
noted that there are tiro different types of adjectives, and the following 
may shed more light chi the subject. (In the predicate adjectives, as we 
have elected to call them, #r- in class 6 is an agreement marker; -a- is 
a relative marker, and -ka- is some kind of tense marker.) 

(13) the r«d book buku dzuku 

the book is red buku rakatsuku 

the red books mabuku matsuku 

the books are red mabuku akatsuku 

the white book buku ^ena 

the book is white buku rakacena 

the white books mabuku macena 

the books are white mabuku akacena 

an agreeable voice 

the voice is agreeable '^ izwi rakatendeka 

the voice that is agreeabli 



agreeable voices "^ 

the voices are agreeable < raazwi akatendeka 

the voices that are agreeable ) 

the broken bell y ^ 

the bell is broken ^ Jare rakaputsika 

the bell that is broken 

the broken bells j 

the bells are broken '< ma tare akaputsika 

the bells that are broken J 

the sharp hoe ) 

the hoe is sharp I Oadza rakapinza 

the hoe that is sharp ) 

the sharp hoes 

the hoes are sharp ^ mapadza akapinza 

the hoes that are sharp 

In discussing this data, we will refer to adjectives such as - cena 
•white' as "real adjectives," and those such as tendeka, 'agreeable,' 
as "predicate adjectives." (Although they may not function as predicate 
adjectives do in English, this term will be adopted from Heny, 1972.) 

Our class 5 morphophonemic rule as formulated has no trouble explain- 
ing such forms. However, one might also wish to account for why some 
adjectives can never appear as real adjectives, but only as predicate 
adjectives. If one could argue that adjectives which occur only as 
predicate adjectives were underlyingly verbs, then this distribution 
could be accounted for, and the lack of stem-initial voicing of stops 
would be doubly explained: first, because the stem- initial stop of a 
predicate adjective is never in word-initial position, and second because 
the initial stop of a verb is never affected by our class $ morpho- 
phonemic rule. 

While there appears to be sufficient evidence to argue for the 
latter solution, this is not within the scope of our paper and we will not 
present the details of the argument. Were such an argviment to be accepted, 
hcwever, we would have more choice in formulating our rule. Instead of 
specifying [+5j, we could choose to note [-Vj. 

We now come to the question of implosives in our analysis. It 
appears that there a?e only two implosive consonants in Shona , cT* and £ . 
There are thus no non-stop, no velar, and no voiceless implosives that 
appear in the phonetic representation of the language. 

In classes 5 and 6, these voiced implosives alternate either with 
voiced implosives, e.g. , 

(lit) class 5 class 6 

Buri ma&uri 'hole' 

cTutu mac^^tu 'strong wind' 


or with voiceless non-implosives. 

(15) class $ class 6 

bLpito mapipito 

cTura matura 


In the cases where voiced implosives alternate with voiceless non- 
implosives we are faced with five basic alternatives to account for the 

(l6) a) [voiced ~1 _, voiceless 1 ^ [voiceless 

[implosivej [implosivej [non-implosivej in class 6 

e.g.* . 

b) fvoiced ~1 

Lnon -implos i ve 

in class 6 


c) jvoiceless 
[implosive J 



in class 5 



d) pvoiceless ~^ 



voiced I 

[ji on -implosivej Lnon-implosivej ' ^implosivej 

in class 5 



[n on -implosive 

[implosive . 

in class $ 


Alternatives a and b are ruled out becuase, as we have argued before, 
to devoice in class 6 rather than voice in class 5 puts an added weight on 
the granitnar of explaining why all class 5 stop-initial words begin with 
a voiced segment and under just what circumstances the stops devoice, 
everything being much more predictable under our analysis of voicing in 
class 5. Alternatives a and c, whereby we come about the alternation via 
a step including a voiceless implosive are a priori infeasible and ruled 
out because there are no such phonetic entitle* as voiceless implosives, 
and we prefer not to have any level in a derivation containing an un- 
pronounceable item. Since we can avoid the postulation of so abstract a 
phoneme as a voiceless implosive, we have chosen to do so. This is not, 
however, to say that we are totally against the use of abstract entities 


in a solution. 

Of the two remaining alternatives under our analysis j 

d) p: 

e) p 

it appears that d would be the most attractive, in that the voicing and 
implosivizing are broken up into two clear steps rather than voiceless 
non-implosives becoming voiced implosives in one fell swoop. However, 
if one were to simply state that all instances of non-implosive /b/ and 
/d/ becOTie implosive, i.e. something like 




as a general clean-up rule for. 





we appear to run into trouble with borrowed words where A>/ and /d/ are 
never implosivized.7 


class 5 

class 6 




It thus appears that the proposed ioplosivizing rule does not apply 
to the entire lexicon of the language, but rather only to native Shona 
vocabulary. We are not splitting the vocabulary in half just at whim, 
however. The native Shcna lexicon also differs from the borrowed vocab- 
ulary in that borrowed words never undergo our proposed morphophonemic 
voicing mle, e.g.. 


class g 
pur is a 

class 6 


Thus it appears that not only our morphophonemic voicing rule, but also 
the implosivizing rule, must be "marked" as applying only to native Shona 
vocabulai7. In other words, we are saying that at least some rules must 
be sensitiTB to a [t native] feature. Our rules now look like the 







(20) C - (*impl] 

+3 top ' 


-delayed release 

(Note that none of the other analyses of voicing in class 5 make any c 
about the state of implosives whatsoever.) 

We also might, stLthough it is not integral to do so, posit an under- 
lying implosive 3 for native Shona, even though it is abstract in that it 
never shows up on the surface. An advantage and reason for doing this is 
to give more regularity to the underlying phonemic system in respect to 
the set of in^jlosives. That is to say, under such an analysis all native 
Shona voiced stops would be underlying implosives rather than just the 
non-velar ones. As such, the final version of our implosivizing rule 
would be. 


C ^ 1+impll 

+stop ~] 

with a general mile that de-implosivizes all velar implosives, 


l+impl j 


Thus far, we have given the impression that our native Shona morpho- 
phcwieraic voicing rule as it stands is exceptionless. Unfortunately, this 
is not the case, because we have forms like the following. 


class $ 



class 6 



where we would expect voicing in class 5 but do not get it. 

In the cases of the affricates of class 5 we seem to be encroaching 
upoi another morphophonemic process of sorts whereby forms like the 
following are acceptable. 


class $ 



'big well' 
'big spear' 

A voicing of the stops is permissible to give the noun the added dimension 
of ugliness, bigness, clumsiness, or some pejorative quality of this 
nature, although the extent of this process is not yet clear to us. And 

this happens not only in classes 5 and 6, but also, for example, in 
class 21. 

(2$) class 21 

kranana 'child' 

zigomana 'big child' 

If these initial segments voiced in class 5, this would lead to ambiguity; 
that is, dzime could mean either 'well,' or 'big well,' It appears that 
this process, whatever the full ramifications of it are, is not reversible 
as such in that the devoicing of a regularly voiced stop does not give 
us a less ugly, large, or clumsy connotation, and in fact gives us 
ungrammatical forms. 

(26) bveni / mapfeni 
but, «pfeni 

bvupa / mapfupa 
but, »pfupa 

An amazing fact about all of this is that only pf and ts are ever 
exceptions to our morphophonemic rule of voicing in class $, For this 
reasoi, these and only these phonemes maj be overtly affected by the 
pe JorativizatL CXI rule. That is to say, /p/, /t/, /k/, /8f/ are never 
overtly affected by this pejorativization rule, as they are always 
voiced in class $ by the morphophonemic rule. 

One analysis that could bring some sort of order to this chaos 
involving the class 5 and 6 affricates ta and pf la t« elaiji that those 
root-initial affricates that show the voicing alternation and are normally 
voiced in class 5 are also inherently big, ugly, clumsy, or whatever; 
those that are inherently voiceless are not and only acquire voicing when 
a special quality of largeness or the like is attributed to them. Under 
such an analysis, it is proposed that our morphophonemic voicing rule never 
cones into play with ts and pf . 

Such facts aeem to be borne out to some extent, becavuse we have 

(27) class $ 

where the affricate T0lea8,as opposed to class 5 pfumo 'sp>ear', tsime 
•footprint', pfuro 'pond', where the affricate does not voice. That is 
to say, to call a baboon ugly is to carry coals to Newcastle and the pf 
of pfeni thus is naturally voiced, whereas semantic forms like 'pond' 
and 'footprint' are more innocuous and only acquire such pejorative meanings 

class 6 












In claims whereby the affricates ts and pf are never affected by our 
rule of morphophonemic voicing, one must block it from applying in some 
fashion. The simplest manner would seem to be as follows. 

(28) /# 

In such a claim, one might be imply^Lng that ts and pf consist of two 
different phonemes each, whereas /J/ and /c/ are single phonemes and under- 
go the rule as if they were just one phoneme. However, to make such a claim 
that ts and pf never undergo our morphophonemic voicing rule, along with 
the claims, implications, and necessary finaglings needed to show that there 
is a semantic division in the lexicon corresponding to the division between 
those ts and pf that voice in class 5 and those that do not, is unsupportable 
for several reasons, among which are the following. 

(29) a) There is no independent evidence to support the claim that 
te, dz, 
c and 

te, dz, pf , and bv are of differend phaieme status frcm 

b) Tlisre is no natural phcmological reason why the condit- 
ioning environment for our voicing rule should be /# ^V 

instead of merely /# . 

c) The semantic differentiation between "normally" voiced 
affricate -initial words which alternate with a voiceless 
affricate is not seen as cold hard fact but only as a 
tendency of sorts oa the basis of the limited forms 
available . 

As such, we are unfortunately left to mark those affricates ts and 
pf that, in a handful of forms, do not voice root-initially in class $,as 
exceptions to our morphophonemic irule and seek perhaps an historical 
explanation for the peculiar phenomena surrounding these two phonetic 

Our discussion, and the formulation of our rule, have led the 
reader to believe that the voicing of root-initial stops in class 5 
occurs caily when those stops are also word-initial. There appear to be 
two exceptions to this, however, and examples of each follow. 

(30) claas $ class 6 

^anga mapar)ga ' knife • 

zibarjga mazipa^ga 'huge knife' 

danga matanga 'corral' 

zif'anga mazitar)ga 'huge corral' 

bveni mapfeni 'baboon' 

zibveni mazipfeni 'huge baboon' 

bvupa mapfupja 'bone' 

zlbvupa Bazipfupa 'huge bone' 


ganda ibvawi wakanda mapfawi 'the skin is tender' 

3faya ibenzi majaya mapenzi 'the young man is 


There are several possible explanations for both of these apparent 
counter-exanqples to our claim, and we shall examine each alternative in 
turn, eventually choosing one as being more probable than the others. 

The first and most obvious way of accounting for the above forms is 
to say that the affixes #i- and -zi- are exceptions to our proposed rule 
in that they allow voicing of a root-initial stop that is not also vord- 
initial. According to this solution, then, these prefixes would have to 
be marked as being able to optioiuilly intervene, i.e.. 



While this analysis works —there are, after all, no other affixes which 
allow root-initial stop voicing in class 5— we will eventually discard 
marking them as exceptions in favor of another solution. 

In examining the above forms which contain #i-, we see two 
possible explanations for the data other than marking #i- as an exception. 
The first possibility is to claim that our rule may be altered so that 
root-initial voicing could take place even if that root-initial stop were 
not word-initial, but were preceded by an optional vowel. Under such a 
solution, our rule would look like the following. 

(32) C > [♦▼cjMV) 



We do not accept this reformulation, however, for it appears to be an 

ad hoc way of accounting for the data and has virtually no natural phonological 

explanation. Since there seem to be no other prefixes which consist of 

a single vowel, this change in our proposed rule cannot be supported 

by any other forms and is merely a way of "doctoring it up," 

Ihe third possible explanation for why root-initial stops voice even 
when they are preceded by the prefix #i- is the analysis we will accept 
as being preferable to the others. Notioe that when #i- is prefixed onto 
a noun, the result is always a sentential element. We will argue, there- 
fore, that the #i- which is prefixed onto these nouns is not to be 
considered to be a prefix in the generally -accepted sense of that word, but 
that the element should be analyzed as a word. In this manner, the under- 
lying form for i^adza would appear as #i#padza and not as #i-»padza . 
Our rule of voicing stops in word-initial position in class 5 would then not 
be blocked from applying, and will produce the correct surface form, i£adza . 


Earlier in the paper proof was offered of a rule that voices initial 
stops to change the lasaning of a word (phonemic as opposed to our non- 
phonemic Tolcing). In such cases, the voiced alternant had the added 
meaning 'big, huge, ugly' or sosnething similar. Now we see that there 
is an infix, -zi-, which also means 'big' when affixed to a noun, and 
appears to be an exception to our voicing rxile, in that in class 5, the 
stem-initial stop voices even though it is not in word-initial position. 
Our explanation of this unexpected voicing of class $ root-initial stops 
is the following: Since stem-initial voiced stops are obligatory when 
-zi- is added to forms that would otherwise be affected by our morpho- 
phcaiemic voicing rule, but voiceless stem-initial stops are obligatory 
when -zi- is added to forms not normally affected by our rule, the 
simplest method of accounting for the facts is to posit the insertion 
of -zi- 'big' at a synchronic level later than that of the noun class 
marker; otherwise, we would have no practical means of explaining why 
stops voice after #zi- (class 5) but not after #raa-zi- (class 6), Below 
is a sample derivation. 


class marker 
our rule 

class 5 




class 6 




It has been shown that the two apparent counter-examples to our 
proposed rule for class $, 






are really not counter -exan^iles at all. The sentence -forming element 
#i- is probably a word in and of itself, and therefore the consonant 
iiranediately following it is word -initial and subject to our morpho- 
phonemic rule voicing initial stops in class $. The infix -zi-, meaning 
•big, huge', can co-occur with root-initial voiced stops because it is 
introduced aftero ur rule of morphophonemic voicing in class $ has applied. 

We can offer yet another piece of evidence as to the validity of 
our claim that word-initial position is the only environment in which a 
stem-initial stop of a class 5 noun or adjective is subject to our 
voicing rule. There is a small class of nouns ( pei^o 'insane', tendi 
'believing', dzidzi 'student', fuai 'rich', ronbo -^oor', fara 'sickly') 
that, when used to modify a class 5 noun, show a #mu- prefix and the stem- 
initial stop is not affected by our class $ voicing rule. 

(3U) class 5 

Jay a mipepgo 
jaya mutendi 
J ay a mudzidzi 

class 6 

ma Jaya mapen go 
majaya raat^di 
ma^aya madzidzi 

'insane young man' 
'believing young mi 
•student young man 


This prefix #inu- appears to be a marker of personification, even thongh 
non-human nouns can also appear in these constructions: 

(35) class $: ruwa mufara 

6: maruwa mafara 'sickly flower' 

class 9: nika mufara 

10: nika mifara 'sickly country' 

Note, however, that if we speak of a rich country, 

(36) class 9: nika fumi (*nika mufumi) 

10: kanika kafumi 

we are not personifying, and therefore we do not use #mu-. This is evidence 
that the #mu- of class 5 is not a "fixed" prefix for any certain adjectives. 

Ibtil now, all nouns in class $ have tseen affected by our morphophonemic 
rule of voicing, because their stem-initial stops also happened to be word- 
initial. Now we have evidence that nouns in class $ whose stem-initial stop 
is not word -initial are not affected by our voicing rule for class $ nouns 
and adjectives. Therefore, it is imperative that we include word- initial 
position in the formulation of the rule. 

To summarize, we have seen that our analysis of the voicing phenomencsi 
in class $ is by far preferable to the other analyses proposed to date, 
because we are able to account for the peculiarities surrounding the voicing 
in class 5 in a natural phonological way, something that has been impossible 
under the proposals of Fivaz, Heny, etc. That is to say, we have shown that 
the voicing phenomenon is due to a morphophcmemic process which is limited 
to word -initial stops in class 5, rather than to a purely morphological 
"prefixing" process, as the other analyses would have it. Within such a 
standpoint, a prefixing process, whatever the exact proposed ramifications 
are, fails to explain why this ••prefixing" only affects some root-initial 
ccHisonants some of the time and vowels not at all, whereas all other noun 
class prefixes of both series affect, or are present with, aU. root-initial 
segments all of the time. 

We have also delved to some extent into the status of implosives in 
the language, the existence of a "pejorativizing'^ root-initial segment 
voicing^rule, the necessity of allowing phonological rules to be sensitive 
to thel^ native/ feature in Shona, and the like. Above all, we have seen 
that there is no prefix in the class $ noun/adjective series, and that the 
voicing alternations that occur in classes 5 and 6 are due to the following 
morphophonemic rule. 

(3) C _^ 7 [+vc]/#. 





This paper was presented at the Ninth Annual African Linguistics 
Conference held in East Lansing, Michigan, in 1978. We would like to thank 
our language consultant, Kokerai Rugara, for his patience and help, and 
Dr. Charles Kisseberth for his comments on an earlier version of this paper, 
Apecial thanks go to Lee A. Becker, for the time he gave us in revising our 
earlier paper. 


^6' represent implosires, b, d, etc., represent non-implosives. 

Since tones in Shona have no bearing on our analysis, we will not mark 
our exaii?>les for tone, 


Note that although there is a tendency for borrowed words with root- 
initial voiced stops to enter into classes 5 and 6, but those with voiceless 
stops to fall into class 9, 

class 5/6 

damu/ madarau 'dam' 

darama^i/ madarama^i ' dynamite • 

buku/ mabuku 'book' 

biza/ mabiza 'horse' 

bara/ mabara 'wheel baiTOw' 

juzi/ majuzi 'jersey' 

class 9 

tafura 'table' 

tii 'tea' 

kati 'coffee' 

pen 'pen' 

kapu 'cup' 

I this is not always the case, e.g., 

class 5/6 

pur is a/ mapurisa 'police • 

peni/ mapeni 'penny' 

Since this tendency could be looked upon as being either the result of only 
voiced stop-initial words going into class 5, OR as the result of analogy to 
the end product of our rule of morphophonemic voicing, this is a moot point. 

The class 9 "prefix" comes from /ni/ historically* 

This i- is prefixed onto mono-syllabic words. 



All cases of A)/ and /d/ in native Shona vocabulary are phonetically 
realized as implosives {{JcP), 

It would be intei^sting, of course, if the historical explanaticn has 
the pejorative connotation rule come into play. 


Biehler, E, 1950. A Shona dictionary. Cape Town: Longmans, Green and 

Co., Ltd, 
Chomsky, Noam, and Halle, Morris. 1968, The sound pattern of English. 

New York: Harper and Row, Publisherf. 
Doke, Clement M. 1931, A comparative study in Shona phonetics, Johannesburg: 

University of the Wirwarsrand. 
Fivaz, Derek . 1970« Shcaia morphophonemics and morphosyntax. Johannesburg: 

University of the Wirwarsrand. 
Fivaz, Derek, and Ratzlaff, Jeanette, 1969. Shona language lesscMS. 

Salisbury: Word of Life Publications, 
Fortune, G, 19$5. An analytical grammar of Shona. London: Longmans, Green 

and Co, 
Gregersen, Eagar A. 197^, Prefix and pronoun in Bantu, International Journal 

of American Linguistics 33:3 Part 2 (supplement), 1-69, 
Hannan, M. (compiler), 197U. Standard Shcaia dictionary, second edition, 

Salisbury: Rhodesia Literature Bureau. 
Heny, Frank, 1972, Bantu lexical classes and semantic universals (with 

some remarks on how not to write phonological rules). Studies in 

African Linguistics 3.207-258, 
Welmers, William E, 1973. African language structxxres, Berkeley: 

Iftiiversity of California Press, 

studies in the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 


S. N. Sridhar 

It is well-known that over three millennia of language 
contact on the Indian subcontinent has resulted in the conver- 
gence of linguistic features of the four language families of 
the area. A major part of this ongoing process is the adap- 
tation of Indo-Aryan characteristics by the twenty-three 
or so Dravidian languages. The present paper is an attempt 
to study the various aspects of this phenomenon in broad 
outline. Following the introduction, the language contact 
situation between Sanskrit , Prakrit , and the modern Indo- 
Aryan languages on the one hand, and the literary and non- 
literary Dravidian languages on the other, is described. 
Next, variation in the nature and extent of Indo-Aryanization 
both within and across the Dravidian languages is outlined. 
Section four deals with the impact of Indo-Aryan on Dravidian 
phonology, morphology, lexicon and syntax; this is followed 
by a discussion of the processes of nativization of borrowed 
elements. Section six describes the sociolinguistic impli- 
cations of Indo-Aryanization, especially the emergence of 
the manipravala style , the reinforcement of caste dialects 
and diglossia, and language attitudes. The conclusion raises 
some issues concerning the role of Indo-Aryanization in the 
evolution of Dravidian languages. 

1. Introduction 

It is often remarked that the Indian subcontinent has a genius for 
assimilating foreign influences without losing its essential character.* 
Nowhere is this resilience demonstrated more clearly than in the way 
the Dravidian languages have absorbed what, by any count, must be 
regarded as massive Indo-Aryan influence and yet retained their essential 
Dravidian character .-'- 

In discussing the Indo-Aryanization of Dravidian languages, we are 
talking about an ongoing phenomenon that must have started about 3,500 
years ago. The presence of linguistic features of possible Dravidian 
origin in the Rig Veda (e.g., retro flex consonants) suggests Aryan 
contact with the Dravidian-speaking people as early as about 1500 B.C. 
This possibility is confirmed by the geographical distribution of the 
Dravidian languages — e.g., Brahui in Balochistan, Gondi in Madhya Pradesh. 
Malto in Bihar-West Bengal, among others. Ignoring the possibility 
of reverse-migration, this scatter suggests the presence of Dravidian- 
speaking people over most of India at one time, and their gradual 
recession before the advancing Aryans. However, our recorded history of 
Dravidian contact with Indo-Aryan goes Tsack only to the early centuries 

ITie present topic may be — and has been-studied from many points of 
view. 3 In the past, the main concentration has been the phonetic and 


semantic modifications undergone by Indo-Aryan words in Dravidian 
languages. However, there is much more to the topic than that. In the 
present paper, my aim is to present an overall picture of the many 
facets of Indo-Aryanization of Dravidian. Following the introduction, 
section 2 discusses the reasons, sources and mechanisms of Indo- 
Aryan borrowings; section 3 outlines the variation in the extent of 
Indo-Aryanization; section U is a survey of the Indo-Aryan influence 
on Dravidian on various levels; section 5 is devoted to nativization; 
section 6 discusses the sociolinguistic implication of Indo-Aryanization, 
and section 7 comprises the conclusion. 

Before proceeding further, I would like to point out that what 
follows is more in the nature of a bird's eye view than a microscopic 
examination of minutiae. Due to the paucity of materials and my own 
inexperience, I have not been able to do equal justice to all the 
languages involved. Yet, I hope the following discussion may serve 
at least as a programmatic sketch for a much needed full-length 
treatment of this important topic. 


The reasons for the prodigious influx of Indo-Aryan features into 
Dravidian are fairly well-known. First, Sanskrit has traditionally 
occupied an exalted position as the language of the elite in India, a 
position analogous to that of Latin in Europe until a few centuries ago. 
Sanskrit was not only the language of the scriptures, the epics and 
classical literature, it was also the repository of a vast body of 
philosophical, scientific and technical literature in every field. 
It enjoyed royal patronage, and the prevalence of Sanskrit-based 
education largely under the control of the Brahmans reinforced its 
prestige. All this led to the creation of a vast body of literature 
in the Dravidian languages, composed in conscious imitation of the Sanskrit 
models , bringing in its wake a veritable flood of Sanskrit words . 
While many of these words were introduced to meet the technical needs of 
specialized language-types (or "registers", for example, in grammar) 
and often to suit the demands of intricate versification (for example, 
adiprasa and antyaprasa) , a large number must have been introduced 
simply because it was prestigious to do so. 

Secondly, Prakrit (and to a lesser extent, Pali) played a crucial 
role in the Indo-Aryanization process. In some languages (e.g., Tamil), 
the words that came in via Prakrit far outnumber those borrowed directly 
from Sanskrit. There are two major reasons for this Prakrit influence. 
It was the court language during some periods in the Dravidian South, 
for example, during the Satavahana and the early Pallava periods. Also, 
the South, especially Karanataka, was the haven of refuge for the Jainas 
and a large number of very prominent authors in the Dravidian languages 
were Jainas who often composed in Prakrit as well (e.g., Nemichandra) . 
Many prominent Prakrit authors also wrote in Dravidian language^, for 
example, Pushpadanta, Trivikrama, and Sakat"ayana, who wrote in Kannada. 
Many of Indo-Aryan loan-words in the non-literary languages are also 
traceable to Prakrit forms. 

The next major category of Indo-Aryan loanwords in Dravidian comes 

from modern Indo-Aryan, especially Hindi -Urdu and Marathi in the 
literary languages of the South, Oriya in Pengo and Kui , Sindhi in 
Brahui , Marathi 'In Kolami , Hindi in Kurukh, etc. 

Parts of South India cqjne under Muslim rule in various periods , 
especially during the reign of the Baharaani kings in Andhra and North 
Karnataka, and that of Hyder Ali and Tippu Sultan in South Karnataka. 
D\iring the latter period, Persian was the official language of the 
princely state of Mysore, with official records being kept in that 
language. The prolonged Muslim rule has contributed a large niamber 
of words in the fieldsof law, administration, and land-management, as 
well as those relating to Muslim culture (see Sridhar, 1975). Finally, 
the increased commerce between the North and South since the beginning 
of the Independence movement has resulted in the introduction of more 
Hindi words (e.g., kalapanT , gh^Vao, bandh, ^arkha , etc.). 

As for the influence of Marathi, both geography and politics 
have been contributing factors . Maharashtra shares borders with both 
Karnataka and Andhra Pradesh, and there is intensive code-mixing in 
these border areas ( see, Gumperz and Wilson, 1971; Upadhyaya, 1971). 
Politically, much of North Karnataka came under the Bombay Presidency 
until the reorganization of the states in 1956, and the administrative 
language as well as the language of education in this region was 
Marathi. Further^ South, the Tanjavur area in Tamil Nadu was iinder 
Maratha rule for nearly two centuries \intil l800 A.D. (Subramonian and 
Ganesha Sundaram, 195*+). The Desastha Madhwa brahmins in the South 
have traditionally been bilingual in Marathi and Kannada, Telugu or 

The non-literary languages of the South (Toda, Kota, Kodagu, 
Badaga, etc.) have also numerous Indo-Aryan borrowings in them, but 
these seem to have resulted from diffusion from their literary neighbors 
(Emeneau and Burrow, 1962). As for the non-literary languages outside 
the major Dravidian belt, their language contact with Indo-Aryan needs 
no explanation. Speakers of these languages often indulge in such 
occupations as wood-gathering, basket-weaving, and perform menial jobs, 
and in general are crucially dependent on their Indo-Aryan neighbors 
for economic survival. Bilingualism in the dominant language of the 
region is for them a cardinal necessity. Under these circumstances, 
it comes as no surprise that some of these languages show much more 
radical changes in response to Indo-Aryan influence than their literary 

The preceding discussion assumes, following the long recognized 
theory, that the main agency responsible for the transfer of traits in 
a language-contact situation is the bilingual individual. Several 
recentsociolinguistic studies of language use in bilingual communities 
(e.g., Gumperz and Wilson, 1971, and Upadhyaya, 1971, in the Maharashtra- 
Karnataka border) give us an insight into the processes by which the 
present day convergence of features in Indian language must have come 
about. In particular, Kachru (1978) argues that code-mixing may be a 
crucial factor in the kind of language change we have been discussing. 



The Dravidian languages show a great deal of inter-language and 
intra-language variation in the extent of Indo-Aryan influence. The 
non-literary languages spoken in the Indo-Aryan region seem to have 
undergone a greater amoiint of structural (morphological, syntactic) 
influence (see section U.3), whereas the literary languages of the 
South have been influenced predominently on the lexical level. Second, 
the non-literary languages contain few unassimilated Sanskrit lexical 
items while this category plays an important role in the lexicons of 
the literary languages. Third, it is suggested (Andronov, 196h) that 
the dialectal dissolution of Dravidian was already well under way before 
Indo-Aryan began to exert its influence. If this is correct, ^ then 
that fact could account for some of the variation. Finally, the extent 
to which a given language has been influenced by Indo-Aryan also 
depends on the language attitudes of its speakers. For example, Tamil 
has traditionally been more conservative than its neighbors in this 
regard — a fact commonly attributed to the fierce loyalty of the Tamils 
to their language and their desire to preserve its "purity"' (see 
section 6.3 for a detailed discussion). In contrast with this, consider 
Bhattacharya's observation that speakers of Ollari (a non-literary 
Dravidian language spoken in the Koraput district of Orissa ) disclaim 
proficiency in their mother tongue (Bhattacharya, 1957). 

There is also a great deal of intra-language variation with regard 
to the extent of Indo-Aryan influence. In this respect, the chief 
variables involved are (i) style, (ii) caste, and (iii) topic of 

Dravidian languages show more Indo-Aryan features in written style 
than in speech, and within the written style, more in poetry than in 
prose. It is well known that during the various periods in the history 
of Dravidian literature, a style of writing characterized by heavy 
Sanskrit borrowings, often together with the original Sanskrit declensions 
and long compounds (e.g., Te. dharmanivahanodyogamummanaku ) became 
fashionable (see Kunjunni Raja, 1972). This style, commonly referred 
to as Manipravala (combination of gems and corals) reached its peak 
in Malayalam (e.g., Chandrotsava ) , but it is met with in the other 
three literary languages as well (for example, the inscriptions of 
Raja Raja Chola and in the Jaina and Vaisnava philosophical literature 
in Tajnil, in the work of Nannaya and Potana in Telugu, and to give a 
modern example, Kuvempu's Ramayapadarsana in Kannada) . Like the epic 
sytle of Milton, Manipravala never found universal favor and is largely 
a literary and linguistic curiosity now, being used marginally in 
orthodox Brahman doctrinaire tracts, if at all. 

The extent of Indo-Aryan features also varies according to the caste 
of the speaker (writer). Bright (1966), Bright and Ramanujan (196H), 
and Ramanujan (1968), among others, have observed that one of the 
variables distinguishing the Brahman dialects5 in Tamil and Kannada 
from the non-Brahman is the presence of Sanskrit phonemes and clusters 
and vocabulary. Here we see what the anthropologist Srinivas calls 
'Sanskritization' (Srinivas, 1966) operating in a literal sense, the 
Sanskrit features serving as a symbol of caste identity and distinctness. 


The emergence of hypercorrect forms with unnecessary aspiration may 
also be regarded as a manifestation of this phenomenon. 

The third parameter of variation is the topic of discoxirse: Indo- 
Aryan features are more likely to be present in the discussion of some 
topics than in others. For instance, religious discourses, in 
particular, explication of scriptures, known as upayasas in the South, 
contain heavy Sanskrit borrowings, as do texts and oral discourses on 
poetics, aesthetics and philosophy, as well as those on the social 
sciences in general. By the same token, one finds extensive use of 
Hindi-Urdu words (ultimately of Perso-Arabic origin) in discussions of 
law, land-revenue, Hindustani (or North Indian) music, wrestling, 
etc. (for a detailed discussion of this register-oriented code- 
mixing, see Sridhar, 1975 and 1978). In modern times, Sanskrit has 
been serving as the principal source of new coinings in all the 
scientific and technical fields. (Tamii is, once again, an exception 
to this statement, the emphasis there being on the exploitation of 
native resources.) 

The foregoing discussion of variation in the extent of Indo- 
Aryanization should suffice to make clear that any facile generalization 
about Indo-Aryanization of Dravidian languages, and especially the 
non-literary ones, is likely to prove wrong. Here, one can only note 
some general tendencies, always keeping in mind the cowplex inter- 
and intra-language variation. With this caveat, let us now turn to 
an examination of the influence of Indo-Aryan on various levels. 


U.l Vocabulary 

Anyone who looks at the dictionaries of Dravidian languages cannot 
but be struck by the extravagant proportion of Indo-Aryan words in 
general, and Sanskrit words in particular, contained in them. Emeneau 
and Burrow (1962) rightly attribute this to ''the tendency for all four 
of the Dravidian literary languages in the South to make literary 
use of the total Sanskrit lexicon indiscriminately (p.l)." Even 
in the colloquial language, so many words of Indo-Aryan origin have found 
their way and been assimilated that it is hard to speak even a few 
sentences without using some of these words. It was no doubt partly 
because of this high proportion of Sanskritic elements (and the 
similarity in the sound values of the graphemes in the scripts Ccoxmting 
the grantha script in Tamil]) that, until just over a century ago, it 
was generally believed that the Dravidian languages were descended 
from Sanskrit. Even Western scholars like Pope, Colebrooke, Carey, 
and Wilkins s\abscribed to this view. Caldwell (1903) spends a 
considerable part of his classic work dispelling this misconception. 

It is impossible to give a list of borrowings in Dravidian (for 
a partial list, 5 see Emeneau and Burrow, 1962) or even to specify the 
conceptual fields they denote, for there is no field of activity which 
has not been influenced by them. In many cases, the native terms have 
been replaced by the borrowed items , even in the areas generally 
regarded as part of the basic vocabulary. A case in point £s the 
set of direction terms in Kannad©, where the native words mu^al, 
paguval , ba^agal, and tenkal have been replaced by the Sanskrit 


purva , pascima , uttara, dakgir?a ,even in everyday colloquial use. 
Sjoberg and Sjoberg (1956) estimate that about twenty percent of the 
"non-cultural" part of the basic vocabulary in literary Dravidian now 
consists of loan-words from Indo-Aryan, creating problems for 
glottochronology (for a detailed discussion, see Sjoberg and Sjoberg, 
1956, and Sreekantaiya, 1956). The proportion of borrowings in those 
areas that are more prone to cultural influence is, obviously, much 
higher, probably as much as fifty to sixty percent. Among the non- 
literary languages, Emeneau ( 1962a: 20) estimates that the vocabulary 
of Kolami (Wardha dialect) contains approximately thirty-five 
percent of words with Marathi etymologies. The case of other non- 
literary languages is perhaps not much different. 

Certain salient features of the Indo-Aryan loan-words in Dravidian 
may be noted. First, the loans are not restricted to any one register 
or set of registers, they are everywhere — in basic as well as tech- 
nical vocabulary, in the written and the spoken language, in the language 
of intimate conversations and personal letters as well as that of news- 
papers and public speeches. Second, the loans are not restricted to 
content words alone, but include a large number of function words such 
as ouantifiers ( kevala , baha^a , alpa ) ,intensifiers ( atyanta , matra ) . 
adverbial conjunctions ( akasmat , bahu/ah ) , etc., as well as bound 
morphemes of various kinds (see next section). Third, a large number of 
Sanskrit words have been borrowed both in their root form (e.g., jiva ) 
and in their derived forms (e.g., jivanta , saktiyuta , aprakai;ita , 
taratamya , etc.). Fourth, the Dravidian languages, especially the 
South Dravidian languages, have a substantial number of Indo- 
Aryan loan words in common (e.g., nadi 'river', candra 'moon', 
ratri 'night', bhumi 'earth', etc.) and these words often undergo 
similar phonetic and semantic changes in these languages (e.g., 
■anumana(m) (u) ' suspicion 'QSkt. id. 'inference, guess '^, see section 
i+.5 below). Fifth, all the literary languages have sets of two 
forms, one taken from Sanskrit and employed with minimal modifi cation, 
and the other either borrowed from Prakrit or modified in the borrowing 
language, or both. Often these pairs are distributed in a diglossic 
fashion (see discussion in section 6.2). Finally, the majority of 
Indo-Aryan loans undergo various kinds of phonetic and semantic 
changes, some of which are common to all the Dravidiaji languages and 
some language-specific (see section 5 for more details). 

^.2 Phonology 

This large-scale infiltration of Indo-Aryan words has resulted in 
the introduction of several phonemes alien to the Dravidian phonological 
structure. Among these, the most important are the aspirated stops 
(kh)Ch,th, th , ph , and their voiced counterparts) as in, for example, 
Ka. khara 'spicy hot', chatri 'umbrella', thaqe 'station', kathe 'story', 
phala 'fruit', and ghan^e 'bell', 'hour' jhari 'centipede', mudha 'fool', 
nidhana 'slow', bhara 'heavy'. Other new introductions are the 
distinction between s_, s_, and s_ as in sasegha 'yet to be finished', 
and the diphthong au as in gau^a 'village chief. The scripts of all 
the literary languages except Tamil now possess characters for the 
full complement of Sanskrit phonemes. Even in Tamil the grantha script 
accomodates these phonemes. 


The Indo-Aryan loanwords have also resulted in the loss or 
violation of some positional constraints in Dravidian. Thus, the 
presence of voiced consonants in the word-initial position as in 
gamaka , initial consonant clusters and non-homorganic plosive clusters 
as in pr'arabdha , word-final obstruents, as in fakat ("*H-U) , among 
others, may be attributed to Indo-Aryan influence. 

However, it is important to note that thesenon-native phonemes 
and consonant clusters are not found in all speakers and in all styles 
(see Sjoberg, 1962, and Fowler, 195^). As already noted, these features 
mark certain styles and their presence is determined by socio- 
linguistic variables (see section 6.1 for further discussion). 

The non-literary languages have also adopted many phonological 
features from Indo-Aryan. For example, Brahui and Kurukh have developed 
nasal vowels of the Indo-Aryan type. Brahui has also lost the short 
e_ and o_ characteristic of Dravidian (Emeneau, 1962). Similarly, Pengo 
(another non-literary Dravidian language spoken in Orissa) has lost 
the distinction between j_ and z_ that it inherited from Proto-Central 
Dravidian, presumably under the influence of Oriya, which lacks that 
distinction. Pengo has also developed the feature of aspiration, 
although it is confined mostly to loanwords from Oriya and used only 
sporadically even there (Burrow and Bhattacharya, 1970). 

it.2 Morphology 

On the morphological level, the most striking characteristic is 
the productive use of a large number of Indo-Aryan derivational affixes. 
Sanskrit suffixes such as - ant a , - igtha , - kara , are freely used (e.g., 
Ka. hanavanta 'rich', Te. k'opigthi 'irascible'. Ma. vasajkaram 
'imprecation'), as are the Hindi-Urdu ones like - dar (Ka. patt'edara 
'detective') - kh'or (te. lancakoru 'corrupt man'), - ya1 a (Ka. mTsevSl a 
'the moustached man'), and - giri (Ka. , Te. giimastagiri 'clerk-ship'). 

The employment of prefixation as a productive work-forming device 
in Dravidian must be attributed to the Indo-Aryan impact and is seen 
in the niimber of forms involving, for example, Sanskrit apa -, ava- , 
ati- , SU-, pari- , as well as the Hindi-Urdu prefixes, b"- , maj7^ , gair- , 
etc. Consider, for example, the following. Ka. apakTrti 'ignominy', 
Te. apa d~ru 'insinuation', Ka. , Te. apar'u'pa (m) 'rare', Ka. ati matu 
'garrulousness' , Ka. sulocana 'eyeglasses', Ka. pari vik^a^e 'survey', 
Ka. proJ:i obba 'each one', as well as Ka. , Te. b'eniyat(tu) 'disloyalty', 
gairuhajari 'absence', maji adhyak^a 'eK-president ' , etc. Most 
of these affixes freely collocate with native roots as well , although 
it is more common to attach them to other borrowed words and use 
them in a sense not attested in the source language (as in the case 
OS the Ka. patte'dara. ) 

In a different vein, Andronov (196U) has observed that the gradual 
loss of many negative-incorporated forms in the major Dravidian languages 
and the use of analytical modes of expressing negation by means of a 
positive verb followed by a negative word may be due to the influence 
of Indo-Aryan. Thus, in Keinnada, for example, forms such as hoga , 
mai^a have gone out of use , having been replaced by hogoci-illa and 


Turning to the non-literary languages, Bloch (195^) has pointed 
out that the infinitive suffix - na in Gondi is indubitably borrowed 
from Hindi-Urdu. Another infinitive suffix - le in the same language 
is attributed to the influence of Eastern Marathi . 

Pengo occasionally employs the Oriya genitive morpheme in -r_ 
as in rajar kuvar 'king's courtyard', and kumar ar 'potter's wife'. 
In Pengo, one also finds a marginal occurrence of the Oriya locative 
suffix -ne as in barhane 'in the rain'. More regular and widespread 
is the use of a number of Oriya post-positions, such as the instrumental 
hudun , san , 'with', the dative kajin 'for the sake of, and the locative 
bitre 'in^, tarento 'beneath' (Burrow and Bhattacharya, 1970, iil ff.). 

Emeneau ( 1962b) has brought to light an interesting case of 
morphological borrowing in Brahui , namely the allomorphs kar and 
kam ('to do, make') from Sindhi and Balochi , respectively, on the 
lines of the established Brahui irregular structure with -nm and -r_. 
Brahui has undergone several other significant morphological changes 
not attested elsewhere in Dravidian, but these changes, as Emeneau 
rightly suggests, are probably directly traceable to the influence 
of Balochi rather than of the Indo-Aryan Sindhi. 

U.3 Syntax 

The influence of Indo-Aryan is least pronounced on the level of 
syntax. Although a number of syntactic patterns modelled on those 
of Sanskrit were introduced by various authors from time to time,"^ 
by and large, few syntactic traits in contemporary Dravidian can 
be traced to Indo-Aryan influence. There are a few notable exceptions, 
however. Modes of compounding such as the avyayibliavg (where the 
first component carries the major semantic burden , e.g., yath'a 
sakti 'according to one's ability') seem to go against the general 
Dravidian tendency to place the determiner before the determined and 
are probably due to the influence of Sanskrit. Similarly, the passive 
construction, (for example, Ka. hu(^ugiyinda blgilu tereyalpa^ttitu 
'girl- instr .-door-to open (inf . )-experienced' "the door was opened 
by the girl"), seldom used as it is, is also likely to have developed 
under the influence of Sanskrit, a tendency further reinforced by 
the influence of English. The third major syntactic trait in 
Dravidian traceable to the impact of Indo-Aryan is the clausal mode 
of relative-clause formation, (described below), which is widely 
attested in all the literary Dravidian languages. 

As is well-known, the preferred mode of relative clause formation 
in Dravidian is by turning the main verb into a participle and deleting 
the coreferential noun, as in Ka. hasiru sire u'^ti^^uva- hengasu nanna 
hegdati 'green-saree-wearing-woman (is) my-wife'. In addition to 
this construction, we now find another construction, closely modelled 
on the Indo-Aryan, e.g., Ka. , yava hengasu hasiru sTre uij-^iddalo 
ava^u nanna heQ^ati , 'which-woman-green-saree-is-wearing+ indefinini- 
tizer-she (is) my-wife'. "The woman who is wearing the green saree is 
my wife." It may be noted that this construction presents one of 
the few cases in Dravidian where the main verb in a subordinate clause 
is not changed into a non-finite form, and a cftreferential subject 
is not deleted. Nadkarni (l9T0) convincingly argues for the Indo- 


Aryan origin of this construction. 

The non-literary languages in day-to-day contact vith Indo-Aryan 
show greater receptivity to syntactic influence. For example, Brahui 
has developed enclitic pronoxons which occur suffixed to verbs and 
nouns (Emeneau, 1962a:6U). Brahui has also lost the inherited Dravidian 
system of natural gender in the demonstrative pronoims. Both of 
these traits are, once again, probably due to Balochi influence. However, 
a third trait, the use of ki^ as a complementizer may be attributed as 
much to Sindhi as Balochi influence. Pengo has borrowed the morpheme 
ki from Oriya, which is used as an interrogative particle appended 
to the end of the sentence, as in venvatay ki? 'did you not hear?' 
Ki is also used in the sense of 'or'. Pengo also employs the Oriya 
conjunctive particles ar_, are ( 'and' ,' again' ) (Burrow and Bhattacharya, 
19T0:6l). Gondi has lost the relative participle typical of Dravidian 
and has adopted the Hindi relative prono\in (Caldwell, 1903:50). 

In concluding this section of the influence of Indo-Aryan on 
different levels, it must once again be emphasized that the prece ding 
characterization is by no means intended to be exhaustive. It is 
merely illustrative of the kind of influence Indo-Aryan has had on 


There has been a long tradition, in the native grammatical 
literature, of distinguishing two broad classes of borrowings from 
Indo-Aryan, namely tatsama ('same as') or words which undergo little 
(final vowel or consonant) or no modification, and tadbhava ('derived 
from' ) or words which have been nativized or assimilated. For 
example, Kannada has both the \inassimilated pak§i (<Skt. pakgin 
'bird') and the assimilated hakki . In this context, "assimilation" 
refers only to phonological modification and does not include semantic 
changes that the Indo-Aryan loans undergo in Dravidian. 

5.1 Phonetic Adaptations 

Some sound changes in borrowed words are common to all four 
major languages. For example, the doubling of the final obstruent 
followed by the addition of the enunciative -u (e.g., sampattu 'wealth', 
hukummu 'order', tejassu 'halo'), the substitution of unaspirated stops 
for aspirated ones (e.g., parike 'broomstick'[<Skt . parikriaj), the 
dropping of a final consonant (e.g., raja 'king'£<Skt. raj an^ ) , the 
substitution of s_ for £ and ^ (ka. hasu (<"Skt . pasu ' cattle '3 ), simplification 
of consonant clusters by anaptyxix (e.g.. Ma. kariya £<Skt . karya 'work'3), 
intervocalic voicing of voiceless plosives (Ta. vigadanf^ Skt . vika^a ) , 
and so on. Many of the changes are, however, language specific. 
In Kannada, for example, the final -a of loan words changes to -e_ 
(e.g., matS 3 >taate , ilakha :? ilakhe , etc.) while in Telugu, such a 
change occurs only marginally. Similarly, Malyaylam speakers substitute 
1^ and ^ for the Sanskrit t(d) and t(4) in certain contexts, e.g., 
when followed by a plosive ( talparyajn , tal^bhavam , vasalkaram , khalgam , etc.), 
whereas Tamil substitutes an r_ (Aiyar, 1973) . Similarly, the regular 
substitution in Tamil of voiceless consonants for the voiced ones in 


Sanskrit (e.g., cinkaram < Skt . sringara ) occurs much less frequently 
in other languages . 

By and large, most of the phonemic suhstitutions in Dravidian 
follow the regular phonological rules of the languages concerned. 
This adaptation, together with the fact that a majority of Indo- 
Aryan loans commonly used in Dravidian have come via Prakrit, makes 
some forms hardly recognizahle as originating from Sanskrit. A 
case in point is Ka. habba 'festival' (Skt. parva ) which comes from 
Pkt. pawa (or pabba ? ) and, with the mutation of v and b in Kannada, 
enters as pabba in old Kannada, and with the change of initial £ to 
h which took place around the 10th century, becomes habba in Middle 
and Modern Kannada. 

At this point, it is worth mentioning that this tendency to 
regularize loanwords to make them conform to the native phonological 
structure operates more strongly in Tamil than in the other Dravidian 
languages. As a result, Tamil has fewer unassimilated Sanskrit words 
than the others. At least two reasons may be advanced to account 
for this. One is the influential injunction in Tolkappiam (Sutras 
397,^01-2) according to which only those Sanskrit words could be 
used which either had letters common to Tamil and Sanskrit or had 
been modified in such a way that they could readily be represented 
in the Tamil script. This rule was scrupulously adhered to for over 
l600 years \mtil the great inflxix of Sanskrit words (including tatsamas ) 
diiring the Vijayanagar period (l6th century). The second, related reason 
may have been spelling pronunciation, the Tamil script lacking 
characters for voiced consonants and apirates , and palatal and 
retroflex sibilants. Note that the other literary languages have 
been more lax in this regard and have moved in the opposite direction, 
extending their phonemic (and graphemic) inventory to accomodate the 
foreign sounds . 

5.2 Semantic Changes 

Many of the loan-words from Indo-Aryan \indergo various kinds of 
meaning change when used in Dravidian. Some of these changes are 
common to South Dravidian as a whole, while others are innovations 
peculiar to one or two languages. The following table, based on 
Aiyar 193^-35, illustrates some common semantic developments. 

Source Word Meaning in Sanskrit Meaning in Dravidian 

anumanam inference; guess suspicion 

antahstah in the middle state story of a building; 

high rank, 
asahya unbearable loathsome 

avasth'a state miserable condition 

p'apam sin, evil sin, pity 


Both Aiyar (op.cit) and Eraeneau and Burrow (1962) contain many 
more examples of this phenomenon, traceable probably to diffusion 
within South Dravidian. Similar changes may be noticed in Hindi- 
Urdu borrowings as wall, e.g., Ka. Te., khayam (<id. 'stand erect') 
'permanent', dubara(i ) (<id. 'twice') 'excess , wastage ' . 

On the other hand, cases in which the same Sanskrit word has 
developed different meanings in the Dravidian languages are more 
frequent. For example, "arambham in Telugu has the meanings, 'pride' 
and 'killing' (literary), as well as the original Sanskrit meaning 
of 'beginning'. The same word in Kannada means 'cultivation' as well 
as 'beginning'. Similarly, 'amodamu in Teluga means 'dalliance' as 
well as ' f rangrance ' ; in Kannada, the word is used only in its Sanskrit 
meaning of 'enjoyment'; piramatam in Tamil means 'excessive; as 
well as 'excellence'; the derivative of the same source in Kannada, 
pramada has the meaning of 'blunder'. The Malayalara word gog^hi means 
'prank, contort ions, gesture', while in Kannada and Telugu, the same 
word has the original Sanskrit meaning of 'assembly'. 

The semantic changes that take place in Indo-Aryan words when 
used in Dravidian comprise of virtually every known type of deviation 
discussed in the linguistic literature. There is, for example, restriction 
of meaning in S. Dr. udyoga 'employment; (<Skt. 'work'); extension in 
Ka. anubhava 'mystical experience' (<Skt. ' '); speciali- 
zation in Ka. j angama 'a Virasaiva saint' (<Skt. 'mobile'); ame- 
lioration in Te. alaya 'temple' (-CSkt. 'house'); deterioration in 
Ka. adhvana , Te. adhvanamu , Ta. attuvanam (<Skt . adhvan 'path, way') 
'destruction, disarray', and so on. Similarly, Hindi-Urdu and 
Marathi words also change their meaning in various ways, e.g., Ka. 
cadi (<Ma. cahad ' ') 'petty complaint, informing on someone', 
Te., Ka. usf5.d (^-H-U ust'adi 'teacher') 'wrestler, gymnastics teacher', 
Ta. att'avanai (Ma. a^havana 'recollection') 'index, ledger', Te., 
Ka. sa^a ^-U §ara ' condition ' ) 'post script', etc. 


It was remarked earlier (section 3) that all the Dravidian 
languages are not uniformly Indo-Aryanized, nor are all varieties in 
a given language influenced to the same degree. This variation is 
sociolinguistically conditioned to a large extent, and, in turn, 
contributes to an already complex sociolinguistic situation in the 
Dravidian languages. In this section we shall briefly examine 
this aspect. 

6.1 Caste Dialects 

The existence of caste-related differences in Dravidian 
languages has long been recognized. In particular, sharp differences 
are noted between the Brahman and non-Brahman dialects on the levels 
of phonology, morphology, and lexicon. While many of these 
differences have nothing to do with borrowing (e.g., Kannada ide (B) 
versus aite (NB) for 'to be 3rd sing, neut.), quite a few of the 
variables involve elements introduced in the process of Indo- 
Aryanization, in particular, Sanskritization. The most important of 


these variables is aspiration in loanwords from Sanskrit. Even 
in these secular days, it is not \incommon to find parents and teachers 
upbraiding an unaspirating urchin for "speaking like a sudra "' Another 
important variable is the existence of separate sets of lexical 
items in the Brahman ajid non-Brahman dialects. As Aiyar (1932) has 
pointed out with reference to Tulu, words borrowed directly from 
Sanskrit are far more numerous in the Brahman dialects, while words 
borrowed or adapted from Middle Indo-Aryan (Jaina Prakrit or Pali) 
are common to both Brahman and non-Brahman dialects (e.g., Tu. 
upavasa (B) vs. nompu (NB) 'fast'; Ta. tirtho" (B) vs. tanni (NB) 
'water'). Even when the same Indo-Aryan word has been borrowed into 
both dialects, it \indergoes different modifications: (Tu. s'avira (B) vs. 
sara (NB) 'thousand'; Ka. baha^a (b) vs. b'a^.a (NB) 'much, very'). 

The reasons for this conservatism" of Brahmaxi dialects are not 
hard to find. The preservation of Sanskrit features in their speech 
gives Brahmans a social identity which their insignificant number 
denies them. As a byproduct of the elitist position of Sanskrit in 
India, this feature gives them prestige as well. Moreover, as 
Ramanujan (1968) correctly points out, part of the reason may also be 
the high premium placed in that community on the accurate rendering 
of the texts and their conservative outlook toward cultural innova- 
tions in general. 

The symbolic value of Sanskritic features in the Dravidian socio- 
linguistic context has also led to the inevitable phenomenon , of. .hyper- 
correction. Thus we find scores of native Dravidian words restructured 
to incorporate prestige features such as aspiration. Kannada offers 
many examples of this sort: bhatta 'paddy', cha^i 'cold', jhari 
'centepede', etc. (cf. also, Te. iddharu 'two persons'). These 
hypercorrections are manifestations, in a literal as well as in the 
sociological sense in which Srinivas (1966) employs the term, of 
Sanskritization . 

6.2 Diglossia 

Indo-Aryanization — in particular, Sanskritization — has also contri- 
buted to the age-old cleavage in Dravidian languages between the 
literary and the colloquial styles. In this context, the term' literary 
style' is used in its broad sense, to refer to all styles of insti- 
tutionalized use of language including that of the public platform, and 
is not restricted to the written form. There have been a number of 
studies of diglossia in Dravidian (e.g., Caldwell, 1903:78ff; Shanmugam 
Pillai, i960; Radhakrishna, 1971; Nayak, 1967), all of which point 
out that one of the distinguishing characteristics of the 'High' style 
is the preponderance of Sanskritic features. The 'High' style is 
characterized by the presence of non-Dravidian features such as 
aspiration, and the use of unassimilated borrowings, as well as struc- 
tures such as the passive construction and the clausal mode of relative 
clause formation. This penchant for mixing non-native elements in the 
literary style gives it a rather artificial and stilted character, 
and when carried to an extreme, as in the Manipravala style, often 
becomes unintelligible to the uninitiated layman. It is not surprising, 
therefore, that from time to time, there have been movements, in all 


the literary languages , to bring the literary style closer to the 
language of the common man, and a recurrent theme in these programs 
has been a call for reducing the Sanskritic component in the literary 
style (see the next section). 

The association of Sanskrit with the High style in Dravidian has 
resulted in an interesting phenomenon: the diglossic distribution of 
multiple derivatives on the borrowed words. Dravidisin languages 
provide many instances of two forms of the same Indo-Aryan loanword 
being used in different senses, one, typically the unassimilated 
Sanskrit word, retaining its original meaning or undergoing ame- 
lioration, and the other, typically the form borrowed or adapted from 
Prakrit, undergoing deterioration. The follwoing examples from 
Kannada are relevant: samsthe 'organization'; sante 'weekly village 
market'; parigattu 'institution'; parse 'annual fair'; dhana 'wealth'; 
dana 'cattle'; arya 'respected person, an Aryan'; aj j a 'grandfather, 
an old man'; vandhye 'infertile woman'; banje 'id (pej)'. 

In this connection, mention must also be made of the resort to 
Sanskrit terms for euphemisms in polite conversation and in writing. 
Thus, the preference for terms like jaghana 'thighs', kuca 'breast', 
mardana 'manipulation of the breasts', rati 'sexual intercourse ' asana 
'erotic position' etc, makes even erotic meinuals read like learned 
treatises! Turning to a sombre vein, terms like vidhave 'widow', 
nidhana 'death', sivaikya 'death', divan gat a 'dead* are generally 
preferred, in writing at least, to their native-language counter- 

6. 3 I n do -Aryan iz at ion and Language 

As pointed out earlier, the large-scale importation of Indo- 
Aryan features into the Dravidian languages has not gone unopposed, at 
least in the literary languages . However , the reasons for the oppo- 
sition, its intensity, and scope vary from language to language. 

In Tamil Nadu, it has turned out to be an emotional issue. Even 
from the dawn of Dravidian history, the attitude of Tamilians to this 
issue has been different from that of the speakers of other literary 
languages. It will be recalled (section 5.1) that To^kappiam sanctions 
the borrowing of only those words that fit into the phonological 
structure of Tamil, This restriction is not met with in the grammars 
of the other three literary languages , which have traditionally adopted 
a more liberal attitude to loan words , welcoming tatsamas and 
tadbhavas alike, often asserting that theaspirated sounds (mahaprana's) 
are integral to their language, and recognizing the expeinded scripts 
resulting from incorporation of non-native sounds. All of these 
grammars do, however, recognize the distinction between native and 
foreign ( d'esya and anyadesya ) words and enumerate eleborate rules for 
deriving the native forms of the loan-words, but they do not proscribe 
the use of tatsamas . In fact , they even go one step further and impli- 
citly encourage the use of Sanskrit compounds in their criticism of 
the employment of hybrid compounds (called arisamasa in Kannada and 
vairisajiiasa in Telugu) . Further, some grammarians even go out of 


their way to show the similarity between the native language Eind 
Sanskrit by concocting examples of constructions never used in the 
language. These grammars are also unlike Tolkappiam in their 
adoption of the Paninian model and terminology. (Some of these grammars 
are, in fact, written in Sanskrit.) Thus the difference in the attitude 
of the Tamilians on the one hand and the speakers of the other literary 
languages on the other toward the borrowing of Indo-Aryan features has 
been pronoimced and long-standing. 

While the Andhras , the Kannadigas , and the Malayalis have wel- 
comed Indo-Aryan borrowings in principle, they have often reacted to 
the indiscriminate use of these loans. As noted earlier, the Mar}iprava].a 
style never found general grass-roots appeal. On the contrary, there 
arose a number of counter-movements seeking to bring the literary 
language closer to the common colloquial idiom. In Kannada, for 
example, there have been at least three such movements. Andayya 
(c. 13th century) composed his Kabbigara Kava . using only native words 
and assimilated loans; the Virasaiva protest (l2th century) against 
the tyranny of Brahmanism found its literary counterpart in the vacanas 
whose language is relatively free from Sanskrit borrowings. In modern 
times, there has been a sharp reaction against the Sanskritiz^d fet^^-l^e 
of Kuvempu and others, and a call to eschew unfamiliar Sanskrit words. 
Similar trends are found in Telugu and Malayalam as well, for example, 
in the work of Somanatha (who advocated janu Telugu 'pure Telugu') 
and Ezhuttacan (a contemporary Malayalam poet). 

The target of these movements, however, is only the excessive use 
of Sanskritic elements — unfamiliar words, long compounds, etc. — but 
the forms already present in common parlance — both assimilated and 
unassimilated — are accepted as a proper part of their respective 
languages. Thus their attitude toward In do -Aryan iz at ion is essentially 
rooted in considerations of general intelligibility rather than any 
philosophical objection to borrowing per se. The latter attitude, 
however, has played an increasingly important role in Tamilian 
glossopolitics in this century. 

In his excellent discussion of the purism movement in Tamil, 
Annamalai (to appear) observes that the movement was fostered by 
several factors: the desire to assert the identity of Tamils following 
the Independence movement which stressed national unity, the discovery 
of the antiquity of Tamil and its self-sufficiency as demonstrated in 
the ancient Sangam literature, the change of power-structure from the 
domination of Brahmans to that of non-Brahmans , among others. 

The scope of the "pure Tamil" movement was quite comprehensive. 
It covered all styles of language, and affected the script,, the voca- 
bulary, and even proper names. Most of the borrowed words (including 
assimilated words) were excised and native Tamil words from the early 
literature revived to take their place. An informal estimate (quoted 
in Annamalai, op.cit.) is that the Sanskrit words in use in Tamil 
writing have come down from fifty percent to twenty percent in the last 
fifty years. However, there do seem to be more Sanskrit words used 
in spoken Tamil. This situation is similar to early modern Icelandic, 
but we know where modern Icelandic has gone. 



As we conclude this 'broad survey of the influence of Indo-Aryan 
languages on Dravidian, it may not be out of place to step back, as 
it were, and ask, what in the final analysis has been the role of 
this process of Indo-Aryanization in the history of Dravidian. This 
is a very difficult question, not only because it can quickly get 
transformed into an emotional question, but also because the process 
is an ongoing one. However, trying to steer clear of value-judge- 
ments, one may attempt a partial answer. 

