Current Approaches to African Linguistics: Vol 5 Vol. 5 [Reprint 2020 ed.] 9783112420089, 9783112420072

Citation preview

Current Approaches to African Linguistics (vol.5)

Publications in African Languages and Linguistics The aim of this series is to offer, at regular intervals, carefully worked out studies or collections of papers in African linguistics. In this way, it is hoped that important work can be brought to the attention of a wider linguistic audience including scholars who are not primarily concerned with African linguistics but who wish to keep abreast of recent advances in the fields of historical linguistics, theoretical phonology and linguistic typology, all of which, incidentally, tend to rely rather heavily on African language data. At the same time, the editors intend to include occasional volumes which offer general accounts of major language groups, hoping that such surveys will be of assistance not only to linguists working on African languages but also to those who specialize in other areas. Throughout the series an attempt will be made to strike a healthy balance between theory oriented and data oriented research. Editors:

Advisory

George N. Clements Didier L. Goyvaerts

John Goldsmith (Bloomington, Indiana) Claire Grégoire (Tervuren) Frank Heny (Middlebury, Vermont) Larry M. Hyman (Los Angeles, California) William R. Leben (Stanford, California) Thilo C. Schadeberg (Leiden)

Other

books

in this

board:

series:

1. Ivan R. Dihoff (ed.) Current

Approaches

to African

Linguistics

(vol.

1)

2. Gerrit J. Dimmendaal The Turkana

Language

3. G.N. Clements and J. Goldsmith (eds.) Autosegmental

Studies

in Bantu

Tone

4. Koen Bogers, Harry van der Hulst en Maarten Mous (eds.) The Phonological

Representation

of

Suprasegmentals

5. Jonathan Kaye, Hilda Koopman, Dominique Sportiche and André Dugas (eds.) Current

Approaches

to African

Linguistics

(vol.

2)

Linguistics

(vol.

3)

6. Gerrit J. Dimmendaal (ed.) Current

Approaches

to African

7. David Odden (ed.) Current

Approaches

to African

Linguistics

(vol.

4)

Paul Newman and Robert D. Botne (eds.)

1989 FORIS PUBLICATIONS Dordrecht - Holland/Providence RI - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Distributor for the U.S.A. and Canada: Foris Publications USA Inc. P.O. Box 5904 Providence R.I. 02903 U.S.A. CIP-DATA KONINKLIJKE BIBLIOTHEEK, DEN HAAG Current Current Approaches to African Linguistics. - Dordrecht [etc.] : Foris Vol. 5 / Paul Newman and Robert D. Botne (eds.). - (Publications in African Languages and Linguistics ; 8) A selection of papers presented at the 17th Annual Conference on African Linguistics, Indiana University, Bloomington, April 3-5, 1986. - With index, ref. ISBN 90 6765 323 3 bound ISBN 90 6765 324 1 paper SISO.afri 830 UDC 809.6(063) Subject heading: African Linguistics.

ISBN 90 6765 323 3 (Bound) ISBN 90 6765 324 1 (Paper) © 1989 Foris Publications - Dordrecht

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in the Netherlands by ICG Printing, Dordrecht.

Preface

This volume consists of a selection of papers presented at the 17th Conference on African Linguistics, held at Indiana University, Bloomington, April 3-5, 1986. The editors selected these papers on the basis of their general linguistic appeal and as representative of current research being done in various language families and linguistic areas. We thank the conference participants who cooperated so willingly in submitting their papers and regret that all could not be accommodated in this volume. The 17th Conference was a lively, well-attended meeting. More than 160 participants attended and over a hundred papers were presented. We wish to express our appreciation to Roxana Ma Newman for her fine job as organizer of the conference. For their generous assistance and financial support in hosting the conference, we would like to thank Patrick O'Meara and the staff of the African Studies Program, the Office of the Vice-President (Kenneth Gros-Louis, Vice-President), the Linguistics Department (Harry Gradman, Chair), the Computer Science Department (Edward Robertson, Chair), the Indiana University Linguistics Club, and the IU Conference Bureau. PN & RDB

Contents

Preface

v

General 1. Ulrike Claudi and Bernd Heine On the Nominal Morphology of "Alienability" in Some African Languages

3

2. Linda Schwartz Asymmetrical Syntax and Symmetrical Morphology in African Languages

21

Afroasiatic 3. Russell G. Schuh Long Vowels and Diphthongs in Miya and Hausa

35

4. William R. Leben, Sharon Inkelas, and Mark Cobler Phrases and Phrase Tones in Hausa

45

5. Ismail Junaidu On the Nature of Topicalization in Hausa

63

6. Maria-Rosa Lloret Final Vowels and Grammatical Marking in Oromo

73

7. Ishetu Kebede The Copula in Oromo

85

8. S. Filipovich and M. Tremblay The Case of U in Tigre: Towards an Extended Theory of Phonological Government

95

Niger-Kordofanian 9. Remi Sonaiya Wh-Movement and Proper Government in Yoruba

109

10. Ronald P. Schaefer Typological Mixture in the Lexicalization of Manner and Cause in Emai

127

11. G. Tucker Childs The Extra-High Tone of Kisi: Just Another Tone or a New System of Prominence?

141

12. John Goldsmith, Karen Peterson, and Joseph Drogo Tone and Accent in the Xhosa Verbal System

157

13. José I. Hualde Double Object Constructions in KiRimi

179

14. Rick Treece Kiswahili Agreement for Kinship Terms

191

15. Bebra

Spitulnik

Levels of Semantic Structuring in Bantu Noun Classification

.

207

Nilo-Saharan 16. Barbara Levergood Arusa (Maa) Phrasal Tonology

221

Creoles 17. Rose-Marie Déchaine An Account of the Serial Verbs in Haitian Creole: Argument Structure Coalescence

237

18. John Victor Singler Topics in Liberian English Modality

253

Chapter 1

On the Nominal Morphology of 'Alienability* in Some African Languages Ulrike Claudi and Bernd Heine 1.

INTRODUCTION Morphosyntactic distinctions between "alienable" and "inalienable" concepts

have been observed in many African languages. According to a typological survey carried out within the Cologne project on language universals (see Reh, Heine, and Lamberti 1982), more than half of the 153 languages studied show such a distinction in some way or other. In Hyman, Alford, and Akpati (1970:85) a "universal semantic dichotomy" has been proposed with regard to relational concepts such as body parts - and perhaps directionals and kinship terms - on the one hand and non-relational concepts on the other, the first referring to "inalienable" and the second to "alienable" or "non-inalienable" possession. While it is obvious that there exists some connection between the relationality of nominal concepts and their specific morphosyntactic behaviour, it seems doubtful whether it is a semantic differentiation between relational and non-relational concepts involving features like [+INALIEN] and [-INALIEN] (see Hyman, Alford, and Akpati 1970) which accounts for different types of possessive constructions. Among the many types of nominal possessive constructions found in African languages, two appear to be particularly widespread, one exhibiting a juxtaposition of the possessor and the possessed NP without any kind of formal linking, and another containing some linking element or structure explicitly marking the relationship between the two NPs involved. In most cases, inalienability provides the morphologically unmarked structure, whereas "expressions manifesting the establishing principle ('alienable') were found to be longer, more complex, more explicit, both formally and semantically" (Seiler 1983:68). In fact, although inalienability can manifest itself in a number of morphosyntactic ways (e.g., in the obligatoriness of possessor presentation, accessibility to possessor promotion, etc. (cf. Seiler 1983:69)), it tends to be coded without any linker in nominal possessive constructions. Exceptions to this general rule are rare. As has been repeatedly noted (Reh, Heine, and Lamberti 1982; Seiler 1983:68), Dizi, an Omotic language of Western Ethiopia, provides one of them (cf. also Allan 1975/76). As shown in Claudi and Serzisko (1985), however, it

4

U. Claudi and B. Heine

is unwarranted to cite Dizi as a counterexample.

2.

CONVERGENT STRUCTURES One peculiarity concerns a striking instance of "convergence" between differ-

ent languages with regard to the way alienable possessive constructions are formed. Two examples may suffice to outline the problem involved, one relating to Kabiye, a West African, and the other to Acholi, an East African language. 2.1 Kabiye In Kabiye (Kabre, Tem)1 a Gur language of Niger-Congo spoken in Northern Togo, there are two types of genitive constructions. These differ from one another essentially in that one uses a genitive particle te between the possessor and the possessed noun phrase, while the other does not, the two noun phrases being simply juxtaposed.2 Note that Kabiye, an SVO language, consistently places the possessor before the possessed constituent.3 These constructions are employed, inter alia, to egress an alienable/inalienable distinction, as can be seen in (1).

(1)

a.

kolu te plya blacksmith of children 'the blacksmith's children (e.g., those living in his compound, but not his own) '

b.

kolu piya 'the blacksmith's own children'

Among the various characteristics that te constructions have, the following deserve mention: (i)

The use of this construction is ruled out when inalienable concepts

are to be expressed, e.g., when the possessed phrase consists of a body part, kinship or relational noun. (ii)

This construction may only be used with human possessor nouns.

(iii) The genitive particle is "homophonous" with the locative postposition te 'at, to' and the relational noun -te 'home', as can be seen in sentences like (2) and (3).

"Alienability" in Some African languages (2)

5

ma a -woki man-tjani té I:NEG-go my -ahe:friend to 'I don't go to my girl friend'

(3)

pe -te we cjéu their-home be beauty 'their home is beautiful'

Among young Kabiye speakers, especially those living in Lomé, the Togolese capital, there is a tendency to generalize the té construction, i.e. to eliminate the constraints expressed in (i) and (ii) above and to use té as a general genitive marker (Lébikaza, p.c.). 2.2 Acholi A strikingly similar situation is found in Acholi, a Western Nilotic language spoken in Northern Uganda, and this in spite of everything else that separates these two languages. Kabiye is a West African language while Acholi is an East African language; Kabiye belongs to Niger-Congo, but Acholi to Nilo-Saharan; Kabiye has a possessor-possessed syntax while the order in Acholi is possessedpossessor; and Kabiye is a gender language while Acholi is not. With regard to Kabiye genitive constructions, the similarities can be summarized thus: (i)

Acholi has two different genitive constructions, which differ from one

another in that one uses a genitive particle (pà) between the possessed and the possessor noun phrase, while the other simply juxtaposes the two phrases.1* As in Kabiye, these constructions express a distinction alienable/inalienable where the marked form denotes alienability, e.g. (4)

a.

wiic pà làtëën head of child 'the head (e.g. sheep's head) of the child'

b.

wii làtëën 'the child's (own) head' (Crazzolara 1955:47)5

(ii) The pà construction is used only with human possessor nouns (see Crazzolara 1955:46).6 (iii) pà is invariably referred to by all writers on Acholi grammar as a "preposition". We have not been able to find convincing evidence of a prepositional use of it. There is, however, a noun paàco in Acholi meaning 'village,

6

U. Claudi and B. Heine

homestead'. This noun may be reduced to pa in what Savage (1956:54) refers to as "geographical" terms, e.g. (tones not provided), (5)

a.

pa neya home uncle

'uncle's home'

b.

pa yweca

'my sister-in-law's home'

Note that in phrases like (5) involving pa as the head noun there is no linking particle. That Acholi pa is in fact derived from the noun paaco is also suggested by the fact that this noun has a suppletive plural form in the languages closely related to Acholi, e.g. (6)

Luo:

paco, pi. mier 'homestead, village' (Stafford 1967:125)

Shilluk:

pac , pi. mySr 'village' (Heasty 1937:80)

In Alur, this suppletive plural has been retained in pronominal possessive constructions: The marker pa of Alur has a suppletive plural form mi (see Ringe 1948:6), e.g. (7)

a.

tòm pa-r -e 'his chair' chair of-GEN-s/he

b.

tòni ml-gì chair of-they

'their chair'

The form mi shows exactly the same kind of word-final erosion (see Heine and Reh 1984:21) as pa: both have been reduced to a segment consisting of one open syllable only.

3.

FROM SPACE TO POSSESSION The possessive structures of Kabiye and Acholi are similar to a degree that

would seem to require some explanation. Historical relationship can be ruled out: The two languages belong to different families,7 and the areas where they are spoken are separated by several thousands of miles. Rather, it seems that there is another source which is responsible for the structural similarity between Kabiye and Acholi. This source relates to the way possession is concept-

"Alienability" in Some African Languages

1

ualized and linguistically encoded. As has been demonstrated with reference to predicative expressions of 'to have' in African languages (Claudi n.d.), possession does not appear to constitute an independent domain; rather, it tends to be conceptualized in terms of other cognitive domains. Very often, it is space which seems to be employed to render possessive notions. Underlying the POSSESSION IS SPACE metaphor, as it has been called in Claudi and Heine (1985:34), there appears to be an assumption to the effect that what is at someone's place is at one's disposal. The conceptual transfer from SPACE to POSSESSION has a number of implications for the grammar of the language affected. First, it constitutes a long, complex process. For example, during the initial phase the spatial and the possessive interpretations co-exist side by side, i.e. there is some ambiguity as to whether a given phrase is to be understood in its spatial or its possessive sense. Second, some of the characteristics of the spatial/locative structure tend to survive even after the transfer has been concluded. The result is that the relevant possessive construction may exhibit some characteristics which are meaningless on synchronic grounds and may be understood only with reference to its historical source. Finally, the already existing possessive structure may be redefined in the light of a new, emerging possessive construction. It is such an evolution which likely accounts for the structural similarities between Kabiye and Acholi. In both languages a nominal 'home, homestead' was used to introduce a possessive concept, in accordance with a "metaphorical equation of the kind "what is at my home belongs to me". This metaphor may be held responsible in particular for the following features encountered with the resulting possessive forms: a.

Since the nominal 'home, homestead' is associated with human beings, the

use of this nominal as a grammaticalized possessive marker is confined to persons. b.

The locative notion "what is at my home" implies that the item referred

to is not an inherent part of myself, it can be separated from me as when I leave my home. The possessive construction introduced by the POSSESSION IS SPACE metaphor therefore is likely to be of the alienable type. c.

With the emergence of a new possessive construction typically associated

with alienable concepts, the already existing pattern of expressing nominal possession tends to receive a specialized function, especially that of referring to inalienable concepts such as body parts, kinship relations and other relational entities, thereby leading to a grammatical distinction alienable versus inalienable.

8

U. Claudi and B. Heine

4.

SOME ADDITIONAL EVIDENCE Examples like the ones discussed in section 2 are not isolated cases. A genit-

ive marker which can be traced back to a noun belonging to the same cognitive domain as in Kabiye and Acholi via the POSSESSION IS SPACE metaphor can be found in some Central Sudanic languages as well. Tucker (1940:73), for example notes, that in Bongo "the presence"or absence of a linking particle depends upon the degree of intimacy in the possession". Thus, there is a linker 6a which is omitted in what Tucker refers to as the "possessive intimate", e.g. (8)

a.

soka fodu smoke fire

'smoke of fire'

(inalienable)

b.

bihi 6a nytre dog of chief

'the dog of the chief' (Tucker 1940:73)

(alienable)

According to Santandrea (1963:39), this rule is occasionally disregarded, e.g. when speaking of a man's relative. Similarly, in Mamvu the structure without a genitive linker is confined to the expression of "more intimate" relations, while the linker 6à occurs with "less intimate" ones: (9)

a.

mi-ngodo my-daughter

'my daughter'

(inalienable)

b.

mu-qaru my-foot

'my foot'

c.

ini -6a qlnga you(sg)-of bicycle

'your (sg) bicycle' (alienable) (Vorbichler 1971:229)

(inalienable)

When a kinship term is used with the 63 construction, it does not refer to a kinship relation, the relevant construction expressing alienability. Accordingly, a sentence like the following presupposes a context where for instance a child is brought back to the village it belongs to (Vorbichler 1971:230) : (10)

amu-b& ungu we of child

'our child'

In Mangbutu, it seems doubtful whether the genitive linker 6a can be omitted at all. The only examples found are:

"Alienability" in Some African Languages (11)

ktni 6a tibe -na chief of spear-DET 'chief's spear1

or

9

tibe-na keni 6a

(Tucker and Bryan 1966:56)

There exists a Bongo noun 6a 'country' (Vocabulary 1963:12), and throughout Tucker's Moru-Madi group, including Mangbetu and Lendu,® there is a noun 6a with the meaning 'house' or 'village'. Thus, a noun 6a is found in at least four out of seven sub-groups of Central Sudanic according to the Greenberg (1963:109) classification. In all cases this noun refers to a place where people live. Therefore, the "non-intimate", alienable construction can also be regarded as being based on the conceptual metaphor POSSESSION IS SPACE, i.e. "What exists where I am is mine". All examples available on these Central Sudanic languages show a human possessor. Furthermore, as one would expect on the basis of the Kabiye and Acholi data (see section 2), possessed nouns usually lack the genitive marker when body part nouns or kinship terms are involved, since a locative conceptualization would be redundant in the former and usually meaningless in the latter case. Whether the element 6a has developed into a general marker of possession in Mangbutu remains to be investigated. The seemingly odd exception noted by Santandrea concerning the occurrence of the "non-intimate" construction when speaking of a man's relatives could possibly be explained by referring to the origin of this construction: Since in a patrilocal society the relatives of a man are supposed to live at his place rather than somewhere else, a locative association between a man and his relatives is not meaningless in this case.

5.

INALIENABILITY AS THE MARKED STRUCTURE?

THE CASE OF DIZI

It has been pointed out repeatedly in works on nominal possession in African languages that the alienable/inalienable distinction tends to be paralleled by one of morphological marking: alienability is associated with the marked and inalienability with the unmarked structure. In sections 2 and 3 an attempt was made to account for such a situation. There are, however, a few languages exhibiting the opposite situation. Dizi, alternatively called Maji, an Omotic language of Western Ethiopia, has been claimed by Allan (1975/76) to be one of them. In certain contexts, inalienable nominal concepts like body parts are marked by placing an element kii after them, while alienable concepts are

10

U. Claudi and B. Heine

characterized by the lack of such a morpheme, e.g. (12)

a.

dad-h -kh bo -3 -s zuutu-no boy-DEF-? liver-DEF-ACC kick -lsg 'I kicked the boy's (personal) liver'

b.

dad-h bb-h-s zuutu-nO 'I kicked the boy's liver (i.e. an animal's liver which, for instance, he was about to eat)'

Further analysis indicates that the Dizi case is much more complicated and that inalienability does not form a crucial aspect of constructions involving the marker kh (see Claudi and Serzisko 1985). There is evidence to show that peculiar cases of marking possession in languages like Dizi can be accounted for by using a dynamic approach which incorporates both psychological factors, such as conceptual displacement, and diachrony. As in Kabiye, Acholi, etc., the Dizi case is one which essentially involves a transfer from a spatial domain to one which includes possession as one of its manifestations.® Basically, the marker kh can be regarded as a locative case particle, as it is used in sentences like the following: (13)

a.

wuurzln Inc -3 -kh A2 -de -ge cat tree-DEF-LOC sit-PRGf-3sgf 'the/a cat is sitting at/near/by the tree•

b.

d&d-k -kh h -sOr -a -s zoaagr3-no boy-DEF-LOC lsgPOSS-trousers-DEF-ACC tear -lsg 'I tore my trousers at the boy's (place)1

The rather vague locative relation expressed by kh can be specified by the additional use of postpositions, e.g. (14)

a.

wuurein inc -k -kh h&cit A1 -de -ge cat tree-DEF-LOC on sit-PRGf-3sgf 'the/a cat is sitting on the tree'

b.

11 -a -kh tsant maakin aaz -de -ge house-DEF-LOC before car stand-PRGf-3sgf 'before the house the/a car is standing'

The locative function of the element kh is also responsible for its occurrence in 'have'-constructions:

"Alienability" in Some African Languages (15)

a.

11

f) -uu-kO -kft dadv Ss-ii-go10 lsgPOSS-uncle-LOC child ? -be-3sgra 'my uncle has a child/children' (Lit., "there is a child/are children at/near/by my uncle")

b.

maakln inv-kh Bs-ki-ge car I -LOC ? -be-3sgf 'I have a car' (Lit., "there is a car at/near/by me")

This locative construction is employed for the purpose of possessor promotion. In such constructions, the possessor is encoded as a verbal complement other than the direct object. This not only means that the verb has to be an "affective verb" (cf. Voeltz 1976:260) but also that there exists the logical possibility of the promoted possessor being affected by the action. This possibility is ideally met in the case of actions directed to body parts since the "owner" of the respective body part is necessarily affected by such an action. It is exactly this type of context which explains the conceptual base for employing a locative phrase for the purpose of possessor promotion: The "possessor" is perceived as an "animate locality" where the action takes place. Accordingly, (16a) and (16b) differ in terms of "foregrounding", i.e. affectedness, rather than in terms of (in)alienability: (16)

a.

dad -Sn dcid-a gel -a -s cuur-e child-f boy-DEF head-DEF-ACC wash-3sgf 'the girl washed the boy's head'

b.

d&d-Sn dkd-ci-kh gtli cuur-e 'the girl washed the boy the head'

In cases where the action does not affect the direct object in any way, the possessor is not affected either and, hence, possessor promotion is ruled out: (17)

a.

yaab person-f

saar-a -s utts-do -go hair-DEF-ACC like-PRGm-3sgm

'he likes the woman's hair' b. *yaab-en-kn saarv uttB-dO-gO The promoted construction is clearly preferred, on the other hand, when actions involving physical violence are expressed:

12

U. Claudi and B. Heine

(18)

a. ?yaab-a aasO-3 -s caan -ki -no man -DEF leg -DEF-ACC break-PRF-lsg 1

b.

1 have broken the man's leg1 yakb-a-kh 5as0 caan-ki-no

'I have broken the man the leg' Now we come to the point which is immediately relevant to our discussion: There are certain contexts where the pragmatic dichotomy between promoted and non-promoted constructions has semantic implications. These are cases of possible ambiguity, as in (19):

(19) [

a.

=(12)

1 b.

dad-k-kh bo-k-s zuutu-no dad-k bd-k-s zuutu-no 'I kicked the boy's liver'

In (19a) the boy's liver is the goal of the kicking, the boy naturally being affected by it, while in (19b) it will be an animal's liver which the boy perhaps wanted to eat. The only means of distinguishing these two meanings is by using kh in (19a). Thus, taking (19) as an isolated example, kh could be interpreted as a possessive marker indicating inalienability. But (19a) is not an inalienable construction, and Allan's claim (1975/76) that kh is an inalienability marker has to be rejected. What appears to happen in this language is that an essentially locative structure, NP + kh, undergoes a conceptual transfer of the following kind: It is used for possessor promotion and reanalyzed as an oblique case expression marking "affectedness" when used with affective verbs. Dizi offers an example where a case marker employed for possessor promotion might develop into a marker of possession, and it is typically inalienable notions like body parts which tend to be concerned in such cases. An evolution of this type, if it indeed takes place, would offer an example of syntactic reanalysis (cf. Heine and Reh 1984:95-111), where clausal syntax develops into nominal syntax, i.e. two complement phrases governed by one and the same verb come to be reinterpreted as one complex noun phrase. 11

6.

CONCEPTUAL DOMAINS OF INALIENABILITY In the preceding sections the discussion was presented as if "inalienability"

formed a monolithic domain of conceptualization which turned out to be treated

"Alienability" in Some African Languages

13

similarly in all those languages which have a morphological distinction for it. This is by no means the case: Languages differ considerably with regard to what they treat as inalienable, and it remains doubtful whether there is any justification at all for assuming that there exists a cross-linguistic category of "inalienability". On the basis of typological observations made in African languages we may say that this term essentially relates to three different kinds of relational concepts which are described in Reh, Heine, and Lamberti (1982:5) as being either (a) socially determined, (b) partitive, or (c) localizing. Socially determined inalienability involves a social or kinship relationship and a syntactic context where both the possessor and the possessed NPs are typically human and denote physically separate but genetically and/or socially related entities. Apart from kinship roles it may include concepts such as 'friend', 'neighbour', 'chief/leader', 'teacher'. Partitive inalienability refers to part-whole relations which are perceived as being physically or otherwise inseparable. Body parts represent typical instances of partitive inalienability, other examples being concepts typically requiring either an animate possessor like 'name', 'voice', 'property', or an inanimate possessor like 'door (of house') or 'branch (of tree)'. Physical inseparability is even more strictly observed in the case of localizing inalienability, which involves spatial-relational concepts like 'top, up', 'bottom, below', 'back, behind', and may acquire a temporal, manner or other significance via metaphorical transfer. For the purposes of the present discussion we will refer to these three types of inalienable concepts by their most prototypical instances, i.e. kinship, body part, and relational-locative concepts, respectively. As was observed above, there are considerable differences between languages as to which of these domains, and which individual concepts, are treated morphologically as inalienable. In Kabiye and Acholi, for example, all domains mentioned above are associated with inalienability morphology; but a more peculiar situation of inalienability marking is found in languages like Ewe (Gbe), a Kwa language spoken in Southern Togo and Southeastern Ghana (see Claudi and Heine 1985). Whereas kinship terms and locative concepts are treated as inalienable, i.e. the possessor precedes the possessed NP without any linking device, body parts receive the "alienability" morphology, i.e. a genitive marker

is in-

serted between the possessor and the possessed NP. As in other cases discussed above, the POSSESSION IS SPACE metaphor can be held responsible for this situation. In accordance with this metaphor, the relational noun $e 'place' was used as a vehicle to denote possession and developed into a marker of nominal possession.1^ Thus

came to mark body parts, as in (20) , in accordance with the con-

14

U. Claudi and B. Heine

ceptual equation "What is at my place is in my possession/is part of me". (20)

Kofi te ta' Kofi of head

'Kofi's head'

There are obvious reasons, on the other hand, why ie did not spread to kinship terms and locative concepts: Kinship relationship typically does not imply a locative notion, that is, the king's brother is not necessarily associated with the place where the king is or lives. Thus, in genitive constructions having a kinship term as their head, the POSSESSION IS SPACE metaphor is not made use of, and hence

does not appear in such constructions. Quite a different situation

is found with relational-locative concepts. Nouns denoting such concepts already contain a locative reference, and there is a rule in Ewe according to which two relational notions must not immediately follow each other (cf. Westermann 1907: 54). The use of the above metaphor, and hence the use of $e, was therefore ruled out since it would have resulted in an ungrammatical sequence of two locative nouns. Finally, one may wonder why with body part nouns one and the same categorial metaphor, POSSESSION IS SPACE, is responsible for an alienable morphology in Ewe but an inalienable morphology in languages like Kabiye and Acholi. It would seem that this is due to the particular conceptual metaphors applied.^ The concept PLACE used in Ewe is compatible with any locative notion, whereas 'home', figuring in Kabiye and Acholi, is hardly compatible with body parts. It makes sense metaphorically to refer to 'John's spear' as "the spear at John's home" but much less so to denote 'John's head' as "the head at John's home". On the basis of evidence such as the above, there is reason to assume that synchronic semantics is not sufficient to account satisfactorily for the nature of "inalienability" in African languages.

7.

CONCLUSION It would seem that examples like the ones presented above have implications

for a theory which aims at understanding, rather than merely describing, language structure. Such implications are in particular: 1. Grammatical categories of the kind discussed here are the result of cognitive processes such as conceptualization and transfer from one cognitive domain to another. For example, the conceptual shift accompanying metaphorical transfers appear to be responsible for the re-interpretation of "lexical" as

"Alienability" in Some African Languages

15

"grammatical" concepts (see Claudi and Heine 1985). 2. A rigidly synchronic analysis is inadequate for capturing some essential characteristics of language structure, especially those relating to the dynamic nature of grammar. 3. An understanding of this dynamic nature leads to observations like the following: Taxonomic structures such as word classes or syntactic categories are to some extent artificial entities; there is a lot of linguistic behaviour taking place between, rather than within, such classes or categories. Furthermore, established terms of linguistic nomenclature, such as "selection restrictions", "homophony", "polysemy", and the like may be empty labels which tend to hide more than they reveal. Rather than describing the relationship between linguistic structures in terms of taxonomic statements like "X is a Y" or "X and Y are mutually exclusive", it would seem more profitable in many cases to use evolutional predicates such as "X is derived from Y". 4. A linguistic analysis which is confined to synchronic observations may lead to inadequate results by ignoring certain asymmetries which form an inherent part of linguistic evolution. The fact that conceptual shifts precede associated morphosyntactic and phonological shifts, for example, may create a situation where there is no longer a one-to-one correspondence between the relevant conceptual structure on the one hand and its morphosyntactic or phonological equivalent on the other. The main purpose of the present paper is to point out an area where linguistics might profit from including some explanatory parameters which hithero have either been ignored or declared irrelevant to linguistic description. ^14 It remains to be investigated how, and to what extent, such parameters should be incorporated within a theory of grammar. The morphology of "inalienability" provides an example of the kind of problem involved. In the preceding section it was demonstrated how conceptual transfer can be held responsible for the peculiar behaviour of "inalienability" in Ewe. This transfer also accounts for another characteristic of Ewe grammar: Genitive constructions involving verbal nouns as heads are treated as "alienable" when the possessor is the semantic agent (= genitivus subiectivus), but as "inalienable" when the possessor represents the semantic patient (= genitivus obiectivus), e.g. (21)

a.

ame lo-lo person of loving

(genitivus subiectivus) 'a person's love1

16

U. Claudi and B. Heine b.

ame lo-lo person loving

(genitivus obiectivus) 'loving a person'

(Westermann 1907:93)

In the former case, 'love' is conceived of as being at the place where the person feeling it is, hence the "alienability" morphology. In the latter case no spatial notion is involved: what is expressed is the object of 'love', rather than where love is located. The POSSESSION IS SPACE metaphor therefore was applied in phrases like (21a) but was ruled out in cases like (21b). Thus, "inalienability" in Ewe cannot be defined in terms of the semantics of possession; rather, it is the result of metaphorical usage. Whenever the POSSESSION IS SPACE metaphor applies we are dealing with "alienable" possession; elsewhere we are confronted with an odd class of "inalienable" concepts. On the other hand, an equally plausible account may be offered by having recourse to descriptive synchronic techniques. On the basis of prevalent constituent analysis of the kind presented in (22) , a verb is syntactically more closely related to the sentence object than to the subject. (22)

S

NP

V

NP

This can be related to the fact that "inalienability" expresses a closer semantic relationship between the possessor and the possessed than "alienability". Assuming that there exists some inconicity between morphosyntax and semantics, it is not surprising then that the distinction verb-object vs. verb-subject is encoded morphosyntactically in the same way as the semantic distinction "inalienable" vs. "alienable", i.e. that a genitivus obiectivus receives the "inalienable" and a genitivus subiectivus the "alienable" morphology.15 Thus, in this case the morphology of the two genitive constructions concerned can be interpreted equally by means of a dynamic-conceptually oriented approach as by means of a synchronic morphosyntactic analysis. It is unclear how in such cases processes like conceptual transfer and grammaticalization are related to the synchronic logic of the speech community concerned. For the time being we assume that the latter provides the necessary frame-work for the former to operate, but much further research is required on this issue. Nevertheless, we hope that the preceding discussion has made it clear that synchronic logic on its own is not a sufficient basis for understanding grammar.

