Experimental linguistics : integration of theories and applications 9789027270863, 9027270864

Table of contents :
AFRICAN LINGUISTICS Essays in Memory of M.W.K. Semikenke
Editorial page
Title page
Copyright page
PREFACE
OBITUARY SEMIKENKE MUZEYI WA KAZOZA
Table of contents
ON THE 'COMPOUND' TONE IN MIGAAMA
REFERENCES
VERBALAGREEMENTAS A NONCYCLIC RULE IN BANTU
1.0 BACKGROUND
2.0 GRAMMATICAL AGREEMENT IN BANTU
3.0 CONCLUSIONS AND THEORETICAL IMPLICATIONS
FOOTNOTES
REFERENCES
AKAN VOWEL HARMONY: A NONLINEAR ANALYSIS
APPENDIX
ACKNOWLEDGEMENTS
NOTES
REFERENCES
TONS ET ACCENTS EN KINYARWANDA
1. LES TONS ET LEUR REALISATION EN FONCTION DES CONTEXTES TONALS ET DE LEUR POSITION DANS L'ENONCE
1.1. Deux tons phonologiques
1 .2. Un ton par more
1.3. Principales variantes de réalisation des tons
1.3.3. La réalisation de tons successifs à l'initiale d'énoncé
1.3.2. La réalisation de tons hauts successifs après un ton bas
1.3.4. Le rabaissement d'un ton bas en finale d'énoncé
1.3.5. La réalisation descendante d'un ton haut en finale d ' énoncé
1.3.6. La réalisation descendante du ton haut préfinal devant un ton bas en finale d'énoncé
2. SCHEMESTONALS NOMINAUX ET ACCENT
2.1. Les schèmes tonals
2.1.1. Les schèmes tonals des noms à thème d'une more
2.1.1.1. Les schèmes tonals des noms à thème d'une more de ton bas
2.1.1.2. Les schèmes tonals des noms à thème d'une more de ton haut
2.1.2. Les schèmes tonals des noms à thème de deux mores
2.1.2.1. Les schèmes tonals des noms dont le thème est de scheme bas-bas
2.1.2.2. Les schèmes tonals des noms dont le thème est de schème haut-bas
2.1.2.3. Schèmes tonals des noms dont le thème est de schème haut-haut
2.1.3. Les schèmes tonals des noms à thème de trois mores
2.1.3.1. Les schèmes tonals des noms dont le thème est de schème bas-bas-bas
2.1.3.2. Les schèmes tonals des noms dont le thème est de schème haut-bas-bas
2.1.3.3. Les schèmes tonals des noms dont le thème est de schème haut-haut-bas
2.1.3.4. Les schèmes tonals des noms dont le thème est de schème haut-haut-haut
2.1.4. Les schèmes tonals des noms dont le thème dépasse trois mores
2.1.4.1. Schèmes des noms dont le thème ne comporte que des tons bas
2.1.4.2. Schèmes des noms dont le thème comporte un nombre impair de tons hauts
2.1.4.3. Schèmes des noms dont le thème comporte un nombre pair de tons hauts
2.2. Schèmes tonals nominaux et accent
2.2.1. Schèmes des thèmes et accent
2.2.2. Schèmes tonals des noms et accent
2.2.2.1. Formes entièrement à tons bas et absence d'accent
2.2.2.2. Formes ne comportant que le thème seul
2.2.2.1. Formes nominales comportant soit un préfixe
2.3. Conclusion
2. SCHEMESTONALS DES INFINITIFS ET ACCENT
3.1. Les schèmes tonals des infinitifs
3,1.1. Les schèmes tonals des infinitifs sans dérivatif ni infixe
3.1.1.1. Les schèmes des infinitifs formés sur des thèmes sans ton haut
3.1.1.2. Schèmes des infinitifs dont le thème comporte un ton haut
3.1.2. Les schèmes tonals des infinitifs contenant des dérivatifs
3.1.3. Les schèmes tonals des infinitifs avec infixes
3.1.3.1. Les schèmes des infinitifs avec infixes dont le thème ne comporte que des tons bas
3.1.3.2. Les schèmes tonals des infinitifs avec infixes dont le thème comporte un ton haut
3.2. Les schèmes tonals des infinitifs et l'accent
3.2.2. Schèmes tonals des thèmes et accent
3.2.3. Schèmes des infinitifs et accent
3.2.3.1. Les formes à tons bas
3.2.3.2. Le ton bas des dérivatifs
3.2.3.3. Les formes d'infinitifs comportant des tons hauts et accent
3.3. Conclusion
4. SCHEMES TONALS VERBAUX ET ACCENT
4.1. Schèmes tonals des verbes à l'immédiat, au récent et au prétérit de l'indicatif affirmatif et accent tonal
4.1,1. Les schèmes tonals des verbes a l'immédiat, au récent et au prétérit de l'indicatif affirmatif
4.1.1.1. Schèmes tonals des verbes à l'indicatif immédiat imperfectif disjoint
4.1.1.2. Les schèmes tonals des verbes à l'indicatif immédiat perfectif disjoint
4.1.1.3. Les schèmes tonals des verbes à l'indicatif immédiat imperfectif conjoint
4.1.1.4. Les schèmes tonals des verbes à l'indicatif immédiat perfectif conjoint
4.1.1.5. Les schèmes tonals des verbes à l'indicatif récent imperfectif conjoint ou disjoint
4.1.1.6. Les schèmes tonals des verbes à l'indicatif récent perfectif conjoint
4.1.1.7. Les schèmes tonals des verbes à l'indicatif prétérit imperfectif disjoint
4.1.1.8. Les schèmes tonals des verbes à l'indicatif prétérit perfectif disjoint
4.1.1.9. Les schèmes tonals des verbes à l'indicatif prétérit imperfectif conjoint
4.1.1.10 Les schèmes tonals des verbes à l'indicatif prétérit perfectif conjoint
4.1.2. Schèmes tonals et accent
4.1.2.1. Schèmes tonals des thèmes et accent
4.1.2.2. Schèmes tonals des formes en ra et accent
4.1.3. Conclusion
4.2. Schèmes tonals des verbes au subjonctif (immédiat) et accent
4.2.1. Les schèmes tonals des verbes au subjonctif (immédiat)
4.2.1.1. Schèmes tonals des verbes à thème d'une more