First of all, there is no denying the fact that the impact of 
Indo-Aryan has been both profound and widespread. There is no 
Dravidian language that has not been touched by this influence, nor is 
there any aspect ■ ;. of a given language. The Dravidian languages 
have stood to gain from this process in many ways — their vocabularies 
have been enriched and their style-range widened. They have developed 
countless new registers and added to their metrical repertoire. It 
should not be forgotten that many of the Indo-Aryan words were first 
introduced by way of literary compositions, some of which are among 
the finest in Dravidian literature. The Indo-Aryan influence came 
at a time when the Dravidian languages were beginning to assert 
their separate identities and aided them in this process. It may 
be recalled that Indo-Aryanization was probably the key factor in 
the separation of Malayalam from Old Tamil, the former having 
adopted a more liberal attitude towards the importation of Indo-'Aryan 

Yet, one does not have to be a purist to concede that Indo- 
Aryanization has not been an unmixed blessing. Among what in some 
quarters may be regarded as not quite wholesome effects of this process 
are the following: the creation of an elite variety based on the 
incorporation of foreign elements; the emergence of a pedantic 
literary style, based on the same principle; widespread linguistic 
slavery manifested in the preference for borrowings over exploitation 
of native resources; the disappearence of scores of perfectly adequate 
and expressive native term.s under the onslought of the borrowings; 
and so on. 

Granting that the tendency to borrow has been excessive and indis- 
criminate, it may still be maintained that Indo-Aryanization per se 
need not take all the blame for the phenomena just mentioned ^ Socially 
defined variation is an integral feature of language whether borrowed 
features serve as the relevant variable or not; variety in literary 
style is also equally inherent in the medium, literature would be of 
little interest any other way; words do change their meaning, fall 
into desuetude, and their place is not always taken by borrowed items. 
The question, then, is, would the Dravidian languages have developed 
the variety of registers that they now have, without the benefit of 
Indo-Aryanization? Probably yes; but the more relevant question is, 
whatever be our language policy for developing technical terms to serve 
in contemporary registers, must we get rid of the terms that are already 
in use, just because they did not once belong to our language" The 
Tamiliams say yes, the other Dravidians say no. So each group goes 


in its own way, the former resorting to ancient Tamil, the latter to 
classical Sanskrit, but the two have one thing in common — they are 
both attempting to blow new life into antique forms, while a 
relative newcomer to the scene, English, knocks persistently at 
the door. 

*I am grateful to Professors Braj Kachru and Hans Hock for 
commenting on an earlier version of this paper. 

^The last part of this claim is not strictly correct. Burrow 
and Bhattacharya (l9T0:ix) note that in many areas, the Pengos 
have lost their own language and have become Indo-Aryan speakers . 
The authors cite the example of the agricultural tract west of 
Nowrangpur and adjoining the Bastar border, "where the Pengos no 
longer speak their own language, though they are aware of its 
one-time existence." Nevertheless, in so far as the true character 
and identity of a language resides in its morphology and syntax 
(apparently basic vocabulary is less impregnable than commonly assumed 
if the Dravidian case is any proof Csee section k.l belowti) , Dravidian 
languages may be said to be still pretty much intact. 

^For a cautious questioning of this traditionally accepted 
position, see Hock (1975)- 

■^See References at the end of this paper. The items marked by 
as asterisk were not available to me at the time of writing. 

^However, the reliability of the method of glottochronology on 
which Andronov's claim is presumably based, has been called into 
question; see, e.g., Bergsland and Vogt , 1962; also note Hock 19T6. 

5l am aware that the use of the term 'caste dialects' has been 
criticized recently, on the basis of the observation that education 
rather than caste is the relevant variable. However, if we recall that 
until recently, the term' educated' could be used interchangeably 
with 'Brahman', much of the ground for objection disappears. Moreover, 
sociolinguistic stereotypes are not always commensurate with socio- 
logical reality. 

°This list is limited to those borrowings from Indo-Aryan which 
have either l)undergone extensions of meanings or far-reaching 
phonetic changes or both; or 2) penetrated into non-literary languages, 
whether or not special meanings are found (Emeneau and Burrow, 19^2 ) . 
Preface) . 

^One can cite numerous examples of this tendency. Sekhar (1969: 
l6l) has observed that Malayalam inscriptions of the 11th and 12th 
centuries contain examples of concord between modifiers and substantives. 


This phenomenon seems to have existed in Telugu as well (Sastri, 
1969:271). According to Sekhar, Lilatilakam , the lUth century grammar 
disapproves of such imitations of Sanskrit syntax yi Malayalam. The 
highly respected 13th century grammar of Kannada, Sabdamapidarpapa 
contains an "extrapolated"' (an\ikta) sutra in which it is claimed that 
even intransitive verbs can be passivized in Kannada. 

"'Conservatism' as applied to Brahman dialects in Dravidian 
may be confusing. It is true that the Brahman dialects have 
traditionally been innovative in acquiring non-native phonology, 
lexis, etc. However, in so far as the borrowed elements do not 
undergo the processes of nativization that apply generally in their 
respective languages, the Brahman dialects are relatively more 
conservative (see Bright and Ramanujan, op.cit.. and the references 
cited in that work for further discussion). 


ANDRONOV, M. I96H . On the typological similarity of New Indo-Aryan 

and Dravidian. Indian Linguistics 25. 119-26. 
AIYAR, L.V.RAMASWAMI. 1932. Tulu prose texts in two dialects. 

Bulletin of the School of Oriental Studies 6.897 -931. 
AIYAR, L.V. RAMASWAMI. 193^^-35. Semantic divergencies in Indo-Aryan 

loanwords in South Dravidian. Journal of Oriental Research. 

Madras, 8-9. 
AIYAR, L.V. RAMASWAMI. 1973. A south Indian (Malayalam) evaluation 

of Sanskrit t(d) and t(d). International Journal of Dravidian 

Linguistics. 2,1,119-26.* 
ANNAMALAI, E. (to appear). The movement for linguistic purism: 

the case of Tamil. Language Movements of this Century in 

India. Mysore: Central Institute of Indian Languages. 
BERGSLAOT), KNUT and HAUS VOGT. I962. On the validity of glottochronology , 

Current Anthropology 3. 115 -53. 
BHATTACHARYA, SUDHIBHUSAN. 1957. Ollari: a Dravidian speech. 

Calcutta: Anthropological Survey of India. 
BLOCH, JULES. 195^. The grammatical structure of Dravidian languages 

(English translation by R.G. Harshe). Poona: Deccan College, 

Post-graduate and Research Institute. 
BRIGHT, WILLIAM. I96O. A study of caste and dialect in I^ysore. 

Indian Linguistics 21. 1+5-50. 
BRIGHT, WILLIAM and RAMANUJAN, A.K. 196U . Sociolinguistic variation 

and language change. Proceedings of the Ninth International 

Congress of Linguists. Cambridge, Mass. 
BURROW, T. andBHATTACHARYA, S. 1970. The Pengo language. Oxford: 

Clarendon Press. 
CALD'.rELL, ROBERT. 1903. A comparative grammar of the Dravidian 

or South Indian family of languages. 3rd edition. (Reprinted 

197^. Delhi: Orient Publications.) 
*DONAPPA, T. I96U. Middle Indo-Aryan loanwords in Telug . Waltair: 

(Unpublished Ph.D. Thesis) Andhra University. 


EMENEAU, MURRAY B, 1962: a. Dravidian and Indian linguistics. Berkeley: 

Center for South Asian Studies, University of California. 
EMENEAU, MURRAY B. 1962 :b. Bilingualism and structural borrowing. 

Proceedings of the American Philosophical Society. 106:5 

EMEMEAU, MURRAY B. and BURROW, T, 1962. Dravidian borrowings from 

Indo-Aryan. UCPL 26, Berkeley: University of California. 
FOWLER, MURRAY. 195^. The segmental phonemes of Sanskritized 

Tamil. Language 30. 360-T. 
GANESHSUWDARAM, P.O. and VAIDYANATHAN , S. 1958. An evaluation of 

the Sanskrit loanwords in Tamil from the point of view of 

Nannul. Indian Linguistics. 19. 63-70. 
GODA VARMA, K. 19^6. Indo-Aryan loan-words in Malayalam. Mavelikkara: 

Rama Varma and Bros. 
GUMPEREZ, JOHN J. and WILSON, R. 1971. Convergence and creolization: 

a case from the Indo-Aryan/Dravidian border in India. Dell 

H. Humes, ed. Pidginization and Creolization of Languages. 

HOCK, HANS HENRICH. 1975- Substratum influence on (Rig-Vedic) 

Sanskrit. Studies in the Linguistic Sciences 5:2. 1975. 

Urbana: Department of Linguistics, University of Illinois. 
HOCK, HANS HENRICH. 1976. Review article on R. Anttila: An 

introduction to historical comparative linguistics. Language 52. 

KACHRO, BRAJ B. 1978. Toward structuring the form and function of 

code-mixing: an Indian perspective. Braj B. Kachru and S.N. 

Sridhar eds. Aspects of Sociolinguistics in South Asia. 

Special Issue (No. 16) of International Journal of the Sociology 

of Language, 21-UO. 
KITTEL, T. I89U. A Kannada-English dictionary. Mangalore. 
KUNJUNNI raja K. 1972. The influence of Sanskrit on the Dravidian 

literatures with special reference fo Malayalam. Arabinda 

Poddar, ed. : Indian Literature. Indian Institute of Advanced Study. 
NADKARNI, MANGESH V. 1970. NP embedded structures in Kannada and 

Konkani. Ph.D. dissertation. University of California, Los Angeles. 
NAYAK, H.M. 1967. Kannada: literary and colloquial — a study in 

two styles. Mysore: Rao and Raghavan. 
* PILLAI, ANAVARTAVINAYAKAM S. 192^. The Sanskrit element in 

the vocabularies of the Dravidian languages. Mark Collins, ed. 

Dravidic Studies III. Madras: University of Madras, 27-^8. 
RADHAKRISHNA, B. 1971. Diglossia in Telugu. Proceedings of the 

First All India Conference of Linguists. 218-26. 
RAMANUJAN, A.K. I969. The structure of Variation: a study in 

caste dialects. In Milton Singer and Bernard S. Cohn, eds.: 

Structure and change in Indian society, pp. U6l-7^. Chicago: 

Idine Pub. Co. 
SASTRI, KORADE MAHADEVA. I969 . Historical grammar or Telugu— 

with special reference to Old Telugu, c, 200 B.C. -1000 A.D. . 

Anantapur: Sri Vendateswara University Post-graduate Centre. 
SEKHAR, A.C. 1959. Evolution of Malayalam. Poona: Deccan College 

Post-graduate and Research Institute. 
SHANMUGAM PILLAI, M. 1972. Tamil—literary and colloquial. 

Linguistics 81. 83-91. 


SJOBERG, MDREE F. 1962. Co-existant phonemic systems in 

Telugu. Word l8. 269-79- 
SJOBERG, AWDREE F. and GIDEON SJOBERG. 1956. Problems in 

glottochronology: culture as a significant variable in 

lexical change. American Anthropologist 58:2, 296-300. 
SREEKANTAIYA , T.N. 1956. Notes on loans and native replacements 

in Kannada. American Anthropologist 58. 306-8. 
SRIDHAR, S.N. 1975. On the Indo-Aryanization of the Kannada 

lexicon: a study of the Hindi-Urdu loanwords in Kannada. 

In H. Van Olphen, ed. : Language borrowing. 99-108. Austin, Texas: 

Department of South Asian Languages and Literatures, 

University of Texas. 
SRIDHAR, S.N. 1978. On the functions of code-mixing in Kannada. 

In Braj B. Kachru and S.N. Sridhar, eds.: Aspects of 

sociolinguistics in South Asia. Special issue (No. l6) 

of International Journal of the Sociology of Language. 

SRINIVAS, M.N. 1966. Social change in modern India. Berkeley: 

University of California Press. 
SUBRAMONIAM, V.I. and GANESHSUNDARAM , P.C. 195^. Marathi loans 

in Tamil. Indian Linguistics. 1^+. lOU-23. 
UPADHYAYA, U.P. 1971. Effects of bilingualism on Bidar Kannada. 

Indian Linguistics. 32.2. 132-138. 
*VAIDYANATHAN , S. 1971. Indo-Aryan loanwords in Old Tamil. 

Madras: Raj an Publishers. 

studios in the Linguistic Sciences 
Volume 8, Number 1, Spring 1978 



Susan U, Stucky 

This paper discusses the status of the noun class 
system in a Bantu lingua franca, Kituba, spoken In 
southeim Zaire. Section 1, outlines the general character- 
istics of Bantu noun class systems. Noun classes are 
marked morphologically by prefixes on the noun In both 
the singular and the plural. Typically, these prefixes 
trigger agreonent for a variety of syntactic categories 
(i.e. verbs, pronouns, and adjectives). The data In 
section 2. shows that Kltuba has lost many of the plural 
prefix markers and, of those that remain, many appear 
to be unstable. In addition, one of the plural markers 
is being generalized. It is argued that the language 
is moving towards a system in which the singular is 
unmarked while the plural Is marked. This analysis Is 
supported by evidence from agreement patterns and the 
behavior of loan words. Section 3» compares the Kltuba 
noun system with that of other Bantu lingua francae. It 
is pointed out that while Kltuba shares many patterns 
with the other trade languages. It has gone further In 
the elimination of a noun class system. The claim is 
then that Kltuba Is moving towards a nonclass system 
with an unmarked singular and a marked plural. Further- 
more, It is concluded that this change is spreading 
through the lexicon item by item and proceeds through 
a variety of mechanisms. 

1.0. Introduction 

This paper presents a discussion of the status of the noun class system 
in Kituba» a Bantu language which has undergone considerable simplification 
over the past several hundred years. Today, Kituba is used primarily as a 
means of ccxnmunication between various Bantu language speaking peoples in 
Bas Zaire, In the Kwangu-Kwilu region of Zaire, and In the southwestern 
part of Congo-Brazzaville. The data in this paper were obtained from 
Sallkoko Mufwene, a native speaker of Kltuba as it is spoken In Bandundu 
In Zaire. Sallkoko refers to his language as either Klkongo ya leta or 
Kikongo ya bulamatari, which are the two names applied to Kituba In the 
western Kltuba speaking area. 

While simplification has taken place in several parts of the grammar, 
as is typical of trade languages such as Kltuba, It is the reduction of 
the noun class system which is the topic of this paper. 


The noun classes in Bantu languages are marked by prefixes on the noun 
stem. They typically fall into singular and plural pairs so that the noun 
has an overt morphological marker for the singular noun and a separate overt 
marker for the plural. It will be shown that the Kituba noun class system 
has reduced the number of singular -plural prefix pairs significantly and, 
in addition, that the system is moving towards an unmarked singular^narked 
plural system through the generalization of one of the existing plural 
markers. Another characteristic of Bantu noun class systems, agreement, 
will also be shown to have entirely disappeared. Taken together, 
these two facts do suggest that very little is left of the noun class 
system. It is concluded that the noun class system in Kituba is barely 
functional and is in the process of being eliminated entirely. 

1,1, Background 

Some background information about Bantu noun class sjrstems will be 
helpful in following the discussion. As mentioned above, the noun in Bantu 
languages belongs to a noun class which is marked by a prefix (sometimes 0) 
on the noun itself. While the number of classes varies from language to 
language, there are rarely fewer than ten or more than eighteen in any given 
language (Guthrie 196?: 13) • TWenty-three proto classes are proposed by 
Cole (1971), the largest number of classes reconstructed so far. Within 
Bantu linguistic tradition, the classes are numbered with reference to the 
proto-classes. This system will be followed here so that the Kituba 
evidence may be more easily compared to other Bantu languages. 

One of the clearest functions of the noun class syston is marking the 
difference between singular and plural. Often there is a pairing of the 
singular-plural prefixes and when the term gender is applied to a language 
of this type, it refers to a pair of prefixes, one which marks the singular 
and one which marks the plural. Certain behavior of the nouns, however, 
precludes treatment of the entire system as pairs of singular and plural 
prefixes. Some classes, such as the locatives, the abstract class, and the 
liquid class, are not markee for either singular or plural. In addition, 
two singular prefixes may take a single plural prefix. For eocample, Proto- 
Bantu Class 10 (henceforth marked Class*10), a plural marker, is often 
found as the plural of both Class*9 and Class*ll in a given language. 
Furthermore, a single prefix in a modern language may be a reflex of two 
prefixes historically. Finally, there is no necessary correlation between 
what constitutes a singular-plural pair in one language with this distinc- 
tion in another. Thus, each prefix must be handled separately. What is 
significant for this paper is that number is overtly marked in the Bantu 
languages because both singular and plural forms e^diibit a prefix which 
may be clearly identifiable as separable frcsn the noun stem, 

A further characteristic of Bantu noun classes is that they govern 
agreement in a wide variety of constructions. The categories most often 
exhibiting agreement include 1) the genitive marker, 2) subject and object 
pronouns (both clitics and independent forms), 3) adjectives, and I*) verbs. 
In some cases, a single phonetic form of a prefix may govern two types of 


agreement , This is generally taken to be evidence that this prefix is not 
the marker of a singular class but of two separate classes. 

A third feature of noun class systems in Bantu languages is the semantic 
domain assigned to many of the classes. Th« correlation between a given noun 
class marker in a Bantu language and a semantic domain is not as consistent 
as in the instance of number, so caution is reeonnended in treating the 
classes as representative of semantic daaains. Nouns do not often fall into 
categories assigned to the prefix."' 

In this analysis, nouns in Kituba will be evaluated along the two most 
consistent parameters, number and agreement, with only slight reference to 
semantic domain. 

2.0. The analysis 

To perform the analysis of the Kituba noun systaa, a list of approxi- 
mately 450 singular nouns was submitted to the consultant who then provided 
the corresponding plural forms. The original list was extracted from three 
sources J ELiet (1933), Swift and Zola (I963), and Fehderau (I966). The 
first two sources provide a vocabulary from the northern Kituba region. It 
is interesting to note that there was almost no difference between the sing- 
ular and plural forms given by the consultant and those provided in the 
three other sources. Thus, western and northern Kituba are fairly cohesive 
with respect to the basic vocabulary. 

2.1,0. The expression of number in Kituba nouns 

This section establishes the relationship of the noun class prefixes 
exhibited in the languages Fehderau (I966) established to be most closely 
related to Kituba, These languages include the central Kikongo dialects 
(Kimanyanga, Kinyombe (north), Ladi and Kisikongo). The Kituba nouns were 
checked against the central Kikongo vocabulary noted by Laman (193^) to 
determine what changes in class affiliation have taken place. The resulting 
system is outlined below. Note that because no agreement remains, the only 
criterion on which to base the existence of a class will be its pairing with 
a plural prefix. Class 1 mu- 

Very few examples of Class 1 nouns remain in Kituba. Earlier changes 
in the Kikongo languages show that many of the nouns reconstructed to be in 
Class*l have as a prefix only a homorganic nasal. Several scenarios have 
been proposed for the loss of the vowel of Class i mu- and the subsequent 
assimilation of the nasal. (See Bell 1972, for example). None of these 
seem to account for this phenomenon in Kituba, however. What is important 
here is that the appearance of many Class 1 nouns in Kituba with a prefix 
N- is not an innovation in Kituba. There is another class of nouns which 
take N^ as well. They belong to Class*10. In Kikongo, Class 1 and Class 10 
nouns were distinguished by their concord. However, since concord has been 


lost in Kituba, the two classes of nouns are indistinguishable. The 
result is a scission in Class 1 such that some nouns originally in Class 1 
now appear in Class 10. 

Only two nouns out of the entire lexicon elicited behave as though 
they belong to Class 1. These are the following! 

1. mu-ana /pwani7 'child' 
ba-ana ^axi&J 'children* 

2 . mu -ntu ' person ' 
ba-ntu 'people' 

Class 1 nouns typically take their plurals in Class 2, These no\ms 
show no sign of change in the plural. 

Several other mass nouns which cannot have plurals but which do exhibit 
a mu- sequence word initially are the following i 

3. mwamba 'sauce' 
mutoki 'perspiration' 
mubisu 'rawness' 

Since these forms do not govern agreement nor take plurals, there is no 
evidence that they are analyzed as containing prefixes. Neither is there 
any direct evidence that they do not. There is a possibility for mutoki 
•perspiration' to be analyzable into a prefix and stan since a verb ku-toka 
'to sweat' does exist in Kituba,^ Class la 0- 

Class la is a class that did not have a reconstructed prefix. It is 
usually paired with a Class 2 ba- plural. Class la is t3rpically restricted 
to kinship terms. In Kituba, this class is only of historical interest 
since other pairs of nouns with no singular marking and a ba- prefixed to 
the sg. stem have been added to the system. Examples include the following: 

bakala/babakala 'man, husband' 

nkuluntu/bankuluntu 'older sibling' 

mama /bamama ' mother ' 

tata /batata 'father, sir' 


Class 2 ba- has already been mentioned with reference to Class 1 and la, 
for which it functions as a plural marker. Ba- has beooaie a plural marker 
for many noxms in Kituba and it is generally prefixed to the entire noun as 
in the examples in {k) above. This is the form which, it will be argued, is 
being generalizea as a plural marker. 


2,1.1.3 Classes 3 "m- and k mi- 

Class 3 mu- was originally distinguished from Class 1 mu- in Klkongo 
for two reasons. First, Class 1 mu- takes as its plural Class 2 ba- , while 
Class 3 mu- takes as its plural Class k mi- . Secondly, Class 1 and Class 3 
goTemed different agreenent forms, a distinction which can no longer exist 
in Kituba since all noun class agreaaent has been lost. In addition, many 
of the nouns in Class 3 merged with nouns in Class 9 Hh ^®* ^^® vowel of 
the prefix was lost. Thus, some nouns which originally belonged to Class 3 
now belong in Class 9 and the rest are indistinguishable from Class 1 nouns 
except for their plural pairings. 

In Kituba, some change is occuring with respect to the plural marker, j 
In Kikongo, the normal plural form for Class 3 mu- would have been Class k 
ml-. Nouns in Class 3 in Kituba taking Class ^ mi- plurals are the following: 



nutunga /mi tunga 
mudinga /midinga 
mukanda /mikanda 
fflunoko /minoko 

'breeze, wind' 
'finger, toe' 

•variety, sort' 
'needle, hospital* 

'voice, throat' 

For a group of nouns in Class 3 there are alternative plural forms, 
Note that the first alternative plural consists of ba- prefixed to the 
singular noun stem. The second plural form is made up of ba- plus the 
original plural marked by mi-» These are as follows! 

6. mungamba/mingamba 

'day laborer' 


'spokesman, speaker' 










mundele/aindele 'white person, European' 

mukombi/mlkombi 'sweeper* 


rawinda /minda ' light ' 


Note that the alternative plural forms were all human nouns but one 
(the last). Since Class l/2 typically contains human nouns, it might be 
argued that these forms were derived by the addition of ba- marking humans. 
Additional support for this analysis comes from the fact that no alternative 
agreement forms now exist for distinguishing Class 1 mu- from Class 3 mu-. 
This seems to be a plausible explanation for the ba+stera forms. The gener- 
alization of the ba+mi+stem forms could also be a result of the human marking 
of ba- . However, we will see later that ba- has been generalized to marking 
the plural for many nouns not of the human class. In light of this general 
pattern in the language, forms like ba+mi- stem suggest that the prefixes 
mu- and mi- are no longer separable and that ba- was prefixed to an already 
existing plural form for which the mi- was no longer separable. An alterna- 
tive explanation, that these forms were formed by analogy to French loans 
where ba- was prefixed to French plurals, is less easy to argue for or 
against since such an analogical process requires that the speaker be aware 
that the French loans were plural to begin with, A neighboring lingua franca, 
Lingala, does, however, Bokamba (personal cc»nmunication) has pointed out 
that Lingala (a neighboring lingua franca also has such ba-mi- forms. He 
suggests that historically the Lingala ba-mi plurals were derived by analogy 
to the Kituba forms since such plurals are found in only Kinshasa Lingala 
(the area of greatest contact between Kituba and Lingala speakers) and since 
these fonns are most common among Kituba speakers of Lingala. 

2.1.1.^ Class 5 di- 

Class 5 has been retained from Kikongo with no change. It has the 
prefix di- in the singular and takes the plural in Class 6 ma-« There are 
no exceptions in the data. Class 5 often contains body parts and names of 
food and plants. Here is a sample: 






'arm, hand' 


•foot, leg' 





Class 6 ma- 

This plural prefix is more versatile than sc«ne plural prefixes because 
it functioned historically in Kikongo (the base language) as the plural of 


nouns In three classes: Classes 5f 11> &nd 1^, and also for 8<»te mass novms, 
particularly those referring to liquids. The Class 5/6 pairing was noted in 
(7) above. Examples of the Class 11/6 pairing include only onei lupangu/ 
mapangu 'fence,' The other two found in the vocabularies ( lukaya 'leaf,' and 
Iwinda 'cockscomb') were not used by the consultant. Only one Class 14 bu- 
noun was in the list. However, unlike the Kikongo plural which was in Class 
6, the plural in Kltuba has shifted to a ba- plural which is prefixed to 
the singular noun. 

8. bu-ala (bwala)/babwala 'village' 
The liquid ma- nouns are given below: 

9, malafu 'wine, liquor' 
maza 'water, river' 
mafuta 'oil' 

A number of mass nouns also take the Class 6 form. 

10. makasi 'anger' 
madidi ' coldness ' 
raakelele 'noise' 

A few nouns which are not in the liquid class also take a raa- plural prefixed 
to the singular form, 

11. raasua/bamasua 'boat' 
f alanga /maf alanga ' franc ' 
fofolo/mafofolo 'match* 

talatala/matalatala 'glass' 

The first of these is undoubtedly of Bantu origin. The second is a 
borrowing from French but has been widely used in many of the Bantu languages 
in the area. The second two are more difficult to identify and no explana- 
tion is offered. Note that one form has an alternative ba -plural. Another 
item (clearly a borrowing) for which Laman (I936), Swift and Zola (I963), 
and Eliet (1953) recorded a ma-plural and which has shifted to a ba-plural 
in Salikoko's speech, is the form papayi /bapapayi 'papaya.' 

Thus Class 6, like Class 4, exhibits a certain degree of instability. 
Some nouns originally taking a plural in Class 6 have shifted to an unmarked 
singular form and a marked plural ( ba- ) form. Others have not. The Class 
5-6 pairing, on the other hand, appears to be relatively stable, 

2,1.1.6 Classes 7 ki- and 8 bi- 

Class 7 ki- nouns took their plural in Class 8 bi- in Kikongo, This 
pairing still holds for Kituba. Approximately twenty nouns out of the list 
show this alternation. Several examples follow: 


12. kibakaMbaka 'wall' 
klfu/bifu 'habit, custom' 
kisalu/bisalu 'work, job' 
kisanunu/bisanunu 'ccxab' 

Certain items have been borrowed from Lingala into Kituba. The 
Lingala reflex of Class 7 is, however e^ and not ki- . Class 9 in Lingala 
remains bi- » The prefixes of such nouns have not been changed on the 
borrowed nouns although some alternation between £ and u is noted. Two 
such examples were noted in the list and are given below. 

13. el^agi/bilungi 'face' (Ling, elongi) 
ekeku/bikeku 'statue' (ling, ekeko) 

One could perhaps account for these forms by saying that since Salikoko 
knows Lingala he uses these forms when speaking Kituba, Bokamba (personal 
communication) has remarked that a u-o alternation is common in Lingala as 
spoken by Kituba speakers and furthermore that Kituba u often correlates to 
Lingala o. Thus, such a correlation seems to be common and accoiints for 
the alternation in (13). 

2,1,1,7 Class 9 N- 

The reflex of Class *9 ni- in Kikongo is a nasal consonant (N^) which 
assimilates in place of articulation to a following consonant. Various 
nasal consonants ( n , g ) appear before vowel initial stems as a result 
of a number of historical changes which need not concern us here. Reflexes 
of Class *9 are numerous. This class is often cited as the animal class 
and there are ample examples of these in the list. In Kikongo (as in many 
other Bantu languages), Class 9 took a plural form in Class 10. However, 
Class 9 nouns in Kituba now take a plural ba- prefixed to the singular form, 

14. mbulu/barabulu 'jackal' 
mbuluku/bambuluku 'antelope' 
ngombe/bangombe ' cow ' 
mfwenge/barafwenge 'civit-like animal ' 
mf inda /bamf inda ' forest ' 

ndunda/bandunda 'vegetables /varieties of vegetables' 
nioka/banioka ' snake ' 

nuni/banuni ' bird ' 

In Kikongo, monosyllabic nouns in Class 9 appear with a prefix vowel 
i- appearing before the nasal. In Kituba these have all but disappeared. 
Their plural forms are of the same type of those in (I'*) above, 

15. (i)mbwa/ba(i)mbwa 'dog* 
( i )mpu/bampu ' hat ' 

mvu/bamvu 'year' 

nzo/banzo 'house' 


Since Kituba nouns do not govern agreement, the Class 1 ma- and 
Class 3 nouns with reduced prefixes have fallen together with nouns of 
Class 9« Both sets now take a marked plural fonn where ba- is prefixed 
to the singular. Nouns containing a nasal word initial in the singular 
and a ba- plural marker prefixed to the singular form constitute, statist- 
ically speaking, one of the largest groups of nouns in the list, Class 10 N- or zi- 

Class 10, which was either N^ or zi- in Kikongo depending on the 
dialect, was, at first, ai- in Kituba. Shepherd's dictionary (1926) 
listed Class 10 ai- as the plural of Class 9 but by the time Fehderau did 
his research (1960* s) ai- had disappeared entirely and had been replaced 
by the prefixation of ba- to the singular form. Class 10, then, has 
entirely disappeared; its pluralization function has been replaced by the 
ba- prefix. Class 11 lu- 

Class 11 lu- has a variety of plurals associated with it. Class 6 
ma- has already been mentioned (See section Class 13 tu- was 
at one time a regular plural for Class 11. However, there are only a 
very few examples of Class 11 in the data. Regardless of their plural 
forms in Kikongo, each instance has been replaced by the marked plural form. 

l6, lusambu/balusarabu 'prayer' 
lusadisu/balusadisu ' aid ' 
lutu/balutu ' spoon ' Class 13 ka- 

Class 13 ka- apparently has disappeared entirely, there being no 
examples in the list having this form. Class 14 bu- 

Class 14 bu- is attested in only two words. The first bwala/babwala 
•village' was discussed in (2,1,1.5). Its plural has been reassigned from 
Class o ma- to the marked plural ba- form. Abstract nouns often appear in 
this class. By virtue of their abstract meaning they do not have plural 
forms. One such example is buaoba 'stupidity,' It is related semantically 
and perhaps morphologically to Kituba aoba 'stupid,' another noun of Bantu 
origin. Thus it may be that the prefix bu- of buaoba 'stupidity' is still 
analyzable as a prefix. Still, Class 14 as a plural marker has all but 
disappeared. Class 15 ku- 

Class 15 ku- is the infinitive form. Its status as a noun is not known 
since no examples appear in the sample. In Kikongo, the ku- form can be 


used as a noun and has as its plural Class 6 ma-. As a verbal form In 
Kituba, it appears in a past progressive construction with an auxiliary 

2,1,1,13 Classes l6 va-. 1? loi- and 18 mu- 

The status of Classes 16-18 is likewise unclear. In Kikongo they 
had two functions, one to mark a noun as a locative and the other to 
construct locative phrases consisting of one of these prefixes and a full 
noun. The status of the locatives in this second function is not made 
clear in any of the sources. According to Fehderau ",,,Lingala forms are 
more current in the case of these prefixes" (Fehderau 1966:55). That is, 
Lingala na 'in, at, to, on, by' is preferred to Kikongo (Niraanyanga dialect 
rau 'in,' ku 'at, to,' va- 'on,' kwa 'by,' The choice of na over the loca- 
tives irould result in the elimination of Classes 16-18, Na in Kituba is 
an all purpose particle covering comitative, instrumental, conjunctive, 
and locative constructions. While locatives were not investigated to any 
degree in Salikoko's speech several examples using rw as a locative turned 
up in the data. Here is an example » 

17. familia na mono meme bika na bwala 

family of me past stay loc. village 
'my family stayed home* 

Locative prefixes also appear attached to noun stons, as noted above, to 
form locative nouns. None of these were obtained in the sample either. It 
may be concluded, then, that Classes 16-18 are either not very productive 
in Kituba or have disappeared entirely. Class 19 fi- 

This class is a diminutive in Kikongo, having the plural in Class 8 
bi-. In Kituba, according to Swift and Zola (I963), all take a ba- prefixed 
to the singular to form a plural. The few examples cited by Swift and Zola 
were not known by Salikoko the Kituba consultant so that no statement can ^ 
be made about the status of this class. 

2.2 Conclusions and hypothesis 

Several generalizations can be drawn from the examples just discussed. 
First, a new method of plural formation has emerged. By marking the singu- 
lar noun with a ba- to form the plural, part of the system has shifted 
away from a marked singular /marked plural. Into this group of nouns fall 
all nouns formerly taking a class 10 plural prefix and those nouns in Classes 
1 and 3 which already had a reduced form of the prefixes (a nasal C), The 
remaining plural classes. Classes k, 6, 8, and 13 all show some degree of 
instability. That is, some nouns which took one of these class markers for 
the plural now take ba- exclusively (prefixed to the singular form) or 
show some degree of alternation between the newer ba-stem form and the 
original plural class marker. Thus, it appears that the system is in a state 

of flvix. For some classes, the movement to a morphologically marked plural 
is conplete. For others (4,6,8, 13), it is not. This change would conform 
to Greenberg's Universal 35 irtiich claims that there are no languages in 
which the nonsingular would be (moirphologically) marked as opposed to the 
singular (Greenberg 1966i9^), since in every case, marking has taken place 
in the plural and not the singular when such marking has taken place (e»g. 
plurals for 3, 9, H, and lU), 

Furthermore, the change has apparently taken place not by a single 
mechanism, but by several. Thus, the shift of most of Class 3 from a Class 
k plural to a double prefix took place for more human nouns than for non- 
human nouns. This leads one to suppose that the semantics of Class 2 ba- 
may have SOTie significance for the change. That is, the addition of ba- 
may have been, at least in scxne cases, for semantic reasons. Or, alterna- 
tively, it may have been by analogy to class la where no singular prefix 
existed. Secondly, the change in Class 9/10 nouns may have been internally 
motivated as well. In the original system. Class 9 and 10 nouns would not 
have been distinguished in the singular and the plural after the elimina- 
tion of concord. Thus, there may have been pressure to mark plurality. 
Finally, the classes structurally most vulnerable (i.e. 11, ih) have been 
affected. Thus, while the overriding reason for the change in the noun 
class system may have been a drive towards a non-class system, or, at least 
a simplification of it, the mechanisms are diverse. 

Two explanations for the current system in Kituba are possible. The 
first is that the language is moving towards a non-class system where 
singulars are unmarked and plurals are marked by ba-. Furthermore, if it 
is correct that the noun class system is in the process of disintegrating, 
this data would indicate that the change spreads slowly through the lexi- 
con, item by item and not by wholesale generalization of ba- plural marking 
to a given class. 

Finally, since agreement with noun prefixes has been entirely elimi- 
nated, the system has no grammatical function other than to mark singular/ 
plural , 

The alternative proposal would be to claim that the noun classes are 
still present and that the correct interpretation of the unmarked singular 
is that it contains a 0- prefix for those noons. The rest of the forms 
which conform to the Kikongo pairings (Classes 5/6 and 7/8, for example) 
would thai be evidence for the retention of the noun classes. One problem 
with this alternative e:q>lanation is that it does not explain why there 
should be variation in the plural forms. More importantly, it does not 
explain why the variation is consistent. 

The first proposal, on the other hand, explains the variation in the 
plurals. The presence of some nouns which have not changed, this hypothesis 
would claim, is because the process is not yet ccHopleted. Further evidence 
that this proposal is the correct one (i.e. that the noun class system is 
disappearing in favor of an unmarked singular-marked plural system) comes 


18. are/bare 

•(bus) stop' 












' response ' 











from the treatment of loanwords and from the absence of agreement forms, 

2,3 Loanwords 

By far the largest number of loanwords come from French. French is 
widely used in Zaire and has been used since the arrival of the Belgians 
in the 19th century. A large sample (approximately 120 items) is available. 
In every instance but one, the plural form involves the prefixation of ba- 
to the singular form. Here is a partial list of forms used by Salikoko: 

(Ft, arret 'stop^) 
(Ft. banque 'bank') 
(Fr, brique 'brick') 
(Fr. bouteille 'bottle') 
(Fr. dessein 'design') 
(Fr. clerque 'slerk') 
(Fr. reponse 'response') 
(Fr. malade 'sick') 
(Fr, pardon 'excuse me*) 
(Fr. taxi 'taxi') 
(Fr. velo 'bicycle') 
(Ft. voiture 'car') 

The exception noted above is the word for ' bread t dimpa/mampa from the 
French du pain 'some bread.' It is clearly an exception. 

Since alternative loanword adaptation strategies are available for 
Kikongo languages, the choice of this particular one for Kituba lends „ 
support to the claim that an unmarked singular-marked plural is developing. 
In other dialects of Kikongo, the strategy for adapting French loans is 
different. If the first syllable of the French noun resembles one of the 
noun class prefixes, then the noun is assigned to that class regardless of 
whether the noun was singular or plural in French, Thus, the French biscuit 
'cookie' nas become bisikiti 'cookies' and its singular form is kisikiti 
•cookie' where the sequence bi- of biscuit has been interpreted as the 
plural prefix bi-. In Kituba on the other hand, the plural is babisikiti 
'cookies' where the bi- was not interpreted as the Class 8 prefix. Another 
example is the French loan for canife 'knife' which became in some Kikongo 
dialects kanifi/tunifi 'knife' where the ka- of canife was reinterpreted as 
Class 13 ka^. and consequently takes its plural in Class 1^, In Kituba, the 
same noun has been borrowed but it turns up as bakanifi 'knives' in the 
plural. The fact that the preferred strategy for French loans in Kituba is 
that of unmarked singular-marked plural lends credence to the claim that the 
noun class system is no longer very functional in Kituba, 

Z,k Agreement 

Agreement (or concorej as it is usually referrsd to in the Bantu lan- 
guages) has been completely eliminated from the Kituba system. In Kikongo, 
each prefix on the noun had a corresponding concordial prefix (or prefixes). 


often phonetically similar, which marked agreement on several classes of 
words, the main ones being verb, adjective, and the genitive marker. All 
such agreements no longer appear in Kituba. The following examples illus- 
trate the cfflnplete lack of agreement in Kituba verbs. In Kikongo, on the 
other hand, a prefix would appear on the verb marking agreement with a 

19* mwana mene vila 

muteki mene Vila 
miteki/ " 
bamiteki " " 

ndeke mene tina 
bandeke " " 

diboko mene bukana 
maboko " " 

kibulu mene vila 
bibulu " 

avio mene pela tia 
bavio " •• •• 

bilo mene pela tia 
babilo " " 

•the child has been lost' 
'the seller has been lost' 

•the bird has flown away' 
•the arm has broken • 
'the beast has been lost' 
'the airplane has caught fire' 
'the office has caught fire' 

Adjective agreement has also been lost. Normally (in Kikongo) an adjective 
follows the noun and receives an agreement prefix corresponding to the class 
of the head noun. In Kituba, adjectives do not follow the noun directly, 
but are preceded by an associative or genitive marker. Here are some examples! 

20, mwana ya fioti 
bana " " 

muteki ya masonga 
miteki/" " 

ndeke ya kitoko 
bandeke " " 

diboko ya mbl 
maboko " " 

kibulu ya nene 
bibulu " " 

'small child' 
'honest seller' 

•pretty bird' 
'bad arm* 
•big beast' 


avio ya nene 'big airplane' 

bavio " " 

bllo ya fioti 'small office' 

babilo " " 

Note in the examples above that the genitive marker (ya) exhibits no 
evidence of concord either. In Kikongo, on the other hand, there would 
have been sane agreement morphone attached to this marker, marking class. 

Thus the concordial system, by being fully eradicated, contributes 
nothing to the retention of noun classes, Furthermore, the one syntactic 
function assigned to noun classes has entirely disappeared. If the defini- 
tion of a noun class system includes the process of agreement, then the 
Kituba system by definition is no longer a noun class system, 

2.5 Pronominal forms 

The pronominal system does not reflect any difference in noun class 
either. Instead, there is a tendency towards an inanimate/animate distinc- 
tion in the third person. Third person animate pronouns are yandi 'he, 
she, it and b£ 'th^, ' The inanimate pronoun 22. serves both the singular 
and plural functions for inanimates. However, this distinction was one 
already made in the independent pronominal system of Kikongo and does not 
reflect an innovation in Kituba, 

3. Noun Class Systems in Other Bantu Pidgins 

3.1 Reduction of the Noun Class Systems in Other Bantu Pidgins 

Previous attempts at sximmarizing the nature of the reduction of noun 
classes in Bantu pidgins include Pierre Alexandre's article, "Note sur la 
reduction du syst^me des classes dans les langues v^hiculaires "a. fonds ban- 
tu" (1968), parts of Bemd Heine's book Pidgin Sprachen im Bantu Bereich 
(1973). and Eyamba Bokamba's article "The impact of multilingualism on 
language structure 1 the case of central Africa" (1977). 

Alexandre cites several characteristics of the reduction of the system 
of nominal classes based on an examination of Up-Country Swahili, Lingala 
and Pidgin A-70. He notes the retention of independent nominal prefixes to 
show number, the loss of determinatives and concord, with fewer derived 
prefixes (i.e. mass and abstract markers). However, he also notes a replac- 
ing of the grammatical agreement (concord) by a notional one, animate/in- 
animate. All grammatical agreement in Kituba has been lost. As to the 
tendency towards an animate/ inanimate distinction, the only case found in 
the pronominal system existed previously in Kikongo. It is not an innova- 
tion of Kituba. 

Heine notes several other tendancies in data gleaned from Fanagalo, 
Lifigala, several dialects of Swahili, Pidgin A-to, Town Bemba and Kituba. 



In addition to the retention of nuaiber and loss of concord mentioned above, 
Heine mentions a tendency to mark the plural iihile leaving the singular 
unmarked. In Kitnba, this is evident in nouns in the 0-/ba- classes. The 
distinction of number is often lost. An example of this is found in the 
use of absolute pronouns where both 'it' and 'they' are now a single form 
yawtt-yo , Heine also remarks on the loss of class pairs. In Kituba, not 
only were two plurals lost, but a total of five pairs were eliminated in 
favor of 0-/ba-, 

Prcfflu both Alexandre's and Heine's presentation of other Bantu pidgins, 
it appears that Kituba is now very similar to them, but differs in the 
extent to which the system has been rcKiuced, Bokamba (1977) in discussing 
the noun class systems of several dialects of lingala and Swahili notes 
several kinds of simplification. In particular he notes for Kinshasa 
Lingala the prefixation of ba- to already existing plurals. Thus, the 
language Is moving towards a unified plural marking but in a way different 
from Kituba and all other Bantu lingua franoae. 

k. Summary 

The data In the paper show that the noun classes in Kituba have been 
severely reduced. It was argued that the language is moving away from a 
class system towards a system in which the singular is unmarked and the 
plural marked by an existing plural marker ba- in the langusige. In addi- 
tion, it was suggested that the change is proceeding slowly through the 
lexicon by means of various mechanisms. 


*Thl8 paper represents a c<»itlnaation of research begun on this topic 
in 1975 at the University of Kansas. However, the first part of the re- 
search was based on secondary sources rather than on informant work. 
Results of that research were reported on at the 7th African Linguistics 
Conference (Stucky 1976). I would like to thank Professor Frances Ingenann 
of the University of Kansas for her careful criticism of that earlier work. 
Thanks are also due Professors Hans Hock and Charles Kisseberth, and E^araba 
Bokamba at the University of Illinois for their help on this more recent 
research. Any errors are, of course, solely my own responsibility. This 
research was supported, in part, by a NDFL Title VI Fellowship through the 
auspices of the African Studies Center at the University of Illinois. 

Fehderau, in his dissertation "The Origin and Development of Kituba 
(Lingua Ft*anca Kikongo)" (I966), tries to demonstrate that Kituba originated 
as a pidgin arising from a contact situation due to commercial trade and 
slave traffic beginning in the fifteenth century. Although the impetus for 
trade may be attributed to the arrival of the Europeans, largely the Portu- 
guese, Fehderau discredits theories pointing to a linguistic basis in Portu- 
guese, Baglish or French (Fehderau 1966i89-92). A further proposal, that 


Kituba in its various forms arose by independent parallel development from 
local languages, is likewise untenable, he maintains. The structural simi- 
larities are too numerous for all the dialects of Kituba, and the structure 
of the local languages is so different from that of Kituba that they must 
have come from a single source (Fehderau I966193). Fehderau then posits 
as a point of origin, the trading center of Manianga in central Lower 
Zaire, He maintains that a period of creolization evidenced particularly 
in urban centers has extended over the last thirty years. 

The simplification of the noun class system has, however, taken place 
within the last thirty or forty years (See Shepherd's 1926 dictionary, for 
instance). If Kituba is really a pidgin, one wonders why it has taken so 
long for the noun class system to break down. Furthermore, this breakdown 
has taken place during Fehderau 's period of creolization. Since reduction 
is not a characteristic of creolization, while expansion is, the evidence 
from the noun classes mitigates against the classification of Kituba as a 
pidgin. More internal evidence is needed to support Fehderau 's arguments 
which are based largely on external evidence. 


Special thanks are due to Salikoko Mufwene, the consultant. He is a 

native speaker of Kituba, a rarity, I might add. In addition to Kituba, he 
speaks lansi, Lingala, Franch, and English. He reports that he used Kituba 
in non-academic and non-familial social interaction. 

•^elmers (1973) lists the following categories as representative of 
what one finds synchronically in many of the Bantu languages. (The most 
recent attempt to define the semantics of the Proto-Bantu noun classes is 
found in Denny and Creider (I976), 

1/2 most personal nouns 

3/4 trees, plants, other inanimates 

5/6 misc., may be augmentative j nouns of style 

7/8 misc., nay be either augmentative or diminutive; 8 is 
sometimes used adverbially 

9/10 animal names, personal nouns and inanimates 

11 attenuative, abstracts 

12/13 diminutive 

14 abstract nouns 

15 infinitive 
16-18 locative 
19 diminutive 

A much larger vocabulary is available in Swartenbroeckx's (1973) 
Dlctionnaire Kikongo et Kituba - Francais. Unfortunately, it is difficult 
to determine which nouns are exclusively of Kituba origin or in which 
Kituba speaking area the nouns are used. Another, undoubtedly excellent, 
source would be Fehderau 's Kikongo (ya Leta)-francais-anglais , a dictionary 
recently published in Zaire (I969). This was not obtainable. 

In many Bantu languages, the noun prefixes participate in morphologi- 
cal word formations. Whether these processes are still productive in Kituba 
is difficult to ascertain. One would suspect that if the prefix system is 


no longer mat-king noun class that It would be unlikely for the prefixes to 
participate in word formation as well. Given the tremendous difficulties 
in formulating the notion of productive, no attanpt was made to test word- 
formation in Kituba, 

I would like to thank Professor Hock for calling to ny attention 
the possibility of internal motivation for the change in Classes 9/10. 
In addition, Bokamba has pointed out that Classes 9/10 are generally the 
first to in most Bantu lingua francae to adept a plural, 

A framework for the discussion of the allocation of loanwords within 
the African class languages can be found in Heine (I968), Heine proposes 
three different criteria. The first is automatic allocation. All nouns 
are assigned to a given class because they are loanwords. The second is 
phonological allocation. If one of the segments is phonologically similar 
to that of a nominal affix then it is assigned to that class (as in Kikongo), 
Semantic allocation, assigning a noun to a class because of its common 
semantic d<»iain, is the third type. 


Alexandre, P, I967 (I968), "Note sur la reduction des systames do classes." 

Collogue sur la classification nwainale dans les langues negro-afri- 

caines. F'arisi Centre Nationale de la Recherche Scientifique. 

Colloques intemationaux. Sciences Humalnes 30»277-290. 
Bell, A, 1972, "The development of syllabic nasals in the bantu noun 

class prefixes MD-, MI-, and MA-," Anthropological Linguistics 

Bokamba, Eyamba, 1977. The impact of multilingual ism on language stimc- 

turesj the case of central africa. Anthropological linguistics 

Cole, Desmond T. 1971. The history of african linguistics to 19'f5, In 

Cmrent Trends in Liqpiistios. 7ol. 7. edited by Thomas A, Seboek, 

ThB Hague* Mouton, 
Oeauiy, J. Peter, and Chet A. Creider, 1976. The semantics of noun classes 
, In^roto-bantu, Studies in African Linguistics 7«lt 1-30, 
Ellet, Biouard et. al, 1953. Les langues spontan^es dites Sommerciales dtt 

Congo; le raonokotuba comjaaj^ lingala et au lari de la r^ion dn 

Pool, Brazaaville: Editions Victor Sammarro, 
Fehderau, H.W. I966, The Origin and Development of Kituba (Lingua Franca 

Kikongo ) . Doctoral dissertation (I966;, Cornell University, Ithaca, 

New York, 
Fehderau, H.W. I969. Kikonga (ya Leta) . Kinshasai I£CO 
Greenberg, Joseph H, 1966, Universals of Language . Cambridge! The 

M.I,T. Press, 
Guthrie, M, 1967. Comparative Bantu t An Introduction to the Comparative ' 

Linguistics and Prehistory of the Bantu Languages . Famborough (Hants), 

England} Gregg Plress Ltd, 


Heine, B. I968, "The allocation of loan-words within the nominal class 

systems of some togo reomant languages." Journal of African Studies .. 

Heine, B. 1970. Status and Use of African Lingua Franoas . Monchenj 

Weloform Verlag. 
Heine, B. 1973. Pidgin Sprachen ia Bantu Bereioh. Berlin: Dietrich Reiner. 
Laman, K. E. 193^^ Dictlonnaire Kikongo-Francais. Birussels. (Reprinted 

196^, Gregg Press), 
Nida, Eugene A. and Harold W. Fehderau, 1970. Indigenous pidgins and 

koines. UAL 36:1^-155. 
Polome, E. 1968, "Lubumbashi swahili." Journal of African Languages, 

Stucky, Susan U, 1976. Reduction of the noun class system in kituba 

(Lingua Franca Kikongo) . Present at the Seventh African Linguistics 

Conference held at the University of Florida, Gainesville, Florida, 

April 1976. 
Swartenbroeckx, Pierre. 1973* Dictionnaire Kikongo et Kituba-Francais . 

Bandundut CEEBA. 
Swift, L. B. and Zola, E. W. A. I963. Kituba Basic Course . Washin^on, 

D. C. Foreign Service Institute. 
Welmers, W. 1973. African Language Structures . Berkeley! Iftiiversity of 

California Press. 

Forthcoming Special Issues 

Fall 1978 Vol. 8, No. 2, Linguistics in the Seventies: 
Directions and Prospects 
(Forum Lectures delivered at 
the 1978 Linguistic Institute 
of the Linguistic Society of America) 
Editor: Braj B. Kachru $10.00 

Fall 1979 Vol. 9, No. 2, Relational Grammar and Semantics 

Editor: Jerry L. Morgan $5.00 

Fall 1980 Vol. 10, No. 2, Papers on Diachronic Syntax 

Editor: Hans Henrich Hock $5.00 

Fall 1981 Vol. 1 1, No. 2, Studies in Language Variation: 
Nonwestern Case Studies 
Editor: Braj B. Kachru $5.00 

Fall 1982 Vol. 12, No. 2, South Asian Linguistics: Syntax 
and Semantics 
Editor: Yamuna Kachru $5.00 


The following issues of SLS are out of print: 
Spring 1971, Vol. 1, No. 1, Papers in General Linguistics; Fall 1971, Vol. 1, 
No. 2, Papers on Hindi Syntax, Editor: Yamuna Kachru; Spring 1972, Vol. 2, 
No. 1, Papers on Syntax and Semantics, Editor: Georgia M. Green; Fall 1972, 
Vol. 2, No. 2, Studies in Baltic Linguistics, Editors: Hans H. Hock and 
Michael J. Kenstowicz; Spring 1973, Vol. 3, No. 1, Papers in General 
Linguistics; Fall 1973, Vol. 3, No. 2, Papers on South Asian Linguistics, 
Editor: Braj B. Kachru: and Spring 1976, Vol. 6, No. 1, Papers in General 
Linguistics, Special section: Topics in relational grammar. Editor: Jerry L. 

The following issues are available: 
Spring 1974 Vol. 4, No. 1, Papers in General Linguistics 


Fall 1974 Vol. 4, No. 2, Papers on Phonetics and Phonology 
Editors: Charles W. Kisseberth and 
Chin-W. Kim 


Spring 1975 Vol. 5, No. 1 , Papers in General Linguistics 


Fall 1975 Vol. 5, No. 2, Papers on Historical Linguistics: Theory and Methoc 
Editors: Ladislav Zgusta and 
Hans H. Hock $3.50 

Fall 1976 Vol. 6, No. 2, Papers on African Linguistics 

Editors: Eyamba G. Bokamba and 

Charles W. Kisseberth $3.50 

Spring 1977 Vol. 7, No. 1 , Papers in General Linguistics $3.50 

Fall 1977 Vol. 7, No. 2, Studies in East Asian Linguistics 
Editors: Chin-chuan Cheng and 
Chin-W. Kim 

Spring 197S Vol. 8, No. \, Papers in General Linguistics 


Orders should be sent to: 

SLS Subscriptions, Department of Linguistics 

4088 Foreign Languages Building 

University of Illinois 

Urbana, Illinois 61801 

Linguistics in the Seventies: 
Directions and Prospects 

Language Change 

William S-Y. Wang Paul Kiparsky 


Charles A. Ferguson Roger W. Shuy 

Functional Phonologx; 

Morris Halle Jonathan Kaye 

Theories of Meaning 

John Searle Edward Keenan 

Language Uniuersals 

James McCawley Bernard Comrie 

Language Acquisition 

Susan Ervin-Tripp Neilson V. Smith 

Collitz Lecture 

George Cardona 

Linguistic Societi; of America Lecture 

Charles Osgood 

Edited by 
Braj B. Kachru 

Special Issue of 

Studies in the Linguistic Sciences 

Department of Linguistics 

University of Illinois 



EDITORS: Charles W. Kisseberth, Braj B. Kachru, Jerry L. Morgan 

REVIEW EDITORS: Robert N. Kantor and Ladislav Zgusta 

EDITORIAL BOARD: Eyamba G. Bokamba, Chin-chuan Cheng, Peter 
Cole, Georgia M. Green, Hans Henrich Hock, Yamuna Kachru, Henry 
Kahane, Michael J. Kenstowicz, Chin-W. Kim and Howard Maclay. 

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. Especially invited 
papers by scholars not associated with the University of Illinois will also be in- 

SPECIAL ISSUES: Since its inception SLS has been devoting one issue each 
year to restricted, specialized topics. A complete list of such special issues is 
given on the back cover. The following special issues are under preparation: 
Fall 1979: Relational Grammar and Semantics, edited by Jerry L. Morgan; 
Studies in Arabic Linguistics, edited by Michael J. Kenstowicz; Studies in 
Language Variation: Non western Case Studies, edited by Braj B. Kachru; 
South Asian Linguistics: Syntax and Semantics, edited by Yamuna Kachru. 

BOOKS FOR REVIEW: Review copies of books (in duplicate) may be sent to 
the Review Editors, Studies in the Linguistic Sciences, Department of 
Linguistics, University of Illinois, Urbana, Illinois, 61801. 

SUBSCRIPTION: There will be two issues during the academic year. Requests 
for subscriptions should be addressed to SLS Subscriptions, Department of 
Linguistics, 4088 Foreign Languages Building, University of Illinois, Urbana, 
Illinois, 61801. 

Price: $5.00 (per issue) 



Braj B. Kachru 


FALL, 1978 





GEORGE CARDONA: Relations between causatives and passives in 

Indo-Iranian 1 


CHARLES OSGOOD: Conservative words and radical sentences 

in the semantics of international politics A3 


WILLIAM S-Y. WANG: The three scales of diachrony 63 

PAUL KIPARSKY: Analogical change as a problem for linguistic 

theory 77 


CHARLES A. FERGUSON: Multilingualism as object of linguistic 

description 97 

ROGER W. SHUY: Multilingualism as a goal of educational policy . . . 1^7 


MORRIS HALLE: Formal vs. functional considerations in phonology. . . 123 
JONATHAN ICAYE: Functionalism and functional explanations in 

phonology 135 


JOHN SEARLE: Intentionality and the use of language 149 

EDWARD KEENAN: On surface form and logical form 163 


JAMES McCAWLEY: Language universals in linguistic argumentation. . . 205 
BERNARD COMRIE: Linguistics is about languages 221 


SUSAN ERVIN- TRIPP: Whatever happened to communicative competence?. . 237 
N. V. SMITH: Lexical representation and the acquisition of 

phonology 259 



The present volume of Studies in the Linguistic Sciences (SLS) is 
devoted to the Forum Lecture Series presented at the 1978 Linguistic 
Institute of the Linguistic Society of America, hosted by the University 
of Illinois from June 12 to August 5, 1978. 

At the outset, as the director of the 1978 Institute, I felt that 
the Forum Lecture Series should focus on a few selected topics which 
would present "the state of the art" in our discipline during the 1970s. 
I was pleased to find that the Organizing Committee for the Forum 
Lectures agreed with this proposal. We therefore selected, after long 
and serious consideration, the following seven areas as the topics for 
the lectures: language change, multilingualism, functional phonology, 
theories of meaning, language universals, language acquisition, and neuro- 
linguistics. The idea behind this structure was to provide a synthesis 
of contemporary approaches to the seven most debated areas in the lin- 
guistic sciences. 

The papers included here present such an overview. And, as Henry 
Kahane, Chairman of the Organizing Committee for the Forum Lectures, said 
while introducing the Series, we wanted to "view each of our seven themes 
with two sets of eyes, through the medium of two intellects." That 
explains why we invited two distinguished scholars to discuss each topic. 
But as we know, in our field — perhaps as in other fields — an overview 
generally results in a presentation of one's own view. If that sub- 
jectivism is present in any of these "state of the art" papers, it is 
unavoidable, and we did not discourage it. 

This volume, however, does not include all of the fourteen Forum 
Lectures. The two papers on neurolinguistics presented by Harry Whitaker 
(University of Rochester) and Grace Yeni-Komshian (University of Maryland) 
could not be included since the authors had other plans for publication. 


In addition to the Forum Lectures, this volume includes the Herman 
Collitz Lecture and the Linguistic Society of America Professor's 
Lecture. The Collitz Professorship was held by George Cardona (Univer- 
sity of Pennsylvania) and the Linguistic Society of America Chair by 
Charles Osgood (University of Illinois) . 

There are several agencies and a number of individuals to whom we 
owe thanks and gratitude for making the Forum Lectures a success, and 
making it possible to publish them in book form. The very successful 
and stimulating Lecture Series would not, of course, have been possible 
without the cooperation of the distinguished scholars who took part. We 
are indebted to the Forum Lectures Organizing Committee, who planned the 
Series meticulously, giving due consideration to every subfield of our 
discipline. The Committee consisted of William 0. Dingwall, Georgia M. 
Green, Braj B. Kachru, Henry Kahane (Chairman), and Michael J. Kenstowicz. 
Josephine L. Wilcock, a moving spirit behind every undertaking on which 
our department embarks, took care of every minute detail. In this she 
received excellent help from Marjorie S. Olson, Secretary of the 1978 
Linguistic Institute. Hans Henrich Hock, Assistant Director of the 
Institute, provided valuable advice for coordinating the lectures with 
the Institute's heavy academic and social schedule, and in innumerable 
other ways. William D, Wallace assisted in the editing of the papers. 
The support of the Division of Applied Linguistics, the Office of 
International Programs and Studies, and the Department of Linguistics 
at UI-UC, made it possible to publish this volume. 