"Alienability" in Some African Languages

17

ABBREVIATIONS Accusative case Definiteness Determination Feminine gender Genitive case Locative case Masculine gender

ACC DEF DET f GEN LOC m

NEG POSS PRF PRG

Negation Possessive pronoun Perfect aspect Progressive aspect

1 3

1st person 3rd person

NOTES *This paper is based on research carried out within the project "Grammatikalisierung afrikanischer Sprachen". We wish to express our gratitude to the Deutsche Forschungsgemeinschaft (German Research Foundation) for having sponsored this work. Furthermore, we are obliged to Mechthild Reh and Fritz Serzisko for valuable suggestions. 1.

The following discussion on Kabiye is entirely based on Lebikaza (1985).

2.

The matter is slightly more complicated. There is an underlying possessive pronoun between these two noun phrases which is coreferential with the possessor noun, and which is reduced to a non-segmental morph, a low tone, unless the possessor noun is followed by a demonstrative or relative clause marker (see LSbikaza 1985:269-271). We may ignore this for our purposes.

3.

Like all other Gur (Voltaic) languages, Kabiye belongs to type B languages on account of its word order characteristics (see Heine 1976:4If.).

4.

In addition, there is a third construction in Acholi, involving the "preposition" mS (see Crazzolara 1955:45ff.; Malandra 1955:25f.; Savage 1956:13ff.).

5.

Concerning the variation between wiic and wii, see Crazzolara (1955:43).

6.

The following note can be taken to be more of a confirmation of this than anything else: "Some writers have stated that pa should only be used when the possessor is a person. This is not correct. It is true that possessors are usually persons, but in some cases animals may also be thought of as possessing things, and even inanimate objects.... The rule seems to be that pa must be used with personal possessors ...." (Savage 1956:71; emphasis added).

7.

According to Gregersen (1972), Niger-Congo and Nilo-Saharan belong to one genetic stock, and Kabiye and Acholi would therefore be genetically related. Even if such a hypothesis were shown to reflect historical reality, the chances that it accounts for such a structural similarity are remote.

8.

Lendu shows the form ba or 6a; cf. Tucker (1940:348f.; Tucker and Bryan 1966:31).

9.

The following account is based on Claudi and Serzisko (1985).

10.

Concerning as, see Claudi and Serzisko 1985:fn 5.

11.

Reanalysis of this type appears to have happened in certain varieties of non-standard German, e.g. the Cologne dialect, Where promoted dative case expressions as in (i) have developed into possessor NPs in genitive con-

18

U. Claudi and B. Heine structions, as in (ii): (i)

Ich habe meinem Freund sein Fahrrad repariert I have my:DAT friend his bike repaired 1

1 have repaired my friend his bike1

(ii) mingem Frund sing Fahrrad is kapott my:DAT friend his bike is broken 'my friend's bike is broken' 12.

This noun has also been grammaticalized to a kind of derivative suffix designating 'place of', e.g. do' -wo-i>5 work-do-place 'place of work'

13.

Concerning the difference between categorial and conceptual metaphors, see Claudi and Heine (1985).

14.

Concerning other, equally important parameters, see Giv6n (1979).

15.

Note also the following generalization proposed by Bird (1972:8): "Wherever alienable and inalienable distinctions are overtly made in the noun phrase, and where those same distinctions will be used to differentiate subjects and objects in nominalizations, it will always be the case that inalienable possession is structurally associated with the object and alienable possession with the subject".

REFERENCES Allan, Edward J. 1975/76. Inalienable possession in four Ethiopian languages. Afrika und Übersee 59:300-307. Bird, Charles S. 1972. The syntax and semantics of possession in Bambara. Paper given at the Conference on Manding Studies, London 1972. Claudi, Ulrike, n.d. To have or not to have: On the conceptual base of predicative possession in some African languages. Unpublished ms., Cologne. Claudi, Ulrike, and Bernd Heine. 1985. From metaphor to grammar: Some examples from Ewe. Afrikanistische Arbeitspapiere 1:17-54. [A revised and enlarged version of this paper will appear in Studies in Language 10 (1986).] Claudi, Ulrike, and Fritz Serzisko. 1985. Possession in Dizi: Inalienable or not? Journal of African Languages and Linguistics 7:131-154. Crazzolara, J.P. 1955. A Study of the Acooli Language. London: Oxford University Press. Givön, Talmy. 1979. On Understanding Grammar. New York: Academic Press. Greenberg, Joseph H. 1963. The Languages of Africa. The Hague: Mouton. Gregersen, Edgar A. 1972. Kongo-Saharan. Journal of African Languages 11 (1): 69-89. Heasty, J.A. 1937. English-Shilluk Shilluk-English Dictionary. Dolieb Hill: The American Mission. [Reprinted 1974]. Heine, Bernd. 1976. A Typology of African Languages Based on the Order of Meaningful Elements. (Kölner Beiträge zur Afrikanistik, 4.) Berlin: Reimer. Heine, Bernd, and Mechthild Reh. 1984. Grammaticalization and Reanalysis in African Languages. Hamburg: Buske. Hyman, Larry M., Danny Keith Alford, and Elizabeth Akpati. 1970. Inalienable

"Alienability" in Some African Languages

19

possession in Igbo. Journal of West African Languages 7:85-101. Löbikaza, KSziö Koyenzi. 1985. Phonologie, Tonologie und Morphosyntax des Kabiye. Ph.D. dissertation, University of Cologne. Malandra, Alfred. 1955. A New Acholi Grammar. 2nd ed. Kampala: The Eagle Press. Reh, Mechthild, Bernd Heine, and Marcello Lamberti. 1982. Inhärenz in Possessivkonstruktionen afrikanischer Sprachen. Unpublished ms., Cologne. Ringe, Peter C. 1948. A Simple Alur Grammar and Alur-English- Alur Vocabularies. Eagle Language Studies Series (Uganda). Santandrea, Stefano. 1963. A Concise Grammar Outline of the Bongo Language. (Museum Conibonianum, 14.) Rome: St. Peter Claver. Savage, G.A.R. 1956. The Essentials of Lwo (Acoli). Nairobi/Kampala/Dar es Salaam: East African Literature Bureau. Seiler, Hansjakob. 1983. Possession as an Operational Dimension of Language. (Language Universals Series, 2.) Tübingen: Narr. Stafford, R.L. 1967. An Elementary Luo Grammar With Vocabularies. Nairobi: Oxford University Press. Tucker, A.N. 1940. The Eastern Sudanic Languages. Vol. 1. London: Oxford University Press. Tucker, A.N. and M.A. Bryan. 1966. Linguistic Analyses. The Non-Bantu Languages of North-Eastern Africa. London: Oxford University Press. A Small Comparative Vocabulary of Bongo Baka Yulu Kara. 1963. Pome: St. Peter Claver. Voeltz, Erhard. 1976. Inalienable possession in Sotho. In Papers in African Linguistics in Honor of Wm. E. Weimers, ed. by L.M. Hyman, L.C. Jacobson, and R.G. Schuh, pp. 256-266. (Studies in African Linguistics, Suppl. 6.) Los Angeles: University of California. Vorbichler, Anton. 1971. Die Sprache der Mamvu. (Afrikanistische Forschungen, 5.) Hamburg/Glückstadt: Augustin. Westermann, Diedrich. 1907. Grammatik der Ewe-Sprache. Berlin: Reimer.

Institut für Afrikanistik Universität zu Köln

Chapter 2

Asymmetrical Syntax and Symmetrical Morphology in African Languages Linda Schwartz 1.

INTRODUCTION This paper addresses the question of how thematic coordination is expressed

morphosyntactically in human languages. 'Thematic coordination' is used here to refer to instances where more than one participant is understood to bear the same thematic relation to the state or event expressed syntactically in a single clause. I will not be concerned with other types of syntactic coordination such as sentential coordination or verb phrase coordination. African languages exhibit a striking areal characteristic in that many of them express thematic coordinations with what I will call 'asymmetrical syntax' supplemented by symmetrical morphology.

2.

SYMMETRICAL SYNTAX In languages like English, thematic coordination is typically expressed with

what I will call 'symmetrical syntax', as in (1). (1)

Sam and I are laughing

In this sentence, Sam and I are understood to be in the same thematic relation to laughing. 'Symmetrical syntax' is when the thematic coordination is expressed by a single continuous syntactic constituent (Sam and J is an NP), where each member of the thematic coordination is expressed as an independent syntactic constituent with the same morphosyntactic status as all other members, except for temporal/linear precedence (Sam and I are both independent NPs).

3.

ASYMMETRICAL SYNTAX IN AFRICAN LANGUAGES In this section, some examples of the range of asymmetrical constructions

expressing thematic coordination which are found in African languages will be examined, and a general typology of these will be provided.

22

L. Schwartz

3.1 NP-marking languages In Logbara, as shown in (2), the thematic coordination of hare and fowl1 is expressed by marking the first NP (hare) with a plural marker; fowl is not contiguous to the first. (2)

LOGBARA (Nilo-Saharan: C. Sudanic [Crazzolara 1960]) etoo pi mû dî a'uâ ie hare PL went then fowl & 'the hare and a small fowl went along together1

The structure is thus asymmetrical because the NPs representing the participants in the thematic coordination do not form a continuous syntactic constituent corresponding to the thematic coordination. In (3), the sequence twe na Yohani in Kirundi is asymmetrical because the pronoun twe 'we' is plural, even though the relevant interpretation identifies only two participants in the thematic coordination: John and J. (3)

KIRUNDI (Niger-Kordofanian: Bantu) twe na Yohani twagiye kwishure 1PL & Y lPL+go school 'John and I went to school'

Linguists describing such languages sometimes refer to the plural marking in such structures as "anticipatory" (Crazzolara 1969, Hoffmann 1963), because the plural marking on the NP or pronoun carries the number marking of the thematic coordination rather than the number marking of the first participant in the coordination. Thus, the participants here do not have equal syntactic status: one, John, is thematically and syntactically independent, while the other, I, is thematically independent but not syntactically so because it is absorbed into the plural representing the set whose members are John and J. In Schwartz (1985) I refer to structure of this type as "Plural Pronoun Constructions".2 Structures like those exemplified in Kirundi and Logbara will be referred to here as 'NP-marking structures' because the morphology which signals a thematic coordination interpretation is syntactically associated with an NP corresponding to one of the participants. The Plural Pronoun Construction like that in Kirundi appears to be very widespread; extensions to non-pronominal NPs like that in Logbara may be less common (at least, thus far I have found less information on them in grammars). In (4) is a list of African languages which have some variat-

Asymmetrical Syntax and Symmetrical Morphology

23

ion of the NP-marking structures. (4)

NP-marking Languages Nilo-Saharan: Acholi, Bari, Logbara, Nuer Niger-Kordofanian: Babungo, Diola-Fogny, Ewe, Kirundi, Kpelle, Mende, Sango, Vei Afroasiatic: Margi, Ngizim, Tera

3.2

Clause-marking languages

The structures in the examples given below show another asymmetrical means of expressing thematic coordination. In the Kanuri example in (5), the expression of the thematic coordination he and his son is syntactically asymmetrical because one participant, his son, is manifested as a free NP, while the other, he, is encoded only in the third person plural verbal morphology, so the morphosyntactic elements corresponding to the participants are neither a single constituent nor of equal syntactic status. (5)

KANURI (Nilo-Saharan: Saharan [Hutchison 1980]) tâdànzéa kâsho child+POSS+ASSOC came+PL 'he and his son came1

In the Tera and Chicaranga examples in (6) and (7), the NPs that represent participants in the thematic coordination again do not form a continuous syntactic constituent.3 (6)

TERA (Afroasiatic: Chadic [Newman 1970]) d'a vanda ndB woya Ali wà A PERF ran EMPH+PL & boy+DEF 'Ali and the boy ran away*

(7)

CHICARANGA (Niger-Kordofanian: Bantu [Louw 1915]) ha i siyani na'ye NEG 1 leave+RECIP &+3 •he and I do not leave one another'

In (6), the boy is interpreted as thematically coordinate with Ali because of the plural emphatic reflexive marker vanda. In (7), he and I are interpreted as

24

L. Schwartz

thematically coordinate because of the reciprocal marker on the verb. Languages with expressions of thematic coordination in asymmetrical constructions like those in (5)-(7) will be referred to as "clause-marking structures", because here the morphology which is crucial to the thematic coordination interpretation is not associated with an NP or sequence of NPs, but is either part of the verbal complex, as in Kanuri and Chicaranga, or may be found somewhere else in the clause, as in Tera (see further examples below). In (8) is given a list of languages which have constructions with some version of clause-marking structures. (8)

Clause-marking Languages Nilo-Saharan: Kanuri Niger-Kordofanian: Babungo, Chicaranga, Kirundi, Sotho Afroasiatic: Tera, Hausa

3.3

Typology of asymmetrical syntax for thematic coordination

The classification of the expression of thematic coordination with asymmetrical syntax presented here leads to the following kinds of languages: (9)

Type 1: languages with NP-marking only: Sango etc. Type 2: languages with clause-marking only: Hausa, etc. Type 3: languages with both: Tera, Kirundi, etc. Type 4: languages with neither: English (symmetrical)

It is quite possible that there is more overlap between Types 1 and 2 (that is, more languages in Type 3) than is apparent from the descriptions of the languages included in this sample. Before proceeding further, several points should be made about the distribution and function of these asymmetrical structures. First, as the lists in (4) and (8) indicate, these types of asymmetrical syntax associated with thematic coordination are widely distributed in Africa geographically and genetically. Second, syntactically asymmetrical expression of thematic coordination is not confined to the language families of Africa. Rather, its genetic and geographical distribution is very wide-spread. Within the Indo-European family, for example, it is found in the Slavic languages (Type 1: Russian, Type 3: Bulgarian) and in some Spanish dialects (Type 2: Chilean Spanish). It is found in American Indian languages of various families; the languages include Dakota (Siouan, Type 2) and Ojibwe (Algonquian, Type 1).

Asymmetrical Syntax and Symmetrical Morphology

25

It is present in Uralic-Altaic languages, including Hungarian (Type 3) and Mongolian (Type 1), and it appears to be quite common in Austronesian languages, as in Yapese (Type 3) and Tagalog (Type 1). Last, in terms of both function and structure, these constructions differ from comitative expressions. In general comitatives differ syntactically from the corresponding asymmetrical coordinations in that the "comitative" participant (that participant manifested as an NP appearing with a connector morpheme) is syntactically optional and when absent does not function as an "understood" semantic participant. In (10) and (11) are contrasted the asymmetrical constructions expressing thematic coordination with those expressing comitatives. (10)

HAUSA (Afroasiatic: Chadic) a.

mun jee kaasuwaa (taare) da k'aneenaa 1PL+PAST go market together & younger brother+lPO "my younger brother and I went to the market'

b.

naa 1+PAST

jee kaasuwaa (taare) da k'aneenaa

'I went to the market with my younger brother' (11)

TERA a.

Ali wk d'a vanda nds woya A PRF go EMPH+PL & boy+DEF 'Ali and the boy ran away'

b.

Ali wa

d'a varan nd3 woya EMPH+SG

'Ali ran away with the boy' In the Hausa example in (10a), the thematic coordination has a plural marker in the inflected auxiliary (glossed as PAST), while in the comitative in (10b) the inflected auxiliary is singular. In the Tera example in (11a), the emphatic reflexive is plural in the thematic coordination while in the comitative construction in (lib) the emphatic reflexive is singular.

4.

THE CONNECTOR MORPHEME

Some general characteristics of NP-marking and clause-marking constructions have been discussed in Schwartz (1985, in press a,b). One point which is of particular relevance here is the nature of the connector morpheme which is

26

L. Schwartz

associated with one of the NPs in these constructions. This connector is glossed as '&' in the examples, to overt connector is not found in all cases where the NPs corresponding to members of the thematic coordination are contiguous. It is not found in Plural Pronoun Constructions in Kpelle and Mende, for example, as shown in (12) and (13). (12)

KPELLE (Niger-Kordofanian: Mande [Welmers 1973]) ta Sumo 3 PL S 'he and Sumo'

(13)

MENDE (Niger-Kordofanian: Mande [Sumner 1917] ) taa ngi nyahzl1* 3PL 3-POSS wife •he and his wife'

However, whenever it does occur, it is not an exclusively coordinating conjunction in the sense of English 'and'.5 Rather, it is a general connector which functions to associate a noun phrase with a clause, and its interpretation includes comitative and generally includes instrumental, manner, sometimes possession, incidental association (like he came [with food] for the children), means (like fill the jar [with water]), or other adjunct roles, so it corresponds roughly to the usage of English 'with', with the specific relation subject to the context. Representative examples to illustrate the general range of the use of the connector are given for Tera in (14) (from Newman 1970) and for Margi in (15) (from Hoffmann 1963). (14)

TERA a.

woy wa d'B gar nds 'jim ye wara 'a boy went to the farm with water for them'

b.

tern 3 njezl bufu ncfc wada 'we are filling a sack with peanuts'

c.

Ali wa jokta nda nda slugu 'Ali stabbed him with a knife'

(15)

MARGI a.

ni ga shili iga Ssi'i dza makSr 'I came with a three fathom rope'

Asymmetrical

b.

Syntax

nhj

kgk

and Symmetrical

tlh

mota

k3 nylny

27

'ddi

•he has many cows' c.

Morphology

kga

(with many cows.) mji

'this car is full of people'

(filled-with)

Schuh (1972) suggests that there is a typological split between languages which have syntactic coordination and those which do not, and that those which do not will utilize other means available in the language to connect phrases. The distinction which I am making here between the use of symmetrical syntax vs. asymmetrical syntax to express thematic coordination would seem to be in the spirit of his suggestion, where in the languages in which thematic coordination is expressed through asymmetrical syntax, a general connector is being used which links NPs expressing a variety of different relations. Furthermore, there seems to be a natural relation between the use of this general connector and the fact that in many of these constructions there is no continuous constituent: this is simply not required by the syntactic device used, which in its other syntactic uses is not confined to [NP _ NP] structures. That is, given the general use of the connector, there is no reason to expect that the NPs being linked thematically should necessarily form a constituent. I am taking as evidence of (at least surface) non-constituency here the fact that other elements which are not part of the thematic coordination may intervene (and sometimes typically do; see remarks on Tera in note 3) between members of the thematic coordination. There are other tests for constituenthood which need to be explored more fully, such as potential for extraction. From what little information I have on this point, it is not clear that all of the languages in this sample will show similar behavior on this point. However, if the connector were all there is to the expression of thematic coordination in these languages, we might expect systematic ambiquity between those asymmetrical constructions expressing thematic coordination and those expressing comitative relations, but this does not generally occur. Rather, languages like those examined here employ additional morphological means to distinguish a thematic coordination interpretation. The next section will examine this crucial morphology.

28 5.

L. Schwartz SYMMETRICAL MORPHOLOGY Tera has at least three distinct ways of expressing thematic coordination.

These are illustrated in {16) — (18). (16)

tem wa d'3 Kanu ndS Dala 1PL PRF go Kano & D 'Dala and I went to Kano'

(17)

Ali wk d'B Kanu ku nd3 Dala PRF go Kano PL & D A 'Ali and Dala went to Kano1

(18)

Ali wk d'9 vanda ndB woya A PRF go EMPH+PL & boy+DEF 'Ali and the boy ran away'

In (16), there is a discontinuous Plural Pronoun Construction, and it is the plural marking on the first person plural pronoun which expresses exclusively thematic coordination. In (17), the morpheme ku, which is the normal plural marker for nouns, occurs after the verb and its locative goal, again indicating a thematic coordination. It has already been shown that Tera also uses an emphatic reflexive marker for this purpose, as in example (11a) repeated here as (18). It is the fact that the emphatic reflexive is marked for number which is significant in this construction as well, as is shown by the contrast with the comitative structure in (lib). In all these examples, then, it is plural morphology which ensures than an asymmetric syntactic construction will receive an interpretation of thematic coordination. This would seem to be very natural, since the semantics of plural morphology is essentially to signal that the same thematic relation is distributed over a set of participants. Another type of morpheme which is used to indicate unambiguously a thematic coordination interpretation is the reciprocal marker, as shown in the example in (7) for Chicaranga and in the Sotho example in (19). (19)

SOTHO (Niger-Kordofanian: Bantu [Doke & Mofokena 1957]) ke-tsebana Ie-uSna 1-know+RECIP &-2 'you and I know each other'

Asymmetrical Syntax and Symmetrical Morphology

29

Although Doke and Mofokena also give examples where a plural subject marker on the verb as well as the reciprocal morpheme indicate thematic coordination, in (19) this interpretation comes from the reciprocal marker alone. Reciprocal morphology, like plural morphology, distributes thematic relations over sets of participants. The semantics of reciprocal morphology is of course more complex than that of plural morphology. It is interpreted to mean that each participant in the set bears two thematic relations and that these relations are reversed pairwise between members of the set (ignoring the more complex semantics of plurals in reciprocal constructions like 1

In other words, 'A & B V^-RECIP

1

the boys and the girls saw each other').

is equivalent to ' A V^ B and B V^ A'.

The general point to be made, then, is that there is a relation between the syntax and the morphology of these asymmetrical constructions used to express thematic coordination. Specifically, although the syntax of the construction is used generally to encode a range of relations besides thematic coordination, morphological markers are available which are used to express that sets of participants are assigned a given role, supplementing the general syntax of these constructions by providing a uniquely asymmetrical interpretation. While the relation between plural marking, including reciprocals, and thematic coordination seems like a very natural one, given the semantics of plurals, it is apparently not the only device used in these constructions. In Babungo (Schaub 1985) , Imperfective aspect used narrative internally indicates change of actor/subject. Consider now a sentence like (20). (20)

BABUNGO (Niger-Kordofanian: Bantu [Schaub 1985]) Lkmbl tli wuumba wi, ghD nws gS taa muu L called+PERF friend his, & 3 went+IMPEKF to river •Lambi called his friend, and they went to the river'

In the first clause Lambi and his friend are both mentioned, but Lambi alone is the actor/subject. If Lambi alone were the actor/subject of the second clause, the verb should be in the perfective. The reason that it is not in the perfective must be because the subject is not Lambi alone but is Lambi and him. Thus, again, the language uses a clause-marking device, verbal morphology, to enforce a thematic coordinate interpretation on a more general syntactic structure.

30 6.

L. Schwartz PRONOUNS The discussion here of the interaction of the syntax of constructions ex-

pressing thematic coordination, which is general and asymmetrical, and the specific symmetrical morphology does not explain a number of significant characteristics of these constructions. One very salient characteristic is the affinity of these constructions for pronouns. For example, the Plural Pronoun Construction as found in the Kirundi example in (3) is often found when there is no corresponding general (non-pronomomal) NP-marking morphology to support the thematic coordination interpretation.6 In Schwartz (in press a) I suggest that motivation for this affinity to pronouns in NP-marking languages may have to do with the function of first and second person plural pronouns. These pronouns identify heterogeneous sets including one speech act participant and one or more other members not of this status. Thus, a first person plural identifies a set such that one of its members is the speaker, and a second person plural identifies a set such that one of its members is the hearer. This is significantly different from the typical function of plural morphology on nouns, which refers to groups of heterogeneous members.7 That is, in the case of a plural noun like 'cats', each member of the set specified by this noun must be a cat, where in the group identified by 'us', one member must be the speaker, but the others are not. Thus, the plural marking of the initial pronoun of the Plural Pronoun Construction is not just a feature of the construction attached in an arbitrary location within the construction. Rather, plural first and second person pronouns are used independently to specify heterogeneous sets, and their function in Plural Pronoun Constructions can be seen as a straightforward extension of their typical use. A different kind of extension of this is the system like that of Logbara in example (2), where NPs other than pronouns are marked as plural to encode a thematic coordination interpretation. In Schwartz (in press b) I address the question of pronominal affinity with respect to that set of constructions in which the set-marking (i.e. the plural marking which indicates that the thematic relation is held by a set of participants) occurs on the verbal complex, as in the Hausa example in (21). (21)

mun jee kaasuwaa (taare) da kai 1PL+PAST go market together & 2 'you and I went to the market'

Asymmetrical Syntax and Symmetrical Morphology

31

In (21) , the plural marking is on the inflected auxiliary (glossed as PAST) and is not a free pronoun. The use of an asymmetrical construction with plural verbal morphology in a language with free pronouns (such as Hausa) has two effects. First, these constructions register the thematic coordination in the clause by the verbal marking (or marking elsewhere in the clause, as in Tera), and second, they avoid a symmetrical syntactic structure with pragmatically imbalanced conjuncts, where a free pronoun (as opposed to the verbal marking only) is used only for focus. It is thus the general syntactic properties of languages with verbal encoding of pronominal information, specifically, the fact that free pronouns are not obligatory and are pragmatically very salient which may disfavor a mapping between thematic coordination and symmetrical syntax in cases where one of the participants is expressed pronominally and is thus inherently highly focussed.

7.

CONCLUSION The individual constructions cited in this paper all require a much more

detailed investigation into their syntax and semantics and into the interaction of the morphological devices in those languages where more than one is available. However, I hope to have made the general point that the asymmetrical constructions examined here represent a trade-off between syntax and morphology, where the syntax is underspecified with respect to the expression of a certain concept, namely, thematic coordination, while the morphology not only specifies more fully the expression of this concept but does so by utilizing devices which are already present and whose use in these constructions can be seen as a natural extension of their use in other constructions in which more than one participant is in the same role.

NOTES *I am grateful to Ismail Junaidu and Sammani Sani for their help in providing the Hausa data, to Firmard Sabimana for the Kirundi data, and to Gerald Sanders for his comments on an earlier draft of this paper. 1.

I will indicate the parts of a thematic coordination by using italics in the English gloss.

2.

Before leaving this construction, it is important to make a general point about it. Due to morphological issues to be discussed later, it is crucial in all of these examples that there be an interpretation where the thematic

32

L. Schwartz coordination has exactly two members, like Sam and I. In many of the languages which have these constructions, such structures could also have a nondual plural interpretation like we and Sam. Various morphological devices can be used to distinguish the dual and plural readings in some languages; some of these devices are identified in Schwartz (1985). It is dual interpretation which is crucial because it shows that the syntactic plurality of the first pronoun in the construction is due to the number of the set representing the thematic coordination rather than to a plurality that does not necessarily include the other member.

3.

Information is not readily available in all grammars regarding the syntactic separability of NPs in thematic coordinations. Newman reports that in Tera, forms like (8) are preferred except under certain syntactic conditions such as heavy predicates or potential ambiguity.

4.

This example is particularly interesting because the singular possessive pronoun shows clearly that the number of the first pronoun in the construction is syntactic rather than semantic.

5.

Vei may be an exception to this. Koelle (1854) gives two conjunctions, be and féra, which he claims are identical in force, with the distinction that féra conjoins only two participants. Both are used in Plural Pronoun Constructions: mú be ánümu 'it is I and they' (lit. 'we and they'); wú féranú 'you (sg.) and they' (lit. 'you pi. and they'). No more general usage than this usage as an NP coordinating conjunction is indicated for either morpheme.

6.

It should be mentioned, though, that this does not mean that there can be no morphological marking for thematic coordinations in these languages for non-pronominal NPs -- there can be, but it is on the clause rather than on the NP. (Recall, however, that languages may have NP-marking but not clausemarking. )

7.

In Logbara, however, plural marking on nouns could be used to encode thematic coordination, so it is not the case that the plural marking on nouns can have only the homogeneous interpretation mentioned here. The point is that the homogeneous interpretation is its typical use, while the heterogeneous interpretation is the typical one for first and second person plural pronouns.

REFERENCES Crazzolara, J.P. 1969. A Study of the Logbara Language. London: Oxford University Press. Doke, Clement, and S. Mofoke'na. 1957. Textbook of Southern Sotho Grammar. Johannesburg: Longman. Hoffmann, Carl. 1963. A Grammar of the Margi Language. London: Oxford University Press. Hutchison, John. 1980. The Kanuri associative postposition, Studies in African Linguistics 11:321-351. Koelle, S. 1854. Outlines of a Grammar of the Vei Language. London: Church Missionary House. Republished in association with the African Languages Review of the Fourah Bay College. Luow, C.S. 1915. A Maunual of the Chicaranga Language. Bulawayo: Philpott and Cottins. Newman, Paul. 1970. A Grammar of Tera. Berkeley & Los Angeles: University of

Asymmetrical Syntax and Symmetrical Morphology

33

California Press. Schaub, Willi. 1985. Babungo. London: Croom Helm. Schuh, Russell. 1972. Aspects of Ngizim Syntax. Ph.D. Dissertation, UCLA. Schwartz, Linda. 1985. Plural pronouns, coordination and inclusion. In: Proceedings of the Tenth Annual Minnesota Conference on Language and Linguistics, ed. by Nancy Stenson, Minneapolis: University of Minnesota. Schwartz, Linda. In press a. Asymmetric feature distribution in pronominal coordinations. In: Agreement in Natural Languages, ed. by M. Barlow and C. Ferguson, Stanford: CSLI. Schwartz, Linda. In press b. Conditions for verb-coded coordinations. In: Studies in Syntactic Typology, ed. by M. Hammond, E. Moravcsik and J. Wirth. Philadelphia: John Benjamins. Sumner, A. 1917. A Hand-book of the Mende Language. Freetowh: Government Printing Office, Sierra Leone. Weimers, William. 1973. African Language Structures. Berkeley and Los Angeles: University of California Press.

Department of Linguistics Indiana University

Chapter 3

Long Vowels and Diphthongs in Miya and Hausa Russell G. Schuh 1.