4.2.1.2. Schèmes tonals des verbes à thème de deux mores (avec noyau de deux mores)
4.2.1.3. Les schèmes tonals des verbes à thèmes de trois mores (avec noyau de trois mores)
4.2.1.4. Les tons des thèmes et les tons des dérivatifs
4.2.2. Les schèmes tonals et l'accent
4.2.2.1. Les schèmes tonals des noyaux et l'accent
4.2.2.2. Le ton bas des dérivatifs
4.2.2.3. Le noyau: unité accentuelle
4.2.2.4. Les schèmes tonals des formes au subjonctif et accent
4.2.2. Conclusion
4.3. Schèmes tonals des verbes à l'indicatif immédiat, récent, prétérit négatif et accent
4.3.1. Les schèmes tonals des verbes à l'indicatif immédiat, récent prétérit négatif
4.3.1.1. Les schèmes tonals des verbes à l'indicatif immédiat imperfectif négatif
4.3.1.2. Les schèmes tonals des verbes à l'indicatif immédiat perfectif
4.3.1.3. Les schèmes tonals des verbes à l'indicatif récent imperfectif négatif
4.3.1.4. Les schèmes tonals des verbes à l'indicatif récent perfectif négatif
4.3.1.4. Les schèmes tonals des verbes à l'indicatif prétérit imperfectif négatif
4.3.1.5. Les schèmes tonals des verbes à l'indicatif prétérit perfectif négatif
4.3.2. Les schèmes tonals des verbes à l'indicatif immédiat, récent et prétérit négatif et accent
4.3.2.1. Les schèmes tonals des thèmes et l'accent
4.3.2.2. Schèmes tonals de ces formes verbales négatives et accent
4.3.3. Conclusion
4.4. Schèmes tonals des verbes au subjonctif négatif et accent
4.4.1. Les schèmes tonals des verbes au subjonctif négatif
4.3.1.1. Les schèmes tonals des verbes à thème d'une more
4.3.1.2. Les schèmes tonals des verbes à noyau de deux ou trois mores
4.4.2. Schèmes tonals et accent
4.4.3. Conclusion
5. CONCLUSION SUR L'ETUDE DES SCHEMES TONALS VERBAUX
6. CONCLUSION GENERALE
FOOTNOTES
REFERENCES
REMARQUES SUR LES SYSTEMES DE TRANSCRIPTION DE CES AUTEURS
ON TONE IN SUKUMA
1. INTRODUCTION
2. A FIRST LOOK AT THE SUKUMA TONAL SYSTEM
3. VERBAL INFLECTION PATTERN
4. LINKED TONES IN THE VERBAL MORPHOLOGY: ΚΑ-PAST
5. TONOLOGY OF NOUNS
6. TENSE MARKERS AND FINAL MORPHEMES
7. SPECULATIONS ON THE HISTORICAL EVOLUTION OF SUKUMA, DISJUNCTIVITY, AND LEXICAL PHONOLOGY
8. CONCLUSION
FOOTNOTE
REFERENCES
L'EXPRESSION DU PASSIF EN MANINKA
III. OBSERVATIONS GENERALES
NOTES
NOTE BIBLIOGRAPHIQUE
THE AUTOSEGMENTAL AND METRICAL NATURE OF TONE TERRACING
1. INTRODUCTION
2. IMPORTANT PROPERTIES OF TONE TERRACING
3. AN AUTOSEGMENTAL-METRICAL ACCOUNT
4. HOW THE PROPERTIES FALL OUT
5. COMPARISON WITH ALTERNATIVE TREATMENTS
6. REFINEMENTS AND EXTENSIONS
FOOTNOTES
REFERENCES
GLOBALITY IN THE KINANDE TONE SYSTEM
0. INTRODUCTION
1. THE VERB RADICAL (R)
2. THE FINAL VOWEL (FV)
3. H TONE ANTICIPATION
4. RULE ORDERING PARADOX
5. CAPTURING GLOBALITY
6. THE RULES
7. SAMPLE DERIVATIONS
8. CONCLUSION
FOOTNOTES
REFERENCES
LE MOT CHOMAGE ET SES IMPLICATIONS SOCIO-POLITIQUES DANS LA VOIX DU CONGOLAIS (1955-1959)
NOTES
A NON-LINEAR ACCOUNT OF THE SYLLABLE IN LUGANDA
1.1. INTRODUCTION
1.2. STRONG CONSONANTS
1.3. VOWEL LENGTH
1.4. INTERACTION BETWEEN CONSONANT STRENGTH AND VOWEL LENGTH
2.1. THEORETICAL FRAMEWORK
2.2. THE SYLLABLE HIERARCHY IN LÜGANDA
3. CONCLUSION
FOOTNOTES
REFERENCES
ON THE SYLLABLE STRUCTURE OF CERTAIN WEST AFRICAN LANGUAGES
FOOTNOTES
REFERENCES
ON JUNCTION IN TWO KISIU NASAL CLASSES
THE PHONOLOGY OF STATE IN KABYLE BERBER
FOOTNOTES
REFERENCES
ON THE CORRESPONDENCE BETWEEN LINGUISTIC TONE AND MUSICAL MELODY
1. INTRODUCTION.
2. A STUDY OF WAKAR INDEFENDAN, "THE SONG OF INDEPENDENCE"
3. SURVEY OF OTHER SOURCES
4. INTONATION IN HAUSA
5. DISCUSSION
REFERENCES
AN ACCENTUAL APPROACH TO TONE IN KIMATUUMBI
1. PREVIOUS WORKS ON ACCENT
2. ACCENT IN NOUNS
2.1. A first approximation
2.2. H tone assignment and unaccented nouns
2.3. Accented long vowels
2.4. Post Prefixal H Tone Assignment
2.5. Retraction
2.6. Tone copy
2.7. Tone Copy and Retraction
2.8. Leftward Shift
2.9 Deaccentuation
2.10 Nominal Retraction
2.11 CVVCV stems
2.12 A Reconsideration of phrase-medial position
2.12 Doubly-accented stems
2.13 Summary
3. THE VERBAL TONE SYSTEM
4. COMPARISON OF TONE AND ACCENT IN KIMATUUMBI
5. ACCENT AND TONE IN BANTU LANGUAGES
5.1 Distinguishing tone and accent
5.2 Historical change
5.3 Accent and Stress
6. SUMMARY
FOOTNOTES
REFERENCES
A LEXICAL TREATMENT OF TONE IN TIV
0. LEXICAL PHONOLOGY
1.0 The cycle
1.1.1. Downstep
1.1.1. H-spread
2.0. Automatic spreading
2.1. Association conventions and the WFC
2.2. Past tense forms
2.3. H-spread
2.4. Habitual 3
2.5. Association conventions
2.5.1. Floating tone
2.5.2. The Elsewhere Condition
3.0. Lexical constraints
3.1. Contour tones
3.2. Lexical constraints
4. ORDERING
5. DERIVED VS. NONDERIVED
6. DEFAULT TONE
7. LEXICAL EXCEPTIONS
7.1. Future
7.2.!-deletion
8. MORPHOLOGICAL ENCODING
8.1. Boundary symbols
8.2. Variables
9. CONCLUSIONe
FOOTNOTES
APPENDIX
REFERENCES
SWAHILI QUASI-PASSIVES : THE QUESTION OF CONTEXT
FOOTNOTES
REFERENCES
THE STATUS OF LEXICAL ASSOCIATIONS AND THE OBLIGATORY CONTOUR PRINCIPLE IN THE ANALYSIS OF TONE LANGUAGES
FOOTNOTES
REFERENCES