Braj B. Kachru 

October 1, 1979 



George Cardona 
University of Pennsylvania 

The Sanskrit of Plnini's time (ca. 500 B.C.), described in his 
Astadhyayi, has a fully developed system of causative verbs. In Panini's 
description, causative bases are derived from primitive ones with the 
affix Nz'C {i with two markers), which appears in the form au in certain 
contexts. For example: 

gam 'go' (3 sg. pres. gaochati) : gam-i (gamayati) 'cause to go' 

budh (budhyate) 'perceive' : bodh-i {bodhayati) 

bhuj (bhufikte) 'eat' : bhoj'-i ibhojayati) 

path ipathati) 'recite, study' : path-i (pathayati) 

as (aste) 'be seated' : as-i (asaya'ti) 

hr (havati) 'take, carry' : har-i _{hd.rayati) 

kr (karoti) 'do, make' : kar-i (karayati) 

p'ac ipacati) 'cook' : pao-i {pacayati) 

Since the Sanskrit Panini describes corresponds closely to the lang- 
uage of Brahmana and sutra'texts (Cardona 1976a:238), this means the full 
causative system had developed by late Vedic times. Thieme (1929:17-23) 
argued that early Vedic lacked true causatives. According to him, the 
derived bases usually called causatives served originally only as transi- 
tives corresponding to intransitives (including passives). Kuryiowicz 
(1956:88-9, 1964:87) accepted Thieme's thesis in his own attempt to recon- 
struct the prehistory of this formation. More recently, Jamison (1976) 
has once more reiterated this position. I shall argue that this generally 
held view does not withstand scrutiny, hence must be abandoned. 

1.0 Let us begin with causatives in the Sanskrit described by Panini, 

1.1 The following examples illustrate syntactic features concerning 
these causatives. 


(1) a. gaochati manavako gvamam 'The boy is going to the village.' 
b. devadatto manavakam grSmarn gamayati 'Devadatta has the boy go 

to the village. ' 

(2) a. budhyate putro dharmam 'The son perceives his duty.' 

b. pita putram dharmam bodhayati 'The father makes his son 
perceive his duty.' 

(3) a. devadatta odanam bhuhkte 'Devadatta is eating rice.' 

b. yajnadatto devadattam odanam bhojayati 'Yajnadatta feeds 
Devadatta rice. ' 

(4) a. Hsyo vedam pathati 'A student is reciting the Veda.' 

b. adhyapdkah sisyam vedam pathayati 'The teacher is having his 
student recite the Veda.' 

(5) a. devadattdh kata aste 'Devadatta is seated on a mat.' 

b. devadattam kata asayati "... has Devadatta sit on a mat.' 


(6) a. harati bharam manavakah 'The boy is carrying a load.' 

b. lidrayati bharam manavakam '. . .is having the boy carry a 
load. ' 

c. harayati bharam manavakena (idem) 

(7) a. devadattah katam karoti 'Devadatta is making a mat.' 

b. devadattam katam karayati '. . .is having Devadatta make a 
mat. ' 

c. devadattena katam Karayati (idem) 


(8) a. odanam pacati devadattah 'Devadatta is cooking rice.' 

odanam paoayati devadattena ' . . .is having Devadatta cook 
rice. ' 

The non-causal sentence (la) has the verb gam together with the nomi- 
native manavakah/- and the accusative gvamam, respectively denoting the 
agent and goal of going. The causal sentence (lb) has the causative gam-i 
in construction with gramam, which again refers to the goal of going, and 
a second accusative: manavakam, which refers to the agent of going, now 
the object of causation. Similarly, (2a), (3a), (4a), (5a) have nominative 
forms that refer to agents of perceiving, eating, reciting, sitting: 
putrah 'son', devadatta!]. 'Devadatta', sisyah 'student', devadattah. In 
addition, (2a), (3a), (4a) contain accusative forms used with reference to 
objects of actions: dharmam 'duty', odanam 'rice', vedam 'Veda'. The 
verb as of (5a) occurs with the locative kate 'mat', denoting the locus 
where Devadatta is seated. To the nominative forms of all these correspond 
accusatives in the related causative sentences: (2b) putram, (3b) devadat- 
tam, (4b) sisyam, (5b) devadattam. Note also that (1) - (4) involve verbs 
of specific semantic groups: movement, perception, consuming, and verbs 
such as path, which take objects that are speech elements, things said or 

(6a) and (7a) are like the non-causal sentences of (1) - (4) in that 
they too contain verbs (Jir, kc ) used with nominative forms referring to 
agents (manavakah, devadattah) and accusatives referring to objects (bharam, 
katam). Moreover, (6b), (7b) resemble the causative sentences of (1) - 
(4) in that they too have causative verbs construed with two accusatives. 
However, (6c), (7c) have no counterparts in (1) - (4). To the nominatives 
in the non-causal sentences (6a), (7a), correspond here instrumental forms 
(manavakena, devadattena) instead of accusatives, as in (6b), (7b). The 
same is true of (8b), which has the instrumental devadattena corresponding 
to the nominative devadattah of (8a) . 

Instrumental forms are also used in passive sentences related to the 
non-causal sentences (6a) - (8a) : 

(9) manavakena bharo hviyate 'A load is being carried by the boy.' 

(10) devadattena katah kriyate 'A mat is being made by Devadatta.' 

(11) odanah paayate devadattena 'Rice is being cooked by Devadatta.' 

The primitive verbs of (1) - (4) also enter into passive constructions, 
and these too have instrumental forms denoting agents. The passive sen- 
tences corresponding to (la) - (4a) are: 

(12) manavakena gramo gamyate 

(13) putrena dharmo budhyate 

(14) odano bhujyate devadattena 

(15) sisyena vedah pathyate 

in which the singular passive forms gamyate (gam-ya-te) , budhyate, bhujyate, 
pathyate agree in number with the nominative singular forms gramah, dharmah, 
odanah, vedah. 

There are thus two types of causative constructions. One is a double 
accusative construction: a causative verb occurs with an accusative form 
referring to the goal, object of the act denoted by the primitive verb and 
another referring to the agent of this act, the object of causation. In the 
second construction, an instrumental form is used, as in a passive sentence, 
with reference to the agent of the act denoted by the primitive verb. The 
double accusative construction is regular with causative verbs derived from 
primitive verbs of particular semantic groups, as shown in (1) - (4). The 
second construction is regular with causatives formed to other transitive 
verbs. In addition, as shown in (b) , two verbs have causatives that are 
used in either construction. Finally, causatives formed to intransitive 
verbs such as as follow the pattern shown in (A), with a single accusative 
referring to the agent of the primitive verb, object of causation. 

1.2 I have thus far spoken of transitive and intransitive verbs with- 
out specifying criteria whereby given verbs in given contexts may be so 
classed. There are straightforward criteria for determining in Sanskrit 
whether or not a verb is transitive, and it will be useful to consider these. 

It can be seen from (1) - (4) that the primitive verbs gam, budh, bhuj , 
path can take accusative complements, as can also hr, kr of (6), (7) and 
pac of (8). Moreover, these verbs are used in true passive constructions: 
(9) - (15). Similarly, to 

(16) manavako gramam agamat 'The boy went to the village.' 

(17) abuddha putro dharmam 'The son perceived his duty.' 

(18) devadatta odanam dbhukta 'Devadatta ate rice.' 

(19) sisyo vedam apathTt 'A student recited the Veda.' 

(20) manavako bharam ahdrsTt 'The boy carried a load.' 

(21) devadattah katam akdrsTt 'Devadatta made a mat.' 

(22) devadatta odanam apaksTt 'Devadatta cooked rice.' 

with the preterital (third singular aorist) forms agamat, abuddha, dbhukta, 
apathTt, ahdrsTt, akdrsTt, apdkszt, correspond the passive sentences 

(23) agami mdnavakena grdmah 

(24) abodhi putrena dharmah 

(25) abhojy odano devadattena 

(26) apdthi vedah sisyena 

(27) ahdri bhdro mdnavakena 

(28) akdri devadattena katah 

(29) apdcy odano devadattena 

These contain the third singular passive aorist forms agami, abodhi, 
abhoji, apdthi, ahdri, akdri, apdai, which agree in number with nominative 
forms denoting objects, just as in (9) - (15). In the same manner, the 
participial forms gatah 'gone to', buddhah 'perceived', bhuktah 'eaten', 
pathitah 'recited', hrtah 'carried', krtah 'made', pakvah 'cooked' in 

(30) mdnavakena grdmo gatah 

(31) putrena dharmo buddhah 

(32) odano bhukto devadattena 

(33) sisyena vedah pathitah 

(34) mdnavakena bhdro hrtah 

(35) devadattena katah krtah 

(36) odanah pakvo devadattena 

agree in number with nominatives grdmah etc. Again in 

(37) devadattena kalxiv akrsdtdm 'Two mats were made by Devadatta'. 

(38) devadattena katd akrsata 'Several mats were made by Devadatta'. 

the aorist forms akrsatam (3 du. mid.), akrsata (3 pi. mid.), used pas- 
sively, agree in number with the nominatives katau (du.), katah (pi.), 
as do the participles krtau, krtah of 

(39) devadattena katau krtau 

(40) devadattena katah krtTUi 

Such participles also agree in gender with nominatives referring to objects; 
gatah etc. in (30) - (36), (39) - (40) are masculine, as are gramah etc., 
but krta of 

(41) devadattena stKdlT krta 'A pot has been made by Devadatta.' 

is feminine, in agreement with sthalT. 

Let us agree that a verb which occurs in constructions as shown is 
transitive. A verb such as as, which does not occur in these constructions, 
is then intransitive. Thus, of itself this verb does not occur with object 
accusative complements.-^ It does form a passive, as in 

(42) devadattena kata asyate 

the passive counterpart to (5a) , but asyate fails to show number agreement 
in sentences such as 

(43) purusabhyarn kata asyate 

(44) purusaih kata asyate 
corresponding to 

(45) purusau kata asate 

(46) purusah kata asate 
Similarly, to 

(47) devadatta kata asista 

(48) purusau kata asisatam 

(49) purusah kata asisqta 
correspond passive sentences 

(50) devadattena kata asi 

(51) purusabhyarn kata asi 

(52) purusaih kata asi 

all with the third singular form asi, and 

(53) devadattena kata asitcon 

(54) purusabhyam kata asitam 

(55) purusaih kata asitam 

have the neuter singular participial form asitam. 

That a verb such as pao in (8a), (11), (22), (29), (36) is transitive 
does not mean active forms like paaati must always occur with an object 
accusative complement or indeed that pao must always be transitive. Con- 
sider, for example, 

(56) odandh pacyate 'The rice is cooking,' 

(57) odanena paoyate [passive equivalent of (56)] 

(58) apacy odandh 'The rice cooked.' 

(59) apacy odanena [passive equivalent of (58)] 

(60) odanena pakvam [alternative to (59)] 

A condition for the use of (56) as opposed to (11) is that an agent perform- 
ing the act of cooking on rice not be mentioned. Moreover, since (57), (59), 
(60) are formally like (42), (50), (53). one can rightly consider that pac 
of the set (56) - (60) is intransitive.^ Nevertheless, the fact remains 
that examples of this type regularly have specific verbs which are other- 
wise transitive, namely those denoting actions objects in which are produced 
(e.g. kr^ 'make') or undergo a change of state (e.g. pao 'cook', hhid 
'split, break', lu 'cut', muc 'loose'). Verbs like as, which are regularly 
intransitive, do not enter into this construction. 

Also excluded from constructions of the type (56) - (60) are verbs of 
movement such as gam. These verbs have, in addition, other syntactic fea- 
tures which set them apart. As shown in (la), the goal of going can be 
signified by an accusative like gramam. This can also be signified by a 
dative form such as gramdya:^ 

(61) gaochati mdnavako grdmdya 

In (30) the participle gatah is coreferential with grdmah, both used with 
reference to the goal of going. However, gata 'gone' can also be agentive, 
referring to an agent of going, as in 

(62) mdnavako gramam gatah 'The boy went to the village.' 
where gramam denotes a goal. 

The verb smr 'remember, yearn for' behaves in general like budh. Thus, 

(63) mdtaram smarati devadattah 'Devadatta remembers his mother with 
longing. ' 

has the accusative matavam 'mother', and the passive 

(64) niata smavyate devadattena 

contains the nominative niata, with which the passive verb smavyate agrees 
in number. However, something one yearns for can also be denoted by a 
genitive in construction with smv, as in 

(65) matuh smarati devadattah 

with the genitive matuh. ^ Moreover, (65) has a passive counterpart 

(66) matuh smaryate devadattena 

of the type (42): there is no number agreement between smaryate and the 
genitive. If one substituted the dual svasdrau 'sisters' or the plural 
svasdrah for mdta in (64) , one would also have to substitute the dual and 
plural verb forms smaryete, smaryante for the singular smaryate, but sub- 
stituting svasroh (gen. du.) or svasrnam (gen. pi.) for matuh in (66) 
entails no change in the verb smaryate. Further, since the type (42), in 
which a passive verb shows no number variation, is typically intransitive, 
one may rightly consider that smv is intransitive in (65), (66).' This 
does not alter the fact that smr in (63), (64) is transitive. Consequently, 
it is altogether proper to say that the causative smav-i of 

(67) matavam smavayati devadattam 

(68) matuh smavayati devadattam 

is derived from both an intransitive and a transitive base verb. For (67) 
and (68) correspond to (63) and (65), respectively. 

1.3 The causative constructions spoken of can be reduced to the fol- 
lowing schemata, together with related non-causal and passive present 

(69) a. V Ni-nom. N2-x8 [non-causal] 

b. V-pass. Ni-instr. N2-nom./x9 [passive] 

c. V-i^ N]^-acc. N2-X N3IO [causative] 

(70) a. V N]^-nom. N2-acc. [non-causal] 

b. V-pass. N]^-instr. N2-nom. [passive] 

c. V-^ N^-instr. N2-acc. N3 [causative] 

(69c) is the norm for transitive verbs of group (A) and intransitive verbs, 
(70c) for transitive verbs other than those of (A) , and the two construc- 
tions are alternatives with causatives formed to the two verbs of (B) . 

1.4 Though causatives with the affix i {ay) are freely derived from 
primitive verbs, it is nevertheless not the case that all bases derived 
with this affix have only causative value in the Sanskrit of Panini's time. 
For example, there are non-causative derivates in -i such as: 

ava-i (aroayati) 'praise' 

garh-i (garhayati) 'revile, deprecate' 

ohdd-i (ahadayati) 'cover' 

dhars-i (dharsayati) 'dare' 

para-i (paraayati) 'blend' 

mars-i (jnavsayati) 'tolerate' 

rmrj-i (marjayati) 'cleanse' 

yoj'-i (yojayati) 'join' 

vavj-i (varjayati ) ' exc lude ' 

vdr-i (vdrayati) 'keep from, ward off from' 

srdth-i (srathayati) 'loose' 

sah-i (sahayati) 'endure' 

It is true that most such bases are not directly cited in the AstadhyayT 
but are known from recensions of the dhatupatha (catalog of verbs) which 
date from centuries after Panini.H Nevertheless, it is certain that 
Panini indeed used such derivates without causative value. His rule 
1.4.27: varanarthanam Tpsitah (sampradanam 24) provides that with respect 
to acts denoted by verbs meaning 'keep from' (varanartha) , that which 
someone wishes to reach (tpsita) is classed as sampradana. The purpose of 
this classification is to let a rule apply which introduces an ablative 
ending after a nominal when a sampradana is to be signified. For example, 

(71) agner manavakam vdrayati "... keeps the boy from the fire. ' 

(72) kupdd andham varayati '. . . keeps the blind man from the well.' 

have the ablatives agneh 'fire', kupat 'well'. The action noun varana of 
vdranartha in Panini 1.4.27 must be derived from var-i, showing that 
Panini knew and used such non-causative derived verbs. 

1.5 It is important to my subsequent discussion briefly to summarize 
the general pattern of uses for atmanepada and parasmaipada affixes — 
medio-passive and active affixes — in the Sanskrit described by Panini. 
Passives, whether of the type shown in (9) - (15) or of the type illustra- 
ted by (42) - (44), have atmanepada affixes, and endings such as te are 
regularly preceded by the affix yaK.^^ So far as concerns verbs with 
agentive endings, these are of three general types. 

Some verbs take only atmanepada affixes; for example, the intransitive 
verb as (dste , [5a]) and the transitive verb bhuj {bhw'ikte , [3a]). 13 

Other verbs take either parasmaipada or atmanepada affixes, depending 
on a semantic contrast. If the result of the act signified by the base is 
intended for the agent, atmanepada affixes occur; for example, (7a), (8a) 
are used if Devadatta is making a mat for someone else or cooking rice for 
someone else to eat; if he were making a mat or cooking rice for himself, 
one would use kurute , paeate A^ 

Still other verbs regularly take parasmaipada affixes; for example, 
the intransitive verb sad (stdati) and the transitive verb path (pathati , 
[4a]). 15 

Causative bases generally follow the second pattern: karayati is 
used, as in (7b, c), if the causal agent causes something to be made that 
is not for himself, but karayate is used if this is intended for the causal 
agent. 16 However, atmanepada affixes are regularly used in reflexive 
causatives. For example, corresponding to 

(73) bhrtya rajanarn pasyanti 'The king's subjects see him.' 
there is a causal sentence 

(74) raja bhrtyan darsayate 'The king shows himself to his subjects. ' 

in which the causative darS-i 'cause to see' is followed by the atmanepada 
ending te and construed with the accusative bhrtyan 'subjects. '1^ 

2.0 With this brief summary of causative and related constructions in 
the Sanskrit of about 500 B.C. as a background, let us now consider earlier 
Vedic usage. For the syntax of causatives in Panini's time certainly has 

a prehistory and reflects changes that occurred in a living language. 
Moreover, as I have noted, semantic differences play a role in the syntax 
of causatives in the Sanskrit of Panini's time, so that I shall consider 
such differences also in connection with causatives of earlier Vedic. 

2.1 A goal one reaches is signified by an accusative or a locative in 
construction with a verb of movement. For example: 

(75) RV 1.145.3ab: torn id gacahanti juhvas tarn arvatzr visvany ekah 
srmavad vaoamsi me 'To him (Agni) go the ladles (i.e. oblations), to 
him the speedy hymns; he alone will hear all my utterances.' 

(76) RV 1.1.4: agne yarn yajnam . . . visvatah paribhur asi/ so. id 
devSsu gacohati 'Agni, that sacrificial offering which you encompass 
from all sides goes to the gods.' 

In (75), gacahanti (3 pi.) takes the accusative complement torn 'him', and 
in (76), gacchati (3 sg.) is construed with the locative devisu 'gods'. 
Both these complements refer to goals. The causative gam-i is also used 
with an accusative or a locative denoting a goal; for example: 

(77) RV 10.145.4cd: param eva paravatam sapdtntm gamayamasi 'We make 
the cowife go to the very farthest distance.' 

(78) RV 5.5.10c: tatra havyani gamaya 'Make the oblations go there.' 

The accusative param . . . paravatam in (77) refers to where the wives make 
a cowife go, the farthest distance, and the locatival tdtra 'there' of 
(78) denotes where oblations are supposed to be made to go. 

The goal one attains or is made to attain can also be an abstract 
property, a state, and gam-i is used with an accusative referring to such 
a state. For example, in 

(79) TS so. evainam bhutim gamayati 'He (Vayu) makes him 


(the sacrificer) attain prosperity.' 

gamayati occurs with bhutim 'prosperity'. The same is said in TS, 
where it is Indra who causes one to attain prosperity. In a similar 

(80) MS 2.5.4. (p. 51.15): yathaisa sviyam asnuta evcan evainam sriyam 
gamayati 'As he attains wealth, thus does he cause him to attain 
wealth. ' 

gamayati is used with sriyam 'wealth'. A parallel passage, 

(81) KS 13.6 (p. 188.7-8): yathaiva sa sriyam prapnoty evam enojn} 
sriyam prapayati 

has the causative prap-i {prapayati) 'make to attain' with sriyam, which 
is also construed with the non-causative prap 'attain' (prapnoti) . 

Although in early Vedic verbs of movement such as gam usually occur 
with accusative or locative forms denoting goals, the construction type 
(61), with a dative referring to a goal, is also used. For example, 

(82) AV 16.6.4: yam dvismo yds aa no dvesti tdsmd enad gamayamah 
'We make this (bad dream) go to him whom we hate and who hates us.' 

has the causative gamayamah construed with the accusative enad 'this' and 
the dative tdsmai 'to him', the latter referring to that person to whom a 
bad dream is made to go. This causative construction implies a non-causal 
sentence in which the goal is also denoted by a dative such as tdsmai .'^^ 

2.2 In (75), the subjunctive srnavat 'will hear' occurs with the 
accusative vaaamsi 'utterances, hymns'. A verb of perception such as sru 
'hear, listen' is also used in construction with a genitive (cf. [65]). 
In particular, sru commonly occurs with a genitive denoting a person to 
whom one listens. For example, 

(83) RV 4.22.10a: asmakam it su srnuhi tvdm indra 'Indra, listen 
well to us. ' 

has the imperative srnuhi construed with the genitive asmakam 'us'. " The 
verb ait 'perceive, know' also is used with either an accusative or a 
genitive denoting what is perceived. For example, 

(84) RV 1.10.2c: tdd indro drtham oetati 'Then Indra perceives the 
purpose. ' 

has the accusative drtham 'purpose' with aetati , but in 

(85) RV 1.2.5ab: vdyav indras ca cetathdh sutdndm . . . 'Vayu, you 
and Indra perceive the Soma juices that have been pressed.' 

cetathah (2du.) occurs with sutdnam (gen. pi.) 'pressed Soma juices'. 


Other verbs of perception usually occur with accusative complements. 
This is true of drs 'look, see', as in 

(86) RV 10. 71. 4a b: utd tvah pdsyan na dadarsa vaoam utd tvah hrnvan 
na srnoty eriam 'The one, while looking, has not seen speech, the 
other, while listening, does not hear her.' 

where the perfect dadarsa is construed with the accusative vacam 'speech'. 
Similarly, tks (tksate) 'look, see' takes an accusative complement, as in 

(87) VS 5.34a: mitrdsya rrid cdksuseksadhvam 'Look at me with the eye 
of Mitra. ' 

where the imperative Tksadhvcav (2 pi.) occurs with the accusative wa 'me'. 
Causatives are formed to all the verbs mentioned: srav-i 'make to hear, 
to listen', cet-i 'make to perceive', tks-i 'make to see'. In 

(88) RV 4.29.3a: sravayed asya kcomd vajayddhyai 'Make his ears 
hear, so that treasure (booty) be won.' 

the imperative sravdya (2 sg.) is used with one accusative: a chanter is 
told to make Indra's ears (karma) hear. What is supposed to be heard is 
not specified, but this is known from context. The previous verse, 

(89) RV 4.29.2ab: a hi srrid yati ndryas cikitvan huyamdnah sotrhhir 
upa yajndm '(Indra,) manly, knowing, comes to the sacrificial offer- 
ing when he is called by the Soma pressers.' 

speaks of Indra coming to the sacrifice when he is called {huyamdnah 'being 
called'), and Indra, characterized as having attentive ears (dsrutkaima) , 
is elsewhere asked to hear the call of those who call him: 

(90) RV 1.10.9a: asrutkarna srudht hdvam 'Attentive-eared (Indra,) 
hear my call. ' 

The imperfect dcetayat (3 sg.) is used with the accusative aaztah 'who 
do not perceive, imperceptive' in 

(91) RV 7.86.7c: dcetayad acito devo arydh 'The noble god (Varuna) 
made the imperceptive to perceive.' 

Gramatically, acit is an agent noun that refers to one who does not per- 
ceive {na oetati) . The causative cet-i is thus related to the active 
eetatt. 20 

In addition, the causative imperatives darsdya, sdmtksayasva (2 sg.) 
'make to see' are found in double accusative constructions: 

(92) AV 4.20.6ab: darsdya md ydtudhdndn darsdya ydtudhdnydh 'Let 
me see the warlocks, let me see the witches.' 

(93) AV 4.15.3a: sdmtksayasva gdyato ndbharnsi 'Let (us) who are 
chanting see the clouds.' 


The accusative plural forms yatudhandn 'warlocks', yatudhanyah 'witches', 
ndbhamsi 'rain clouds' refer to objects of seeing, and rria 'me', gayatah 
'those chanting' signify objects of causation, those caused to see these. 

2.3 The verb pa (3 sg. pres. pibati) 'drink' is used with an accusa- 
tive or a genitive denoting what is drunk. For example, 

(94) RV 1.15.1a: indva somam piba rturia 'Indra, drink the Soma in 
due time. ' 

(95) RV 1.4. 2b: somasya somapah piba '(Indra,) drinker of Soma, 
drink the Soma. ' 

both have the imperative piba. Indra is asked to drink Soma juice, and 
this is referred to by the accusative somam or the genitive somasya. The 
causative pay-i (2 sg. imper. payaya) takes two complements: an accusative 
(e.g. madhuni 'sweet drinks') or a genitive (e.g. gen. sg. madhvah 'sweet 
drink') denoting what is drunk and an accusative (e.g. ydjatran 'those 
worthy of awe-inspired worship'), as in 

(96) RV 3.57.5cd: . . . ihd visvam dvase ydjatran a sadaya payaya 
da. madhuni '(Agni,) have all the worshipful ones sit here and drink 
the sweet drinks.' 

(97) RV 1.14.7: tan ydjatram . . . madhvah sugihva payaya 'Wonderful- 
tongue (Agni,) have the worshipful ones drink the sweet drink.' 

2.4 The verb vao (vivakti, vakti) 'speak, say, utter' occurs with an 
accusative denoting what is spoken, said, uttered, or said to be and either 
a dative or an accusative signifying one to whom something is said. For 

(98) RV 1.74.1b: mdntram vocemagndye 'We would utter a mantra to Agni . ' 

(99) RV 10.80.7b: agnim maham avocama suvrktim 'We have uttered a 
great hymn to Agni.' 

In (98), the optative vocema occurs with the accusative mdntram (sacred 
formula) and the dative agndye 'to Agni', but the aorist avocama of (99) 
is construed with two accusatives: suvrktim 'hymn' and agnim. 

The causative vdc-i is common in Brahmana and srautasutra passages, 
and this can occur with more than one accusative. For example, 

(100) TS ta yd.jamanam vdoayati '(The Adhvaryu priest) has 
the patron of the rite say (formulae) to them (the deities) . ' 

has vaoayati construed with ydjamanam, referring to the person made to 
utter formulae, and tdh 'them', which refers to those to whom these are 

2.5 In (96), sdd-i (a sadaya with the preverb a) 'have . . . sit' 
occurs with the accusative ydjatran and the locatival ihd. The base sad 


'sit', from which sad-i derives, is also used in this way, as in 

(101) RV 1.188.6c: usasav ehd stdatam 'May the two dawns (i.e. Dawn 
and Night) sit here.' 

where iha is used with a . . . stdatam (3 du. imper.). 

2.6 As was shown in 2.2, sru and ait with active inflexion are used 
transitively. There are also middle forms of these verbs used intransi- 
tively or with passive value. Middle forms of sim have the senses 'be 
heard' and 'be heard of, be famous', as in 

(102) RV 1.74.7: no. yor upabdir asyvah srnve rathasya kaaaand/ yad 
ague yasi dutyam 'No hoofbeat sound of horses (as they pull) a 
moving chariot is ever heard, Agni, when you go on your mission (to 
the gods) . ' 

(103) RV 8.6.14c: vrsa hy ugra srnvise 'For, mighty (Indra,) you 
are famous as a bull.' 

where srnve 'is heard' is used with reference to the sound of hoofs 
(upabdin) and srnvise 'you are heard of (as being . . .), you are famous' 
is used in connection with Indra. Similarly, aite 'is perceived' occurs in 

(104) RV 10.143.4ab: aite tad vam suradhasa ratih sumatir asvina 
'Generous Asvins , your generosity, your benevolence is perceived then 
(at the time you come to the seat of the ritual) . ' 

Corresponding to such middle forms, srav-i, cet-i are used in the 
senses 'make to be heard, to be heard of, to be famous' and 'make to be 
perceived, to appear' . For example, 

(105) RV 8.96.12d: sravaya vaaam kuvid a'lga vedat '(Chanter,) make 
heard your speech (so that) perhaps (Indra) will be aware of it.' 

(106) RV 7.62.5c: a no jane sravayatam yuvana 'Young (Mitra and 
Varuna,) make us famous among the people.' 

(107) RV 1.3.12ab: maho drnah sdrasvatt prd cetayati . . . 'The 
Sarasvati makes visible her great flow of water.' 

(108) RV 3.34.5c: daetayad dhiya irria jaritre '(Indra) made these 
hymns (thoughts) perceptible to (to be perceived by) the chanter.' 

These are of the type (69c), though (105) lacks N2-X. In (108), moreover, 
the value of N2-X is jaritre 'to the chanter', a dative. To (108) would 
correspond a non-causal sentence 

(109) dcitran dhiya irria jaritre 'These hymns became perceptible to 
the chanter. ' 


(110) AV 14.1.31c: brahmanaspate patim asyai rocaya 'Brahmanaspati, 
make (the bride's) husband please her.' 

corresponds to a non-causal sentence 

(111) brahmanaspate patir asyai rocatam 'Brahmanaspati, may (the 
bride's) husband please her.' 

with a well-known construction: the verb rue (roeate) meaning 'be pleas- 
ing' occurs with a dative denoting one whom another pleases. 21 

Now srav-i, cet-i in the functions under discussion differ in an 
obvious way from sad-i (2.5), though all three correspond to intransi- 
tives. The verb sad is regularly intransitive (see 1.2), but sru and ait 
can also be transitive, and it is middle forms of these that have intran- 
sitive or passive value, while sad is regularly inflected actively. The 
derivate sad-i can rightly be considered a transitive counterpart to sad. 
However, srav-i, cet-i cannot be treated so simply without doing injustice. 
These have two functions. First, they play the role of transitive- 
causative verbs in relation to middle forms such as srnve , cite. In 
addition, they function as causatives to transitives {srnoti , cetati) . 

2.7 There are other sets such that an intransitive verb with middle 
inflexion is paired with a derivate in -i , which takes active inflexion. 
A well-known example is vrdh: vardhate 'grows (in strength, age), 
increases' and vardhayati 'makes to grow, to increase'. For example, 

(112) RV 6.37.5b: indro gzrhhir vardhatam vrddhamdhah 'May Indra, 
who is of grown might, (still) grow strong through our hymns.' 

has the middle imperative Vardhatam, and 

(113) RV 5.11.5cd: tvam girah sindhum ivdvam-r mdhtr a prnanti 
savasa vardhdyanti ca '(Agni,) hymns fill you with might, as great 
streams fill up the Sindhu, and make you grow in strength.' 

has vardhdyanti. Nevertheless, pairs such as vardhate, vardhayati differ 
significantly from pairs like srnve, srdvayati , considered in the preceding 
section. To begin with, sravayati 'makes to be heard' corresponds to a 
passive srnve 'is heard', and the relation between vardhayati and vardhate 
is not comparable. At best, one might suppose from comparing (112) and 
(113) that indro gTrbhir vardhatam could mean 'May Indra be made to grow 
by our hymns', so that vardhate could be a quasi-passive to vardhayati. 
However, vardhate is an agentive form; compare, for example, 

(114) RV 10.161.4ab: satdrn jTva sarddo vdrdhamanah satam hemdntan 
ohatam u vasantan 'Live a hundred autumns, a hundred winters, a 
hundred springs, increasing in strength.' 

which has the participle vdrdhamanah (nom. sg. masc.) 'growing, increas- 
ing.' It is, then, proper to say that gtrbhih of (112) refers to hymns as 
causes, not causal agents. In addition, sravayati 'makes to hear' 
contrasts with smoti 'hears', but there is not a comparable contrast 

between vardhayati 'makes to grow' and vardhati . On the contrary, these 
are used in an equivalent manner. For example, 

(115) RV 7.99.7c: vardhantu tva sustutayo giro me '(Visnu,) let my 
hymns, of good praise, make you grow strong.' 

has the active imperative vardhantu used in the same manner as vardhdyanti 
in (113). 

Similarly, the intransitive verb ni vartate (3 pi. imper. ni vartan- 
tani) 'turn back' contrasts with the transitive verb ni vartayati (2 sg. 
imper. ni vartaya) 'make to turn back', but there is no contrast between 
this and ni vartati (3 sg. in j . ni vartat) : 

(116) RV 10.19.3ab: punar eta ni vartantam asmin pusyantu gopatau 
'Let these (cows) again turn back, let them thrive in (the possession 
of) this cow owner.' 

(117) RV 10.19.2ab: punar end ni vartaya punar end ny a kuru 'Make 
these (cows) turn back again, make them be here again.' 

(118) RV 1.121.4cd: ydd dha prasdrge trikakum ni vartat dpa dri'tho 
mdnusasya duro vah 'When the three-humped (bull Indra) turned back 
(the cows, i.e. waters) in their flow, he opened the doors of the 
deceiver of man. ' 

2.8 There are other verbs from which are derived bases in -i that 
are equivalent to active forms of the primitive bases but which differ 
from vrdh, vrt and such in that their middle forms can have passive value. 
Consider, for example, mrj 'wipe, clean, polish, curry' (cf. Thieme 1929:19, 
22). This verb is transitive by the criteria available in Sanskrit (see 
1.2). The active mdrsti (3 pi. mrjdnti) takes an accusative complement 
denoting an object of cleansing, as in 

(119) RV 9.8.4a: mrjdnti tva ddsa ksipah '(Soma,) the ten fingers 
cleanse you. ' 

which has the accusative tvd 'you'. The middle mrje (pres. ptcple. mrjdnd) 
is used with passive value, and there is a marked passive mrjyate: 

(120) RV 9.109.17: sd vdjy aksdh sahdsraretd adbhir mrjdno gobhih 
srindndh '(Soma,) this steed of a thousand seeds, has rushed, being 
cleansed by water, blended with cows' milk.' 

(121) RV 9.3.3: esd devS vipanyubhih pdvanidna rtdyubhihl hdrir 
vdjdya mrjyate 'This heavenly (Soma,) as he is filtered, is cleansed 
by the inspired poets, truth seekers, (like) a bay horse (curried) 
for treasure (booty) . ' 

In addition, the participle mrstd 'cleansed, curried' is used with refer- 
ence to an object of cleansing, as in 

(122) RV 9.82.2b: dtyo nd mrstS abhi vdjam arsasi '(Soma,) you rush 


towards the treasure prize like a horse that has been curried.' 

The base marj-i formed to this verb is used in the same manner as the 
simple active rriarsti; for example, 

(123) RV 9.89.4c: svasara tm jamayo marjayanti 'The twin sisters 
(i.e. the fingers of the hand) cleanse him (Soma).' 

which in content is fully comparable to (119) . 

2.9 It is true, as Thieme emphasized, that derived verbs such as 
marj-i, formed to transitive bases other than those considered in 2.1-2.4, 
are regularly equivalent to simple transitive verbs in the Rgveda. It is 
also true, as has been pointed out (see Delbrlick 1888:224-6), that con- 
structions of the types shown in (B) , (C) of 1.1 occur in Vedic texts. 

The causative grah-i 'cause to seize' occurs frequently. For example, 

(124) TS etad va asyai nirrtigrhTtam/ nivrtigrhTta evainarn 
nirrtya grahayati 'That (part) of this (earth where one practicing a 
spell against another offers) is seized by Destruction; (the one 
practicing the spell) causes Destruction to seize him (against whom 
the spell is intended) at the place seized by Destruction. ' 

(125) MS 2.5.6 ^p. 54.17-8): prajapatih praja asrjata/ to. enam srsta 
City canany cental ta atimdnyamaria varuneriagrahayat 'Prajapati created the 
creatures. When they had been created, they became uppity towards 
him. He had Varuna seize them, being uppity.' 

The causative dap-i 'cause to give' occurs in the same construction: 

(126) TS dhnaivasmai ratrim pro. ddpayati ratriyahali '(The 
Adhvaryu) causes the day to give the night to him (the sacrif icer) , 
the night (to give him) the day.' 


(127) Ait.Sr. 5.1.5: rajaputrena carma vyadhayanti 'They have a 
Ksatriya pierce the skin. ' 

has vyddh-i, the causative of vyadh (vidhyati) 'pierce', in the context of 
part of the Vajapeya ceremony, when Ksatriyas shoot arrows at a skin. 
These are instances of construction (70c), with instrumental forms signi- 
fying agents of non-causal acts: nirrtya 'Nirrti, Destruction', vdru- 
nena 'Varuna', ahrid 'day', ratriya 'night', raj aputrena 'king's son, 
Ksatriya' . 

Causatives in a double accusative construction of type (69c) also 

(128) Ait.B 3.46.2: tad dhaitad eva jagdham yad asamsamanam drtvij- 
yam karayata uta va me dadyat uta va rrid vrntteti 'That is (called) 
jagdha ('eaten', i.e. similar to leftovers) when (the patron of a 


sacrifice) has carry out the office of officiating priest (a Brahmana) 
who wishes, "May he give me (some money) or select me".' 

(129) S5r. 5.8, KBU 3.3.8: esa hy eva sadhu karma karayati tarn yam 
ebhyo lokebhya unninTsata esa u evasadhu karma karayati tarn yam 
adho ninzsate 'For this one alone causes him to do good whom he 
wishes to lead up to those worlds; this one alone causes him to do 
bad whom he wishes to lead down. ' 

The accusatives drtvijyam 'office of Rtvi j ' , sadhu karma 'good work, deed 
(sanctioned in traditional lore)', asadhu karma 'bad work, deed (forbidden 
in traditional lore) ' refer to objects of the non-causal act denoted by the 
primitive verb k£^ 'make, do', as in Ait.B 5.34.3: artvijyam karoti 'carries 
out the office of Rtvij ' . The accusatives asamsamanam 'who wishes', tarn 
yam 'him whom' refer to objects of causation, signified by the derivate 
kar-i (karayate, karayati). 

Such causatives can also be used in elliptical sentences, where a 
term of a full construction is to be supplied. For example, SB 
athottamam dohayitva 'Then, after having the last (cow) milked . . .', 
has one accusative, uttamam 'last (cow)', but lacks a term referring to 
the agent of milking. This is known from the sacrificial context; the 
Adhvaryu, his speech held in check (vagyatah) , has a non-Sudra (asudrena) 
milk (dohayati) cows: 

(130) K^S A. 2. 22: vdgyato dohayaty asudrena 

(130) is itself elliptical, since the objects of milking are not here named 
and this does not specify the causal agent. 


(131) a. AV 3.20.8c: utadistantam dapayatu . . . 'And let him cause 
to give one who does not wish to give.' 

b. VS 9.24c: dditsantam (Hpayati ... 'He causes to give one 
who does not wish to give.' 

c. TS dditsantam dapayatu . . . 

dap-i occurs with one complement, the accusative dditsantam 'who does not 
wish to give'. Of course, dd 'give' is regularly a transitive verb, used 
with accusative complements denoting things given. But the non-causal 
sentence corresponding to (131a) would have ditsati used absolutely: 
dditsantam dapayatu = yo nd ditsati torn dapayatu. It is understandable, 
then, that ddp-i of (131) occurs in the construction type (69c) , with 
Ni-acc. corresponding to N^-nom. of the non-causative sentence. The verb 
ji too can be used absolutely: jayati, vi jayate 'is victorious, wins'. 
But ji also can be used with an accusative denoting that over which one is 
victorious, which one conquers, wins. And jdp-i, the causative to this 
verb, is found in a double accusative construction: 

(132) VS 9.11: brhaspate vdjam jaya brhaspdtaye vaaam vadata 
brhaspdtim vajam jdpayata/ indra vajam jayendrdya vacam vadatendram 
vajam japayata 'Brhaspati, win the prize; (drums,) make your sound 

to Brhaspati, make Brhaspatl win the prize. Indra, win the prize; 
(drums,) make your sound to Indra, make Indra win the prize.' 

(133) TS indraya vacarn vadatendvam vajam japayatendro vajam 
ajayit 'Make your sound to Indra, make Indra win the prize. Indra 
has won the prize.' 

(134) MS 1.11.3 (p. 163.9-10): indraya vacam vadatendraya vaaam sam 
vadatendram vajam japayatendra vajam jayeyam 'Make your sound to 
Indra, make your sound together to Indra, make Indra win the prize. 
Indra, I would win the prize. ' 

2.10 As I noted in 2.6 srnve can be used as a passive ('is heard') 
or in a stative sense ('is heard of, is famous'). Consider now 

(135) TS yah papmaria grTrttah syat so. agneycm kvsndgrxvam 
a labhetaindram rsahham/ agnir evusya svena bhagadheyenopasrtdh 
papmanam dpi dahaty aindrenendriydm atmdn dhatte/ muayate papmano/ 
bhavaty eva 'One who should be seized by evil should immolate a 
black-necked animal offered to Agni, a bull offered to Indra. Agni, 
approached with his share (of the offering), burns his evil. Through 
the (bull) dedicated to Indra, he puts pov7er in himself. He gets 
loose from evil. Indeed he prospers.' 

(136) TS yah papmdna grhttdh syat tasmd etam aindravarunTm 
payasyam niv vapet/ indra evasminn indriyam dadhati/ vdruna enam 

' varunapasan munaati 'For one who be seized by evil, one should set 
down this offering of curdled milk dedicated to Indra and Varuna. 
Indra puts in him power. Varuna looses him from the noose of Varuna. ' 

The verb of mucyate papmdnah 'He gets loose from evil' in (135) is compara- 
ble to mrjyate of (121) in that both have the affix -ya- . However, in 
(121) the passive is accompanied by an agentive instrumental, which is 
absent in muayate papmdnah. Moreover, the affix -ya- bears the high pitch 
in mrjyate, while in muayate it is the verbal base that is so accented. 
This form thus has the accent pattern of jayate 'is born' and statives such 
as pusyati 'thrives'. It is licit, then, to consider that muayate in (135) 
is not a true passive, as I have indicated with my translation. This 
intransitive is related to the transitive munaati 'looses', as in (136). 

(137) TS vi vd etdsya yajnd rdhyate ydsya havir ati riayate 
'His sacrifice fails to prosper the oblation of which is excessive.' 

has ati riayate 'is excessive'. This is accented in the same manner as 
muayate. Moreover, its meaning precludes treating it merely as a passive 
of, ria 'leave behind'. This too is a stative, like rdhyate 'prospers', 
vi . . . rdhyate 'fails to prosper'. To the intransitive ati riayate cor- 
responds a transitive ati reaayati 'makes excessive' (3 sg. opt. ati 
reaayet) , as in 

(138) TS ydd ava dyed ati tad reaayet 'If he were to cut 
into (a pregnant animal), then he would make (the offering) excessive.' 


This Is of the type srav-i 'make to be heard, to be famous'. Thus, in the 
language of Brahmana texts also there are found derivates in -i of this 

As one would expect from what was pointed out in 1.4, there are also 
found here derived verbs such as niarj-i 'cleanse', without causative 
value. For example, 

(139) ^B dtha pavitrayor marjayante 'Then they wash (with 
water) in two filters.' 

has marjayante, and this is not causative: participants in the rite are 
said to wash themselves, not to have others wash them. In his commentary 
on this passage, Sayana notes that rriarj-i is here a verb of the tenth 
class (cf. 1.4 with n 11) and glosses sodhayanty atmanam 'cleanse them- 
selves' . 

2.11 There is evidence to indicate that in early Vedic there was, in 
addition to the morphologically derived causative discussed thus far, a 
periphrastic causative construction in which a dative of an action noun was 
used with kr or dha 'make'. For example, 

(140) RV 1.112.8b: prandham sronam aaksasa Stave krthah '. . . (As- 
vins,) you make the blind to see, the lame to go.' 

(141) RV 3.31.19cd: druho vi yahi hdhuVd adevTh svas ca no maghavan 
tsataye dhW^ 'Generous (Indra,) avoid the numerous godless deceivers, 
and make us to win the sun.' 

(142) RV 7.79.5c: vy uaahantT nah sandye dhiyo dhah 'As you shine, 
(Dawn,) make our hymns to win.' 

However, no periphrastic causative construction was generalized in Indo- 
Aryan. Only morphologically derived causatives remained the norm, 22 so 
that these alone will concern us. 

2.12 It will be useful now to summarize what I have sketched out thus 
far and to make an additional point before considering how to account for 
historical developments. As shown, we have to do with the following kinds 
of relations. 

(a) A transitive verb is related to a derived causative in early Vedic 

gacchati 'goes' gamayati 'causes to go' 

srnoti 'hears, listens' sravayati 'makes to hear, listen' 

ceiati 'perceives' cetayati 'makes to perceive' 

pasyati 'sees' darsayati 'makes to see' [suppletion] 

"vksate 'sees' Tksayate 'makes to see' 

pihati 'drinks' payayati 'has . . . drink' 

vivaktij bravTti^ aha 'says' vacayati 'has . . . say' [suppletion] 

(b) An intransitive verb is related to a transitive-causative in 


-i (2.5): 

stdati 'sits' sadayati 'seats, sets' 

(c) A middle intransitive/passive is related to a transitive-causative 
in -i (2.6 cf. 2.10): 

srnve 'is heard, is famous' sravayati 'make to be heard, famous' 
cite 'is perceived, percep- oetayati 'makes to be perceived, 
tible' perceptible' 

(d) A middle intransitive is related to a transitive active, which is 
equivalent to a derivate in -i (2.7 cf. 2.10): 

vardhate 'grows' vardhati, vardhayati 'makes to grow' 

ni vartate 'turns back' ni vartati, ni vartayati 'makes turn 


(e) A passive, either unmarked, with raedio-passive endings alone, or 
marked, with ya before these endings, is related to a transitive active and 
a derivate in -i which are equivalent (2.8 cf. 2.10): 

mrje, mrjyate mdrsti, marjayati 

(f) In later Vedic, a transitive verb has a passive and a causative 
in -i related to the transitive (2.9): 

grhyate 'is seized' grhnati 'seizes' grahayati 'has ... seize' 

dtyate 'is given' dadati 'gives' ddpayati 'has ... give' 

vidhyate 'is pierced' vidhyati 'pierces' vyadhayati 'has ... pierce' 

kriyate 'is made, krmoti 'makes, karayati 'has ... make, do' 

done' kavoti does' 

duhyate 'is milked' dogdhi 'milks' dohayati 'has ... milk' 

[dugdhe 'gives milk,' n 4] 

The verbs of (a) are transitive in the Sanskrit of Panini's time (1.2). 
Verbs of the semantic groups in question were used transitively earlier too. 

As shown, srnve in (102) is passive. The participle srutd 'heard' 
also has passive value, referring to an object heard: 

(143) RV 6.49.3d: mdnma srutdm naksata raydmdne '(Night and Day) 
come to the devotional hymn, heard (by them) as they are celebrated 
in verse. ' 

Though srutd occurs also in the sense 'heard of, famous', this does not 
alter the fact that it can be passive. The passive drsydte 'is seen' 
occurs in 

(144) AV 7.101: ydt svdpne dnnam asriami nd prdtdr adhi gamydte/ 
sdrvam tad astu me sivdm nahi tad drsydte diva 'l"/hat food I eat in a 
dream is not found in the morning; may that be propitious to me; for 
it is not seen by day.' 

(145^ RV 10.88. 7ab: drsenyo yo mahirid samiddho rocata diviyonir 
vihhava '(Agni,) worthy of being viewed, through his greatness, whose 
womb is in heaven, who shines afar, has shone when kindled.' 

contains drsenyah 'worthy of being seen, viewed'. This is comparable to 
usenya 'desirable' (RV 7.3.9c) from the verb vas 'desire', which takes 
object accusative complements. Of the same type is tksenya 'worthy of 
being seen' (RV 9.77.3c). That middle forms of drs may be considered to 
mean 'appear' does not change the fact that this verb has passive forms, 
as does any transitive verb. The passive prydte 'is drunk' occurs in 

(146) AV 5.19.5cd: ksTram yad asyah pTyate tad vai pitrsu kilbisam 
'If its milk (the milk of a Brahmana's cow) is drunk, that is a 
stain on the Manes.' 

(147) RV 1.175. lab: matsy ccpayi te . . . matsaro madah '(Indra,) get 
drunk, for you have drunk the intoxicating drink. ' 

has apayi 'has been drunk', a passive aorist, and 

(148) RV 8.32.16c: no. sbmo apvata pope 'Soma was not drunk without 
reward. ' 

contains the passive perfect form pape . In addition, the participle pTtd 
can refer to an object that has been drunk, as in 

(149) RV 6.47.3a: ayam me pxtd ud iyarti vaoam 'This (Soma) arouses 
my speech when drunk (by me) . ' 

It is true that, like English 'drunk', pvtd can also refer to one who has 
drunk, but the fact remains that it is used passively. As shown in 2.4 
vac takes an object accusative complement referring to what one says. 

(150) RV 1.77. lab: katha daserriagndye kasmai devdjustoeyate bhamine 
gzh 'How may we honor Agni? What hymn, pleasing to the god, the 
shining one, is (to be) uttered?' 

has the passive uayate 'is uttered, said', and in 

(151) RV 6.34.5b: indraya stotrdm matibhir avaai 'A hymn of praise 
has been uttered to (this) Indra, through (our) thoughts.' 

the passive aorist avaai 'has been uttered, said' occurs. Again, in 

(152) RV 4.41.1: indra ko vam varuna surnndm apa stomo havismcffh .../ 
yo vam hrdi krdtumWfi asmdd uktdh paspdrsad indravaruna ndmasvan 
'Indra and Varuna, which praise with oblation has attained your 
benevolence? (Which is it,) Indra and Varuna, that, full of insight, 
obeisance, touches your heart when uttered by us?' 

uktdh 'uttered, said' is used passively. And 

(153) RV 6.45.6c: nrbhih suvTra uoyase '(Indra,) you are said by 

noble men to have a wealth of heroes.' 

contains the passive ucyase 'you are said ..." 

With the above, one may compare similar forms of transitive verbs 
such as mrc (2.8) or kr: 

(15A) RV 5.29.15ab: indra bvahna kr^yamand jusasva ya te savistha 
navy a akarma 'Indra most powerful one, take pleasure in the poetic 
formulae being produced, which we have made, new, for you.' 

(155) RV 4.6.11a: akari brdhma samidHdna tubhyam '(Agni,) you who 
are being kindled, a poetic formula has been made for you.' 

(156) RV 7.61.6d: kvtdni brahma jujusann imdni 'May these poetic 
formulae produced (by me) please (you, Mitra and Varuna) . ' 

By the criteria which serve to show that a given verb is transitive, 
then, the above verbs are indeed transitive. Consequently, srdvyati etc. 
in (a) are to be considered causatives to transitive verbs. That these 
transitive verbs do have such causatives in early Vedic distinguishes 
them from others, like mrj (2.8). But then, some of these verbs have an 
additional property that distinguishes them from other transitive verbs: 
they can take object complements other than the accusative. This group of 
items thus forms a fairly well defined subset of transitive verbs. 

Verbs of movement such as gam also share these features. The verb 
gam occurs with nominals denoting goals, in the accusative, locative or 
dative, and gamayati is related to gacchati as its causative. However, 
in earliest Vedic gam and similar verbs had not developed a full passive 
system. In the Rgveda, the participle gata. usually refers to an agent of 
going, and the passive gamyate does not occur. Nevertheless, it is also 
patent that from early times this verb was developing a full passive 
system. The Rgveda itself has an example of gata used with reference to 
an object, a path one has gone on: 

(157) RV 7.58.3: brhad vdyo maghdvadbhyo dadhdta jujosann in marutdh 
sustutim ndh/ gatb nddhva vi tivdti jantum pro. nah spdrKabhir 
utzbhis tireta '(Maruts,) grant great life force to the generous 
patrons. Let the Maruts enjoy our praise. It will take a living 
one across (life) as (does) a path that has been traversed. May you 
bring us to the fore with your desirable aids.' 

Moreover, in 

(158) MS 1.6.12 (p. 105.1-2): jdnam bhdgo ' gacchat tdsmdd ahur jdno 
gantavyas tdtra bhdgena sdm gacchatd iti 'Bhaga went to the people. 
Therefore, they say, "One should go to the people; there one comes 
together with ones share".' 

the gerundive gantavyah 'to be gone to' is used of a goal to which one 
should go, the people \jdnah) . Such usage is understandable only if gam 
was indeed treated as a transitive verb. Of course, in the Sanskrit of 


Panini's time gata could still refer to an agent or an object, as in (30), 

Now, gatah in (157) does not refer to a goal. It is used here of a 
path, road (adhva) one has taken on the way to a goal. The passive parti- 
ciple gata is thus used here of an object. In this context, it is worth- 
while to consider details about the Paninian rule (2.3.12: gatyarthakarmani 
dvittyaoaturthyau cestayam anadhvani) which serves to account for pairs of 
sentences such as (la) gaachati mdnavako gramam, (61) gaachati rrianavako 
gramdya. Another rule (2.3.2: karmani dvitTyd) says a nominal base is 
followed by an accusative ending (dvittyd) to signify an object (karman) . 
According to 2.3.12, a nominal base is followed by either an accusative 
or a dative ending (dviftyd, oatupth%) to signify an object {karman) with 
respect to the act denoted by a verb of movement (gatyartha) , unless the 
object is a path, road (anadhvani). In his second varttika on 2.3.12 
(dsthitapratisedhas ca), Katyiyana remarks that this negation ipratisedha) 
applies with regard to a road that one has taken (dsthita) . That is, the 
option provided by 2.3.12 applies with reference to a road that is a goal, 
not a road one takes; the latter is referred to by an accusative, as pro- 
vided for in 2.3.2. Thus, one says 

(159) adhvdnam gaachati 'He is going on the road, to the road.' 

(160) adhvane gacakati 'He is going to the road.' 
The passive counterpart of (159) is 

(161) adhvd gamyate 

and (157) contains such a passive turn, with the participle gata. 

The verb r {rcchati 'gets to, reaches') also takes accusative comple- 
ments. For example, in 

(162) TS yan nyanaam ainuydt prsthita enam dhutaya rccheyuh 
'If one were to set up (the fire) face down, oblations would get to 
it from the back. ' 

rcoheyuh (3 pi. opt.) occurs with an accusative (enan 'it') referring to 
the fire into which oblations go. As shown in 2.1, (82) has the causative 
gam-i, used in speaking of making a bad dream go to enemies. 

(163) RV 10.164.5de: pdpS yarn dvismds tarn sd vcahatu yo no dvesti 
tarn rcchatu 'Let this evil (intent) get to him whom we hate, let it 
get to him who hates us.' 

contains roahatu 'let ... get to' with an accusative {torn 'him') in a 
similar context. More figuratively, raahati occurs with an accusative 
(etam 'him') in 

(164) TS aputd va etdm vag rcchati yam djaghnivdmsam abhi 
sdmsanti 'Impure speech gets to him whom people calumniate without 
his having killed. ' 


referring to one who is the object of calumny, whom impure speech (aputa 
vak) gets to. In addition, rcchati frequently occurs with suk 'pain, 
affliction' and an accusative denoting that which pain gets to, as in 

(165) TS pa&or va aLahdhasya hvdayam sug rcohati 'Pain 
gets to the heart of an immolated animal. ' 

Compare a passage like 

(166) ^B pasov ha va alabhyamanasya hrdayam suk sam abhy 
dvaiti 'Pain gets into the heart of an animal being immolated.' 

which has the movement verb i 'go' (with preverbs sam abhy dva) . Further, 
the past participle rtd is used with reference to an object to which pain, 
affliction has gotten. For example, 

(167) TS tdsmat sarriqvat pasunam prajayamananam aranydh 
pasdvah kdntyamsah/ sued hy rtdh 'It is for this reason that, among 
animals born otherwise the same, wild ones are smaller. For they 
are afflicted by pain. ' 

Given suoa ... rtdh 'gotten to by pain', one must accept that accusatives 
such as hrdayam of hrdayam sug raehati 'pain gets to the heart' in (165) 
are object accusatives. 

The available evidence thus indicates that the primitive verbs of 
(a), including a verb of movement like gam, are transitives. Consequently, 
gamayati etc. of (a) must be considered causatives to transitive verbs. 
In addition, some of these verbs have passive forms such as srnve , and the 
causatives of (c) are related to these. 

Now, by the criteria mentioned, sad of (b) is intransitive. Of 
course, there is no true passive to such a verb, so that the contrast 
between transitive and passive does not obtain. Moreover, there is here 
no possible contrast between a transitive and a causative. The same is 
true of type (d) . However, in this case there is also a redundancy in 
that, despite the absence of an opposition between transitive and causa- 
tive, one has pairs like vardhati, vardhayati . In type (e) , transitive 
and passive do contrast, but there is no contrast between transitive and 
causative. In addition, we have here redundancy both in the passive 
(jnrje, mrjyate) and in the transitive (mdrsti, marjayati^ . Finally, type 
(f) is like (a) in that there is here a three-way contrast of passive, 
transitive, and causative. However, causatives of (f) are demonstrably 
later than those of (a) . 

3.0 There are, then, two major questions to be considered: How 
were causatives of type (f) introduced into the language? How is it that 
causatives take part in different syntactic constructions? The data 
sketched above suggest answers to these questions. 

3.1 Consider now a development involving passives. From examples I 
have given, it can be seen that simple middle forms could be used with 
passive value in early Vedic. Thus: (102) srnve 'is heard', (120) adbhir 


mrjanah 'while being cleansed by water', (148) pope 'was drunk'. Simi- 

(168) RV 10.112.3cd: asriabhir indra sakhibhir huvanah sadhrToTnS 
mddayasva nisadya 'Indra, sit down and get yourself drunk with us 
as you are called by us, friends.' 

has huvanah 'being called (by us, your friends)', a participial form to 
have 'is called', and 

(169) RV 1.92.7b: divah stave duhita gotamebhih '(Dawn,) the daugh- 
ter of heaven is praised by the Gotamas.' 

contains stave 'is praised'. In addition, there are present and imperfect 
forms that I call marked passives, in which the affix ya precedes middle 
endings or a participial affix. Thus: (121) mrjyate 'is cleansed', 
(89) huyanianah 'being called'. Similarly, 

(170) RV 3. 22. led: sahasrinam vajam ... sasanvan tsan tstuyase 
Ja.tavedah ' (Agni) Jatavedas, you are praised, being one who was 
gained treasure in the thousands.' 

has the marked passive stuyase 'you are praised'. In early Vedic, then, 
there was a competitive situation, which involved redundancy: either 
marked or unmarked passives could be used. However, by the period of 
Into-Aryan represented in Brahmana texts, the marked passive had all but 
completely ousted the unmarked passive (see Speyer 1896:49, §167). One 
is thus practically to the system of the Sanskrit described by Panini , 
wherein a present or imperfect passive regularly has atmanepada affixes 
preceded by yoK and third singular aorist forms of the type akari 'has 
been made', opposed to akrta 'has made for himself, are the norm in pas- 
sive function (see 1.5). 

I think it can be considered no accident that, in addition, the stage 
of Indo-Aryan represented in Brahmana texts has a fully developed causa- 
tive system, including causatives of type (f) in 2.12. Not only this, but 
causative bases have passive forms, as do any transitive verbs. Thus, 

(171) TS upopte 'nye gvahah sadyante'nupopte dhruvah 'Other 
cups are set on a place where (earth) has been strewn, the Dhruva 
cup on a place where it has not been strewn. ' 

(172) MS 2.3.9 (p. 37.9): brahmanah payayitavyah 'The Brahmana is 
to be given to drink. ' 

contain sadyante 'are set' and payayitavyah, a gerundive of pay-i used 
with reference to one who is to be made to drink. Further, 

(173) ^B 12.4.1,12: tad ahuh/ ydsyagnihotrt dohySanand vasyeta kirn 
tdtra karma Kd prayascittir vti 'They say: what is to be done, 
what is the expiatory procedure, on the part of one whose cow for 
the Agnihotra ceremony should bleat while one is having it milked?' 


has the passive participle dohyamana 'while being caused to be milked', 
to the base doh-i , as in 

(174) K^S 4.14.1: agnihotrtm dohayati 'He has (someone other than 
a Sudra) milk the cow for the Agnihotra' (cf. [130]). 

Such passives are also part of the verbal system in the Sanskrit of 
Panini's time. For example, as (la) gacohati rridnavako gvamam has a pas- 
sive counterpart (12) manavakena gramo gamyate, so also does (lb) devadatto 
nidnavakam gramam gamayati: 

(175) devadattena rridnavako gramam gamyate 

3.2 The elimination of unmarked passives in Indie had systematic 
repercussions in the rest of the verb system. So long as unmarked passive 
forms of the type mpje coexisted with marked passives of the type mrjyate, 
the contrast between transitive and passive could be correlated directly 
with the contrast between active and middle, the affix ya being redundant. 
Once the type mrjyate has become the norm, however, this is not so. Now 
if a verb has both active and middle endings contrastively , this is cor- 
related only with the distinction between performing an act for ones own 
benefit and doing this for someone else. Both active and middle endings 
of the present-imperfect system now can be said to signify agents, as 
opposed to objective middle endings preceded by ya, now required. 