INTRODUCTION Miya and Hausa, two West Chadic languages spoken in northern Nigeria, both

have vocalic inventories containing long and short vowels as well as two rising diphthongs. (1)

MIYA

HAUSA

l:

l:

u:

u: u o: o

i e: e

3 a: a

a: a

The pronunciation associated with the symbol B for Miya is most frequently a high, central vowel [±] , but it also has conditioned variants [i] and [u] , found, in particular, when contiguous to the glides /y/ and /w/ respectively. Miya has no underlying mid vowels, though phonetic mid vowels do occur as conditioned variants of the low vowels and diphthongs (see Schuh (n.d.) for a detailed account of Miya phonology). All the vowels listed for Hausa are contrastive in at least some environments. In this paper, I concentrate on the question of whether the phonetic long high vowels, [i:] and [u:] , found in both languages should be analyzed as syllable rimes composed of two vocalic moras (/¿i/ and /uu/) or as rimes composed of a short vocalic nucleus plus a glide coda l / i y / and / u w / ) . I will conclude that Hausa requires the former analysis and Miya the latter. Although I will consider only the long high vowels in detail, the conclusions have similar implications for the analysis of diphthongs, i.e. whether [a1] and [au] should be interpreted as /ai, au/ or /ay,

2. 2.1

aw/.

DISTRIBUTION OF RIMES Consider first the distribution of the syllable rimes. In Miya, the follow-

36

R.G. Schuh

ing rimes may appear in word final position: (2)

-i:/ -u:, -a, -a, -a1, -a U , -aC, -aC

"C" in the last two must be a sonorant consonant (nasal, liquid, or, I will argue, glide). Note that long -a: is not a possible word final rime. Moreover, the presence of -3 as a rime is always predictable: in word final position -9 only appears where the preceding consonant is not a sonorant, e.g. tal 'beer' (but no *talS) vs. tasB 'Acacia albida' (but no *tas). If [i:] and [u:J were considered long vocalic rimes, the distribution of rimes would be strangely skewed: long high vowels, but not their short counterparts, could appear word final, whereas for low vowels, only the short one could appear word final. If, on the other hand, we analyze [i:] and [u:] as /iy/ and /uw/ respectively, these rimes fall into the -9C type, with vowel quality accounted for by an independently needed rule which assimilates a to a contiguous glide (cf. tiyim 'tooth' but no *t3y3m, tuwun 'mine (f)' but no *t3wan). In fact, we can reduce the system of underlying vowels in Miya to /a, a:/, since we have eliminated long high vowels as underlying, and we can account for all short high vowels as being variants of a single vowel whose quality and positioning are entirely determined by sequence structure.1 Similarly, the diphthongs will be -aC rimes, i.e. /-ay, -aw/ rather than /-ai, -au/, the latter being vocalic sequences not otherwise found in Miya. 2.2 Turning to Hausa rime distribution, in word final position we find that all rime types are possible, i.e. all vowels (long and short), the diphthongs, and -VC (where V must be short, since Hausa does not allow long vowels in closed syllables, and C is generally a sonorant, though this condition is not so rigidly observed as in Miya, cf. takwas 'eight'). Thus, one can find such minimal pairs or near minimal pairs as ji-.bi 'day after tomorrow' vs. ji-.bi: 'any meal', ya: sa:mu: 'he got (it)' vs. ya: sa:mu 'it was obtained', dashe: 'transplanting' vs. (a) dashe 'transplanted', ba:ko: 'stranger' vs. Ba:ko man's name, da ra:na: 'with the sun' vs. da ra:na 'at midday'. Thus, there is no distributional skewing as in Miya. If we consider Hausa long vowels and diphthongs to be underlyingly / W / , permissible Hausa syllable structure can be summarized very simple as CV( Ivl ) . lc| If, however, we were to adopt the V + G analysis of long vocalic rimes proposed for Miya, a syllable such as [Ci:] would be underlyingly /Ciy/, syllables of

Long Vowels and Diphthongs in Mi ya and Hausa

37

the type /Ca:/ would be underlyingly /Caa/, with no counterpart to the claimed /Ciy/.2 The same might be said for Miya, but as we will see, in Hausa, the long vocalic rimes [i:, u:, e:, o:, a:] share a number of structural properties, whereas in Miya they share no properties other than phonetic length, as far as I know.

3.

EVIDENCE FOR OR AGAINST CONSONANTAL GLIDE CODAS IN RIMES

3.1 The regular plural formation rule for Miya is a reduplication pattern of the shape -aCaw, where C is the last consonant of the noun stem: (3)

atar maadzB kwakwi y muduw varay saw

'squirrel1 'river' 'granary' 'python' 'seed' 'tail'

pi.

atararaw maadzadzaw kwakwiyayaw muduwawaw varayayaw sawawaw

By analyzing the rimes of long high vowels and diphthongs as underlying /VC/, no special rules are needed to account for the consonant in the reduplicated part of the plural. Were these rimes analyzed as / W / , we would need a devocalization rule to convert the second V to C before applying plural formation, a complication of the phonology with no independent justification. 3.2 Hausa has a number of vowel shortening rules, some of which are illustrated in (4). The examples in (4a) illustrate a general rule which shortens a vowel in a closed syllable, in this case where a suffix -n indicating previous reference is added to a noun. The examples in (4b) illustrate a fairly productive process which allows a proper name to be created from a common noun by shortening the final vowel. (Most common nouns end in long vowels). The examples in (4c) illustrate a non-productive relation between verbs and their pluractional counterparts, the latter being formed by reduplicating initial CVC of the stem.3 (4)

a.

Closed syllable shortening:

gidaa riimii tsuntsuu birai kibau

'house' 'kapok tree' 'bird' 'monkeys' 'arrows'

gidan riimin tsuntsun biran kiban

'the 'the 'the 'the 'the

house' kapok tree' bird' monkeys' arrows'

38

R.G. Schuh b.

Proper name formation: markaa

'height of rainy season' baakoo 'stranger' bakii 'black' kumaatuu 'cheeks'

c.

Maria Baako Baki Kumaatu

name of woman born at this time man's name "Blackie" name of someone with fat cheeks

Verbs and related pluractionals: taaraa kiiraa duukaa

'gather' 'forging' 'stoop'

tattaraa kirkiraa duddukaa

'gather many' 'invent' 'crouch down'

If we analyze long vowels as underlyingly / W / , the rule(s) needed to account for such processes can be simply formulated as V + 0 in the appropriate environments. An analysis where long high vowels and diphthongs are underlying /VC/ would require either a complication of the shortening rule or an extra rule vocalizing the glide before shortening in the environments in question. 3.3

Hausa singular/plural pairs like those illustrated in (5) provide further

data showing the undesirability of the /VC/ rime analysis. (5) a. birnii kaskoo

'city' 'clay bowl'

pi. biraanee kasaakee

b. wurii damoo

'place' 'monitor lizard'

wuraaiee damaamee

c. haalii reemaa zoomoo suunaa kiifii

'circumstance' 'rock dassie' 'hare' 'name' •fish' 'fig tree'

haalaayee reemaayee zoomaayee suunaayee kiifaayee bauraayee

Elauree

The plural types illustrated in (5a-c) are in complementary distribution. For words with a medial CC sequence, such as those in (5a), long -aa- is inserted between the two consonants, and the final vowel is -ee; for those with an initial CV syllable (V being short), long -aa- is added, the second consonant is reduplicated, and the final vowel is -ee; for those with initial syllables having long vocalic rimes, including diphthongs, the plural is marked by a suffix -aayee. (See Schuh 1972, Newman 1972, Leben 1980, and references cited therein for more information on these plurals.) As in the statement of the vowel shortening rule(s) above, the statement of the plural formation rule is straightfor-

Long Vowels and Diphthongs in Miga and Hausa

39

ward if we analyze all the long vocalic rimes as / W / , but would be considerably more complex if they were analyzed as /VC/. It is worth noting here that the schema presented above for permissible Hausa syllable structure, viz. CV( j v j

), leaves open the possibility of CVG

(G = glide) syllables, since there is no restriction stated against the second C being G. Thus, we would predict rimes of the type /iy, uw, ay, aw/ alongside /ii, uu, ai, au/. Hausa does, in fact, seem to require recognition of such rimes. Note the following singular/plural pairs: (6)

miikii duutsee kaimii kyauree

'ulcer' 'stone' 'spur' 'door panel'

pi. miyaakuu duwaatsuu kayaamee kyawaaree

These nouns have plurals which suggest that the rime of the initial syllable should be /VC/, e.g. /rniykii/ inserts -aa- between the medial CC sequence to give plural miyaakuu. (The first two examples in (6) belong to a class ending in -uu rather than -ee in the plural, but otherwise the distribution of the inserted long -aa- is the same.) While we may want to allow such underlying structures to account for singular/plural pairs like these, they represent an unproductive historical remnant from a time when Hausa phonology may have been more like that of Miya. In modern Hausa the productive pattern is to treat long vowels and diphthongs as / W / , even where historically they derive from *VC sequences, as in (7). (See Schuh 1972 and references cited there for the sound change which vocalized syllable final velars.) (7)

buuzuu


eta-0

From the above conjugations, we can see that Oromo uses Copular -Da . [y] and [u] < — > [w] are different realisations of the same segment, we can derive the correct forms, as illustrated in (12).

The Case of U in Tigre (12)

99

u w

N

Ax x \ busV / /

x

x

[bus]

N X

I

X

I A

N X

X

X

1/ w

?A +

X

X

^

bus

['Abwas] In the singular, the [u] is associated to the Nucleus and is therefore realised as [u] . In the plural however, as the U occupies a [-syllabic] position (the Onset), it is realised as [w] . This analysis implies that the hybrids are part of the consonantal root. In (12), the root is bu£. Moreover, assuming the representation of the vocalic system in (3), by which [e] and [o] are complex segments constituted of I-A and U-A respectively, we can also derive the plural forms of nouns containing [e] and [o] , as in (13). In this example, when the vocalic melody [a] occupies the Nucleus with the hybrid [u] of the root sUrt the resulting segment is [o] . (13)

o w

N

X

A

X

N X

x

\\/ / sur [sor]

X

I

X

1/ ?A +

I A N

X

X

X

X

/ s u r

X

[?Aswar]

Thus, the data observed so far seem to sustain an analysis according to which the hybrids observed in the singular-broken plural pairs are part of the consonantal root. Such an analysis predicts that the order in which the components of the root appear will remain constant throughout the derivation of the word. Given that the order of the root consonants is constant and that line crossing as in (6) is forbidden, any derivation where the position of the hybrid in relation to the other members of the root is not constant is predicted to be ungrammatical. This prediction, however, is not borne out, as shown by the examples in (14), all involving the hybrid [u] .

S.

100

(14)

Filipovich

S M.

Tremblay

Singular

Plural

Gloss

fanus babur birob qeto_ tiso

fawanis bawabir bAwarib qkwhtti tAwAssi

'lamp' 'train, ship1 'grass sack1 1 arm-chair' 'pot'

We note a regularity in the distribution of the hybrids in the forms in (14) : in all cases, the hybrid appears in the third position of the root in the singular (as in fanus) and in the second position in the plural (as in fawanis). (15)

Singular

Plural

CCUC

CUCC

We should also point out that the data in (14) involve forms which contain two long vowels in the singular. In addition to the above, another fact distinguishes a round vowel from the rest of the Tigre vocalic system. The vowel [u] , manifesting the hybrid u, never appears in the first vocalic position in bi-syllabic nouns. On the other hand, as can be seen in (16) , the vowel [o] , formed of an U and an A, is free to occupy either the first or second vocalic position. (16)

* CuCV(C) CoCV(C) CVCu(C) CVCo(C)

To summarise so far, our examination of singular-broken plural pairs reveals that given the correspondence between the vowels in the singular and the glides in the broken plural, the hybrids are part of the consonantal root. This analysis predicts that the position of the hybrid in relation to the other consonants of the root should remain constant. This prediction, however, is not borne out as there exists a restricted class of forms in which the hybrid [u] that occupies the third root position in the singular appears in the second root position in the plural.

3.

A POSSIBLE SOLUTION: GOVERNMENT These observations have led us to believe that the behaviour of the hybrid

The Case of V in Tigre

101

[u], within the domain of the word, follows from an intrinsic property of the element U. To account for the distribution of the hybrid [U] and the apparent methathesis observed in the forms in (14), a way must be found to express the interaction between segmental and prosodic levels of representation. More specifically, we require a principle governing structural relations in a supra-syllabic domain. Consider for a moment how structural relations are expressed elsewhere in linguistic theory. In syntax, hierarchical relations existing between constituents in a given structural context are defined in terms of "government". Aoun (1985:23) defines this relation as "a locality condition imposed on contextual dependencies". More precisely, it defines the structural relation which exists between a governor (the head) and a governee (the dependent category). In phonology, the distribution of segments within the domain of the syllable has been formulated in terms of Prosodic Government, according to which the head of the syllable, the Nucleus, governs the other constituents of the Rime. (17)

Prosodic Government: The Nucleus must c-command all the constituents of the Rime. (Lowenstamm & Kaye, in press)

The last syllable on the right in (18) is ruled out as the nucleus does not c-command the coda.

3.1 Supra-Syllabic Government The notion of "government" in phonological theory thus far provides a means to constrain the output of phonological processes at the syllabic level. It does not, however, impose any restrictions on structural relations existing at a supra-syllabic level, and does not make reference to constraints applying to the segmental content of a domain larger than the syllable. We would like to propose that such restrictions do exist, and can be accounted for if we postulate a wellformedness condition expressed in terms of "government" which applies at the level of the word. Following Chomsky (1981), in order to formulate a theory of government of this type, we must establish (1) the class of governors, (2) the class of governed items, and (3) the structural context involved. According to our theory, in a domain composed of syllables (the phonological word), one syllable is identif-

102

S. Filipovich S M. Tremblay

ied as being dominant. This choice is parametrically determined for each language, domains being typically head intitial or head final as shown in example (19) . (19)

Head Initial

Head Final

The head of the dominant syllable, the Nucleus, is the governor of its domain, and as such governs all other projections of N in its realm. We can define SupraSyllabic Government as in (20) and the Head Parameter as in (21). (20)

Supra-Syllabic Government: A Nucleus identified as being the head of a phonological word governs all other syllables in its domain.

(21)

Head Parameter: The head position is parametrically determined (head initial/head final).

In the next part of this paper, we will show how an extended theory of government can provide em interesting account for the behaviour of U in Tigre nouns. Our theory must be able to account for two facts:

3.2

(i)

Firstly, the distribution of [u] - more precisely, why do we never find [U] in the VI slot of a bisyllabic noun?

(ii)

Secondly, we must explain the apparent metathesis of root consonants in nominal forms with two long vowels: given that the order of root consonants is constant, how can we explain the derivation of the forms in (14)?

The Head Parameter in Tigre Lowenstamm & Prunet (1985) have observed that harmonic processes in Tigre

are right dominant. In example (22) , when the vowel [a] , which is capable of triggering vowel harmony, is suffixed to the noun shlshlAt, harmony spreads from right to left, lowering all mid vowels in the harmonic domain to [a] . (22)

sAIsAIAt + a

salsalata

We can infer from this observation that the phonological word in Tigre is head final. According to our definition of government in (20) , the Nucleus of the

The Case of U in Tigre

103

final syllable of the phonological word in Tigre is the governor of its domain. Given the distribution of [u] in Tlgre nominal forms, the fact that we never find [u] in the VI slot would indicate that this vowel must occupy the head position, and therefore be excluded from appearing in governed positions. The hybrid [u] in Tigre can be thought of as conveying dominance properties, for when [u] appears in its [+syllabic] form, its distribution is restricted to ungoverned syllabic positions.2 We have formulated this constraint in (23). (23)

U

cannot be governed

[+syl] If this is indeed the case, it would seem that segments which are characterised as being [+dominant] in a system are compelled in some way to occupy a dominant or governing position within the domain of the word. The matching of dominance properties of prosodic and segmental levels of representation in Tigre may be stated as a wellformedness condition applying to the output of phonological processes. This condition is formulated in (24). (24)

3.3

Wellformedness Condition: A segment which is characterised as being [+dominant] must occupy a dominant or governing position in the domain in which it appears. The Metathesis of Root Consonants

Having shown that the distribution of [u] follows from an intrinsic property of this segment, let us now turn to the question of the metathesis of root consonants in the forms in (14). As our analysis presupposes that the hybrid is part of the root, our first task is to establish the underlying root form for these nouns. For example, for a form like (25), does the ordering of root consonants correspond to the singular, in which case the root would be BRUB, or does the plural reflect the underlying root form, BURB? (25)

Singular birob

-

Plural BRUB

bhwarib

-

BURB

We suggest that the order of the consonants in the root corresponds to the surface order of the plural for the following reasons. If we take the arrangement of root consonants of the singular (CCUC) as underlying, we would have to explain why [u] appears in the C2 slot in the plural.

104

S. Filipovich s M. Tremblay

As opposed to the [+syllabic] realisation of [u] , which is subject to the constraint in (23) , we can find the [-syllabic] [w] in governed and ungoverned syllables. Moreover, [w] is free to occupy any onset or coda within the domain of the syllable.3 As the distribution of [U] while [-syllabic] is unconstrained at both the syllabic and word levels, any explanation we could provide for the methathesis given an underlying root CCUC would amount to a stipulation that in certain contexts - namely the forms in (14) - [u] could not appear in the C3 slot in the plural. If, on the other hand, we assume that the root forms in (14) correspond to the plural CUCC, we can account for the appearance of [U] in the V2 slot in the singular by appealing to independently motivated observations, such as the distribution of [u] in underived forms not involving methathesis and the direction of government as shown through harmonic processes.1* In other words, the forms displaying methathesis are precisely those in which we would expect metathesis, and this follows from the wellformedness condition in (24).

4.

SUPRA-SYLLABIC GOVERNMENT AT WORK In order to derive the forms in (14) we must know that (i) Tigre is head

final; (ii) there is a wellformedness condition in Tigre stating that U must occupy the head position within the domain of the word; and (iii) the underlying order of the root consonants of the forms involving methathesis is reflected in the plural. For example, in the case of birob 'grass sack', whose root is BURB, the derivation of the plural bKwarib is straightforward, as can be seen in (26a). The vocalic and consonantal melodies are associated to the prosodic template of the broken plural, the [u] occupying the C2 slot as would be expected. (26) a.

A

a

N

N

N

x x x x x x x x b

U

r

b

[bAwartb] As for the singular, we can see in (26b) that following normal autosegmental association principles, we derive *burab.

The Case of U in Tigre

105

(26) b.

a

N

A \ V I

X

X

X

X

U

b

I A N

X

r

X

X

^

b

* [burab] This derivation constitutes a violation of the wellforniedness condition in (24) , as [u] appears in the Vl slot, a governed position not dominated by the head of the word. If, on the other hand, the [u] is associated to the V2 slot as in (26c), we would violate one of the association conventions in (6) which prohibits crossing of association lines.

[birob] Consequently, given the behaviour of [u] in Tigre nominal forms and the types of roots involved (CUCC), we must suppose that [u] does not occupy the same autosegmental plane as purely consonantal root segments. In (26d) we see that by positing a different level of representation for

U

we can derive the

correct form without violating any principles. Under our analysis, [u] is attracted to the governing position and can freely dock here because it occupies a separate autosegmental tier. (26) d. N

X

A

X

\b

N

X

X

X

A

X

X

\ rVi b/

U — [birob]

106

S. Filipovich S M. Tremblay

As our analysis neatly accounts for the unique behaviour of 0 in nominal forms in Tigre based on the assumption that U should be represented as occupying an autonomous autosegmental plane, two questions immediately come to mind: Why should this be so? And is the behaviour of U (and the representation proposed here) an idiosyncratic property of Tigre, or can it be related to more general /

principles which cover similar phonomena found in other languages? Although we can provide only partial answers to these questions, it would seem that supportive evidence can be found in favour of the analysis proposed here. In response to the second question, the unique autosegmental behaviour of U in Ethiopie languages has been noted abundantly in the literature (Polotsky 1938, 1951; Leslau 1941; Ullendorff 1955; Hetzron 1971, 1972; Hetzron & Marcos 1966; MacCarthy 1983). As U in Ethiopie languages has "long distance" effects (as for example in Chaha (Hetzron & Marcos 1966; McCarthy 1983) it would appear that an independent autosegmental representation for U in Ethiopie is warrranted. As for the question "why should this be so?", the data reported here are but another example of what has been referred to a "The Back/Round Asymmetry" (Kaye, Lowenstamm, & Vergnaud 1985) reflecting a fundamental distinction between the elements associated with these lines, I and U. Although this distinction is not explanatory, its existance reflects the intrinsic difference between these two elements.

5.

CONCLUSION We have shown that it is possible to account for the unusual distribution of

[U] in Tigre nominal forms by extending the notion of government to the domain of the word. Our analysis makes explicit the need to be able to express restrictions on structural relations beyond the level of the syllable. We have chosen to formulate our analysis in terms of a wellformedness condition rather than making reference to phonological rules. This paper is in keeping with the research programme developped by Kaye, Lowenstamm & Vergnaud (1985) which favours a principles and parameters approach to phonological analysis.

NOTES *We would like to thank the following people for their invaluable comments on earlier drafts of this paper: Jean Lowenstamm, Rose-Marie Déchaine, Jean-François Prunet, and Monik Charette.

The Case of U in Tigre

107

1.

By convention, elements are represented in capital letters, (I, U, A etc.), and their segmental realisation in small letters ([a], [e] , [i] etc.) Hybrids will be represented as [u], [I]. See Kaye, Lowenstamm & Vergnaud (1985) for a detailed account of the internal representation of segments.

2.

We find a similar situation in Kpokolo (a Kru language spoken in the Ivory Coast). Kpokolo, like Tigre, is head final with respect to government. What's more, the distribution of round vowels in Kpokolo is sensitive to government, that is, they are restricted to the governing position within the domain of the word. As Kpokolo is head final, round vowels are found freely in final syllables, and may be found in non-final syllables (i.e. governed syllables) only if "licensed" by the presence of a round vowel in the head position. For details, see Kaye, Lowenstamm & Vergnaud (1985).

3.

Within the domain of the syllable, the onset may be governed by a nonbranching N' (if we adopt Reinhart's (1983) definition of c-command). The coda is always a governed position.

4.

Another piece of evidence in favour of this analysis was brought to our attention by J. Lowenstamm (personal communication). If the underlying root for these forms were CCUC, in the case of babur we would have a root BBUR which is unattested in Semitic, as no root has identical consonants in the first and second consonantal positions. In view of this observation, the root for babur would therefore have to be BUBR.

REFERENCES Aoun, Joseph. 1985. A Grammar of Anaphora. (Linguistic Inquiry Monograph 11.) Cambridge: MIT Press. Clements, George N., and Kevin Ford. 1979. Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10:179-210. Goldsmith, John. 1979. Autosegmental Phonology. New York: Garland. Hetzron, Robert. 1971. Internal labialisation in the tt-group of south Ethiopie. Journal of the American Oriental Society 91:192-207. Hetzron, Robert. 1972. Ethiopian Semitic: Studies in Classification. (Journal of Semitic Studies Monograph 2.) Manchester University Press. Hetzron, Robert, and Habte Miriam Marcos. 1966. Des traits pertinents superposes en ennemor. Journal of Ethiopian Studies 4:17-30. Kaye, Jonathan, and Jean Lowenstamm. 1984. De la syllabicité. In: Forme sonore du langage, ed. by F. Dell, D. Hirst and J-R. Vergnaud, pp. 123-160. Paris: Hermann. Kaye, Jonathan, Jean Lowenstamm, and Jean-Roger Vergnaud. 1985. The internal structure of phonological elements: A theory of charm and government. In: Phonology Yearbook 2, ed. by Colin Ewen and John Anderson, pp. 305-328. London: Cambridge University Press. Leslau, Wolf. 1941. Documents tigrigna. Paris: Klincksieck. Levin, Juliette. 1983. Reduplication and prosodie structure. Talk given at GLOW, York, England. Lowenstamm, Jean, and Jonathan Kaye. 1986. Compensatory lengthening in Tiberian Hebrew. In: Studies in Compensatory Lengthening, ed. by Engin Sezer and Leo Wetsels, pp. 97-132. Dordrecht: Foris. Lowenstamm, Jean, and Jean-François Prunet. 1985. Tigre vowel harmonies. Paper presented at the 16th Conference on African Linguistics, Yale University, March, 1985.

108

s. Filipovich s M. Tremblay

McCarthy, John. 1981. A prosodie theory of non-concatenative morphology. Linguistic Inquiry 12:373-417. Palmer, F.R. 1962. The Morphology of the Tigre Noun. London: Oxford University Press. Polotsky, H. 1938. Labialisation et prépalatalisation à distance. Bulletin de la Société linguistique de Paris 39. Polotsky, H. 1951. Notes on Gurage Grammar. Jerusalem. Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. London: Croom Helm. Ullendoff, Edward. 1955. The Semitic Languages of Ethiopia: A Comparative Phonology. London: Tailor's foreign Press.

Département de Linguistique Université de Québec à Montréal

Linguistics Department McGill University

Chapter 9

Wh-Movement and Proper Government in Yoruba Remi Sonaiya 1.

THE MOVEMENT FACTS Yoruba has a rule which moves a wh-word or phrase from subject, object, or

adjunct position in underlying structure into COMP, as illustrated in (l)-(3). This syntactic wh-movement is obligatory in questions, and the wh-word is followed obligatorily by the focus particle, ni.

(1)

ta who

ni o wa ? FOC 3s come

who came?

(2)

ki ni o ri ? what FOC 2s see

what did you see?

(3)

nibo ni o ti ri Bgla ? where FOC 2s V see

where did you see B9la?

Movement out of subject position (1) requires the insertion of a resumptive pronoun at the extraction site, while no such insertion is needed for object and adjunct position, as seen in (2) and (3) respectively. Movement out of adjunct position requires the insertion of a preverb ti, fi or

and it is possible

for two of these preverbs to co-occur.1 The focus construction using the ni particle is not limited to wh-movement alone, but is very widely used in a variety of constructions. It allows any sentence constituent to appear sentence-initially. An object can be focused:

(4)

awo ni Ade fi fo dishes FOC PROG wash

it is dishes that Ade is washing

or a subject: (5)

ade

ni 6 A fo awo FOC 3s PROG wash dishes

or a prepositional phrase:

it is Ade that is washing dishes

110

Remi Sonaiya

(6)

ori tabili ni iwe wa head table FOC book be

it is on the table that the book is

or an adjunct: (7)

ni at

oja ni Ade market FOC

ti V

ri Boia see

it is at the market that Ade saw Bpla To question a genitive NP, two formats are available in Yoruba: either the use of a resumptive pronoun (8) or pied piping (9). Thus we have: (8)

ta ni olukç na çmç rç ? who FOC teacher beat child his/her

(9)

omo ta ni oluko na ? whose child did the teacher beat?

Wh-words in Yoruba can be divided into two classes: those that require the insertion of a preverb (class I) and those that do not (class II). The two classes are: Class I

Class II

ki - what ta - who nigbawo - when2

bawo nibo nitori ki kilode

- how - where why

For "where" questions, certain asymmetries are noticeable. Consider the three sentences: (10)

nibo ni Ade lo ? where FOC go

where did Ade go?

(11)

nibo ni Ade where FOC

fi iwe si ? put book on

where did Ade put the book?

(12)

nibo ni Ade where FOC

ti V

where did Ade see Bola?

ri Bola ? see

Wh-Movement and Proper Government in Yoruha

111

The difference between (10) and (11) on the one hand and (12) on the other lies in the fact that movement in both (10) and (11) is out of a subcategorized complement, while it is out of an adjunct position in (12). That is, the traces of movement in (10) and (11) are properly governed by the verb and preposition respectively, in the sense of GB theory, while the trace in (12) is not properly governed. Hence, the preverb must be inserted, I claim, in order to properly govern that trace and prevent violation of the Empty Category Principle (Chomsky 1982) which states that [e] or [e] must be properly governed. NP XP Proper government itself may be defined in the following terms: A properly governs B if A governs B and (i) A is a lexical category, or (ii) A is coindexed with B. This is one of two strategies for avoiding violations of the ECP. In the case of extraction from subject position and genitive position, both of which are improperly governed because they are not complements, the violation is avoided by "covering up" the offending trace with a resumptive pronoun. In English, it is possible to extract from subject position without violation of the ECP, as in: (13)

who did you say [_e [ g e came]]?

This has been attributed to the second clause of the definition of proper government above, under which a trace is properly governed if locally bound by a c-commanding antecedent—here, the trace in the intermediate COMP. Why does Yoruba lack this possibility for antecedent government? Note that in (14)

*who did you say [- that e [ e came]]?

proper government also fails in English—presumably because the trace in the branching COMP (branching because it contains a complementizer) does not ccommand the subject trace. This observation, in conjunction with the fact that the focus particle ni causes COMP to branch in multi-clause questions as well

112

Remi Sonaiya

as in simple sentences in Yoruba, explains the absence of antecedent government from COMP as a possibility.3

2.

SUBJECT EXTRACTION AND RESUMPTIVE PRONOUNS Yoruba can be said to have an Empty Subject Filter (ESF) stipulating that

any surface structure containing an empty subject position is ill-formed. This filter may itself be thought of as a direct consequence of the ECP, already mentioned. Again, my claim is that there is no antecedent government in Yoruba, and this fact is borne out by the need for a resumptive pronoun in subject position. There is no other obvious way to explain the fact that "who t came?" is grammatical in English, while its equivalent in Yoruba (i.e., whithout a resumptive pronoun) is not. It must be that in English, a trace can be governed by its antecedent (in COMP), while this option is not available in Yoruba. Certain facts about resumptive pronouns, the third person singular pronoun, and the subject position come together in an interesting way in Yoruba, and I wish to look at these in some detail here. The point of departure for the discussion here is to note the existence of certain apparently subjectless wh-questions in Yoruba: (15)

ta nl n pe iy ? who FOC PROG call you

who is calling you?

(16)

ki nl sele ? what FOC happen

what happened?

It would seem then, that resumptive pronouns are not obligatory, as I have claimed, for subject extractions, and that the ESF stipulated above is unnecessary. The crucial factor here has to do with the status of the third person pronoun, which is also used as a resumptive pronoun. Carstens (1986), on the basis of data similar to (15) and (16), concludes that resumptive pronouns are, in fact, optional— "the pronoun may be omitted". She states: These facts are significant for our investigation, since they indicate that subject interrogation may somehow be licensed without the RP presumptive pronoun). Determining what makes this possible should allow us insight into why proper government from COMP might fail, and thereby into which aspect of it fails (lexical or antecedent

113

Wh-Movement and Proper Government in Yoruba government), if the evidence confirms the hypothesis that there are two aspects to consider.