Citation preview

AFRICAN LINGUISTICS

SSLS STUDIES IN THE SCIENCES OF LANGUAGE SERIES

General Editor Didier L. GOYVAERTS

Volume 6

Didier L. Goyvaerts (ed.) African Linguistics Essays in Memory of M. W. K. Semikenke

AFRICAN LINGUISTICS Essays in Memory of M.W.K. Semikenke

Edited by DIDIER L. GOYVAERTS

JOHN BENJAMINS PUBLISHING COMPANY Amsterdam/Philadelphia 1985

Library of Congress Cataloging in Publication Data Main entry under title: African linguistics. (Studies in the sciences of language series, ISBN 90 6439 212 9; 6) Bibliography: p. 1. African languages — Addresses, essays, lectures. 2. Linguistics -- Addresses, essays, lec­ tures. 3. Semikenke Muzeyi wa Kazoza, d. 1982 -- Addresses, essays, lectures. I. Semikenke Muzeyi wa Kazoza, d. 1982. II. Goyvaerts, D.L., 1946. III. Series. PL8003.A35 1985 496 85-26968 ISBN 90-272-2322-X (alk. paper) © Copyright 1985 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.

PREFACE

The idea of bringing out the present volume of essays arose during my stay in Zaire, a couple of months after Semikenke's death. Upon my arrival I was shocked to learn that after her return to Bukavu, Semikenke's widow had been blamed for her husband's death which led to rejection by the tribe and separation from her little daughter. Such harsh treatment may be incomprehensible to most westerners but is none the less in line with deep-rooted African tra­ dition in which curses, maledictions, evil spirits and the like still play a prominent role. Subsequent to her ostra­ cism, Semikenke's widow found herself in a desperate sit­ uation. In order to be able to provide some financial as­ sistance, I eventually decided to edit the present book the proceeds of which will go entirely to Mrs. Semikenke. I am extremely grateful to the many friends and col­ leagues who responded to my request with considerable en­ thusiasm and managed to send in their contributions well ahead of the proposed deadline. Special thanks are also due to John and Claire Benjamins for their generosity. Having prepared the present volume for publication, I cannot but be impressed by the quality of mosot of the papers. I am convinced the contributors have not only ren­ dered a service to Mrs. Semikenke but also to Linguistics, for it is difficult to see how, for example, future work in the field of theoretical phonology might progress with­ out taking into account several of the essays that appear in this book.

D.L.G. February 1983

OBITUARY SEMIKENKE MUZEYI WA KAZOZA

In the early morning of January 25, 1982, Mr. Semikenke Muzeyi wa Kazoza passed away after a short but tragic ill­ ness in a hospital in Leeds, England. His premature death at the age of 28 came as a shock to his friends in England, Belgium and Zaire. I first met Semikenke in 1976, when I was in Africa to teach a few linguistics courses at the I.S.P. (Univer­ sité Nationale du Zaire) in Bukavu. At that time, Semi as he liked to be called - was an undergraduate student in the Department of English. His work in both the fields of literature and language teaching was simply outstanding as most of his teachers would freely admit. Somehow, Semi and I were soon on the same wavelength: he showed a keen inter­ est in my teachings and I immediately noticed what a fine mind he had. Perhaps it was under my influence (sorry Luk De Vos and Bob Whitehead !) that Semi decided to embark upon a career in linguistics. The fact that he managed to assim­ ilate a lot of rather technical material in a fairly short time and in circumstances that were far from conducive to scientific research is just remarkable - few Europeans are in a position to realize what it is like to merely read a book or write a letter in one of the dwellings in Kadutu, Bukavu's shanty town. After graduation, Semi was not to disappear into the jungle to teach at some remote settlement where he would be completely cut off from the world of learning for the rest of his life but, instead, he remained in Bukavu as an assistant-teacher at the I.S.P. thanks to the support he received from Father Milani and Didy Naeyaert. Being an exemplary teacher, he obtained, in 1979, a grant from the British. Council to spend one year in London. He followed

AFRICAN LINGUISTICS

courses in TEFL (an obligatory condition set by the Brit­ ish Council and on which the grant depended) but kept up his interest in linguistics through conversations with Geoff Pullum who at the time was lecturing at University College. The following year Semi went on to the Universi­ ty of York (thank you Bob Le Page and all the other mem­ bers of staff at the Language Department !) where, under the guidance of John Kelly, he started work on the tonology of Kinyabwisha - a dialect of Kinyarwanda - and where in all likelihood he would have obtained his Ph.D. within the next year or so. It is terribly sad that he was not able to finish his Ph.D. on which he had been working so hard, so consci­ entiously and so cheerfully despite the financial and climatological difficulties and the many personal problems he almost constantly had to face. Those who knew him will always remember his irresistible charm, his enthusiasm and his intellectual brilliance. It is no exaggeration to state that his death will leave an indelible scar in the hearts of all the people who had the privilege of knowing him. All of us sincerely hope that his wife, Beá, and their two-year-old daughter Stella, will gather the neces­ sary strength to overcome this tragedy and to continue their lives which, it is to be feared, must surely come to them as unfair interludes on this earth.

Didier L. Goyvaerts University of Brussels (V.U.B.) University of Antwerp (U.I.A.) 30-1-82

TABLE OF CONTENTS

Preface Obituary Table of Contents On the ' compound' tone in Migaama D. Abdoullaye and J. Kelly Verbal Agreement as a noncyclic rule in Bantu E.G. Bokamba Akan vowel harmony: a nonlinear analysis G.N. Clements Tons et accents en Kinyarwanda R. Furere and A. Rialland On tone in Sukuma J. Goldsmith L'expression du passif en Maninka Cl. Grégoire The autosegmental and metrical nature of tone terracing C.T.J. Huang Globality in the Kinande tone system L.M. Hyman and N.Valinande Le mot chômage et ses implications socio-politiques dans la Voix du Congolais B. Kadima-Tshimanga A nonlinear account of the syllable in Luganda F. Katamba On the syllable structure of certain West African languages J. Kaye

v vii ix 1

9 55 99 167 189

209 239 261

267 285

TABLE OF CONTENTS On junction in two Kisiu nasal classes J. Kelly The phonology of state in Kabyle Berber M. Kenstowicz, . Bader and R. Benkeddache On the correspondene between linguistic tone and musical melody W.R. Leben An accentual approach to tone in Kimatuumbi D. Odden A lexical treatment of tone in Tiv D. Pulleyblank Swahili quasi-passives; the question of context J. Russell The status of lexical associations and the obligatory contour principle in the analysis of tone lan­ guages J.V. Singler

x

309 319 335

345 421 477 491

ON THE 'COMPOUND' TONE IN MIGAAMA D. Abdoullaye and J. Kelly (university of York)