In this context, consider again a verb such as vrdh, of type (d) in 
2.12. The middle vardhate was retained in Sanskrit, as a deponent. On the 
other hand, the active vardhati was gradually eliminated and does not have 
a place in the Sanskrit of Panini's time. In the emerging system, a 
transitive vardhati could be paired with a middle vardhate of the type 
kurute 'makes for himself or with a true passive vrdhyate of the type 
kriyate 'is made'. Understandably, neither of these was used. Conse- 
quently, vardhati ceased to be used. In the later system, then, one has 
only vardhate, in which the middle ending signifies an agent of growing, 
opposed to the causative vardhayati . Similarly, the later system has 
vartate and vartayati. Note, in addition, that in this way a redundancy 
was eliminated. In early Vedic, there was no contrast between Vardhati and 
vardhayati or ni vartati and ni vartayati. In the emerging system, this 
lack of contrast is reflected formally, with only Vardhayati, ni vartayati. 

3.3 Consider now the consequences of such developments with regard 
to transitive verbs like mrj, of type (e) in 2.12. As has been shown, in 
the early system such verbs entered into constructions which are schemati- 
cally as follows: 

(176) a. V- S *.j r N.-instr. N.-nom. [passive] 

[ya-mxd.J 1 2 

b. V-act. N -nom. N -ace. [transitive] 

c. V-i N -nom. N„-aee. [transitive] 

With the elimination of the type rrrrje from the present passive system, we 


(177) a. 

V-ya-mid. N -instr. N.-nom. 



V-act. N -nom. N -ace. 



V-i N^-nom. N.-acc. 


That is, there are now two transitive constructions associated with a 
single passive construction. Moreover, one of these transitive construc- 
tions, (177c), has a verbal base of the type which functions as a causative 
to a particular subset of transitive verbs, type (a) of 2.12. 

I consider that causatives of type (f) had their origin in the trans- 
fer of construction (177c) to causative function: (177c) was shifted to 
(69c) or (70c). 

Now, from all that is known about historical developments in language, 
one must assume that for some time a situation persisted such that, while 
the verb of (177c) was being used as a new causative, (177c) continued to 
function also as a transitive construction equivalent to (177b). Indeed, 
it is difficult to account for the evidence otherwise: in later Vedic, 
right up to the time of Panini, there are derived verbs such as marj-i, 
var-i, which do not necessarily function as causatives (see 1.4, 2.10). 
In this manner, verbs in construction (177c) could still be related to 
passives in construction (177a), just as an old causative like sravayati 
could still be related to a passive; see 2.12(c). Recall now that there is 
a straightforward relation between sentences in which old causatives of 
type (a), 2.12, were used and the non-causative sentences connected with 
them: to a nominative of the non-causative sentence (Ni-nom.) corresponds 
an accusative (N^-acc.) in the causative sentence, and other case markings 
remain the same. By the same relation, a new causative sentence related 
to (177a) would have N2-acc. and N^-instr. In other words, the new causa- 
tive construction would be (70c). 

In the new causative construction, the verb stands as one of three 
related structures, together with an active and a passive. Hence, there 
could be additional influence of the old causative construction: the new 
causative too might enter into a double accusative construction, type 
(69c) . There is evidence to show this occurred at a fairly early time (see 
2.9). However, separate constructions were still maintained, and, as 
things sorted themselves out, by Panini 's time only kar-i and Kdr-i among 
the new causatives occur in the double accusative construction, and then 
only optionally. 

Of course, old causatives that were in use from earliest Vedic 
remained, and in constructions proper to them. In connection with these, 
recall what I pointed out in 1.2 about (63)-(66). The verb smr can be used 
transitively. It can also take a genitive complement and occur in a pas- 
sive construction with such a complement, so that the passive verb shows 
no variation for number agreement and is of the type asyate in (42) . As 
shown earlier (2.3), verbs of consuming too take accusative and genitive 
complements in Vedic. And they have passive forms. Thus, asyate 'is 
eaten' in 

(178) ^B patresu hy asanam asyate 'For food is eaten in 
vessels. ' 


is a passive like pTydte in (146): it agrees in number with asanam 'food'. 
On the other hand, asyate in 

(179) tdsniad etesam pasundm ndsyate 'Therefore, one does not eat these 
animals. ' 

which corresponds to 

(180) 5b tdsniad etesam pasundm ndsitdvyam 'Therefore, one 
is not to eat these animals.' 

is like smaryate of ^(66). It does not agree in number with the genitive 
plural etesam pasundm, just as the neuter singular gerundive asitdvyam of 
(180) shows no gender-number agreement with this genitive. Let us now 
agree, as I think we must, that sentences such as (la) and (61) or (94) 
and (95) are equivalent to each other so far as concerns the relation 
between the acts in question and the goal or object: in both instances, 
someone is going to a village and Indra is asked to drink Soma. Further, 
as shown, verbs of the type (a) in 2.12 are transitive. However, as also 
shown, the goals, objects of these verbs could be denoted by non-accusative 
forms. And, as can be seen from (179), (180), such forms could be used 
also in passive sentences. To such sentences, one could not have a new 
causative of the type (70c) , with an accusative corresponding to the nomi- 
native of the passive, for the simple reason that the passive lacked such 
a nominative. Moreover, as noted in 2.12(a), (c) , though causatives such 
as srdvayati could indeed be related to both an active and a passive, there 
was a clear semantic difference: srnoti 'hears, listens^ : srdvayati 'causes 
to hear, listen' as opposed to sruyate 'is heard': srdvayati 'causes to be 
heard'. These facts suffice, I think, to account for the fact that con- 
struction (70c) was not fully extended, to include also old causatives 
that entered into the double accusative construction, eliminating (69c) . 

One should, nevertheless, expect some encroachment of (70c), under 
specific conditions: if a verb of the semantic groups for which (69c) was 
the norm were used, possibly with specialized meaning, regularly with an 
accusative complement. There is evidence to show such extension. For 

(181) TS sa hrdayasuldm ahhi sdm eti/ yat prthiv^am 
hrdayasuldm ud vdsdyet prthivxm suodrpayed ydd apsv apdh sucarpayet/ 
suskasya cdrdrdsya ca samdhav ud vasayaty ubhdyasya sdntyai/ yam 
dvisydt tarn dhydyet/ sucaivainam arpayati 'It gets to the heart spike. 
If (the Adhvaryu) were to remove the heart spike onto the ground, he 
would cause pain to get to the earth, if onto water, he would cause 
pain to get to the water. He removes it to a juncture of dry land 
and wet land, to let both have peace. Let him (then) think of one he 
hates. (Thereby,) he causes pain to get to him.' 

has prthivTm sucarpayet 'he would cause pain to get to the earth', in which 
the causative arp-i (3 sg. opt. arpayet) , seen also in suca ... enam 
arpayati 'he causes pain to get to him', is construed with an accusative 
(,prthiv%m, enam) that refers to the object afflicted by pain and an instru- 
mental {suca) referring to pain, affliction, which is caused to get to 


these. As can be seen from sa 'it (i.e. pain)', this passage is part of a 
larger context. It continues immediately after (165), where hrdayam sug 
voahati occurs. Patently, arp-i is the causative to r, a verb of movement. 
The primitive verb vochati in usages such as (165) regularly takes only an 
object accusative complement, which implies a nominative in a passive 
construction together with an instrumental referring to pain which gets to 
someone (cf. [167]). Accordingly, one finds arp-i in construction (70c). 

There is also evidence to show that, at least in some areas, the 
causative construction (70c) was extended to other verbs which originally 
entered into double accusative causatives. In varttika 5 on Panini 1.4.52 
{adikTmdintvahTnam pratisedhah) , Katyayana remarks that the agent of 
certain acts, among them those denoted by ad, khdd 'eat', should not be 
classed as object with respect to causatives. In his discussion of this 
varttika, Patanjali gives the following examples (Abhyankar 1962:337, 
lines 15, 18): 

(182) a. atti devadattah 'Devadatta is eating.' 

b. adayate devadattena '... is feeding Devadatta.' 

(183) a. khadati devadattah 

b. khadayati devadattena 

Further, in virttika 7 on 1.4.52 (bhakser ahimsarthasya) Katyayana notes 
that the same prohibition should be made for the causative of bhaks meaning 
'eat'. By the third century B.C., then, these three verbs of consuming 
formed causatives which could occur in construction (70c) . And by this 
time these verbs regularly took only accusative complements. 

However, so far as I know, this construction had not been fully gener- 
alized by Panini 's time. Indeed, verbs of the semantic groups which had 
construction (69c) in early Indo-Aryan continued to enter into a causative 
construction different from that of other transitive verbs, and do so to 
this day (see note 22) . 

3.4 In sum, the available evidence requires one to accept that earli- 
est Indie had true causatives, formed to particular transitive verbs. New 
causatives were then developed at a fairly early time, certainly by the 
times of Brahmana works, leading to the system known from Panini 's era. 23 
I suggest the new causative construction is to be accounted for as shown 
in 3.3. 

4.0 At the beginning of this paper. I pointed out that according to 
the view now prevalent early Vedic did not have true causatives. Let me 
now turn to the arguments advanced in favor of this position. 

4.1 Jamison (1967:126) states the following: 'Though in early Skt. 
(= RV, AV, early Brahmanas) there existed no specifically causative means 
of expression, either morphologic or periphrastic, in the later language 
(= later Brahmanas, Epic, and Classical) causativity could be expressed 
morphologically by means of a verbal formation with -aya-suffix. The -aya- 
causative of this later period is the formal continuation of the -aya-verb 
of the older Vedic period. However, this earlier formation was not a 

causative in function but merely had tr. [transitive, GC] value." Later 
(1976:127), Jamison remarks that the fundamental property of the formation 
in question in the later language is causativity, not transitivity. One 
of the pieces of evidence which leads to this conclusion is that, according 
to Jamison, non-causative transitive derivates such as ahad-i in 

(184) RV 6.75.18a: mamiani te varmand chadayami 'I cover your vul- 
nerable areas with armor.' 

cannot be formed in the later language. The evidence fails to support 
Jamison's claims. In view of even the select examples given in 2.9, her 
statement concerning the absence of causatives in the early language must 
be considered false, so far as concerns the Atharvaveda and Brahmanas . 
Moreover, since derivates in -i were not made exclusively causative in 
Sanskrit, right up to Panini's time, her claim about the later language is 
also factually unacceptable. 

Jamison's major claims are two. First, she says derivates in -i were 
formed in the early language only to intransitive verbs, so that the central 
function of these derived bases was then transitivity. Second, true causa- 
tives arose, according to her, through a syntactic analogy due to 'the 
inherent ambiguity' (1976:133) of accusatives in sentences with two accusa- 
tive forms. To support her claim that early derivates in -i could only be 
formed as transitives to fundamentally intransitive verbs, Jamison has of 
course to do something about the verbs of my type (a), 2.12. Her procedure 
is simple. She merely says (1976:130) that verbs which, though they do 
take accusative complements, also take others are thereby intransitive, 
and concludes: 'Thus, according to my definition, when a noun can occur in 
the AC [accusative case, GC] or in another case in the same meaning, the 
noun is not a DO [direct object, GC], and the verb with which it appears 
is not transitive, even at those times when the non-object complement is in 
the AC' Further, Jamison strongly emphasizes the type Vardhayati (see 
2.7), which occurs with a single object accusative, and says (1976:126), 
'In the early language the defining syntactic feature of the -aya-transitive 
was that it could be construed with a single AC direct object . . . ' She 
also remarks (1976:126) that a verb such as vardh-i in RV 1.36. lid: 
agnim vardhayamasi 'We make Agni grow strong' had 'another secondary 
syntactic feature' in that such a sentence could 'be perceived to be func- 
tioning as causative to a corresponding intr. sentence built to the same 
underlying root . . . ' According to Jamison, then, a sentence such as AV 
12.3.34d: enam [var. etdm] ... gamayantam (which she translates [1976:132] 
'Make him go to the end') has only one object accusative: enam 'him'. The 
accusative antam is not objective, either here or in the corresponding 
non-causative sentence sd gacohaty dntam, which Jamison treats as intran- 
sitive. However, she goes on to say (1976:133), sentences like the last 
cited have surface structures identical to sentences with transitive verbs, 
both containing accusatives. And due to the ambiguity of the accusative, 
a sentence such as enam . . . gamayantam could then be reinterpreted as con- 
taining two object accusatives. In this manner, she concludes (1976:133), 
' . . . causativity, came to be perceived as the central function of the 
-aya-formation, ' although earlier this was merely (1976:128), '. . . an 
accidental byproduct of the relation between a fundamentally tr. and a 
fundamentally intr. verb. ' 


It is crucial to Jamison's thesis that the accusatives in sentences 
such as (75) tarn ... gaaohanti .... visvany ekah srnavad vacamsi me, (84) 
... indro artham cetati, (86) ... tvdh dadarsa vacam, !.. tvah 
. . . na smoty eriam, (87) ... rria ... Tksadhvam, (94) ... somam piba ... 
not be object accusatives. Moreover, she rightly stresses (1976:130) the 
need to distinguish between object accusatives and others 'without appeal- 
ing to semantic intuitions,' that is, on formal grounds. However, the 
single formal criterion she uses is not sufficient. Simply to say that a 
verb which can take a non-accusative complement in addition to an accusa- 
tive is thereby intransitive by definition is not justifiable. Recall now 
that (114) contains the participial form vdrdhamanah, to the intransitive 
verb vrdh (2.7), the intransitive verb JTv (2 sg. imper. JTva) 'live', and 
the accusatives katam ... saraddh 'a hundred autumns', satam hemantan 'a 
hundred winters', satam ... vasantan 'a hundred springs'. Similarly, 

(185) TB so'svatthe samvatsaram atisthat 'He (Agni) remained 
in the Asvattha tree a year. ' 

has the intransitive stlia (3 sg. impfct. atisthat) 'stand, remain', the 
locative asvatthe, and the accusative samoatsaram 'a year.' Again, in 

(186) 5b tan ma ekam ratrim ante sayitase 'Then you will 
lie near me one night . ' 

the intransitive st (2 sg. fut. sayitase) 'lie, sleep' occurs with the ^ 
locative phrase me ... ante 'near me' and the accusative phrase ekam vatvim 
'one night'. That jTu, stHd, s~ occur with accusatives in such examples 
does not, or course, suffice to make them transitive verbs. The accusa- 
tives found here have specific semantic features. In (114), (185), (186), 
they refer to times associated with the acts in question, during which 
these take place or are supposed to, and in 

(187) RV 2.16.3cd: na te vdjram anv asnoti kascana ydd asubhih 
pdtasi ySjana puru '(Indra,) no one reaches your vajra when you fly 
many yojana with your fast (horses) . ' 

the accusative yojana puru 'many yojana (measures of distance) ' used with 
pat 'fly' refers to a distance. In effect, these are adverbial accusatives, 
traditionally called accusatives of extent, and they are neutral in the 
sense that they can be used equally with intransitive and transitive verbs. 
Thus, prathamam ratrim 'the first night' in 

(188) SB asvinena pdyasa prathamam ratrim pari sinaati 'He 
sprinkles (the wine) the first night with the milk dedicated to the 
Asvins. ' 

occurs with the transitive verb pari sinaati 'sprinkles round'. Similarly, 

(189) RV 1.98. 2d: sd no diva sd risdh patu ndktam 'May he (Agni) 
protect us from harm day and night.' 

has the transitive verb pa 'protect' with the object accusative nah 'us' 
and the adverbial phrase diva ... ndktam 'day and night', in which the 


accusative naktam is coordinated with the instrumental diva. Such adverb- 
ial elements remain unchanged in passive sentences. For example, 

(190) RV 10,95.4cd: astam nandkse yasmin aakan diva naktam snathitKd 
vaitasena ' (UrvasT) came to the house (of Pururavas) , in whom she 
took pleasure, pierced by his rod day and night.' 

has the passive participle snathita 'pierced' with the adverbial phrase 
diva naktam. The situation is quite different concerning the accusatives 
which occur with gam, sru, cit, drs, ~ks, pa ('drink') in (75), (84), (86), 
(87), (94), and similar sentences. To begin with, these accusatives do not 
fit into any one semantic category; they are not like adverbial accusatives 
of time and distance. In addition, the verbs with which these accusatives 
occur are found in passive constructions such that passive verb forms show 
number and gender agreement with nominative forms corresponding to accusa- 
tives in non-passive sentences (see 2.12). Thus, if one considers the 
criteria which should be considered, one must conclude that these verbs 
have to be treated as a subset of transitive verbs. By singling out one 
syntactic feature and omitting others, Jamison fails to take into account 
the necessary facts. Hence, the crucial part of her thesis fails. Given 
this, there is no need to go into her explanation concerning how speakers 
are supposed to have perceived things. 

Jamison also suggests an explanation for the use of instrumental forms 
in the construction type (70c). She says (1976:134, n 3) , 'I consider 
this type of expression a secondary development of the one with double AC, 
designed to avoid the ambiguities inherent in having both DCs in the same 
case.' This too I consider unacceptable. After all, despite possible 
ambiguity, the double accusative causative construction was maintained. 
Moreover, the distribution of the two causative constructions used in the 
Sanskrit of Panini's day shows a distinction among transitive verbs that 
supply causatives for these that is remarkably systematic, not what one 
might expect if, as Jamison says, one of the constructions was used to 
avoid ambiguity. Under her explanation, one should expect a more random 
distribution, the potentially ambiguous construction steadily losing 
ground. However, what is potentially ambiguous in isolation is not so in 
context, and speakers use sentences in contexts. For example, consider 

(191) TS 

a. yajna yajnam gaeaha yajndpatim gacahety aha (TS 
'(The Adhvaryu) says "Sacrifice, go to the sacrifice (i.e. Visnu) , 

go to the patron of the sacrifice".' 

b. yajndpatim evdinam gamayati 'He makes it go to the patron 
of the sacrifice.' 

Taken out of context, (191b) could be ambiguous, but, as can be seen from 
the anaphoric pronoun enom (ace. sg.), this is embedded in a context, where 
it is unambiguous. 

4.2 In an earlier study of Sanskrit causatives, Thieme also tried to 
demonstrate that early Vedic had no true causatives, formed to transitive 
verbs. I agree with Thieme (1929:19) that sentences such as (7b) result 
from influence of causatives formed to intransitive verbs, that is, from an 


extension of the double accusative construction. Moreover, Thieme's 
treatment of the Issues shows the mastery of Vedic and Paninian materials 
so typical of this scholar. Nevertheless, I regret I cannot consider his 
arguments fully cogent. In essence, Thieme tries to show that all verbs 
which formed derivates in -i that one might call causative were intransi- 
tive in early Vedic. He takes up (1929:21-2) the causative construction 
for verbs of movement etc. in the Sanskrit of Panini's time (see 1.1, [A]), 
refers to Vedic passages where this is found, and then says (1929:22) 
concerning the verbs in question: 'Zu alien diesen Verben ist ein Intran- 
sitiv denkbar, das ausdrllckt, dass die vom Verb bezeichnete Handlung am 
Subjekt in Erscheinung tritt. Dieses Intransitiv kann entweder durch das 
blosse Medium ausgedruckt werden (vardhati "starkt," vardhate "wird 
stark") Oder — was besonders haufig ist— durch das Medium mit -ya {rrw.jyate 
"wird sauber"), oder durch das Verbaladjektiv mit -to. (vrddhd "einer der 
gewachsen hat," gegenuber ptta, das auch heissen kann "einer der getrunken 
hat"). Tritt zu diesem Intransitive der Agens im Instrumental, so sprechen 
wir von Passiv. Z.B. harsate "freut sich," aber X.30.5 yabhih ... harsate 
"durch die er er freut wird." mrjyate "wird sauber, lauter," z. B. in 
IX. 86. 6 ycdi pavitre adhi mrjyate harih, aber IX. 76.4 yah suryasyastrena 
mrjyate "der gelautert wird durch • • •'" [Emphasis Thieme's]. 

The above actually involves two points. Let me begin with the claim 
that an intransitive middle is conceivable for all the verbs in question. 
The verb gam regularly takes middle endings in Panini's time if it is 
accompanied by the preverb sam (Panini 1.3.29: samo gamyrcchibhyam 
[kartojn 14, atmanepadam 12]): sam gam 'come together'. This is so also 
in earlier Vedic, as can be seen from examples such as (158) and RV 
1.119.3a: sam yan mithah pasprdhariaso agmata '. . . when they came together 
competing one'with the other'. And ud gam 'go up' also is found with 
atmanepada affixes (e.g. RV 8.69.7d). However, such uses of the middle of 
gam can hardly account for causatives like gamayamasi in (77), at least not 
in the way required by Thieme. It is true that sravayati 'make to be 
heard' is paired with a middle like srnve 'is heard', but sravayati 
'makes to hear' cannot be considered merely a transitive related to an 
intransitive middle. It must be paired with svnoti 'hears' (see 2.2, 2.6). 
Similarly, though oetaryati 'makes to be perceived, to appear' is a trans- 
tive-causative related to aite 'is perceived, appears', aetayati 'makes 
to perceive' must be coupled with cetati 'perceives'. It is also unac- 
ceptable to consider that dnrsayati 'makes to see' is merely a transitive 
to a middle drsyate 'is seen, appears', and for the same reason. Under 
Thieme's thesis, darsayati would have to be a transitive to drsyate mean- 
ing 'appears', but a transitive of the type aetayati with an accusative 
denoting what is made to appear and a dative signifying one to whom this 
is shown, as in (103), simply will not account for sentences like (92), 
(93). As shown in 2.12, pa 'drink' is transitive. This verb usually has 
active inflexion except in passive usage. It is true that early texts also 
have non-passive forms of this verb; for example, RV 10.114.7cd: apnanam 
ttrtham ka iha pro. vocad yena pafKd pra pihante sutasya 'Who will proclaim 
here the Apnana ford, the path on which (the gods) drink the pressed Soma?' 
Note, however, that pra pibante occurs with sutasya here, and is thus 
syntactically like piba in (95). It is also true that such middle f orms ^ 
of pa can be used absolutely, as in RV 10.135.1b: devaih sam pibate yamdh 
'. . . Yama drinks with the gods'. However, the active of ^a is similarly 

used, as in RV 1.23.18b: yatra gavdh pihanti nah '. . .in which our cows 
drink'. Consequently, it is not possible to account for the causative 
pay-i in the manner Thieme suggests. The verb vaa too is transitive 
(2.12). To be sure, there are non-passive middle forms of this verb, but 
they do not conform to the requirements of Thieme 's thesis, since they are 
regularly transitive. For example, in RV A. 5. 11a: rtam voce namasa 
praahycmidnah 'I will tell the truth when asked with a show of obeisance' 
the middle form voae (1 sg. aor. inj . ) occurs with the object accusative 
rtcim 'truth'. No matter how conceivable it might at first appear, then, 
the fact is that saying derived bases such as gam-i, srav-i, dars-i , 
pay-i, vaa-i originally were nothing more than transitive derivates cor- 
responding to intransitive middle forms cannot be justified from the 

Thieme makes his second point by bringing in sentences of type (56) 
odandh paoyate 'The rice is cooking', when he speaks of mrjyate used with 
or without an agentive instrumental. ^This sentence type does occur in 
Vedic. For example, TS sa yad hhidyetavtim araahed yaja/nano 
hanyetasya yajndh 'If it (the ukha pot) were to break, the patron of the 
sacrifice would undergo ruination, his sacrifice would perish' has 
bhidyeta, hanyeta, optative forms corresponding to intransitive middles 
bhidydte 'breaks', hanydte 'perishes'. However, bringing in this type 
does not overcome the objections set forth above. In fact, this raises 
additional problems. The sentence type (56) regularly has verbs which 
can and do take object accusative complements and have true passives. The 
feature particular to this type is that an external agent performing an 
act on an object may not be specified, and the verb behaves like an 
intransitive verb. However, not all verbs enter into this construction, 
and verbs of movement such as gam do not. Moreover, if a form like 
pacyate, bhidyate, hanyate is used passively ('is cooked', 'is broken', 
'is killed'), an external agent is signified by an instrumental. Conse- 
quently, under Thieme 's thesis, one should expect only construction (70c) 
in causative sentences, not also (69c). 

Of course, Thieme is well aware of the problem, and he anticipates 
Jamison's position, by arguing that verbs which have causatives with 
construction (69c) are actually intransitive. Thus, he says (1929:20) 
that pdrdm ... paravdtam of (77) is an accusative of direction (Richtungsak- 
kusativ), vajam of (132) an accusative of content (Inhaltsakkusative) , so 
that the primitive bases gam, ji from which gam-i, jdp-i derive are intran- 
sitive. He does admit that accusatives used with verbs like Tks, budh 
are object accusatives. For he calls these transitive verbs (1929:20). 
Still, he hesitates to give them full transitive status, saying (1929:20), 
'Transitive Verben der Art wie zks, budh jedoch sind ihrem Ursprung nach 
intransitive Verben des Vorgangs, die mit einem freieren Akkusativ, dem 
Akkusativ der Evstreakung , konstruiert werden' [Emphasis Thieme's], and 
in a footnote (1929:20 n 2) he further suggests these verbs may originally 
have been construed with an accusative of goal (Akkusativ des Ziels) , of 
direction. Later (1929:22), Thieme reiterates that gam takes an accusative 
of direction, a verb of perception an accusative of extent, though here 
he puts 'der Erstreckung' within inverted commas. Further, according to 
Thieme (1929:20-1), a genitive used with a verb of consuming like pa 
'drink' is a partitive genitive, and an accusative construed with such a 


verb is one of extent: 'Ahnlich llegt der Fall bei pa "trinken", das mit 
dem acQ . aber auch mit dem ^erz . part, konstruiert warden kann, wenn 
nahmlich die Handlung sich nur auf einen Teil der bezeichneten Sache 
erstreokt^ [Emphasis Thieme's]. The above does not withstand scrutiny. 
One must obviously recognize an accusative of goal, which alternates with 
other case forms (2.1), as one has to distinguish, with respect to verbs 
of movement, an accusative denoting a road taken and one denoting a road 
to which on^ goes, as in (159), (160). Moreover, one might claim that 
raahati, rtah in examples such as (165), (167) are forms of a verb that 
has undergone semantic specialization, meaning 'afflict, attack', so that 
it would not be strictly a verb of movement. However, there is no escaping 
the fact that an accusative of goal construed with gam is treated in the 
same manner as an object accusative, witness (158), and whatever one says 
about raahati, the fact remains that it is transitive. Hence, gam-i, 
arp-i can be nothing other than causatives to transitive verbs. Whatever 
be one's opinion concerning the possible origins of accusatives used with 
verbs of perception, it is difficult to see how these could be equated 
with adverbial accusatives in examples such as (114) , (185) , which can 
rightly be termed accusatives of extent. Similarly, though a genitive 
like somasya in (95) be labelled partitive, there is no obvious semantic 
difference between (94) and (95), and there is no justification for claim- 
ing the accusative somam of (94) is of the same type as accusatives of 
extent found elsewhere. It is also difficult to justify the claim that 
ji and oap-i in examples like (132)- (134) are construed with an accusative 
of content, a cognate accusative. A phrase such as jitim jigyyh in ^B yajnena vai deva imam jitim jigyuh "The gods won this victory 
through the sacrifice' clearly contains a cognate accusative^ j'lttm 'vic- 
tory'. This is not the case in RV 8.76.4ab: ayam ha yena va idam svar 
mavutvata jitam 'This (is Indra,) by whom, with his troop of Maruts, this 
sun was won', since svar 'sun' is totally unrelated to the base ji of the 
participle jitam 'won conquered', nor is vajam in ^ (132)-(134) a cognate 
accusative. It might well be that phrases like vajam jaya svar jitam 
could represent an extension of an original construction with a cognate 
accusative. However, this cannot justify arguing that ji, incontrovertibly 
transitive in Vedic, is somehow not a true transitive verb in examples 
like (132)- (134) because what is spoken of as won is a thing and not a 
person conquered. 

4.3 Two major attempts to show that early Vedic did not have true 
causatives, formed to transitive verbs, thus fail. 24 Given what has been 
shown, then, I consider inescapable the conclusion that earliest Indo- 
Aryan did indeed have true causatives, derived from transitive verbs of 
particular semantic groups. 

5.0 The early Iranian evidence is, as one might expect, more sparse 
than that of Indie. However, as one might also expect, it does not disa- 
gree with the Indie evidence. 

As the Indo-Aryan verb gam occurs with an accusative, a locative, or 
a dative denoting a goal (2,1), so too does the Avestan verb gam. Thus, 
para hyat ... jimaitT 'before ... come' occurs with the accusative ma 'to 
me' in 


(192) Y 48.2ab: vaoea nidi ya tvam vTdva dhura/ para hya-j^ ma ya mang 
paraQa jimaitT 'Lord, tell me what you know, before the retribution 
in your mind come to me . ' 

and in Y 49.1c, gaidt. rrioi 'come to me', gaidi (2 sg. imper.) is construed 
with the dative nidi 'to me'. Further, in 

(193) Y 51.15b: gavd damane ahurd mazda jasat pouruyd 'The Wise 
Lord came first into the house of chant.' 

jasat occurs with the locative garo damane 'house of chant'. From this 
verb~is derived the causative jamayeiti, as in 

(19 A) Yt. 17.20: raekd me haaa aqha zamat vaqhd karanaoiti yd mam 
aevd Jamayeiti yd spitamd zaraQustrd 'He makes it better that I 
leave this earth, Spitama Zarathushtra, who alone makes me leave.' 

where it is used with one accusative, mam 'me'. 

The verb fra par (preverb fra) 'cross over' occurs with an accusative 
denoting what one crosses; for example, 

(195) Y 46.10e: fro tais vlspais cinvatd fra fra paratum 'With all 
these would I cross the bridge of the elector.' 

where fra fra (1 sg. aor. subj.) is used with the accusative paratum 
'bridge' . On the other hand, 

(196) Y 19.6: yasca me aetahmi ar)hvd yat astvainti spitama zaraQustra 
ha'iam ahunahe vairyehe ... fra va... srdvaydt ... Qrvsci-^ tard 
parartumcit Ke urvanam vahistam ahum fraparayeini 'Spitama Zarathu- 
shtra, I will make his soul thrice to go across the bridge (of the 
elector), to the best life, who in this corporeal world recites out 
loud the Ahuna Vairya prayer.' 

contains the causative fraparayeini 'I will make to cross' in construction 
with tard paratumcit 'across the bridge' and two accusatives: urvanam 
'soul' and vahistam ahum 'the best life (paradise)', of which the latter 
denotes a goal. In addition, the causative niparayeinti 'they cause to 
come, bring' in 

(197) Yt. 17.54: md ois ar^ham zaoBranam vindita yd mavdya niparayeinti 
'No one shall have part in these oblations which they bring me . . .' 

is construed with the accusative yd '(the oblations) which' and the dative 
mdvdya 'to me' . 

An object of perception can be denoted by an accusative or a genitive 
in Indo-Aryan (2.2), and this is so also in Iranian. Thus, the Avestan 
cognate of Sanskrit sru, the verb sru (2 sg. opt. surunuy&) takes the 
accusative complement yasnam 'prayer' in Y 68.9: surunuyd. no yasnam 
ahurdne 'AhuranT, hear our prayer', but in Yt 10.32: surunuyd. no miQra 
yasnahe 'Mithra, hear our prayer', the same verb occurs with the genitive 


yasnahe. In addition, middle forms of sru are used in the same manner as 
those of sru, meaning 'be heard', 'be heard of (2.6). For example, 

(198) Y 34.12ab: kat toi razav^ kat vaH kat va stuto dat va 
yasnahya/ sruidyai mazda fra vaoca ... 'I'/hat is your command, what 
do you wish in praise or prayer of veneration? Wise one, proclaim 
it that one might hear ..." 

contains the medio-passive infinitive svuidyai 'that ... be heard' (cf. 
Benveniste 1935:76), and Y 46.14b: fea va frasruidyai vast~ 'Who wants to 
be famed?' has frasruidyai 'to be heard of, famous'. Further, the finite 
form spuye (1 sg. mid.) is used in this sense in 

(199) Y 33.7ab: a ma idum vahista .../ asa vohu manaqha ya sruye 
pard magaorio 'Come to be, best ones, with Truth and Good Conception, 
that I be heard before the priests.' 

Corresponding to this medio-passive, there is a causative srav-i', for 
example, Y 49.6cd: ... yaOid z sravayaenid tarn daenam ya xsmavato ahurd 
'. . . how we may make heard this your religious insight. Lord'. This 
causative is also used specifically with reference to reciting sacred 
texts, as in (196). 

The Avestan verb vard is cognate with Sanskrit vrdh (2.7) and used in 
the same meanings. First, it means 'grow, increase': 

(200) Yt. 13.68: x'^aepaiQe rid dair^hus fraSataeaa varsSdtae ca 'Our 
land is to flourish and grow. ' 

In addition, the active (3 sg. varadaitz) is transitive, meaning 'make 
grow' ; for example, 

(201) Y 28.3: ya va asa ufydni manasad vohu .../ mazd^cd ahuram 
yaeibyo x^aQramcd .../ varadaitT drmaitis a nidi rdfa&rdi zavang 
jasatd 'Come to my calls, to help me, I who will praise you in chant 
with truth: Good Conception, the Wise Lord, and ruling power, for 
whom Devotion makes grow their ruling power.' 

Moreover, the derivate yorJ-i is used in the same sense as the active: 

(202) V 2.4: dat me gaeQa vara6aya 'Then, (Yima,) make my world 
increase. ' 

Similarly, the primitive verb paiti.irinaxti 'abandons' and the derived 
verb paiti ... raecayeinti 'abandon' are equivalent: Yt. 14.47: ko 
rasnOm paiti.irinaxti 'Who abandons Rasnu?' Yt. 10.41: paiti Qrdtdra 
yazata ... vasmano raecayeinti 'The gods who protect them abandon the regi- 
ments'. This too is paralleled in Indo-Aryan (2.8, 2.10). 

6.0 Since the early Iranian data are in harmony with those of Indie, 
I think it safe to say that the causative system found in earliest Vedic 
is inherited from Indo-Iranian. Indeed, even if there were no Iranian 
evidence, one should be led to this conclusion. For, as has been shown. 


the trend within Indie is for causative formations to be extended from an 
earlier limited domain, so that the early Vedic system, in which true 
causatives were formed to transitive verbs of particular semantic groups, 
could hardly be an Indo-Aryan innovation. What I call the old system, 
then, had the following network of contrasts. 

II. Verbs of movement, perception, consuming, and verbs meaning 
'say' or similar: For these, there was a three-way contrast among tran- 
sitive, passive, and causative (2.1-2.4, 2.12). In addition, a causative 
could be related to either an active or a passive, so that two causative 
constructions could contrast, meaning 'C causes A to V B' and 'c causes 

B to be 7-ed by /I ' (2.6). 

Ill- Other transitive verbs: For these, there was a two-way contrast 
between transitive and passive. The contrast between causative and tran- 
sitive did not obtain (2.8, 2.10). 

III. Intransitive verbs: There could be no contrast between transi- 
tive and passive, nor was there one between transitive and causative (2.5, 
2.7, 2.10). 

In the innovating system, there is a contrast between transitive and 
causative for all transitive verbs. Moreover, by the time of Panini the 
contrast between causative sentences meaning 'C causes A to V B' and 'c 
causes B to be V-ed by A' was being obliterated. However, the old system 
leaves its traces in syntactic dintinctions : 

J9, Verbs of movement, perception, etc. (J^ above) as well as intran- 
sitive verbs: Causatives of these enter into construction (69c), in which 
an accusative denotes the agent of the non-causal act. 

JJ5 . Other transitive verbs: Causatives of these regularly enter 
into construction (70c) , in which an instrumental refers to the agent of 
the non-causal act. 

The old system leaves another trace in that some derived verbs such as 
rriarj-i, var-i are also used without causative value (1.4, 2.10). 

My theses, then, are these. First, earliest Indo-Aryan had true 
causatives, formed to verbs of particular semantic groups (J^ above), and 
this state represents an inheritance from Indo-Iranian. Secondly, there 
took place in Indo-Aryan at a time certainly prior to Panini a series of 
syntactic and morphologic changes, involving middle and passive forms as 
well as causatives, which led to the generalization of causative forma- 
tions (3. 1-3. 3). 25 


Ait.Sr. : Aitareyaranyaka; Ait.B: Aitareyabrahmana; AV: Atharvaveda; 
KBU: Kausitakibrahmanopanisad ; KS: Kathakasamhita; K^S: Katyayanasrau- 


tasutra; MS: Maitrayanisamhlta; RV: Rgveda; ^Ar.: Sankhayanaranyaka; 
5B: ^atapathabrahmana; TB: Taittirlyabrahmana; TS: Taittiriyasamhita; 
V: Videvdat (Vendidad); VS: Vajasaneyisamhita; Y: Yasna; Yt.: Yast. 

*Thls is a revised version of the Herman Collitz Lecture delivered 
at the 1978 Linguistic Institute on July 25, 1978. An abbreviated version 
is also to appear in the Golden Jubilee Volume of the Vaidika Samsodhana 
Mandala, Poona. I with to thank H. H. Hock and Bh. Krishnamurti for 
helpful discussions on the matters treated here and G. Dunkel for valuable 
comments on an earlier draft. I am also grateful to the National Endow- 
ment for the Humanities for support I received while writing this. 

For a brief presentation of Panini's derivational system, see 
Cardona 1976b (pp. 146-8 on causatives in particular). 

I cite such isolated forms in their pre-pause variants, with -h etc.; 
different ones may appear in sentences cited, due to phonologic environ- 

I say 'of itself because combinations such as adhi as 'remain in, 
rule over' are transitive. 

Thus Panini, who treats the rice spoken of in these sentences as 
agent (Cardona 1974:241-3). Forms of the type paayate 'is cooking' are 
not used for certain verbs; e.g. dugdhe 'gives milk' (duhyate 'is milked'), 
namati 'bends'. 

a locative is also used; see (76). In the 
Sanskrit of Panini's time, a verb such as vis 'enter' could still take a 
locative complement, but not if used with the preverbs abhi, ni; abhi vis, 
ni vis govern accusatives. 

The use of a genitive in construction with certain verbs is more 
widespread in earlier Sanskrit (see 2.2, 2.3). I have omitted many 
details concerning locative and genitive complements with particular verbs 
in later Sanskrit. 

^Thus Panini; see Cardona 1974:289 n 49. 


That is, the nominal N2 can be followed by different endings, as 


N2 in b is followed by a nominative ending if it has an accusative 
ending in a,; otherwise, the same ending occurs with N2 in both. 

V-i is a causative base derived from the primitive verb V of a, b^. 

according to which iiiC is optionally added to bases which follow, up to 
and including dhrs (see Pathak-Chitrao 1935:668 [immediately preceding 
verb 1807]), without causative value; ara-i etc. given above are among 
these verbs. 

241. Forms of the type (50) asi, (58) apaci have 
the affix CtN, which in Panini's derivational system is introduced before 
the ending ta. This ending is then zeroed. Passive forms other than the 
ones noted simply have middle endings. 

'protect', bhuj has agential forms with parasmai- 
pada affixes, bhunakti etc. 

I have of course given only a bare outline of the distribution, 
and cannot consider more details here. 

Here again, I must omit many details. Note nevertheless that bodh-i 
([2b] bodhayati) does not have a middle under the condition given. Nor 
does a causative like bhoj-i ([3b] bhojayati) , to a verb of consuming, 
though pay-i ([96] paydya) can have a middle. 

I do not imply, by speaking of reflexives, that one should consider 
a reflexive pronoun implicit in such sentences; see Cardona 1976b: 148, 

For Iranian parallels, see 5. 

19 f 

In Vedic, genitives construed with sru regularly denote persons to 

whom one listens, but in Iranian a genitive can also signify a thing 
listened to; see 5. 

20 » 

That a comparable passage (RV 7.60.6b) has acetasam (ace. sg.) 

'lacking insight' does not affect the grammatical relation noted. 

21 _ . _ _ . 

Cf. also aviv bhu 'become apparent', avis kr 'make apparent', con- 
strued with a dative denoting one to whom something becomes or is made 


This is true also of modern Indo-Aryan, where distinct causative 

constructions still exist for different types of transitive verbs. On the 

other hand, in major languages such as Gujarati and Hindi the double 

accusative construction is not used. Instead, an indirect object form 

denotes the agent of the non-causal act. I cannot discuss this or facts 

from middle Indo-Aryan in the present paper. 


I have not discussed here the aorist type which is regularly paired 

with causatives, the reduplicated aorist, but this does not affect my 

argument . 


I have concentrated on Jamison's and Thieme's views because these 

are the only scholars who have truly attempted to justify their positions. 


Earlier, Gaedicke (1880:277) proposed an explanation which was refuted by 
Thieme (1929:20). Delbriick (1888:226) suggested that an instrumental 
appears in a casuative construction because the subject appeared so much 
to be the main participant that the second agent was demoted to being a 
means. This fails to explain why double accusative constructions also 
were used. Most recently, Haudry (1977:382-9) has treated causative 
constructions in a manner which is vague, ill-founded, and fails to 
account for the chronology of different constructions. I discuss his view 
in a review of his book, to appear in Kratylos. 


In order to avoid ending on a note of speculation, I have not gone 

on here to discuss the question of causatives in a broader Indo-European 

context, though I plan to take this up. For the moment, let me only note 

that there is sufficient evidence from subgroups such as Germanic and even 

Italic to suggest that in Proto-Indo-European too causatives may have 

been semantically restricted, at least so far as concerns morphologically 

derived causatives. In addition, pairs such as niarsti , marjayati (2.8, 

2.10, 5) doubtless represent a merger of forms that originally belonged to 

distinct categories, which brings up the question of iteratives. For the 

instant, I think it sufficient to emphasize that Indo-Iranian definitely 

had true causatives formed to transitive verbs of perticular groups, so 

that any future study of the Proto-Indo-European type *g^om-ey-e-t-i 

(Skt. gamayati, Av. jamayeiti) , *louk-ey-e-t-i (Skt. rocayati , Av. rao- 

cayeiti, Lat. luoere , as in Plautus, Casina 118 luaebis ... faaem 'you'll 

light the bridal torch') must start with this as a datum; cf. also Lat. 

disco 'learn', doceo 'teach', a causative found in a double accusative 

construction, and a causative such as Gothic dragkjan, which corresponds 

to the transitive drigkan 'drink'. Moreover, one must keep in mind that 

Proto-Indo-European could well have had periphrastic causatives such as 

shown in 2.11. 


Abhyankar, K. V. 1962. The Vyakarana-Mahabhasya of Patanjali, ed. by 

F. Kielhorn. 3rd ed., vol. 1. Poona: Bhandarkar Oriental Research 

Benveniste, Emile. 1935. Les infinitifs avestiques. Paris: Adrien- 

Maisonneuve . 
Cardona, George. 197 A. Panini's karakas: agency, animation and identity. 

Journal of Indian Philosophy 2. 231-306. 

. 1976a. Panini, a survey of research. The Hague: Mouton. 

. 1976b. Some features of Paninian derivations. History of 

Linguistic Thought and Contemporary Linguistics ed . by H. Parret, 

137-58. Berlin-New York: de Gruyter. 
Delbriick, Bertold. 1888. Altindische Syntax. Repr . 1968. Darmstadt: 

Wissenschaf tliche Buchgesellschaf t . 
Gaedicke, Carl. 1880. Der Accusativ im Veda. Breslau: Koebner. 
Haudry, Jean. 1977. L'emploi des cas en vedique. Introduction a 1' etude 

des cas en indo-europeen. Lyon: Editions L'Hermes. 
Jamison, Stephanie. 1976. Functional ambiguity and syntactic change: 

the Sanskrit accusative. Papers from the Parasession on Diachronic 

Syntax, 126-35. Chicago: Chicago Linguistic Society. 
Kuryiowicz, Jerzy. 1956. L'apophonie en indo-europeen. Wroclaw: 

Polska Akademia Nauk. 

1964. The Inflectional Categories of Indo-European. Heidelberg: 

Pathak, Shridharshastri and Siddheshvarashastri Chitrao. 1935. Word 

Index to Panini-sutra-patha and Parisistas. Poona: Bhandarkar 

Oriental Research Institute. 
Speyer, J. S. 1896. Vedische and Sanskrit-Syntax. Strassburg: Trilbner. 
Thierae, Paul. 1929. Das Plusquamperfektum im Veda. Gottingen: Vanden- 

hoeck und Ruprecht. 


Charles E. Osgood 
University of Illinois 

Albert Einstein, with the foresight characteristic of true genius, 
had this to say at the dawn of the nuclear age: 'The unleashed power 
of the atom has changed everything except our ways of thinking. Thus 
we are drifting toward a catastrophe beyond comparison. We shall require 
a substantially new manner of thinking if mankind is to survive.' It 
will be my thesis that a new manner of thinking about our world will 
first require a revolution in how we talk about it. 

I shall try to demonstrate that certain principles of psycholin- 
guistics have relevance to the science, and art, of politics — particularly 
international politics. Psycholinguistics is that facet of the human 
sciences which deals with relations between the characteristics of 
messages and the states of the (human) organisms which exchange them. 
We shall be concerned primarily with the semantic aspects of these 
relations — with the meanings of word forms, with the rules by which words 
are combined into phrases and sentences, with the effects of one's utter- 
ings and scribblings upon others, with the semantic and valuative 
constraints upon what we ordinarily say with our words and sentences, 
and with how we can get out of this bind by some calculated rule-breaking. 

You may well be wondering already just what all this could possibly 
have to do with politics, particularly international politics. To satisfy 
this wonderment is, or course, the task I have set for myself. Through- 
out this talk I shall introduce each new notion with illustrations 
drawn from ordinary language and only then move into illustrations and 
analyses in the realm of international politics. I adopt this strategy 
for two reasons: first, because we are all most familiar with the 
ordinary language we use in communicating with family, friends, and 
storekeepers, and the examples are therefore more compelling; second, 
because examples drawn from contemporary international relations, neces- 
sarily reflecting my personal political values, are liable to be 
controversial and hence lose their force. 

Personification of international relations is a common practice of 
social scientists and statesmen, just as it is among laymen. It is even 
to be found regularly in the writings of those scholars who would be the 
first to deny, as a matter of disciplinary principle, that laws of 
individual behavior apply to social groups. Such implicit personification 
— and consequent uncritical use of folk psychology in the interpretation 
of international relations— is both deceptive and dangerous. But there 
is a paradox here: although one may deplore such naivety, if personifi- 
cation is in fact common practice in thinking and decision-making in 


international affairs, then it must be considered part of the subject 
matter of political science and made as explicit as possible . I shall 
assume that principles of semantics discoverable at the individual level 
can be transferred to the national level — because individual humans do 
participate in the 'behaviors' of nations, because certain individuals 
do play key roles in the 'decisions' of nations, and because principles 
of information-processing (perceiving, interpreting, decision-making, 
executing) do have their analogues at all levels. 


Words are conservative. Although they adapt to changes in the real 
world, they do so very slowly if left to themselves. Assuming a lawful 
and immutable relation between words and things, we humans keep trying 
to force things to conform to the meanings their words have acquired in 
the past. Thus were the semantics of World War I applied in preparing 
for World War II, and the French built a completely useless Maginot Line 
and a false sense of security behind it. So do we seem to be preparing 
for World War III, despite desperately not wanting it, by applying the 
semantics of World War II. 

Wrapped up in the whole process as we are, it is difficult to 
perceive the lag between our language and the world it is supposed to 
describe and interpret. Occasionally a dramatic failure of events to 
conform with our expectations may jolt us into awareness that our world 
is changing — a rash of race riots or a score of people suffocating in 
the smog of an American city. The brighter among us may note that there 
is a constant conservative bias in our predictions about the future — 
sputniks appeared in our skies and the Chinese tested deliverable nuclear 
weapons long before these things were supposed to happen. 

In periods of history when the rate of cultural change was slow, 
words could almost keep up with things and their mismatch was barely 
perceptible. But in the present age, with its exploding population and 
its exploding technology — which means increasing human interaction and 
accelerating cultural change — our semantic maps of the real world become 
outmoded more and more quickly. Add to this the fact that most decisions 
in today's world are being made by men in their fifties, sixties, seven- 
ties, or even eighties — men whose maps were outlined at least 30 years 
ago — and the language gap threatens to become an abyss. Led by old men 
reading old maps, we are trying to find our way through the Wonderland 
of the Twentieth Century, having more contact with the phantasy of words 
than with the reality of things. 

The gap getween word and thing also increases with the remoteness 
of things from immediate, individual experience. As one moves outward — 
away from the intimacy of family, possessions, and the ordinary round of 
living, toward the community, the nation, and the world — the map becomes 
a less and less reliable guide to the territory. For one thing, we 
become less able to correct our verbal maps against the facts of the ter- 
rain. For another, the meanings of our words become increasingly 
dependent upon the words of others. The words with which we talk about 
the most critical issues of our time — nuclear overkill, arms control. 


population control, and Palestinian Coiranandos — are what in the Business 
of Semantics are termed 'assigns'. Their meanings are literally assigned 
by association with other words put together by other people (usually in 
the mass media) rather than from direct experience with things and events. 
We have immediate knowledge about what a knife can do: our knowledge 
about what a nuclear missile can do is rather remote and gutless. Thus, 
irrationally, knowing that one's city is targeted by a 20-megaton missile 
some ten thousand miles away is not as threatening psychologically as 
knowing that a madman with a carving knife is loose in the city streets. 

The Power of the Word 

Words may be conservative, but they can also be powerful. Wherein 
lies the power of the word? It obviously doesn't lie in the noises and 
squiggles per se. It lies, rather, in a very remarkable relation between 
these physical manifestations, called 'signs', and certain processes in 
language-users, called 'meanings'. This is the representing relation . 
Words come to evoke in their users some distinctive representation of the 
things referred to, and these representations can be manipulated symboli- 
cally much more easily than the things themselves. Thus we can behave — 
appropriately or otherwise — with respect to the not-here and the not-now. 
A Jinrniy Carter can devise strategies in which nation-names move about 
his mind like chessmen on a board; words about things remote can give 
him hope (President Nasser makes a peace initiative by going personally 
to Israel to talk peace in the Middle East) or they can give him fear 
(the possibility that Communist Cuba and Russia may threaten the dominance 
of the West in Africa and threaten 'balance of power' stability). 

Exactly what noises and squiggles will be used to represent what 
things and events is largely arbitrary — which is one reason why human 
languages are mutually unintelligible. When one is born into a given 
language-culture community, he unwittingly assumes a social contract to 
use noises and squiggles the way others in his community do — to use the 
noise man to refer to adult male humans, for example. But once the myriad 
contracts of lexical and syntactic usage have been agreed upon, the rela- 
tion of meanings to things is not arbitrary — which is one reason why 
human languages are mutually translatable. 

The meaning of a word can be conceived as a simultaneous bundle of 
distinctive semantic features. These features represent those differences 
in reactions to things and events which, in human experience, have been 
found to make a difference in adjustment to the physical and social 
environments. Thus the representing or symbolizing process utilizes a 
kind of code for highlighting what properties of things and events have 
been found to be important in experience. Recent studies in comparative 
psycholinguistics are beginning to make it clear that human groups, 
regardless of differences in language or culture, tend to use very similar 
systems of semantic features — as if, in general, they have found the same 
kinds of distinctions important to make. 

In some of our own research, for example, we have been finding that 
humans highlight certain properties of interpersonal behavior, but not 
others, as semantic features in their interpersonal verbs. I'fhat is also 


interesting is that they use many of the same verbs with the same features 
to talk about international relations, as if nations were people. There 
is an Associative-Dissociative feature (help vs. hinder, proteat vs. 
attack), an Ego- vs. Alter-oriented feature (exploit vs. aid, manipulate 
vs. advise), a Superordinate-Subordinate feature (dominate vs. submit to, 
lead vs. follow), a Future-Past feature (promise vs. apologize, threaten 
vs. retaliate), and so on. Needless to say, there is considerable danger 
of misr epresentation when we apply, willy-nilly, the semantics of inter- 
personal relations to the things and events of inter national relations. 

Since the semantic features on which words are coded can never 
exhaust the properties of things which might be represented, it is inevita- 
ble that words must be abstractions. In this sense, words are caricatures 
of things. Although this is most obvious for big, impressive symbols like 
Colonialism or Blaak Power, it holds also for little ordinary words like 
dog and pencil. It is precisely in this selective abstraction from reality 
that the power of words lies . Those properties of things that are seman- 
tically coded are sharpened in thinking; those properties of things that 
are not so coded are blurred in thinking. A man can freeze to death for 
lack of kindling wood and yet be found with a pocketfull of wooden 
pencils — because pencils are coded 'to write with', not 'to make fires 
with' . 

What facets of reality are sharpened or leveled by words depends on 
what properties of things have made a difference in the past use of the 
language. This is the conservatism of words, and it can make us attend to 
features of events which are now irrelevant and be oblivious to features 
which are now critical. Bound by an antiquated semantics of Power 
Politics, diplomats keep wary eyes on the shifting patterns of alliances 
and scurry about trying to maintain mutual security pacts — NATO's, SEATO's, 
CENTRO's, and PREPOSTRO's. The semantic coding of mutual security pact 
simultaneously fixes thought on the solidarity of one group and on its 
antagonism toward another. But the differences within the groups so 
segregated and the similarities between their particular members are 
obscured. Within the so-called Free World there are members which, by 
most criteria, are more similar to members of the so-called Communist 
Bloc, and vice versa. 

The semantic features of words may either define or imply. For 
English speakers, the term father includes Male, Adult, and Parental among 
its defining features, and Dominance and Goodness among its implicative 
features. Any human we call a father must be an adult male who has 
progeny, or else we are breaking our linguistic contract, but obviously 
he need be neither good nor dominant. Substitution of implicative for 
defining features — that is, the use of emotive language — is the trademark 
of propaganda. Buck teeth and a spider body obviously did not define a 
Japanese human, but it did make him more killable in World War II. Just 
as the same odor may be called an aroma or a stench, depending on how one 
wants the listener to feel about it, so may the same guerrillas be called 
freedom fighters, rebels, or terrorists, depending upon how one wishes 
the listener to feel about them. 

Words of Power 

Among the few remaining places where myth and fantasy still have 
power for modern man are advertising and international relations. 
Although we cannot eat words, smoke words, or shave with words, the 
advertising industry makes a valiant effort to achieve that end; incan- 
tations, spells, contagious and analogic magic are part of the trade. 
International politics has its incantations ('the right to self- 
determination'), its spells ('we have a commitment'), its contagious 
magic ('a threat to freedom in Asia is a threat to freedom in Podunk, 
U.S.A.'), and its analogic magic ('another Munich'). Both advertising 
and international relations have everything it takes for myth-making — 
remoteness from the individual, conflict between ultimate Goods and Evils, 
and near-omnipotence of the storytellers. And, in both, the myths created 
do serve many social functions: they provide an illusion of understanding, 
simple prescriptions for ordering a complex world, and the promise that, 
if the illusions are believed and the prescriptions followed, everyone 
will live happily ever after. In the political arena, at least, the 
storytellers are often their own most gullible audience. 

The words of international power politics are typically analogic. 
There are analogies from interpersonal relations; nation-states become 
We' s and They's who assume military postures, who glare at each other eye- 
ball to eye-hall, who harass or intimidate each other, who trust or are 
suspicious of each other. There are analogies from 19-century physics, 
with balances of power or power vacuums, with slippery slopes or rows of 
dominos along which vague forces operate with some natural inevitability, 
with attractions and repulsions, and centers toward which power gravi- 
tates. Analogies have the same sharpening and blurring effects upon 
thinking that ordinary words do, only more so; they raise the feature 
analogized to a dazzling prominence which obscures other, often more rele- 
vant features. 

The pseudonyms which pepper dialogue in the nuclear age carry this 
attenuation of word from thing even further, adding semantic features 
which do not exist in the thing and neatly countermanding features which 
do exist. The title CAMELOT conferred a romantic, even chivalrous, tang 
to an ill-fated U.S. Army project designed to study the causes of revolu- 
tions. To name an anti-ballistic missile system SAFEGUARD certainly must 
make its possessors feel more secure. A touch of nobility is added to 
raw power when intercontinental ballistic missiles are named THOR, 
JUPITER, ATLAS, ZEUS, and POLARIS — although I miss the ultimate in seman- 
tic deception which would be a missile named VENUS! 

Acronyms — short-cuts which compress the forms of phrases into their 
initial letters — carry the attenuation of word from thing even further. 
They also put a great burden on human memory. Debates on the status of 
NATO and on the merits of SALT negotiations can be carried on with only 
remote reference to the things involved. They become 'things' of sorts on 
their own, things to be defended or attacked by partisans or critics with 
more reference to personal and political loyalties than to the complex 
events and arrangements in the real world they actually represent. When 
an acronym is devised before a thing is given an ordinary name, contact 


with reality Is liable to disappear entirely. A playful example from the 
argot of military space scientists is EGADS, which names the system used 
to destroy a malfunctioning missile after it has been launched — 
translation: 'electronic ground automatic destruct sequencer'! 

Much of military jargon is designed to dehumanize things and thus 
protect the speaker or listener from feeling moral qualms. In The Domes- 
day Dictionary of Kaplan and Schwerner (1963), for example, we find that the 
tertiary effects of nuclear blast are 'damage received by displacement of 
the biological target, or blast flight' and this is further analyzed into 
'differential displacement' (which translates as violent separation of 
hand, arm, leg, head, etc.) and 'total displacement' (which really 
translates as being tossed into the next block in one chunk) . . . The 
military do have a way of juggling semantic features. We never bombed a 
simple village hut in South Vietnam, it was always a 'VC structure'. 

Here are a few words of power to conjure with: The word power 
itself is often confused, by those who think they have it, with the word 
forae. The meaning of power, unlike force, includes the feature of being 
able to produce a desired result. According to political scientist Murray 
Edelman, 'force signals weakness in politics, as rape does in sex.' If 
the use or threat of force serves to create the very outcome it was 
designed to prevent — for example, setting up military bases in Thailand 
resulting in a sharp increase in communist counteractivity — then it is 
self defeating and certainly not a demonstration of power. 

Peace and freedom are nice words, but they are otherwise rather 
empty semantically. And being empty, they lend themselves to bizarre 
uses. Is peace a world spinning toward eternity triggered for mutual 
annihilation and kept from it only by mutual fear? This is what the 
phrase 'peace through military strength' really means today. The name of 
freedom, along with the resounding phrase 'law and order', is often used 
to justify its exact opposite — to rationalize subservience to authority 
and to condone suppression of the freedom of others. 

The term coexistence does reflect an awareness of the irrationality 
of mutual nuclear annihilation — but what more than this does it mean to 
people? At an International Convocation on the Requirements of Peace, 
held in 1965 by the Center for the Study of Democratic Institutions, 
Paul-Henri Spaak of Belgium said: 'coexistence means only that you and 
we renounce war. For the rest we fight, trying to make the best demon- 
stration we can that one system or the other is best.' 

In is interesting that most participants in the conference used the 
phrase peaceful coexistence to describe this state of affairs. Surely, 
competitive coexistence would more aptly represent what they had in mind. 
There are many ways one can enliven the word coexistence, and by so doing 
point forcefully to quite different real-world possibilities: one could 
talk about fearful coexistence, pointing to the inherent nature of mutual 
nuclear deterrence; one could talk about tolerant coexistence, pointing 
to acceptance of differences without trying to change them, and even 
co-operative coexistence, pointing to co-operation rather than competition 
despite differences in systems. 

In Defense of Ambiguity 

Ambiguity can hurt. It does when the same word is used in the same 
context , but with different meanings for different people. Such is the 
case with socialism and capitalism — to say nothing about democracy as 
used by Americans vs. Russians, for example. But for most ordinary words, 
potential ambiguity is effectively eliminated by context . I'Jhen someone 
says duck in the barnyard, I will not crouch as I will when the same word 
is uttered in a sand-lot ball park. 