Carstens summarizes Pulleyblank1s (1985) analysis of the subject extraction facts of Yoruba, claiming that "the general lack of proper government from COMP is caused by a failure of c-command". The argument is that the resumptive pronoun is obligatory only when COMP branches, and that ki ni and ta ni (the two whwords with which resumptive pronouns are used), are "structurally ambiguous". Carstens claims that these wh-words "incorporate the focus marker into their structure as single words", but if they are analyzed as two constituents, there can be no c-command, and the use of a resumptive pronoun becomes obligatory. I wish to show that there is a better analysis (Carstens herself notes that "there are a few problems with this account"), an analysis which hinges crucially on what the status of the third person singular pronoun is. Stahlke (1973), has suggested that the third person singular pronoun in Yoruba be analyzed as a zero morpheme, given the following facts about its distribution: (i)

Preterite Affirmative mo wa o wa o wa

(iii)

Future Affirmative mi (y)ö wa 0 d wa 0 yd wa

(ii)

I will come you will come he will come

Preterite Negative mi (k)d wa o (k)o wa 0 kd wa

I came you came he came (iv)

I did not come you did not come he did not come

Future Negative mi ki yd wa o ki yd wa 0 ki yd wa

I will not come you will not come he will not come

As can be seen, in all cases except the preterite affirmative, the third person pronoun is zero. The morpheme o appearing in the exceptional case is there due to a phonological rule in Yoruba that third person pronouns, in the context of preterite affirmative sentences, must be marked by a high tone. This rule applies to all third person pronouns, as demonstrated by the following: won wa - they came won kd wa - they did not come won yd wa - they will come

om6 wa - the child came omo kd wa - the child did not came omo yd wa - the child will come

Only in the preterite affirmatives do we have high tones—there are mid tones in the other cases. Therefore, for the sentences involving wh-movement [see (15) and (16)], I claim that the resumptive pronoun is, indeed, present. Normally, the focus part-

114

Remi Sonaiya

icle bears a mid tone, but in these examples, it is marked by a high tone—the high tone of the third person pronoun which is required in preterite affirmative sentences. Since all that is required is a high tone, Yoruba thus employs two alternatives for marking resumptive pronouns: (a)

insertion of the morpheme o to bear the high tone (e.g., ki ni 6 sele? — what happened?)

(b)

transfer of the required high tone onto some element in the sentence that does not already bear a high tone (e.g., ki nl sele?1*

With this analysis, we need not postulate any ambiguity in the structures of the wh-words ki (what) and ta (who). There is no idiosyncratic behavior on their part that allows antecedent government in certain cases, while disallowing it in others. Furthermore, it follows from the foregoing that the focus particle cannot be analyzed as combining with these wh-words to form single words: I have shown that a resumptive pronoun is always present, and this means that antecedent government is always blocked, due to the branching structure of COMP, as in (17) COMP

/X ki

3.

S

ni

THE PREVERBS In multi-clause questions, an adjunct phrase may be extracted either from the

lower or from the higher clause. The presence of the preverb in the embedded clause alone means that that clause is the clause in which the question phrase originates. For illustration, I will use the same sentences as used by Carstens (1984). (18)

nibo ni Ajike where FOC

so fun tell

e pe oun you that 3S

ti V

ri Tolu ? see

where did Ajikf: tell you that she saw Tolu? Thus, for (18), a sentence which is ambiquous in English, the only meaning available in Yoruba is one in which it is the place where Tolu was seen that is being questioned (and not where you and Ajike were).

Wh-Movement and Proper Government in Yoruba

115

There are two other possibilities for multiple-clause questions: having the preverb exclusively in the matrix clause, and having one preverb each in both matrix and embedded clauses.5 Carstens claims that if a preverb is present only in the matrix clause, the sentence is ambiguous, so that two different interpretations are available for the sentence in (19). (19)

nibo

ni

Ajike

ti_ so

fun

e

pe

oun

ri

Tolu ?

The two interpretations are: (a)

for which place x, [Ajike told you at x] that she saw Tolu

(b)

for which place x, Ajike told you [she saw Tolu at x] .

The claim here, then, is that question (19) corresponds to two different structures, one in which the adjunct extraction is from the matrix clause (a) and the other in which the extraction is out of the embedded clause (b). From the subjects interviewed, however, I found that there was a very strong preference for the (a) interpretation. In fact, one person claimed outright that this was the only possible interpretation. Therefore, the facts are not very clear. The obvious point that should be taken into account is that there is an apparent tendency, when the preverb is in the matrix clause, to take that clause as the locus of the question. The final possibility is to have the preverb in both matrix and embedded clauses, for example: (20)

nibo

ni

Ajik$

ti

sg

fun

f

pe

oun

ti ri

Tolu ?

Carstens claims that this sentence can have either of the interpretations (a) or (b) above, or even a third one, (c) below: (c)

for which places x and y, [Ajike told you at x] that [she saw Tolu at y] .

My own finding, again, is that it is easier to note preferences rather than clearcut judgments.6 The preferred reading, here too, is still that one in which it is the matrix clause that is being questioned, and not the embedded clause. We will not discuss, in this section, the consequences that these facts have for the analysis of government in Yoruba. We will leave that for the section on

116

Reml Sonaiya

government. But before we proceed to do this, another fact deserves mention. There are two different "why" words in Yoruba. The first one is nitoii ki ni (meaning literally 'because of what'), and the other is kilode (literally 'what has come /happened/'). There is a behavior that is peculiar to kilode among whwords due to its unique internal structure. We will discuss this structure by way of leading up to the discussion of the structure of COUP in Yoruba in the following section. We have claimed that when moving from adjunct position it is necessary to insert a preverb because the branching of COMP blocks antecedent government. But what about: (21)

kilode why

tl o lg ? you go

why did you go?

Here, tl is required, even though COMP doesn't seem to branch, i.e., if we take kilode to be a Q-word. But if tl is a preverb, two puzzles arise: 1. Why does it have a high tone (a mid tone is normal), and 2.

Why is it in a pre-subject (rather than post-subject) position?

There are two possible analyses of this phenomenon. The first is the following: tl is a preverb which is obligatorily moved into COMP, due to the absence of the focus particle ni, presumably by a rule prohibiting an empty c in the following structure: * [= kilode [- [e] [ o ti lo?] ] ] c c s obi. This analysis, however, still leaves the puzzle as to why a preverb bearing a mid tone in post-subject position changes to a high tone in pre-subject position. The second analysis, which is the correct one in my opinion, takes tl to be a complementizer, and this fact can be demonstrated by comparing kilode questions with the following similar ones. (22)

ta

lo

wa

(=ta ni o wa) tl o fun ni owo ? who FOC 3S come that 2S give PFEP. money

who came that you gave money to?

Wh-Movement and Proper Government in Yoruba (23)

nibo

lo

117

(=nibo ni o) n lo ti o WQ aso funfun ? where FOC 2S PROG go that 2S wear garment white

where are you going all dressed in white? (24)

kilode

(=ki ni o de) ti e pe de ? what POC 3S happen that 3PL late arrive

what happened that you came late?

(=why did you come late?)

From these facts, we pan conclude that the ti in (24) is a complementizer, just as in (22) and (23), and that no preverb is required for such questions. Thus, the structure of "kilode" questions cannot be the following:

But how must such questions, then, be analyzed? This takes us directly into our next section, dealing with the structure of COMP and government in Yoruba.

4.

THE STRUCTURE OF COMP AND GOVERNMENT There are still two options for the analysis of kilode: it can either be

considered a single word, or it might be thought of as a full sentence. A consideration of both types of analysis will give us insight into the nature of COMP and how government works in Yoruba. (a) The Single-Word Analysis: If kilode is a single word in COMP and it is directly followed by a complementizer, then we must make the claim that there are two COMP positions in Yoruba, i.e., (26)

The trace of the moved wh-word would be in adjunct position, and since the COMP

118

Remi Sonaiya

does not branch, kilode would be able to govern its trace in adjunct. But there is still a problem: Why should this Q-word alone not require a focus particle (i.e., allow antecedent government)? (b) The Full-Sentence Analysis: Kilode, under this analysis, must actually be considered as a complex phrase containing the (subject) question-word, ki (what), which does not require a preverb, because its trace is replaced by the resumptive pronoun, o, as is usual for subject traces. The structure we have is:

Movement here has taken place out of subject position, and a resumptive pronoun has been inserted because the branching COMP blocks antecedent government. (Phonological processes yield kil6de from ki ni 6 de). From this, it seems that kilode should, in fact, not be considered a Q-word at all. Indeed, this would appear to be perfectly reasonable, since no one has ever suggested that the very similar structure, "ta lo wa", (see (22)), is a Q-word. This analysis requires no special assumptions or idiosyncratic behavior about kilode.7 What falls out clearly from the foregoing discussion is that the WH-COMP has a branching structure, while the COMP for complementizers does not. As stated earlier, a branching COMP blocks antecedent government, and this explains why there are specific constraints on traces in subject and adjunct positions: a resumptive pronoun must be inserted in the subject position, while a preverb is obligatory for extraction out of adjunct position. The facts remain the same for both simple and complex sentences, and the complementizers (pS and ti) do not seem to play any role with respect to proper government. Since we have claimed that proper government in Yoruba is restricted to government by lexical categories only (i.e., no antecedent government), then the issue of successive cyclic movement of the wh-phrase in complex sentences does not have much relevance, and we will, therefore, not consider it. For example, take the sentence:

Wh-Movement and Proper Government in Yoruba (28)

119

[_ki ni [s01u so [_pe [ s o 5§1?] ] ] ] ? what FOC say that 3S happen what did Olu say happened?

Even if the wh-word (originating from the subject position of the embedded clause) left a trace in lower COMP, the criterion for proper government still would not be satisfied for Yoruba, and that is why a resumptive pronoun in that position remains obligatory. However, if we must take into account successive cyclic movement, our analysis must be that the trace left in the lower COMP turns it into a branching structure, which blocks proper government.

5.

GOVERNMENT FROM INFL Preverbs and resumptive pronouns are needed in cases where the wh-trace can-

not be properly governed, either by the verb in the sentence, or from COMP. In this section, we will consider the use of preverbs in INFL for proper government. The trace left in adjunct position after movement of the wh-phrase cannot be properly governed by the verb in the VP. However, we may assume that extraposition of the adjunct phra,se out of VP is forced by the Case-Resistance Principle (Stowell 1981). An inserted preverb then lexically governs the adjunct trace: (29)

S

/

WH

\ni

I

Preverb

It

The analysis of sentences with a preverb only in the embedded clause of a complex structure is quite straightforward. The question bears upon the embedded clause alone, and the preverb properly governs the trace in adjunct position. Thus, for sentence (23), the structure will be:

120

Remi Sonaiya

For sentences with the preverb only in the matrix clause, the interpretation in which that clause is the locus of the question poses no analytical problems. Recall, however, that for such sentences, like (19), repeated here for convenience, it is possible for some speakers to get an interpretation in which the embedded clause is the locus of the question. (19)

nibo .ni Ajike where FOC

ti V

so fun tell

§ pe oun you that 3S

ri Tolu see Tolu

where did Ajik? tell you that she saw Tolu? This gives us the following structure:

Thus we have a preverb in the main clause which must govern a trace in the adjunct position of the embedded clause. As noted earlier, a consideration of successive cyclic movement of the wh-phrase through COMP does not adequately

Wh-Movement and Proper Government in Yoruba

121

account for the facts, and therefore does not promise to yield any convincing explanation in this case either. As a matter of fact, one can conclude from the

discussion so far that no element in COMP can properly govern a trace in Yoruba. We are thus forced to look elsewhere for an explanation. However, before doing this, let us consider the other problematic case, namely, that of preverbs in both clauses. Again, if the question bears upon both clauses, there is no analytical problem (although, as we have said, this would be an unusual interpretation). But it has been claimed that there are two other possible interpretations for such sentences. (20) is here repeated: (20)

nibo ni Ajikg ti where FOC V

SQ fun $ pe oun tell you that 3S

ti V

ri Tolu ? see Tolu

where did Ajike tell you that she saw Tolu? The two other interpretations bear upon either one of the clauses, that is, either where the telling took place, or where Tolu was seen. As we mentioned earlier, the judgments on these facts are far from being clear-cut. All the same, even if only a few speakers are able to get these "problematic" interpretations, it still remains that an explanation must be sought. We agree with Carstens that one should probably assume that movement of the preverb proceeds through INFL. In that case, the analysis of (19) will proceed in the following way: the preverb in the matrix clause originates in INFL of the lower clause from where, through INFL-to-INFL movement, (a movement which one might claim is motivated by the general preference for preverbs in the matrix clause), it is moved to the higher clause, leaving a trace in the lower clause: (32) COMP nibo

ni

NP

INFL

I

I

Ajike ti

NP

INFL

I

I

oun

VP

t V I

ri

NP !

Tolu

122

Remi

Sonaiya

It is not known whether verb traces need to be properly governed as well, and if so, in what terms such government should be defined. We, therefore, leave the matter as it stands. For the case of a preverb in each clause, the non-problematic interpretation, i.e., that involving two different questions, gives us the following structure: (33)

INFL Ajike

ti

VP V

sg fun

NP

COMP pe

NP I

oun

INFL I

ti

VP

XP

/ X V NP I I ri

Tolu

t

There are two traces, each of which is properly governed by the preverb in its clause. If the interpretation of (20) bears upon the embedded clause alone, the analysis must be that rather than move the preverb from lower INFL to upper INFL and leaving a trace at the source, such speakers prefer to insert another preverb in the higher INFL, an insertion, again, which would be motivated by the preference for a preverb in the matrix clause. It is the interpretation in which the question bears upon the higher clause that seems to defy analysis. If the intent of the speaker is to question where the telling took place (the higher clause) and ti has been adequately inserted in the INFL of that clause, it is not clear at all why another preverb should be inserted in the lower INFLexcept that the speaker is one that mechanically puts a preverb in every clause of his/her sentences, as long as there is a preverb-requiring Q-word in initial COMP.

6i

CONCLUSION I have shown in this paper that the use of resumptive pronouns and preverbs

is obligatory in preventing violations of the ECP. In particular, we have been able to show that the cases of subject extraction that seem not to use resumptive pronouns actually do, and that, following Stahlke (1973), the third person

Wh-Movement and Proper Government in Yoruba

123

singular morpheme should be analyzed as being 0, and that the o that appears in certain contexts is a result of the stipulation that the pronoun be marked by a high tone in affirmative sentences. The o is therefore inserted to bear the high tone. However, if the high tone can be transferred on to some other element in the sentence, this is also acceptable. Therefore, it is possible to have kilo sele (=kini 6 sfle) or kin! sele. In the case of preverbs, I have demonstrated that these are required to properly govern adjunct traces, since there can be no antecedent government in Yoruba, due to the structure of COMP which branches. Preverbs are, therefore, required as lexical governors. However, the interpretation of multi-clause sentences with preverbs leaves some unresolved questions, as noted in the preceding section. I have agreed with Carstens in assuming INFL-to-INFL movement of the preverb as a way of handling cases of a preverb in one clause acting as a proper governor for a trace in another, but some aspects of this still seem to defy convincing analysis.®

NOTES 1.

Usually it is se that occurs with either of the other two. e.g.,

bawo ni how

o

ti

FOC 2S V

se

fe

V

want go

lo ?

how do you want to go? Also, ti is optionally followed by gbe in some dialects: e.g., nibo ni o ti gbe ri Bola ? where FOC 2S V V see where did you see Bpla? 2.

One may wonder why a time adjunct, nigbawo, should pattern with class I rather than class II preverbs. An explanation might be that being a time adjunct, nigbawo is coindexed with TENSE, and presumably properly governed by it.

3.

Finally, wh-words are not used in indirect questions in Yoruba. Rather, the non-wh equivalents are used, along with the relative clause marker, ti. For example: mo fe mo eni ti 6 wa I want know person that 3S come *mo fe to) ta ni 6 wa I want to know who FOC 3S come I want to know who came

4.

This analysis also adequately accounts for the "Ekiti facts" discussed by Carstens. In this dialect of Yoruba, the following can be observed:

124

Remi Sonaiya a.

Ies6 ri

Nike

(=lesl o who

who b.

ri

Nike ?)

3s see

saw Nike?

*lesi ri Nike ? who saw Nike?

That is, while in Standard Yoruba it is possible to mark the resumptive pronoun either with an overt pronoun (bearing a high tone) or simply by transferring the high tone onto some other element, for Ekiti, only the first alternative is possible—that is, morphological marking. The reason for this is the fact that the word that would otherwise be qualified to receive the high tone, in the case of transference, already possesses a high tone underlyingly: lesl. This fact therefore disqualifies the word from receiving the transferred high tone of the resumptive pronoun. Overt marking is thus obligatory in this case. 5.

However, before we go into their discussion, it is necessary to make clear that even at the level of (18)-type of sentences, all the wh-words do not behave in an entirely similar manner. For example, while "where" questions with no preverb in the matrix clause are readily accepted by all speakers, as in (18), the same is not true for "why" questions. There seems to be a very strong preference for the preverb in the matrix clause of "why" questions, so that even if, in the sentence below, the intended question is why Ajike went to the market (and not why Ade told Olu this fact), there is sti,ll a clear reluctance among speakers to accept the sentence. ??nitori ki why

ni Ade FOC

so fun tell

Olu pe Ajike that

fi v

lo go

si to

oja ? market

why did Ade tell Olu that Ajike went to the market? 6.

The first point to note is that there is a general dislike of more than one preverb in any given question. Thus, for example, for two sentences involving two different wh-words with preverbs in both clauses, one speaker put a question mark in front of one (a "why" question) and an asterisk in front of the other (a "how" question). Other speakers were equally reluctant to accept sentences with two preverbs, and some of them thought it was an unnecessary repetition.

7.

The other "why" word (or phrase), nitori ki, which can be literally translated as "because of what", needs a preverb to govern its trace in adjunct position. Thus, given the sentence nitori ki why

ni Ade FOC

fi V

why did Ade beat Bola? the structure we have -is: COMP nitori ki

na Bgla ? beat

Wh-Movement and Proper Government in Yoruba 8.

125

The author is grateful to Wayne Harbert of Cornell University for the help and encouragement received in writing this paper.

REFERENCES Carstens, Vicki. 1985. Wh-movement in Yoruba. In: Studies in African Linguistics. Supplement 9: Précis from the 15th Conference on African Linguistics, UCLA, March 29-31, 1984, pp. 39-45. Carstens, Vicki. 1986. Proper government in Yoruba. M.A. thesis, UCLA. Chomsky, Noam. 1982. Lectures on Government and Binding. Dordrecht: Foris Publications. Stahlke, Herbert. 1973. Ross" constraints and related problems in Yoruba. Studies in the Linguistic Sciences, 3(1):192-230. Stowell, T.A. 1981. Origins of Phrase Structure. Ph.D. dissertation, MIT.

Department of Modern Languages and Linguistics Cornell University

Chapter 10

Typological Mixture in the Legalization of Manner and Cause in Emai Ronald P. Schaefer

The structural underpinning of serial verb sentences has perplexed investigators of the Kwa languages of West Africa for some time (Christaller 1875, Westermann 1930, Welmers 1973). In the classic definition of these structures two verbs, sharing a common subject, occur in unmarked series in a single sentence. Over the years analysis of serial constructions has been subject to various theoretical interpretations, some incompatible with one another. Nonetheless, most analyses of this topic are restricted to relatively narrow syntactic concerns. Ignored for the most part are representations in the lexicon and the lexicalization processes they presume. In order to explore the possible role of these processes in specifying the structural configuration of serial sentences, the behavior of particular verbs in Emai, an

Edoid language of

south-central Nigeria, will be examined.1 As a prelude to further discussion, let us distinguish two classes of serial sentences identified in previous studies. Sentences of one class are frequently claimed as derivatives from an underlying series of sentence structures (Stewart 1963, Williamson 1965, Bamgbose 1966, Awobuluyi 1967, Bendor-Samuel 1968). Accordingly, a Yoruba serial structure like that in (la) with the verbs se and ta is analyzed by Awobuluyi (1967) as the outcome of a derivation in which a deletion rule has applied to a conjoined, multi-sentence base of the type underlying (lb), which employs the conjunction marker si and specifies the distinct clausal origins of the verbs se and ta. (1)

a.

Bola Bola

se eran ta cook meat sell

'Bola cooked some meat to sell and. sold it' b.

Bola Bola

se eran, o si ta a cook meat, he and sell it

'Bola cooked some meat to sell and sold it' Though a multi-sentence derivation of this first class has not always been accepted (Schachter 1974), sentences belonging to the second class definitely cannot be derived via this route (Bamgbose 1974). If a sentence like (2a) from

128

R.P. Schaefei

Yoruba, for example, is derived from a base incorporating the structures in (2b), then the rule deleting the second occurrence of obe naa would change the meaning of this original structure, for the form to 'be enough' specifies the state of obe naa 'the stew' in the second clause of (2b), whereas in (2a) it modifies the relationship expressed by obe naa dun, i.e., the stew's being delicious is enough. (2)

a.

obe naa stew the

dun to be delicious enough

'the stew is delicious enough' b.

obe naa stew the

dun •, be delicious

'the stew is delicious'

obe naa to stew the be enough ; 'the stew is enough'

The predicament raised by this second class of serials is caused by a partial change in the meaning of the second verb. In a number of prior studies, this grammatical situation has been viewed as reflecting a change in syntactic function: the second verb is interpreted as a syntactically reanalyzed form, either a preposition (Lord 1973), an adverb (Stahlke 1974), or a verbid (Ansre 1966). Perhaps the most carefully and clearly stated of these positions is Ansre. According to Ansre, differences in the distributional pattern of a single phonological shape allow one to postulate parallel differences of syntactic category. The examples in (3) from Ewe nicely illustrate this point, since the form le occupies distinct syntactic positions: in (3a) le follows the subject and functions as the main verb, whereas in (3b) it follows the direct object and marks location, the form fie assuming the function of main verb. (3)

a.

agbalea le kplga dzi book-the be-at table top 1

b.

the book is on the top of the table'

me I

fie buy

afbale le Keta book at Keta

'I bought a book at Keta' To motivate further a distinction of syntactic class between occurrences of a single phonological shape, Ansre considers differences with respect to syntactic and morphological rules. For example, only in (3a) does le accept tense/ aspect marking and attract negation. Ansre thus postulates the le of (3a) as

Lexicalization of Manner and Cause in Bmai

129

a verb and the le of (3b) as an instance of the category verbid, a syntactic category akin to preposition (Lord 1973). A crucial aspect of this analysis is the assumption that sets of constraints pertaining to the behavior of a phonological shape are best accounted for by assigning the corresponding forms to distinct syntactic classes. There is, however, another solution which acknowledges these syntactic differences but accounts for them through alternating semantic representations, senses, of a single lexical entry, i.e., a single syntactic category. To illustrate this interpretation in detail, we concentrate on the relationship between lexicalization processes and the behavior of surface level morphemes in expressions of motion in Emai. Limiting analysis to a single domain will allow us to scrutinize more carefuly the contribution of each morpheme to the surface level expression of fairly similar events. For the examples that follow it should be borne in mind that though only a few Emai verbs are utilized; they are representative of their class. Linguistic expression of motion in Emai reveals sentences of the different types shown in (4).2 (4)

a.

òli ómóhè ó vbl ukpà-ódè the man move-onto at road 'the man moved onto the road'

b.

òli ómóhè là ó the man run 1

c.

the man ran onto the road1

òli ómóhè là vbi ukpa-ódè the man run at road 1

d.

the man ran on the road' òli ómóhè sua the man push

1

e.

vbi ukpà-ódè at road

èkpètè ó stool -

vbi ukpà-ódè at road

the man pushed the cart onto the road1 òli ómóhè suà ékpètè the man push stool

'the man pushed the stool' With respect to each of the sentences (4a), (4b) and (4d), notice that the final three lexical forms are identical, as are the first two.3 It is only the internal forms, la and sua ekpete, respectively, that distinguish among them. Of these common forms, that of o will be the principal focus of analysis, since

130

R.P. Schaefer

it either exhibits or sheds grammatical properties in a fashion consonant with Ansre's distinction between verbid and verb. In (4a), the form o functions as a verb, main verb in fact, but in the other two sentences, which are serial in nature (Ia...o, sua...o), the syntactic category assigned to o is not so readily apparent. That form o continues to function as a verb in these serials is supported by the fact that its tone value in different tense/aspect conjugations co-varies with the value assigned the first verb. Both la and o realize the low tone of continuous aspect in (5a) and both sua and o realize the high tone of the completive in (5b). However, the two can never differ, la being low and o high, as in (5c) , or sua high and o low, as in (5d). (5)

a.

§11 §m$h£ $ lk 6 v M u kp3t-6d£ 'the man ran onto the road'

b.

oil omohd sua 6 vbl ukpM-odS 'the man pushed the stool onto the road'

c. *dll omoh& 6 lh 6 vbl ukpk-6d£ 'the man is running onto the road' d. *Q11 gmoh£ sua SJtpStS d vbl ukp£-6d£ 'the man pushed the stool onto the road' That o may not be a verb in both (4b) and (4d), however, is suggested by its syntactic position and the fact that la and sua do function as the only verbs in (4c) and (4e) , respectively. For the moment, therefore, let us assume that o in (4b) and (4d) is a verbid. As a consequence, the behavior of o in these sentences should contrast with its behavior in (4a), i.e., it should shed some of the properties defining its verb status, in (4a) . Initial inspection of the grammatical properties of the o forms in these different construction types reveals that our expectation is partially fulfilled. Syntactic operations applying to the structure of (4a) and (4d), for instance, demonstrate that form o is indeed more constrained in the latter. Supporting this contention is the fact that o in (4a), but not in (4d), is subject to the verb focus rule of Emai. This rule, first of all, copies the verb in its gerundive form to a position leftmost in a sentence structure and marks it, in the affirmative, with the form li. Application of verb focus is seen in (6a), where the only verb in the sentence, sua, is focussed astisdSmland marked by 11. Successful application of this rule is also evident in (6b) and (6c), where the sua of (4d) and the o of (4a) occur in focussed position.

Lexicalization of Manner and Cause in Emai

131

It is particularly in contrast to the latter that verb focus is relevant, for as (6d) demonstrates, o in (4d) cannot undergo this rule.4 (6)

a.

usuami 11 òli pushing FOCUS the

ómóhè suà èkpètè man push stool

'what the man did to the stool was push it' b.

usuami 11 §11 ómóhè suà èkpètè 6 pushing FOCUS the man push stool -

vbl ukpà-ódè at road

'what the man did to get the stool onto the road was push it' c.

uòml 11 òli gmghè 6 vbl ukpà-ódè moving-onto FOCUS the man move-onto at road 'what the man did with the road is move onto it'

d. *uòml 11 e.

òli

ómóhè

suà

èkpètè ó

uòml 11 §11 Qm$hè là ó moving-onto FOCUS the man run -

vbl ukpà-ódè vbl ukpà-ód$ at road

'what the man did by running was move onto the road' How do these facts compare to the behavior of form o in (4b)? Since it is syntactically positioned rightmost in the verb series, one would legitimately expect its behavior to be as it was in (4d). That is, just as o in (4d) is preceded by a verb, sua, so in (4b) it is preceded by a verb, la, Xn either sentence, o finds itself in a serial context of greater syntactic complexity relative to the simpler environment of (4a). Despite this similarity in serial position, form o in (4b) is not constrained compared to its mate in (4d). Application of verb focus shows this to be the case, as comparison of (6e), the focused counterpart of (4b), with the ungrammatical (6d), the counterpart of (4d) , establishes. Dissimilar grammatical behavior, therefore, characterizes the phonological shape o in (4b) and (4d), which the distributional criterion of Ansre assigned to the category verbid. Presented with these facts, one could argue that too much emphasis is being accorded the verb focus construction. After all, the facts above, though true, may be isolated and reflect idiosyncratic properties of the lexical items involved. A more secure foundation would rest on a wider pattern of behavior, one encompassing a number of construction types. In fact, examination of the verbal constituents of sentences like (4b) and (4d) in still other construction types allows the differential behavior of o within serial structures to emerge with greater clarity. As a first step toward supporting this position, consider that the o forms

132

R.P. Schaefer

in (4b) and (4d) behave differently with respect to the process of consecutivization, which carries the meaning 'and then'. For instance, the consecutive structure corresponding to (4b) allows the form o to be preceded by the consecutive marker

as (7a) indicates, but there is no consecutive structure

corresponding to (4d), as revealed by (7b). (7)

a.

SII the 1

ómóhè là $ ó man ran CONS -

vbl ukpà-ódè at road

the man ran and then moved onto the road'

b. *òll ómóhè sua èkpètè ò ó the man push stool CONS -

vbl ukpà-ódè at road

Likewise, the o forms in (4b) and (4d) do not behave similarly with respect to negation. When the potential ambiquity of the simple negative forms corresponding to (4b) and (4d) are fleshed out in dual clause structures, only the former alloVs the form o to function as the sole verb of the first or second clause. In this regard, examine first the simple negative structures corresponding to (4b) and (4d), (8a) and (8d), respectively, where the form 1 in post-subject position marks sentence negation. Next, consider the corresponding disambiguated dual clause structures, where the second clause is marked for negation and the two clauses are joined by the marker bl khl 'with that, but'. (8)

a.

òli the

ómóhè 1 là ò man not run -

vbl ukpà-ódè at road

' the man did not run onto the road ' b.

òli ómóhè là bl khl the man run but

6 1 ò vbl ukpà-ódè he not move-onto at road

' the man rein but he did not move onto the road ' c.

òli ómóhè ó vbl ukpà-ódè bl khl the man move-onto at road but

ó 1 là he not run

'the man moved onto the road but he did not run' d.

òli ómóhè 1 sùà $kpètè ò the man not push stool

vbl ukpà-ódè at road

'the man did not push the stool onto the road' e.