Migaama, sometimes called Djonkor Abutelfan, is spo­ ken in the central part of Chad. Migaama, in a narrow sense, is one of four main dialects (viz. Migaama, Doga, Gamiya and Dambiya) which constitute a single larger unit itself called Migaama. The data that is used in this paper is taken from the speech of the first author above, who is a native speaker of the Migaama dialect. This dialect, and this alone, is spoken in the villages of Baro, Gurbiti, Sisi, Mabar, Bo, Golaw, Tabo, Tyororo and Padong in canton Abou Telfane. Other villages in the same canton and in neighbouring cantons speak partly the Migaama dialect, partly either other dialects of Migaama, other Chadic lan­ guages, such as Bidyo, or non-Chadic languages, such as Dajo. The present co-author was born in Baro, lived there up to the age of twelve and has visited regularly up to the present. In Newman and Ma's (1966) classification of Chadic languages, Migaama is listed in the Plateau-Sahel sub­ group, and, more specifically, in Group 9. More recently, Barreteau and Newman (1978) have produced a revised and more complex classification in which Migaama appears in sub-group (a) of Group 1 (Dangla) of sub-branch East  of the Eastern branch of Chadic. For all that a good deal of work has recently been done on Chadic, both qua individual languages and qua group, it is still true to say with Terry (1971) that: 'Our ignorance of the Chadic family is enormous. The needs are obvious as well. Extensive grammatical descriptions need to be written for well over one hundred languages.' Our paper makes a small contribution to this need by examining the nature of what has been called 'le ton mo­ dule' (Abakar et al 1975; Fédry 1977) or the 'fallton'

AFRICAN LINGUISTICS

2

(Jungraithmayr 1975) in Migaama. These terms we shall gloss here as 'compound tone'. Both Fédry and Jungraithmayr imply that this item in the tonal repertoire is marginal or secondary, Jungraithmayr by his representation of it on long vowels, written as two vowel letters, with a sequence of one high plus one low tone mark, Fédry by his remark, à propos four 'Mubi-Karbo' languages, that: 'Nos quatre langues se repartissent en deux types de système tonal: -système tonal à trois tons (sans modulé), Didiyo, Dangaléat de l'est -système tonal à deux tons (avec modulé) Dyongor, Dangaléat de l'ouest.' But neither sets out the evidence for looking at this to­ nal item in this way. We have no quarrel with their anal­ ysis, as our use of the term 'compound' shows: what we propose to do here is to present some evidence in support of such an interpretation in at least some cases. The Migaama compound tone, presented in pitch terms typically as a short high rise preceding a relatively steep fall from high to low, has a number of distribution­ al characteristics: it occurs on single syllables con­ taining either (a) a long vowel, or (b) a short vowel followed by (i) a long consonant, or (ii)a liquid (= nasal, lateral, tap or flap) in the same syllable. If, provisionally, we take a long consonant closing the syllable it stands after as well the one it precedes, we can re-state to say pound tone is limited to occurrence in long or short closed ones. Examples are:

to act as as opening that the com­ open syllables

(a) nî: (we), mî:ca (son, child), i:na (goat), â:ru (cane), bó:lâ: (axe), hàj:â:-j-à (fan), dâ:po (to crawl around) (b) i.

èrêj:o (to cry out happily), àmâr:o (to rec­

o n c i l e ) , pàΓâj:ó (to be i l l ) , dêb:ó (bow) ii. dambu (stone, mountain), sók:OQ (tomorrow), gûΓbu (stone, rock), jîmbè (bee, honey) It cannot occur, then, for instance, on a short vowel in an open syllable. Nor, interestingly, is a sequence of two or more compound tones ever found in words of one mor­ pheme .

'COMPOUND' TONE IN MIGAAMA

A kind of assimilatory coalescence is one important factor in the occurrence of the compound tone in Migaama. So far we have been able to identify two ways in which this takes place, both as the result of deletions. One of these is, broadly speaking, a synchronic process, the oth­ er a diachronic one. The former is the simpler and more readily attestable of the two, and will be dealt with first. Two variants of the verb 'to do something early in the morning1 are heard, namely, àrimo and armó, the second having no syllabic between the r and the m. Both forms are in use, but the second is much more common. In parallel cases àcficé alternates with àctcé (has/have fallen down) and kúcdisé with kûcdsé (has/have warmed up) . This suggests that, to have a compound tone in words that, like these, have i-deletion, the tone of a deleted short syllable must be low, and the preceding and following ones high. The i-deletion phenomenon happens frequently when the con­ sonant preceding the i is one of the three post-alveolar items d, r, ç. In all these cases, and those listed below, the trisyllabic forms are still used by some people, but these are getting fewer and fewer. Other examples: has/have nipped wild melon has/have knotted pestle has/have stung has/have contrac­ ted (of muscles) The diachronic deletion process is much more complex and occurs under different conditions. It is more complex as it is best treated as a number of (posited) stages in sequence, demonstrated here by our supposed derivation of the masculine singular form for 'small1, which is rô:t:à. The feminine form corresponding to this is rom:atà, and there is no possibility of interpreting the masculine in terms of deletion of the second syllable of this, as in the forms discussed in the previous section, since the second syllable here is high. The two forms for 'small' are to be seen, we believe, against the background of a set of extant masculine/feminine forms exemplified by:

AFRICAN LINGUISTICS

4

m. (i)

f, (ii)

meat-eating naughty inquisitive Both masculine forms are current, the first, longer, form, being now obsolescent. In the case of the word for 'small1, two forms paral­ lel to those above are posited. These have, we hypothe­ size, undergone various deletions, and other changes in­ volving shortening and lengthening. The stages we posit ~ are as follows: For the feminine *rómCátà *romà:Catà → róm:áta For the masculine *roma:Cata → *romà:Ctà * r ôra ta *rómàctà → ô:t:à The first two suggested stages of the masculine derivation match the actual forms of the masculine adjectives above. For 'small' only the final masculine and feminine forms are in existence. We suspect that the extended development of the 'small' forms has to do with the presence of two nasals in the stem, as nasals seem particularly suscepti­ ble to deletion in this language, giving rise to long vow­ els, as in the final stage of the masculine derivation above, and would, accordingly, suggest that the  of these forms was a nasal, though there is no real evidence for this. What particularly concerns us in these forms is the presence of the compound tone in the masculine as a result of the deletion of a short low-toned syllable after a pre­ ceding high-tone syllable. In the feminine form a long low-toned syllable is deleted, and its tone is not merged with that of preceding high-toned syllable. Some stages in the derivations are motivated by general conditions in Migaama, such as that a cluster cannot be preceded by a long vowel, and that deletion of one member of a cluster produces lengthening of the other. That this latter is so is shown by sets such as gurùstù ~ gúrùntù ~ gúrùt:ù for 'my money', all currently attested: 'money' is gúrus. These two pieces of evidence suggest to us that at least some cases of the compound tone result from histor­ ical deletion, just as others result from synchronic dele­ tion. It is commonplace, of course, that particular phono­ logical items in language can arise from dual, or several,