Proponents of One World Language as the salvation of Mankind often 
include elimination of ambiguity as one of the planks in their platform 
— one word one meaning! Not only would this increase the size of our 
working vocabulary inordinately — the little word play, for example, has 
some 40 different meanings, at least 15 of which are quite common — but a 
language free of ambiguity would be like sand in the fluid coupling of 
man-to-man and man-to-government. Some degree of ambiguity is necessary 
for a smoothly functioning society. 

Political scientist Murray Edelman also makes the insightful observa- 
tion that administration of laws governing ordinary social behavior tends 
toward a compromise between the letter of the law and the vagaries of 
human nature — too loose administration sets man against man and too strict 
sets man against government. When law is enforced as if it were a command 
rather than a virtuous generalization, around which a game can be played, 
it becomes a trial of force rather than a trial of wits. 

In international politics there appears to be little tolerance of 
ambiguity. Negotiators strive for months to overspecify the inherently 
unspecif iable, to wring all possible ambiguities out of every statement in 
every language. At the Peace of Westphalia in 1648 it took six months for 
delegates to decide in what order they should enter and be seated in the 
negotiating chamber! How many weeks did it take to decide on the shapes 
and markings of tables for the U.S. /Vietnam negotiations in Paris? Inter- 
national politics ^ more a trial of force than a trial of wits, and 
ambiguity is therefore more threatening than challenging. 

A prime example of intolerance of ambiguity is reification of The 
Nation as the unit in international affairs. Since the flowing forests 
and oceans do not recognize these creations of the human mind, we put up 
boundary markers, erect walls and fortification, establish bordercrossing 
restrictions, try to impose language homogeneity within boundaries, define 
invisible territorial extensions into the seas and brightly color our 
maps in different hues so that children can learn just what is really 
where — all to reaffirm that The Nation is indeed a unitary thing like other 
things that have names. It then becomes easier to personify nations as 
Actors in a great global game, and harder to appreciate either the simi- 
larities across boundaries or the differences within. 

Part of any semantic revolution must be an increase in the ambiguity 
of nation names, by using language which deliberately levels distinctions 
between nations and sharpens differences within them. This implies gradual 
dissolution of the nation-state as the prime political unit — which, surely. 

is heresy. But in a nuclear age there may be more security in disunity ; 
out of creative chaos there could come both a greater tolerance of ambigu- 
ity and more accurate semantic mapping of the complexities of the real 
world — paradoxical as that may seem. Fortunately, human languages contain 
the seeds for their own revolutions, so let us turn now from words to 


We refer to our species as homo sapiens, but homo loguens would be 
more accurate. Other animals have been shown to symbolize the not-here 
and not-now to some degree — witness the way honey bees can represent the 
distance and direction of new food sources by their dance; other animals 
certainly manipulate the environment toward their own ends — witness the 
dams built by beavers. But no other animal really talks, in the sense of 
using a shared set of abstract rules for making propositions about the 
world. This is at once our great advantage as a species and, potentially, 
our downfall. Ability to propositionalize about things leads to control 
over them, but control over things without understanding and control over 
ourselves invites disaster. 

Humans share many needs with lower organisms — needs to procreate, to 
feed, to avoid pain, for example. But the fact of propositional language 
creates uniquely human needs, particularly the need to know and the need to 
do something about what is known. When one can conjure with the not-here 
and not-now, one often assumes some responsibility toward it. Myths, 
religions and ordinary daily news are testimony to the need to know, or at 
least have some illusion of knowing, and knowing impels doing things in 
anticipation of events — making an offering at the shrine of one's God, 
planting crops at a time deemed propitious by omen or science, or even 
building a fall-out shelter in anticipation of a nuclear attack — which has 
now become sheer nonsense. 

Valid propositions about the environment, either informal ('common- 
sense') or formal ('science'), have given humans considerable freedom from 
its harsh constraints. But scientific control of the environment has 
already reached the point where we now have a new kind of freedom — a 
freedom ^, which is positively embarrassing in its riches: freedom to 
make deserts into gardens, freedom to manipulate our own numbers and even 
our own genetic makeup, freedom to bring any part of the world to any 
doorstep by satellite communication. The possibilities are staggering to 
the imagination, but choice among them demands fresh and apposite proposi- 
tions about the world. 

Rules for Creating and Understanding Sentences 

We make propositions by combining words into sentences of a particular 
type. Propositional sentences make assertions about their topics, and they 
can be signaled by I state that, I assume that, I claim that, and the like 
— but we are rarely given this warning. Unlike words, propositional 
sentences can be tested for their truth value. Take, for example, the 
sentence Tom is a thief. One can ask significantly, is it true that Tom 
is a thief? One cannot ask significantly, is it true that Tom? or ask 
is it true that thief? 


When prepositional sentences come from prestigious persons and make 
assertions whose truth values cannot be readily tested — either because of 
the remoteness of the things referred to or because of the inherent 
ambituity of the concepts involved — their power of conviction is compelling. 
Faced with the assertion of a Pentagon expert that a missile gap exists 
and threatens our security, what is hapless Mr. Congressman to do but vote 
for a bigger defense budget — even though this assertion eventually proves 
to be false? Interestingly enough, resistance to such assertions, when it 
comes, is usually not so much based upon the truth-value of the assertions 
per se as upon the veracity of the speaker as a source. The peculiar 
power of sentences comes from the fact that they can assert something about 
the topic to be true — even this sentence, what I say (topic) is not true 
(commentary) . 

The meanings of words are constrained by the sentences in which they 
appear. Let us represent, for convenience, the meaning of a word as a 
string of pluses, zeros, and minuses — a strip-code for its bundle of 
semantic features — and then ask ourselves what may transpire when two or 
more words are forced to interact within phrases or sentences. (1) If the 
coding of one word is the same on a certain feature as another word with 
which it is pressed into syntactic combination, it will intensify that 
feature of meaning — as the adjective sudden does for the noun surprise in 
the phrase sudden surprise. (2) If one word adds distinctive codings where 
the other has zeros, then it serves to modify the meaning of the whole — 
as in the phrase sudden anger (one can also experience slow, burning 
anger) . (3) If one word has codings opposed to those in another word on 
the same semantic features, then what are called 'semantic anomalies' are 
produced — as in sudden aomplaaency or meek contempt. This 'feeling of 
incongruence' is a most significant human capability, of which I shall 
have more to say momentarily. 

The assertion Tom is a thief momentarily punches some of the semantic 
codings of thief onto the topic of the sentence, Tom, thereby modifying 
the meaning of Tom — quite congruently for the arresting officer who caught 
him with his hand in the till but completely incongruently for Tom's 
Mother! Yet they both understand what the speaker intends, whether believ- 
ing or not believing him. This momentary fusion of semantic features is 
essential for understanding phrases and sentences, but it would be disas- 
trous if the change in word meanings were permanent. Fortunately, words 
tend to snap back into their normal semantic shape after being bent to 
the purposes of sentences. Otherwise our semantics would become a murky, 
meaningless shambles. 

The influence of word upon word within sentences can be quite subtle. 
In English the word neighbor is not coded for sex. Yet if I say My new 
neighbor is pretty, you automatically assume that I am referring to a 
female, and if I say My new neighbor is handsome, you assume that it is a 
male. The sex codings of pretty and handsome are being momentarily con- 
ferred upon neutral neighbor. Sentences graced by the phrase most people 
carry a sense of ubiquity that far exceeds its casual method of calcula- 
tion. Use of the phrase Administration Spokesman confers an aura of 
intimate access and sophistication upon the source which is hard to escape 
(be it news secretary, just secretary, or even chambermaid). Conversely, 



one of the many ways of 'belittling' sources is to make the propositional 
nature of their assertions explicit. To report that The Russians claim 
that the Communist Summit Conferenae was a demonstration of unity within 
the world of socialist societies is to cast the shadow of doubt upon what 
they are claiming. Propositional sentences are most compelling when 
their propositional status remains implicit. 

Students of the Oxford School of Philosophy, notably Wittgenstein 
and Ryle, have suggested that words can only have meaning when used in 
sentences. But surely this is a two-way street. Just as sentences 
constrain the meanings of words, so do the meanings of words constrain the 
sentences that can be produced in ordinary language. 

At the grossest level there are syntactic rules : correctly speaking 
English, I cannot create sentences of words coded grammatically as Object- 
Subject-Verb in that order — for example, Garlic I taste — although ordinary 
speakers often do produce just such non-sentences when highly motivated 
by the logical object {garlic). At the finest level there are strictly 
semantic rules : it is semantically anomalous to say She plead with him 
tolerantly, because plead with is coded for a Subordinate relation of 
she to him whereas tolerantly implies a Superordinate relation of she to 
him. This is the level at which language can most subtly constrain how 
we think about things, and some semantic juggling is often necessary to 
make true propositions. I think it would have been accurate to say in 
January 1973 that Secretary Henry Kissinger plead tolerantly with Presi- 
dent Thieu of South Vietnam to accept the terms of the Paris Accords — 
tolerantly because of relative power status of the nations involved. 
There are, finally, pragmatic rules of usage — rules which reflect what we 
know, or think we know, about the real world. Thus the sentence He shot 
at the car with a Reagan sticker is unambiguous only because we know that 
you can't use bumper stickers to shoot with. 

Why do I claim that sentences are potentially radical? For one thing, 
whereas the words available to us at any time are finite, the number of 
sentences we can create with them is potentially infinite. For another 
thing, most sentences we encounter are novel as wholes; I know that most 
of the sentences in this paper were novel to me as I wrote them, and I 
assume they are novel to you, yet I have reason to hope that we both 
understand them. Furthermore, sentences can be whimsically nonsensical 
or deliberately nonf actual — we can lie with sentences. But most impor- 
tantly, sentences 'do things' to the words they use, forcing new and 
sometimes revealing interpretations of reality. When, of a certain woman, 
I say She will make someone a nice husband — or, so that Woman's Lib. won't 
come after me. He will make someone a nice wife! — I am breaking semantic 
rules, to be sure, but I am also offering a thumbnail personality sketch 
of the lady — or gentleman — in question. 

Sources and Effects of Sentences 

Now let's return to the world of international politics. Precisely 
what sentences about this world are we likely to create and how are people 
likely to interpret them? The actual sentences about nations and their 
relations that we hear are a vary small subset of the possible sentences 


which could be produced, even within the rules of sentencing just outlined. 
The reason is simply that we usually say 'what is in our minds' — but this 
will require a bit of elaboration. In my analysis of why we create the 
sentences we do and interpret them the way we do, I shall emphasize only 
one semantic feature — the evaluative or attitudinal feature. But this 
aspect of meaning certainly carries the most weight in politics. 

Try to imagine a space of some unknown number of dimensions within 
which the concepts we talk about appear as points. The dimensions of 
this space are the semantic features already discussed. The most promi- 
nent dimension of this space — let's say the up and down of it — is the 
evaluative, or good-bad , dimension. This dimension of meaning is based 
upon the human bedrock of pleasures and pains, of rewards and punishments, 
of threats and promises. 

One's own Ego, or self concept, is good by virtue of biological 
imperative — at least until one is taught about sin or becomes old and 
cynical — and this Ego early-on becomes the arbiter for evaluating other 
concepts. Signs of immediate things like PAIN, DIRT, and DANGER, which 
hurt, shame, and threaten Ego, tend to move downward toward maximal 
Badness. These concepts in turn become pivotal evaluators, pushing and 
shoving other, more remote things , like FRIEND vs. STRANGER and GOD vs. 
DEVIL, into appropriate positions along this evaluative dimension — simply 
by virtue of being associated, positively or negatively, with them in 
propositional sentences. Exposed to assertions like My Daddy says Blacks 
are lazy or My teacher says policemen protect us from thieves, race 
concepts, occupational concepts, nation concepts, and all the rest gradu- 
ally acquire values. The underlying principle in this primitive mapping 
of semantic values onto things is pressure toward cognitive consonance 
(or, if one prefers the negative, avoidance of cognitive dissonance). 
It is by such a basic process that the Unfamiliar is assimilated to the 
Familiar and thereby given meaning. The Cold War is steadily fired by 
application of its own tired rhetoric, which regularly appeals to the 
pivotal concepts of GOOD and EVIL for its sustenance — thus being a kind 
of Holy War. 

Assertions connect topics with commentaries. The connectors (verbs) 
may be either associative (A is B, A likes B, A helps B) or dissociative 
(A is not B, A dislikes B, A hinders B) . If both topic and commentary 
have the same evaluative sign (both good or both bad ) , associative asser- 
tions will be consonant and dissociative assertions dissonant. If both 
Russia and China are held to be communist and therefore evil, it is 
cognitively consonant that Russia should support China — and needless to 
say, from the State Department on down, many Americans found the late 
President de Gaulle's associative moves toward both Russia and China back 
in the mid-1960' s most unsettling. It is also unsettling for some folks 
now when there is a developing detente in the relations of the U.S. and 
mainland China. 

If topic and commentary have opposed signs (one is good and the other 
bad), then dissociative assertions become consonant and associative ones 
dissonant. In human conflict situations it is easy to believe aggressive 
dissociative assertions made by one's opponent {W will bury you) and 


hard to believe his conciliatory associative assertions {We can coexist 
peacefully) . It just simply is congruent in Black-and-White thinking for 
BAD GUYS to threaten GOOD GUYS, and vice versa. This creates a constant 
bias in oredihitity which encourages escalation of conflicts. 

Most people, most of the time, create sentences that are consonant 
with their own systems of attitudes and beliefs. This is one reason why 
they produce only a small subset of possible sentences about the world. 
If someone believes that Communism is evil, he is not likely to produce 
sentences like I favor the Communist system or like Communists can be 
trusted, and so forth for myriads of other potential sentences. But 
what happens when a person is constrained to produce assertions that are 
inconsistent with his own attitudes and beliefs? What happens when he is 
exposed to assertions made by others that are inconsistent with his own 

This is where what I call psycho-logic enters the picture. This is 
an entirely illogical, but very potent psychological, process. It is 
designed to restore the comforting state of mental equilibrium — with all 
the Goods and Bads in their right places — whenever this state has been 
disrupted by incongruity. Incongruity creates cognitive stress, and the 
victim is driven to do something about it; when he does something to the 
world or to himself which restores cognitive balance, it gratifies him 
and makes him feel secure. Psycho-logic may not follow logical principles, 
but it is nevertheless peculiarly significant for an understanding of the 
dynamics of international politics. 

When a person is constrained by his own selfinterest to talk and act 
as if he agrees with certain people and believes certain things, his 
behavior is inconsistent with his own values and he is under constant 
cognitive stress. One resolution is simply to tolerate the dissonance, 
rationalizing it as leading to a larger goal — but people differ in their 
tolerance of ambiguity. One wonders if the late Mr. Adlai Stevenson 
suffered just such prolonged cognitive stress while functioning as U.S. 
Ambassador to the United Nations. Another resolution is to change one's 
behavior, get out of the situation, say what one really thinks; people in 
public life do commit suicide sometimes, politically as well as physically. 
The most likely resolution, however, is a form of psycho-logic: the 
person gradually and unconsciously changes his values and beliefs toward 
consistency with what he feels he must say and do. Men on the accession 
routes to power are peculiarly susceptible to such self-induced 'brain- 
washing' . 

It can be hypothesized that the same dynamics influence relations 
among nations. If the people of nations A and B really dislike and dis- 
trust each other, but they are constrained by their mutual selfinterest 
to repeatedly behave as if they liked and trusted each other, their mutual 
perceptions should become more consistent with their mutual behaviors. Is 
there any evidence in recent history for such processes at the international 

In a significant paper titled 'The Kennedy Experiment', Amitai Etzioni 
documents a real-world test of this hypothesis. Beginning with the late 


President Kennedy's speech at the American University on June 10, 1963, 
in which he announced the first unilateral initiative (that the U.S. was 
stopping all nuclear tests in the atmosphere) , a series of reciprocal 
unilateral steps were taken (the Soviets stopping production of strategic 
bombers, the Americans approving the sale of $250 million worth of wheat 
to the Russians, and so forth), culminating with conclusion of the Test Ban 
Treaty — clearly to the self interest of both sides. Did the predicted 
psychological side-effects occur? A correspondent to the New York Times 
had this to say on June 16, near the beginning of the experiment: 
' . . . there was a new threat of international peace in the air this 
week, the kind of threat that leaves sophisticates smirking and the rest 
of us just dumbfounded' ; but on September 22 the same correspondent had 
changed his tune: 'We have cleared the air and cleared the atmospheres 
and warmed the climate and calmed the winds . ' The Kennedy Experiment came 
to an abrupt end — in Dallas, Texas. 

When a person is exposed to assertions from others which cause him 
cognitive stress, psycho-logic offers a number of defensive mechanisms. He 
may simply deny the assertions and stop thinking about them. More subtly, 
he may accept the assertions intellectually but deny their emotional impli- 
cations; many Americans became quite callous about the use of napalm, 
crop-destroying sprays, and other indescriminate weapons against living 
things in South Vietnam. For insurance against cognitive stress, one may 
selectively expose himself to information that is consistent with his own 
belief system and selectively avoid information that is inconsistent. 

By attributing different motives to ourselves and our opponents — 
benevolent motives to i'TE and malevolent motives to THEY — exactly the same 
behavior by both can be rationalized into the value system. Psycho-logic 
thus creates a double standard of national morality. During the Cuban 
missile crisis, the American government defined the weapons being implanted 
in Cuban soil as 'offensive' in nature, whereas essentially similar Ameri- 
can weapons in West Germany, Italy, and Turkey were defined as 'defensive' 
— and the Soviets, of course, applied exactly the same definitions, but in 
reverse. After the signing of the Paris Accords, both the South and the 
North Vietnamese flagrantly violated its provisions — yet, when the l^fE's 
did it, it was practical military realism, but when the THEY's did it, it 
was either crass deceitfulness or naked aggression. 

Once unleashed, sentences fly about willy-nilly in all directions. 
This is praticularly true with modern electronic communications and it 
poses particular problems for political messages. From the viexv^joint of 
the political speaker , he is always dealing with multiple audiences which 
have multiple interests and multiple relations to him. When a head of 
state makes a speech, he can be sure that his own party, the opposition 
party, labor union members and industrialists, his o\m military and the 
opponent's military, to say nothing of general publics both at home and 
abroad, are all listening in, and what falls harmoniously on the ears of 
one is bound to fall gratingly on the ears of another. Occasionally one 
may be able to use a private code for the benefit of some segment of his 
audience. Ex-Secretary Rusk used to refer to the capital of Mainland China 
not as Peking but as Peiping, the name favored by the Nationalists and 
therefore an insult to the Communists — of which the American public was 


quite unaware. From the viewpoint of the political listener , there is a 
continuous bombardment of conflicting assertions. Within the U.S., one 
expects this, and indeed it is one index of democracy. But since the 
ordinary citizen, no matter where he may be, tends to identify other 
nations as monolithic personalities, conflicting voices from abroad become 
most confusing. When the hard-line assertions of a militarist are inter- 
laced with the soft-line assertions of a U.N. diplomat, and both are 
attributed to the same national entity, that nation obviously is quilty 
of deliberate doubletalk. 

Now, it may be argued that human individuals of even ordinary intel- 
ligence, although susceptible to psycho-logic, certainly are more 
sophisticated in their thinking than this analysis implies. This is quite 
correct. But when all of the complex interplays among the individuals 
who make up nations have worked themselves into the amalgam of national 
policy, what appears in the dialogue between nations seems to represent 
something close to the lowest common denominator of its individual ingradi- 
ents. But even for a sophisticated individual, the remoteness of the 
events talked about and his nearly complete dependence upon the mass media 
render him peculiarly susceptible to psycho-logic when he thinks and talks 
about international affairs. 


Rules are made to be broken. It is only through breaking them when 
necessary that we can force our language into a more faithful portrayal of 
things as they are. When a young lad, accused of a felony, exclaims, 'Ah 
ain't nevah done nothin' to nobody nohow!', he is guilty of a quintuple 
negative at the very least — but his claim to honorable charactet is being 
vividly made. Winston Churchill's coinage of the phrase Iron Curtain 
broke pragmatic, if not semantic, rules of English, but it certainly pro- 
vided an apt characterization of the Cold War situation at the time. We 
need a more lively language to revitalize the stale rhetoric of inter- 
national politics. 

But if rules are made to be broken, then there must also be rules for 
breaking rules — otherwise we have not revolution but chaos. '^^Jhen I use 
a word,' Humpty Dumpty said, in rather a scornful tone, 'it means just 
what I choose it to mean, neither more nor less.' 'The question is,' said 
Alice, 'whether you can make words mean so many different things.' 'The 
question is,' said Humpty Dumpty, 'which is to be master, that's all.' 
There are grains of both truth and untruth here. Being master of one's 
words is not synonymous with being arbitrary in one's use of them. Adap- 
tation of a language to the rapidly changing world is essential, but so is 
continuity of a language with its past. 

As to novelty or innovation in language, a continuum can be traced 
from monotony, through coinage and metaphor, to verbal magic, and ulti- 
mately into chaos. The stale rhetoric of international politics approaches 
the nadir of novelty; words and phrases predict each other redundantly, 
semantic features are all in rapport, and little new information is commu- 
nicated by the parading sentences. If there is too much unpredictability 
— if too many semantic features are in conflict — a pinnacle of novelty may 


be approached, but again little is communicated because of sheer chaos. 
Just as there is an optimum degree of unpredictability in the arts — 
variations on themes which maximize aesthetic pleasure — so is there an 
optimum degree of rule-breaking in language which maximizes communication 
of fresh ideas, whether it be in poetry or politics. 

Effective metaphors break semantic rules optimally. In the literal 
use of English, one cannot say the panther shouted (only humans can shout). 
Speaking poetically, however, one might say the thunder shouted down the 
mountainside, thereby humanizing the thunder; but I, at least, could not 
say the brook shouted down the mountainside — without making it a torrent. 
Thunder and shouting share enough features to allov? overriding of the 
Humanness distinction, but brook and shouting do not — the brook chuckled 
down the mountainside would be fine. I use the term 'verbal magic' to 
refer to the case where a false context is used to support a semantically 
anomalous assertion. My favorite example is a TV beer advertisement: 
After dropping bottles of the brew from skyscrapers, running them over 
with steamrollers, and flinging them against brick walls — with nary a 
scratch to the glass bottles — the assertion is brightly made that this 
beer has indestructible flavor! 

One ingredient of a semantic revolution is a healthy suspicion of 
familiar words and phrases . By way of example, let us analyse the familiar 
phrase, NUCLEAR DETERRENCE. The assertion A deters B carries the implica- 
tions that B must be aggressive (since he requires deterring) and that A 
is both nonaggressive (since he merely deters) and actually benevolent 
(since he is opposed to a potential aggressor) . The phrase >nJTUAL NUCLEAR 
DETERRENCE, demanded by the real-world fact of Soviet nuclear capability, 
is a bit more sophisticated — for one thing, it neutralizes A and B with 
respect to just who is benevolent and who is potentially aggressive — but 
this phrase also has subtle connotations. It has a stable, reassuring 
feel to it — almost like being in a medieval suit of armor — but given its 
foundations on the shifting psychological sands of mutual fear and dis- 
trust, nothing could be much less stable or reassuring. In other words, 
the semantic implications of this phrase simply do not fit its referent. 

Another ingredient of a semantic revolution is imaginative flexi - 
bility with, and tolerance for, unfamiliar words and phrases . Returning 
to the concept of MUTUAL NUCLEAR DETERRENCE, with its illusory assurance 
of stability, it is refreshing to note that one well-known strategist 
dubbed it 'the delicate balance of terror'. In coining this phrase our 
strategist was deliberately violating semantic proprieties; terror is 
certainly not something our language implies can be delicately balanced , 
and yet this is precisely the state of affairs, as the Cuban missile 
crisis so amply demonstrated. Creating radical phrases is not restricted 
to academic types, of course, l-rtien asked by CBS News on February 20, 
1967, to explain the latest military escalation in Vietnam, an Adminis- 
tration spokesman said it represented a 'calculated outburst of impatience' 
on the part of President Johnson! 

New words and phrases rush into the spaces left by the mismatch of 
language to reality. The greater the rate of culture change the greater 
the pressure on speakers to innovate. The nuclear age is creating many 

empty spaces in our semantic maps of the world, but linguistic innovation 
is more apparent in its technological than its political aspects. In The 
Doomsday Dictionary, we find that a heach is a unit of fission energy 
equaling 3000 billion tons of dynamite, or enough to kill half the earth's 
population by fallout (a term derived from the motion picture. On the 
Beach), and that a 'kahn is the more 'modest' quantity of fission energy 
required to liquidate the entire population of one major country without 
any shelters — 300 'kahn equals one beach. 

The term kahn, of course, is taken from the name of one of the most 
provocative, innovative — and macabre — writers about strategy in the nuclear 
age, Herman Kahn. His several books are replete with lively concepts 
like aatalytia war (conflict between two major nuclear powers deliberately 
started by a third power), exaalation ladder (steps ranging from subcrisis 
disagreements to all-out nuclear war and its aftermath) , and stark deter- 
rence (capacity of overkill by a factor of 10 or more, so that miscalcula- 
tion or wishful thinking by an opponent becomes most unlikely). Of course, 
there are apposite coinages, which highlight the truly significant proper- 
ties of things, and inapposite ones, which obscure them. The term overkill 
is not particularly apposite, suggesting degrees of killing humans when 
humans can only die once; it is actually used to refer to the extra inten- 
sities of attack required to liquidate a defended target as compared with 
an undefended one. 

Another ingredient of a semantic revolution is respect for the 
distinction between words and things . This, of course, is the central 
theme of General Semantics. Actions should speak louder than words, but 
in international politics actions are often ambiguous as to interpretation. 
They may be expressive as well as instrumental. The assassination of 
John F. Kennedy was interpreted variously as a communist, a military- 
industrial, a right-wing conspiracy, or a CIA plot — depending on where the 
observer stood politically himself. The conflict in South Vietnam, at 
the time of U.S. escalation in February 1965, was officially interpreted 
as 'aggression from the North' but unofficially interpreted by many 
unattached experts as 'civil war'. Nevertheless, we would be wise to pay 
more attention to what nations d£ and less attention to what they say . 
As long as the feet of Chinese soldiers and their weapons stayed on 
Chinese soil, for example, we could afford to shrug off their blustering 

Yet another ingredient is substituting empathy for projection in the 
effects of our words upon others . Since most of us do most of our talking 
to people in a community which shares a common set of meanings and values, 
we have usually been successful in predicting their reactions to our words 
simply by projecting on to them how we would react ourselves. When we 
extend this normal process to people of other cultures using other lan- 
guages, however, we inevitably get into trouble. Projection made it easy 
for Americans to assume that the South Vietnamese people saw the conflict 
there as we did — as part of a massive confrontation between the Free 
World and Communism. However, in terms of Vietnamese experience it was 
probably, first and most immediately, a struggle for sheer survival in a 
war they hardly comprehended and second, but more remotely, a struggle for 
national independence from colonialism. Developing empathy for others 


requires exposing oneself to their conditions and experiences — directly 
if possible but mediately through well selected descriptions if not — and 
then role-playing the effects of our messages while standing in their 
mental shoes. 

UTien we label other people's needs, the words we choose may reflect 
how much projection or empathy is involved. A former official of the 
Agency for International Development, Byron Johnson, called the acronym 
AID a hypocritical misnomer, charging that American 'aid' consisted 
mostly of self-promoting, arrogantly-administered loans that offended the 
pride of the people in other nations. Despite the benevolent intent, it 
is depressing how often AID has been given or taken away for military 
and political reasons, how often it has foisted marginally useful goods 
on bewildered people in order to relieve gorged American markets, how 
often it ended up in the pockets of the relatively well-to-do in foreign 
countries, thereby actually widening the gulf between Have and Have-not 
people within the countries being 'aided'. The very term underdeveloped 
is an insult to people whose written histories may go back much further 
than our own, for it equates material, technological progress with cultural 
development in general. We would be wiser to speak about unevenly devel - 
oped countries . This phrase at least admits of the possibility that we 
Westerners are also underdeveloped in some respects, as well as definitely 
overdeveloped in others — to the extent that Mankind could perish from it. 

Another revolutionary phrase I would recommend is benevolent subver- 
sion: there is no question but what the West — and particularly the United 
States in the past few decades — has been the most effective force in 
history for subverting the governments, the economies, and the whole ways 
of life of other peoples throughout the world. And there is no question 
but what we have often been benevolent in intent, at least as we see 
ourselves. But when forced into anomalous confrontation, the words 
benevolent and subversion bring into mind obscure but significant features 
of each — the potential selfishness of benevolence and the potential 
altruism of subversion. 

Perhaps the most important ingredient in a semantic revolution is 
using radical sentences to compensate for conservative words . Although 
new words can be coined and old words can change their meanings, they 
remain essentially conservative and inept for the fresh purposes of the 
moment. Sentences can crunch words together into phrases which highlight 
new features and eradicate old; they can formulate assertions about the 
world which question tired assumptions and suggest new truths. Of course, 
the phrases may be bizarre and the assertions incredible upon close 
inspection, but the potential for fresh ways of talking, and hence think- 
ing, is there. 

For tv/enty years I have been espousing a general strategy for inter- 
national behavior modeled upon what I think I know about inter personal 
behavior — a strategy which seems appropriate for a nuclear age. Its 
formal and descriptively adequate name is graduated and Reciprocated 
Initiatives in Tension-reduction. But early-on I discovered that very few 
people, including myself, could easily remember this complex nominal com- 
pound correctly. While doodling at a conference one day I came up with 

the acronym, GRIT — which not only provided cues for the salient features 
of the strategy but also seemed quite appropriate, since grit is exactly 
what it requires. 

This strategy is a form of calculated de-escalation of conflicts: a 
nation caught in a conflict spiral deliberately initiates sequences of 
small, tension- reducing moves, these moves being well within its margin 
of security and being designed to elicit reciprocating steps from the 
opponent. To the extent that reciprocation ±s^ obtained, somewhat larger, 
more significant steps can be taken, and both parties move cautiously 
toward a political rather than a military resolution. 

My original paper on this (Osgood 1962) , appeared in a book sponsored by 
a group of Democratic congressmen, titled The Liberal Papers; the Republican 
National Committee described nr^. contribution as 'surrender on the installment 
plan — to which, in the present context, I can only say 'touche'I Debating the 
feasibility of such a novel approach to conflict resolution — at least as 
far as sovereign nations are concerned — forced me to question certain 
unquestioned assumptions about national security in this nuclear age, and, 
in doing so, to create a few radical sentences of my own. They will pro- 
vide a final illustration of my thesis. 

I had to assert that the primary motive behind the threatening 
behavior of nations toward others is usually fear . The traditional assump- 
tion, consistent with the dynamics of psycho-logic, is that one's opponent 
is always motivated by aggressive impulses. Accepting even the possibility 
of fear motivation enables one nation to respond more rationally (and less 
fearfully) to the blustering behavior of another. 

1 had to assert that there is no real security in military superiority 
in a nuclear age . The traditional assumption is that a nation is secure 
only when it is so strong militarily that no other nation, or combination 
of nations, would dare to attack it. But, of course, the possessor of 
such superiority becomes the focus of fear for others and thus the stimulus 
for continuing the nuclear arms race — which in our time is an invitation 
to commit suicide. I would even go so far as to recommend a strategy of 
calculated nuclear inferiority : when two (or more) nations can destroy 
each other ten times over, as far as being a viable society is concerned, 
a calculated degree of inferiority probably offers more security (by 
virtue of being less fear-producing) and certainly gives one an advantage 
in initiating arms controlling and limiting agreements, l^ether because of 
political intent or economic necessity, I think that this was the strategy 
of the Soviet Union vis-a-vis the United States until the past decade or so. 

I had to assert that having an invulnerable nuclear deterrent makes 
it possible for a nation to take calculated risks, risks designed ulti- 
mately to eliminate the need for the deterrent itself . The usual assumption 
is that such capability is merely a deterrent against others, and the 
possessors stand frozen in their mutual threat. But the simple fact of 
deterring others also provides a nation with a margin of security within 
which to take calculated risks. Rather than being seen as simply a continu- 
ation of the military tradition of discovering bigger and better ways of 
destroying opponents, the grisly fact of nuclear weaponsy could be viewed 
as the impetus toward eradicating the very tradition that gave rise to it. 


I had to assert that in the reduction of international tensions , 
POST commitment by reciprocation can be substituted for PRIOR commitment 
by negotiation . Although in our ordinary relations with wives, children, 
neighbors, and colleagues we almost never negotiate a prior commitment 
before starting tension-reducing actions ourselves, it has been tradition- 
ally assumed that under conflict conditions one nation cannot be decent 
to another without first obtaining from the other an iron-clad prior 
commitment that it will be decent in return. Not only can patterns of 
initiatives and reciprocations (post-commitments) reduce mutual hostili- 
ties, they can also create an atmosphere within which serious negotiations 
on critical issues can be more successfully undertaken. Both of these 
propositions were demonstrated in The Kennedy Experiment, and there have 
been some other demonstrations in human history. 

And, finally, I had to assert that goodwill among nations is a 
result of, rather than a prerequisite for, de-escalation of tensions . 
The traditional assumption has been that a nation cannot risk making even 
conciliatory gestures toward another unless some degree of mutual goodwill 
and trust exists between them — since otherwise such gestures are likely 
to be interpreted as 'signs of weakness'. Quite to the contrary, as I 
argued earlier, given even a modicum of mutual self-interest , the recipro- 
cal actions taken in its service can literally create mutual goodwill 
where little or none of it existed before. 

Interlocking assertions like these can provide a framework for rethink- 
ing our relationships to each other on this shrinking little planet at the 
dawn of its nuclear age. The assertions may be proven false, of course, 
but so may the tenets of any theory. The important thing is to use the 
potential radicalness of our sentences to at least reach for a clearer 
perspective on our own time and place. Revolutions do not guarantee a 
better world — they only make it possible. 

*An earlier version of this talk, with the same title, was published 
in Social Psychology and Political Behavior, ed. by G, Abcarian and 
J. W. Soule, 1971, Columbus, Ohio: Merrill. 


Kaplan, D. M. and A. Schwemer. 1963. The domesday dictionary. New 

York: Simon and Schuster. 
Osgood, C. E. 1962. 'Reciprocated initiative.' The liberal papers, ed. by 

J. Roosevelt. New York: Anchor Books. 


William S-Y. Wang 
University of California at Berkeley 

As always, it is both an honor and a pleasure to speak at the forum 
of the Linguistic Institute. In the present case, my pleasure is very 
much enhanced by the theme of this particular Institute, which is 'Language 
form and Language function: A Western and Nonwestern Perspective'. 

It is all too tempting for us to draw conclusions about what human 
language is like based on the samples of language we most commonly 
encounter here in the West. 

Let me give just one example of this western bias, one with immediate 
importance. Under the impact of Western languages, it is often assumed 
that alphabetic orthographies, constructed on phonetic segments, are the 
optimal system for representing language. This assumption is mostly made 
tacitly, to be sure, and thus it may be all the more misleading. On the 
other hand, orthographies based on other units, found outside the West, 
such as syllabaries or logographs, are felt in some way to be under- 
developed or retarded, and doomed to fall in time to the ultimate and 
inevitable triumph of the alphabet. 

But clearly this is an issue that is much too important to be assumed 
a priori on the skewed sample of just the Western languages. It is 
encouraging to note, though, that serious work is beginning to be done on 
the question of optimal orthography, based on considerations of language 
structure and psychological experiments on reading. 2 Hopefully, an 
increasing amount of involvement will be forthcoming from linguists in 
such research. 

So the moment is long due that we extend our focus beyond the familiar 
scene of Western languages in a serious way. In this respect, it is 
reassuring to look through the catalog of this Linguistic Institute, with 
its rich array of courses on the languages and linguistics of Africa and 
Asia. The planners of this Institute deserve our warm thanks for their 
effort in highlighting a more balanced perspective on the study of human 
language . 

The topic I was assigned to discuss this evening is 'Language Change'. 
Change involves the differences between two or more states in time. 
Presumably the goal here would be to reveal the principles according to 
which these differences are implemented, and to discover the mechanisms 
which brought about these differences. I hope to make some observations 
on both of these issues of implementation and actuation. But first it is 
necessary to distinguish change along several time scales, for it seems 
that the questions, the data, and the methods would not be the same for 
all these scales. 



For convenience of exposition, I will refer to the three time scales 
as microhistory, mesohistory, and macrohistory . They all deal with 
change across time and therefore are all aspects of the diachronic study 
of language. The microhistory of language is reckoned across a very thin 
slice of time, in years or decades. It is concerned with what Wm. Labov 
(1972) calls 'change in progress', which offers a diachronic way of look- 
ing at synchronic variation. 

In the microhistory of language the interests of several research 
areas converge. The sociolinguist focuses his attention on groups of 
people, as these are divided by age, by sex, by region, and by social 
background. Typically, within so short a period we find the model and 
the copies, the unchanged and the changed, existing side by side in an 
orderly profusion, vying for survival. By correlating the social para- 
meters with the noted language differences, we may hope to make some 
short-term predictions on which of the various usages of today will 
continue into the language of tomorrow (Weinreich et al . 1968). 

Closely related to these questions is the study of language contact. 
With the rapid increase in population and mobility, the notion of a pure 
and homogeneous speech community, whatever credibility it may have once 
held, cannot realistically be maintained. Diversity is the fundamental 
ingredient of change, as is well recognized for biological systems; 
this dogma is no less true for linguistic systems at every level. The 
further heterogeneity that contact produces adds yet new dimensions to 
the challenges of the student of language change. 

Among these various groups of people there is a very special class, 
with a unique biological and social status — and these are the very young. 
How children learn language — the transmission across generations — is 
clearly one of the vital questions in the whole of language change. 
Unlike that of the other groups, the language children use changes not 
only as a function of differences in environment, but also as a function 
of an increasing biological capacity due to neural and motor-sensory 
maturation. Ontogeny hardly ever recapitulates phylogeny with fidelity. 
Nevertheless, some insights can be gained here that may prove useful 
toward understanding the processes of language evolution. 

Until recently, theorizing on language acquisition, especially within 
the framework of generative grammar, has been often cast in rather global 
terms — in terms of whole categories of sounds and the addition and reorgan- 
ization of rules. However, more fine-grained work done in this area, by 
Ferguson and Farwell (1975), and Hsieh (1972), and others, clearly show 
that the real situation is much more complex, both within the case his- 
tories of each individual child, and in the strategies and development 
across different children. 

Some children, for instance, are extraordinarily pliable and are 
eager to experiment with new sounds and new sequences. Others are more 
conservative in their phonological behavior, and produce only forms that 
they have a good chance of getting right; they seem to prefer learning 


and using those words in which they have phonetic confidence, while avoid- 
ing others. Linguistic personalities, it would seem, are manifested at 
quite an early age, and these differences in strategy may very well have 
lasting influences on their adult language behavior. There is reason to 
expect that some of the individual differences in the language of adults 
can be traced back to the early years. It is an important finding that 
these differences can be detected even during the learning of the first 
several consonants, as is shown in some recent work of Ferguson for the 
acquisition of English laterals, reported in Fillmore, Kempler, and Wang 

From such fine-grained research, it emerges clearly that there is a 
primacy of lexical learning during phonological development. Even a 
relatively small scale sampling of the sort done by Ferguson and Farwell 
shows that the child does not progress by learning units like phonemes 
or allophones, but rather by gradually adding lexical items to his 
repertoire. The same sound in similar contexts may undergo altogether 
different histories, as this sound appears in different words. This is 
clearly demonstrated in Table 1. These data in the table are extracted 
from the study of Ferguson and Farwell, who followed with great care the 
development of three children who were learning their first fifty words. 
The data are of one child's progression in the acquisition of initial b- 
across a span of some four weeks. They show how different words exhibit 
different patterns of variation according to different schedules. The 
unity of the phoneme only emerges at the end of the acquisition process 
when mastery is complete. 












































Table 1 



The basic unit of acquisition, then, is something like the word. The 
awareness, conscious, or subconscious, of phonetic identity or similarity 

between portions of words, is something that comes to different children 
in completely idiosyncratic ways, and probably cannot be attributed to 
any single uniform stage of development. This theme of lexical primacy 
is central to my remarks there, and I will return to it later. 


As opposed to microhistory, the great bulk of the literature of 
language change actually deals with the middle time scale. Historical 
linguistics has traditionally concerned itself with changes that occur 
across centuries or millenia. Since it reaches further back in time, 
the primary data for language history are much more uneven and uncertain. 
Written records of early languages are low in both quality and quantity, 
and only in a handful of cases do they extend for more than 2 or 3 thous- 
and years. The methods of reconstruction have their intrinsic limitations, 
Most of us would probably agree with Kiparsky (1976) when he set the outer 
limits at 10 to 20 thousand years 'over which we can hope to reconstruct 
anything at all about Proto-language' . 

A classic question in language mesohistory, dating back to at least 
the neogrammarians, has been the manner or means by which a change is 
implemented. This question has caught the attention of a wide gamut of 
scholars through the decades, from Henry Sweet to Alf Sommarfelt and 
Sapir, to Hoenigswald and Halle. 

The received doctrine on this question has long been one which may 
be characterized as phonetically gradual and lexically abrupt . The idea 
is, that once a phoneme changes, it changes in all the relevant words 
according to the same schedule. By the claim of lexical abruptness, the 
regularity of the change, the so-called neogrammarian hypothesis, would 
follow as a consequence by definition. But since a change may involve 
hundreds of words, and since languages do not seem to effect such whole- 
sale changes in pronunciation overnight, the phonetic gradualness becomes 
a necessary corollary of lexical abruptness. 

So the notion of lexical abruptness is motivated by believing that 
changes are always regular, and the notion of phonetic gradualness is 
necessitated by believing in lexical abruptness. And it is essentially 
this doctrine of historical change, with minor variations, that has 
retained the most widespread acceptance in the work in generative phonol- 
ogy as well as the structuralist phonology that preceded it. 

But here again, as was the case in the microhistory of language, 
things are not that simple. Empirical investigations over the past 
decade or so on a variety of languages, using large quantities of data, 
have shown that there are changes which are implemented in a manner that 
is lexically gradual. That is, a change may initiate on a handful of 
words in the lexicaon, where these words do not constitute any natural 
phonological or morphosyntactic class. Then the change affects an increas- 
ing sector of the lexicon in time, perhaps eventually completing its 
course on all relevant words. Whether or not a particular change ever 
completes its course depends on a whole host of factors, some of which 
are outside the linguistic system. We can get at least a rough idea of 

the degree of regularity that sound changes exhibit from a study that 

C. C. Cheng and I (Cheng and Wang 1971) did on the development of initial 

consonants in Chinese over the past 1400 years, shown in Table 2. 

Time and again, scholars of sound change have observed that exceptions 
to correspondences are embarassingly numerous, frequently the irregular 
reflexes outnumbering the regular ones. Table 2 gives a rough quantitative 
index for these observations. Each cell in the table refers to the devel- 
opment of one Middle Chinese initial consonant (ca. 600 A.D.) into one of 
the contemporary dialects. When the development is indeed perfectly 
regular, the cell is marked with an 'x'; otherwise it is left blank. Each 
blank cell in Table 2, then, is one of the 'embarassingly numerous' cases 
where regularity was not reached — and there are indeed many of them. 

A particular change, for various reasons, may even reverse its 
course. Tore Janson (1977) has studied such a case in the deletion of 
final -d in the Swedish of Stockholm, where he attributes the reversal to 
the influence of orthography. At any rate, this picture of implementation 
which involves a gradual change across the lexicon, which may or may not 
result in complete regularity, has been called lexical diffusion (Wang 1969). 

Since lexical diffusion is a process via which a change is imple- 
mented , it is in principle accessible to any change, however the change is 
actuated , i.e. whether the actuation is external or internal to the system 
under change, or whether it is phonetically or analogically triggered, etc. 

An interesting case of lexical diffusion has been reported by Lyovin 
(1977) , where the actuation is due to the avoidance of homophony within 
the inflectional paradigm of Tibetan verbs- The study is of particular 
theoretical significance because of the intricate relations between a 
change that reduced initial consonant clusters, which causes homophony, 
and the hierarchy of verb categories which controls the schedule according 
to which periphrastic expressions are created to avoid the homophony. The 
present tense dominates the past tense, for instance, so when the two 
forms become homophonous due to cluster reduction, it is the past tense 
form that gets displaced from the inflectional paradigm. Similarly, the 
past tense form dominates the future tense form, and it is the latter that 
gets displaced when homophony occurs. In each case, the meaning of the 
displaced form has to be expressed periphrastically . 

Lyovin' s examination of the data across the various categories of 
verbs in Classical Tibetan led him to observe that 'homophony-inducing 
sound changes will be blocked from applying to certain lexical items or 
classes of lexical items until compensatory developments permit the further 
diffusion of the shifts in question' (p. 129). 

Another study in lexical diffusion that raises important theoretical 
issues is the investigation of 0. Robinson (1977) on Swiss German vowels. 
Two historical changes are relevant, as shown in the diagram. 

umlaut ^ 

a ■*■ ae 

rounding i 

I umlaut " 

PC w 

H hJ 60 

M < C 

S M (0 

H O 


















































































vo 00 00 <r 


>< X >< >< X 



X X X X X X X 


xxxxxxx xxxxxx 


O M N O X! 60 C 

C C X 

N C X 3 

•H bO 0) 



•S 3 

O X U P^ 


Robinson notes that the 'rounding rule was diachronically much later 
than umlaut ' . Both rules continue to operate synchronically in morpho- 
phonemic alternations. 

Now if the rules applied synchronically in the order in which they 
entered the language diachronically, then the /a/'s which are subject to 
the umlaut rule should change to /ae/'s, and thus escape the rounding rule. 
But in almost all of the northern dialects, these /a/'s actually umlaut to 
/o/'s, presumably via /o/. Furthermore, a synchronic rule that rounds 
/ae/ into /o/ would not be possible; it would be contradicted by other data. 
For this interesting phenomenon, Robinson adopts Kiparsky's interpretation, 
that the synchronic rules have reversed their diachronic order, that is, 
from umlaut-rounding to rounding-uralaut . 

The data of relevance here have to do with certain northern dialects, 
e.g. Kesswil, where many forms exhibit the morphophonemic alternation 
/a~ae/, when the reordering hypothesis would expect /a~o/. Robinson's 
explanation of these ae-alternations is that rule reordering, though an 
abstract change, is diffusing across the lexicon, much as concrete changes 
like d-deletion or vowel shift. These ae-alternations, then, are forms 
which the reordering has not reached as yet. The theoretical significance 
of this explanation is the extension of the process of lexical diffusion 
to operate at a relatively abstract level of sound change. To support 
his explanation, Robinson gives some exemplary arguments against some 
alternative solutions, including the possibility of borrowing these ae- 
alternations from other dialects. 

Work on lexical diffusion over the past decade has shown that the 
traditional view of sound change has been excessively restrictive, and that 
the implementation may very well proceed along other paths than one which 
is lexically abrupt. This work has been surveyed in Chen and Wang (1975) 
and Wang (1976, 1979), and partly anthologized in Wang (1977). At this 
stage, there is less need to document additional cases of lexical diffu- 
sion. Rather, our next challenge, it seems to me, is to solve the puzzle 
of what kind of sound change would travel along which path for its imple- 
mentation. In addition to the lexical and phonetic parameters, there is 
the additional statistical parameter that must also be considered in 
investigating the many paths of sound change. 

Two other directions of current research in this area should also be 
mentioned, in view of the very promising results obtained so far. One is 
the demonstration of the relationship of the relative frequency of words 
to their schedules of change, as discussed by Hooper (1977) for English, 
and Gerritsen and Jansen (1978) for Dutch. Although such a relationship 
has been hypothesized at least since a century ago, it is only recently 
that it can be objectively tested as a result of the availability of 
frequency dictionaries. It is of special importance that Hooper has been 
able to show that high frequency words in English are leaders in a phonetic 
change like schwa-deletion, but are laggers in an analogic change like the 
weakening of strong verbs. If we can continue to identify independent 
variables for the chronological profiles of change (relative frequency 
being one such variable) , then perhaps the goal of being able to make short- 
term predictions in sound change will be eventually attainable, after all. 3 

The other direction that I'll briefly allude to has to do with sub- 
grouping languages. Since it has been assumed in the bulk of historical 
work that sound changes are all lexically abrupt, it follows that for any 
change in question, a particular group of languages either shares the 
change or not — a binary yardstick. But once we recognize that diffusion 
processes may be at work, then some changes at least can be exploited for 
much more information. Indeed, as Hsieh pointed out in his seminal 
article (1973), this was the insight underlying Swadesh's method of lexi- 
costatistics. Instead of a binary yardstick, quantitative measures can 
be made on various subgroupings of the languages on the basis of both 
changed and unchanged cognates. Hsieh demonstrated the effectiveness of 
this approach on the Wu dialects of eastern China by using a single tone 

More recently, Bh. Krishnamurti (1978) and Krishnamurti and his col- 
leagues, in some yet unpublished work, have made important progress in 
developing a suitable methodology for the subgroup ing problem. Using the 
Dravidian Etymological Dictionary as the starting point, they have 
analyzed a large amount of computerized data to determine the genetic 
relationships among various languages of South India. There is every 
indication that their procedures will prove fruitful when validated 
against linguistic situations where the answers are known, at which point 
they can be used in situations where no answers are available (for unwrit- 
ten languages, for example). 


In considering language change within the largest time perspective, 
its macrohistory, we are faced with the greatest challenges — since the 
relevant primary data are mostly not available. Furthermore, the necessary 
backgrounds and methods are frequently outside of the typical domain of 
linguistics. So we need to look into neighboring disciplines that work on 
related questions, such as anthropology, biology, ethology, psychology, 
zoology, and so on. An example of the interdisciplinary convergence of 
interest in this area is the very successful conference in the origin and 
evolution of language and speech, sponsored by the N.Y. Academy of Sciences 
in 1975 (Harnad et al 1976) . See the excellent discussion of this confer- 
ence by Hill and Most (1978) . 

A useful assumption to make here, one which most investigators accept, 
is that language did not abruptly burst into the course of human evolution 
fully in the state of intricacy and complexity that we find it today. To 
believe the so-called discontinuity theory, that there was some sudden 
wholesale genetic mutation that in a fell swoop transformed muteness into 
eloquence is to relegate macrolinguistic research to the mystical regions 
of miracles and spontaneous creations.^ 

Rather, there must have been a very long course of emergence, much 
longer than the span of time over which current methods of historical 
linguistics can take us. During this emergent state, protolanguage 
probably at different times shared various developments, by either homology 
or analogy, with the communication systems of other species. So it is 
likely that sorting out the ingredients of these other systems will 


contribute to our understanding of how the ingredients of human language 
were accumulated piece by piece over the hundreds of millenia. 

One category of ingredients that has been present since the early 
stages of emergence is the use of gestures. Although gestures are used 
across a wide range of species, we share many specific ones with other 
primates, including several facial expressions, presumably because of the 
shared anatomy. Gestures accompany speech in all languages, time-locked 
to various extents; they also play a relatively more prominent role in 
the early communications of very young children. Phylogenetically , Gordon 
Hewes (1973), among others, has championed the view that human language 
originated in gestures. 

Another category of ingredients that is also widespread in communi- 
cation systems is the use of prosodic features — fundamental frequency, 
duration, and intensity. These features are probably exploited more 
fully in bird song than anywhere else, though recent investigations 
on primate calls are revealing a variety of new ingredients, such as 
amplitude modulation, frequency modulation, and the use of formants. 
In human evolution, there was a correlated development of the increased 
use of the hands for fighting and carrying with the increased use of the 
mouth for communicating. There are certain obvious advantages that pro- 
sodic features have over gestures: Communication can take place over 
larger territories requiring no visual contact (such as in darkness or 
across foliage) , while the rest of the body can be engaged in other 
simultaneous activities (such as running or fighting) . So it is rea- 
sonable to expect that the prosodic communication gradually expanded 
in importance relative to the gestures. 

As the civilization of early man became increasingly complex, 
however, he needed larger sets of signals with which to communicate 
the expanding vocabulary of messages. Since communication facilitates 
the planning and execution of group activities, for purposes such as 
hunting for food or defense against predators or other tribes, we can 
easily see how groups that have developed better language would have 
better chances of survival. Everything else being equal, better language 
enables the formation of larger (and hence stronger) groups, the trans- 
mission of more precise and varied information across both space and time, 
and the perpetuation of the gene pool of the group of users of that 
language. Biologically speaking, then, selectional pressures favored 
those populations with better tools for communication in probably much 
the same way that they favored better tools for digging or fighting. 5 

As the message set grew, the three prosodic parameters became insuf- 
ficient as carriers for the signals. A phonology that was largely prosodic 
(and laryngeal) eventually changed into a segmental one, making use of the 
richer possibilities of the supralaryngeal gestures. Whereas other pri- 
mates share with us the use of gestures and prosody, it is the transition 
from prosody to segmentals that is the critical step which our ancestors 
alone have taken. It has been proposed by P. Lieberman (1975) that this 
transition was made possible by the phylogenetic descent of the larynx, 
and that the descent was a special adaptation for speech. A more plausible 
view to my mind is that the descent of the larnyx, as well as a whole host 


of other skeletal restructurings in the head and neck, was one of the 
many mechanical responses that the human body had to make in assuming 
erect posture. In this latter view, where the structure was largely 
available before the linguistic use, it would be more accurate to call 
it a case of preadaptation . 

In any case, segmental phonology grew dominant over prosody, just as 
prosody in an earlier stage of emergence dominated gestures. All three 
systems continue to co-exist in every language today, functioning in 
mutually supplementary ways. The importance of the transition into seg- 
mental phonology is highlighted by recent researches on communicating 
with apes. Whereas chimpanzees show remarkable cognitive abilities at 
symbolization, including the processing of complex sentence types, they 
have next to zero ability at producing controlled segmental sounds. From 
an evolutionary perspective, that critical step which early man took in 
pairing messages with the particular medium of consonants and vowels was 
the start of the journey that led to speech, the 'indispensible foundation' 
upon which language is built. Judging from the amount of new brain tissue 
in the cortex that appears to be involved with language, that step must 
have been taken quite some time ago (cf. A. Liberman 1974). 

Although there may never be any way of documenting the transition 
directly, we can imagine how it could have come about by referring to 
processes of phonological change that can be observed in recent times. 
These are of the type that has been called 'phonologization' . Typically 
what happens is as follows. At time ti, one group of words is distinguished 
from another group in that the former group has the phoneme X while the 
latter has the phoneme Y. For reasons of coarticulation, X and Y act upon 
their environments in different ways, i.e. XE' but YE", even though at t^ 
the difference between E' and E" is not considered distinctive, and may not 
even be noticed. But should a merger take place by the time t2, when X > Y, 
then the two groups of words are no longer distinguished by X and Y, but 
now by E' and E". The difference between E' and E" becomes 'phonologized ' 
at t2 — it has become phonemic. 

There are numerous such cases in the literature. Chinese, for one 
example, had distinct voiced and unvoiced stops in syllable-initial posi- 
tion. Because of coarticulation, the syllables with voiced stops must 
have had a different pitch contour from those with unvoiced stops. (Com- 
pare the different pitch contours for the English words bin and pin.) 
Some centuries ago, however, the voiced stops merged into the unvoiced 
ones. Now the two groups of words are kept apart no longer by the voicing 
in the initial stop, but by the tonal difference that was once caused by 
that voicing. 

The transition from prosodic utterances to segmental utterances could 
not have been this straightforward, of course, and must have taken place 
over a much longer timespan, probably with many false starts in the process. 
The segments started as conditioned or spurious accompaniments of the 
prosodies, and were only eventually 'phonologized' by transference. Once 
the pairing between messages and the segmentals got established, however, 
language must have evolved at an explosive pace, given the much greater 
signalling potential and physiological economy of a segmental phonology 
over alternative media. 


Over the past several years there has been some significant research 
into the sign language of the deaf. This work is of fundamental importance 
for the light it sheds on language viewed from a different modality, and 
thereby giving us a broader perspective on human communication. Experi- 
ments have been done on comparing the rates of production and perception 
of speech with those of sign language. While such experiments are of great 
intrinsic interest, it should be clear that certain questions of linguistic 
macrohistory cannot be directly answered from their results. For even if 
it can be completely demonstrated that sign language is as effective as 
speech in every conceivable way, which I would find totally surprising, 
the point remains that the user of the sign language is the direct benefi- 
ciary of numerous millenia of language evolution and elaboration that had 
in fact taken place via speech. The deaf is deprived of only a minor, 
peripheral part of that total neural machinery. The equivalence between 
modalities, if it can be demonstrated, tells us that once the full symboli- 
zation system has been developed in our brain, a surrogate medium may serve 
as well as speech. It tells us nothing on whether the surrogate medium 
had enough to recommend it for language to have evolved into its present 
state of intricacy and richness, which is clearly a stronger requirement. 

In contrasting the three scales of diachrony, I have perhaps put too 
much emphasis on the differences between them. Clearly the three must be 
related to each other in intimate ways, since they are but three different 
time windows through which we are viewing the same phenomenon. At present, 
the facts are scanty. There are huge gaps in our knowledge on the steps 
in which language evolved through the successive emergent states into the 
present steady state. It is obvious that the study of linguistic univer- 
sals must be centrally relevant here in providing a base line for the 
steady state. Similarly, our understanding of the relation between micro- 
history and the classical concepts of linguistic change is spotty at best. 
A great deal of basic research needs to be done before a plausible scenario 
can be provided for what has been called 'the life and growth of language'. 

This paper is prepared while I am a Guggenheim fellow visiting 
Osmania University in India. Thanks are due both to the Guggenheim Founda- 
tion and Bh. Krishnamurti of Osmania University for their support and 
encouragement. I would also like to acknowledge the many years of camara- 
derie and collaboration of C. C. Cheng, which were critical ingredients 
in the progression from idea to data to knowledge on language change. 

A good beginning is the Cross-Cultural Conference on Language, 
Reading and Orthography, sponsored by the National Institute of Child Health 
and Human Development, Bethesda, Md . , September 1978. The proceedings are 
being edited for publication by J. Kavanagh and R. Venezky. An extensive 
discussion of Chinese orthography from several viewpoints is available in 
the Journal of Chinese Linguistics 6.2, 1978; see especially the contribu- 
tion by Ovid Tzeng et al. for recent psycholinguistic work. (See also 
Hardych et al. 1977, 1978.) 


Of all the independent variables that influence change, probably 
the least stable and therefore most difficult ones to capture are the 
social ones. Note the perceptive counsel of a poet on this point, Alexan- 
der Pope: 

In words, as fashions, the same rule will hold 

Alike fantastic, if too new or old 
Be not the first by whom the new are tried 

Nor yet the last to lay the old aside. 

There was, of course, no dearth of proponents for such discontinuity 
theories among philosophers working before evolutionary thinking was 
developed in biology. See Stam (1976) for a useful historical critique, 
especially from mid-18th century through mid-19th century in Europe. 

Once languages have reached a more-or-less steady state, these 
selectional pressures no longer operate as they did during the emergent 
state. Labov's observations on language diversification being dysfunctional 
(1972:273) are presumably based on the steady state changes, as indeed 
were the remarks by Charles Darwin and Max Muller that Labov referred to. 


Chen, M. and W. S-Y Wang. 1975. Sound change: actuation and implementa- 
tion. Language 51:255-81. 

Cheng, C. C. and W. S-Y. Wang. 1971. Phonological change of Middle 

Chinese initials. Tsing Hua Journal of Chinese Studies 9, 1. 215-70. 

Ferguson, C. A. 1979. Phonology as an individual access system: some 
data from language acquisition. In Fillmore, Kempler, and Wang. 

and C. B. Farwell. 1975. Words and sounds in early language acquisi- 
tion. Language 51. 419-30. Reprinted in Wang 1977, 7-68. 

Fillmore, C. J., D. Kempler, and W. S-Y. Wang. 1979. Individual differ- 
ences in language ability and language behavior. New York: Academic Press 

Gerritsen, M. and F. Jansen. 1978. Word frequency and lexical diffusion in 
dialect borrowing and phonological change. Studies in Dutch Phonology. 

Hardyck, C, 0. Tzeng, and W. S-Y. Wang. 1977. Cerebral lateralization 
effects in visual half-field experiments. Nature 269. 705-7. 

1978. Lateralization of function and bilingual decision processes: 
Is thinking lateralized. Brain and Language 5. 56-71. 

Harnad, S. et al. (eds.). 1976. Origin and evolution of language and 
speech. Annals of the New York Academy of Science 280. 