òli the

ómóhè sua èkpètè bl khl man push stool but

ó 1 sùà òl he not push it

ò -

vbi ukpà-ódè at road

'the man pushed the stool but it did not move onto the road'

Lexicalization of Manner and Cause in Emai f. *òlx ómóhè sua èkpètè bl khl the man push stool but

133 ó 1 6 vbì ùkpà-ódè it not move-onto at road

g. *òlì èkpètè ó vbl ùkpà-ódè bl khl the stool move-onto at road but

òli the

ómóhè i sùà di man not push it

What is revealing about the resulting disambiguated structures is their nonparallel form. For instance, (8a) is disambiguated in the fashion of (8b) and (8c), o or la occurring as the only verb in either clause and attracting the negative marker 1 in the second clause. In contrast, (8d) cannot be disambiguated in a parallel fashion, since o, as (8f) and (8g) indicate, cannot assume main verb status in the first or second clause. Only (8e) is grammatically correct, where both verbs of this serial configuration, sua and o, occur in the same clause.5 The differential behavior of the o forms in (4b) and (4d) could not be more evident. What are we to make of this? Should only the o form of (4d) be considered a verbid, thus ignoring the distributional criterion which originally established the difference between (4a) and the pair (4b) and (4d)? Advocates of the verbid position, in an attempt to retain their concept, could argue that the o forms in (4b) and (4d) do not in fact exist in comparable syntactic environments, since there are actually two syntactic positions: o follows an intransitive verb in (4b) and a transitive verb and its object in (4d). However, this begs the question of whether it is only a syntactic issue, verbid vs verb, that is being raised. It is important to remember, after all, that the syntactic environment for o is different precisely in that there is a preceding manner conveying form in (4b) and a preceding causative form in (4d). Semantic considerations such as these no doubt led to Lord's (1974) analysis of serial sentences. She introduced two rules for semantically interpreting serial constructions, repeated below for ease of reference. A.

The VP's all refer to subparts or aspects of a single overall event.

B.

The second VP is always in some sense a further development, result or goal of the first VP.

Application of these rules to sentences (4b) and (4d) leads naturally to the recognition of the second verb, o, as a further development, result or goal, of the initial verb, either la or sua. At an intuitive level this is appealing, but the fact of the matter is that o does not express a further development for the

134

R.P. Schaefer

verb la in the same way as it does for the verb sua, since in (4b) the spatial relationship of the subject,

'man', is specified relative to ukp£-6dS

'road', whereas in (4d) it is the relationship of the direct object, 'stool', that is specified relative to ukph-6d&. This is exactly the point, what does "further development" or "result" mean is these instances. One senses that Lord's rules nudge us in the proper direction, but a more precise delineation of the notion "subparts of a single overall event" is required if a more adequate understanding of serial structures is to be attained. In other words, what is required is a finer understanding of how the semantic characterization of non-linguistic events is apportioned among lexical forms. A potentially useful step toward the goal of more precisely delineating motion events at the semantic level is available in Talmy (1975, 1985). For Talmy, a simple motion event is defined as one object moving (or located) with respect to another background object. It is delineated by the semantic components FIGURE, MOTION, PATH and GROUND, where FIGURE refers to the moving object (or object being located) and GROUND refers to the background object against which the FIGURE'S movement or location is being specified. The remaining components can best be illustrated in (9a), where a simple motion event is registered. (9)

a.

The boy moved onto the road,

b.

The boy is on the road.

In (9a), boy and road specify FIGURE and GROUND respectively, onto specifies the PATH, the course which the moving object follows, and the verb moved specifies the component MOTION, the fact that movement is being conveyed. A nonmotion event of positional location like (9b) would be similarly analyzed except that the component BE-LOCATED (BE-L) rather than MOTION is specified by the verb is and the component PATH is realized by the preposition on. As is well known, a simple motion event can be lexically packaged in a different fashion, though as Taimy points out, there are typological limitations to the resulting composition. Central to this system of limitations or constraints is the component MOTION and its tendency to incorporate with other elements in a main clause verb root. For ease of reference, these constraints will be illustrated with English sentences, although the resulting lexicalization patterns can hardly be maintained as its characteristic type (Talmy 1985). An alternative lexicalization of the event encoded by (9a) is (10) , where our analytic focus is the main verb root. In (10a) the components MOTION and PATH are incorporated in the verb enter, as it means 'to move into', the remaining

Lexicalization of Manner and Cause in Emai

135

nominal forms boy and house realizing FIGURE and GROUND, respectively. (10)

a.

The boy entered the house.

b.

It rained into the house.

Still another lexicalization pattern, though one involving a different motion event, is shown in (10b). The verb root rain in this instance incorporates the components FIGURE and MOTION, i.e., the rain itself is the moving object, and the preposition into and the nominal room specify PATH and GROUND, respectively.6 As demonstrated by these examples, a simple motion event can be lexicalized in either of three patterns with respect to the main verb root: specify only MOTION, incorporate MOTION+PATH or incorporate FIGURE+MOTION. Beyond these relatively simple structures, motion events exhibit more complex packaging, since languages lexicalize various manners and causes of movement. With respect to these relatively more complex motion events, Talmy utilizes the components MANNER and CAUSE and finds that across a wide variety of languages they incorporate with the element MOTION in a few consistent patterns and that, typically, a language will treat these two components via the same lexicalization type. In one of these types, either MANNER or CAUSE incorporates with MOTION in the main clause verb root and a minor grammatical category specifies PATH. Languages which characteristically rely on this type of lexicalization are Chinese and most of Indo-European, except Romance. Accordingly, in (11a) the main verb run in English incorporates both the fact of motion and the manner in which that event is carried out, a MOTION+MANNER lexicalization type, while the preposition onto incorporates PATH, the course which the FIGURE'S movement traced. Similarly in (lib) the main verb push incorporates the fact of motion applying to the FIGURE chair, and the cause of that movement, the boy's pushing, while the preposition onto conveys PATH. A schematic representation of this lexicalization type is shown in Figure 1. (11)

a.

The boy ran onto the road.

b.

The boy pushed the stool onto the road.

Figure 1: FIGURE

Lexicalization of MOTION+MANNER

MOTION |

PATH

verb root

GROUND

j MANNER 1 {CAUSE J

136

R.P. Schaefer

A contrasting lexicalization type is illustrated by Romance languages of Indo-European, as well as Samoan and Semitic. Instead of MANNER and CAUSE incorporating with MOTION in the verb root, the former components are realized by separate lexical items of an adjunct character. The main verb of the sentence, meanwhile, incorporates both MOTION and PATH, as the Spanish examples in (12a) and (12b) illustrate. (12)

a.

la botella entró the bottle moved-in

a la cueva (flotando) to the cave floating

'the bottle floated into the cave' b.

quit-é el papel del paquete cortándolo moved off-I the paper from package cutting it 1

1 cut the wrapper off the package1

Thus, while FIGURE and GROUND are specified by the nomináis botella and cueva, respectively, in (12a), the main verb entro incorporates MOTION as well as PATH, and the optional adjunct flotando specifies only the MANNER of the event, 'by floating'. Likewise in (12b) the main verb quit- incorporates MOTION and PATH, 'move off', and the adjunct cortándolo 'cutting' specifies the cause of the motion event, where papel functions as FIGURE and paquete as GROUND. And as was true of the previous pattern, MANNER and CAUSE lexicalize in a similar typological fashion with respect to the verb root of the main clause, as illustrated by figure 2. Figure 2. FIGURE

Lexicalization of MOTION+PATH

MOTION PATH | | '

GROUND

I MANNER I (CAUSE J

verb root This brief digression, though it does not do justice to the extent of Talmy's cross-linguistic investigation, does provide us with a means for more clearly analyzing the semantic role of verbal elements comprising serial configurations for motion in Emai. Most important in this regard is the relationship between the semantic composition of the o forms in the motion events recorded by sentences (4b) and (4d) and the distinguishing sets of constraints. These constraints, that is, correlate with specific lexical structures in a way which argues that motion events in Emai are lexicalized by a mixture of typological patterns governing the incorporation of the component MOTION: both of the typological

Lexicalization of Manner and Cause in Emai

137

patterns outlined by Talmy and depicted above in figure 1 and figure 2 are utilized. To be more specific, motion constructions expressing manner, i.e., (4b) with la, allow the component MOTION to incorporate with PATH in form o, as is typical of languages like Spanish. As a consequence of incorporating MOTION, this form, in agreement with simple motion expressions like (4a), exhibits the properties of a verb with respect to focus, consecutive marker placement and disambiguation of negatives. The resulting analysis of semantic components among the forms la and o would be as in (13). (13)

òli ómóhè là- ó vbl ukpà-ódè the man run move-onto at road 'the man ran onto the road1

The contrasting lexicalization pattern, the one also found in English, characterizes motion constructions expressing cause. For these, the component MOTION incorporates with CAUSE in the verb root sua, while the more constrained behavior of form o in these structures derives from its specification of only the component PATH. It is thus the failure of o to incorporate MOTION in the sua type structure which leads to constraints with respect to verb focus, consecutive marker placement and negative disambiguation. For these causative constructions, a more precise specification of semantic components for the forms sua and o is reproduced in (14). (14)

òli ómóhè suà èkpètè ó vbl ukpà-ódè the man cause-by-pushing-to-move stool onto at road 'the man pushed the stool onto the road'

As the immediately preceding examples have attempted to illustrate, the verbid quality of o in motion constructions is due to a mixture of lexicalization processes rather than differences of syntactic category. This is not to say that syntactic reanalysis does not occur, only that greater caution should be employed in reaching such a conclusion. For a form like o, the lexicalization approach has the advantage of capturing the semantic similarity between the senses 'move onto' and 'onto' within the lexical representation of a single syntactic form. Moreover, since this analysis is not confined to o, occurring with laqaa 'to move around', raa re 'to move past', fan ze 'to move across', among others (Schaefer 1986), it captures a significant lexical generalization

138

R.P. Schaefer

in the grammar of Emai.7 Typological mixture, therefore, is the characteristic means for lexicalizing different types of events in the motion domain of Emai. Analyzing the distributional and grammatical peculiarities of forms like o in terms of lexicalization rather than a reanalyzed syntactic category may have wider application. Though it is difficult to determine without further analysis, alternating incorporation patterns may underlie Ansre's (1966) analysis of the verbid le in Ewe, for in simple sentences of positional location this form appears to incorporate (BE+L+PATH) and in serials simply PATH. At a very tentative level, one is then able to offer a lexical generalization for Kwa which would recognize that when verbs of position, incorporating BE-L+PATH, and of motion, incorporating MOVE+PATH, participate in particular serial configurations, there is a tendency for only the PATH component to be retained. A question which immediately arises from this generalization and the particular lexicalization facts found in Emai concerns why the other possible typological mix of lexicalization processes did not occur. One imagines that the incorporation patterns could just as well have been reversed, i.e., manner constructions utilizing the English pattern, (MANNER+MOTION)...PATH, and causative constructions utilizing the Spanish pattern, (MOTION+PATH)—CAUSE. Perhaps these may even exist within other languages of the Kwa family. Such a wideranging examination obviously cannot be attempted here, but let us suppose for the moment that there is no Kwa language which employs the Spanish pattern for causative and the English pattern for manner. If this were true, one could look for its antecedents in historical reconstruction of domains such as motion. Future research may profitably be directed toward such an examination. On a larger scale encompassing languages of the world, if only the Emai class of typological mixture existed, one would be inclined to postulate a hierarchical relationship between the components MANNER and CAUSE and their incorporation with MOTION. A simplistic formulation of this hierarchy would hold that if a language incorporates MANNER with MOTION, then it also will incorporate CAUSE with MOTION, but not necessarily the converse. The Emai pattern, at least, suggests the plausibility of such a hierarchy and of the propensity for CAUSE to incorporate with MOTION. In summary, serial verb constructions in Emai were explored in order to demonstrate that typological mixture characterizes its lexicalization of motion events. Of particular focus was a class of Emai forms, to a degree akin to what have been called verbids, which in non-serial sentences exhibit verb-specific properties, maintain those properties as the second verb of a manner serial

Lexicalization of Manner and Cause in Emai

139

construction but shed those same properties as the second verb of a causative serial. To account for the resulting system of constraints, a mixture of lexicalization processes, each characteristic in a different language family, was invoked. In one process, typical of languages like Spanish, MOTION and PATH incorporate in the second verb in series, the preceding serial verb specifying only MANNER. In the other, typical of languages like English, the second verb in series specifies only PATH, the preceding verb incorporating MOTION and CAUSE. Differences in lexicalization rather than syntactic category were thus postulated to account for the peculiar syntactic behavior of the second verb in serial constructions.

NOTES •Preparation of this study was supported at different stages by a Faculty Research Grant from the University of Benin, Nigeria, and NICHHD Postdoctoral Training Grant #HD07255 administered by the University of Kansas. Special thanks are due my Emai assistants, Francis Egbokhare and Gabriel Egeruan. 1.

Emai is classified as a member of the North Central Branch of Edoid, in turn a branch of Benue-Kwa, by Hansford, Bendor-Samuel, and Stanford (1976).

2.

The blank (-) in (4b) and (4d) is intentional, as it is our purpose to determine what semantic specification is appropriate in the respective syntactic environments.

3.

The Emai data are presented in an orthographic representation suggested in Schaefer (n.d.), which follows the general conventions for Edoid by using "vb" for a voiced bilabial approximant. Lax mid vowels are represented as "e" and " 9 " .

4.

It is also not possible for o to adjoin with sua in focused position, i.e., *usuao, though it can adjoin with la, i.e., ulao.

5.

In order to place only the verb sua 'pushing' under the scope of negation an explicit verb of causation like nwun 'put' is required in the first clause in combination with o, and sua would occur as the only verb in the second clause and attract negation.

6.

Navajo and languages of the Hokan family in Northern California such as Atsugewi employ this as their characteristic lexicalization pattern.

7.

These processes also occur with other causative forms like si 'to pull', re gbe 'to kick' and manner forms like sio 'to crawl', tin 'to fly'.

140

R.P. Schaefer

REFERENCES Ansre, Gilbert. 1966. The verbid - a caveat to serial verbs. Journal of West African Languages 3(1): 29—32. Awobuluyi, Oladele. 1967. Studies in the Syntax of the Standard Yoruba Verb. Ph.D. dissertation, Columbia University. Bamgbose, Ayo. 1966. A Grammar of Yoruba. Cambridge, England: Cambridge University Press. Bendor-Samuel, John T. 1968. Verb clusters in Izl. Journal of West African Languages 5 (2):119-138. Christaller, J. G. 1875. A Grammar of the Asante and Fante Language Called Tshi. Basel. (Republished 1964, Ridgewood: Gregg.) Hansford, Keir, John Bendor-Samuel, and Ron Stanford. 1976. Studies in Nigerian Languages No. 5: An Index of Nigerian Languages. Ghana: Summer Institute of Linguistics. Lord, Carol. 1973. Serial verbs in transition. Studies in African Linguistics 4:269-296. Lord, Carol. 1974. Causative constructions in Yoruba. Studies in African Linguistics Supplement 5:195-204. Schachter, Paul. 1974. A non-transformational account of serial verbs. Studies in African Linguistics, Supplement 5:253-270. Schaefer, Ronald P. 1986. Observations on reference object geometry in Emai path expressions. In: Proceedings of the Twelfth Annual Conference of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society. Schaefer, Ronald P. n.d. Suggestions for an Emai orthography. Manuscript prepared for Ministry of Education, Owan Local Government Area, Bendel State, Nigeria. Stahlke, Herbert. 1974. Serial verbs as adverbs: a reply to Paul Schachter. Studies in African Linguistics, Supplement 5:271-277. Stewart, John M. 1963. Some restrictions on objects in Twi. Journal of African Languages 2 (2):146-149. Talmy, Leonard. 1975. Semantics and syntax of motion. In: Syntax and Semantics Volume 4, J. Kimball (Ed.), pp. 181-238. New York: Academic Press. Talmy, Leonard. 1985. Lexicalization patterns: semantic structure in lexical forms. In: Language Typology and Syntactic Description Volume III: Grammatical Categories and the Lexicon, T. Shopen (Ed.), pp. 57-149. New York: Cambridge University Press. Welmers, William. 1973. African Language Structures. Berkeley & Los Angeles: University of California Press. Westermann, Diedrich. 1930. A Study of the Ewe Language. Translated by A.L. Bickford-Smith. London: Oxford University Press. Williamson, Kay. 1965. A Grammar of the Kolokuma Dialect of Ijo. Cambridge, England: Cambridge University Press.

Department of English Language and Literature Southern Illinois University at Edwardsville

Chapter 11

The Extra-high Tone of Kisi: Just Another Tone or a New System of Prominence? G. Tucker Childs 1.

INTRODUCTION Prominence systems, particularly those of Bantu, have been the subject of

considerable descriptive and theoretical interest. This paper devotes itself to one of the central issues in the discussion of African prominence systems, the relationship between tone and accent. In that both tone and accent may draw on the same phonetic material, the basic distinction that is usually made between the two is that tone is paradigmatic and accent is syntagmatic. In a paradigmatic system, any number of units can have equal prominence, e.g., a series of high tones is possible. The VALUE of the tone unit is crucial. In a syntagmatic system, on the other hand, the LOCATION of the accented unit is crucial; only one unit in a phonological phrase is allowed to have prominence, e.g., a sequence of two accented syllables in a phonological phrase is ungrammatical. It is not totally unexpected that an accentual system would develop from a tonal system. The origin of a tonal system has elsewhere been attributed to the re-interpretation of a segmental contrast, as noted by researchers in Asian languages, e.g., Matisoff (1973). An accentual system, however, usually develops from other suprasegmental patterns: from a re-analysis of tone, or from the lexicalization of intonation (Beckman 1984). In the case of Kisi, an Atlantic language spoken in the Republic of Guinea, the tonal system, at the least, is undergoing realignment. An extra-high tone, which cannot be easily interpreted as resulting from tone-splitting (Haudricourt 1972) or from some other form of rephonologization (Hyman 1976) , is significantly higher in pitch than the regular high tone and is severely restricted in its distribution. The evidence supporting the claim for accentual reanalysis in Kisi consists of comparative evidence, both external and internal, as well as acoustic evidence. When closely related languages display different systems of prominence, the languages can be arrayed in an order representing probable stages in the evolution of the given system. This "processural" approach (Greenberg 1978) coupled with the technique of "intra-genetic synchronic comparison" (Greenberg 1976) allows us to posit stages in the change of a language from one system to another.

142

G. Tucker Childs

Occasionally similar variation occurs internally, and the same sort of analysis can be applied.

2.

BRIEF REVIEW OF THE LITERATURE In this section I discuss recent work on historical tonology, essentially

with reference to African languages. I next discuss the experimentalist methods that have been used to investigate changes in prosodic systems, particularly the work of Hombert and Lehiste. 2.1

Historical tonology

Recent reviews of work on historical tonology are contained in Maddieson (1974, 1978), Hyman (1978), Clements and Goldsmith (1984), and Hombert (1984b). The types of development that are discussed most frequently and that occur most commonly involve rephonologization, either increasing or decreasing the number of tones in a system (Hombert 1984b). The particular phenomenon that I discuss, the rise of an "extra-high" tone, is not an uncommon one. Hombert suggests that a third tone often arises as a variant of a high tone (1984b:179). Clements also remarks on such developments and notes that these inventory expansions are "usually regarded by investigators as late developments" (1977:178). That is, the extra-high tones are additions to a system already installed. Hyman (1976) and Clements (1977) both document the rephonologization of a sequence of high-downstep-high as extra-high-high. Many others have discussed the importance of the loss of a prevocalic voicing distinction on the tone of the following vowel (Hombert et al. 1979), thereby precipitating the addition of another tonal contrast, occasionally in the extrahigh range. 2.2 Experimental historical tonology The program of experimental historical phonology advocated in Ohala (1974) has been followed with respect to tonogenesis in the work of Hombert, e.g., Hombert (1975). Surveying the wide incidence of high tones after what were historically voiced consonants and low tones after historically voiceless consonants, these tones appearing after the disappearance of a voicing contrast, Hombert presents production and perception data to outline how such a change could have occurred. Acoustic measurements show that such differences can be perceived. Although Hombert has somewhat altered his opinion concerning tono-

The Extra-high Tone of Kisi

143

genesis, he continues to tout experimental methods. In his more recent work Hombert has called for renewed efforts in using laboratory techniques to develop theories of language change (1984a). Lea (1973) also uses experimental techniques to investigate tonal developments; Lea seeks to explain tonogenesis, tonoexodus, and downdrift in terms of the effects of suprasegmental and segmental influences on FO contours. Maddieson (1977) also relies on experimental means, seeking to explain universal tendencies in terms of perception. The work most closely resembling the study described in this paper is Lehiste (1983), which similarly examines an innovation in the prosodic domain. In Estonian a third contrast in length, "overlong" (versus short and long), has been described by other researchers. Lehiste finds that the contrast is actually signaled by differences in FO, and she calls this innovation evidence for the emergence of an accentual system in which stress, quantity, and tone all play a part, rather than simply quantity (in a three-way paradigmatic contrast). In support of her hypothesis she adduces evidence from synchronic and diachronic studies of related languages, as well as evidence from perceptual experiments. It is this sort of treatment that will be pursued below.

3.

EVIDENCE FOR ACCENTUAL RE-ANALYSIS

3.1

Internal evidence

The peculiar characteristics of the extra-high tone were apparent when it was contrasted with the "regular" H and L tones of the language and the system that they formed. In such sentence pairs as those analyzed below, the difference between the high and extra-high tone is clear. The extra-high tone is higher in pitch, louder, and longer than the high tone, and informants adamantly proclaim the two non-low tones different. That the extra-high tone represents an innovation, promoting a re-analysis of the system is suggested by several factors. The first is the "extra" phonetic substance it carts around, making it more prominent than the other tones. Furthermore, that the extra-high tone is not a new mid-tone or an extra-low tone is also significant; as an extra-high tone, it has greater perceptual saliency. "Extreme tones are highly marked" (Maddieson 1978:342). As has been remarked more than once, marked tones are generally found in the upper part of the pitch scale used for tones (Maddieson 1978:343). This is particularly true with respect to a language such as Kisi, noted for its dramatic downdrift (Welmers 1976:148).

144

G. Tucker Childs

We see, then, that the extra-high tone has considerable prominence. The second internal fact that suggests re-analysis is the severely limited environment in which the extra-high tone appears. The extra-high tone functions concomitantly with a negative particle to signal Habitual/Past negation. I give an example below. (1)

Habitual

a data ya L LL H you condemn me

Past

a data ya

Hab/Past Neg

a data ya le L LH+ H H

Note:

H+ = extra-high tone

L LH H 'You never condemned me./ You didn't condemn me.1

Disregarding the negative particle le H for the moment, we see that the only difference between the Affirmative Past and the Negative Past is the contrast between the /H/ and /H+/ tones. The two tone sequences associated with the disyllabic verb data above are /LH/ (Past) and /LH+/ (Negative Past/Habitual). It is this contrast that will be evaluated below. A final internal fact which suggests that re-analysis is taking place comes from the difference between noun and verb tonology. The basic fact is that verbs generally have no lexical tone while nouns do; verbs are assigned tones by the morphology, and the contrasts there have a syntagmatic flavor.1 Tonal contrasts on verb stems are limited in that only three of four possible sequences are allowed (excluding the imperative), as shown below. (2) LL LH HL HH

Nouns pele lomso lolo masa

'leg' 'insect' 'bladder' 'yaws'

Verbs data data data *

'condemn (Habitual)' 'condemn (Past)' 'condemn (Neg Hortative)

There are further restrictions on allowable tone sequences on Kisi verbs. As shown in the examples below, there is only one tonal change per verb; in this case the second and third syllables always have the same tone. (3)

a a a a

fasanga L LHH mangala L LHH fasanga le L HLL H mangala le L HLL H

'they 'they 'they 'they

splashed' played magic' should not splash' should not play magic'

(Note: "ng" stands for the velar nasal)

The Extra-high Tone of Kisi

145

That only one pitch change (H to L or L to H) appears on verb stems recommends an accentual treatment.2 The same restrictions obtain in other verbal environments. In sum, then, we see considerable reduction in possible forms in the verbal tonology, allowing for the possibility of an accentual reanalysis. 3.2

Intra-genetic evidence

Comparative evidence suggests that the Kisi phenomenon may be part of a larger trend in Atlantic. It has been claimed that high and low tones are genetic in Atlantic and that the language group is unusual in that the tone system seems to be disappearing (Hombert 1984b:119). Nothern Branch languages, e.g., Fula and Wolof, differ from Southern Branch languages, e.g., Kisi and Temne, in that the Southern Branch languages are predominantly tonal, while the Northern are not. That tone may be disappearing even in the Southern Branch is suggested by a comment on Sherbro. In Sherbro, the Southern Branch language most closely related to Kisi, it has been reported that tonal contrasts are becoming "less prominent" (Hansen 1979:34). Despite the obvious trend, two Northern Branch languages have apparently developed a three-tone system. N1Diaye-CorrSard (1970) uses three tones in describing Balante, a Northern Branch language. Ferry (1981) reports that another Northern Branch language, Basari, has three level tones.^ It is not clear whether or not another analysis, e.g., one involving downstep, is possible for these languages. Such a treatment has been postulated for Temne (Wilson 1968). Nonetheless, we must recognize that integrating a third tone into the language, as the third member of a paradigmatic opposition, represents a viable alternative to developing a new prominence system. A third attested possibility, a combination of tone and stress, is claimed for Gola, another Southern Branch language (Westermann 1921). Westermann requires two level tones in his treatment, a high and a low (or sometimes a mid-low). Interestingly, the environments in which he invokes stress are comparable to those where the extra-high tone is needed in Kisi. In Gola the negative has high tone and stress (1921:66f). "Stress" never appears with low-toned vowels and thus seems quite similar to accent. Unfortunately, Westermann does not provide much discussion of the phonetics of stress in Gola, and thus it is impossible to know how comparable the situations actually are. Another feature shared by stress and high tone in Gola and by Kisi extra-high tone is limited distribution. In Gola the combination of stress and high tone is found only in the negation of Perfect, Progressive, and Future; more commonly, negation is accomplished by tonal changes and the negative particle go. Thus it seems that a

146

G. Tucker Childs

look at the phonetics of Gola stress (or accent) would certainly be instructive as regards the situation in Kisi. Nonetheless, on the basis of developments elsewhere, the evidence is relatively straightforward and favors seeing Atlantic as becoming accentual. The Northern Branch languages, comprising the preponderance of languages belonging to Atlantic, are overwhelmingly accentual (Sonja Fagerberg-Diallo, p.c.). Fula, Balanta (Ganga), Diola-Fogny, and Wolof, the major languages of the Northern Branch are all reported as having dominant initial stress (Hyman 1977:59). Evidence from outside Atlantic but within Niger-Congo furnishes further justification for our seeing accentual reanalysis as taking place. Proto-Benue-Congo has been reconstructed as being tonal with a high and a low tone. In Bantu it has often been the case that in purely tonal languages, accent first appears in the verbal morphology, primarily due to the fact that verbs are monosyllabic and allow for the addition of verbal extensions. On the basis of evidence from the verb alone, one could not conclusively identify proto-Bantu as a tone language, since the verb could be described in terms of a system which assigns an underlying feature of accent to certain formatives and not to others. The evidence that Proto-Bantu was tonal rather than accentual resides largely in the nominals (nouns and adjectives). (Clements and Goldsmith 1984:3) They further claim that it is primarily through the verbal morphology that the accentual system first enters Bantu. Thus it may be that tone becomes accent in Atlantic via the same route. Before pronouncing judgement on Kisi, however, we need to look at some instrumental measurements.

4.

METHOD In this preliminary study the focus is on the high (H) and extra-high (H+)

tones of Kisi. The data consist of fifty sentences recorded in the Phonology Lab at the University of California, Berkeley using a sound booth and high quality recording equipment. Vowel quality was held constant; all syllables carrying the focus tones contain only the short vowel /a/. As much as possible, sentences were chosen so that the syllable of focus was open and has as onset a voiceless obstruent. The verbs are all monomorphemic, i.e., have no verb extensions. The sentences consist of three types, each type containing sentences with

147

The Extra-high Tone in Kisi the high or extra-high tone, as shown in Figure 1. Figure 1. Type I

Types of sentences used for acoustic measurements Disyllabic verbs. Affirmative Past L L H H a data ya

Type II

'they condemned me'

Disyllabic verbs, Negative Habitual/Past L L H+ H H a data ya le

Type III

'they don't/didn't condemn me'

Trisyllabic verbs. Negative Habitual/Past L L H+H+ H H a fasanga ya le

'they don't/didn't splash me'

The tonological environments in which each token (of high or extra-high tone) occurred are schematized in Figure 2. "S2" stands for the second syllable of the verb, and "S3" for the third. "SI", the first syllable of each verb, is always low in these tenses. Figure 2.

Tonological environments of H vs H+ contrast S2

Disyllabic:

H /

# L _I

H+ /

"

*

S2 Trisyllabic:

H

# L # H S2

S3

H /

L

H #

H+ /

L

H+ #

/

H H / H+ H+

S3 # # # #

L H L H

In disyllabic verbs every high or extra-high tone (S2) is preceded by a L tone (SI) and can be followed by either a H or L tone. With respect to trisyllabic verbs, the high or extra-high tone is assigned to both the second and third syllables. This means that in S2 the high or extra-high tone is still preceded by a low tone, but in this case can be followed only by a copy tone, that is, a high by a high and an extra-high by an extra-high. Restrictions on the environments in which tones on S3 occur obtain in a mirror-image situation; the high or extra-high can only be preceded by a copy tone.

148

G. Tucker Childs

Once the sentences were recorded, they were digitized and analyzed using the Chi Program developed by Steve Pearson of Chi Systems. The Chi Program allows readings for fundamental frequency, duration, and loudness every 10 milliseconds. For this part of the experiment only readings of the high versus the extra-high tone were used.

5.

RESULTS AND DISCUSSION In Table 1 are summarized the results of measuring F0, duration, and loudness.

The average values are given for the second vowel in disyllabic verbs. The differences in the average values for the three parameters, as given in the last line above, are all well above the level of just noticeable difference for each parameter (Lehiste 1970). Table 1.