'COMPOUND1 TONE IN MIGAAMA

5

sources in this way. Other cases of the compound tone are to be found which arise in yet another way, and are to be treated differently. Migaama exemplifies the not unusual situation of a phonological item, here the 'compound1 tone, being assigned to loan words, as the nearest 'copy1 of the phonetics of the loan-items. No deletions are here in­ volved, of course. The great preponderance of Migaama loans are from Ar­ abic and French. Neither of these is a tone language, but in words borrowed from both, compound tones are found in the syllables which correspond to the accented syllables of the originals, provided that other conditions, already familiar, are fulfilled. So, in Arabic loans, the compound tone is found when the Arabic has : (a)

a long vowel in the accented syllable

(b)

a long consonant following the accented syllable

(c)

two consonants, unless the first is s or f, fol­ lowing the accented syllable

The following list shows the 'compound1 tone occurring un­ der some of these conditions, in a set of borrowings from Arabic :

Arabic ga'?aada sa'laam 'waadi tal'faan ha'waan 1 suuk na'kuur 'kaatum dza'niyya ha'lib

Migaama gà?â:dà sàlâ:m wâ:di:né tàlfâ:n hàwânné sûnné nàkûrré kâ:tùmné jàni:né hàli:p

bench greeting flood area idiot pus, dirt market shy ring fault fresh milk

(-né, -ré and -lé are three variants of a Migaama suffix frequently found in non-French loan-words.) Loan words from French show a similar treatment. As French accent is frequently realised as falling pitch, Migaami (Migaama speakers) use a compound tone in the cor­ responding syllable, with, if necessary, appropriate changes such as lengthening, to accommodate items to Migaama syllabic structures. Examples are:

6

AFRICAN LINGUISTICS

French

Migaama

mê:tèr master pê:r father lik:ô:l school bùrwê:t wheelbarrow sèrbê:t towel sàlbê:t chemise sàmî: s shirt ceinture sàntî:r belt valve bâlbé valve film fîlmè film sùsê:t sock chaussette When a French word ends in a short open syllable, the ac­ cent often corresponds to a Migaama high tone, as in certificat sèrtifiká certificate and some words have doublets, one in each system, such as pake or p :t In all these loan cases it will be noted that non-accented syllables in the source languages are matched by lows in Migaama, providing again an imitation of the phonetics of those languages; accordingly we might with some justifica­ tion say that loan-items have no tonal phonology in Migaa­ ma, though the phonetics of these items sounds the same as the phonetics of Migaama items having the same array of tones. This case might be described as a 'pseudo-compound1 tone, arising from a quite different source. For items of indigenous vocabulary we cannot demon­ strate at this time that all compound tones in Migaama are derived from the deletion of short low-toned vowels. Nom­ inais with suffixed first-person singular possessives are not immediately treatable in this way, and are not, in any case, directly comparable with any of the items considered above. Such sets as maître père 1'école brouette serviette

my carrying-pole my wild pigeon my hut are not immediately susceptible to treatment in these same terms. It is not unusual in the analysis of the tonal sys­ tems of Chadic languages to come across circumstances that call for the analysis of a falling or fall-like tone as a sequence of two tones, high and low, on one syllable. Schuh (1981) has such an item, for which he uses the term 'contour tone', and Kraft and Kraft (1973) tell us that 'occasionally a single Hausa syllable contains both tone-

'COMPOUND' TONE IN MIGAAMA

7

phonemes - the lower following the higher [ . . . ] We may r e ­ fer to t h i s combination of the two tone-phonemes as f a l l ­ ing tone but recognise i t i s not a t h i r d phoneme, but merely a complex of the two tone-phonemes f a l l i n g on the same s y l l a b l e . ' I t i s not perhaps s u r p r i s i n g then t o find t h a t in Migaama, t o o , such an a n a l y s i s i s i n d i c a t e d , in a t l e a s t some c a s e s .

REFERENCES

Abakar, A. e t a l . 1975 Kltipiyne Migaama, manuel de Dyongor (Abu Telfan) P .  . S . , Sarh. larisés,

pour

sco­

B a r r e t e a u , D. avec l a c o l l a b o r a t i o n de P. Newman 1978 Les langues Tchadiques in Inventaire des études linguistiques (ed. D. B a r r e t e a u ) , 291-330, Conseil I n t e r n a t i o n a l de l a langue f r a n ç a i s e , P a r i s . Fédry, J. 1977 Aperçu sur la phonologie et la tonologie de quatre langues du groupe 'Mubi-Karbo', Études

phonologiques

Paris. Jungraithmayr 1975 Der Imperfektiv-stamm in Migaama, Folia 100.

Orientalia

tchadiennes,

XVI, 85-

Kraft, C.H. and Kraft, M.G. 1973 Introductory Hausa, U.C.P., Berkeley and Los Angeles. Newman, P. and Ma, R. 1966 Comparative Chadic, Journal of African Languages 5, 218-251. Schuh, R.G. 1981 A dictionary of Ngizim, U.C.P., Berkeley and Los Angeles. Terry, R.R. 1971 Chadic, in Current Trends in Linguistics 7, 443-454, Mouton, The Hague.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU* Eyamba G. Bokamba (University of Illinois)

1.0

BACKGROUND 1.1 I n t r o d u c t i o n . Subject-verb agreement (hereafter VA) is perhaps the most universal characteristic of human languages; yet there has been very little attention given to it in the generative transformational literature. As a result, many of its formal properties continue to be mis­ understood. The few studies that exist on the subject, however, suggest that verbal agreement (VA) is a relative­ ly simple cyclic rule which is triggered by either logic­ al or cyclically derived subject NPs (Chomsky 1965; Gregersen 1967; Givón 1972; Vanek 1977). These conclusions, which appear to have been largely based on analyses of simple sentences, have been ques­ tioned in at least two recent studies (Morgan 197 2; Lawler 1977) . Morgan shows on the basis of English data that VA is a complex transformational rule whose structural de­ scription requires varying semantic specifications, and whose mode of application is anything but cyclic. In par­ ticular, Morgan (197 2) argues that there are three criti­ cal problems that must be dealt with in any analysis of VA: (1) the problem of ascertaining the level at which VA applies; (2) the problem of selecting which NP triggers VA; and (3) the problem of determining which properties of the selected NP will control VA. Morgan's contention is essentially that when these problems are dealt with, many of the assumptions associated with the currently ac­ cepted theory cannot be maintained. Lawler (1977) seems to have taken up this challenge in his study of VA in Achenese (an Indonesian language) where he attempts to show in effect that this transformation is always trig­ gered by the logical subject NP regardless of whether or not it has been demoted. Lawler's data, if correct, would