Hewes, G. W. 1973. Primate communication and the gestural origin of 
language. Current Anthropology 14. 5-24. 

Hill, J. H. and R. B. Most. 1978. Review of Origin and Evolution of 
language and speech, ed. by Harnad et al. Language 54. 647-60. 

Hooper, J. 1977. Word frequency in lexical diffusion and the source of 

morphophonological change. Current progress in historical linguistics, 
ed. by W. Christie, 95-105. Elsevier: North Holland. 

Hsieh, H. I. 1972. Lexical diffusion: evidence from child language 

acquisition. Glossa 6. 89-104. Reprinted in Wang 1977, 133-47. 
1973. A new method of dialect subgrouping. Journal of Chinese 

Linguistics 1. 64-92. Reprinted in Wang 1977, 159-97. 
Janson, T. 1977. Reversed lexical diffusion and lexical split: loss of 

-d in Stockholm. In Wang 1977, 252-65. 
Kiparsky, P. 1976. Historical linguistics and the origin of language. 

Annals of the New York Academy of Science 280. 97-103. 
Krishnaraurti, Bh. 1978. Areal and lexical diffusion of sound change. 

Language 54. 1-20. 
Labov, W. 1972. Sociolinguistic patterns. Philadelphia: University of 

Pennsylvania Press. 
Liberman, A. M. 1974. The specialization of the language hemisphere. 

The neurosciences: third study program, ed. by F. 0. Schmitt and 

F. G. Worder. Cambridge, Mass.: MIT Press. 
Lieberman, P. 1975. The origins of language: an introduction to the 

evolution of human speech. New York: MacMillan. 
Lyovin, A. 1977. Sound change, homophony, and lexical diffusion. In 

Wang 1977, 120-32. 
Robinson, 0. W. 1977. Rule reordering and lexical diffusion. In Wang 

1977, 69-85. 
Stam, J. H. 1976. Inquiries into the origin of language: the fate of 

a question. New York: Harper and Row. 
Tzeng, 0., D. Hung, and L. Garro. 1978. Reading the Chinese character: 

an information processing view. Journal of Chinese Linguistics 6. 

Wang, W. S-Y. 1969. Competing changes as a cause of residue. Language 

45. 9-25. 

1976. Language change. Annals of the New York Academy of 

Science 280. 61-72. 
(ed.). 1977. The lexicon in phonological change. The Hague: 


1979. Language change: a lexical perspective. Annual Review 

of Anthropology 8. 353-71. 
Weinreich, U., W. Labov, and M. I. Herzog. 1968. Empirical foundations 

for a theory of language change. Directions for historical linguis- 
tics, ed. by W. Lehmann and Y. Malkiel, 97-195. Austin: University 

of Texas Press. 


Paul Kiparsky 
Massachusetts Institute of Technology 

1. Simplifiaation vs. proportions. Recent linguistics has moved 
from the structuralist segregation of synchronic and diachronic study of 
language towards a reintegration of the two, based on the idea that 
structure can determine change, with the corollary that change can therefore 
be diagnostic of structure. This program has had consequences on both 
sides of the equation. On the synchronic side, it is mainly through evi- 
dence from change that phonologists have argued for such factors as opacity 
and recoverability in the evaluation measure, and for the imposition of 
restrictions on the abstractness of underlying phonological representations. 
On the historical side, the new conception of linguistic structure evolved 
through efforts to construct explicit formal generative grammars has pro- 
voked some rethinking of the nature of both sound change and analogy. 

As an example of the relevance of synchronic work for historical 
linguistics we may take the classic issue of whether sound change admits 
nonphonetic conditioning. Formal grammar as well as sociolinguistic 
research on linguistic variation has shown that phonology is intimately 
connected to higher linguistic levels. This undermines the structuralist 
concept of autonomous phonology, and thereby in turn casts doubt on the 
companion concept of 'blind' sound change. Morphological and other non- 
phonetic conditioning of sound change was previously ascribed to grammatical 
analogy secondarily acting upon the results of a quasiphysiological process 
of sound change. It is now more simply taken to be part and parcel of the 
phonological process itself, as a reflection of the inherent contextually 
determined and functionally governed variability characteristic of all new 
rules. This permits a more restricted and therefore more predictive theory 
of analogy. 

In general, the theory of grammar has opened up the possibility of 
structural explanations as alternatives to be considered in many cases 
where so far only a historical explanation was available. Noting a change 
(A > A' , B > B') in two classes of items, the traditional historical 
linguist would tend to seek its origin in one class and postulate its 
analogical spread to the other: 

(1) A , B 

A', B 

4- analogy 

A', B' 

In many cases, however, there turns out to be a structural explanation in 
the form of a shared property of A and B which allows us to assume a common 

origin for the phenomenon in both classes: 

(2) A , B 
A', B' 

Thus the investigation of clusters of co-occurring changes becomes an 
important method of probing into the organization of linguistic phenomena. 

The issue to which I should like to address myself here is analogical 
change. 2 Traditionally visualized as the extension of surface patterns (in 
terms of proportional schemata) it has more recently been given another 
interpretation as the elimination of arbitrary complexity in the linguistic 
system (in terms of the evaluation measure developed in the theory of gram- 
mar) . At the back of this lies a new view about the nature of the concrete 
process of analogical change and how it relates to the acquisition and use 
of language. A proportional view of analogy fits naturally into a theory 
of language acquisition based on substitution-in-frames techniques and 
equivalent 'taxonomic' devices. The idea that analogy is simplification 
of grammar jibes better with the idea that language acquisition is based on 
a general rule schematism in conjunction with an evaluation measure which 
selects the simplest grammar from the set of alternatives compatible with 
the data encountered by the learner, and assumes that learners progress to 
the adult grammar through a series of intermediate stages of increasing 
adequacy and complexity. Analogical changes are then interpretable as 
retentions of features of these Intermediate stages into the adult system. 

The difference between the views is not so radical in practice as it 
seems from the above formulation. Both approaches, to be at all plausible, 
require some elaboration, which may take and in fact has taken different 
terms. Other factors than simplification have been introduced by propo- 
nents of grammar-based theories. An example is transparency (Kiparsky 
1971, 1973), which determines the directionality of rule reordering, 
indifferent as far as simplicity is concerned. Proposals have been made to 
bring in paradigmatic factors independently of simplicity considerations 
(Harris 1973; Kiparsky 1972). The general effect of these modifications is 
to bring the theory into closer agreement with proportional theories. 
Under the radically 'concrete' analyses of Natural Generative Phonology 
and Upside-Down Phonology the treatment of analogy as simplification yields 
predictions which resemble those of proportional theories. 

On the other hand, diverse proposals exist to replace bare proportions 
by interpreted ones. Typical is this passage from von Wartburg (1969:59): 
'. . . analogy does not operate blindly. The Neogrammarians, and Saussure 
also, made the mistake of regarding analogy as a purely proportional 
process. Paul (p. 116) simply asserts that, for a new form to arise, there 
must be three elements present in the equation. He cites as an example 
animus : animi = senatus : x; x = senati (instead of Class. Lat. senatus) . 
This conception accounts for only a limited number of instances of the 
working of analogy. In most cases, analogy does not arise merely as the 
result of a formal resemblance; the two words so linked are also closely 
linked by their semantic environment.' Similar ideas are found earlier in 
Hermann (1931). Another kind of 'annotated proportion' was developed by 


Kurlowicz (1956, 1964, and many other works). Kurylowicz argued that both 
the terms of the proportions and the relation between the terms have to 
be interpreted in terms of certain abstract structural notions. For 
example, the relation between a and b and between a and x in the propor- 
tional schema could not be just any relation, but only that between a 
'forme de fondation' and a 'forme fondee*. And the terms themselves may 
have to be different from what is actually pronounced in the language, as 
in the following proportions: 

(3) a. R : eR = e : X (X = ee = i) 


b. ReT : RuT = RewT : X (X = RuwT = RuT) 

(3a) was proposed to account for the rise of vrddhi (long grade) in nouns 
(Kuryiowicz 1968:298). (3b) was proposed as the explanation of the Germanic 
verbs of the type lukan (Kuryiowicz 1970) . These proportions critically 
depend on the equivalence within the early Indo-European phonological 
systems of syllabic and nonsyllabic resonants (/? = R) and long vowels and 
geminate vowels (e = ee) . 

In this way, analogical proportions come look in some ways rather like 
the rules of a generative grammar. The relation between forme de fondation 
and forme fondee is something like the relation between input and output of 
a rule, and proportions where terms are 'virtual' in that they abstract 
away from certain overtly present surface features are comparable to rules 
which apply to abstract representations at nonterminal stages in deriva- 
tions. For example, (3a) and 3b) can be construed as representing the 
generalization of ablaut rules at a point in the synchronic derivation of 
Indo-Iranian and Germanic which precedes the application of vowel contrac- 
tion and of the rules which determine the distribution of syllabic and 
nonsyllabic sonorants. Indeed, Kuryiowicz (1964) has explicitly justified 
the abstract character of analogical change by the principle that redundant 
features or elements are to be disregarded in setting up proportions. In 
our terms that translates into the claim that rules can generalize even 
when followed by other rules which add those redundant features or elements. 
Thus, Kuryiowicz (followed by Watkins (1969)) considers the proportion (2) 
a valid account of the replacement of Indo-Iranian mid. 3. pi. ra by nta: 

(4) act. t : 3. pi. nt = mid. 3 sg. a : X (X = nta) 

Here the suffix -t of the unmarked finite verb form (act. is to be 
stripped away, as it were, in order for the terms to match properly. 

Still, even in their modified forms, both theories retain many 
inadequacies. Either side can easily adduce examples of indubitable ana- 
logical changes which do not agree with the claims of the other. In fact, 
much of the debate around analogy over the last ten years or so has con- 
sisted of generative linguists showing up various problems for the 
proportional theory (such as analogical changes that depend in various 
ways on abstract properties of grammatical structure — significantly more 
abstract than even the work of Kuryiowicz assumed) and the opposition 
countering with cases of analogical processes which complicate grammars 
rather than simplifying them. A detailed assessment of some of this 

evidence is presented in the fuller version of the present work. It fully 
confirms Hock's (1975) conclusion that analogy can involve different 
levels of abstraction: there are some changes 'which can be plausibly 
explained or motivated only in terms of (the generalization of) sometimes 
fairly abstract rules, and not (or only with great conceptual difficulties) 
in terms of classical surface-oriented analogy', and other changes which 
show a sovereign disregard of those abstract rules and are 'explainable 
only in terms of surface analogy'. 

Because of space limitations I shall omit entirely any discussion of 
the first type of case, which contradicts recent proposals to revive 
surface proportions and reaffirms the fundamentally grammar-based, simpli- 
ficatory character of analogical change. What follows is a brief outline 
of the problems posed by cases of the second type, cases which appear to 
show just the opposite. I shall argue that they too can be analyzed as 
simplifications provided we take into account certain kinds of interaction 
between the language learner and his linguistic environment and certain 
kinds of inertia inherent in the process of language acquisition. That is, 
the difficulties turn out to be artifacts due to the simplifying idealiza- 
tions of the ideal speaker-hearer and instantaneous language acquisition, 
which are appropriate in the study of formal grammar but should not be 
imported into historical linguistics. 

Cases where analogy produces complication in the grammar can be grouped 
into two main types, not necessarily mutually exclusive or in all cases 
clearly separable, but still broadly distinct in their symptoms and causes: 
partial analogy and false analogy . 

2. Partial analogy. The characteristic feature of partial analogy 
is that an analogical process which would be a simplification if it took 
place across the board ends up actually complicating the grammar because it 
is not fully carried through. Each type of simplification can be partial 
in this sense. As a counterpart to the loss of a rule (and in fact usually 
as the first step in that process) we find its curtailment, i.e. partial 
loss . For example, Sadock (1973) finds that the loss of the word-final 
devoicing rule in Yiddish went via an intermediate stage at which it 
applied in a phonologically restricted context. 'First, it was lost every- 
where except in the final cluster- -nd, and then after the loss of final 
schwas, the rule was dropped entirely.' Thus 'the Yiddish word-final 
devoicing rule in all likelihood was made considerably more complex (i.e. 
it became more restricted in its application) before it was lost' (p. 793). 
Other cases have been presented in Andersen (1969) , Dressier (1972) , Ralph 
(1975:161 f f ) . As a counterpart to the generalization of a rule, we find 
its partial generalization in some subenvironment only (Thomason 1976; 
Ralph 1975:172ff). As a counterpart to reordering , we find a new order of 
application introduced for only some of the cases (Kiparsky 1973:93-101; 
Robinson 1976; Bley-Vroman 1975). 

In such cases, the old form of the rule, or (in case of reordering) 
the old order remains in the grammar and continues to be applicable to a 
subset of forms, which must be identified by a contextual restriction 
added to the rule, or by an arbitrary rule feature marked on some lexical 
entries — and at worst, by a combination of both. The regularization may. 


for example, go through only: 

— in inflection as opposed to derivation 

— in some morphological category (e.g. plurals but not singulars) 

— in some inflectional subclass (e.g. the most productive declension 

or conjugation) 
— in some subset of the vocabulary (often the less common words) 
— in some phonologically definable circumstances (often in cases 

which are in various ways relatively less salient) 
— in just one sense of a word (often e.g. in metaphorical use) 

In all such cases, the result is a more complex grammar. 

Can even such changes be due somehow to the mechanism of simplifica- 
tion? The obvious possibility to consider is to reassess the form of 
grammars or the evaluation measure (or both) in the hope of discovering 
that the grammars which are actually learned by speakers are such that even 
in the apparently 'bad' cases the new grammar turns out to be more highly 
valued after all. Methodological considerations would recommend this 
approach because it would give both the simplest and the most restrictive 
theory of analogy — provided of course that the required kinds of synchronic 
graimnars could be justified in their own right independently of the histori- 
cal changes which we wish to explain. 

Several such attempts have in fact been made. One is to maintain the 
evaluation measure unchanged while assuming grammars of a much 'shallower' 
type than generally envisaged. The best-known solution of this type is 
embodied in the theory of Natural Generative Grammar (Vennemann 1974; 
Hooper 1976; cf. the related ideas of Skousen 1975). As Hooper notes, the 
gradual spread of certain kinds of leveling (rule loss) through the lexicon 
could be characterized as simplification if phonological rules and repre- 
sentations were made very 'concrete'. If, for example, a morpheme was 
entered in the lexion as the set of its allomorphs, then any analogical 
change which reduces the number of allomorphs — all cases of rule loss, for 
example, are of this type — would entail a simplification in the phonological 
representation of any item subject to the analogical change. If marked 
ordering of phonological rules did not exist, then for example what now are 
treated as counterfeeding orderings would really be lexical exceptions to 
the counter fed rule, and a 'reordering' into feeding order would really be 
the elimination of this exception feature, which one would expect to take 
place word by word, as does any idiosyncratic lexical marking. If we could 
assume that subrules are not collapsed as is claimed in the theory of 
generative grammar, but are to be kept separate, then obviously the inde- 
pendent generalization of those subrules would not be a problem. In short, 
the fragmentation observed in the problematic cases of change could be 
ascribed to the synchronic system in which the change takes place. 

This line of thinking, however, destroys the basis for understanding 
the normal situation where a rule is generalized in the same way in differ- 
ent subenvironments . The splits effected by partial analogy are so varied 
that in order to turn them all into simplifications we end up with no 
synchronic generality at all, so that each individual case of alternation 
is governed by a 'rule' of its own. The result is a synchronically wrong 
description and a treatment of analogy which suffers from the same kinds of 
defects as the proportional model. 


A more promising-looking attempt to motivate partial analogy in 
synchronic theory is sketched out in Andersen (1969) . Though he too pro- 
poses fairly shallow underlying forms, he explicitly assumes single 
(nondisjunctive) representations related to the phonetics by extrinsically 
ordered rules, among which he distinguishes morphophonemic and phonological 
rules. The key to his solution is not so much the 'concreteness ' of gram- 
mars as the idea that at the stage of the language which immediately 
precedes the appearance of the analogical forms, speakers learn as it were 
the 'wrong' grammar, which is such that the subsequent spread of the analogy 
in fact constitutes a simplification. The scenario for word-by-word 
extension of a rule would then be as follows: at some point language 
learners acquire the rule in over generalized form. 'When this is the case, 
lexical items which for historical reasons are not subject to the rule (i.e. 
do not constitute an environment for its application) have to be specified 
as exceptions to the rule. The extension of the domain of the rule con- 
sists in gradual elimination of such exceptions. '. . .'The first phase 
can be understood as motivated by the need to formulate grammatical rules 
in as simple (general) terms as possible. The second phase follows as the 
traditional forms are imperfectly passed on to succeeding generations of 
learners. It entails a simplification of the lexical component of the 
grammar. The first phase is necessarily abrupt — but it is covert and goes 
unnoticed by the speakers. The second phase is gradual and may be more or 
less apparent to the speakers as older and newer forms are utilized as 
indices of style' (p. 822). 

In order to account for word-by-word curtailment , Andersen then sup- 
poses that speakers can also undergeneralize ; 'If we suppose that this 
rule in its original formulation applied only to morphemes specifically 
marked as subject to the rule, then we can understand its curtailment — 
and eventual elimination — as the result of a simplification of the lexicon, 
exactly analogous to the one we defined for the extension changes. Here 
then, as in the extension changes, we have two phases: first a morpho- 
phonemic rule is formulated in such a way that individual lexical items 
have to be marked with respect to that rule; then the marking of these 
items is gradually omitted. The difference is that where the marked items 
are exempt from the rule, the simplification of the lexicon results in an 
extension of the domain of the rule; but where the marked items are the 
ones that are subject to the rule, the result of a simplification is a 
curtailment of the domain of the rule' (p. 823). The idea is, simply put, 
that language learners failed to discover the context of the rule and 
consequently had to mark the forms subject to it with an ad hoc rule fea- 
ture, whose elimination case by case curtails and eventually liquidates 
the rule. 

The interesting feature of this theory is that it claims to predict 
the direction in which a rule will change. Whether a rule is extended 
(eliminating the exceptions to the rule) or curtailed (eliminating the 
rule) is determined by its status (as major or minor, motivated or unmo- 
tivated, etc.). As things stand, this apparent explanatory power is 
dubious because there is no theory that tells us independently of the 
change itself what the relevant character of the rule in fact is. It is 
well enough recognized that the productivity of a rule is not at any rate 
a simple function of the frequency with which it is applicable (whether 


counted by types or tokens) . Rules which are originally restricted to a 
handful of forms may spread, and conversely, dominant rules may lose out. 
Andersen also appeals to the naturalness of a rule: phonetically motivated 
rules are extended but phonetically unmotivated rules are curtailed. This 
is beset with the same circularity because the status of a rule as motivated 
or unmotivated may depend on the choice of distinctive features, which in 
turn is made by looking at the subsequent changes that depend on it. 

It would appear that this solution does not remove the indeterminacy 
but simply shifts it: instead of determinate synchronic systems but 
unpredictable alternative possibilities of eliminating points of linguistic 
complexity, we have indeterminate synchronic systems of which each alterna- 
tive sets a fixed path for future analogical change. In effect, this is 
an attempt to reduce partial analogy to false analogy. 

There are difficulties with extending Andersen's theory to partial 
analogy in general. It does not work where the subset of cases affected is 
structurally (phonologically or morphologically) rather than lexically 
defined. At the intermediate stage of the loss of final devoicing in 
Yiddish, if Sadock is right, the domain of the devoicing rule was cut back 
from all obstruents to just -nd. The rule at this stage has added on a 
phonological context which unarguably makes it more complex than before. 
Even if we suppose that the rule had, before its curtailment, somehow 
become a minor rule (which in itself does not seem a justifiable assump- 
tion), it surely cannot be maintained that adding contexual restrictions 
even to minor rules makes them simpler! Only the kind of curtailment that 
comes from eliminating features on lexical items that mark them as subject 
to minor rules can be plausibly regarded as simplification. It is also 
impossible to suppose that the devoicing rule itself remained in its 
original general form and the curtailment was effected by marking lexical 
items as exceptions to it. Apart from the difficulty that the prior status 
of devoicing as a minor rule again cannot be justified, this would amount 
to saying that it is only accidentally true of the words which maintain 
devoicing at the intermediate stage that they all end in -nd. How could 
such a regularity have arisen if it was not apprehended as a rule by 
language learners at some point? The possibility of a purely item-by-item 
specification is excluded still more clearly in Thomason's Slavic example. 
The class of lexical items to which the rule applies at the two stages 
must be specified in a general way in the rule itself since the class of 
Animate nouns is an open one. It then follows that the generalization of 
the Accusative = Genitive rule results in a more complex rule. 

A second general type of approach is keeping more or less orthodox 
assumptions about the form of grammars and instead looking for some other 
version of the evaluation measure which would reduce instances of partial 
analogy to legitimate optimalization. 

A plausible scheme for revising the evaluation measure might be to 
keep simplicity in it but to have it interact with other parameters in 
some way. One such proposal, due to Ralph (1975:179), is to adopt Chen's 
(1973, 1974) device of 'metarules' as a regulator of partial analogy. 
These are akin to the 'marking conventions' introduced by Chomsky and Halle 
(1968:Ch. 9) to account for the 'intrinsic content' of the features, and 


still more to the scales of environments utilized by Foley (1977 and many 
earlier writings), but they implement the idea in a rather different way. 
Suppose we assign a higher 'index' to the fronting of long and rounded 
vowels than to the fronting of the corresponding short and unrounded (or 
lower) vowels, e.g. for the special case that is relevant to Old Swedish 
palatal umlaut as discussed by Ralph: 

(5) 2 o o 
la a 

Then multiplying (or adding) the indices for each feature will give a 
ranking of multifeature contexts which corresponds to their historical 
order of appearance: 

(6) a 1 palatalized first 
a,o 2 palatalized next 
5 4 palatalized last 

Given a general theory of 'metarules', the order of generalization would be 
predicted by the principle that rules are generalized in the order of 
increasing indices. 

In one respect this approach represents a major advance over the 
others we have considered so far: it implies, correctly, that the pattern 
of partial generalization is not random. It makes the interesting and, I 
suspect, correct prediction that no language will be found in which such 
a fronting process has unfolded in the reverse order, although from the 
viewpoint of the classical simplicity measure that order would be exactly 
equivalent to what actually happened in Old Swedish. 

No doubt the metarules as conceived by Chen and adopted by Ralph are 
lacking in generality. To assume that a table like (5) needs to be specially 
set up to index the possible contexts of every phonological process is to 
assume that there are no general principles behind the ordering of contexts, 
which is scarcely credible. Many cases fall under the rather simple over- 
arching principle that assimilation proceeds in the order of increasing 
phonetic 'distance' between focus and environment, as measured by the 
number of relevantly shared feature specifications between them. The more 
alike two segments are, the more prone they are to assimilate, other things 
being equal. For example, velars will be assimilated (palatalized) by an 
adjacent [-back] segment in the following order: palatalized consonants, 
y, i, e, ae. By the same principle, i fronts a before it fronts o. I do 
not wish to claim that this is the only principle at work: I would suppose, 
for example, that there are other reasons why the short vowels are fronted 
before the long vowels. Perhaps the explanation for the ordering of 
contexts will turn out to be exceedingly complex. This would still be 
better than a list of 'metarules', which can at best claim to be a summary 
of the observed facts but not an explanation. 

Moreover, neither metarules nor even a more general set of principles 
which would predict the order of contexts would yet address the entire 
problem of partial analogy. It is hard to see how these ideas could be 

extended to cases of partial analogy in morphological subcontexts, and the 
lexical cases are evidently quite impossible to handle that way. 

For the morphological and lexical cases there does exist another 
proposal which is entirely different from metarules in content but resem- 
bles it formally in that it involves enriching the evaluation measure. 
This is the idea of Harris (1973), Kiparsky (1972) and others that paradig- 
matic uniformity has to be recognized as an independent factor in change, 
not instead of but alongside simplicity (cf. Schindler 1974 for discussion). 
In effect, it says that the evaluation measure is to somehow check the 
outputs of grammars (it is impossible to tell just from the rules) and 
assign a 'cost' to allomorphic variation in paradigms. That is, it claims 
that the language learner preferentially selects grammars whose outputs 
have certain ' transderivational' properties. This is a retreat to a much 
weaker theory, since evaluation measure otherwise is defined purely on the 
grammar. It is also distressingly vague because crucial aspects of the 
proposal were left in the air: is allomorphy to be quantified by types or 
tokens? What are paradigms? How are the relative contributions of allo- 
morphy and complexity in terms of the number of feature specifications 
(not to speak of opacity) to be weighted relative to each other (King 1972, 
Kiparsky 1974)? In itself, that does not mean that the idea of a complex 
evaluation measure relying on allomorphy or some equivalent concept as one 
of its several components is necessarily wrong. The vagueness could be 
eliminated by exact definitions of 'allomorphy', 'paradigm', and other key 
notions of the theory, and some of it may indeed have to be retained as an 
inescapable correlate of the indeterminacy of analogical change itself. 
A priori there is no reason why there should not be independent criteria 
of evaluation in as functionally complex a system as language is, and why 
their relative weight should not be in certain respects indeterminate — all 
the more so if it can be shown that the weighting varies with linguistic 
function, so as to create styles adapted to the special requirements of 
speaker, hearer, and learner (Kiparsky 1975). 

There is another general consideration which ought to give one pause. 
The status of 'paradigm uniformity' and other such criteria in the theory 
of generative grammar would be quite different from the role of the cor- 
responding notions of analogy in the neogrammarian theory. In the latter, 
proportional analogy is assumed to underlie language acquisition and 
speech production, and its rule as a mechanism of change is simply a 
reflection of its more basic function. In the theory of generative gram- 
mar, simplicity has a corresponding role in language acquisition, and its 
relevance to analogical change would therefore be expected. Proportions, 
paradigmatic uniformity etc. do not figure in the theory of generative 
grammar and it would therefore be surprising to find them directing change. 
It would mean that there would be no essential relation between the way 
language changes and anything else about it. 

Although some authors (e.g. Lass 1976) have welcomed the proposal to 
introduce paradigmatic notions into the evaluation measure, I believe Bever 
and Langendoen (1971) were closer to the mark in criticizing it, or rather 
the program of which it formed a part, for treating function as structure: 
that is, for attempting to explain within the linguistic system a phenomenon 
which originates in the interaction of that system with other systems. The 


commonsense explanation for the uneven progress of analogy is that speakers 
resist some kinds of innovations and readily adopt others. To make some- 
thing of this idea we surely have to go beyond the formal theory of 
grammar to the theory of language use , specifically to that part of it 
which concerns the way speakers deploy the variation deposited in their 
grammars by ongoing linguistic change (Labov 1972) . I shall consider here 
one way in which such a program might be worked out. 

One could picture partial analogy arising through the interplay of 
language acquisition and synchronic variation by a process somewhat 
analogous to that by which the conditioning of sound changes is crystal- 
lized out of their initially variable application. A three-part scenario 
for it would be: 

(1) Language learners can acquire different grammars even when exposed 
to the same language, because of random differences in the speech data 
they are exposed to, in the order in which they are exposed to it, in the 
way they analyze indeterminate cases, and possibly because of other reasons, 
Such 'imperfect learning' (along with language contact and other sources 

of linguistic innovation) constantly replenishes the language with a 
fresh supply of coexisting variants. 

(2) Some of the variants arising this way will tend to be avoided in 
speech because they are dysfunctional or because they become stigmatized. 
Others will be favored in speech because they can be used to make things 
easier to say or to understand. These effects will show up in varying 
degrees depending on the circumstances. 

(3) Some of the bias in the deployment of variants will in turn be 
grammaticalized, with favored variants being acquired as obligatory and 
disfavored variants being lost. 

The assumption is, in short, that the variation due to 'imperfect 
learning' which we postulate as the source of analogical change is chan- 
neled jointly by the conservative influence of the established norm and by 
the functional needs of the system. Their combined effect should, if the 
assumption is right, predict whatever is predictable in the patterns of 
partial analogy. Let us consider the two factors in turn. 

The first way an analogical innovation might end up incompletely 
effected would be, then, that it gives way in particular subenvironments 
and/or in particular lexical items to the older standard, whether through 
self-monitoring by language learners, through explicit correction by other 
speakers, or both. Would any general type of pattern be likely to emerge 
from such a process? Presumably there could be no completely predictable 
pattern, if only because the linguistic data which bring about the partial 
adaption of the older system will come to the attention of different 
individuals in different orders and at different stages in their evolving 
grammars. Still, one would expect some innovations to stick out against 
the norm more than others, and thereby to have a diminished life expec- 
tancy. The innovations most likely to succeed would be the least 'blatant' 
or. 'salient' ones, for they have the best chance of surviving the language 
learners' own efforts to accommodate their speech to the community norm. 


and are also least likely to provoke outright correction or ridicule by 
other speakers . 

Intuitions on what exactly defines the crucial dimension of relative 
saliency are bound to be vague. Fortunately we have some slight empirical 
foothold on it, for dialectologists have long operated with the notion of 
relative saliency in studies of interdialectal adaptation. 

The relevance of certain kinds of saliency in syntactic change has 
recently been demonstrated in an important article by Naro and Lemle (1977) . 
They investigate the loss of subject-verb agreement in Brazilian Portuguese 
and conclude that the process spreads through the language in order of 
increasing saliency. They take the relative saliency of an innovation 
against the background of the old form to be a composite function of 
distinct factors, of which they name two: 

(1) An innovation which is more differentiated from the old form is 
more salient than one which is more similar to it. 

(2) An innovation is more salient in monitored (formal) speech than 
in unmonitored speech. 

It is plausible to add to this the following factors: 

(3) An innovation in a frequent form is more salient than an innova- 
tion in a rare form. 

(4) An innovation which alters surface structure or phonotactics is 
more salient than one which does not. 

This permits us to interpret many types of partial analogy: the tend- 
ency for 'small' alternations to be eliminated before 'big' ones (1), 
stylistic differentiation (2), frequency (3), and various 'structure- 
preserving' effects (4). 

In addition to an overall normative pressure which pushes back par- 
ticularly the most salient innovations, we are also supposing as another 
selective process, in line with traditional functionalist thinking, that 
innovations which serve the needs of speakers well or ill will be resisted 
or favored accordingly. Here is where the familiar functional factors 
find a natural interpretation. 

Obviously the factors that control the use of variants in speech, 
whatever they turn out to be, can have nothing to do with analogy per se, 
for variation from any source must be patterned in the same way once it 
is in the language (Samuels 1972:87). This is borne out by the resemblances 
in the characteristic profiles of partial analogy and variable rules 
(Kiparsky 1972) . We have in effect ended up dividing the explanation of 
analogical change between two different theories, one dealing with the 
source of the innovations (imperfect learning) and the other with their 
selection in speech and eventually in grammar. (A vague parallel might be 
the theory of evolution as a function of genetic variation and natural 
selection.) The constraints on analogical change are, then, jointly 

determined by the properties of imperfect learning and of variation. 

This part of my proposal is close in general spirit to Samuels (1972) , 
who also distinguishes between the components of 'variation' and 'systemic 
regulation' in change (see especially p. 135 f f ) . He remarks that 'neither 
sound-change nor analogy can be regarded as providing any more than the 
raw material of change; the process of change, from its initiation to its 
acceptance in a system, is considerably more complex than this ..." 
(p. 136). Samuels devotes his attention mainly to 'systemic regulation' 
and rather underestimates the structural conditions of change. I cannot 
agree with him, though, that new variants always reduce information-content 
and that subsequent selection is a 'compensatory mechanism' which 'consists 
of the restoration of the distances lost in [variation]' (p. 178). 

A certain amount of discussion has centered on the question whether 
change originates in competence or performance (see Dressier 1976 for 
recent discussion) . From the present point of view, the question is rather 
misleadingly posed, since change involves both competence and performance. 
This is also the conclusion reached through a different line of reasoning 
by Vincent (1974:436): "... analogy is neither a competence nor a 
performance phenomenon, but rather a sort of line linking these two aspects 
of language. ' 

3. False analogy. There is a second fallacy in assuming that the 
retention of forms from intermediate stages of language acquisition must 
simplify the grammar: they might be residual projections from an analysis 
which, even though optimal for the restricted linguistic data available to 
the learner at some stage, is not part of the optimal grammar of the whole 
language. If they are retained anyway, the result is analogy which creates 
a complication of the grammar. Everyone knows how the crucial data that 
may decide between competing analyses can lurk in obscure corners and 
elude the linguist working from a limited corpus. It would not be surpris- 
ing to find the language learner occasionally in the same predicament. 
Accordingly, the assumption that analogy must represent simplification with 
respect to the adult grammar is entirely unwarranted. The most that we can 
legitimately claim is that analogical innovations represent analyses which 
are optimal at the particular stage of language acquisition at which these 
analyses arise and for the particular set of primary linguistic data which 
is under consideration by the language learner at that state. Even a 
partial retention of these innovations may force a complication on the 
grammar when it is reanalyzed on the basis of a fuller set of data. 

It is obvious that the sort of paradigmatic leveling which fails to 
simplify and possibly complicates the adult grammar is often ascribable to 
a simplification at an earlier stage in language development. For a 
learner of Latin whose relevant data includes only the oblique case forms 
honorem, honoris etc., honor is the simplest projection for the nominative 
singular. Never mind that having later learned honestus and acquired an 
intervocalic s ->■ r rule anj^ay for cases like flos'^flDris he might be 
better off with the older honos after all: the horses are already out of 
the barn. For a learner of Canadian English who has not acquired the 
ay '^ By and vowel shift rules, write would simply be underlying /rayt/, 
and the simplest projection for the suffixed forms write + er,write + ing 


would be [raydar], [raydlQ] — though once the dependence of ay vs. ey on 
the voicing of the following consonant is learned, the transparent order- 
ing which yields the older forms [raydar], [raydlQ] might be preferable. 

Similarly, cyclic ordering can be a structural accommodation of 
analogy which imposes on derived words the surface features of stems they 
contain. It is not necessary, if this is so, to suppose that cyclic order- 
ing is to be assigned preferred status by the evaluation measure, for the 
tendency of rules to shift into the cycle can be explained without it. 

This inertia of the language learner's own mistakes is perhaps the 
minimal assumption which can be made to account for 'false' analogy. It 
should be distinguished from McCawley's (1975) more radical, by no means 
implausible idea (cf. Fodor 1978:62) of an inertial effect in the language 
learner's own grammars . One version would be that language learners 
resort to restructuring only 'under duress', perhaps if the grammars they 
have formulated cannot be patched up to incorporate new data in a way 
which satisfies some criteria — which remain to be specified by a theory of 
language acquisition. Alternatively, one could imagine the degree of 
resistance to restructuring depending how great the restructuring would 
be, perhaps as weighed against the amount of simplification it would 
afford. These hypotheses would entail that the evolving grammar is not 
necessarily at any particular stage, including the finished grammar, 
optimal for the data it covers, though it of course might be. Even more 
severe would be the assumption that all language acquisition is, in effect, 
a patchup job, with new data always being absorbed by means of the small- 
est possible grammatical adjustments, without restructuring (cf. also 
Haber 1975 for some related observations) . That would mean that the 
child's first hypotheses always fix the future course for the evolving 
grammar. The optimal grammar would on this view be an ideal construct 
which was perhaps never realized by any actual language learner. 

Similar in spirit is Andersen's (1973) idea of 'adaptive rules'. It 
amounts to assuming that the learner may be stuck with a nonoptimal 
analysis which in itself generates the wrong output and is mapped into the 
right one by 'adaptive' rules. Andersen illustrates his concept with an 
interpretation of a sound change which takes palatalized labials to den- 
tals. He assumes that at some point language learners began to interpret 
palatalized labials as being dentals ([p'] as /t/ etc.). To account for 
their actual pronunciation as [p'] rather than [t] they therefore required 
an adaptive rule which turned dentals to palatalized labials (/t/ -^ [p'l 
etc.). Having thus become 'virtual' dentals by reanalysis, the palatalized 
labials changed to dentals phonetically by the loss of the adaptive rule. 
I admit to finding this unconvincing as a proposal about sound change. The 
thought that language learners would somehow come to analyze palatalized 
labials as being really dentals, without any phonological motivation for 
doing so, and later start to pronounce them in accordance with this reanaly- 
sis is too idealist for my taste. For analogical change Andersen's notion 
of adaptive rules might make more sense. Even there, however, they seem 
to me to underestimate the flexibility of the language learner. 

These hypotheses make different but largely overlapping predictions 
about analogy. They share the essential property of allowing for the 


real-time course of language acquisition to determine aspects of the system 
that eventually emerges. The precise causes of the inertia, and the degree 
to which it is mediated by the structure itself, could well be some 
combination of those mentioned here. The essential empirical difference 
between my assumption that the language learner freely restructures his 
grammar when encountering new data that motivates it, and McCawley's and 
Andersen's assumption of more limited adaptive measures for coping with 
new data, would seem to be the following. On the assumption of continuous 
restructuring, a certain amount of selection and organization would be 
expected to be imposed on the innovations by the adaptive system, due to 
the efforts of the learner to integrate the new data into his grammar in 
the optimal way. If on the other hand the new data is accommodated by 
tacking on 'patchup' rules or 'adaptive' rules, to the already formulated 
grammar, no such structural effects should presumably show up. Herein 
lies the relevance of cases where the result of analogical leveling is 
systematically adopted into the language precisely where it can be accom- 
modated in a simple way into the system, and suppressed elsewhere. This 
would be incomprehensible if we did not assume grammatical reintegration 
rather than makeshift 'adaptive' rules. 

'Adaptive' rules do appear to be the right mechanism for certain 
sorts of cases, hypercorrect forms, for example. They show precisely the 
distinction between the ordinary result of language acquisition, where 
reanalysis takes place, and the special situations caused by limited access 
to the relevant linguistic data, which lead to hypercorrection. 

If this view of paradigmatic leveling is right, then it follows that 
efforts to combine paradigmatic leveling and 'distinctness' effects in 
terms of a unified set of 'paradigm conditions' (Kiparsky 1972) or 
'Humboldt's Universal' (Vennemann 1974) were fundamentally misguided. 
Paradigmatic leveling is a direct reflex of language learners' tendency 
to simplify grammar, whereas avoidance of homonymy, distinctness of gram- 
matical categories etc. are imposed on the system by the purposive selection 
of speech variants from whatever source, including ordinary sound change. 
As a result of their different etiology, they are also implemented in the 
grammar in different ways. It is striking that 'distinctness' may be 
effected by the most varied structural means. Even for one category within 
one grammar it is possible to demonstrate the characteristic 'conspiracy' 
situation where the distinction between morphological markers is maintained 
by numerous rules and restrictions on rules acting in concert (cf. Kiparsky 
1972:214). Hogg (1975) has pointed out that such cases do not seem to 
occur with paradigmatic leveling. This is predicted from our assumption 
that the selection of variants favors category distinctness but is neutral 
as to paradigmatic leveling. 

We have argued that 'paradigm' conditions can be eliminated from the 
evaluation measure. It is tempting to try a similar reduction for opacity. 
Can we interpret shift into transparent rule order as the result of incor- 
porating into the language the projections of the simplest grammar for 
some set of core data, as we did for paradigmatic leveling? That would 
let us pare down the evaluation measure even further by taking even 
opacity out of it, essentially restoring it to its original form of a pure 
simplicity measure. 


Let Ki, R2 be two opaquely ordered rules which are reordered into the 
transparent order R2, Rl- Consider now how the reordering can be analyzed 
as the grammatical reintegration of the projection from Ri at a stage when 
R2 has not been acquired. Taking first reordering into feeding order, 
suppose we have the rules 

(7) i. a -»■ b / _ c 
ii. d -*■ c / _ e 

where (ii) counterfeeds (i) , e.g. /ade/ ^ aee, not *bee . Suppose that the 
primary linguistic data under consideration by the language learner at 
stage Sjj is enough to motivate (i) but not (ii) . Such instances of the 
output ae from (ii) as are encountered are therefore incorrectly analyzed 
as derived from Ice/ rather than /de/ at S^. From a+oe the grammar there- 
fore projects the 'wrong' output boe by (i) . Suppose that at stage S^+l, 
rule (ii) has been learned. The wrong output hae of Sn can be reintegrated 
into Sj^+1 simply by ordering the rules in the transparent order (ii,i). 
If the analogical bee wins out, reordering has taken place. If correction 
to ace prevails, normal language transmission has taken its course. So 
it is possible to account for reordering into feeding order by the process 
of grammar simplification (imperfect learning) without supposing that 
feeding order, or transparency, is in itself more highly valued, if the 
entirely natural additional assumption of learners' output inertia is 
made and we further postulate an order of acquisition. Note that our 
scenario predicts the asymmetry of reordering. A switch from feeding to 
nonfeeding order would be inexplicable by the process we have outlined. 

The above proposal is extreme in that it reduces assumptions about 
the factors governing language learners' structuring of data to the SPE 
minimum (simplicity) and puts the whole burden of explaining transparency 
effects, as was done with leveling, on the postulated inertia of learner's 
own earlier speech forms and on assumptions about the order of acquisition 
of rules. There is also a more moderate version of the same general idea, 
where the role we have so far assigned to the order of acquisition of 
rules is given to a minimal additional assumption about the language learn- 
ers' structuring of data: 

(8) For any set of data D, learners select the shallowest underlying 
representations compatible with the simplest grammar of D. 

All that this says is that speakers do not indulge in gratuitous abstrac- 
tion over and above what is required for linguistic generalizations. 
(8) is of course quite distinct from such constraints as the Alternation 
Condition, which limit the abstractness of underlying representations 
regardless of any resulting grammatical complications. It has, I think, 
been generally accepted even by those who reject these severer constraints, 
and would seem to be independently necessary even if these are accepted, 
in order to predict the restructuring on nonalternating outputs of neutrali- 
zation rules (for example, word-final obstruents restructured as unvoiced 
in isolated forms in languages like German, Yiddish, etc., with final 
devoicing rules, where the Alternation Condition does not predict the 
choice of the devoiced obstruents as basic because the devoicing rule 
applies in a derived environment) . 


Most of the predictions about transparent ordering can be had this 
way too. Instead of assuming a stage where only (i) has been learned, 
suppose that (ii) has been learned but that those outputs of it (ce) for 
which the learner is not (yet) in a position to motivate an underlying 
source /de/ are systematically represented by him as /ce/, even though the 
more abstract /de/ would be equally simple. (So far we have done without 
this assumption) . Then the grammar projects in such cases from a+ae the 
output bee rather than aae. If retained, it must be structured in terms 
of feeding order in all cases where ce is reanalyzed by the learner as 
/de/ on the basis of data encountered later. 

Another likely victim of our reductionist rampage is the revised 
Alternation Condition (Kiparsky 1973), which stipulates that obligatory 
neutralization rules (or, in a milder formulation, obligatory nonautomatic 
neutralization rules) apply only to derived inputs. 

4. Conclusions. To summarize, we have discussed two recalcitrant 
types of analogy, partial and false analogy, and concluded that they too 
yield to the interpretation of analogical change as simplification under a 
more realistic interpretation of change which allows for inertia in the 
acquisition process and for interference between the learner's system and 
the target norm of the speech community. This is partly good news and 
partly bad news. The bad news is that linguistic change can no longer be 
used as a probe into linguistic structure in the direct and naive sense 
that many of us used to think. Before we can exploit historical evidence 
for synchronic purposes we need a firm theory of the intervening factors — 
the effect which the concrete process of acquisition itself and differen- 
tial saliency can have on what is acquired. It follows for example that 
the arguments concerning abstractness will have to be rethought in so far 
as they are based on historical evidence. The good news is that we can 
restore the integrity of the evaluation measure by eliminating from it the 
accretions which it had acquired. These accretions have two suspicious 
properties in common. First, they require reference not to the form of 
the grammar (as simplicity does) but to the relationship of the output 
of the grammar to something else — either to other output forms (paradigm 
conditions), to rules (opacity), or to underlying forms (recoverability, 
Hale's deep/surface disparity avoidance). Second, they were motivated 
practically exclusively from historical evidence. Arguments that they are 
required to select the synchronically correct grammar have not been forth- 
coming. The elimination of these considerations from the evaluation 
measure would therefore be a gain rather than a loss. 

We wind up with a surprising conclusion. There are two assumptions 
of the theory of generative grammar that have been subjected to strong 
criticism: the simplicity metric, and the idealizations of instantaneous 
language acquisition and the ideal speaker-hearer. What we have seen is 
that by eliminating the second from historical linguistics we overcome 
the objections to the first. We seem to have found a way of retaining 
the simplicity measure without giving up what is right about traditional 
functionalist thinking. Structural and functional considerations both 
play an essential role but as part of separate systems. This relieves 
the evaluation measure of the burdens which purely structural theories 
unjustly placed on it and eliminates the unfortunate ' teleological ' 


element of traditional functionalism, as well as its inability to account 
for structural patterning in analogical change. The ' teleological' effect 
can now be seen to follow from the functionally and structurally determined 
'filtering' of the variation generated by the process of imperfect learning, 
while the structural patterning follows from the source mechanism of imper- 
fect learning as well as from the grammaticalization of the functionally 
patterned variation. 

This essay is extracted from a monograph on analogy to be published 
separately. It presents the main line of argument in a condensed form, 
omitting most of the critical discussion of alternative views and nearly 
all supporting data. This work was supported in part by the National 
Institutes of Mental Health Grant No. 5 POl MH13390-12. 

There is a terminological confusion to be cleared up. The term 

analogy has become established in reference to both a particular type of 
phenomenon in linguistic change, whose existence is not in dispute, though 
there may be disagreement about where its exact boundaries lie, and a 
particular theoretical interpretation of this phenomenon, namely the pro- 
portional schema or its equivalent, whose correctness has certainly been 
cast into doubt. My own practice earlier was to simply avoid the term, 
and to replace it, in both senses mentioned above (the phenomenon and the 
process I was proposing to account for it) by simplification . The term 
is also avoided by Andersen (who likewise objects the traditional inter- 
pretation and speaks of simplification (1969 :82A), though from a somewhat 
different point of view): 'The changes with which we will be concerned 
would be called analogical by some. We will not use this term, for terms 
like "grammatical analogy" as traditionally used are highly ambiguous, 
and we need to be specific' (p. 807). Others have continued to use the 
term analogy to refer to the mechanism as traditionally understood, that 
is, to the extension of surface patterns by the proportional schema or its 
equivalent (e.g. Chomsky and Halle 1968:156, fn. 12; Isacenko 1970). 
Finally, King (1969) and others maintain the term in ambiguous use both 
'as a cover designation for . . . instances of change' (p. 128) and as a 
term for the proportional mechanism (e.g. p. 182). The ambiguity would 
seem to be easily resolved by context but has nevertheless spawned a 
fatuous line of argument where generative criticisms of the surface pro- 
portional theory of analogy are called 'attacks on analogy', and their 
authors are then, absurdly, denounced for failing to distinguish between 
analogy and sound change ! 

None of these ways is satisfactory. It is best to have a separate, 
neutral term, for the type of historical change whose interpretation is 
at issue, and distinct terms for the various processes which have been 
proposed to account for it. I will here use analogical change (or some- 
times just analogy ) for the historical phenomenon and proportional 
analogy for the general class of processes by which it is traditionally 
explained (suitably qualified when necessary to devote specific versions 
of it), and simplification for the general class of processes which has 


been more recently suggested in its stead (again suitably qualified when 
necessary) . 


Andersen, H. 1969. A study in diachronic morphophonemics: The 'Ukrain- 
ian prefixes'. Language 45, 807-30. 

1973. Abductive and deductive change. Language 49. 765-93. 
Bever, T,, and D. T. Langendoen. 1971. A dynamic model of the evolution 

of language. Linguistic Inquiry 2. 433-64. 
Bley-Vroman, R. 1975. Opacity and interrupted rule schemata. Papers 

from the Eleventh Regional Meeting, 73-80. Chicago: Chicago 

Linguistic Society. 
Chen, Matthew. 1973. On the formal expression of natural rules in 

phonology. Journal of Linguistics 9. 223-49. 

1974. Natural phonology from the diachronic viewpoint. Papers 
from the Parasession on Natural Phonology, 43-80. Chicago: Chicago 
Linguistic Society. 

Chomsky, N. , and M. Halle. 1968. The sound pattern of English. New 

York: Harper and Row. 
Dressier, W. 1972. On the phonology of language death. Papers from the 

Eighth Regional Meeting. Chicago: Chicago Linguistic Society. 
1976. How much does performance contribute to phonological 

change? Wiener Linguistische Gazette 13. 3-18. 
Fodor, J. D. 1978. What kind of exception is have gotl Linguistic 

Inquiry 9. 45-65. 
Foley, J. 1977. Systematic morphophonology. Cambridge: Cambridge 

University Press . 
Haber, L. R. 1975. The muzzy theory. Papers from the Eleventh Regional 

Meeting, 240-56. Chicago: Chicago Linguistic Society. 
Harris, J. 1973. On the order of certain phonological rules in Spanish. 

A festschrift for Morris Halle, ed. by S. Anderson and P. Kiparsky, 

59-76. New York: Holt, Rinehart, and Winston. 
Hermann, E. 1931. Lautgesetz und Analogie. Abhandlungen der Gesellschaft 

der Wissenschaften zu Gottingen. Phil. -Hist. Klasse, Neue Folge 

23,3. Berlin: Weidmannsche Buchhandlung. 
Hock, H. H. 1975. Current trends in historical linguistics. Proceedings 

of the first meeting of the Kentucky Interdisciplinary Conference on 

Hogg, R. M. 1975. The place of analogy. Neophilologus 59. 109-13. 
Hooper, J. 1976. An introduction to natural generative phonology. New 

York: Academic Press. 
Isacenko, A, V. 1970. East Slavic morphophonemics and the treatment of 

the jers in Russian: A revision of Havlik's Law. International 

Journal of Slavic Linguistics and Poetics 13. 73-124. 
Kenstowicz, M. , and C. Kisseberth. 1977. Topics in phonological theory. 

New York: Academic Press. 
King, R. D. 1969. Historical linguistics and generative grammar. 

Englewood Cliffs, N.J.: Prentice-Hall. 

1972. A note on opacity and paradigm regularity. Linguistic 

Inquiry 3. 535-39. 

Kiparsky, P. 1971. Historical linguistics. A survey of linguistic 

science, ed. by W. 0. Dingwall, 576-649. College Park, Md.: Univer- 
sity of Maryland Press . 

1972. Explanation in historical linguistics. Goals of linguistic 
theory, ed. by S. Peters, 189-227. Englewood Cliffs, N.J.: Prentice- 

1973. Phonological representations. Three dimensions of linguis- 
tic theory, ed. by 0. Fujimura, 1-136. Tokyo: TEC Company. 

1974. On the evaluation measure. Papers from the Parassession 
on Natural Phonology, 328-37. Chicago: Chicago Linguistic Society. 

. 1975. Comments on the role of phonology in language. The role 

of speech in language, ed. by J. F. Kavanagh and J. E. Cutting, 271- 

80. Cambridge, Mass.: MIT Press. 
Kuryiowicz, J. 1956. L'apophonie en indo-europeen. Wroclaw: Polska 

Akademia Nauk, 
. 1964. The inflectional categories of Indo-European. Heidelberg: 


1968. Indogermanische Grammatik, Band 2: Akzent-Ablaut . 

Heidelberg: Winter. 

1970. Les verbes germaniques du type lukan. Melanges Fourquet, 

Labov, W. 1972. Sociolinguistic patterns, Philadelphia: University 

of Pennsylvania Press. 
Lass, R. 1976. English phonology and phonological theory. New York: 

Cambridge University Press. 
Mascaro, J. 1976. Catalan phonology and the phonological cycle. Ph.D. 

dissertation. MIT. 
McCawley, J. 1975. Acquisition models as models of acquisition. Pro- 
ceedings of the 1974 NWAVE Conference. Washington, D.C.: Georgetown 

Naro, A., and M. Lemle. 1977. Syntactic diffusion. Ciencia e Cultura 

29. 259-68. 
Paul, H. 1920. Prinzipien der Sprachgeschichte. Fifth edition. Tubingen; 

May Ni emeyer. 
Ralph, B. 1975. Phonological differentiation: Studies in Nordic language 

history. University of Gothenburg. 
Robinson, 0. W. 1976. A 'scattered' rule in Swiss German. Language 52. 

Sadock, J. M. 1973. Word-final devoicing in the development of Yiddish. 

Issues in linguistics: Papers in honor of Henry and Renee Kahane, 

ed. by B. B. Kachru et al., 790-97. Urbana, 111.: University of 

Illinois Press. 
Samuels, M. L. 1972. Linguistic evolution. Cambridge: Cambridge 

University Press. 
Schindler, J. 1974. Fragen zum paradigmatischen Ausgleich. Die Sprache 

20. 1-9. 
Skousen, R. 1975. Substantive evidence in phonology: The evidence from 

Finnish and French. The Hague: Mouton. 
Thomason, S. 1976. Analogy as grammar complication. Proceedings of the 

Second International Conference on Historical Linguistics, ed . by 

W. Christie. 401-9. Amsterdam: North Holland. 
Vennemann, T. 1974. Words and syllables in natural generative grammar. 

Papers from the Parasession on Natural Phonology, 346-74. Chicago: 


Chicago Linguistic Society. 
Vincent, N. 1974. Analogy reconsidered. Proceedings of the First 

International Conference on Historical Linguistics, ed. by J. Anderson 

and C. Jones, Vol. II, 427-45. Amsterdam: North Holland, 
von Wartburg, W. 1969. Problems and methods in linguistics. New York: 

Barnes and Noble. (Translated from the 2nd French edition of 1963. 

Originally published in German in 1943.) 
Watkins, C. 1969. Indogermanische Grammatik, III.l, ed. by J. Kurylowicz. 

Heidelberg: Winter. 


Charles A. Ferguson 
Stanford University 

In this paper I intend to take a reactionary point of view and say 
something quite old-fashioned. I will say that linguists should write 
grammars. Then I will add the somewhat radical thought that they should 
write multilingual grammars. It may take a considerable number of words 
to say that, but you have fair warning that that is what I am going to say. 

Writing grammars is a traditional task of linguists, in fact it is 
almost a defining task of linguists. It is not the only one, but it is 
certainly something linguists are known to do. They attempt to provide 
a principled description of a particular language; either the whole lan- 
guage or some part of it. They have been doing that for a long time, and 
I am suggesting that they should not lose heart, but should continue doing 
it, because it is a good thing. To be sure, there are other tasks, such 
as the formulation of theories about language in general and theories about 
how to write grammars. Some of these tasks are just as important or pos- 
sibly more important than writing grammars. But the writing of grammars 
is a good thing. 

Why do linguists write grammars? I would like to think that the first 
reason they do so is simply to extend our knowledge about human languages, 
so that we know more things about languages. I realize that in some quar- 
ters this is a fairly old-fashioned view of why one would write a grammar, 
but it is what comes to my mind first. Second, linguists write grammars 
in order to codify languages. That is to say, to prescribe the form of a 
standard language for its users. This is also supposed to be not such a 
good thing according to the ethos of modern linguistics, but if we think 
back to the great linguists in the history of our science, most of them 
have written prescriptive grammars. Panini was certainly a prescriptivist, 
no matter how else we may characterize him, and he was also one of the 
great linguists and grammar-writers of all time. Third, linguists write 
grammars in order to have a basis for pedagogy, so that people can teach 
the language more effectively and once again, although modern linguists 
belittle such a purpose, some of the great grammarians have had exactly 
this aim. Sibawihi, the most famous Arab grammarian, certainly wrote his 
grammar, at least in part, so that Persians would know how to get their 
Arabic right and so that Arabic could be taught properly to Persians. 
Next, many linguists are interested in writing grammars so that they can 
compare one grammar with another, whether for typological, historical, or 
areal comparison or for pedagogically oriented contrastive analysis. In 
other words, an important purpose of writing grammars is to learn something 
about the nature of language in general by comparing grammars. And then 
finally, I put it in fifth place, you can write a grammar merely to check 


out some theory, that is merely to see if you can confirm or disconfirm a 
particular theory or theoretical model that you are interested in, or per- 
haps modify the theory or extend it in some way to take care of other 
kinds of material. People often put this purpose at the head of the list 
and say that grammar writing is not worthwhile unless you are checking a 
theory. In the present state of linguistics, I prefer grammars. Good 
grammars last well and are used by all kinds of theorists, but theories 
come and go quickly and too often interfere with balanced coverages in 

If we agree that linguists, for these various reasons, do write gram- 
mars, whether attempts at whole grammars or partial ones, why is it becoming 
less fashionable to do so today? I think the chief reason is that the task 
is getting harder and harder to do. That is, the more that linguists study 
language, the more they find there is to put in grammars. There is also the 
basic problem of how to go about writing a grammar, which is becoming 
increasingly troublesome, too. But the first problem is that there is just 
so much there. Language is a very complex phenomenon, and linguists are 
dealing with more and more aspects of it. All you have to think of is how 
linguistics programs today have full courses in pragmatics, child phonology, 
social dialects, and a host of topics which a few years ago were not in the 
catalog. Presumably, these things have to get into grammars, somehow, and 
it seems to have become an impossible task to write a grammar. In fact, a 
philosophy of despair has emerged. Some linguists say, 'The most you can 
do is write some fragment of some possible grammar of some part of some 
language (preferably with theoretical significance) . ' 

Another problem is the old-fashioned question of what constitutes a 
language. How do you know when you have one? This is an important part 
of my topic: natural language as the object of description. It is not 
easy to define what we mean by a language, i.e., the structural or func- 
tional criteria we need for determining when we have a language. Linguists, 
by and large, have put that problem aside. It comes up every once in a 
while as something that we have to struggle with, but the fact is we have 
ready at hand no way of deciding when the object we have is a language, and 
not a dialect or two languages or something else. 

Let us assume that the notion 'a language' has a natural status of some 
kind as an object of description which is highly productive for understand- 
ing the nature of human language behavior in general, how it changes, and 
how it is processed individually and socially. Such an assumption does not 
exclude the study of partial, restricted and marginal language behavior 
which is abnormal or pathological, but it accords with the view that whole 
languages are the normal object of linguistic description. The criteria for 
delineating this natural unit would include autonomy, stability, and func- 
tional range. By 'autonomy' I mean the degree to which the object in 
question is structurally and functionally distinct from neighboring 
phenomena; this may involve such measures as structural coherence, mutual 
intelligibility, extent of superposed variety, and attitudes of speakers 
(cf. Ferguson & Gumperz 1960). By 'stability' I mean the degree of internal 
variation; extremes of diachronic or functional flux in language behavior 
seem to represent atypical language situations different in kind from normal 
use of language (cf. Ferguson & DeBose 1977). By 'functional range' I refer 


to the degree of restriction in semantic range and acceptable uses; an 
object restricted to religious use or lingua franca functions and not 
serving as the mother tongue of a social group is not a full language 
(Ferguson & DeBose 1977). These criteria are mentioned as illustrations, 
not as offering a solution. The identification of the linguist's object 
of description is not a trivial issue; it is a tough problem and one that 
will not go away just by our ignoring it. 

Linguists choose different ways of getting around that problem. One 
alternative is simply to select a corpus of some kind, and do your best to 
write a grammar of that. This is usually done not just to describe the 
corpus itself, but to provide a description that will apply to a larger 
body of material. Or you can take the way out of just describing one idio- 
lect, and regarding that as the normal object of linguistic description, 
without attempting to characterize the larger unit. Or you can try to find 
out what langue really is, i.e., find out what is shared by the particular 
community, and leave out things that are not shared. 

There are other ways: you can even back out all together and say 'the 
language that I am writing a grammar of is whatever it is that my grammar 
describes', and then the grammarwriter 's decisions are inexplicit but not 
subject to debate. Another thing we can do is to take selected fragments 
of language and say, 'This is what we are going to write about." Following 
this approach, it may be possible to find some kind of useful core fragments 
that we would all agree are useful information about any language we wanted 
to describe. Examples of this method are the efforts by Bernard Comrie or 
the Stanford Phonology Archive to produce a checklist about things you would 
like to find out about any language that you start to describe (Ferguson 

The next issue is variation. We have all come to recognize the fact 
that languages are full of variation. There is just no way that we can 
avoid it. An object is not a human language if it does not vary along 
many dimensions. There must be dialect variation depending upon where the 
users come from or what part of society they are in. There must be register 
variation conditioned by the use, the occasion, the addressee, the topic, 
or whatever. Also, many languages have superposed varieties which we find 
a little awkward to call dialects or registers, as when a standard variety 
or even a national standard on top of regional standard is part of a par- 
ticular language situation. There are actually a number or recognized 
types of language situations, such as diglossia, standard language with 
dialects, decreolization continuum, and so on, as well as many situations 
which do not fit into those types. Most linguists are getting reconciled 
to the fact that they must include an account of variation in writing the 
grammar of a language. 