Avg N H+ - H =

Average values for FO, duration, and loudness for high and extra-high tones, S2 in disyllabic verbs. Differences in averages (H+ - H) FO

Duration

Loudness

(Hz)

(msec)

(dB)

H

H+

H

H+

H

H+

144 204

155 306

193 27

199

80.0 204

82.9 306

11.6 Hz

21

6 msec

2.9 dB

It should be mentioned that the last syllable of the verb for both two- and three-syllable verbs seems overly long in comparison to other syllables. The vowels measure almost 200 msecs while average vowel length is approximately 110 msecs. The lengthening is probably a function of the careful pronunciation of a laboratory situation. This contamination, however, should not affect the differences between the two values. In Table 2 are given z-values for the differences between the means for the three parameters of F0, duration, and loudness. The z-value is a statistical measurement of whether or not the difference in means is great enough to be accepted at a predetermined significance level, here 0.05. For a p < 0.05, a z-value of 1.96 is needed. As expected the z-value for fundamental frequency is well above the z-value of 1.96 needed for significance. It is notable that the z-value for loudness is also well above the significance level. The differ-

149

The Extra-high Tone in Kisi ence in duration clearly is not great enough. In that the number of measurements for duration is an order of magnitude less than the number for FO and

loudness, with more measurements of duration the confidence level could possibly be reached. Table 2.

z-values for parameters of FO, duration, & loudness H versus H+ in disyllabic verbs FO

Duration

Loudness

15.96

.708

4.31

In Table 3 are shown average values of fundamental frequency, duration, and loudness for trisyllabic verbs. All of the averages are as expected with one exception. The one unexpected feature of this display is that the second syllable of the verb with an extra-high tone is actually shorter than the second syllable of the verb with a high tone, i.e., 61 msecs for the extra-high-tone versus 95 msecs for the high-tone verb. This anomaly is shown in the display of differences in the values in Table 4 and is schematized in Figure 3. Table 3.

Average values for FO, duration, and loudness H vs. H+ in trisyllabic verbs FO (Hz)

Avg N =

Duration (msec)

H S2 S3

H+ S2 S3

144 140 57 100

153 153 73 161

Table 4.

S2

H S3

9 5 165 18 18

S2

Loudness (dB)

H+ S3

8 1 195 12 12

H S2

S3

81.2 78.8 56 100

Differences betwfeen H+ and H for F0, duration, & loudness, (H+ - H), trisyllabic verbs F0 (Hz)

(H+) - (H)

S2 S3

8.3 13.4

Duration (msec) -14 30

Loudness (dB) 2.1 1.7

S2

H+ S3

83.3 80.5 72 173

150

G. Tucker Childs

Figure 3.

Schematic representation of syllable shortening and lengthening in tri-syllabic verbs. VI 95

C —

V2 —

165

VI + V2 — 260 msec

—

81

—

—

195 —"

H+

276 msec

Note also that the F0 difference between H+ and H is greater for the third syllable (13.4 Hz) than it is for the second syllable (8.3 Hz). This fact may give S3 even greater prominence than S2. Statistical analysis performed on the differences between these three sets of values show them to be significant. Z-values are given in Table 5 for the three parameters for trisyllabic verbs. All but the difference in means for the duration of the second syllable are significant above the 1.96 z-value needed for a 0.05 level of confidence. That the duration values are once again lower than those for the other two parameters can be explained by reference to the low number of tokens for this measurement. Nonetheless, the values for duration and loudness are significantly different, as would not necessarily need to be the case in a system relying solely on differences in F0. Table 5.

Z-values for parameters of F0, duration, and loudness H vs H+ in trisyllabic verbs FO

S2 5.61

S3 11.26

Duration

Loudness

S2

S3

S2

S3

2.16

1.74

3.58

2.90

A number of conclusions can be drawn on the basis of the above data. First of all, fundamental frequency is clearly important in the signaling of grammatical differences represented by the contrast between the high and extra-high tone. It also seems clear that amplitude is involved, and probably duration as well, in a rather more complex manner than was originally envisioned. In trisyllabic verbs the lengthening of the third (and final) syllable and the shortening of the second may be functioning to give the final syllable even greater prominence than would be achieved by lengthening both syllables.

The Extra-high Tone of Kisi

151

The differential effects suggest that we no longer have a strictly paradigmatic system. A similar effect is achieved by raising the second of the two extrahigh syllables more than the first. Instead of giving equal prominence to two syllables, dominant prominence is given to only one, suggesting that only one syllable is allowed to receive prominence. It is here that the indications of a changeover to a culminative and syntagmatic system are clearest. Thus, it appears that the phonetic "seeds" (Hombert et al. 1979) for a changeover to a new system have been sown. The next step in the investigation will involve perceptual experiments, testing whether or not contrasts of duration and intensity have yet to become significant for Kisi speakers. Further instrumental analyses will determine if indeed this is the only place in the prominence system where such contrasts are exploited, that is, whether or not differences in intensity and duration are used to signal other "tonal" differences. In order to fully substantiate this position, it needs to be shown that the contrast between the basic high and low tones is not also one cued by duration and intensity. The question arises as to whether the nascent system is one with three tones, or is one with two tones and accent, with the accented syllable employing contrasts other than those of fundamental frequency. In that the extra-high tone is higher than the high tone, a paradigmatic system is suggested. In that it is also longer and louder hints that there might be differences other than FO registering the contrast. In that one syllable is shortened and another lengthened and FO builds up to the final syllable, i.e., one syllable of the extrahigh sequence stands out more than the other, suggests a syntagmatic relationship. This relationship holds only with respect to the extra-high tone and thus represents a QUALITATIVE change in the prominence system of Kisi.

6.

CONCLUSION In conclusion I would like to sketch out the stages I project in the change-

over from a tone system to an accentual one. By doing so, I do not mean to suggest any inevitability about the process, nor to claim any exclusiveness about either system. Many languages exploit the resources of accent and tone simultaneously. The first stage in such a changeover is a situation in which both verbs and nouns are assigned lexical and grammatical tone; at this stage we have a fully tonal system, labeled "Stage 1" in Figure 4. That Kisi verbs were once

152

G. Tucker Childs

tonal is indicated by irregularities in the assignment of tones to a small group of monosyllabic verbs (see Note 1). Stage 2 is characterized by the morphology taking over the assignment of tone on verbs, with nouns maintaining their lexical tone. Contrasts are still paradigmatic. It is during Stage 2, at least with regard to Kisi, that not all tonal oppositions are manifested; as shown in (3) above, we do not have sequences of HH(H) on verbs. In Stage 3 syntagmatic oppositions appear, similar to the appearance of the extra-high tone in Kisi. In fact there is already some evidence that the extra-high tone is spreading to other environments. In the Negative Hortative there is a tone melody of HL(L), and the first high tone sounds much like an extra-high. It is at Stage 3 that tonal (paradigmatic) contrasts are reanalysed as accentual (syntagmatic). Figure 4.

Stages in the evolution from tone to accent

Stages:

2

1 N

V

N

+ +

+ +

+

3 V

N

"Pure accent" V

N

V

Contrasts: Paradigmatic lexical grammatical Syntagmatic lexical grammatical

+

+ Only one accent on all phonological words

How this system would extend itself to nouns is not clear, but there are hints that nouns are losing one paradigmatic contrast. Nouns commonly appear with a suffixed noun-class marker, which always has a high tone. Speakers might interpret this suffix as part of the phonological word and reinterpret it as accent. This scenario is not so unlikely if one considers that it is the final syllable in trisyllabic verbs that receives extra prominence by the shortening of the preceding syllable. Furthermore, Mukarovsky (1976:180-81) reports that the attachment of high-tone affixes causes stem vowels to lower. "In several languages [of the Atlantic Group] noun or verbal affixes with high pitch clearly account for a lowering of the basic high pitch of the first stem syllable. This occurs likewise in CB [Common Bantu] word stems." As a final comment it should be noted that these remarks need considerable testing before they can be asserted with any authority. Nonetheless, the importance of phonetic detail in analyzing language change is clear. In fact, such close analysis allows us to capture language change in progress.

153

The Extra-high Tone in Kisi NOTES

*I am thankful for comments on an earlier draft from Karen V. Beaman, John J. Ohala, William S.-Y. Wang, and W.A.A. Wilson, none of whom should be held responsible for the present contents. 1.

There is one exception to this generalization, monosyllabic and disyllabic verbs with low stems in the Past. The irregularity here suggests the residue of a tonal contrast between classes of verbs. The regular tonal pattern for verbs in the past is LH(H). o o o o

tol tol tuu tuu

L L L L

L H LL LL

1

she 'she 'she 'she

planted' ascended' buried' measured'

o o o o

hulu hulu bEndu bEnda

L L L L

LL LH LL LH

'she threw out' 'she jimped up' 'she missed1 'she's fine'

These low-toned verbs are irregular in no other environment, not even in the negation of these sentences. This situation is significantly different, then, from the case in Bantu, where all verbs were originally monosyllabic (Clements and Goldsmith 1984). Kisi historically could be distinguished as a tone language on the basis of both its nouns and verbs. 2.

This holds true for all extended verbs but one, the Middle Extension (Childs 1987).

3.

Ferry's comments suggest that the high tone may be something close to the extra-high in Kisi, "Le ton haut étant le mieux distingué: il semblerait que l'écart de hauteur entre le ton moyen et haut est plus important qu' entre moyen et bas (1981:57).

REFERENCES Beckman, Mary Esther. 1984. Toward Phonetic Criteria for a Typology of Lexical Accent. Ph.D. dissertation, Cornell University. Childs, G. Tucker. 1987. Verb extension renewal in Kisi. In: Proceedings of the 16th Annual Conference on African Linguistics, ed. by David Odden, pp.73-91. Dordrecht: Foris Publications. Clements, George N. 1977. Four tones from three: the extra-high tone in Anlo Ewe. In: Language and Linguistic Problems in Africa (Proceedings of the VII conference on African linguistics), ed. by Paul F.A. Kotey & Haig DerHoussikian, pp.168-81. Columbia, SC: Hornbeam Press. Clements, George N. and John Goldsmith. 1984. Introduction. In: Autosegmental Studies in Bantu Tone, ed. by G.N. Clements and John Goldsmith, pp.1-17. Dordrecht: Foris Publications. Ferry, M.P. 1981. "Le basari". In: Les langues dans le monde ancien et modern: les langues de 1/Afrique sub saharienne, ed. by G. Manessy, pp.55-64. Paris: Centre National de la Recherche Scientifique. Greenberg, Joseph H. 1976. Circumfixes and typological change. In: Papers from the 4th International Conference on Historical Linguistics, ed. by Elizabeth C. Traugott et al. (Amsterdam Studies in the theory and history of linguistic science IV, Current Issues in linguistic theory. Volume 14). pp.233-241. Greenberg, Joseph H. 1978. How does a language acquire gender markers? In: Universals of Human Language, ed. by J.H. Greenberg et al., pp.48-82.

154

G. Tucker Childs

Stanford: Stanford University Press. Hansen, Walter, 1979. Collected Notes Toward a Phonology of the Sherbro Language. Freetown, Sierra Leone: The Institute for Sierra Leone Languages. Haudricourt, André G. 1972. Two-way and three-way splitting of tonal systems in some far-eastern languages. In: Tai Phonetics and Phonology, ed. by Jimmy G. Harris and Richard B. Noss, C.I.E.L., O.S.U., Bangkok. Hombert, Jean-Marie. 1975. Toward a theory of tonogenesis: an empirical, physiologically and perceptually based account of the development of tonal contrasts in language. Ph.D. dissertation, University of California, Berkeley. Hombert, Jean-Marie. 1984a. Réflexion sur le mécanisme des changements phonétiques. PHOLIA 1:87-112. Hombert, Jean-Marie. 1984b. Les systèmes tonals des langues africaines: typologie et diachronie. Pholia 1:113-64. (Extrait de Phonétique expérimentale et diachronie: application à la tonogénêse [Thèse d'Etat, 1984]). Hombert, Jean-Marie, John J. Ohala and William J. Ewan. 1979. Phonetic explanation for the development of tones. Language 55:37-58. Hyman, Larry M. 1976. D'où vient le ton haut du fe?fe?-bamlleke? In: Papers in Honor of Wm. E. Welmenrs, ed. by L.M. Hyman et al., pp.123-34. (Studies in African Linguistics, Supplement 6). Hyman, Larry M. 1977. On the nature of linguistic stress. In: Studies in Stress and Accent (Southern California Occasional Papers in Linguistics, 4), ed. by Larry M. Hyman, pp.37-81. Los Angeles: Department of Linguistics,

use.

Hyman, Larry M. 1978. Historical tonology. In: Tone: A Linguistic Survey, ed. by V. Fromkin, pp.257-269. New York: Academic Press. Lea, Wayne A. 1973. Segmental and suprasegmental influences on fundamental frequency contours. In: Consonant Types and Tone (Southern California Occasional Papers in Linguistics 1), ed. by L. Hyman, pp.17-70. Los Angeles: Department of Linguistics, USC. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, MA: MIT Press. Lehiste, Ilse, 1983. Prosodie change in progress: evidence from Estonian. In: Language Change, ed. by Irmengard Rauch and Gerald F. Carr, pp.10-27. Bloomington: Indiana University Press. Maddieson, Ian. 1974. A possible new cause of tone-splitting: evidence from Cama, Yoruba, and other languages. Studies in African Linguistics, Supplement 5, pp.205-221. Maddieson, Ian. 1977. Tone-spreading and perception. In: Studies on Tone (UCLA Working Papers in Phonetics, 36), pp.84-90. Maddieson, Ian. 1978. Universals of tone. In: Universals of Human Language, ed. by Joseph H. Greenberg et al., pp.335-65. Stanford, CA: Stanford University Press. Matisoff, James. 1973. Tonogenesis in Southeast Asia. In: Consonant Types and Tone (Southern California Occasional Papers in Linguistics 1), ed. by L. Hyman, pp.71-96. Los Angeles: Department of Linguistics, USC. Mukarovsky, Hans G. 1976. A Study of Western Nigritic, Vol. 1. Vienna: AfroPub. N'dlaye-Corréard, Geneviève. 1970. Études fes ou balante (dialecte ganja). Paris:SELAF. Ohala, John J. 1974. Experimental historical phonology. In: Historical Linguistics II. Theory and Description in Phonology, pp.353-89. Amsterdam: Noth Holland Publishing Company. Weimers, William E. 1976. A Grammar of Vai. Berkeley & Los Angeles: University of California Press. Westermann, Diedrlch. 1921. Die Gola-Spraché in Liberia. Hamburg : L. Friederichensen Co.

The Extra-high Tone in Kisi

155

Wilson, W.A.A. 1968. An interpretation of the Temne tone system. Journal of West African Languages, 5. ( 1):5-11. Wilson, W.A.A. In Press. The Atlantic languages. In: Current Trends: NigerKordofanian Languages, ed. by John Bendor-Samuel. The Hague: Mouton.

Department of Linguistics University of California, Berkeley

Chapter 12

Tone and Accent in the Xhosa Verbal System John Goldsmith, Karen Peterson and Joseph Drogo 1.

INTRODUCTION This paper presents work in progress involving the interaction of tone and

metrical accent in Xhosa, a Bantu language of the Nguni group spoken in South Africa. Our data are taken almost exclusively from Claughton's (1983) monograph; our general approach is based on the idea that tone and accent exist in Xhosa and operate autonomously. To say that the tone and accent systems function autonomously is, to be sure, a fashionable position currently; we mean by it not that the two systems are independent (for it is their dependence - or "interaction" - that is the point of this paper) but rather that the two are formally distinct and that each behaves as students of tone and accent systems would expect. For example, the most common locus for accent to fall is on the penult or antepenult; the familiar Latin stress rule, with its echo in so many related and unrelated languages, illustrates this point. Within the framework of metrical phonology, this is viewed as the combination of extrametricality assigned to the last syllable, together with a quantity-sensitive binary foot assigned to the end of the word. It is this pattern, distinct in several ways from the familiar character of tonal systems, that we find in the Xhosa verb. At the same time, the Xhosa verb is possessed of a tonal system, straightforwardly analyzed with the tools of autosegmental phonology. This is hardly surprising, in light of our understanding of Bantu tone systems (as recently summarized in Clements and Goldsmith (1984), for example). The difficulty of analyzing the Xhosa system, however, derives from the fact that it is "perturbed", we might say, by the accentual system that we hinted at in the preceding paragraph. This, then, is the goal of this paper: to introduce and justify an accentual system whose existence is ultimately motivated by the way in which it modifies and affects the autosegmentalized tones. The analogy that springs immediately to mind may not be too trite to mention: although the planets of the Solar System function autonomously, they affect each other with their gravitational pull, and the astronomer may reasonably infer the existence of an unobserved planet on the basis of the perturbing effects on observed planets,

158

J. Goldsmith, K. Peterson and J. Drogo

more especially if the hypothesized perturbing planet is calculated to move in an elliptical orbit about the Sun, i.e., to behave just the way known planets do. This is the kind of case that we shall build for the accentual system of Xhosa.

THE XHOSA VERB

2.

The structure of the Xhosa verb is as in (1): (1)

subject marker

[radical + (extensions) + FV]

The subject marker (SM) varies in form depending on the noun class of the subject; tense marker (TM) is absent in the Infinitive and absent in the Present Indicative when the focus is off the verb; the object marker (OM), like the SM, varies in form depending on the noun class of the object; verb radical and verbal extensions; and the Final Vowel (FV), which varies with tense. The radical, extensions (if present) and FV constitute the stem of the verb.

3.

PRESENT INDICATIVE In the Present Indicative forms presented in (2), there is no TM (i.e.,

focus is off the verb) and no OM. They therefore include the bare minimum necessary for a verb: SM - radical - FV. The SM is toneless (third person animate (Class 1/2) SMs are High-toned, while first and second person SMs are toneless). (In the data presented here, as well as in our analysis, we will abstract away from a general rule that lowers sequences of High-toned syllables that result from a High tone spread over the vowels; we indicate the possible surface form in which all the vowels are High.) (2)

Present Indicative, no TM, no OM, toneless SM a. Toneless radical 1. ni-wis-a 'you 2. ni-bal-a 'you 3. ni-balis-a 'you 4. ni-sukumis-a 'you 5. ni-namathelis-a 'you

b. High-toned radical drop' count' narrate' narrate shake' cement'

1. ni-bèk-a ni-b&c-a 'you 2. ni-mém-a ni-m6m-a 'you 3. ni-bùlis-a ni-bttlis-a 'you 4. ni-b6nls-a ni-b6nls-a 'you ni-b6nlsls-< 5. ni-j^Sjitsls-a 'you 6. ni-nlnlthSkis-a 'you

put' invite' greet' show' show clearly' make slippery'

Tone and Accent in the Xhosa Verbal System

159

The forms in (2.a) have no High tone, while those in (2.b) do, a difference directly ascribable to the fact that in (2.b), the verb radicals are underlyingly High-toned, while in (2.a) they are underlyingly toneless. In (2.b.4, 5, and 6) the High tone associated with the radical has spread to some syllable further to the right (associating with an intervening syllable in the case of (2.b.6))). This spreading is the result of a rule which we will call Attraction to Accent. Looking for the moment only at the last two forms in column b, we see that the High tone has spread to the antepenultimate syllable of the word. (In 2.b.4), ni-b6nis-a, a High tone appears on the penult. This and similar forms will be discussed below.) As will become clear later, when we examine more forms, accent (*) is generally assigned to the antepenultimate syllable of a verb. A subsequent rule, Attraction to Accent, associates the nearest High tone to the left of the accent with the accented syllable, and with all intervening syllables, when the accented vowel is not already associated with a tone (indicated by the circle around the vowel). So far, then, we have the following rules:

(3)

Accent Placement (first approximation) V V V]

(4)

•*• V V V

Attraction to Accent

(3), the Accent Placement rule, will need further modification, but it will suffice in order to illustrate Attraction to Accent in the following form, (2.b.5): (5)

ni-bonisis-a H *

ni-bonisis-a

Accent Placement

H *

ni-bonisis-a I'''

H

Attraction to Accent

160

J. Goldsmith, K. Peterson and J. Drogo

In forms (2.b.l) and (2.b.2), the High tone is associated with the penultimate syllable, which is also the first syllable of the stem. These forms suggest a modification of the Accent Placement rule, to the effect that if the penultimate syllable of the verb is associated with a High tone, it will be assigned an accent. If not, accent will be assigned to the antepenultimate syllable. We will discuss the status of this observation in some detail below. (31)

Accent Placement (second approximation) a.

V V]

I H

IH

V V]

b.

V V V]

-»• V V V]

*

In (2.b.l), then, accent is assigned to the penult, because it is associated with a High tone, and this blocks the application of (3'.b). Since the accented syllable is already associated with a High tone. Attraction to Accent will have no effect. We return now to (2.b.4), ni-bonis-a, where the High tone of the radical has spread to the penult, rather than to the antepenult, as would be predicted from the rules which have been shown so far. The verb -bon- is one of a group of verbs which undergo a lexically-governed Accent Hopping rule, which, after placement of accent, shifts that accent one syllable to the right. (6)

Accent Hopping *

V V

*

-»• V V

As we will see, the stems that trigger Accent Hopping may come from either the High-toned class or the toneless class. It is the independence of the two characteristics that we take to be one of the major reasons not to view the Accent Hopping property as being fundamentally a property of tone. Although Accent Hopping eventually has a surface effect on tone placement, the diacritic that triggers it is not technically or formally a tonal property, and our proposal to analyse it as an accentual property seems to us to be the most natural way to treat it in the grammar. The derivation of (2.b.4) thus includes the following steps:

Tone and Accent in the Xhosa Verbal System (7)

161

ni-bonis-a H *

ni-bonis-a

Accent Placement

H ni-bonis-a

Accent Hopping

H ni-bonîs-a

Attraction to Accent

H Looking ahead, we will see shortly that (2.b.2), -mem-, is also a member of the class of verbs which undergoes Accent Hopping. However, (b.l) and (b.2) are identical in their surface tones. This is because, in the Present Indicative, there is a restriction on the Accent Hopping rule that it cannot move an accent onto a FV. Therefore, if accent is already on the penult, it will stay there. Up to this point, we have dealt only with the forms in (2.b). Since there are no High tones at all in (2.a) , we cannot see the effects of Accent Placement, Accent Hopping and Attraction to Accent. Now let us look at the same forms, but with an OM. All OMs in Xhosa are underlyingly High-toned. (8)

Present Indicative with OM, no TM, toneless SM a. toneless radical 1. 2. 3. 4. 5.

ni-wâ-wis-a ni-wa-bâl-a ni-wa-bâlis-a ni-wa-sûkûmis-a ni-wa-nâmâthélis-a

b. High-toned radical 1. 2. 3. 4. 5. 6.

ni-wa-bék-a ni-wa-mém-a ni-wa-bulis-a ni-wa-bânîs-a ni-wa-bônîsis-a ni-wa-iiikîthékis-a

The forms in (8.a) show that the same accent rules which were posited for the High-toned verbs in (2.b) also apply in the toneless verbs. Accent placement assigns accent to the antepenult (in none of these forms is the penult already associated with a High tone; therefore, (3'a) does not apply), and a High tone spreads to the accented syllable. (8.a.2), -bal-, like the High-toned verbs -mem- and -bonis-, is a member of the verb class which undergoes Accent Hopping. in (8.a.2-5) and in (8.b), the High tone has spread from the OM -wa- to the accented syllable, but is no longer associated with -wa-. In the forms of (2.b),

162

J. Goldsmith, K. Peterson and J. Drogo

when a High tone spreads to an accent, it remains associated with the syllable it was originally associated with. However, in those cases, the syllable originally associated with a High tone was the first syllable of the verb stem. We propose, therefore, a rule which disassociates a High from -wa- if that High is also associated with another syllable to the right. (9)

Disassociation from OM OM

W H

V

Thus, after a High tone from -wa- has spread to the right by Attraction to Accent, it disassociates from -wa-. (10)

ni-wa-sukumis-a

!

H

*

ni-wa-sukximis-a

Accent Placement

I

H *

m-wa-sukumis-a H

Attraction to Accent

*

ni-wa-sukumis-a

Disassociation from OM

H Rule (9) does not apply in (8.a.l), since the High tone from -wa- has not spread (i.e., is not associated with any other syllable), because an accent has been placed on -wa- (the antepenult). (11)

ni-wa-wis-a I

H ni-wa-wis-a

Accent Placement

The forms in (8.b) have two underlying High tones, the High tone from the radical and the High tone from the OM -wa-. However, it is likely that there is only one High tone on the surface. If there were two, then the High tone

Tone and Accent in the Xhosa Verbal System

163

associated with the radical should spread to the accent, while the one associated with the OM should remain where it is, since as formulated, Disassociation from OM applies only when the High tone associated with the OM is also associated with some other syllable(s). (12)

ni-wa-bonisis-a

I IH

H

*

ni-wa-bonisis-a

II H H

*

ni-wa-bonisis-a

I V

H

Accent Placement

Attraction to Accent

H

I 1/

*ni-w&-b6nisis-a H H However, if we suppose that one of the High tones is deleted, then this problem does not arise. It may appear from these data that it is the leftmost High tone, the High tone from -wa-, which is deleted. However, if we look ahead briefly and examine a form with a third person SM (i.e., a High-toned SM), we will see that that approach will not work. Changing ni- to li-, we get ll-w&-b6nlsis-a. If deletion occurred from the left, then we would expect the surface form *li-wa-bonlsis-a, which is incorrect. However, if deletion is from the right, then the leftmost High tone will remain, and will spread to the accent. We will call this rule Meeussen's Rule (cf. Goldsmith 1984a,b and elsewhere) . (13)

Meeussen's Rule V

V

H

The derivations of ni-wa-b6nlsis-a and H-wS-b6nIsis-a are as follows on the next page. Disassociation from OM applies only if the OM is the leftmost association of a High tone.

164

J. Goldsmith, K. Peterson and J. Drogo

(14)

ni-wa-bonisis-a

I I

H

I I I

H

H

ni-wa-bonisis-a

I I

H

li-wa-bonisis-a H

H

li-wa-bonisis-a

Accent Placement

I I I

H

H

ni-wa-boni sis-a I H

-0

I A en-gàrrim Qéjuk

'new truck' (720)

0 H L

H

0

V A

'new garment' (720)

c. Ol-kiia qéjùk

'I see the big chief' (720)

d. à-dól ol-karsis kitòk

'big truck' (720)

e. en-gàrrim kitòk

'big garment' (720)

f. ol-kilà kitòk

The Phrase Final Lowering rule can be written in several ways. Assuming that the L tones are present at the time that lowering takes place, the rule as formulated in (21) states that a phrase final H is lowered when preceded by a H tone. The trigger and victim can possibly be separated by one or more L tones, illustrated by the derivation in (22a). The rule is not applicable to the structure in (22b). (21)

Phrase Final Lowering 1 H

L / H Lq

(L tones present)

//

Lower a phrase final H when it is preceded by a H (possibly separated by one or more L tones). (22)

Phrase Final Lowering derivation L H L

I

I I

H+ L

/\

a. en-gàrrim Qéjùk (=(20b))

(L tones present) L H H L

I

I

II

((21) not applicable)

b. en-ker safiuk (=(17b))

Arusa (Maa) Phrasal Tcnology

229

If we assume that L tones are not present. Phrase Final Lowering 2 in (23) involves deletion of a final H and insertion of L tones to all unassociated skeletal slots by the Default Rule in (24) . This pair of rules was assumed in (16), (18), and (20) and will be assumed henceforth in illustrating derivations. ^ (2 3)

Phrase Final Lowering 2 H

(L tones absent)

->• 0 / H V c 0 //

Delete a phrase final H tone. (24)

Default Rule 0

+

L /

V Associate a L tone with all unassociated slots. That the presence or absence of Ls intervening to the left between trigger and victim is irrelevant to the Phrase Final Lowering rule indicates that it may be more insightful to analyze the rule as operating at a stage when the Ls are absent, i.e. as in (23). Of course, this must be at a stage before L-Raising applies since L-Raising crucially refers to Ls. This presents no problem as the output of Phrase Final Lowering is subject to L-Raising. Indeed, Phrase Final Lowering is more likely a phonological rule since it changés the phonological status of the phrase final H and as such would be ordered before LRaising by a theory distinguishing phonological and phonetic rules.1* On the other hand, the potential victim H of Phrase Final Lowering must be the last tone on the last vowel in the string. That is, the rule is sensitive to the presence of L tones to the right. If Ls are not present the phrase final condition must be encoded by referring to the skeletal tier in Phrase Final Lowering 2 in (23). Just this situation was avoided in the formulation of FSimplification in (5). I have no explanation for why the right and left hand environments differ as to whether they are sensitive to the presence of L tones.5 It is assumed in the formulation of the Phrase Final Lowering rules that the vowels that lower are all linked to the same H tone. We must ask whether there is a contrast in the language between a single tone and multiple adjacent H tones on the tonal tier. Certainly, there is a contrast at word boundary:

230

Barbara Levergood

in the forms in (18), only the final H lowers, never the preceding H across word boundary. A more interesting question is whether H tones in Arusa obey the OCP at the word level. 2.3.2 Downstep There is evidence that suggests that H tones violate the OCP at the word level. There are several words in the language, some of which appear in (25) , which have final L tones in isolation or phrase finally, as the final vowel of en-tit'o in (25a), which alternate with H phrase internally, as the same vowel in (25b-c). There is a H tone vowel within the word immediately preceding the alternating vowels. Following Tucker and Mpaayei's (1955) use of the term, we will call this phenomenon "downstep" and we will use the downstep symbol "!" to mark the alternating vowels. (25)

Downstep H H+ 0 I . I. H1

H

1H

ol-Syy'6-ni 'boy (Nom)1 (721) l-slnk"ir 'fish (PI)' (721) 'girl' (721) i in-tas&t-'in 'old women* (721)

b. en-tit-6 säbuk

il-kilS-n'i 'garments' (721) 'fat girl' (721)

H H H L I ,I / V c. en-tit-6 torrö-nö 'bad girl' (721)

m&nk&l-'l

'careless (PL)' (721)

su&at-'i

'healthy (PL)' (721)

saBiik-'in

'fat (PI)' (721)

In (26a), sidä-y is subject to Phrase Final Lowering triggered by the H on i en-k6r. The downstepped form ol-äyy'6-ni in (26c) is claimed to have an identical tonal structure, two adjacent H tones on the tonal tier, the second of which lowers by Phrase Final Lowering (and later undergoes L-Raising). Phrase Final lowering is correctly not applicable to either of the forms in question in phrase internal position in (26b,d). I thus claim that downstep is just a special case of Phrase Final Lowering. The trigger for lowering is in the preceding word in (26a) and within the same word in the downstep example in (26c). The downstep forms fill the only gap in the distribution of H tones subject to Phrase Final Lowering. (26)

Phrase Final Lowering and downstep H H-t-0 A I a. en-ker sldä-y (722)

H H H 0 I A I b. en-ker sidä-y nalel) (832) '(very) good ewe'

H H*0 I ,A c. ol-äyy'o-ni

H H H+0 I ,A I d. ol-äyy'6-ni dorroip (832)

' (short) boy (Nom) '

Arusa (Maa) Phrasal Tonology

231

The tonal representation of downstep forms is unproblematic if only the phrase level is considered. Recall that Phra3e Final Lowering in (21) or (23) does not care how many, if any, L tones separate the victim H from its trigger. Thus, if there are two adjacent H tones on the tonal tier in phrase final position as in (25a) and (26c), the second will correctly lower by this rule. By the same reasoning, it may as well be that these downstep cases contain an intervening floating L. However, the phrasal tonology cannot provide evidence for such a word internal tone. 2.4

F^L: F-Lowering

We showed in section 2.1 that phrase final Fall alternates with H phrase internally, accounted for by F-Simplification. A Falling tone also alternates with L both phrase finally and internally under a very narrowly defined condition that will be argued for now. In the Maa dialects, nouns, demonstratives, etc. are marked for case solely by tone (Tucker and Mpaayei 1955:175; Heine 1980:108). In particular, in (27), the Nominative /kSr/ has a Falling root; the Accusative is H. Both of these roots alternate. In (27), the H alternate occurs only phrase internally in (27b) as a result of F-Simplification. The L alternate occurs phrase finally in (27c,e) and phrase internally in (27d,f). The demonstrative /elS/ exhibits the same F/H/L alternation in the Accusative in (28) , showing that this alternation is not unique to a particular case. (Here we begin to use tone marks rather than pitch tracks to mark fairly low level values. Slash brackets enclose a more abstract, possibly intermediate, level of representation.) (27)

F^H^Z,: /Jeer/ (Nominative) a. e-dól en-kèr b. e-dól en-kéi kitok c. e-dól enà ker d. e-dól enà ker kltok e. e-tà-àr én-ker /e-tà-àr-à en-kèr/ f. e-tà-àr én-ker kltok /e-tà-àr-à en-kèr kltok/

(28)

•the ewe sees it' (722) 'the big ewe sees it" (722) •this ewe sees it' (512, 722) •this big ewe sees it' (513, 722) •the ewe killed it' (512, 722) •the big ewe killed it' (513, 722)

i^H^L: /elè/ (Accusative) a. à-dól elé mórwo b. e-tà-làm èie mórwo /e-tà-làm-à elè mórwo/

'I see this old man' (723) •he avoided this old man' (547, 723)

232

Barbara Levergood

It is revealing to compare the F/H/L alternation in (27) with the H/L alternation in the Accusative /k£r/ in (29). The H in (29) only alternates with L when it is subject to Phrase Final Lowering in (29b,d), but not when it is phrase internal in (29c,e). Because the L alternate does appear phrase internally in (27d,f), we know that it must be the Fall, not the H, which alternates with L in (27). (29)

H^L: /ker/ (Accusative) a. en-k£r b. e-döl enä ker c. ä-tä-är enä kSr säßuk d. e-tä-är en-ker /e-tä-är-ä en-ker/ e. e-tä-är ¿n-kSr säßuk /e-tä-är-ä en-kSr säfcuk/

'ewe (Acc)1 'he sees this ewe' (512, 722) •I killed this fat ewe* (513,722) 'he killed the ewe' (512, 722) •he killed the fat ewe' (513, 722)

As the H/L alternation is covered by Phrase Final Lowering and the F/H alternation by F-Simplification, we only need to ask what triggers the F/L alternation. When the Fall lowers, there is always an immediately preceding H tone, as in (27c-f) but not (27a-b), and as in (28b) but not (28a). However, not just any immediately preceding H will do, since the Falls in (30) do not lower. (30)

Surface HF sequences H

a.