10

AFRICAN LINGUISTICS

contradict the relational grammar hypothesis that only terms can trigger VA (Perlmutter and Postal 1974). l This paper is a further attempt to characterize VA in natural languages. It examines data involving the interaction of VA and certain inversion phenomena, i.e., passivization, topicalization, relativization, and coor­ dination deletions in three Bantu languages: Swahili, Lingala, and Dzamba.2 The paper considers the three ques­ tions raised by Morgan (1972) in view of the data present­ ed here, and concludes that VA is a complex noncyclic rule, at least in these Bantu languages. This conclusion leads to some speculations on the properties of VA in oth­ er languages, and assesses the theoretical implications of the data discussed. The analysis begins with a sketch of some of the basic syntactic characteristics of Sim­ plexes in Bantu languages, with particular emphasis on Swahili, Lingala, and Dzamba. 1.2 Word Order in Bantu. The common word order in a simple sentence in Bantu languages has been assumed to be Subject-Verb-Object (SVO). This can be seen in the follow­ ing sentences: (1)

Dzamba

a.

b. c. (2)

Lingala

a.

b. c. (3)

(SVO)

(SOV) (VSO)

(Li.)

mw-aná á-tínd-áki ma-loba. child Ag-send-ed words (The child sent a message.) *mw-ăná ma-loba á-tínd-áki child words he-sent *á-tínd-áki mw-ăná ma-loba. he-sent child words

Swahili

a.

(Dz.)

omw-âna a-tom-aki mw-énzi the-child Ag-send-ed a message (The child sent a message.) *ómw-âna mw-énzi a-tom-aki the-child a message he-sent *a-tom-aki ómw-âna mw-enzi. he-sent the-child a message

(SVO)

(SOV) (VSO)

(Sw.)

m-toto a-li-peleka ma-neno. child Ag-Pst-send words (The child sent a message.)

(SVO)

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

11

b.

*m-toto ma-neno a-li-peleka. (SOV> child words Ag-pst-send  *a-li-peleka m-toto ma-neno. (VSO) Ag-pst-send child words If a sentence contains both a direct and an indirect object in these languages, the latter generally precedes the former in a neutral (i.e., non-emphatic) statement. A change in the ordering of these constituents leads to a change in focus and/or meaning, as evidenced in part in the (b) versus the (a) sentences: (4) Li.. a.

b.

c. (5)

mw-ăná á-pés-ákí mbwá biJléi. (SVIoIo) child he-give-ed dog food (The child gave the dog [some] food.) mw-ăná á-pés-ákí biléi mbwá. (SVDoIo) The child gave food, rather than something else, to the dog.) mw-ăná á-pés-ákí biléi na mbwá.(SVDoIo) (The child gave the food to the dog.)

Sw.

a.

m-toto a-li-m-pa mbwa cha-kula.(SVIoDo) (The child gave the dog some food.) b. m-toto a-lipa cha-kula mbwa. (SVDoIo) (The child gave food, rather than something else, to the dog.) The structure exemplified in the Lingala and Swahili sentences (4b) and (5b), respectively, is not possible in many other Bantu languages. What one generally finds in such languages when the direct object precedes the indi­ rect object is a structure like (4c) in which a preposi­ tion intervenes between the two objects. Kikongo and Shona fall into this category, while Dzamba and Kinyarwanda con­ stitute a third category; that is, one which disallows both (4b)and (4c). 1.3 Cliticization. What the facts in (4) and (5) seem to suggest is that the SVIoDo order is the unmarked word order for constructions involving two objects. This observation, however, holds only for sentences in which both object NPs are full nouns; if one of them is an anaph­ oric pronoun, there is no apparent syntactic or semantic difference in many of the languages where such construc­ tions are possible. This is particularly true of central

AFRICAN LINGUISTICS

12

Bantu languages (e.g., Dzamba, Likila, Lonkundo, Kikongo, and Tshiluba), as illustrated by the Dzamba and Lingala sentences (6) and (7), respectively: (6)

Dz.

a. b.



d.

ómw-âna á-éza-aki é-mva bieka. (SVIoDo) (The child gave the dog some food.) ómw-âna a-yé-éza-aki bíéka. the-child he-it-gave food (The child gave it some food.) ómw-âna a-bí-éza-aki é-mva. the-child he-them-gave the-dog (The child gave it [i.e., food] to the dog.) *ómw-âna a-bí-ézá-áki ná é-mva. the-child he-them-gave to the dog

(7) Li. a. mw-áná a-é-pés-ákí biléi. child he-it-gave food (The child gave it [i.e., the dog] some food.) b. mw-áná a-bí-pés-ákí mbwá. (The child gave it [food] to the dog.) c. *mw-ăná a-bí-pés-ákí na mbwá. Sentences (6d) and (7c) are ungrammatical, because the oc­ currence of the prepositions in constructions like these is not permissible. These facts seem to suggest that the grammar treats clitics as part of the morphology of the verb, rather than as reordered constituents, for other­ wise we would expect them to obey the same constraints as nouns. Support for this hypothesis comes from facts like the following where independent pronouns are subject to the same constraints as full nouns in those languages that permit such constructions: (8)

Dz.

ómw-âna á-éza-aki ba:-butú í-bieka. the-child he-gave the-guests the-food (The child gave the guests the food.) ómw-âna á-éza-aki bá:-bee í-bye. the-child he-gave def-them def-it (The child gave it [the food] to them.) *ómw-âna a-éza-aki í-bye bá:-bee.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

(9)

13

Li.

a. mw-ăná a-pés-ákí ba-paya bi-léi. child he-gave guests food (The child gave the guests the/some food.) b. mw-aná a-pés-ákí bangó byango. them it (food) (The child gave it to them.) c. *?mw-ăná a-pés-ákí byango (ná) bango. he-gave it (food) (to) them Swahili permits only sentences in which the indirect ob­ ject is a pronoun: (10) a. m-toto a-li-wa-pa wa-geni chakula child he-ed-them-give guests food (The child gave the guests some food.) b. m-toto a-li-wa-pa wao chakula. (The child gave them some food.) c. *m-toto a-li-wa-pa wageni cho he-gave-them guests it (food) In most Bantu languages, including the three under con­ sideration here, the preferred strategy for the construc­ tion exemplified in (8b) and (9b) is to cliticize one of the pronouns into the verb. In such cases the ordering of the two objects is unconstrained, i.e., either object can precede the other as in (6) and (7) above. Two other elements that need to be mentioned here are adjectives and adverbs. All adjectives (i.e., de­ scriptive, demonstrative, quantifier, possessive) in Ban­ tu languages generally follow the noun that they modify, as illustrated in (11) through (12): (11)

Li, mw-áná óyo mu-néné á-bét-ákí mo-paya child this Ag-big he-hit-ed guest wa ngai. Poss Me/mine (The big child hit my guest.) bá-ná ba-ye ba-néné bá-balé ba-bét-ákí children these Ag-big Ag-two they-hit-ed mo-paya wa ngáí guest Poss me/mine (These two big children hit my guest.)