Let me give one illustration from my own experience in a diglossic 
community: the diphthongs in the kind of Arabic that is spoken in Damascus. 
There are two long mid vowels, i and o, and also two dipthongs ay and au , 
and there is some kind of special relation between them, since many words 
are pronounced sometimes with the monophthongs and sometimes with the 
diphthongs. Standard grammars of Classical Arabic give only ay and aw, 
and grammars of colloquial Damascene Arabic give e and 5 with only marginal 


examples of ay and au, yet speakers of Arabic in Damascus show extensive 
variation in the pronunciation of these sounds. Control of the appropriate 
use of this variation is part of the basic linguistic competence of Damascus 
speakers and belongs in the grammar. Even a person who has very little 
education in Classical Arabic has learned from very early childhood to 
use ay and aw in certain sterotyped expressions, and even the most 
scholarly Arab using Classical Arabic pronounces some proper names 
and foreign loanwords with e and o. What is of greater interest and 
what forces me to say that the phenomenon cannot be described by describ- 
ing colloquial Arabic and Classical Arabic and some type of relationship 
between them, is that it is possible to pronounce most words in the lexicon 
either way, depending on whether speakers want to show that they are 
speaking in a classical style or a colloquial style. There is a very 
complex system of variation in which some words tend to be more classical, 
some tend to be more colloquial and all fluctuate to a certain extent 
within the system. An analysis which gives only e, 5 or only ay, aid is 
missing an important point about the way people speak Arabic in Damascus. 

There are, of course, various ways of including such material in gram- 
mars. I do not have to tell you that people have invented things called 
variable rules as one way of doing this. There is also an oldfashioned 
way of describing one part of a language in the body of the grammar 
and putting the other varieties in footnotes or specially marked pas- 
sages in the grammar. I am referring to such grammars as Smyth's old 
Greek grammar (Smyth 1956), which has 'Classical' Attic in the main 
body of the text and footnotes to tell you the Ionic, the Doric, and 
so on. Or, let us say, Kellogg's Hindi grammar (Kellogg 1965), which 
gives an impressive range of dialect variation in the notes and marginal 

These may be very primitive attempts to cope with getting variation 
into the grammar, but sometimes they are much more informative than a 
grammar which carefully and purposefully removes a great deal of the 
variation and describes only a uniform, isolated part of the language 
behavior. One important part of characterizing a language is identifying 
the nature and extent of variation in it. That is, one does not only 
compare two languages as to phonological structure, syntactic structure, 
lexical structure, semantic structure, or other 'structures' or 'compo- 
nents' or 'levels' that traditional linguists talk about. Two languages 
which are basically similar in phonology and syntax, let us say, but one 
of which has a great deal more variation of a certain kind than the other, 
differ in a linguistically significant way. We need ways of talking about 
the nature and extent of variation in a particular language as a part of 
its grammar. 

Just as variation within a language is universal, we could almost 
say that multilingual ism in speech communities is universal. A great 
many speech communities are characterized by such structural discontinui- 
ties in the kind of language used in the community that by almost any 
criteria one would want to say that several different languages are used 
in the community. There are, of course, monolingual speech communities 
where the variation is so limited that no one would want to speak of more 
than one language. I suspect we would not have to travel very far from 


Champaign, Illinois, for example, to find one of the surrounding towns 
that is a monolingual speech community in this sense. In such communi- 
ties there will be register variation, dialect variation, and so on, 
but it will all constitute unmistakably a single language. However, if 
we look around the world, including many parts of the United States, 
there are speech communities in which a number of different languages 
are part of the linguistic repertoire of the community. Distinct 
languages exist side by side and are part of the whole scheme of vari- 
ation of the speech community. What 1 want to suggest tonight is that 
if we are going to write a grammar of what goes on in a speech community 
that uses — let us say — four languages, instead of writing four separate 
grammars and then writing rules for when people use one language or 
another, we should try to write a unified grammar in which all this '/ 
variation fits somewhere. Instead of four separate structures which are 
variously used, we may have one linguistic structure with complex internal 

Perhaps the easiest way to argue for this suggestion is to give 
examples of multilingual speech communities which linguists have tried 
to describe. Now, as far as I am aware, linguists have not really tried 
to write whole grammars of such linguistic repertoires. In fact, what 
they typically do is to write what is essentially a problem paper. They 
take some set of interesting facts about a multilingual speech community 
and say, in effect, 'See, look how these work in this community, and 
isn't that exciting', or 'interesting', or 'doesn't that have implica- 
tions for this or that?' But they do not try to say, 'These facts are 
part of a whole grammar of several languages operating together. ' The 
examples I will use were chosen more or less at random and are not 
intended to be typical, but they are all possible examples of multi- 
lingual speech communities where repertoires could be analyzed as 
unified grammars. 

Norman Denison has written a number of articles about a little town 
called Sauris or Zahre, an isolated speech community in the Italian Alps 
(e.g., Denison 1968). It is a little town of fewer than a thousand 
people, but it has been flourishing for a long time and presumably will 
continue to flourish for some time to come. Everybody in that town 
speaks three languages. The local variety of Standard Italian is the 
language of education, the language of school, even kindergarten, and 
the language that parents often speak to their children so that they will 
do better in education. It is the language in which almost all writing 
takes place, the language of certain kinds of formal uses and so on. 
Second, there is Friulian, a kind of local regional standard of Italian, 
which is very different from Standard Italian. This is the language 
everybody uses informally when non-co-villagers of the region are present. 
If at least one person from another village in the Carnian region is 
there, everyone switches to Friulian. It is also used under other cir- 
cumstances. For example, some people who have had high school education 
outside of Sauris use it as the familiar conversational idiom of group 
solidarity. Finally, there is the German dialect of Sauris/Zahre, a 
unique variety of German which is limited to that town in its essential 
characteristics and which is used when people are talking to each other 
informally as fellow inhabitants of the town. In a way, you might say 

that it is the nearest thing to a mother tongue Saurians have. This 
brief summary is, of course, an oversimplification. In any case, 
Norman Denison's studies of the Saurian speech community offer very 
interesting descriptions of sample utterances and the occasions for 
code switching. He is concerned with the situational factors which 
account for the use of one variety versus another. He identifies 
thirteen such situational factors and proposes a tentative hierarchy 
of them. His view of the immediately desirable further research is to 
improve on that hierarchy, by finding out when one factor outweights 
another, when different factors are in conflict, and so on. 

Denison takes a big step toward the position I am suggesting 
here. He says, 'For Sauris we have to see the three languages as. .con- 
stituting, for community-internal purposes, one super-code, as it were. 
The choice of one of them as macro-textual vehicle for the linguistic 
component of a communicative event is like the selection in a monoglot 
community of register at the earliest degree of delicacy' (Denison 
1968:585). He even offers some careful parallels between langague 
switching in Sauris and register switching within English in Britain. 
He does not, however, take the final step in attempting to write a uni- 
fied grammar of the phonological or grammatical competence of Saurians. 

As a second example, let us take a study done by Peter Trudgill, 
who is known to us here as a faculty member of the Institute. He has 
written a really excellent article about reduction and simplification 
in a community of Albanian speakers in Greece — a community which has 
been there for several centuries (Trudgill 1976-7) . The members of 
this speech community are all bilingual in Arvanitika and Greek, and in 
discussing the differences between Albanian proper and Arvanitika, Trud- 
gill notes that in some instances the slack is taken up by Greek. The 
simplest example is that of the lexicon; a large niomber of Albanian 
words have been replaced by Greek. Apparently an Arvanitika-speaker who 
does not remember a word feels free to substitute a Greek equivalent. 
The grammatical reductions, substitutions, and adaptations are much 
more complex. Arvanitika and Greek are interwoven in fascinating ways. 
Trudgill 's concern in the article, however, is with the reduction and 
simplification in Albanian under the conditions of restricted use, and 
he makes no attempt to treat Arvanitika and Greek within a single gram- 
matical framework. The closest he comes to this point of view is the 
comment ' . . . the reduction to which Arvanitika is currently subjected 
is simply keeping pace with the extent to which Arvanites are now becoming 
native speakers of Greek' (Trudgill 1976-7:48). 

Another example is Beatrice Lavandera and her study of the kind of 
language spoken by Italian immigrants in Argentina (Lavandera 1978) . 
These Italian immigrants are bilingual. Their children typically become 
monolingual Spanish-speaking, but a large population of bilingual Italian 
immigrants persists because of continued immigration. They speak a kind 
of Italian among themselves, in fact several quite different kinds of 
Italian because they come from several parts of Italy, and they speak a 
kind of Spanish to the Spanish-speakers of Argentina, a 'reduced' form 
of Spanish, that is called Cocoliche. This variety of Spanish is very 
close to the surrounding Argentinian Spanish but Lavandera has shown that 

it differs systematically in certain ways. In particular, the full range 
of stylistic variation which the native speakers of Spanish have, is 
lacking in Cocoliche. For example, whether you pronounce the final s as 
the sibilant, as aspiration, or as nothing at all, is a stylistic device 
which can be used in many ways in Argentinian Spanish. But Cocoliche- 
speakers either have the full s or they don't, as an all or none proposition. 
In fact native speakers of Buenos Aires Spanish use this s variation both 
word-finally and preconsonantally within the word, but Cocoliche-speakers 
typically drop the final s completely and use the full sibilant pronun- 
ciation preconsonantally. Similarly, there are different ways of using 
the subjunctive in indirect discourse which the Spanish-speakers have 
available and which the Cocoliche-speakers do not use. 

In this instance the lack in Cocoliche is not interference from 
Italian, since the speakers use the corresponding subjunctives in their 
Italian. Lavandera recorded artificially elicited material in the 
speakers' 'purest' Italian and 'purest' Spanish, i.e., Cocoliche, ('Tell 
me a story in dialect.' 'Now tell me the same story in Spanish.') She 
also recorded spontaneous conversations in the home where the language 
was sometimes what the speakers later judged to be 'dialect', 'Spanish', 
or a mixture of the two. Her conclusion comes close to the position 
I am advocating here, although she does not try to write a unified gram- 
mar: 'The data indicates that for these speakers Spanish and Italian 
are not independent codes, at least not more so than the different 
registers of a monolingual speaker' (Lavandera 1978:403). 

One final example, the one I find the most interesting, is a quad- 
rilingual Indian community in Canada. Ronald Scollon recently studied 
a speech community in Port Chippewyan, Alberta (several articles and a 
monograph in press; cf. Scollon for the forthcoming) where the people 
have for at least a hundred years, been speaking four different languages: 
Chippewyan, Cree, French, and English — an Athabaskan language, an Algon- 
quian language, a Romance language, and a Germanic language. And over 
the times, these languages have come to play particular roles in that 
community in a very complex pattern. Everybody speaks all four languages, 
but the phonologies and the syntaxes have become remarkable alike. For 
example, the phonologies tend to be like Cree and the syntaxes more like 
Chippewyan. But there are nice examples of patterns that cut across all 
four languages. Particularly interesting is a phonological variable 
described by Scollon. Beginning with some variation previously existing 
in Cree, there is now quite extensive variation between 's' sounds and 
'/' sounds, including affricates of the ts, tj type. It works this way: 
in general, the speakers tend to use the 's' kind of sound when they 
are unconsciously expressing a modernizing tendency, that is when the 
situation is recognized as a modernizing one or a modernizing topic is 
being discussed or a modernizing addressee is involved. And the 'J' 
kinds of sounds are associated with traditional addressees and so on. 
The sound variation which began in Cree has now spread throughout the 
system, and so we find such strange anomalies to the observer, as this 
example which Scollon gives: one particular woman in this speech com- 
munity was heard saying [suz] for English shoes when teaching the 
kindergarten, a modernizing locale, and she was heard saying [Ja] for 
[sa] the word for 'beaver' in Chippewyan, when she was at home talking 

with a group of elders in the community, a traditional setting. So in 
both cases, the pronunciation was 'wrong' in terms of what the original 
forms should be, but it was right in terms of the total grammar of the 
speech community. In this wonderful set of studies by Scollon, does 
he talk about a grammar of this whole system? Not at all. His first 
concern in one of the papers is performance aspects of traditional 
narratives in the society. He is trying to understand the way tradi- 
tional narratives are recited, their functions in communication, how 
they are framed in the discourse, and so on, and he provides a very 
fine exposition of traditional narrative performance. In order to 
talk about that, he has a lot about linguistic convergence. The closest 
he comes to saying that it all could constitute a single grammar is this: 
'The four languages at Port Chippewyan . . . have converged to a con- 
siderable extent. . . . The systematicity of discrete historical varieties 
is subordinated to the pragmatic needs and personal experiences of the 
individual' (Scollon forthcoming). 

Incidentally, I would like to point out that all four of the 
studies have a diachronic flavor. Whether the authors intended it or 
not they all talked about what those languages were like before the 
time of description and what seems to have happened and to be happening, 
and they all make some cautious predictions about what might happen in 
the future. Also, it so happens that these fine linguists in making 
these descriptions, at one point or another in each one of the articles, 
have acknowledged reluctantly, hesitantly, and in a kind of aside, that 
really the linguistic repertoire is one system. What I am saying is, 
let's face that openly. Let's try the job of writing grammars of 
such complex systems. I know it is hard enough to write grammars, 
anyway, and what I am suggesting makes it even harder. But I still 
think this is the job of the linguist. As long as we keep kidding our- 
selves that there are no situations like this, in some way, that they 
are not somehow linguistically significant, then I think we are missing 
an important part of linguistics. We are missing an important part of 
how languages change through time and how language variation functions 
synchronically, essential features about the nature of human language. 
And I have a final quotation, this one from a study of mine. Back in 
1957, I wrote an article called 'Two problems in Arabic Phonology,' a 
somewhat polemic technical description of two questions in Syrian Arabic 
phonology. One of the questions was the ~ lo versus aylcoS) that I mentioned 
earlier. At the end of the article, I apparently lapsed into a kind of 
dreamy mood, because this is the sentence I wrote: 'I like to dream of 
a time when the analysis of a uniform, static, autonomous synchronic 
system will be regarded as merely a special limiting case in a far more 
general linguistics which is prepared to deal with data highly hetero- 
genous dialectally and diachronically ' (Ferguson 1957:478). 

I still agree with the substance of that sentence but the addition 
I am making tonight is that if the variation turns out to include varieties 
so different that we would want to call them different languages, we might 
still have to put it in the grammar. In other words, I am saying that 
multilingualism may be a legitimate object of linguistic description. 


Denison, N. 1968. Sauris: A trilingual community in diatypic perspec- 
tive. Man 3. 578-92. 

Ferguson, C. A. 1957. Two problems in Arabic phonology. Word 13. 460-78. 
1978. Historical background of universals research. Universals 
of human language, ed. by J. H. Greenberg et al. Vol. 1, 9-31. Stan- 
ford: Stanford University Press. 

and C. E. Debose. 1977. Simplified registers, broken language, 

and pidgin and Creole linguistics, ed. by A. Valdman, 99-125. 
Bloomington: Indiana University Press. 

and J. J. Gumperz. 1960. Introduction. Linguistic diversity in 

South Asia, ed. by C. A. Ferguson and J. J. Gumperz (UAL Vol. 26, 
No. 3, Part IV), 1-18. 

Kellogg, S. H. 1965. A grammar of the Hindi language. 3rd ed. London: 
Routledge and Kegan Paul. 

Lavandera, B. R. 1978. The variable component in bilingual performance. 
International dimensions of bilingual education, ed. by J. E. Alatis, 
391-409. (Georgetown University Round Table on Language and Lin- 
guistics 1978.) 

Scollon, R. Forthcoming. Variable data and linguistic convergence. To 
appear in Language in Society 6. 

Smyth, H. W, 1956. Greek grammar. Rev. ed. Cambridge, Mass. : Harvard 
University Press. 

Trudgill, P. 1976-7. Creolization in reverse: Reduction and simplifica- 
tion in the Albanian dialects of Greece. Transactions of the Philogical 
Society 1976-7. 32-50. 


Roger W. Shuy 

Georgetown University & 

The Center for Applied Linguistics 

In his introduction to one of the most recent collections of articles 
on the amazingly diverse and emotional topic, bilingualism in America, 
Hernan LaFontaine notes that educational policy development is difficult 
to articulate and even more difficult to change. In the case of the recent 
flurry of interest in multilingualism in this country, LaFontaine (1978 :xi) 
observes: 'In addition to the normal inertia resisting any educational 
change, we have had to face political, social, and economic issues raised 
as barriers to the effective implementation of bilingual education pro- 
grams' . Such issues are not new to education for this constitues a field 
which has been characterized by decades of searching mindlessly after every 
new method or gimmick which comes along. Now that language has surfaced 
as the focus of the developing policy, the inertia which so characterizes 
education is of immediate concern to linguists.! 

Recent events in educational legislation are instructive about the 
process of change in American schooling. In January of 1974, after four 
years of litigation and appeals, the U.S. Supreme Court ruled in favor of 
a San Francisco Chinese family named Lau who claimed that the local school 
system had violated their constitutional right to access to education by 
providing that education in a language which was foreign to the learner, 
English. By ruling in favor of Lau, the Supreme Court said, in effect, 
that it is the right of every American citizen to expect the strong and the 
educated to deliver learning in such a way as to address the beginning 
points of the weak and the uneducated. Education has claimed to be doing 
this for years, usually referring to it as 'starting with the child where 
he is'. Although this principle is widely proclaimed, in reality it is 
clearly absent, for major American educational policy continues to follow 
a clearcut compensatory education model which places little value on diver- 
sity and great emphasis on making every child as much like main-stream 
children as possible as soon as it can be done. Lau vs. Nichols suddenly 
challenged all this and, with one stroke of legislation, brought legisla- 
tive power to what had been almost empty verbiage in teaching and learning. 

This bilingual education legislation, like much of educational legis- 
lation was actually an expression of a moral value derived from a great 
deal of intuition and from very little empirical research or theoretical 
clarity. Social scientists had not provided Congress with a clear date 
base for bilingual education any more than they had in the past offered 
ample evidence that busing would lead to better learning for the educa- 
tionally disadvantaged. In fact, when confronted with the assertion that 
bilingual legislation preceded the knowledge upon which it could be based, 
the lawyers for both Lau and Aspira2 readily admitted that such was the 


case, noting further that without such legislation, the knowledge base might 
never be started.^ This does not mean that no knowledge base existed prior 
to legislation; only that this knowledge base was far from adequate. Nor 
is this to cast criticism on the legislation, for it was undoubtedly well 
motivated and much needed. It is called to attention here rather as a hum- 
bling reminder that educational theory and research, in many instances, 
tends to follow the legislation rather than precede it. 

In fact, the developmental model might be said to look something like 
the following: 


of Concept 





Theory & 


This model of development in educational programming is not at all 
unfamiliar in our schools. In a recent evaluation of Title I programs in 
America, no effort was made to assess the theoretical difference between 
compensatory and noncompensatory theory.^ Children in Title I programs 
were not compared with non-Title I children nor was there any conscious 
effort to address that most critical aspect of education: the legitimacy 
of what the learner already knows. Instead, the evaluation assumed the 
legitimacy of compensatory theory and began looking for delivery system 
differences which mattered. The value was assumed to be correct. The 
materials and curriculum were developed and by the time the evaluation took 
place, nobody could remember that theory nor basic research had been con- 
sidered. The press of the political, social and economic concerns was too 
great to be delayed by our not having thought about, much less, researched 
the issue. 

The fact that such a procedure is not unusual in education is no excuse 
for linguists throwing up their hands in despair. Educational policy is 
developed no less chaotically than language policy, and we have had far too 
small an impact even on policy issues which bear directly on our own field. 

What linguists know to be true about how language works has not been 
communicated to education and has had little impact on educational policy. 
As a field, we have been busy with our own problems of analysis and theory, 
and we have, with good reason, been cautious about making claims which may 
later turn out to be wrong. Perhaps this caution is good in some ways. 
Linguists should certainly try to avoid the excess claims for our field 
such as those which were made in the fifties, when linguistics was thought 
to be the savior of reading and composition. Vestiges of this overpromise 
remain today in the form of 'the linguistic approach' to reading (which is, 
in actuality, a phonological approach to decoding). Having been stung by 
this and the Paul Roberts composition disaster (which not only promised that 
generative grammar would aid writing ability but also gave an early and 


somewhat inaccurate representation of generative grammar) , educators stopped 
seeking out linguists and most linguists didn't seem to care. We went off 
to our own issues and let education find other easy 'believisms' to adopt. 

It was not to the credit of the field of linguistics that multilinguism 
came into prominence with the passage of the Title VII Amendment to the Ele- 
mentary and Secondary Act. Embarassing as it may be, the turn toward the 
direction of favoring more than one language in American education (and thus 
in American life) did not grow out of suggestions from the field which calls 
such practice its own. Since linguistics also has had little to do with 
stimulating other national language movements in recent years such as Ver- 
nacular Black English and the resurgence of interest in Native American 
languages, one might speculate about why we haven't pushed our own field, 
our own interest, our own expertise area. At least seven factors, individ- 
ually or together, could cause linguistics to be slow in this matter. 

1. Energy. Perhaps linguists have too much to do. We teach, write, 
research, and appear to work very hard but, as fields go, we are few 
in number. Our problems to address are endless. Some will remember 
when an LSA meeting seemed to have participants who believed that we 
were only ten years or so away from writing a complete grammar of 
English. The more we learned, the farther away that point seemed to 
get. Today there is hardly a linguist alive who feels that we are 
anything short of light years away from such a goal. Perhaps we do 
not participate in developing multilingual policy because we lack 
energy, but this does not seem to be the total cause. 

2. Belief. A second possible reason why linguistics has been slow to 
influence educational policy is that we may not really believe in what 
we are doing. This is an extremely difficult suggestion to accept 
since there is probably no more fanatic discipline in all of academics 
than linguistics. Nevertheless, the question remains that if linguists 
practice multilingualism and if we believe in it, why is it that we 
have had so little influence on educational policy? Do we believe it 
in our heads but not in our hearts? 

3. Academic Neutrality. With the academic stance of neutrality we 
may be getting closer to the reason why linguistics is slow to influ- 
ence educational policy. This neutrality gets reflected in many ways. 
Some linguists are above getting involved with making their theory 
move toward practical reality. It smacks of the academic proletariat. 
It seems too much like getting one's hands dirty or working in the 
kitchen. There was a time, not too long ago, when applied linguistics 
was frowned upon because it cast off the stance of neutrality and 
entered the political, social, and economic world. The ineffective- 
ness of such a stance was brought home to me clearly a few years ago 
when a colleague and I had presented to an author cogent, academic 
reasons why the reading test which he had developed was in error and 
dangerous. His response to our linguistic argumentation was direct 
and simple: 'The test sells well and as long as it sells well, I'll 
keep it on the market.' Clearly our academic neutrality would do 
nothing useful for the thousands of children who were being misdiag- 
nosed by this test. The issue was an economic one as far as the author 


was concerned and the best way to address an economic issue is not with 
linguistic reasons but with an economic weapon such as systematic boy- 

Linguists who do not want to get involved, who just want to do 
their specialty and avoid advocacy have a legitimate right to do so, of 
course. And as a field, we can simply describe multilingualism, show 
how it works and develop theories about it, remaining neutral about it 
all. If we do this, however, we will not be doing much toward develop- 
ing educational policy related to multilingualism. We will be leaving 
it to other people, many of whom know little about multilingualism. 

4. Vision. A fourth reason why linguists have been slow to effect 
educational policy of multilingualism stems from a myopic perspective 
of our field. In contrast to the previously noted academic neutrality, 
this position is one of academic isolation which leads to a severe type 
of intellectual myopia. The field of bilingual education in America 
suffers badly from such myopia. Linguists tend to think of bilingual 
education as language. Social scientists think of it as culture. Law- 
yers consider it law. Educationists see it as materials, methods, 
curriculum, and teaching. All are right, of course, but it is clear 
that few see it from the perspective of the other. 

If the lawyers for the Puerto Rican Legal Defense and Education 
Fund had seen bilingual education from the perspective of education and 
linguistics, for example, they would never have argued for a consent 
decree which required the New York City schools to develop, administer, 
and score tests in English and Spanish for over 200,000 Puerto Rican 
children in writing, reading, listening, and speaking to determine in 
which language these children could 'most effectively participate in 
the classroom. ' Linguists could have told the lawyers that such a test 
does not exist and that it would take a great deal of research to dis- 
cover what it means to 'effectively participate in the classroom.' 
They could also affirm that it is rather difficult to test for speaking 
ability in a paper/pencil group test. Educators could have told the 
lawyers that New York teachers do not look with favor on losing a week 
of class for such testing and that the teachers union, being generally 
opposed to bilingual education in the first place, would not neces- 
sarily discourage lost or falsified data. 

It is difficult to imagine how this disciplinary isolation could 
be averted. Universities foster it and many efforts at building inter- 
disciplinary programs have yielded products which know a little about 
the various disciplines but are competent in none of them. It would be 
far better to be fully trained in a given field such as linguistics, 
then work on problems in education along with educators, not in opposi- 
tion to them. Team research, team teaching, and a general cooperative 
spirit seems to be the answer here. It is clear, however, that disci- 
plinary isolation has slowed linguistics in its effort to influence 
educational policy related to multilingualism. 

5. Snobbishness. Closely allied with the vision problem is that of 
snobbishness. Perhaps what we really mean is insecurity since it is 


often quite difficult to tell one from the other. Disciplinary isola- 
tion seems to go hand in hand with one-ups-manship of this sort, 
however, and it is not difficult to see evidences of such unhappy 
behavior among our own numbers. 

Sometimes this snobbishness is not even recognized. Clear evi- 
dences seem to be easier to locate in crimes against linguistics rather 
than crimes of linguists against other fields. It is no secret, how- 
ever, that among the social sciences there are different research 
paradigms and techniques. Many experimental psychologists, for exam- 
ple, seem to be unaware of the legitimacy of such nonexperimental 
approaches as participant observation or ethnographic analysis. 
Recently, in fact, the NIE sponsored a workshop which addressed the 
conflict between quantitative and qualitative approaches to research 
in education. This was a very important first step since quantitative 
research has dominated government research funding in recent years. A 
story is told, in fact, of an experimental psychologist who was intro- 
duced to a classicist. The psychologist asked the classicist what he 
was doing and the latter responded that he was spending the year in 
research. The psychologist looked surprised and replied, 'I didn't 
know that classicists did research.' 

Such snobbishness is not limited to the field of experimental 
psychology or to quantitative techniques. It is safe to say that lin- 
guistics is perceived as abstract, esoteric, jargon-ridden, and 
impractical by many school people. Linguists are often thought to be 
irritable, excited, unhelpful, and one-track minded. It is not dif- 
ficult to see how such an impression could develop if we do no more 
than tell educators that they need linguistics and that, to get it, 
they need to take our introduction to linguistics courses. IVhile 
their concerns are on reading, writing, speaking, and listening, ours 
are on phonology, grammar, semantics, and universals. It is easy to 
be thought of as snobs under such circumstances. 

6. Knowledge. Perhaps linguistics has not affected educational 
policy in the area of multilingualism simply because we do not have 
the knowledge base with which to do it. We know some things about 
first language acquisition (at least up to age ten or so) and a few 
things about code switching and second language learning. But we are 
practically helpless in answering questions of the schools and the 
courts which concern fluency (what does fluency mean anyhow?), the 
effect of cognitive growth and language switching, the exact dimen- 
sions of literacy and the language of literacy. Even more embarassing, 
we do not have answers to the educators' questions on the most basic 
issues such as what really matters most in learning a second language. 
Does a nonnative really have to learn to speak a new language like a 
native? Won't the nonnative lose something important if he speaks the 
target language the way native speakers do? These and many other 
potential research questions are unanswered. The courts believe that 
someone has the answers to these questions. Linguists clearly do not. 
When asked, we admit it. Then the schools have to act as though the 
answer were clear. Since linguists haven't given the answer, the 
schools muddle along as best they can. They make policy without us 
because someone has to do it. 

7. Internal Controversy. However much linguists might consider 
internal controversy to be a healthy thing which enables us to grow 
and expand, we are often perceived by educators and others to be con- 
stantly argumentative and undecided. As we all know, linguistics is 
a field in which people engaged in bitter public disagreement can turn 
around and, unexplainably, go to lunch together as friends. This is 
not always the case, of course, but it is clear that others do not 
always perceive us in the same way that we may perceive ourselves. In 
addition, the fact that our field seems to have theoretical revolutions 
with great regularity leaves other fields in a quandry about how to 
relate to us. 

Of these seven factors which one might suggest to explain why lin- 
guistics has been slow in affecting educational policy with respect to 
multilingualism, five stand out. It is easy to reject the suggestions that 
we lack energy or belief. We can probably be faulted when we attempt to be 
academically neutral. We have undoubtedly been myopic about multilingual- 
ism, seeing it only as a linguistic phenomenon. Accurately or not, we have 
been perceived as snobbish and prone to excessive internal controversy. It 
is the factor of our inadequate knowledge base which should concern us the 
most, however. Multilingualism is clearly within our territory. Linguis- 
tics may not own the field but it can hardly be said to be marginal to it. 
That we have so few answers to offer the courts or the schools as proof of 
what works or even what is^, is no credit to our discipline. 

In general, linguistics has not taken an active role in trying to ini- 
tiate language policy in this country. We have not anticipated what is 
coming next. We have not analyzed our market. We have not tried to fore- 
cast. Indeed, linguistic involvement in language issues has tended to be 
responsive rather than interactive or formulative. We have let other forces 
of society set the stage and make mistakes. Then we come on the scene and 
complain bitterly about how stupid the schools are to have set their current 

Our involvement with educational issues relating to Black English Ver- 
nacular in the sixties is a case in point. School problems of Black 
children were recognized first in the schools by people like Ruth Golden 
in Detroit. It is not possible at this time to know whether or not Ruth 
Golden asked linguists for help when she attempted to develop standard 
English teaching procedures in that city. In a way it is immaterial for, 
if she had asked, she would have found little of use to her. So she went 
ahead on her own. She developed the Golden Tapes, a set of audio tapes 
which addressed certain Black English pronunciations. Mrs. Golden found 
it particularly troublesome that students at Central High School pronounced 
it Cintral High School. So she developed tape drills to correct this. The 
point here is not that Ruth Golden selected a poor feature to focus on. It 
is, rather, that there was no linguistic help for her to rely on. She had 
to go it alone. Five years later, by 1968, linguists could tell her that 
the /I E/ collapse before nasals does not really matter. Other features 
count considerably more heavily. Why couldn't linguists have anticipated 
the Black English development and geared up for it in advance? We had to 
be responsive rather than interactive or formulative. What is worse, we 
were not even anticipatory. 


Other fields have developed an anticipatory capacity with some 
degree of reliability. We are all familiar with difficulties involved 
in weather-forecasting and economic predictions but even something as 
prosaic (is this snobbishness again?) as sales forecasting may be able 
to offer insights. Business is not merely responsive; it interacts to 
change societal conditions to its own liking through advertising, lobby- 
ing, bribing, three-martini lunches, and many other ways. Lest there 
be any confusion on this matter, let me state clearly that 1 am not 
suggesting that linguists begin advertising or bribing, or having three 
martini lunches, but I think it might be useful for us to consider two 
of the common processes of sales forecasting. 

One process simply looks at the past and notes recurring cycles 
and locates the current situation somewhere in that cycle. This is a 
bit like projecting language change based on a developmental line from 
as far back in time as we can go to the present then extended (perhaps 
a dotted line) into the future, based on past development. 

The second process of sales forecasting is the careful analysis of 
market conditions. Sales forecasters predict future conditions by 
matching controllable and noncontrollable factors. The controllable 
factors are things like what we know, how much equipment we have, our 
financial condition, our personnel, and our reputation and image. The 
noncontrollables are the social forces at work, demographic factors, 
ethical forces, political forces, cultural environment, international 
issues, and the development of new technology. For example, if tech- 
nology should suddenly create a way to make congenitally deaf people 
hear normally, the political issues involved in the signing (ASL) vs. 
oralists camps of deaf education would suddenly seem less critical. 
Without such technology, however, linguists have been stoking up the 
controllable resources to try to make oralists understand that signing 
is the native language of deaf people. ^ 

The point here is that linguists will continue to find it difficult 
to affect educational policy involving multilingualism as long as the 
field is merely responsive to developments created outside of linguis- 
tics. In order to be anything other than responsive, linguistics must 
develop ways of anticipating societal needs in advance and gearing up 
for them before they happen. The rest of this paper will provide five 
suggestions about what linguistics must do to become more active in 
developing educational policy involving multilingualism. 

1. Beoome accepted by education groups. Work with educators (in 
team, if possible) . If linguistics is to be recognized as an ally 
of education, we will need linguists who can communicate with 
educators. We should be writing in their journals, speaking at 
their conventions, and serving on their panels. If one wants to 
do something useful about doctor-patient communication, one says 
it where it will be heard by the right people. One writes for 
a medical journal or speaks at a medical convention. If we 
reject the idea of speaking at an education convention on the 
grounds that such conventions are of low calibre, we are either 
guilty of the snobbishness discussed earlier or we have picked 

the wrong convention. Two very important considerations must be 
kept in mind when one engages in this sort of activity. First, 
there is a pecking order in education just as there is in lin- 
guistics. Failure to cite the perceived acceptable literature 
can limit outsiders' positive reactions considerably. Secondly, 
educators will be interested in their own problems primarily and 
it is up to us to show them how linguistics can contribute to 
their problems. Administrators face the problem of how to assign 
pupils to bilingual programs or regular programs. They frequently 
rely on tests, such as language dominance tests, to make such deci- 
sions. Linguists can argue effectively that language dominance is 
a peculiar concept in search of a definition but this will not make 
the administrator's problem go away. Eventually we will need to 
do more than debunk current practice. We will need to offer a 
viable alternative to the administrator's problem. Writing in 
education journals, appearing at education conventions, and serv- 
ing on education boards provide the evidence of our concern. 
Such evidence will enable us to be accepted by the education 
groups through which policy on multilingual ism will likely be made. 

2. Get involved with the power structure. Earlier I commented 
that lobbying might not be appropriate for all people. Neverthe- 
less the business of influencing educational policy requires a 
broker of some sort. It is possible to get oneself appointed to 
textbook selection committees, to write to the NIE Director, to 
read proposals, or to sit on evaluation panels. Those of us lin- 
guists who read proposals in areas related to language in education, 
speech, or humanities realize full well that our job is not only to 
evaluate a proposal but also to represent linguistics to our fel- 
low panelists (who are most likely to be psychologists) and to 
actually teach them about research designs, strategies, and know- 
ledge from our field. 

3. When possible, get involved with advocacy. To make multilin- 
gualism become education policy it would appear that linguists 
will need to become considerably more activist than we ever were 
before. A linguist who is accepted by education and who is 
involved in the power structure stands a better chance of being 
asked to serve as an expert witness in a law suit involving 
bilingual education than one who is not so involved. The decision 
to be an advocate in such circumstances opens one to considerable 
complication, however, as the following incident will indicate. 

In December of 1976, the New York City School System had just 
entered into a consent decree with a New York Puerto Rican rights 
group, Aspira of New York. As noted earlier, the consent decree 
specified that the school system test to determine in which lan- 
guage, Spanish or English, the over 200,000 New York Puerto Rican 
children could most effectively participate in the classroom. The 
test was to be developed, administered, and scored by the following 
June. It was to measure ability in both languages in reading, 
writing, listening, and speaking. The department of research in 
the schools called on the Center For Applied Linguistics to develop 


the test. One quick look at these impossible requirements prompted 
CAL to suggest a different strategy. Rather than to do a job which 
had to be based on an as yet unidentified and unresearched know- 
ledge base, CAL suggested that New York throw itself at the mercy 
of the federal judge and tell the truth: that nobody really knows 
how to do this as yet and that it would take a year or two of 
solid research and development to be even near ready to carry 
out the task. 

CAL was not effective in convincing the director of research, 
however, and New York went at it on their own. Approximately 
two weeks later, the test was completed and the director of 
research happily claimed that he could do in two weeks what CAL 
projected would take a year or so. 

At this point, the Puerto Rican Legal Defense and Education 
Fund, legal council for Aspira, called CAL to request help in 
determining how good this new test, the Language Aptitude Bat- 
tery, actually was. CAL took great delight in this request for 
it gave linguistics a golden opportunity to address the issue of 
bilingual placement. Part of the work was to explain to the 
judge, in court, the weaknesses of a test which claimed that dis- 
crete point items such as the /s/ vs. /c/ contrast in words like 
shoes and choose are the critical measurements of language abil- 
ity. Another task was to explain to the judge that a test written 
in English would not be equally valid when translated into Spanish. 
In addition to these validity issues, CAL pointed out to the court 
many of the ambiguities in the wording of both the questions and 
the instructions. 

A second incident involving CAL advocacy work took place in 
Chicago where, in 1976, the Office of Civil Rights declared that 
the city schools were out of compliance with the Equal Rights 
Amendment, largely in the area of minority language speakers. 
The procedure for identifying bilingual children was again the 
case. Chicago represented itself at the crucial hearing as iden- 
tifying children in need of bilingual education by administering 
a fifteen minute video tape to Chicago teachers. They observed 
three minutes of five different children of varying English lan- 
guage ability. Each child was said to represent one grade on a 
five point scale of English competence. The teachers were to see 
this performance, then use it as a model and guide for classifying 
all of their limited English speakers. Again it was not very dif- 
ficult to write an affidavit describing what was wrong with this 
procedure. The complex part, however, was that the director of 
bilingual education in that school system is an extremely competent 
person who sincerely wanted to improve the situation. To make 
matters worse, the attorneys for the Chicago schools did not put 
Chicago's best foot forward. Chicago was doing better things than 
this court performance showed. The business in advocacy roles often 
puts the linguist into a very touchy situation. Should we continue 
to help OCR attack what Chicago has been doing even though we know 
that they are in competent hands and making steady progress? We 


gambled to go ahead on the assumption that by exposing the school's 
presentation we might ultimately strengthen the program and the 
director's hand at the same time. As it turns out, this is exactly 
how it worked. OCR won the case and the Chicago schools were 
threatened with having all federal funding withheld. There was a 
quick change in strategy shortly after, however, and the program 
became stronger than ever. 

Once one enters into policy development such as this, however, 
one runs certain risks. I was personally frightened that my good 
relationship with the director of bilingual education might dete- 
riorate because of the court case. I gambled that it would not and, 
fortunately, I was right. 

Getting involved with advocacy in this way is something which 
few academics ever think they will do when they first get seduced 
by their field. My own personal dreams of living a tweedy, halls 
of ivy existence vanished like vapor as soon as I realized that if 
I really believed in what I was teaching, I might also be expected 
to do something about it. This doing can be interpreted as talking 
or teaching, at least for a while — or maybe even writing papers 
ang giving speeches. But eventually these things come to seem 
oddly preliminary and removed from the real arena. At such times 
one does not stop being a linguist and start being a salesperson. 
Instead one continues to be a linguist and tries to determine ways 
to influence policy. If the linguist does not do this, others who 
know our field less well most certainly will. And this is exactly 
what has happened far too often in the past. The friends of lin- 
guistics, those who had a little knowledge, are the ones who did 
us in, who promised the moon and who saw linguistics as the answer 
to education's problems. Obviously that tack has not worked. Edu- 
cational policy is no better for it and it has certainly not been 
enlightened by it. Perhaps it is time to try it from the obverse 
side of the coin. If educators do not understand language well 
enough to lead in the areas of educational policy related to lan- 
guage, perhaps linguists can develop strategies toward understanding 
education well enough to do it. To do this there is probably no 
better subject matter than multilingualism. And there is no 
quicker way to find oneself in the middle of the policy issue than 
to become involved with an honest-to-goodness law suit involving 

4. Research needed problems. A fourth way for linguists to become 
involved in the educational policy of multilingualism is to direct 
our research toward the needed problems . At this time we still do 
not know what matters most and what matters least in learning a 
second language. We need a much broader and deeper research base 
for bilingual education than is now available. To date, research 
in bilingual education has been considered more an education issue 
than a linguistic one. The tendency of the research has been to 
compare one methodology with another or one curriculum with another. 
However, we have now reached a period in the development of the 
field when people are demanding evidence of success or failure of 


the AIR evaluation of Title VII programs is that it did not under- 
stand the basic principles of language variability .6 It is difficult 
to put faith in any assessment of the effectiveness of bilingual 
education which does not distinguish between acceptable and unaccept- 
able variation and which does not address language variability as it 
manifests itself in the dimensions of social, regional, age, sex, 
race, education, occupation, intention, genre, topic, etc. It is 
also growing increasingly difficult to accept the discrete-point 
criteria used over and over again as evidence for language ability 
in either language. 

Recent research in linguistics has focused more and more on 
the importance of core syntactic, semantic, and pragmatic pro- 
cesses and less and less the isolable phonological, morphological, 
and lexical items. It has focused on the holistic more than on 
the segmental skills. Bilingual education programs need to focus 
more on this holistic view and more the sort of variability involved 
in asking questions, making requests, seeking clarification, taking 
a conversational turn, and reporting what others have said, for 
example, than on pronunciation accuracy or inflectional perfection. 
This direction is not merely a tag-on to existing programs; it is 
the starting point for language learning. 

Any effective research on language in bilingual education pro- 
grams will need to have, at its base, an adequate linguistic theory 
of language learning. Whether by accident or by design, there is, 
at the core of bilingual education, a satisfying theory of language 
learning. From research on first or native language learning we 
know that infants communicate ideas first, then later they perfect 
their phonology, lexicon, and grammar. Traditional language pro- 
grams cause second language learners to try to reverse this process, 
getting the phonology, lexicon and grammar first and hoping, later, 
for the development of language use. When subject matter is taught 
in a second language, the focus gets put right back on the idea of 
subject again, just as it is for infants. The resulting learner 
language is, in comparison to that of a mature speaker, rather 
sophisticated, just as it is for infants. But it is the process 
that is sophisticated and thousands of years of experience have 
shown it to be the superior method. Thus, bilingual education, 
for this very reason, has at its disposal an adequate language 
learning theory: a deep to surface structure sequence which has 
been proved successful by millions of infants as they learn their 
native tongues. Such a theory contrasts sharply with the surface 
to deep structure sequence found in most second language classrooms. 

With the advantage of this deep to surface theory, bilingual 
education research could easily capitalize on the advances made 
recently by other first language learning research. We know, for 
example, that pronominalization is learned early and easily by 
children whereas preposed adverbials are learned rather late. 
Nancy Yanofsky (1978) has recently shown, however, that reading 
instruction does not take advantage of this knowledge. A number 
of texts analyzed in her research show that preposed adverbs are 

frequent in beginning reading instruction while pronouns are intro- 
duced rather slowly. 

The areas for research into language variation and bilin- 
gualism are many. Careful descriptions of the production of such 
variability need to be made in order to determine how the speakers 
actually talk. Likewise, sociolinguistic attitude studies need to 
be made in order to determine community values toward the various 
language varieties used in school programs. We desperately need 
some clarity in the area of language assessment, particularly as 
it reflects a learner's sequential progress in a number of school 
and non-school settings. We need a dose of realism in the develop- 
ment of curricula in language teaching as well, with a focus on 
the contributions which linguistics can make in bringing about a 
radical restructuring of language learning classrooms by focusing 
on core syntactic, semantic and pragmatic units of language rather 
than the minute, phonological, morphological, and lexical aspects. 

5. Become well-placed. A fifth way for linguists to influence the 
educational policy of bilingualism is to get ourselves well placed. 
One of the few benefits of the depressing job market in linguistics 
today is that it is forcing us to think of working outside of aca- 
demia. And where is educational policy often forged? Outside of 
academia. It will do our field, and the subject of multilingualism 
a great deal of good if folks who know about language are placed in 
government agencies in which language issues are directly or indi- 
rectly addressed. Likewise it may do multilingualism good to have 
linguists working with speech therapists, helping to sort out speech 
problems from multilingual differences and as lawyers, litigating 
cases involving language issues. It may do multilingualism good 
to have linguists working in the advertising agencies in New York, 
monitoring language issues on television ads and it will most cer- 
tainly do the field of multilingualism good to have linguists 
working for publishers. All of these positions, however promising 
they may be for influencing educational policy, require a sacrifice. 
Marcia Whiteman has had to learn to be a bureaucrat in order to 
exercise her linguistic judgment on language issues at N.I.E. Andy 
Garfinckel has had to learn the advertising business as a trainee 
in order to work in television advertising. George Williams, Mary 
Levy, and others have actually gone to law school in order to be 
able to better apply their language knowledge toward policy issues. 
Faye Vaughn-Cooke , a linguist, works in speech therapy saving 
minority speakers from being designated as having severe patholo- 
gies. And as of September we will have, as far as I know, our 
first Ph.D. in linguistics enter Georgetown University's medical 
school where Johnnie Smelz will get the necessary medical creden- 
tial to serve her consuming interest in neurolinguistics . To 
become well-placed will require sacrifices of a sort which linguists 
have not had to make in the past. It may be easy for me to say, 
since I don't have to make such a sacrifice, but I think our field 
will be far better for it in the long run. 

By carrying out these five suggestions: becoming accepted by edu- 
cators, getting involved in the power structure, becoming advocates, 


researching the needed areas, and getting ourselves well placed, lin- 
guistics can improve its anticipatory capability. We need to do more 
than to forecast the next language-related thrust, for there is a mul- 
titude of language-related issues in education already before us. 
Almost all aspects of testing (multilingual or monolingual), language 
learning, reading, and writing are deeply involved with our field. 
The issue of multilingual ism, however, is perceived by the public as 
closer to linguistics than these other matters and for this reason, 
multilingualism is a prime candidate for a strong push from our field. 
Taking our cue from sales forecasting, we can most certainly address 
the controllable issues: what we know and need to know, our equipment, 
materials, personnel, and reputation. But we can also give considerably 
more attention to the uncontrollables : economic factors, political 
developments, demographic changes, ethical forces, and new technology. 
Obvious economic factors may suggest that we gear up for knowing Arabic. 
Demographic changes may suggest that we gear up for teaching Samoan in 
Los Angeles. As social changes go, we addressed issues of language and 
sex a bit faster than we got to language and race. Linguists got a late 
and rather disastrous start in the bilingualism push and we soon found 
ourselves thought to be marginal to the education departments of the 
country. Our inability to affect educational policy is a direct result 
of our failure to anticipate the movement before it happened and our 
lethargy in gearing up after it did. 

Worse than that, linguistics has not articulated some of its own 
roost basic premises in such a way that the education world can see and 
react to. It is patently obvious to anyone who has learned language, 
for example, that there is no way in God's earth to try to speak a new 
language without making mistakes in it. Progress can sometimes be 
noted, in fact, by the increasingly sophisticated kinds of mistakes which 
learners make. Years ago Labov referred to this concept as 'stages in 
the acquisition of Standard English.' All linguists know that children 
learn their native language in much the same fashion. Somehow this con- 
cept has not come through clearly to educators, who seem to feel that 
speakers should be assessed on some sort of right-wrong binary scale. 

Likewise, linguists have not well articulated their own basic pre- 
mise that variation in language is a healthy, necessary, human, and 
beautiful thing. 

In addition, in bilingual education programs, language variability 
will need to be displayed as a resource rather than as a handicap. Edu- 
cators, monolingual or bilingual, have been very slow to grasp this 
principle. Journalists seem to accept it, at least partially, when they 
argue that a good writer must write one way for one audience and a dif- 
ferent way for another. Poets have also seen the principle or they would 
never be able to vary, to twist or to torture language in a poetic fashion. 
The openness of poets to this principle is rather apparent in a poem 
about bilingualism written a few years ago by Ogden Nash: 

There are countries where being multilingual is of no 

advantage to the tourist. 
Because no matter how impeccable his grammar and idiom the 

natives refuse to understand him unless his accent and 

inflection are that of the purist 
They tell of an American diplomat who lost his suavity 

never to be restored 
When a Paris taxi driver denied the very existence of 

his destination because he pronounced Place de la 

Con -corde instead of Place de la Con- corde . 
(At this point I must for the justly maligned Manhattan 

hackers interpolate a kindly word; 
None of them to my knowledge has ever rebuffed a Parisian 

asking to be taken to the corner of Sird Avenue and 

Sirty-sird.) (Nash 1971 ) 

Whether or not Nash is right about the contrastive attitudes of the 
Americans and French concerning language variability, he has captured 
something about what we need to know about bilingualism if we are to 
affect educational policy in any significant way. And there is a 
great deal more. 

One might add, however, that if linguists were on top of the 
situation, issues of reading, language arts, testing, and equity would 
also be our concerns. 

The Aspira Consent Decree is a similar piece of legislation 

relating to New York City Puerto Rican children. 

These assertions were made at the TESOL meeting in Los Angeles in 
March, 1975. 

As a matter of precedent for policy, Georgetown University 
has recently accepted English as satisfying the foreign language 
requirement of an American congenitally deaf student. Likewise 
Georgetown has accepted Sign Language as meeting the foreign lan- 
guage requirement of a native English speaker. 

however, that there are many other reasons 
to criticize this evaluation, some of which have to do with design 
and instrumentation, regardless of language unsophistication. 


LaFontaine, Hernan (ed.). 1978. Preface. Bilingual Education, p. xi. 
Wayne, N.J.:Avery Publishing. 


Nash, Ogden. 1971. Speak to me only with thine eyes, and please, not 

too fast. Look. Oct. 5. 
Yanofsky, Nancy. 1978. Pronominalization and adverb preposing in 

reading texts. Unpublished ms . Georgetown University. 


Morris Halle 
Massachusetts Institute of Technology 

In his highly instructive paper 'Functional Explanations in Genera- 
tive Phonology' Michael Kenstowicz (to appear) distinguishes two kinds of 
explanation in phonology, 'internal' and 'external'. According to 
Kenstowicz, 'an "internal" explanation attributes the presence of some 
(part of a) rule or other aspect of phonological structure to another 
independently motivated portion of the grammar, while an "external" expla- 
nation seeks motivation in terms of something that exists outside of and 
independently from the grammar.' Appeals to simplicity, to feature count- 
ing, or to the fact that certain rules are cyclically ordered would be 
examples of 'internal' explanations for certain phenomena, whereas appeals 
to ease of articulation or to avoidance of ambiguity however defined might 
be viewed as examples of 'external' explanations. It is the latter kind 
of explanations that traditionally have been labelled 'functional'. In 
what follows I examine a number of such 'external' explanations and attempt 
to show that in each case an alternative 'internal' explanation is not 
only available, but also preferable. IJhile I do not wish to be understood 
as claiming that these examples prove that the search for 'external' 
explanations should be abandoned forthwith, the discussion below does cast 
doubt on the claims that have been advanced by proponents of 'functional' 

It is an obvious fact that many languages have rules that neutralize 
contrasts between distinct lexical items or morphemes. Every beginning 
student has no doubt been told several times that Russian, German, and a 
host of other languages have a rule that devoices word-final obstruents. 
As a result, hearing the utterance [zenskij rot], a Russian does not know 
whether the speaker is referring to a female's mouth (/rot/) or to femi- 
nine gender (/rod/) . This ambiguity might in principle present difficul- 
ties for communication, and languages might, therefore, be expected to 
avoid rules that result in the neutralization of distinctive contrasts. 
As a matter of fact, neutralization rules are attested in practically 
every language and the phonetic ambiguities which they produce do not 
appear to have great practical effects. It might, therefore, appear that 
the ambiguity resulting from neutralization of contrasts plays no role 
whatever in phonology. 

This conclusion, however, does not seem altogether justified in the 
light of interesting observations made by Jonathan Kaye (1975) that degree 
of ambiguity (or its complement 'ease of recoverability ') can explain 
certain otherwise quite puzzling phonological phenomena. I discuss here 
the simpler of the two examples analyzed by Kaye, that of Maori. In this 
language word final consonants were deleted in the alternations illus- 
trated in (1) : 












patua etc 

Kaye observes that 'since the consonants showing up in the passive forms 
are not predictable, the standard phonological analysis would posit under- 
lying consonants for the verb stems which are deleted when not followed 
by a suffix' (pp. 248-49). It has, however, been argued by Hale (1973) 
that this is not a correct analysis. Instead Hale proposes that in Maori 
all lexical morphemes end with a vowel and that the passive morpheme has 
eight allomorphs whose distribution is governed by an idiosyncratic 
property of the stem. 

Kaye doubts the explanation advanced by Hale that an 'underlying form 
in Maori with final consonant would violate a surface canonical constraint 
against final consonants . . .' because ' many languages, e.g., Karok, 
Shawnee, Arapaho, Atsina, have rules that insert a prothetic consonant 
(typically ? or h) in front of (underlying) vowel initial words. In such 
languages no word begins with a vowel on the surface, yet . . . there is 
no evidence of restructuring of the vowel initial words in any of these 
languages. A theory based on surface canonical forms cannot distinguish 
between the Maori case and the Karok case. Having vowel initial stems 
violates the surface constraints of Karok just as much as having consonant 
final stems does in Maori' (p. 249). What distinguishes the Karok case 
from that of Itoori according to Kaye is that the consonant deletion rule 
of Maori produces multiply ambiguous strings, whereas the glide insertion 
rule of Karok does not affect the ambiguity of lexical entries. Hence in 
Karok no restructuring is evident, whereas in Maori a very similar change 
resulted in restructuring. 

I am not fully persuaded by Kaye's account. It seems to me that the 
ambiguity of a word such as /awhi/ is unchanged by the reanalysis. Whereas 
on the old account the word /awhi/ is multiply ambiguous — at least, in 
principle — by virtue of the fact that it could arise by word final conso- 
nant deletion from /awhit/ as well as from /awhim/, /awhir/, awhik/, etc., 
/awhi/ is equally ambiguous on the new account because now the lexicon 
can include — at least, in principle — many items represented as /awhi/ which 
are distinguished one from another by the different passive suffixes that 
they require; /-mia/, /-ria/, /-kia/, etc. 

There is, however, an alternative account available. Hale has pointed 
out to me in a recent conversation that the passive and the gerundive are 
the only forms in Maori where consonant alternations of the type illus- 
trated in (1) are to be found. If we attempt to capture this fact with the 
help of a word final consonant deletion rule we are in effect implying 
that such alternations are the rule, rather than the narrowly circumscribed 
special cases that they in fact are. Moreover, this solution fails to 
capture yet another regularity of the language; namely it implies that 
morphemes can end either in vowels or in consonants; whereas if the passive 
and gerundive suffix are reanalyzed as having consonant initial allomorphs, 
all morphemes of the language end with a vowel. Thus, reanalysis does two 
things at once: it eliminates the word final consonant deletion rule from 
the phonological component and it restricts the canonical form of morphemes 


in the language. Arrayed against this is, of course, the additional cost 
of a complicated and ad hoc rule governing the distribution of the two 
suffixes. Although there does not exist at present a proposal of how to 
convert these considerations into a mechanical evaluation procedure, such 
an evaluation is surely not unimaginable and has implicitly been utilized 
quite widely. Consider, for example, the treatment of the verb to be in 
phonological descriptions of English. It is invariably treated as totally 
exceptional and attempts are not made to integrate it into the phonological 
and morphological rules of the language. IJhatever simplicity considera- 
tions underlie this decision — which to most linguists is so uncontroversial 
as not even to merit serious discussion — will also correctly handle the 
case of the Maori passive. 

Thus, it would appear that the Maori case cannot be taken as evidence 
that functional explanations are to be sought in preference to formal 
ones. While this claim was not explicitly advanced by Kaye it is specifi- 
cally made by Zimmer (1975) who draws attention to very interesting facts 
in Turkish phonology which according to him 'highlight a kind of problem 
that the old simplicity measure signally failed to address itself to, and 
that is not solved by most recent proposals' (p. 556). I shall attempt 
to show here that Zimmer is mistaken in this assertion, that the old 
simplicity measure does not only address itself to this problem, but in 
fact forces the solution that Zimmer regards as the correct one. The 
facts of interest are as follows. Turkish has a rule devoicing stops and 
affricates in syllable final position: 










dip- ler 








kanat- lar 




sac- lav 




aaa- lar 



In the Istanbul dialect certain stem-final /k/'s alternate with when 
followed by a vowel, while others do not: 

(2) b. 

i) ayak 

ayak- lar 







ii) ok 



' arrow 





Historically, these alternations between /k/ and are due to the fact that 
in the Istanbul dialect /g/ was deleted intervocalically . A possible 
account for the facts — and one adopted by Lees in his 1961 monograph — is 
basically identical with the historical development. The grammar includes 
a rule ordered after the syllable final devoicing rule which deletes 
intervocalic /g/: 

(2) c. 


/ [+syl] 



It will be noted that this analysis is abstract in the sense that it 
requires underlying representations in which a distinction is made between 
/k/'s that undergo rule (2c) — these are represented as /g/, — and /k/'s 
that do not undergo rule (2c) — these are represented as /k/. 

This analysis of Lees' is contested by Zimmer who argues that an 
observationally adequate description of the Istanbul dialect requires 
instead of (2c) the rule (2d) , which deletes intervocalic velar obstru- 
ents in the third and subsequent syllables of the word: 

(2) d. [-son, -cor, -lab] 

In particular, Zimmer cites such Western loan words as: 

(2) e. bek beki 'back' 

kartotek kartote-i 'cardfile' 

and refers to experiments with nonsense words all of which show that dele- 
tion depends on position in the word rather than on underlying phonological 
representation. Zimmer views this result as evidence in favor of func- 
tional over formal considerations, and he is understood in this way also 
by Kenstowicz, who writes: 'Turkish speakers have thus opted for a more 
concrete analysis of ^-deletion. The reason appears to be that from the 
point of view of the language learner, it is easier to distinguish the 
two different kinds of /c-final morphemes on the basis of an observable 
property (the number of syllables) rather than on the basis of a property 
(underlying /g/) whose existence follows from rather sophisticated reason- 
ing. ' 

I do not believe that this conclusion is justified. What Zimmer has 
shown is not that Actual Speakers, unlike Ideal Speakers or linguists, 
are unable or unwilling to engage in 'rather sophisticated reasoning'. 
Instead, Zimmer has shown that the solution proposed by Lees is wrong, 
simply because it fails to characterize the data, especially the data in 
(2e) . It hardly needs to be said that from an incorrect rule no theoreti- 
cal conclusions whatever can be drawn. 

In his paper Zimmer observes that the abstract analysis (one includ- 
ing rule (2c)) can be made to work by adding a morpheme structure condition 
that 'post-vocalic morpheme-final /k/ occurs only in monosyllabic mor- 
phemes, and post-vocalic morpheme-final /g/ only in polysyllabic ones' 
(p. 561). This condition, which must explicitly be added to the grammar, 
can be formally expressed as shown in (2f) : 


(2) f, 


[-voice] / // ([-syl]) [+syl] -1- 

[+voice] / [+syl] [-syl]Q[-l-syl] 

In other words, Lees' solution can be made observationally adequate if in 
addition to rule (2c) it contains also the morpheme structure condition 
(2f ) . It is this adequate solution which must be compared with Zimmer's 
alternative; i.e. a solution in which deletion takes place in accordance 
with rule (2d). The latter solution will also require a morpheme structure 


condition which, however, will state a simpler fact than (2f ) , namely 
that all morpheme final postvocalic velars are voiceless. We represent 
this condition formally as in (2g) : 

(2) g. [-son, -cor, -lab] ->■ [-voice] / [+syl] + 

We thus have two alternative descriptions that cover the same range of 
facts: one including the rule (2c) and the morpheme structure condition 
(2f); and the other composed of rule (2d) and morpheme structure condition 
(2g) . We observe that while rule (2c) is simpler than (2d) the difference 
amounts to just a single feature. On the other hand, morpheme structure 
condition (2f) is considerably more complex than its counterpart (2g) . 
Thus, on purely formal grounds the solution advanced by Zimmer must be 
chosen over the empirically adequate version of the solution a la Lees. 
In other words, the issue is not between a solution requiring no sophis- 
ticated thought and one that demands elaborate reasoning, but rather 
between two solutions that require the same amount of sophistication, 
where one is somewhat simpler than the other; simplicity being measured 
by the old feature-counting metric of SPE. I conclude that the Turkish 
example provides no evidence whatever for the proposition that purely 
formal considerations may be overridden by considerations of a functional 

It is important to be clear about the fact that simplicity considera- 
tions are relevant only in a restricted class of cases — i.e. in cases like 
the Turkish velar deletion just discussed where the same theory admits 
two or more alternative formulations that cover the exact same body of 
facts. Appeal to the simplicity metric is irrelevant where alternative 
descriptions either cover different sets of facts or where they employ 
different theoretical devices. Thus, appeals to simplicity — in the 
narrow technical sense of the term employed here — cannot decide, for 
example, between a description that admits reference to metrical trees 
of the kind exemplified in Liberman and Prince's (1977) recent article as 
against one based on a theory like that of Chomsky and Halle (1968) that 
has no place for such theoretical constructs. The fact that simplicity 
considerations are irrelevant in such cases does not mean that there are 
no rational ways of choosing between alternative theories but only that 
this choice has to be made on grounds other than simplicity. 