L

A/ n-tempO-n

L

'elephants' (723)

b.

H

L

I / M ol-mäshSr-1

'tick' (723)

Rather, the crucial observation is that a Fall lowers just when it is immediately preceded by a demonstrably separate, i.e. derived, H tone. In (27c-d) the trigger is across word boundary. There are two other similar ways that the H tone trigger can become immediately preceding in the course of the phonological derivation. In (31a), the Fall on the root /tlm/ alternates with L. In the derivation in (31b), only the first of the two consecutive Falls on /en2 tlm/ is simplified to H by F-Simplification. This H tone then immediately precedes the Fall on /tlm/ and serves as the trigger for F-Lowering. Note that F-Simplification must not be allowed to apply to /tlm/ before F-Lowering gets a chance to. (31)

Simplified F trigger a. S-dol ena tim a-ddl ena tim taata /h-dbl ena tlm tkata/

'I see this unoccupied area' (833) 'I see this unoccupied area today' (833)

Arusa (Maa) Phrasal Tonology

233

H L H L H L H L L

b.

A \ V V \\ \ à-dól enà tlm tàata H

L H

L

F-Sim

H L H H L H L L

A I I I II I a-dol ena tim taata

F-Lowering

HI 1

A I I I III à-dól enà tira tàata

(=(31a))

In (27e-f) and (28b) the immediately preceding H tone is also derived, as illustrated in (32) . The derivation begins in (32a) with the Fall on /kSr/ separated from the potential trigger by a L tone vowel. H-Tone Spreading (HTS) to that vowel across word boundary creates a Falling contour that is simplified to H by F-Simplification in (32b-c). This H triggers F-Lowering on /kSr/ in (32d-e). Again, F-Simplification must not apply to the potential victim of FLowering. It must be conceded that the interaction of these two rules is not completely understood. (32)

Derived trigger L

I

H

L

/ K M

HL

H L

a. e-ta-ar-a en-kèr kltok L c.

HL

H

L

I

HTS

H

L

HL

H L

I I I

b. e-ta-ar-a en-ker kitok L H HL H L

I

H L

i /i^ i 11 e-ta-ar-a en-ker kitok

L e.

H

L

II

/l\

I

II

•+ d. e-ta-ar en-ker kitok V-Elision

F-Sim F-Lowering

H L

I /T\ I I I e-tà-àr én-ker kitok

(=(27f))

The rule of F-Lowering is formulated in (33). L tones must be present at the time the rule applies so as to block application to the Fall on /kèr/ in (27a), for example, where the potential trigger and victim are not on consecutive vowels. Because a simplified Fall triggers the rule, F-Simplification must delete the L component of the Fall in (31b) and (32b) to avoid having to explicitly include an optional floating L tone in the structural description of FLowering. As for the H tone in (33), I assume without evidence that it is deleted, not just delinked.® (33)

F-Lowering (L tones present) H

0 / H

L

Delete the H component of a Fall when immediately preceded by a H tone.

234 3.

Barbara Levergood CONCLUSION

By way of conclusion, we summarize what we have deduced from the phrase level rules about tonal representations in Arusa. L-Raising in (11) and H-Downdrift in (14) are rules that do not change phonological values and are good candidates for phonetic rules. L-Raising may be tentatively argued to be conditioned by a singly linked L tone, which would contrast with multiply linked L tones. F-Simplification in (5) is a rule with no surface exceptions that applies only phrase internally. Formal considerations suggest that it applies at a time when all L tones are present. We know that only a demonstrably derived immediately preceding H tone triggers F-Lowering in (33). This requires that all L tones be present at the time it applies. Both phrase level rules known to refer to Falling tones then arguably apply to a representation containing other L tones. Finally, the rule of Phrase Final Lowering in (21) or (23) presents contradictory evidence. Its right hand side is best formulated as though the L tones were present, its left hand side as though they were not. The operation of Phrase Final Lowering provides evidence for a contrast between a single multiply linked H tone and a sequence of H tones on the tonal tier, both within a word and across word boundary. This result is dependent on not finding evidence for word internal floating L tones. That these rules show some sensitivity to L tones is not surprising if one considers that they are all phrase level rules. The true test of the character of Arusa L tones should become apparent from the word internal tonology.

NOTES *The native name of the language is Enkutuk oo Larusa 'language of the Arusha'. My consultant is Naftali Mollel, a graduate student at the University of Illinois. He is fluent in Arusa, Swahili, and English. Although he has not lived in Arusha for some years, his intuitions about Arusa are consistent on points crucial for this paper. There are many similarities between Mollel1s idiolect and the Maasai dialect described by Tucker and Mpaayei (1955) in their important grammar. The debt I owe Mollel for his skill and patience in working with me is immeasurable. I would also like to thank the participants in the Spring 1985 Field Methods class at the University of Illinois for discussing Arusa with me: Laura Downing, Andi Dunn, Omar Ka, Chuck Kisseberth, Elizabeth Piatt, Nalini Rau, and Eluzai Moga Yokwe. The numbers following the glosses in the examples refer to pages in the original field notes.

Arusa (Maa) Phrasal Tonology

235

1.

According to the tonetic representations provided by Tucker and Mpaayei (1955:166ff) for Maasai, a sequence of L vowels is always level in pitch (unless phrase final, when the final vowel may be pronounced as LFalling). Although Maasai also has partial H-Downdrift, a single L vowel is always lower in pitch than flanking H tone vowels. Both of these differences between Maasai and Arusa can follow from the claim that only Arusa has the rule of L-Raising.

2.

Occasionally a single L tone flanked by H tones will be somewhat higher than the following H. This can be analyzed simply as phonetic variability in the pitch of the L.

3.

For completeness, I would like to briefly mention another alternation. In practice, Phrase Final Lowering eliminates virtually all phrase final H tones. However, the gap is filled by an optional simplification of phrase final Falls to H. This is not a case of the F-Simplification rule already discussed as it appears to depend only on how much breath the speaker has left at the end of a sentence.

4.

However, Phrase Final Lowering is suspended in yes/no questions.

5.

A problem that arises when claiming that some level of representation lacks Ls is the treatment of Falling tones. Crucially involved but deferred are the questions of whether the Falls are underlying or derived, and, in a question raised by Chuck Kisseberth, whether the L component of the Fall is subject to a claimed lack of Ls in a representation.

6.

Tucker and Mpaayei (1955:173), in a discussion of downstep, cite examples similar to that in (28b), stating, "the final vowel of the demonstrative, having a lowered high tone, may lose its tonal relationship with the word that follows it". Presumably, they would treat the Falling tone as being subject first to F-Simplification, triggered by its phrase internal position. Then, they seem to consider the lowering phenomenon a case of Downstep, which is otherwise applicable even in Maasai only phrase finally. Their generalization is troubling because a phrase boundary is required in the unlikely position between the demonstrative and its head noun and because this position must be claimed to be phrase internal for the purposes of the first rule and phrase final for the second. One wonders whether Tucker and Mpaayei were momentarily confused and considered this to be a H/L (Nominative case) rather than a F/L (Accusative case) alternation.

REFERENCES

Goldsmith, John. 1979. Autosegmental Phonology. New York: Garland. Heine, Bernd. 1980. Camus, a Maa dialect. In: The Non-Bantu Languages of Kenya, ed. by Bernd Heine, pp.99-140. Berlin: Dietrich Reimer. (Language and Dialect Atlas of Kenya, Vol.2.) Leben, William R. 1978. The representation of tone. In: Tone: a Linguistic Survey, ed. by Victoria Fromkin, pp.177-219. New York: Academic Press. Leben, William R. 1980. Suprasegmental Phonology. New York: Garland. Levergood, Barbara. 1984. Maasai phrasal tonology. Unpublished ms., University of Texas. Liberman, Mark, and Janet Pierrehumbert. 1984. Intonational invariance under changes in pitch range and length. In: Language Sound Structure, ed. by Mark Aronoff and R.T. Oehrle, pp.157-233. Cambridge MA: MIT Press.

236

Barbara Levergood

McCarthy, John J. 1986. OCP effects: gemination and antigemination. Linguistic Inquiry 17:207-63. Odden, David. 1986. On the role of the obligatory contour principle in phonological theory. Language 62:353-83. Prince, Alan. 1983. Downstep in phonetic implementation, with special reference to Kikuyu. Paper presented at The University of Texas, December 5, 1983. Pulleyblank, Douglas G. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Tucker, A.N., and John Tompo Ole Mpaayei. 1955. A Maasai Grammar. London: Longmans, Green and Co. Vossen, Rainer. 1982. The Eastern Nilotes. Berlin: Dietrich Reimer.

Department of Linguistics University of Texas s University of Illinois

Chapter 17

An Account of Serial Verbs in Haitian Creole: Argument Structure Coalescence Rose-Marie Déchaîne The aims of this paper are twofold: first, to give an account of the properties of serial verb constructions (SCs) in Haitian Creole (HC); second, and more generally, to identify how "serializing" grammars differ from "non-serializing" grammars. The distribution of SCs in HC is considered, followed by a discussion of their syntactic and lexical properties and an analysis of SCs in terms of an operation on the argument structures of verbs.

1.

DISTRIBUTION OF SERIAL CONSTRUCTIONS IN HAITIAN Although SCs may involve two or more verbs, I shall limit my attention to

SCs with two verbs. The surface order which corresponds to a SC is as in (1): an NP subject followed by a succession of verb phrases. I refer to the first verb of a serial construction as VI and the second verb as V2. (1)

NP

[Vl

NP

] [V2

NP

]

Semantically, an SC forms a complex predicate whose meaning is determined by the combination of the meanings of VI and V2. One of the verbs is from an ppen class and the other is from a closed class. SCs in HC split into two groups according to whether it is VI or V2 that is from a closed class. 1.1

VI From a Closed Class

The examples in (2) are representative of SCs in which VI is from a closed class. In these cases the verb pra 'take' combines with verbs of trains fer such as bag 'give', pOte 'carry' and mete 'put'.1 (2)

a. za te pra liv - la bay mari Jean PA take book-DET give Marie 'Jean gave-transferred the book to Marie' b. pOl pra pul-yo pOte na mase Paul take chicken-PL carry LOC market 'Paul took-carried the chickens to the market'

238

Rose-Marie Déchaîne c. za pra maze mete na frizidE-a Jean take food put LOC fridge-DET 'Jean put food in the fridge1

It should be noted that the possibility of combining pra with other verbs to form a SC is restricted, as far as I know, to these three verbs. 1.2

V2 From a Closed Class

In the other type of SC found in Haitian it is V2 which is from the closed class. SCs of this type involve either the verb bay 'give' or the directional verbs ale 'go', sOt 'go out', vin 'come', and rive 'arrive' as V2, as shown in (3) and (4) respectively. (3)

a. za tune liv-la bay mari Jean return book-DET give Marie 'Jean returned the book to Marie' b. h pase liv-la bay za 3SG pass book-DET give Jean 's/he passed the book to Jean' c. h vOye balO-a bay za 3SG throw ball-DET give Jean 's/he threw the ball to Jean'

(4)

a. za te pOte pul al(e) na mase Jean PA carry chicken go LOC market 'Jean brought chickens to the market1 b. li te pOte barik-la sot lakay 3SG PA carry barrel-DET go out house 1

s/he carried the barrel from the house'

c. li te pOte 3SG PA carry

barik-la vin isit barrel-DET come here

's/he brought the barrel here' d. li te mEne burik-la rive lakay 3SG PA bring mule-DET arrive house 's/he brought the mule to the house' To recapitulate, there are two types of SCs in Haitian. In one case VI is from a closed class, as in (5). In the other case, which is the most productive paradigm, V2 is from a closed class, as in (6).

239

Serial Verbs in Haitian Creole (5)

(=2)

closed class prk

(6)

(=3,4)

open class pOte tune vOye prEte etc.

(Vl)

open class

'take*

bay mete pOte (VI)

closed class

'carry' 'return1 'send' 'lend'

bay

pOte 'carry' mEne 'bring' kuri ' run' rale 'pull' puse 1push' mbte 'ascend' deskn 'descend' etc.

1.3

(V2)

'give* 'put' 'carry'

vini ale rive sOti

(V2)

'give'

'come' 'go' 'arrive' 'go out'

SCs and Ditremsitive Verbs

The presence of serial verbs in a language has often been attributed to the absence of ditransitive verbs (cf. Givon 1975). This is not the case in HC, since it has a large inventory of ditransitive verbs. Furthermore, all the verbs which occur in SCs also occur as independent verbs.

2.

SYNTACTIC PROPERTIES OF SCs SCs in HC present the same syntactic properties as SCs in other languages:

a single NP subject, restrictions on the distribution of tense/aspect markers, absence of markers of subordination or coordination. It is not possible for a lexical subject (even if it is a co-referential pronoun) to appear in front of V2, as shown in (7). (7)

a. *za± te pra

liv - la

za±

bay mari

*za± te pra

liv - la

li^

bay mari

Jean PA take book-DET Jean/3SG give Marie b. *za, te pOte

pul

za .

ale

*za± te pOte

pul

hj-

ale na mase

Jean PA carry chicken Jean/3SG

go

na mase LOC market

240

Rose-Marie Déchaîne

Tense-aspect markers may appear only before VI, as shown in (8) (8)

a. *za

te pra

liv - la te bay

mari

*za

pra

liv - la te bay

man

Jean PA take book-DET PA give Marie b. *za

te pOte

pul

t- ale na

mase

*za

pOte

pul

t- ale na

mase

Jean PA carry chicken PA

go LOC market

The only allowable configuration of tense/aspect markers is (9a)

(9b) and (9c)

are not possible realizations of SCs. (9)

a.

NP TENSE/ASPECT

VI

(NP)

b.*

NP

VI

(NP)

TEMPS/ASPECT±

(NP)

TEMPS/ASPECTj V2

TENSE/ASPECT-L

c.* NP TENSE/ASPECTVI

V2 V2

(NP) (NP)

(NP)

It has been suggested that SCs fall into the same semantic class as coordinated or subordinated clauses (Williams 1971, Schachter 1974). However, in HC, SCs do not have the syntactic properties of co-ordinated or subordinated clauses nor are they interpreted in the same way. When two sentences are coordinated, the conjunct is optional; there is an NP subject in front of V2, and there may be a tense marker in front of V2 as well. (10a) corresponds to (2a) except that it is a co-ordinated construction and not an SC, and (10b) likewise corresponds to (4a). (10)

a. za Jean

te pra liv - la (epi) li (te) bay mari PA take book-DET (and) 3SG (PA) give Marie

'Jean took the book and he gave it to Marie' b. za te pOte pul (epi) li (te) ale na mase Jean PA carry chicken (and) 3SG (PA) go LOC market 'Jean carried chickens and he went to the market' In an embedded clause, signalled by the presence of the complementizer pu, the NP subject before V2 is obligatory even if it is co-referential with the NP subject of VI and there may be a tense marker before V2, as shown in (11).

Serial Verbs in Haitian Creole

(11)

a.

241 %

v

za

te potè

pul

pu-1

te

ale na mase

*za

te pOte

pul

pu

te

ale na mase

Jean PA carry chicken COMP 3SG PA go LOC market 'Jean carried the chickens in order to go to the market' b.

za

te pra

liv - la

pu-1

te bay mari

*za

te pra

liv - la

pu

te bay mari

Jean PA take book-DET C0MP-3SG PA give Marie 'Jean took the book to give to Marie' Furthermore, the two NP subjects may have disjoint reference: bay mari liv-la mete su tab-la pyB Pierre give Marie book-DET put on table-DET

(12)

'Pierre gave the book to Marie for her to put it on the table' 2.1 Syntactic Analyses of Serial Constructions To my knowledge, three types of analyses have been proposed to account for the properties of SCs: transformational, base-generated, and movement. Under a transformational analysis, SCs are derived from underlying co-ordinate/subordinate clauses through the application of deletion rules. Thus, in (13) the co-ordinate/subordinate marker as well as the NP subject and Tense marker preceding V2 are deleted. (13)

TRANSFORMATIONAL: NP

Tense

[VI NP]

(Williams 1971) COORD/SUB

0

NP + 0

Tense

[V2 NP]

+ 0

Schachter (1974) has demonstrated that such an analysis is inadequate in that it fails to predict the absence of an NP subject and tense/aspect markers before V2. Furthermore, as Williams (1971) observes, filters or diacritic features are required to prevent the deletion rules from applying in an indiscriminate fashion to all co-ordinated or subordinated clauses. Under a base-generated analysis, successive VPs are directly generated by rewrite rules: under S as in (14a), or under VP as in (14b).

242

Rose-Marie

(14)

Déchaîne

BASE-GENERATED:

a.

S

+

b.

VP1

->-

NP

VP1 VI

NP

VP2

(Schachter 1974)

VP2

(Jansen et al. 1978)

In current Government and Binding theory constituent order is determined by higher order principles such as the direction of theta-role and Case assignment. In such a model rewrite rules such as the ones in (14) are no longer a necessary part of the base component. Under a movement analysis, one verb is taken to be the main verb assigning all the thematic roles but only one Case. Constituent order is derived from the movement of one of the NP complements from a non-Case marked position to a Case marked position, as shown in (15). (15)

MOVEMENT: VI TAKE

(Roberts 1985) V2 non-8 GIVE

+Case

c.

NP

NP

8

8

b.

VI TAKE

+Case -Case

VI NP NP V2 CARRY 8 8 GO +Case -Case

non-8 +Case

d. non-8 +Case

NPi

V2 GIVE

NP

8

tA

8

+Case -Case

VI NP t± V2 CARRY 8 8 GO +Case -Case

NP ± non-8 +Case

A movement analysis of this type fails to provide a unified account of SCs in HC since one must posit two distinct derivations according to whether it is VI or V2 which assigns the thematic roles. In (15a), V2 'give' assigns two thematic roles but only one Case, and there is leftward movement of the second NP to a Case-marked position adjacent to VI to derive (15b). In (15c), VI 'carry' assigns two thematic roles but only one Case, and there is rightward movement of the second NP to a Case-marked position adjacent to V2, as in (15d) . Finally, although base-generated and movement analyses both account for the surface syntactic constraints noted in section 2, they fail to account for certain lexical properties of SCs discussed in the next section. 2.2

Lexical Properties

of Serial

Constructions

SCs exhibit certain properties that suggest they are derived from a lexical rather than a strictly syntactic operation. Specifically, there are constraints on the distribution of the semantic roles (I shall refer to them as thematic roles) of complements of verbs appearing in SCs. My use of the notion 'the-

Serial Verbs in Haitian Creole

243

matic role' is informal and I will assume an understanding of what Agent, Theme, Goal, Source, etc, are. Syntactic analyses of SCs have no way of accounting for the fact that the same thematic role cannot be assigned to two different NPs. For example, the verb pra 'take' assigns the thematic roles Theme and Source (1 take something from someone'), The verb bay 'give' assigns the thematic roles Theme and Goal ('give something to someone'). When these two verbs appear together in an SC, the same thematic role cannot be assigned to two different NPs. This rules out a sentence like (16) where the complement of VI and the complement of V2 both have the same thematic role, that of Theme. The only possible interpretation of (16) is 'Paul gave the water to the book'. (16)

* pOl pra dlo bay Paul take water give ""Theme Theme

liv-la book-DET Theme Goal

= 'Paul gave the water to the book' This constraint also applies to SCs Involving verbs of movement, such as pOte...ale 'carry...go'. pOte 'carry' assigns the thematic roles Theme and Goal ('carry something somewhere'), ale 'go' assigns the thematic role Goal ('go somewhere'). When pote and ale combine in an SC, the same thematic role, this time Goal, cannot be assigned to two different NPs. The only possible interpretation of (17), is 'Paul carried the house to the school'. (17)

*

pOl pOte lakay al Paul carry house go •Goal Theme

lekol-la school-DET Goal Goal

= 'Paul carried the house to the school' One might argue that the impossibility of (16) and (17) is due to general semantic constraints; i.e. it doesn't make sense to "take the water and give the book" as in (16) or to "carry to the house and go to school" as in (17).2 However, even when the expression might be semantically wellformed it is excluded, as shown in (18).

244

Rose-Marie Déchaîne

(18)

*p01 pra masan-na bay Paul take merchant-DET give *Source Theme

mari Marie Goal Goal

f 'Paul took from the merchant to give to Marie' = 'Paul gave the merchant to Marie' When pra 'take' and ba(y) 'give' combine it is not possible to assign the thematic roles Source and Goal and not assign the thematic role Theme. The only possible interpretation of (18) is 'Paul gave the merchant to Marie'. These restrictions suggest that an SC is not simply a concatenation of verbs, with VI and V2 independently assigning thematic roles to their arguments. In the following section, I will account for the syntactic as well as the lexical properties of SCs in terms of an operation which combines the argument structures of verbs.

3.

ARGUMENT STRUCTURE Before defining the operation which derives SCs, I will give a brief over-

view of the theory of argument structure which underlies this analysis. The argument structure of a verb consists of the list of its arguments. Arguments are assigned referential indices, and the thematic role of each argument is specified. For example the argument structure of the verb 'take' would be as in (19a). : (19)

a.

TAKE

( Agent,

Themej)

b.

TAKE (Agent

Themej)

Following Williams (1981), I assume that verbs distinguish between two types of arguments: the external argument and the internal arguments. Internal arguments are assigned their thematic roles by co-indexation within (or internal to) the VP, so that in (19b) the thematic role Theme is assigned, by co-indexation, to the NP internal to the VP. The index of the external argument is projected to the VP node. It is assigned to the subject position by predication, so that again in (19b) the thematic role Agent is assigned by means of percolat-

Serial Verbs in Haitian Creole

245

ion and predication to the subject position. Since a node may only be assigned one referential index the VP node can carry only one index and there will never be more than one external argument. I adopt the convention of indicating the external argument by italics. There are two types of operations which affect the argument structure of a verb. An argument can be added ('add x 1 ) or an argument can be deleted ('delete x'). I will be concerned with the addition of arguments which involves the mapping of the argument structures of two distinct predicates onto each other. Coalescence, the operation which derives SCs, involves the binding (by co-indexation) of arguments bearing the same thematic role. (20)

COALESCENCE:

Given the argument structures of two distinct verbs, VI and V2, bind arguments (by assigning them the same referential index) bearing the same thematic role.

The coalescence of pra 'take' and bay 'give' would be as in (21) with the thematic roles Agent and Theme each assigned the same index. (21)

VI TAKE Agent± Theme,

V2 GIVE Agent^ Goalj Theme.

TAKE-GIVE Agent^ Goalj Theme,,

The coalescence of pOte 'carry' and ale 'go' would be as in (22) with the thematic roles Theme and Goal each assigned the same index. (22)

VI CARRY Agent} Themej Goalk

V2 GO Agent± Theme j Goalfc

CARRY-GO Agent^ Themej Goalfc

Under the assumption that the binding of thematic roles suppresses their syntactic realization, the complex argument structure created by coalescing VI and V2 would project the syntactic representation shown in (23). By appealing to the notion of predication, as defined in Williams (1980), we are able to account for the association of a single subject NP with two distinct VPs. The rule of predication co-indexes a subject NP with the predicate that modifies it and is subject to a c-command restriction whereby the subject NP must c-command the modifying predicate.3 By virtue of the operation

246

Rose-Marie Déchaîne

of coalescence VP1 and VP2 both carry the same index which percolates to the upper VP. Since the two VPs both carry the same referential index the rule of predication can apply. Extending the conventions established by Williams (1980), the rule of predication would assign the index m to the two predicates since VP1 and VP2 are contained in and co-indexed with the upper VP. (23)

SERIAL CONSTRUCTIONS S

a.

VI

NP4

I

TAKE (Agi ,Thj )

b.

V2

IGIVE

NPi_

{Agi ,Thj ,Goalk)

S

VI

NPj

CARRY Usri,Thj ,Goalfc)

V2

NP k

GO (Thj,Goaljc)

Such an analysis also accounts for the restrictions on the realization of thematic roles. (24a-c) are excluded by virtue of the way in which the argument structures of the verbs are coalesced. Specifically, all arguments bearing the same thematic role are assigned the same referential index. The impossibility of realizing the same thematic role twice accounts for the unacceptabillty of (7) where the thematic role Agent is assigned two different indices, (16) where the thematic role Theme is assigned two different indices, (17) where the thematic role Goal is assigned two different indices. (24d) is excluded since, for coalescence to occur, the argument structures of the two verbs must have in common at least one internal thematic role.

Serial Verbs in Haitian Creole (24)

BINDING OF THEMATIC ROLES: AgentjVj

VP2

b.

*

VI

Theme j

V2

Theme^

= (16)

c.

*

VI

Goal j

V2

Goaljc

= (17)

d.

*

VI

Source

V2

Goal

= (18)

a. * Agentj^ VP1

4.

247

= (7)

CONSEQUENCES

4.1

The Theta-criterion

At first glance, the operation of coalescence, in which a single argument is assigned the same thematic role "twice", seems to violate general wellformedness conditions on the assignment of thematic roles, and more particularly the Theta-criterion (Chomsky 1981) under which an argument can bear only one thematic role. (25)

= (4) (Chomsky 1981:36) Theta-criterion: each argument bears one and only one theta-role, and each theta-role is assigned to one and only one argument.

However, since coalescence is a lexical operation involving the binding of arguments bearing the same thematic role, the theta-criterion, in a strict sense, is not violated since each argument position bears only one thematic role. 4.2 Serialization Versus Restructuring The analysis proposed here predicts that one should be able to distinguish serialization from other operations that affect the argument structure of verbs. For HC, this prediction is borne out and it is possible to differentiate serialization from restructuring and causativization by means of the order of complements and the extractability of verbs. In a SC, VI and V2 cannot be adjacent, as shown in (26b). (26)

a.

pOl

pOte

pul-yo al

na mase

Paul carry chicken-PL go LOC market 'Paul carried the chickens to the market' ^

V

b.* pOl pOte al pul-yo na mase Paul carry go chicken-PL LOC market

248

Rose-Marie Déchaîne In a restructuring context, however, VI and V2 must be adjacent as shown in

(27). I am assuming a restructuring analysis such as Rizzi's (1982) in which a complex verb is formed. (27)

a.