In several papers which deservedly have attracted wide attention Paul 
Kiparsky (1973) has suggested that certain limitations be imposed on the 
characterization of phonological facts. In the earliest of these papers 
'How Abstract is Phonology?' written in 1968, Kiparsky proposed that 
phonological theory should be so constrained as to rule out 'the diacritic 
use of phonological features and the phonologic use of diacritic features.' 
The latter use, according to Kiparsky, is exemplified by rules that 'have 
the form of phonological rules but operate on diacritic features' (p. 16), 
and I have nothing further to way here about rules of this kind. The 
diacritic use of phonological features, on the other hand, is illustrated 
by cases where two forms are given different underlying representations 
'for the sole purpose of classifying segments into those that do and those 
that do not meet the structural description of a rule' (p. 15). Such 
rules — of which the /g/-deletion rule (2c) in Lees' account of Turkish is 


a typical example — result in what Kiparsky terms 'absolute neutralization', 
and in his 1968 paper Kiparsky proposed to exclude rules of absolute 
neutralization by imposing the so-called alternation condition which 
requires 'that morphemes which are always phonetically identical must have 
the same underlying representations' (p. 18). 

Kiparsky 's condition is said to render grammars more easily learned; 
since the relationship between underlying and surface representations is 
more direct, the learning of the underlying representations requires much 
less of that 'rather sophisticated reasoning' which many linguists appear 
to be uneasy about attributing to the average man or woman on the street. 
I should like to remark that while Kiparsky 's condition renders grammars 
less abstract and hence, perhaps, more easily learned, it does not speak 
directly to the question of what role 'formal' vs. 'functional' considera- 
tions play in the functioning and development of languages. Kiparsky 's 
proposal is a proposal about the form of underlying representation of 
morphemes and is thus on a par with every other condition imposed on the 
form of grammars. What differentiates it from some of those — e.g. from 
the conventions on rule ordering — is that we have difficulty in thinking of 
a functional motivation for the latter, but can readily think of one for 
the alternation condition. 

As is well known, Kiparsky 's paper gave rise to an extended discus- 
sion, and as a result of this discussion as well as of additional independ- 
ent work, Kiparsky has modified his views to a considerable extent. The 
modified position is outlined in the paper 'Abstractness, Opacity and 
Global Rules' (1973). Its most important feature is that whereas previ- 
ously 'the Alternation Condition (had) been formulated ... as a limitation 
on underlying forms,' the new proposal reformulates it 'as a limitation on 
the application of phonological rules' (p. 65). Specifically, Kiparsky 
proposes the requirement (3) : 

(3) Nonautomatic neutralization processes apply only to derived forms. 

One of the most important facts that led Kiparsky to revise the Alter- 
nation Condition in the above manner was his discovery that there exists 
a class of examples where absolute neutralization is blocked in certain 
contexts and admitted in others. In particular, rules of this class apply 
only in 'derived' contexts; i.e. either across morpheme boundary, or 
morpheme-internally, if a crucial part of the context triggering the rule 
is itself derived by an earlier rule. They do not apply morpheme- 
internally where the triggering context is present already in the under- 
lying representation. I regard the discovery of the existence of rules of 
this kind as one of the most significant empirical finds in modern phonol- 

Kiparsky illustrates how such rules operate with the help of the 
following two rules of Finnish phonology: 

(4) a. e ^ i I __ # 

b. t ^ s I __ i 

c. i) /halut+i/ ^ /halus+i/ 'wanted' 


ii) /vete/ ^ /veti/ -»■ /vesi/ 'water' 
iii) /tila/ 'place' 

/itikka/ 'mosquito' 
/neiti/ 'Miss' 

It should be noted here that in Kiparsky's formulation restriction 
(3) does not apply to all phonological rules but only to phonological rules 
which are neutralizing and which, moreover, are nonautomatic, i.e. which 
admit specific exceptions. In their recent book on phonological theory, 
Kenstowicz and Kisseberth (1977) raise cogent questions about this restric- 
tion: 'Why should automatic neutralization rules apply in both derived 
and nonderived contexts, but nonautomatic neutralization rules just in 
derived contexts? . . . Suppose that there are one or two items that are 
exceptions to [a] rule of neutralization . . . Why should the existence of 
such random exceptions have the consequence that "an alternation . . . 
cannot be accounted for" by appealing to a rule of absolute neutralization' 
(p. 212). As a matter of fact, Kenstowicz and Kisseberth adduce two 
examples where a nonautomatic neutralization rule must be permitted to 
apply in nonderived contexts. I briefly sketch one of these examples. In 
the Yawelmani dialect of Yokuts vowels are shortened in closed syllables. 
This shortening rule is a neutralizing process since it results in the 
merger of long vowels into their short counterparts, which otherwise are 
distinct phonemes in the language. The shortening rule is also nonauto- 
matic: it does not apply before the causative-repetitive suffix /Isa/. 
In view of this, restriction (3) should block the application of the 
shortening rule in nonderived contexts. This, however, is not the case. 
Kenstowicz and Kisseberth cite the form [?aml + al] 'might help' which must 
derive from an underlying [?a:ml + al]. Thus shortening, which is a non- 
automatic neutralizing rule, must be allowed to apply in a nonderived 
context, in violation of (3).l Moreover, in his dissertation, Mascaro 
(1976) develops 22 rules for Catalan phonology: he shows that 14 of these 
22 rules apply only in derived contexts (he states that they are cyclic 
rules, but as we shall argue below, cyclic rules are restricted to apply- 
ing in derived contexts only). Each of these 14 rules is neutralizing, 
and several of these are automatic (see pp. 17-18). Thus, (3) cannot be 
understood as a biconditional for there exist automatic processes that 
apply in derived contexts exclusively. 

In the light of the preceding, one is strongly tempted to accept 
Kenstowicz and Kisseberth 's pessimistic conclusion that 'it is not immedi- 
ately clear that there is a way to predict which rules will apply only in 
derived contexts and which rules will apply in nonderived contexts as 
well. If no general principle is at work, the structural descriptions of 
particular rules would have to be global in nature' (p. 214). Fortunately, 
however, there is an interesting alternative to this. It has been argued 
by Mascaro (1976) that given independently motivated constraints cyclic 
rules will be limited to applying in derived contexts only. In order to 
see this we must examine what is meant by the assertion that a grammar 
contains rules which apply cyclically. As Mascaro remarks, at a minimum 
this requires that such rules obey convention (5): 

(5) Given the nested constituents 


n '** n-1 **' 1 ''*1 '"' n-1 * ' 'n 
a set of cyclically ordered rules C will apply to the domain 
[.....] only after having applied to the domain [._i.--. -■ ] • 

Convention (5), however, is not sufficiently restrictive for it allows 
for types of rule interactions that are much more complex than those 
actually observed. For example, convention (5) as stated admits a rule 
that applies in even-numbered constituents but not in odd-numbered constit- 
uents. Perhaps more interesting is the fact that convention (5) allows 
a feature switching rule to reapply on each successive constituent. Thus, 
it is conceivable that in a language with a feature switching rule like 
the vowel shift of Modern English, a word consisting of a single constit- 
uent with the stem vowel /i/ would appear on the surface as [ay]; if the 
same stem were part of a word consisting of two nested constituents, its 
vowel would be manifested as [ey]; if it were part of a word consisting of 
three nested constituents, its vowel would be manifested as [iy] on the 
surface. Moreover, the same series of outputs would be repeated in words 
consisting of four, five, six nested constituents, etc. 2 

Such repetitive reapplications of rules could be blocked by a restric- 
tion that would prevent rules from returning 'to earlier stages of the 
cycle after the derivation has moved on to larger, more inclusive domains' 
(Chomsky 1973:2A3). Chomsky, who encountered a similar problem in his 
work in syntax proposed the restriction (6a) : 

(6) a. No [cyclic - MH] rule can apply to a domain dominated by a 
cyclic node A in such a way as to affect solely a proper subdomain 
of A dominated by a node B which is also a cyclic node. 

The consequences of this restriction for phonology were explored by 
M.-L. Kean (1974). Although one of Kean's two examples was subsequently 
shown by M. Allen (1975) to be amenable to a simpler solution that did not 
require recourse to restriction (6a) , the paper was immensely instructive 
in that it demonstrated in detail how general abstract restrictions on the 
applicability of rules such as (6a) can serve to account for rather subtle 
surface facts. To illustrate what is involved consider the effects of the 
rules (i) and (ii) on the strings (iii) in (6b) below: 

(6) b. 

i) X Y / Z ...] 

iii) [...[PX]Z...] [ 

ii) P 


Q / _ 

.] [. 

_ Y ...] 

i) PYZ 
ii) blocked by (6a) 



Convention (6a) blocks the application of rule (6b ii) to the string 
[[PX]Z] on the second pass through the cyclic rules because the substring 
PY (from PX) is a proper subdomain of the external constituent entirely 
contained within the internal constituent. The same condition does not 
hold in the case of the substring PY in the derivation from the input 

It will be readily observed that rules (6b i) and (6b ii) are struc- 
turally identical with the Finnish rules (4a) and (4b) respectively. 
Suppose that we assumed, therefore, that the Finnish rules are cyclic and 
attempted to apply them to the strings in (6c) in accordance with constraint 

(6) c. [//[vete]#] [//[neiti]//] [#[halut]+i#] 










4a. veti 

4b. blocked by (6a) — halus+i 

Output *veti *neisi halusi 

As shown in (6c) we get incorrect outputs in two out of the three examples. 
The reason for the incorrect outputs is not the rules, but rather the 
constraint (6a) which determines under what conditions the rules will or 
will not apply. As will be shown immediately below, if the constraint on 
cyclic rule application is modified along lines suggested below the two 
rules will produce the correct output. The version of the constraint on 
cyclic rule application that I propose below is a combination of certain 
suggestions made by Kiparsky (1973:60), with others due to Mascaro (1976:9): 

(7) A cyclic rule R applies properly on cycle j only if either a) or 
b) is satisfied: 

a) R makes specific use of information, part of which is available 
on a prior pass through the cyclic rules, and part of which becomes 
first available on cycle j. There are three separate cases subsumed 
under a): R refers specifically to some A and B in: 

i) [jXAY...[^_^...B...]Z]; 

ii) [^Z[^_^...B...]XAY]; 
iii) [.X[._^...A...]Y[._^...E...]Z] 

b) R makes specific use of information assigned on cycle j by a 
rule applying before R. 

These conventions (7) generate the derivations in (8): 

(8) [//[vete]//] [//[neiti]//] [// [halut]+i//] 
4a. — blocked by (7) 


4a. veti blocked by (7) 

4b. vesi blocked by (7) halus+i 
Output vesi neiti halusi 

Like in derivation (6c) the string [//vete//] is turned into [//veti//] 
by rule (4a), but unlike (6c), this string is now subject to rule (4b) by 
virtue of convention (7b). The rule (4b) applies to the string [//halut+1//] 
by virtue of convention (7a). The derivation of /neiti/ is the most inter- 
esting of all for it illustrates the important fact that given the conven- 
tion (7) no cyclic rule can apply to the innermost constituent of a string. 
We note that (7a) requires that material from two distinct constituents be 


specifically mentioned in the structural analysis of a cyclic rule R. Con- 
vention (7b) does not require this, but R must be preceded by a rule which 
applies by virtue of (7a) and which, therefore, cannot apply to the inner- 
most constituent of a string. Thus, since neither of the conditions in 
(7) can be met by the innermost constituent of a string, no cyclic rule 
can apply to the innermost constituent. It follows from this that if a 
particular phonetic property — e.g. stress — is assigned by a cyclic rule, 
the rule cannot assign it to root words; instead stress on root words will 
have to be supplied directly in the lexical representation. It is, there- 
fore, of considerable interest that Mascaro was able to show that in 
Catalan, where stress is assigned by cyclic rules, the attempts to extend 
these rules so that they would also supply stress to root words in their 
lexical representations have been failures and that stress must be supplied 
directly in lexical representations . 3 

Upon reading an earlier version of this paper G. N. Clements has 
suggested to me that convention (7) might be replaced by the following: 

(9) a. A cyclic rule R applies properly on cycle j if and only if 
there is an immediately preceding cycle j-1 such that R must make 
specific use of information not present in the string at the end of 
that cycle. 

This convention differs from (7) in that it requires cyclic rules to apply 
in the following two types of underlying strings which are ruled out by (7) : 

(9) b. i. [.Z[._^...] XABY] 
ii. [^XA [^_^]...BZ) 

I note that since application of rule R to the string AB would be ruled out 
if AB were part of the innermost constituent, it would seem plausible that 
the rule R should not be allowed to apply to the string AB when the string 
first appears on the jth cycle. 

The only difference between the strings (9b i) and (9b ii) is that the 
units A and B to which the cyclic rule applies are adjacent to one another 
in the first, and nonadjacent in the second. I would guess that this 
difference between adjacent and nonadjacent constituents should not make 
an essential difference for a convention on rule applicability such as (7) 
or alternative (9). I, therefore, prefer (7) to the alternative (9). 
Unfortunately, neither Clements nor I have been able to think of actual 
examples of strings of the forms above. The difference between the formu- 
lations (7) and (9) must, therefore, remain moot at this time. 

To conclude it would appear that the special conditions on rules dis- 
covered by Kiparsky are unconnected with the automatic or nonautomatic 
character of the rule. Their relationship to the abstractness of lexical 
representations is indirect: since cyclic rules cannot apply to innermost 
constituents they cannot be the source of phonetic properties within root 
morphemes, such properties will in many instances have to be directly 
present in the lexical representation. What is crucial about the conditions 
discovered by Kiparsky is that they are perfectly natural conditions on 
the application of the class of rules that make reference to the internal 

constituent structure of words and phrases. Such rules have received some- 
what less attention in phonology than they deserve, especially in recent 
years. Mascaro's dissertation is particularly instructive in this respect 
for it demonstrates in detail the important role that cyclic rules play in 
the segmental phonology of Catalan. 

I have examined here three typical examples in which it was claimed 
that 'functional' rather than 'formal' considerations must be invoked in 
order to account for the facts, and have shown that in each case the 
standard 'formal' considerations provide an account that is not only 
adequate, but also superior to the other alternative. It would be hasty 
to conclude from this that 'functional' considerations are totally irrele- 
vant in phonological research. Apart from the fact that negative results 
do not warrant such an inference, there is no question that 'functional' 
factors play a role in linguistics. If their influence is often hard to 
discern clearly, this is due to the fact that they are just one type among 
a host of factors that determine the course of linguistic evolution. By 
disregarding them in our research we, therefore, run the risk of missing 
potentially promising and enlightening lines of inquiry. The search for 
these factors, however, does not warrant a disregard of other — in particu- 
lar, formal — factors in phonology. 


*To be published in a Festschrift for Oswald Szemerenyi, Amsterdam 
Studies in the Theory and History of Linguistic Science: Current Issues 
in Linguistic Theory (John Benjamins B.V., Amsterdam). (Preliminary 
Version). This work was supported in part by the National Institutes of 
Mental Health Grant #5P01 MH 13390-12. 

Kiparsky (personal communication) contests Kenstowicz and Kisseberth's 
analysis of Yawelmani, in particular their rule of /i/-epenthesis on which 
the above example crucially depends. 

That cyclic rules might function in this fashion was suggested by 
T. Bever and T. Langendoen (1963), who attempted in this manner to account 
for the e/o ablaut in Greek and other Indo-European languages. This 
ingenious idea unfortunately failed to be confirmed by the facts and has, 
therefore, been abandoned. 

The above raises questions about the SPE analysis of English stress. 

It should be noted that the only cases where stress needs to be supplied 

to the innermost constituent are the deverbal nouns of the type aondensa- 


Allen, M. 1975. Vowel mutation and word stress in Welsh. Linguistic 
Inquiry 6. 181-201. 


Bever, T. and T. Langendoen. 1963. Reciprocating cycle in the Indo- 
European e/o ablaut. MIT RLE Quarterly Progress Report //69. 202-3. 
Chomsky, N. 1973. Conditions on transformations. A festschrift for 

Morris Halle, ed. by S. R. Anderson and P. Kiparsky, 232-86. New 

York: Holt, Rinehart, and Winston. 
and M. Halle. 1968. The sound pattern of English. New York: 

Harper and Row. 
Hale, K. 1973. Deep-surface canonical disparities in relation to analysis 

and change. Current trends in linguistics. Vol. II, ed. by T. Sebeck, 

401-58. The Hague: Mouton. 
Kaye, J. D. 1975. A functional explanation for rule ordering in phonology. 

Papers from the parasession on functionalism, 244-52. Chicago: 

Chicago Linguistic Society. 
Kean, M.-L. 1974. The strict cycle in phonology. Linguistic Inquiry 5. 

Kenstowicz, M. To appear. Functional explanations in generative phonology. 

Phonology in the 1970's, ed. by D. L. Goyvaerts. 
and C. Kisseberth. 1977. Topics in phonological theory. New York: 

Academic Press. 
Kiparsky, P. 1973. Phonological representations. Three dimensions of 

linguistic theory, ed. by 0. Fujimura, 3-136. Tokyo: TEC Company. 
Lees, R. B. 1961. The phonology of Modern Standard Turkish. Bloomington: 

Indiana University Press. 
Liberman, M. and A. Prince. 1977. On stress and linguistic rhythm. 

Linguistic Inquiry 8. 249-336. 
Mascaro, J- 1976. Catalan phonology and the phonological cycle. Ph.D. 

dissertation. Department of Linguistics, MIT. Distributed by 

Indiana University Linguistics Club, Bloomington. 
Zimmer, K. 1975. Some thoughts on likely phonologies for non-ideal 

speakers. Papers from the parasession on functionalism, 556-67. 

Chicago: Chicago Linguistic Society. 


Jonathan Derek Kaye 
Universite du Quebec a Montreal 

0. In this paper I will do my best to shed a little light on the 
rather murky area of functionalism as it is applied in phonology. In part 
one I will briefly sketch two of the major currents in the functional 
approach. In addition I will attempt to show what relationship each of 
these functionalist approaches bears to current phonological practice. 

The first variety of functionalism to be considered (all too briefly) 
is what I call 'natural functionalism' in the sense of Donegan and Stampe 
(1978). 1 The second type of functionalism is, I will argue, a mixture of 
two rather different elements: questions of substantive conditions in 
phonology and a posteriori functional explanations. This latter sort of 
functionalism is associated with Kiparsky's phonology ca. 1970. It is 
also to be found as applied to syntax in Chomsky and Lasnik (1977) and 
Chomsky (1977) . For identification purposes and for lack of a better label 
I will call this 'a posteriori functionalism' recalling its essentially 
parasitic status with respect to normal linguistic hypothesis formation. 

These approaches will be discussed from the point of view of genera- 
tive phonology. Naturally my conclusions should be understood in this 
context. Other theoretical frameworks may lead to quite different results. 
Because of the above-mentioned orientation I will devote more attention to 
the question of a posteriori functionalism. 

Part two of this paper will illustrate the way a posteriori functional 
explanations figure in phonological analyses. A problem from Algonquin 
(Kaye 1978a, b; Kaye and Nykiel 1978) is discussed and a substantive condi- 
tion on evaluating phonologies is proposed. A functional explanation is 
then suggested to account for the substantive condition. 

1. One of the major problems in discussing functionalism is that 
there are just about as many different functionalisms as there are lin- 
guists who use the term. 2 I do not wish to embark on an exhaustive 
historical study of the origins, varieties and evolution of functionalism 
and so I will sketch what I hope is a reasonably accurate, albeit of 
necessity much abbreviated description of natural functionalism. Since 
definitions of this approach are exceedingly difficult to find I will 
present my own. Natural functionalism is characterized by the assumption 
that there is a direct and clear link between what is judged to be an 
essential feature of the object of study (such as its function) and theo- 
ries about that object. Expressed somewhat differently it is claimed that 
starting from this essential element we may deduce aspects of the theory 
which is to describe it. This position may be summed up by the expression, 

'form follows f unction '.^ 

The functionalism of Andre Martinet seems to me to be a good example 
of this position. Martinet (1972:7) states '. . . le langage humain est, 
au fond, fait pour la communication, et qu'il se faqonne du fait qu'il sert 
a la communication. ' For Martinet the essential feature or function of 
language is communicative. This notion can be seen in his discussion of 
push chains. Potential mergers are avoided by successive phonetic shifts, 
each shift creating another potential merger and provoking a further 
phonetic shift. This state of affairs follows from the idea that the function 
of language is communicative and the fact that a merger would run counter to 
this function to a greater or lesser extent. A phonological opposition once 
lost is no longer available to a language to express semantic distinctions 
and communication is accordingly rendered more difficult. 

A similar situation is found in functional syntax. When English lost 
case endings as a means to express grammatical relations, word order came 
to fulfill this role. It was reasoned that grammatical relations need to 
be expressed in order for communication to take place. If one means of 
accomplishing this (inflexional endings) ceased to be available for one 
reason or another, it was necessary that another mechanism be brought in 
(word order) . 

The theory of 'natural phonology' as proposed by Donegan and Stampe, 
while differing enormously from Martinet's functionalism in both the nature 
of its claims and its internal organization, seems to me to share with it a 
certain philosophical base. In describing their theory Donegan and Stampe 
(1978:1) state that, 'its basic thesis is that the living sound patterns 
of languages, in their development in each individual as well as in their 
evolution over the centuries, are governed by forces implicit in human 
vocalization and perception. ' The essential feature of language mentioned 
in the characterization of natural functionalism given above is not its 
function for Donegan and Stampe. Strictly speaking their theory is not 
functional in the above sense. The term 'essential feature' rather than 
'function' was intended to be broad enough to include natural phonology 
within the domain of natural functionalism. Precisely what this essential 
feature is, is made clear in the following passage: 'The natural subject 
matter of an explanatory theory includes all and only what the theory can, 
in principle, explain. In the case of natural phonology this means 
everything that language owes to the fact that it is spoken' (1978:4). 
\'Jhile communication is the key element in Martinet's functionalism, 
Donegan and Stampe focus on the oral nature of language.'^ The relation- 
ship between these key concepts and the theories associated with them 
appears to be quite similar. As the communicative function of language 
is supposed to be evident in the theory of Martinet so the oral nature of 
phonology should be evident in Donegan and Stampe 's natural phonology. ^ 
In their discussion of the question of innateness they give a further 
indication of the functional nature of their theory: 'The issue of 
innateness, despite all the debate it has aroused is entirely besides the 
point. What we want to know, whether it is innate or whether it is uni- 
versally acquired, is why . . .' (1978:5, original emphasis). 

The natural functionalist seeks to construct a theory which provides 
its key concept as the answer to all the questions of the sort raised in 
the above citation, i.e., natural phonology is expected to explain the 


universal presence of a given trait (innate or not) because of the oral 
nature of language; Martinet would presumably respond to the question with 
a theory offering language's communicative functional as at least part of 
the reply. 

Certain anthropological and sociolinguistic approaches appeal to 
extralinguistic features as being essential to the formulation of adequate 
linguistic descriptions. Thus, it is often claimed that language cannot 
be properly analyzed outside of its cultural context, or else that language 
must be considered as a social instrument and theories which purport to 
explain it cannot ignore this fact. The details of language's social 
function or its cultural context are not always clear, nor is their exact 
status in the formulation of linguistic hypotheses evident. Such notions 
appear to play a similar if not identical role to those mentioned in the 
discussion of Martinet functionalism and natural phonology. 

A number of questions are raised by the sort of functionalism sketched 
above. First, the metatheoretical status of natural functionalism and its 
accompanying essential features is open to at least two interpretations. 
One may view the problem of functionalism as essentially methodological. 
Natural functional considerations of the type discussed above may serve 
as heuristic aids in formulating linguistic hypotheses. Under this inter- 
pretation the functionalism of Martinet, to take one example, would 
instruct us that language's communicative function is a useful concept to 
keep in mind when formulating linguistic hypotheses. The second interpre- 
tation is to consider the function (essential feature) of language to be 
the object of study. A theory is evaluated on the basis of the extent to 
which it explains or reflects the essential feature that is supposed to 
characterize language. Thus, a theory explaining why or demonstrating 
that language is a social, cultural, or communicative instrument would be 
considered as more highly valued than one which treats such notions as 
irrelevant. A phonological theory which does not reflect the essentially 
phonetic nature of language would be regarded as inadequate. I will 
argue that neither interpretation of natural functionalism constitutes a 
viable alternative to standard generative phonological practice. There 
are two general problems associated with both of these interpretations. 
First, it is not at all obvious what function a functional theory of 
language is to be based on. Indeed, one encounters almost as many brands 
of functionalism as there are writers on the subject. Furthermore, if 
the function of a language is all that crucial either for methodologi- 
cal reasons or as an object of study, it follows that functional theories 
should differ wildly according to what has been chosen as language's 
function. But then how are we to evaluate these competing theories? 
Surely this must be done in exactly the same way that any nonfunctional 
theory is to be evaluated. 

It must be remembered that even the most intuitively 'obvious' notion 
of language function is not uncontroversial. Thus, Martinet's position 
that the function of language is communication is contested by Chomsky 
in several recent works. Although not taking a strong position on the 
question, Chomsky has suggested that the expression of thought is at least 
as strong a candidate as communication. It is not my purpose here to 
try and settle this issue, rather I wish to point out that the function 


of language is still a very open question and one is by no means constrained 
to accept any given definition in the absence of supporting evidence. Up 
to now it is just such evidence that has been lacking. 

The second general objection to natural functionalism is more serious. 
Even if somehow we could all agree on what the function of language 
really is, it is not at all clear that this would change anything about 
the way we would do linguistics. There is no reason to suppose that 
there is any clear or direct line between form and function in language. 
If this is so, then using language function as a starting point would 
appear to be methodologically unsound. Evaluating theories on the basis 
of their shedding light on an a priori defined linguistic function (e.g. 
rejecting the EST because it is unrevealing of language's social function) 
seems misguided. 

It is instructive to study the formulation of theories in other 
disciplines to emphasize this point. One can immediately exclude many of 
the 'hard' sciences from discussions of functionalism as defined above. 
It would be absurd to think of a priori functional explanations for astro- 
nomical, physical, or chemical phenomena. These sciences have survived 
without requiring natural functionalism either to guide them to fruitful 
hypotheses or to form the framework for eventual explanations of the 
phenomena within their domain. It is only when we approach the biological 
sciences that a discussion of functionalism makes any sense at all. Here 
the discovery of the structure of DNA is quite illuminating. In reading 
Watson's account of how he and Crick came up with their proposed structure 
(Watson 1968) , it is evident that not only did the function of DNA play no 
role in its analysis but that one of Watson and Crick's preoccupations was 
that its structure would not be 'pretty': 'pretty' in the sense of reveal- 
ing something about how DNA worked. Thus, Watson writes (1968:188), 
"There had been far too many days when Francis (Crick/JDK) and I worried 
that the DNA structure might turn out to be superficially very dull, 
suggesting nothing about either its replication or its function in con- 
trolling cell biochemistry " (emphasis mine/JDK). To the delight of Watson 
and Crick, DNA's structure turned out to be quite pretty indeed. What I 
wish to stress here is that there was a chance that DNA's structure could 
have been 'dull' in Watson's sense of the word. This case is interesting 
because there was general agreement about DNA's function before its 
structure had been discovered and this knowledge was not particularly 
helpful in its analysis. The application of this point to linguistics is 
clear: there is no reason to suppose in advance that linguistic structure 
bears any direct relationship to any function, real or imagined. I will 
now turn to specific criticisms of the two interpretations of natural 

Natural functionalism may be interpreted as requiring or suggesting 
that a language's function be taken into account in formulating hypotheses 
about its structure. From an epistemological point of view the above 
statement is to be rejected. In evaluating hypotheses the method that 
was used to arrive at them is irrelevant. Hypotheses are evaluated as to 
how well they explain the phenomena found within their domain and how they 
compare to other hypotheses with similar domains. How one came by these 
hypotheses does not figure in the evaluation." 


We could also consider this methodological interpretation as a heuris- 
tic device: while not being absolutely necessary to the formulation of 
linguistic theories, it may be a good idea or a useful suggestion to 
follow such a procedure. I have not seen any evidence that such is the 
case. The theory of generative phonology has progressed normally over 
the years without any particular emphasis on functional considerations ,7 
and no clear examples come to mind where functional considerations have 
helped to overcome a particular problem in this framework. I conclude that 
the case for natural functionalisra as a methodological tool remains unproved, 
indeed without any apparent support of any kind. We may now turn our 
attention to the claim that linguistic theory must explain or incorporate 
the particular function or essential feature assigned to language by the 
brand of functionalism in question. Natural phonology requires that 
linguistic theory include 'everything that language owes to the fact that 
it is spoken. ' It is perfectly appropriate that one should choose this 
as the object of study of phonology. I fail to see, however, that one 
should accord this approach some prior claim to legitimacy. It seems to 
me that natural phonology is a reflection of the research that happens 
to interest its adherents. Failure to share one's research interests is 
hardly a crushing criticism. In as much as natural phonology and gener- 
ative phonology share common objectives, one may compare them. If they 
seek to achieve fundamentally different goals, any comparison or criticism 
is pointless. Language can certainly be described from a social, cultural, 
or communicative point of view. It does not follow that such descriptions 
or theories have anything to do with a linguistic theory that seeks to 
describe whatever internalized knowledge characterizes the set of speakers 
of some language and what universal principles underlie the set of all 
languages. The functional approach described by Heath (1978:89) attempts 
to ". . . describe a set of functions which, by means of different combi- 
nations of formal units and with inevitable variations in (sociocultural) 
environmental detail, play fundamental roles in shaping the formal gram- 
mars of individual languages.' There is no more reason to accept this 
view of linguistic theory than any other. Ultimately the results emanating 
from such programs and further lines of research suggested by them will 
determine their success. Again, I can see no reason to accord any prior 
claim to research programs incorporating functionalist orientations. To 
conclude this extremely brief discussion of natural functionalism, I see 
no role for such an approach in the development of phonological theory. 
It appears epistemologically unsound and offers no clear methodological 
advantages. A priori theoretical constraints of this sort seem unduly 
confining or else too vague to be of much use. As for their role in defin- 
ing research areas, natural functionalism has not proved any more fruitful 
than nonfunctional approaches. Finally, functionalism has not played any 
obvious role in the development of other (nonsocial) sciences. 

A rather different functional approach has come into prominence in 
the last ten years. P. Kiparsky has been the dominant figure in this 
approach with C. Kisseberth, M. Kenstowicz, Y.-Ch. Morin, K. Hale, and 
many others normally associated with generative phonology also making 
contributions to this area.^ I will show that this current is radically 
different from natural functionalism and for the most part is really a 
confusion of several distinct questions. ^ 


Generative phonology attempts to explain, among other things, how a 
child exposed to a limited set of primary linguistic data arrives at an 
extremely rich phonological system in a remarkably short period of time. 
It has been assumed that the child had an enormous 'head start' in the 
acquisition process: a way of selecting a phonological grammar from the 
set of all those that are compatible with the primary linguistic data to 
which he/she was exposed. This 'head start' was characterized in genera- 
tive phonology by the evaluation metric: a diverse set of properties that 
were reflected by the theory's notational system such that more general 
or natural rules and systems were less costly (required fewer symbols to 
express) in this system. Towards the end of the sixties Kiparsky came 
up with several properties that were not conveniently reflected in the 
notational system but that also seemed to characterize more natural 
linguistic systems. It was claimed that such properties were also used 
by children acquiring a language in ways analogous to the evaluation 
metric. Kiparsky (1971) termed these conditions substantive as opposed to 
formal. These substantive conditions which included such notions as 
opacity, abstractness, and paradigm regularity were no more functional in 
nature than natural classes, feature hierarchies, and markedness conven- 
tions. The latter could be read off directly from the notational system 
while the former could not. Nothing in principle would prevent the even- 
tual refinement of the notational system so that say, opacity could 
likewise be directly represented. At that point opacity would be a formal 
rather than a substantive condition on phonologies but it would in no way 
change any functional characteristics it might have. It just so happens 
that functional explanations became associated with these substantive 
conditions although there is no logical link between them. To take one 
example, Kiparsky showed that rules became less highly valued to the extent 
to which they had become opaque. Kiparsky's definition of rule opacity 
naturally accompanied this claim. At this point there is nothing whatever 
that would lead one to consider Kiparsky's contribution as functional. 
Kiparsky went on to speculate as to why language should favor transparent 
rules over opaque ones and suggested that transparent rules are easier to 
learn. It is at this point that a functional element entered the discus- 
sion. Since functional notions of this sort are made only after the 
postulation of substantive conditions on phonology, I have called this sort 
of functionalism, 'a posteriori functionalism' . In fact, as I will try to 
show below, this is not a distinct theoretical orientation from normal 
generative phonology. 

To begin with, two points must be made clear: the conjectural nature 
of the functional claims and the fact that the validity of the hypotheses 
on which they are based is completely independent of such functional 
statements. Functional statements along the lines of Kiparsky's assertion 
that transparent rules are easier to learn, are generally offered as a 
sort of 'coda' that follows normal phonological arguments justifying the 
substantive conditions in question (in this case rule opacity) . There is 
an (implicit) appeal to common sense but no effort is devoted to justifying 
these sorts of claims. Notice also that the validity of the functional 
statement has no bearing on the validity of the theoretical status of the 
substantive condition on which it was based. Suppose, for example, that 
we could somehow prove that opaque rules were not more difficult to learn 
than transparent ones. This would in no way imperil the claim that 


grammars with opaque rules are less highly valued than those without them. 
This latter claim was based on evidence completely independent of any 
functional considerations. 10 The data for which rule opacity was set up 
to account would remain untouched by the success or failure of any func- 
tional explanation. The lack of any functional motivation for a substan- 
tive condition would simply indicate that this particular aspect of 
linguistic structure was not 'pretty' in the sense of Watson discussed 

Susumu Kuno was one of the early practitioners of a posteriori 
functionalism. Kuno (1972) suggested that many of Greenberg's word order 
universals follow if language processing involved a push-down storage 
device at some point. Again, notice that this claim was made only after 
Greenberg had postulated his universals. Greenberg's observations about 
word order were not made with functional considerations in mind, nor 
does their validity rest on the success of a posteriori functional explana- 

Chomsky and Lasnik make repeated reference to functional explanations 
of this kind. For example they state (1977:470), 'As we have observed, 
"functional" considerations are quite appropriate in the study of language 
evolution. We have pointed out that these filters have at least one 
striking effect: they sharply reduce the range of possible outcomes from 
well-formed base-generated structures, while still permitting at least 
one in each case. Thus they narrowly restrict the association of deep and 
surface structures.' Chomsky and Lasnik certainly did not set out to 
devise a linguistic system with the above-mentioned property (i.e. reduc- 
ing the range of possible outcomes from well-formed base-generated 
structures). This result was the result of normal syntactic research. 
It just so happened that the resulting model was pretty enough to allow 
these functional properties to be read off of it. Even in this case, 
however, the explanation is not quite functional. One could propose a 
general condition on grammars to the effect that they are more highly 
valued if they possess the property of having reduced ranges of outcomes 
from well-formed structures. Predictions involving linguistic change, 
interference phenomena and perhaps some absolute limits on the range of 
possible grammars could follow from such a condition on grammar. It must 
be noted that there is no a priori reason to believe that such a condition 
exists. It could only be justified in the same way that the opacity con- 
dition, the subjacency condition for movement rules or the filters them- 
selves were justified. Given the existence of this substantive constraint 
one might then offer a functional explanation (it makes processing easier, 
etc.) to account for its presence. In any event the position of those 
functional explanations in linguistic theory is quite tenuous. 'The 
functional explanation . . . holds, if at all, at the level of evolution 
of the species. It does not relate to explanatory adequacy ..." 
(Chomsky and Lasnik 1977:437). 

Returning to phonology, Halle in his contribution to this volume has 
suggested that all functional considerations should reside in the simpli- 
city metric. In other words, the destinction between formal conditions 
on phonology (say, rules involving natural classes are more highly valued 
than those without such classes) and substantive conditions (say, rule 


opacity) should be eliminated. This could be interpreted to mean that all 
conditions ranking phonologies should be formulatable in the notational 
system, either in the traditional way, higher ranking phonologies being 
less 'costly' in terms of the number of symbols needed to express them, or 
by extending the formalism and adding ancillary conditions. 

The former conditions include the natural classes, abbreviatory 
devices and so on. The system was subsequently enriched to include mark- 
edness, rule variables, metric structures, and so on. There appears to be 
no reason to distinguish these two classes and in that sense I agree with 
Halle's suggestion. Furthermore, it seems a reasonable proposal to attempt 
to incorporate substantive conditions such as rule opacity and recovera- 
bility into the notational system. I see no problem in such an endeavor 
and at worst it would have as a consequence the elimination of an arbitrary 
distinction in current generative phonology. Functional explanations may 
be just as easily proposed for, say, markedness, which is currently con- 
sidered a formal condition, as for rule opacity, which is a substantive 
one. Thus Halle's proposal would leave unchanged the functional status 
of generative phonology. It would have the beneficial effect of disentan- 
gling functionalism from substantive conditions on phonology. In the 
following section I will illustrate the sort of a posteriori functionalism 
discussed in the previous paragraphs. 

2. In this section I will discuss a substantive condition on phonology 
and show its role in the analysis of a particular phonological problem. 
I will indicate how such questions relate to the a posteriori functionalism 
discussed in the preceding section. 

In some recent articles (Kaye 1974; 1975) I discussed the notion of 
recoverability in phonology. Very roughly, recoverability is a limit on 
the amount of surface ambiguity of a phonological form. What this means is 
that given a phonetic form and the rules, the inventory of phonemes and 
(deep) phonotactic constraints of the language in question, the number of 
possible sources of the phonetic form may be computed. For example, in 
German given a phonetic string ending in [-t], given the rule of final 
devoicing and given the fact that morphemes may end in -t or -d, the string 
has two possible sources, viz. /-t/ or /-d/. Only comparison with morpho- 
logically related forms containing suffixes will indicate which source is 
to be chosen. In the absence of such related forms one would choose /-t/ 
as the underlying final consonant since German final devoicing is a neutra- 
lization rule. Recoverability then evaluates phonologies as more highly 
valued if such ambiguity falls within certain limits. In the example 
below I will illustrate the application of this principle to (deep) phono- 
tactic constraints. Lac Simon Algonquin contains both a word structure 
constraint (WSC) which is rendered completely opaque by a phonological 
rule. 11 

Lac Simon Algonquin has the WSC given in (1) below. 

(1) Algonquin WSC: No word may begin with a cluster or a voiceless 

In addition the following data (2) indicate that Algonquin has acquired a 


rule of initial devoicing (3). 

(2) pi:goska: 'it breaks ki:bi:goska: 'it broke' 
ka:zo:tiw 'he hides' niga:zo:ta:n 'I hide' 
te:si>biwa:gin 'chair' kide:sibiwa:gin 'your chair' 
sigiswa: 'he smokes' ta:zigi>swa: 'he would smoke' 
6i:ma:n 'canoe' oji:ma:n 'his canoe' 
so:skose: 'it slides' nizo:skose: 'I slide' 

(3) Initial Devoicing C -^ [-voice] /## 

While not altogether a complete justification for the analysis (which is 
irrelevant in any event for the point under discussion), the following data 
indicate that we are dealing with a rule of initial devoicing rather than 
medial voicing. 

(4) a. kitiga:n 'cultivated field' kici 'much, big, great' 

a:sibi: 'net' pi:sa:dik 'in height' 

a:nipi:c 'when?' n4>do :z4,to:n 'I make it' 

b. ni-kin 'my bone' o-se:z-in 'his brother' 
ki-sogona: 'your nose' ni-ka:t 'my leg' 

c. a:sibi: 'net' a:sibi:-ka:k 'in the net' 
pa:pa:se: 'woodpecker' pa:pa:se:-s4>s 'little woodpecker' 

(4a) shows that both voiced and voiceless obstruents occur word-medially 
although only voiceless obstruents occur word-initially at the phonetic 
level. (4b) shows that dependent noun stems may also begin with voiceless 
obstruents. These forms are preceded obligatorily by prefixes and thus 
cannot violate (1). Finally, (4c) shows that suffixes may also have initial 
voiceless obstruents. 

The status of nonalternating (i.e. always initial) morphemes is crucial 
for this analysis. If one assumes that rule (3) is neutralizing, nonalter- 
nating obstruent-initial morphemes as in (5) would be required to have 
initial voiceless obstruents in their underlying representations. If, on 
the other hand, rule (3) is assumed to be allophonic, even nonalternating 
obstruent-initial morphemes would have underlying Initial voiced obstruents. 

(5) Obligatorily initial morphemes with initial obstruents 

ki- '2nd person pa-inime: 'it is ka:- 'relative marker' 

prefix' necessary 

ki:spin 'if te:digo: 'much' pi:a 'and then' 

This situation is somewhat reminiscent of the much discussed German case. 
The two languages are compared in (6) below. 

(6) The status of rule IV 



voiceless voiced /voiced/ /voiced/ voiceless voiced 

po:de: ki:bo:de: vat vzd'dr 

'he makes a fire he made a fire' 'wheel' 'wheels' 

voiceless voiceless /voiceless/ /voiceless/ voiceless voiceless 

NO FORMS EXIST rat retd 

'counsel' 'counsels' 

voiceless always initial /???/ /voiceless/ voiceless always final 
pa:nime: 'it is necessary...' unt 'and' 

Both German and Algonquin have voiced-voiceless alternations and both 
have forms which always occur in the suitable context for the devoicing rule 
to apply (i.e. initially in Algonquin and finally in German). Algonquin 
differs from German in that there are no examples where a voiceless 
obstruent occurring in initial position is also voiceless when the same 
morpheme follows a prefix. This is, or course, simply a reflection of (1). 
The question now centers around the forms of (5) . In (7) the two possible 
solutions are sketched. Solution I assumes that the forms of (5) have 
underlying voiced obstruents initially while solution II assumes they are 
voiceless. Since both solutions are compatible with the data we must look 
elsewhere for evidence in support of one or the other analysis. Further- 
more, we must find a principled basis for preferring or requiring one 
solution over the other. 

(7) Solution I 

/gi:spin/ /de:digo:/ /bi:c/ 

Rule IV: [+cons, -son] — > [-voice] ///// 

Word Structure Constraint: No word may begin with a voiceless 
consonant . 

Solution II 

/kirspin/ /te:d±go:/ /pi:c/ 

Rule IV: [+cons, -son] — > [-voice] ///// 

The principle difference between the two solutions is the presence or 
absence of the WSC. Evidence from the nativization of loan words favors 
the former solution. 

(8) Loan Word Behaviour 

a. pa:na:n 'banana' (<banana) n<i>ba:na:nim 'my banana' /ba:na:n/ 
pa.'stone: 'American' (<Boston) ka:ba: stone :na:goni6 'he who 

looks like an American' /ba: stone:/ 

b. pitik 'potato' (<petac) nibitikim 'my potato' /bitik/ 
toma:do:s 'tomato' (<tomatoes) nidoma : do : zim 'my tomato' 


ko:fi:ke: 'he makes coffee' (<coffee) nigo:fi:ke: 'I make 
coffee'/ go:fi:ke:/ 

In (8a) loan words with an initial voiced obstruent in their source 
language are presented. Iften preceded by a prefix the initial obstruent 
retains its voiced status. In absolute initial position it is realized as 
voiceless. These forms have no bearing on the status of the WSC (1) but 
rather show that rule (3) is still operative. The forms of (8b) begin with 
voiceless consonants in the source language. In Algonquin they are voice- 
less in initial position. In noninitial position, however, they are voiced. 
Since we have already stated that there is no rule of medial voicing, we 
must assume that these forms were interpreted as having an initial voiced 
obstruent. The only apparent reason for such an interpretation is the 

existence of the WSC (1), ruling out word-initial voiceless obstruents as 
possible underlying forms. 

Evidence from loan word phonology favors solution I over solution II. 
What feature of this solution renders it more highly valued than the other 
solution? Both solutions require the presence of rule (3). The only 
difference in the formal structure of these solutions is the presence of 
a phonotactic constraint in solution I. By incorporating a substantive 
condition into our phonological theory we can require the selection of 
solution I over solution II. The substantive condition in question is 
given below: 

(9) All things being equal, phonologies with phonotactic constraints 
on possible underlying forms are more highly valued than those without 
such constraints. 

If this turns out to be a correct observation about ranking phonologies 
one may well turn to the question of why phonological theory has a condi- 
tion like (9) : why should analyses with phonotactic constraints be 
selected in the acquisition process over those without such constraints, 
even in the face of apparently contradictory surface forms? In other 
words, is there a functional explanation for (9)? One striking property of 
this condition is to limit the number of underlying forms and consequently 
to facilitate the task of associating a given phonetic form with its 
lexical representation. Indeed we may view (9) as simply a special case 
of a more general condition of recoverability. Note that the case for 
phonotactic constraints as well as other manifestations of recoverability 
may simply be another aspect of Chomsky and Lasnik's functional explanation 
mentioned above. Filters had the property of reducing the number of out- 
comes associated with a given base-generated structure. Phonotactic 
constraints have the property of reducing the possible underlying sources 
associated with a given phonetic string. It would be most tempting to 
look for a general functional explanation which would incorporate both 

The purpose of the above illustration was to show the role of (a pos- 
teriori) functional explanations in the current practice of generative 
phonology. It should be noted that things go on pretty much as they did 
in the past with functional considerations playing no role in the quest 
for explanatory adequacy. Like Watson and Crick, phonologists will go on 
with their analyses with the hope that what results will be pretty. 

'This is a natural theory in the sense established by Plato in the 
Cratylus , in that it presents language (specifically the phonological 
aspect of language) as a natural reflection of the needs, capacities, and 
world of its users, rather than as a merely conventional institution' 
(1978:2; original emphasis). 

See Baron (1974) for discussion of the problem. 

Heath (1978:92). '(Although) in this framework we take function 
as prior to form . . . ' 

It must be stressed that Donegan and Stampe are only talking about 

the phonological aspect of language. Martinet is dealing with all aspects 

of language. 

If my understanding of natural phonology is correct, anything remotely 
resembling a phonological component in a sign language such as ASL should be 
pure coincidence. 

See Feyerabend (1978) for a discussion of this question. 

Cf. the discussion of a posteriori f unctionalism below. 

See also Morris Halle's contribution to this volume. 

I am one of the parties guilty of this confusion. See Kaye (1975). 

Kiparsky's arguments for this rule opacity involve evidence from 
sound change, specifically rule reordering. 

This analysis has proved to be rather controversial. The reader is 
referred to Kaye (1978a; 1978b) and Kaye and Nykiel (1978) for the 


Baron, N. S. 1974. Functional motivations for age grading in linguistic 

innovation. Historical linguistics I, ed. by J. Anderson and C. 

Jones, 33-64. Amsterdam: North Holland. 
Chomsky, N. 1977. Dialogues avec Mitsou Ronat. Paris: Flammarion. 
and H. Lasnik. 1977. Filters and control. Linguistic Inquiry 

Feyerabend, P. 1978. Against method. London: Verso. 
Heath, J. 1978. Functional universals. Berkeley Linguistic Society 

Kaye, J. 1974. Opacity and recoverability in phonology. Canadian 

Journal of Linguistics 19.134-149. 

1975. A functional explanation for rule ordering in phonology. 

Papers from the parasession on f unctionalism, 244:52. Chicago: 

Chicago Linguistic Society. 

1978a. On the alleged correlation of markedness and rule func- 
tion. Current phonological theories, ed. by D. Dinnsen. Bloomington; 

Indiana University Press. 

1978b. Abstractness recoverability and phonotactic constraints. 

To appear in Phonology in the 1970 's, ed . by D. Goyvaerts. 


and B. Nykiel. 1978. Loan words and abstract phonotactic con- 
straints. Paper read at the 16th International Polish-English 
Contrastive Linguistics Conference, Boszkowo, Poland. 

Kiparsky, P. 1971. Historical linguistics. A survey of linguistic 
science, ed. by W. Dingwall, 576-642. College Park: University 
of Maryland. 

Kuno, S. 1972. Natural explanations for some syntactic universals. 
Mathematical linguistics and automatic translation. Report no. 
NSF-28 to the National Science Foundation. Harvard University. 

Watson, J. 1968. The double helix. New York: Atheneum. 


John R. Searle 
University of California at Berkeley 

It is possible to make a reasonably clear distinction between inten- 
tionality with an s and intentionality with a t. Without putting too fine 
a point on it one can say that intensionality-with-an-s is a property of 
a certain class of sentences. A sentence is intensional if literal utter- 
ances of it have at least one interpretation where they fail to satisfy 
one or more of the standard tests for extensionality. The two tests most 
relevant to the present discussion are these: if existential generalization 
over the occurrence of referring expressions is not a valid form of 
inference or if the sentence fails to allow the substitution salva veritate 
of expressions which normally have the same reference, then it is inten- 
sional-with-an-s. Thus for example, the sentence 'John is looking for the 
lost city of Atlantis' has at least one literal use to make a statement 
which does not entail that there is a lost city such that John is looking 
for it. And the sentence 'The sheriff believes that Mr. Howard is an 
honest man' has a literal use to make a statement which together with the 
true statement that Mr. Howard is identical with Jesse James does not 
entail that the sheriff believes that Jesse James is an honest man. 
Because intensional sentences normally derive their intensionality from 
the occurrence of certain expressions it is possible to speak not only of 
intensional sentences, but also of intensional verbs, intensional contexts, 
etc. It is also possible to speak of intensional statements and inten- 
sional propositions. No doubt these two tests need some refinement to 
enable them to cope with sentences other than those used to make statements 
but the basic idea of intensionality-with-an-s seems reasonably straight- 

When it comes to intentionality-with-a-t it is much harder to say 
exactly what it is we are talking about. It is common to say that inten- 
tionality-with-a-t is not primarily a property of sentences, but a property 
of some or perhaps even all mental phenomena. It is the property which 
mental states have of being directed at objects or states of affairs. Thus 
for example, a belief is always a belief that such and such is the case, 
a fear is always — or at least in general — a fear of^ something, a desire is 
always a desire to have something or that something be the case. Just so 
we have some fairly clear idea of what it is we are talking about, I pro- 
pose the following as a rough preliminary test for intentionality-with-a-t. 
A mental state is an intentional state if and only if the specification of 
the content of that mental state requires the specification of some object 
or state of affairs which is not identical with that mental state. On 
this test pains and aches and at least some cases of anxiety are not inten- 
tional, whereas beliefs, hopes, expectations, and desires are. A specifi- 
cation of the content of my pain is just a further description of the pain; 
but the specification of the content of my belief, hope, expectation, or 

desire must specify what it is that I believe, what I hope for, what I 
expect, and what I desire. All that a pain has to have to be a pain is to 
have a certain feel to it, certain phenomenal qualities, but for belief, 
hope, expectation, and desire, something more is required: the questions 
'what do you believe, what do you hope for, what do you expect, what do 
you desire?' must have answers, if the agent can be said to have a belief, 
hope, desire, or expectation at all; and those answers will specify 
objects and states of affairs that are not identical with the mental states. 

In Wittgenstein's jargon, pains and aches have causes but not targets, 
whereas love and hate, belief and desire have both causes and targets, 
and in each case the cause may or may not be identical with the target. I 
will call such mental states intentional states, and the objects and states 
of affairs at which they are directed intentional objects. The notion of 
an intentional object is a frequent source of confusion in philosophy and 
I will have more to say about it later on. We will see that for the so- 
called propositional attitudes it is not really necessary at all. For 
the present it is important to note that on this criterion not all mental 
states are intentional states; only those which require the specification 
of an intentional object are intentional states. It is also important to 
note that this criterion is not intended as an analysis of the notion of 
intentionality. If so, it would be hopelessly inadequate since it rests on 
several unexplained and obscure notions: in what sense does a mental state 
have a 'content' and what is meant by saying the 'specification' of that 
content 'requires' the specification of something else? 

This test for intentionality-with-a-t is designed to isolate those 
mental states which are in some sense directed at objects and states of 
affairs from those which are not; but even assuming that it is successful 
in doing that, it raises but so far leaves unanswered the hard questions 
about intentionality. The two most pressing questions raised by the 
criterion are these: the statements specifying the object or state of 
affairs in the specifications of the intentional state are not like 
ordinary statements containing specifications of objects and states of 
affairs because they do not require the existence of the objects or state 
of affairs in order to be true. It can, for example, be true that John 
expected it would rain and that he worships God without it being true 
either that it rained or that God exists. Indeed, the sentences which 
specify intentional states are themselves intensional-with-an-s on both 
our criteria for intensionality-with-an-s . So our first question is: why 
is that so? How does intentionality-with-a-t give rise to intensionality- 
with-an-s? The second question is this: what exactly is the relation 
(or set of relations) between intentional states and their objects, and 
what is it about this relation that requires the specification of an 
intentional object in the specification of an intentional state? I think 
an answer to the second question will enable us to answer the first and 
to clarify the notion of an intentional object. 

In discussing the second question, Wittgenstein frequently makes 
remarks like the following 'If someone could see the expectation itself — 
he would have to see what is being expected' and 'The representation of a 
wish is eo ipso the representation of its fulfillment' (cf. Wittgenstein 
1967a, 1967b) • Wittgenstein himself and Kenny in his book on Wittgenstein 

claim that there is an 'internal relation' between the intentional state 
and its intentional object. But this characterization is not much help 
to us unless we are told exactly what is meant by the notion of an internal 
relation. If for example, someone tells us that there is an internal 
relation between being a triangle and being three-sided, it is possible 
to make some fairly clear sense of the notion of an internal relation: a 
logically necessary condition of being a triangle is being three-sided. 
But in that sense there clearly is no internal relation between my belief 
that it will rain on Wednesday and the state of affairs of its raining on 
Wednesday. The proposition that it will rain on Wednesday neither entails 
nor is entailed by the proposition that I believe that it will rain on 
Wednesday. It seems that the only sense we can give to the notion of an 
internal relation in this case is one we have already spelled out in our 
test for intentional states — the specification of an intentional mental 
state requires a specification of its intentional object. But that still 
leaves us with our second question unanswered. What is it exactly about 
intentional states that relates them to their objects? Mackie (1975:48) 
poses our second question as sharply as anyone: 'At last we are beginning 
to get into focus the real puzzle about intentionality, ' he writes, 'Is 
it not strange that there should be one state of affairs (Tom's state of 
believing, hoping, fearing, or whatever it may be) that requires for its 
adequate description the partial, incomplete, selective, indeterminate, 
description of a quite different and so far merely possible state of 
affairs?' I think there is a fairly simple answer to Mackie 's question. 
To see this answer imagine a parallel question about statements and speech 
acts generally. Suppose someone said, 'Is it not strange that there should 
be one state of affairs (Tom's statement that it is raining or his order 
to Bill to leave the room) that requires for its adequate description the 
partial, incomplete, etc., description of a quite different and so far 
merely possible state of affairs (the state of affairs that it is raining 
or that Bill leaves the room) . ' But there is nothing puzzling about this 
requirement once you recognize that a statement is precisely a representa- 
tion of a state of affairs that is asserted to exist and an order is a 
representation of a state of affairs which the hearer is ordered to bring 
into existence. Any speech act with a propositional content contains a 
representation of some object or state of affairs; and in the same way, I 
wish to argue, all intentional states are representations of objects and 
states of affairs. The reason that the specification of my belief that 
it is raining requires a specification of the state of affairs that it is 
raining is exactly the same reason that the specification of my statement 
that it is raining requires a specification of the state of affairs that 
it is raining: in each case the one is a representation of the other. 

The answer to our second question then, how does the intentional state 
relate to the intentional object and what is it about the relation that 
requires that a specification of one involves a specification of the other, 
is simply that the intentional state contains a representation of the 
intentional object in the same sense that speech acts contain representa- 
tions of objects and states of affairs. 

Because an intentional state contains a representation we can now give 
a clear sense to the notion that it is 'internally related' to the object 
it represents: Any representation is internally related to its object in 


the sense that it could not be that representation if it did not have that 
object. Thus for example, an identity criterion of my belief that it is 
raining is that it must have as its intentional object the state of affairs 
that it is raining. 

I don't much like having to invoke the notion of 'representation' 
here because it has so little explanatory power, but it is at least correct 
and it does enable us to answer both our second and now our first question. 
How does intentionality-with-a-t give rise to intensionality-with-an-s? 
Sentences about intentional states are at least in part about representa- 
tions. That being the case, their truth conditions will sometimes depend 
on features of the representation and not entirely on features of — or even 
on the existence of — the object represented. To be looking for the lost 
city in Atlantis is to have a representation that one seeks to instantiate, 
and the statement that one has such a representation can be true even where 
there is no such instantiation. And similarly to believe that Mr. Howard is 
an honest man is in part to have a mental representation associated with 
the name 'Mr. Howard' which may be quite different from the mental repre- 
sentation associated with the name 'Jesse James', hence the failure of the 
substitutability of such names even where Mr. Howard is identical with 
Jesse James. 

This conception of intentionality as representation will also enable 
us to get clear about the nature of intentional objects. People often talk 
as if intentional objects had some peculiar ontological status and had to 
be distinguished from actual objects. But the intentional object of a 
mental state is just the actual object or state of affairs represented by 
an intentional state. If there is no such object or state of affairs then 
the intentional state does not have an intentional object though it does 
still contain a representation. We need therefore to distinguish the 
representative content of a mental state from the intentional object of 
that mental state. If John loves Sally and believes that it is raining 
then the intentional object of his love is Sally, the actual flesh and blood 
Sally and not some mental phenomenon, and the intentional object of his 
belief is the state of affairs in the world that it is raining and not the 
proposition that it is raining. In order that his love should be of Sally 
and his belief be that it is raining he must have some representation of 
Sally and of the state of affairs that it is raining. But the object of 
his intentional states is not these representations, rather the intentional 
states are directed at their objects by way of their representative content. 
A proposition one might say is not the object of a belief, it is the content 
of the belief. The oscillation between the extensional and the intensional 
reading of statements about intentional states is precisely an oscillation 
in the extent to which the statement is committed only to facts about the 
representative content or to facts about the intentional objects.! 

The distinction between the representative content and the inten- 
tional object is parallel to Frege's distinction between sense and reference. 
Just as a definite description refers to an object in virtue of its sense, 
but does not thereby refer to its sense, so an intentional state is 
directed at an object in virtue of its representative content, but is not 
thereby directed at its representative content. Both sentences describing 
intentional states and sentences describing acts of referring are subject 


to extensional and intensional interpretations and both for the same reasons, 
Thus 'John referred to the King of France' and 'John thought about the King 
of France' have both extensional and intensional readings, depending on 
whether they are construed as about objects of representations or solely 
about the representations themselves. 

The reason that many authors have failed to see that the intentional 
object is identical with the actual object is that the specification of 
the intentional object is by way of the aspect under which it is represented 
by the representative content, and for that reason it is an intensional- 
with-an-s specification. To put it crudely, it has seemed to them that 
the intentional object is incomplete in ways that actual objects are not 
incomplete and that therefore intentional objects can never be actual 
objects. How can the actual object be identical with the intentional 
object when the actual object has all sorts of features the intentional 
object does not have? Thus Davidson (1968:81-82) writes: 

What is less obvious, at least until we attend to it is that 
the event whose occurrence makes 'I turned on the light' true 
cannot be called the object, however intensional, of 'I wanted to 
turn on the light'. If I turned on the light, then I must have 
done it at a precise moment, in a particular way — every detail is 
fixed, but it makes no sense to demand that my want be directed 
at an action performed at any one moment or done in some unique 
manner. Any one of an indefinitely large number o