„

pOl al pOte pul-yo na mase Paul go carry chicken-PL LOC market 'Paul went to carry the chickens to the market"

b.* pOl al na mase pOte pul-yo Paul go LOC market carry chicken-PL Moreover, one can extract VI in a SC to form a predicate cleft construction, as shown in (28a). It is not possible to extract V2, as in (28b), and neither is it possible to extract VI and V2 as in (28c). (28)

a.

se pOte pOl pOte pul-yo al na mase FOC carry Paul carry chicken-PL go LOC market 'Paul carried the chickens to the market' v

b.* se ale pOl pOte pul-yo al na mase FOC go Paul carry chicken-PL go LOC market c.* se pOte ale pOl pOte pul-yo al na mase FOC carry go Paul carry chicken-PL go LOC market In a restructuring context VI cannot be extracted as in (29a); V2 can be extracted as in (29b); VI and V2 can be extracted as in (29c). (29)

a.* se ale pOl al pOte pul-yo na mase FOC go Paul go carry chicken-PL LOC market b.

se pOte pOl al pOte pul-yo na mase FOC carry Paul go carry chicken-PL LOC market 'Paul went to carry the chickens to the market'

c.

se al pOte pOl al pOte pul-yo na mase FOC go carry Paul go carry chicken-PL LOC market 'Paul went to carry the chickens to the market'

The differences in the extractability of VI and V2 for SCs and restructured constructions falls out from the fact that restructuring creates a complex verb as in (30) while SCs, under the analysis proposed here, are "co-ordinated" VPs as in (23).

Serial Verbs in Haitian Creole

4.3

249

Serialization Versus Causativization

Causative constructions present many of the same syntactic properties as SCs except for the presence of an NP subject before V2, as in (31a). It is also possible to cleft VI or V2 in a causative construction, as shown in (31bc) . (31)

a.

za vOye pyE mote pye -zavoka -a Jean send Pierre climb tree-avocado-DET 'Jean sent Peter to climb the avocado tree'

b.

se vOye za vOye pyE mote pye -zavoka -a FOC send Jean send Pierre climb tree-avocado-DET "Jean sent Pierre to climb the avocado tree'

c.

se mote za vOye pyE mote pye -zavoka -a FOC climb Jean send Pierre climb tree-avocado-DET 'Jean sent Pierre to climb the avocado tree'

The availabity of V2 for clefting in causative constructions falls out from the fact that one argument structure is embedded in another and not coalesced, as shown in (32).

5.

CONCLUSION The analysis developed here accounts for the syntactic as well as the

lexical properties of SCs. The presence of a single NP subject has been shown

250

Rose-Marie Déchaîne

to be the consequence of a more general restriction on thematic role assignment: all arguments bearing the same thematic role are assigned the same referential index. The absence of tense/aspect markers and co-ordinating and subordinating markers before V2 is predicted by the monoclausal representation adopted in (23) . The asymmetries observed between serialization, restructuring and causativization are predicted by this analysis since these three constructions are derived from three different operations on argument structures, as shown in (33). ADDITION: COMBINATION OF ARGUMENT STRUCTURES

(33)

DOMAIN INTRUSION

COALESCENCE

EXTERNAL ARGUMENT - causativization (Williams 1981, Zubizarreta 1985) - complex verb formation (Grimshaw and Mester 1985) - restructuring (Lefebvre and Muysken, to appear)

INTERNAL ARGUMENT -applied verb formation (Koopman 1983) - instrumentalization (Bresnan 1982, Koopman 1983) - prefixation of 'out' (Bresnan 1982) - SERIALIZATION

More generally, the difference between serializing and non-serializing grammars, under this analysis, is attributable to an operation which coalesces the argument structures of verbs. This analysis, however, does not account for the ordering characteristics of SCs. More specifically, the thematic role Theme always occurs before the other arguments in an SC. This is also a property of purposive constructions (cf. Jones 1985) and it would doubtless be enlightening to investigate the similarities between these two construction types.

NOTES 1.

Throughout this paper [E] and [O] will be used for the [-ATR] versions of [e] and [o] respectively.

2.

This was brought to my attention by Isabelle Halk who noted that purposive constructions are subject to the same semantic constraint. Thus, the sentence 'I went to school to read at home" is also semantically anomalous and is probably excluded because the same thematic role is assigned to two different arguments.

3.

Williams adopts the following definition of c-command: A c-commands B if every branching node which dominates A dominates B.

Serial Verbs in Haitian Creole

251

REFERENCES Bresnan, Joan. 1982. Polyadicity. In: The Mental Representation of Grammatical Relations, ed. by J. Bresnan, pp.149-172. Cambridge, MA: MIT Press. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris Publications. Déchaine, Rose-Marie. 1986. Opérations SUT les structures d'arguments : Le cas des constructions sérielles en haïtien. Master's thesis. Université du Québec à Montréal. Déchaine, Rose-Marie and Claire Lefebvre. In preparation. The Grammar of Serial Constructions. Givôn, Talmy. 1975. Serial Verbs and syntactic change: Niger-Congo. In: Word Order and Word Order Change, ed. by C. Li, pp.48-112. Austin: University of Texas Press. Grimshaw, Jane, and Ralf-Armin Mester. 1985. Complex verb formation in Eskimo. Natural Language and Linguistic Theory 3:1-19. Jansen, Bert, Hilda Koopman, and Pieter Muysken. 1978. Serial verbs in the Creole languages. Amsterdam Creole Studies 2:125-159. Jones, Charles F. 1985. Syntax and Thematics of Infinitival Adjuncts. Ph.D. dissertation. University of Massachusetts at Amherst. Koopman, Hilda. 1983. The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris Publications. Piou, Nanie. 1982. Le Clivage du prédicat. In: Syntaxe de l'haitien, ed. by C. Lefebvre, H. Magloire-Holly and N. Piou, pp.122-151. Ann Arbor: Karoma Publishers. Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris Publications. Roberts, Ian. 1985. Serial Verbs and Government Binding Theory. Unpublished ms., University of Southern California. Schachter, Paul. 1974. A non-transformational account of serial verbs. In: Studies in African Linguistics, Supplement 5, pp.253-270. Sylvain, Suzanne. 1936. Le créole haïtien: morphologie et syntaxe. Reprinted 1979. Geneva: Slatkin Reprints. Williams, Edwin. 1980. Predication. Linguistic Inquiry 11:203-238. Williams, Edwin. 1981. Argument structure and morphology. The Linguistic Review 1:81-114. Williams, Wayne R. 1971. Serial verb constructions in Krio. In: Studies in African Linguistics, Supplement 2, pp.47-65. Wingerd, Judy. 1977. Serial Verbs in Haitian Creole. In: Language and Linguistic Problems in Africa, ed. by P. Kotey and H. Der-Houssikian, pp.452-466. Columbia, SC: Hornbeam Press. Zubizarreta, Maria Luisa. 1985. Morphophonology and morphosyntax: the case of Romance causatives. Linguistic Inquiry 16:247-289.

Department of Linguistics University of Massachusetts at Amherst

Chapter 18

Topics in Liberian English Modality John Victor Singler

In the tense-modality-aspect triad that is referred to in pidgin and Creole studies, the focus has been on tense and aspect. Generally, modality has been neglected if not ignored altogether. The present study will consider modality in non-native Liberian English, looking first at irrealis and then at modality more generally.

1.

IRREALIS Irrealis is the term used by Bickerton (1975) and others to indicate the

combination of future and conditional. That is, in pidgins and Creoles, the same preverbal AUX is used to mark verbs expressing future events or states and verbs occurring in the consequent clauses of conditionals. While this collapse of future and conditional into a single category is characteristic of pidgins and Creoles, that does not mean that it is only found there. Liberian Niger-Congo languages, among others, do the same thing. Singler (1984) argues for the validity of applying Creole models including the continuum, when appropriate, to the stable pidgins of West Africa and Papua New Guinea. In (1) is given the basic model of irrealis along the continuum in non-native Liberian English. (1)

BASIC IRREALIS IN LIBERIAN ENGLISH Future Irrealis

go

we •wud

(we

>

SE

will ; wud

>

SE

Conditional

would)

Some adjustments do have to be made to the model in (1); nevertheless, for most instances of future and conditional, irrealis in Liberian English is as uncomplicated as the model suggests. That is, basilectal speakers (those farthest from Standard English) use go to mark future and conditional.

254

John Victor Singler

More mesolectally, go shifts to we, which - like go - is used to mark both future and conditional. Finally, in the acrolectal extreme (the end of the continuum closest to Standard English), the Standard English distinction between future and conditional is observed, with we restricted to future and wud used for conditionals. The word go is the primary irrealis marker used throughout West Africa and in some New World Creoles as well.1 Accordingly, the model in (1), or something like it, can be extended to other pidgins and Creoles besides Liberian. However, Liberian English would seem to differ from other West African pidgins and Creoles in the pervasiveness of we as opposed to go. Everybody uses we at least some of the time. This statement is borne out by the speakers in the corpus used in Singler (1984), a group of fifty-five speakers whose speech spans the Liberian continuum. As for the other end of the continuum, the use of wud in Liberian English is rare. While the corpus contains more than 2,000 irrealis uses of we, there are only 29 tokens of wud. Of these, 22 express politeness or tentativeness, as illustrated in.(2).2 (2)

JVS: Where do you want to go? CALVIN: a wud lak tu go tu marovia. I MOD like to go to Monrovia. 'I would like to go to Monrovia'.

The model in (1), while basically sound, requires two adjustments in order to give a more nearly complete picture of irrealis in Liberian English: the addition of another irrealis marker (discussed in 1.1) and the addition of another type of irrealis (discussed in 1.2). 1.1 ke The additional irrealis marker is kc (from Standard English 'can'). Apart from its irrealis role, ke has various uses in different subsets of the Liberian English continuum. It is widely used mesolectally to mark habitual and iterative actions and is used rather generally to mark potentiality. Its use as an irrealis marker is limited to the Mande-speaking interior of Liberia, i.e. the northwestern counties of Lofa, Nimba, and Bong. Reflecting this geographical distribution, Singler (1984) refers to this variety of Liberian English as Lofa-Nimba Pidgin. It is essentially the same variety as that which Hancock (1971) calls Soldier English.

Topics in Liberian English Modality

255

In general, irrealis expresses a prediction that something will happen, not merely that it could happen. Thus, to establish that ke is a genuine irrealis marker, it is necessary to demonstrate that its use does not simply express mere potentiality, but rather makes the prediction that the event thus marked will occur. Two examples of the use of ice by Lofa-Nimba speakers make this point. (3)

heloba-heloba ma de, he'. helluva-helluva man there, heyI

efi i ste da chia if he stay that chair

de, pipo kt_ ke, no go-ko. there, people MOD kill, no go-come. 'That was quite a man there, heyI If he had still been in his Cabinet position [at the time of the 1980 coup], they would have executed him. There's no two ways about it'. (Nimba Watchman) The no go-la

at the end of (3) underscores Nimba Watchman's certainty that the

Cabinet Minister he is talking about would have been executed. (4)

de ko, de sem leta fo sail ke, de se, they come, they send letter for Charlie King, they say, "yu mos dravi di pipo wi go fo wia ho. efi sali "you MOD drive the people we go for our home, if Charlie I no sem de, fro hya tu bi tu oklo wi ke ke King you no send them, from here to be two o'clock we MOD fway fo des ta hya". fight for this town here". 'They the Germans came, and they sent a letter to [President] Charlie King, saying, "You must expel the people so that we can take them with us to Germany. If, Charlie King, you don't send them to us by two o'clock, we will start shelling your town".' (Lofa Overseer)

The example in (4) is in reference to an incident that occurred during World War I. A German gunboat sailed into the Monrovia harbor, and its captain sent an ultimatum to the Liberian President: "Hand the British and French consuls over to us, or we shall bombard the city". In this case, the Germans were not merely introducing the possibility of fighting; they were announcing their intention to fight. In fact - as Lofa Overseer points out subsequently in his narrative - when the Liberian government did not surrender the consuls, the Germans shelled the harbor. This use of ke to mark irrealis is also attested in Tok Pisin, spoken in Papua New Guinea. According to Wurm (1971:48) , "Verb forms preceded by ken +

256

John Victor Singler

[ the predicate marker] i indicate actions which will definitely take place at a future time. This form is very commonly used in Highlands Pidgin ..." This is illustrated in (5): (5)

botol i ken i bruk. 'The bottle will certainly break'. (Tok Pisin) In the Lofa-Nimba case, ke is apparently a legacy from the coast. The

evidence for this suggestion comes from Grade (1892). The other Liberian basilect today is grounded in Kru Pidgin English, the speech of the Kru sailors. Grade's article is about the "Neger-Englisch an der Westkiiste von Afrika". From Grade's description of his informants for the article, it seems clear that he is describing a late nineteenth century version of Kru Pidgin English. Presenting the tenses of the speech variety, Grade identifies go as the future AUX (1892:384). In another section on AUX's, however, he says that Standard English 'will' and 'shall' can be "replaced" by can, and he illustrates this point with the example in (6). (6)

him can die. 'He will die'.

(Neger-Englisch [Kru Pidgin English?] ) (Grade 1892:348)

h If ice was still in use by Kru Pidgin English speakers in the first half of this century, it would have been part of the Kru Pidgin English input to the repidginization that occurred in the Liberian Frontier Force and at the Firestone Rubber Plantation, the two principal sites for the development of the Lofa-Nimba Pidgin. That is, it seems that the Lofa-Nimba Pidgin acquired irrealis ke from Kru Pidgin English and then, subsequently, Kru Pidgin English ceased to use ke as an irrealis marker. 1.2

Immediate Irrealis

The model in (1), with the addition to the Lofa-Nimba basilect of ke, accounts for what may be called basic or unmarked irrealis. However, there is a second type of irrealis, illustrated in (7) and (8): (7)

bo a kDme se de jz>s na, me. but I MOD send them just now, man.

se da. sit down.

'But I'm gonna send 'em right now, man. Sit down'. (Charles)

257

Topics in Libetian English Modality a ke se da hya, de se, "de goe pie des I HAB sit down here, they say, "they MOD play this

(8)

soso pies o." such place ." 'When I'm sitting down here, sometimes they come and say, "They're going to dance and sing at such and such a place".' (Rally Time) The modals tome and goe

and the verbal suffix -e (from Standard English '-ing')

are all used to signal events or states that are on the brink of occurring. (Often, an appropriate Standard English gloss for such sentences involves 'about to', rather than 'will' or 'gonna'.) Bickerton's (1975) prototypical Creole tense-modality-aspect system contains no distinction like basic versus immediate irrealis. On the other hand, such a distinction does occur in several of Liberia's Niger-Congo languages. These include Klao (Singler 1979:5), Kpelle (Thach 1981:82), and Loma (Dwyer 1981: 85). In all three of these cases, the linguists describing the language use the term "immediate future". Klao uses a form of 'come' as its immediate future marker, Kpelle uses a form of 'go', and Loma uses forms of both 'come' and 'go'.

2.

MODALITY Futurity and conditionals comprise one type of modality. As to modality more

generally, the fundamental distinctions in the literature involve kind of modality and degree of modality. Kind opposes epistemic modality to deontic, and degree distinguishes possibility from necessity, permission from obligation. Epistemic refers to evaluating the truth of propositions, to knowing. Deontic, on the other hand, refers to actions and to expressing "what is obligatory, permitted, or forbidden" (Palmer 1979:3). (Deontic constructions frequently have a performative character.) That the same modals can be used for both kinds of modality is illustrated by the chart in (9): (9)

Modality in Standard English Epistemic Possible 'I may have left it in the car'. Necessary • It must be there'.

Deontic Permissible 'You may take as many as you like'. Obligatory 'You must do what she tells you to do'.

258

John Victor Singler

In Standard English, there is a wide range of auxiliaries that span the ground between possible and necessary and between permissible and obligatory, including, for example, 2.1

1

should1.

The Hortative

A second perspective on modality comes from Welmers (1973). Though he does not frame it in terms of modality, he notes that in many Niger-Congo languages there is a construction that he labels the "hortative1, i.e. a construction used to 'exhort'. In Kpelle, the hortative is signalled by tone changes in the subject pronoun and verb (1973:133). In Dewoin, a Western Kru language spoken in Liberia, the AUX we marks the hortative in main clauses, while ma marks it in subordinate clauses (Welmers 1977:345). With regard to the sense of the hortative construction, Welmers says that, when used with a first-plural subject, 'let' is an appropriate translation, as in (10a). When the hortative is used with third-person subjects, however, 'should', not 'let', is the appropriate gloss, as in (10b): (10)

a. ku pill 'Let's jump'.

(Kpelle [Welmers 1973:133])

b. 4 pill 'He should jump'. According to Welmers, the hortative is rarely used with first-singular subjects except in questions or embedded sentences, i.e. "Should I jump?" and "He told me I should jump". In.the case of second-person subjects, for many languages the hortative gives way to the imperative. The distribution of the hortative vis-a-vis person and number correlates with the sense of exhortation upon which Welmers has built his description of the construction. One can exhort a group of which he or she is a member (firstperson plural subject), and one can exhort third parties, but one does not ordinarily exhort oneself (first-person singular). As for exhorting addressees (second-person subject), the imperative construction is usually used for this. From Welmers1s description of the hortative, it is clear that it falls into the realm of deontic modality, somewhere between permission and obligation. To return to Liberian English, in terms of modality, particularly modality expressed by AUX's, what shows up is almost all deontic. Epistemic modality is more marginal. Epistemic possibility is most commonly expressed by the % ^ adverb sota, as in (11):

Topics in Libeiian English Modality (11)

259

a se, "sota di pipo si ro, de go". I say» "MOD the people see road, they go". 'I said, "Maybe the people found the road and went ahead".' (Nimba Cook)

Epistemic necessity can be expressed by mos, as in (12):^ (12)

JVS: But the time they came, I wonder they met anyone here. GEDEH CHILDMXNDER: yu mi, you mean, JVS: The time they came from Ivory Coast. GEDEH CHILDMINDER: ye, to da ta de ke e mi, yes, but that time them came and meet, . ^ 'b mebi de we mi sDbadi. e iros tu bi sobadi, bo maybe them MOD meet somebody, they MOD to be somebody, but da pipo da. that people die. GEDEH CHILDMINDER: Yes, when they came, maybe they encountered somebody. There had to have been someone here, but whoever it was is dead now.

However, constructions like the one in (12) expressing epistemic necessity occur only rarely. 2.2

The Deontic Modals: le and nOs

Deontic modals, on the other hand, are quite common. In the Liberian English basilects, le and mos are dominant, other modals occurring infrequently if at all. (Among some Kru Pidgin English speakers, clause-initial mek does appear.) Deontic le and nos are illustrated in (13) and (14). (13)

le wi go wo say, wi se da de. MOD we go one side, we sit down there. 'Let's go over there and settle'.

(14)

(Lofa Tapper)

yu rish fo ta, yu se, "o ray, ta chi. a you reach for town, you say, "all right, town chief. I beg pDta. ge mi pota. des ma BOS go ba". beg porter, give me porter, this man MOD go back". 'When you got to a town, you would say, "All right, town chief. I want a porter; give me a porter. This other man should return to his town".' (Gedeh Soldier)

In examining le and nOs, their meaning, their distribution, and their inter-

260

John Victor Singler

action will be considered. Most speakers use both, with nos occurring more often than le. There are exceptions to this, however. For example, one of the Nimba speakers represented in the corpus uses le 79 times but never uses mos. For the most part, the distribution of le vis-a-vis subject pronouns fits Welmers's description of the hortative. It occurs frequently with first-plural and third-person subjects, as in (13) and (15): (15)

le_ no 1 i gev o di mashe. MOD give us the matches. 'Molley should give us the matches'.

(Lofa Shopkeeper)

Further, le occurs only rarely with second-person subjects. As a rule, when this happens, it is because the imperative is contextually, i.e. socially, inappropriate, as in (16). In the example in (16), the speaker is part of a group of guests, and the guests are eating from a different pan from that used by their hosts. The suggestion in (16) to put all the food in one pan is meant to give the hosts the chance to eat as much as the guests. (16)

wi te we tell 'We said that we

de, "le e pu di wo fu, le wi i". them, "MOD y'all put the one food, MOD we eat". to them, "Y'all should put all the food in one pan so can all eat together".' (Nimba Cook)

The difference in distribution between Welmers's hortative and Liberian English le occurs in the remaining category - first-singular subjects - where le occurs frequently, as in (17) and (18). (17)

le mi ge di asa. MOD me give the answer. 'Let me give the answer'.

(18)

(Lofa Shopkeeper)

we yu rish fo ta, yu se, "we, a tay. when you reach for town, you say, "well, I tire.

ie_ mi MOD me

slip hya". sleep here". 'When you got to a town, you would say, "Well, I'm tired. I ought to sleep here".' (Gedeh Soldier) Almost all of the occurrences of le with first-singular subjects are like the example in (17). A better translation, perhaps, of (17) would be, 'I'll give

261

Topics in Liberian English Modality the answer'. That is, such uses represent an assertive statement by the

speaker. If modal, they are non-deontic. Truly deontie sentences like (18) are not at all common. In contrast to le, mDs occurs most often with second-person subjects, as in (19) and (20): (19)

so hi se di peso wu we men da ge de, so he say the person who MOD marry that girl there, di peso BOS fes go we the person MOD fix gun with

soso TO. rock.

'So he said that whoever would marry that girl had to construct a gun using nothing but rocks1. (Aesop) (20)

di bos, so o pipo de, soso rO de pu esay some of people there, rock they put inside the box, de se "go mos tek e Jce". they say, "yall MOD take and carry". 'As for some of us, they [soldiers] filled boxes with rocks and ordered us to carry them'. (Lofa Tailor)

The modal nos also occurs frequently with third-person subjects, as in (14). As noted above, Welmers's comment that the first-singular hortative forms are usually confined to questions and embedded sentences does not apply to le; it does apply to mos, as in (21) and (22) . (21)

a mos Jco? I MOD come? ' May I come with you?1

(22)

lutene se a mos go. lieutenant say I MOD go. 'The lieutenant said that I must go'.

(Gedeh Soldier)

Almost without exception, uses of nos with first-plural subjects are like the cases with first-singular subjects: that is, they involve either questions or embedding. The relationship to the person and number of the subject is shown in (23) , for the hortative in Niger-Congo, for le, and for mos.

262

John Victor Singler

(23)

sg

pi

sg

pi

sg

Niger-Congo Hortative (Weimers 1973)

B

pi

mos

frequently occurring

r« Q/Erestrictedto questions and embedded sentences

As the chart in (23) indicates, Welmers's Niger-Congo hortative and le are identical in distribution except for the case of first-singular subjects. There, whether one considers sentences like (17) modal or not (and I have not in (23)), ie's behavior differs from the Niger-Congo paradigm. However, that is the only place where it differs. The discussion thus far of le and mOs has followed Welmers in arranging their distribution according to the person and number of the subject. As for the meaning of the two forms, for the most part they both involve Welmers's hortative. In terms of a deontic continuum, the range of le is generally nearer permission and the range of mos nearer obligation. A qualifying word like "generally" is needed because there are ample cases where le is used forcefully and nos tentatively (and where Standard English intuitions about 'let' and 'must' are entirely inapplicable). Consider sentences like (24) and (25): (24)

1utene se, "o ray. go de. le di pipo to", lieutenant say, "all right, go there. MOD the people come". 'The lieutenant said, "All right. Go there and tell the people to come".' (Nimba Watchman)

(25)

i se, "no, a no go gri. le_ hi to e du e koko wDk". he say, "no, I no MOD agree. MOD he come and do a cocoa work". 'He [the headman] said, "No, I won't give my permission. Tell him to come here and work on the cocoa".' (Ghana Steward)

Topics in Liberian English Modality

263

In each case, the person whose speech is being reported has absolute authority: he doesn't exhort; he gives orders. In each case, however, le has been used. The ability of nos to express non-obligatory deontic modality is illustrated in (26) and (27).

(26)

'V 'V 'Xi we wi tra de so de, wi se, when we try them some day, we say, pli tra e hep o we des", please try and help us with this",

"wi jam, "we , de ke they NEG

yo mas y'all MOD du e, de do it, they

grona bo. se wi a say we are boy. 'When we ask [our father's relatives here in Monrovia] for help, we say, "We're really hard up. You should please try to help us", but they refuse. They call us juvenile delinquents'. (Nimba Vendor) (27)

de di oma se hi lak tu ke mi so hi ke res then the oldman say he like to carry me so he can raise mi. so, ma fada se, "o ray. yu nos ke hi". me. so, my father say, "all right, you MOD carry him". 'Then the oldman said that he would like to take me with him so that he could raise me. So, my father said, "All right. You may take him with you".' (Rally Time)

In (26) , Nimba Vendor and his brother are supplicants and not in a position to impose obligations. In (27), Rally Time's father is giving permission, not ordering. In general, the more tentative uses of nos occur with second-person subjects, i.e. the subjects where le is not generally available. A further point with regard to nos and its less forceful uses involves a tendency that has been widely noted in the study of pidgin and Creole languages. That is, the form that is marked or stronger in the target language frequently enters the pidgin or Creole as the unmarked, neutral element. Thus, Standard English 'too' as in 'too tall' loses its sense of excess and means simply 'very' in pidgins and Creoles. Similarly, the Haitian definitizers come not from French definitizers but from French deictics. With regard to the relation of le to nos, there is no neat boundary between the two. This is illustrated by examples like those in (24) through (27), where le expresses deontic obligation while nos expresses deontic permissibility. The overlap of the meaning of le and nos is further illustrated by their appearance in similar sentences from a single speaker, as in (28).

264

John Victor Singlei

(28)

a. go mas ge mi stre. God MOD give me strength. 1

May God give me strength'.

»v. b. _Ie go ge mi lo lay. MOD God give me long life. •May God give me long life'.

(Bettee)

Examples like (28) notwithstanding, le and mos are not simply distributional variants of a single modal sense. The fact that each occurs frequently with third-person subjects suggests that they convey distinct meanings. A final point with regard to le and mos and their meaning is that they can co-occur, as in (29a). Where the context sets up a hortative sense, the removal of one of the modals does not seem to alter the sense of the sentence appreciatively; thus, (29b) and (29c) have the same force as (29a). (29)

a. a se, "di bes te, le_ yu pipo mos go tu sku". I say, "the best thing, MOD you people MOD go to school". 'I tell them, "The best thing for you people would be to go to school".' (Nimba Vendor) 'v b. a se, "di bes te, le_ yu pipo go tu sku". c. a se, "di bes te, yu pipo HPS go tu sku".

Sentence (30) is also hortative. (30)

JVS : You want to wait •:ill we finish? GEDEH GOLDMINER:

i£ a mos feni. MOD I MOD finish. •Yeah, I'll finish, and then we'll do it'.

Here, though, perhaps because the subject is first-singular, removing one or the other of the modals matters and changes the meaning. Finally, there is (31). (31)

"so eni basa ma hu go di gio pipo fo ta, "so any Bassa man who got the Gio people for town, i nos ge tu comeshena denes, le i mos ke". he MOD give to Commissioner Dennis, MOD he MOD carry". '[So the President said:] "So any Bassa man who has Gio people living with him, he must surrender them to Commissioner Dennis so that Dennis can return them to Gio country".' (Gedeh Soldier)

Topics in Liberian English Modality

265

In the relevant clause in (31) , President Tubman is giving an order to the Commissioner - clearly a case of deontic obligation. The instances of lc and nos shown in (24) through (31), examples that show that each occupies a fairly large range of the deontic continuum, represent a parallel to the Niger-Congo hortative. While Standard English modals share a number of distinct functions, the Niger-Congo hortative - and 1E and IDS are largely confined to the deontic continuum. But where Standard English deontic modals are, comparatively speaking, rather precise in their meaning, the Niger-Congo hortative and le and nos each occupy much of the range from permission to obligation.

3.

CONCLUSION Pidgin and Creole languages draw their features from the target language,

from universals including pragmatic universals, from happenstance, and from their substrates. What one has in the case of Liberian English modality is what one has so often in Liberian English: English-based forms representing Niger-Congo-based syntax and semantics.

NOTES *David Peewee, Samson Tiklo, Boakai Zoludua, and Sumoyea Guluma assisted me in collecting the data upon which this study is based, and I am grateful to them for their assistance. 1.

In addition to go, Todd (1984) reports that Cameroon Pidgin English also has the conditional marker fo, illustrated by the following sentence: A fo go if a gzt chans. 'I will go if I get the chance (but it is most unlikely)'. (Todd 1984:143).

2.

The pseudonyms used here are taken from Singler (1984). Speakers who learned Liberian English as adults have been given occupational pseudonyms, e.g. Bricklayer. Speakers who began learning Liberian English as children have been given non-occupational pseudonyms, e.g. Calvin. The geographic terms Lofa, Nimba, and Gedeh refer to the part of the interior that the speaker comes from. The Lofa and Nimba speakers in the corpus have a Mande language as their first language, while the Gedeh speakers have Krahn, a Kru language, as their first language. Ghana is applied to Kru Pidgin English speakers of Liberian English.

3.

With regard to the pronunciation of mos the vowel is pronounced [3] as well as [o], and the final consonant is frequently deleted. All four pronunciations ([ra9 ], [mo], [m3s], and [mOs]) occur frequently. A fifth, [mosi] , is confined to the basilect.

266

John Victor Singler

REFERENCES Bickerton, Derek. 1975. Creolization, linguistic universals, natural semantax, and the brain. Paper presented at the Hawaii Creole Conference. Dwyer, David J. 1981. A Reference Handbook of Lorma. East Lansing: Michigan State University, African Studies Center. Grade, P. 1892. Das Neger-Englisch an der Westküste von Afrika. Anglia 14:362-393. Hancock, Ian F. 1971. Some aspects of English in Liberia. Liberian Studies Journal 3:207-213. Palmer, F.R. 1979. Modality and the English Modals. London: Longman. Singler, John Victor. 1979. The segmental phonology of verb suffixes in Talo Klao (Kru). M.A. thesis, UCLA. Singler, John Victor. 1984. Variation in Tense-Aspect-Modality in Liberian English. Ph.D. dissertation, UCLA. Thach, Sharon, (with David J. Dwyer). 1981. A Reference Handbook of Kpelle. East Lansing: Michigan State University, African Studies Center. Todd, Loreto. 1984. Modern Englishes: Pidgins and Creoles. Oxford: Basil Blackwell. Weimers, William E. 1973. African Language Structures. Berkeley & Los Angeles: University of California Press. Weimers, William E. 1977. Mood in Dewoin. In: Language and Linguistic Problems in Africa, ed. by Paul F.A. Kotey and Haig der-Houssikian, pp. 344-361. Columbia, SC: Hornbeam. Wurm, S.A. 1971. New Guinea Highlands Pidgin: Course Materials. (Pacific Linguistics, Special Publications, D3.) Canberra: Australia National University.

Department of Linguistics New York University