14

AFRICAN LINGUISTICS

sw. a. m-toto huyu mkubwa a-li-piga mgeni Child this Ag-big he-ed-hit guest wa-ngu. Poss-mine (This big child hit my guest.) b. wa-toto ha-wa wa-kubwa wa-wili children these Ag-big Ag-two wa-li-piga mgeni wa-ngu. they-ed-hit guest Poss-mine (These two big children hit my guest.) Whenever a sentence contains an adverb, it generally oc­ curs at the end of the sentence in nonemphatic cases: (13) Li. a. mw-áná óyo mu-néné" á-bét-ákí mo-paya wa ngáí lelo. (This big child hit my guest today.) b. mw-áná óyo mo-néné á-bét-áki mo-paya wa (12);

ngaí na

mw-ete.

(This big child hit my guest with a stick.) In emphatic cases, depending on the language, the adverb (especially time) may occur either sentence-initially or anywhere before the verb (see Bokamba 1976b, 1976c). 2.0

GRAMMATICAL AGREEMENT IN BANTU 2.1 Adjectival and Verbal Agreement in the Simplex. Grammatical agreement is one of the most pervasive char­ acteristics of Bantu languages. It is traditional in Ban­ tu linguistics to distinguish two types of agreement pat­ terns: (1) nominal or adjectival agreement, and (2) ver­ bal agreement. Although this paper will not deal with ad­ jectival agreement per se, it will be instructive to briefly examine this type of pattern in the simple sen­ tence for comparative purposes. Adjectival agreement in most Bantu languages is largely based on the morphology of the noun, whereas ver­ bal agreement is not. In particular, nouns in Bantu lan­ guages are divided morphologically into noun classes in terms of singular/plural pairing of the noun prefixes and in terms of the distinct pattern of VA they exhibit. About 23 noun classes have been reconstructed for the en­ tire group, but in actual fact no Bantu language studied

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

15

so far has all these classes.3 There is a great deal of variation in this respect throughout the Bantu area, with a number of these languages having on the average between thirteen to sixteen noun classes. The languages under con­ sideration here fall within this range: Dzamba and Lingala have each twelve noun classes, and Swahili has sixteen4 (Bokamba 1976b; Ashton 1946). In all three languages adjectival agreement is characterizable as a morphological copying phenomenon that copies the prefix of the modified noun onto the variable adjective, as in (14)and (15): (14) Li . a. mw-äna mu-nέnέ (cl. 1) a big child big children b. bä-na ba-nέnέ (cl. 2) c. mu-nkanda mu-nέnέ (cl. 3) a big/large book d. mi-nkandá mi-nέnέ (cl. 4) big/large books e. li-toko li-nέnέ (cl. 5) a large mat large mats f. ma-toko ma-nέnέ (cl. 6) e-langá e-nέnέ (cl. 7) a large garden g. h. bi-langa bi-nέnέ (cl. 8) large gardens a big house i. n-dáko e-nέnέ (cl. 9) n-dako i-nénέ (cl. 10) big houses j. bo-lingo bo-nέnέ (cl. 14) great love (15)

Sw.

a tall child tall children a long ditch long ditches a long hoe long hoes a long knife long knives (a) tall house(s) a long board/ plank ku-andika ku-zuri (cl.12) good (hand)writing mahali ma-zuri (cl. 16) a good/nice place

m-toto m-refu (cl. 1) wa-toto wa-refu (cl. 2) m-fereji m-refu (cl. 3) mi-fereji mi-refu (cl. 4) jembe refu (cl. 5) ma-jembe ma-refu (cl. 6) g. ki-su ki-refu (cl. 7) h. vi-su vi-refu (cl. 8) n-defu (cl. 9/10) i. nyumba j- u-bao u-refu (cl. 11)

a. b. c. d. e. f.

. 1.

16

AFRICAN LINGUISTICS

The adjectival agreement prefixes in Dzamba are almost identical to those of Lingala (14). In all three languages this type of agreement is obligatory whenever the adjec­ tive is variable. Verbal agreement, in contrast, involves more complex features and subcategorization of the subject noun. Each finite verb in Bantu languages obligatorily agrees with its subject in person and class number by coping the rel­ evant semantic and morphological specifications of that noun. The contrast between adjectival and VA is exempli­ fied in the following Swahili sentences: (16) a. m-toto m-refu a-me-anguka. Ag-tall Ag-Perf-fall (The tall child fell/has fallen.) b. wa-toto wa-refu wa-anguka. (The tall children fell/have fallen.) c. m-nazi m-refu u-me-anguka. (The tall coconut tree fell/has fallen.) d. mi-nazi mi-refu i-me-anguka. (The tall coconut trees fell/have fallen.) e. jembe refu li-me-anguka. (The long hoe fell/has fallen.) f. ma-jembe ma-refu ya-me-anguka. (The long hoes fell/have fallen.) g. ki-su ki-refu ki-me-anguka. (The long knife fell/has fallen.) h. vi-su vi-refu vi-me-anguka. (The long knives fell/have fallen.) i. nyumba n-defu i-me-anguka. (The tall house fell/has fallen.) j. nyumba n-defu m-bili zi-me-anguka. (The two tall houses fell/have fallen.) Notice here that while the adjectival agreement markers can in almost all cases be traced morphologically to the noun prefix in each sentence, the VA prefixes cannot, ex­ cept for classes 2 and (7/8) which are illustrated in (16b) and (16g,h), respectively. The pattern of VA exhibited in (16) is further com­ plicated by the sub-categorization of the Swahili nouns into animate vs. inanimate. Nouns falling into the former category, regardless of their morphological noun class, will take the same agreement prefexes as human nouns, as in (17); whereas those falling in the latter group will

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

17

take the agreement prefixes that are appropriate to their classes, as in (16c-j): (17) a. ki-jana m-refu a-me-anguka. youth Ag-tall Ag-Perf-fall (The tall youngman fell/has fallen.) b. vi-Jana wa-refu wa-me-anguka. (The tall youngmen fell/have fallen.) c. ki-boko m-kubwa a-me-anguka. hippo. Ag-big Ag-Perf-fall (The big hippopotamus fell/has fallen.) d. vi-boko wa-kubwa wa-wili wa-me-anguka. (The two big hippopotamuses fell/have fallen.) e. mbwa m-kubwa a-me-anguka. (The big dog fell/has fallen.) f. mbwa wa-kubwa wa-wili wa-me-anguka. (The two big dogs fell/have fallen.) The expected adjectival and VA patterns for these nouns would be as in (18), but as evidenced here, these sen­ tences are ungrammatical; they would be acceptable only if the subject were considered to be toys.5 (18) a. *ki-jana ki-refu ki-me-anguka. b. *vi-Jana vi-refu vi-me-anguka. c. *ki-boko ki-kubwa ki-me-anguka. d. *mbwa m-kubwa i-me-anguka. e. *mbwa m-kubwa m-bili zi-me-anguka. This type of agreement pattern is not unique to Swahili: it occurs in many Bantu languages which, for convenience, may be termed the Swahili-type. Until the data in (17) and (18) were presented, it appeared that the only relevant features that need to be specified in the adjectival and VA rules are number, per­ son, and class membership. But as attested in (17) and (18), however, these specifications are not sufficient by themselves: the feature [animacy] is crucial in the deter­ mination of the appropriate agreement forms. These data lend support to Morgan's (1972) argument that it is not sufficient to determine which NP triggers VA: one must al­ so ascertain which feature(s) of the selected NP con­ trol (s) that agreement. Diverging from the Swahili-type languages are the Lingala or Dzamba-type which exhibit a largely alliter­ ative VA pattern. This is illustrated by the Lingala Sen­ tences below:

18

AFRICAN LINGUISTICS

(19)

a.

mw-áná mu-néné a-kwéy-akí. child Ag-big Ag-fall-ed (A/The big child fell.) b. bá-ná ba-néné ba-kwéy-ákí. (The big children fell.) c. mu-nkandá mu-néné mu-kwéy-ákí. (The big/large book fell.) d. mi-nkandá mi-néné mi-kwéy-ákí. (The big/large books fell.) e. li-toko li-néné li-kwéy-ákí. (The large mat fell.) f. ma-toko ma-néné ma-kwéy-ákí. (The large mats fell.) g. e-langa e-nene e-zík-akí. (The big farm burned.) h. bi-langá bi-néné bi-zík-ákí. (The large farms burned.) i. bo-lingo bo-nene bo-béb-í. (The great love/romance is wrecked.) Dzamba and other related languages exhibit the same agree­ ment pattern illustrated here, and do not subcategorize nouns into animate/inanimate as the Swahili-type languages do.6 These two agreement patterns largely characterize the type of VA found in the simple sentences in most Bantu languages when the subject noun is third person singular or plural. When it is first or second person singular or plural, the VA prefixes typically expected in each lan­ guage type are those given in (20) for the Swahili-type, and in (21) for the Dzamba/Lingala type: (20) Sw. a.

b. c. d. e. f.

(mimi) ni-li-kwenda nyumba-ni. I Ag-Pst-go home-loc (I went home.) (wewe) u-li-kwenda nyumba-ni. (you (sg.) went home.) ) a-li-kwenda nyumba-ni. (He/she went home.) (sisi) tu-li-kwenda nyumba-ni. (We went home.) (nyinyi) m-li-kwenda nyumba-ni. (You (pl.) went home.) (wao) wa-li-kwenda nyumba-ni. (They went home.)

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

(21)

19

Dz. a. (ó-ngá)

ná-zong-aki o lómbo loo me. Ag-return-ed to home today (I returned home today.) b. (ó-kao) o-zong-aki o lc5mbo loo me. (You (sg.) returned home today.) c. (ó-kei) a-zong-aki o lómbo loo me. (/She returned home today.) d. (ba-bánga) to-zong-aki o lómbo loo me. (We returned home today.) e. (bá-bεnyi) o-zong-aki o lómbo loo me. (You (pl.) returned home today.) f. (bá-bee) ba-zong-aki o lómbo loo me. (They returned home today.) What these facts, along with those in (14) through (18), show is that for each person and number there is a specif­ ic agreement or subject prefix. In some languages the pre­ fix may be zero for some of the cases (e.g., third person human in Chimwi:ni) or identical for two of the independ­ ent subject pronouns in the conjugation paradigm (e.g., 2nd person singular and plural, as in Dzamba sentences 21b,e, or 3rd person singular and plural, as in Makua). The over-all system, however, provides VA contrasts that permit unambiguous reference to the person, number, and noun class of the subject noun. It is for this reason that the occurrence of the independent subject pronouns in sen­ tences like (20) and (21) is optional; they can be omitted freely (after the application of VA) without affecting the grammaticality of the sentence. Insofar as the question of the selection of the NP which triggers VA and that of the semantic features which control it are concerned, it ap­ pears from the data presented thus far that the logical NP and the features person, number, animacy, and noun class control VA in these languages. 2.2 Inversions and Verbal Agreement. It is largely facts such as those presented above that have often led to the conclusion that VA in Bantu languages is defined on underlying subjects and that it operates in a left-toright manner (Gregersen 1967; Givón 1972). 7 As implied earlier, however, this position cannot be maintained in light of data from sentences involving inversion phenome­ na to which we now turn. We will show in this part of the paper that neither the standard theory or the relational

20

AFRICAN LINGUISTICS

grammar analysis of VA will account for all the facts in inverted constructions. 2.2.1

Passive

and

Verbal

Agreement.

Consider in this

regard, first, the passive construction. This rule, as in many other languages, involves inversion: the rule applies to active sentences such as those underlying the (a) and (c) below by promoting the logical object to the subject position and demoting the logical subject to an object po­ sition, marking it with the preposition na, as in (b) and (d): (22) Sw. a. m-toto a-li-vunja ki-ti. child Ag-Pst-break chair (The child broke the chair.) b. ki-ti ki-li-vunj-w-a na m-toto. chair Ag-Pst-break-Pass by child (The chair was broken by the child.) c. m-toto a-li-vunja vi-ti. (The child broke the chairs.) d. vi-ti vi-li-vunj-w-a na m-toto. (The chairs were broken by the child.) e. *ki-ti a-li-vunj-w-a na m-toto. f. *vi-ti a-li-vunj-w-a na m-toto. (23) Li. a. ba-paya ba-somb-akí li-toko lelo. guests Ag-buy-ed mat today (The guests bought a mat today.) b. li-toko li-sómb-ám-ákí na ba-paya lelo, mat Ag-buy-Pass-ed by guests today (The mat was bought by the guests today.) c. ba-paya ba-somb-aki ma-toko lelo. (The guest bought the mats today.) d. ma-toko ma-sómb-ám-ákí ná ba-paya lelo. (The mats were bought by the guests today.) e. *li-toko ba-sómb-ám-áki ná ba-paya lelo. f. *ma-toko ba-sómb-ám-ákí ná ba-paya lelo.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

21

Dz. a. ó-mw-azi a-kpét-ékí í-mw-eté wáabo. the-woman Ag-cut-ed the-tree here (The woman chopped down the tree here.) b. í-mw-eté mu-kpét-ém-ékí n'6-mw-âzi wáabo. the-tree Ag-cut-Pass-ed by-the-woman here (The tree was chopped down by the woman here.) c. ó-mw-azi a-kpét-ékí í-nzeté wáabo. (The woman chopped down the trees here.) d. í-nzeté i-kpét-ém-ékí n'6-mw-âzi wáabo. (The trees were chopped down by the woman here.) e. *í-mw-eté a-kpét-ém-ékí n'ó-mw-âzi wáabo. f. *í-nzeté a-kpét-ém-éki n'ó-mw-âzi wáabo. In addition to the inversion, several other properties ex­ emplified here need to be highlighted. First, it should be noted here that unlike in Achenese, VA in Bantu passives is triggered by the derived subject NP, as illustrated in the (b) and (d) sentences above. Second, the grammaticality of sentences like (e) and (f) demonstrates that pas­ sive precedes VA. Given that VA in Bantu languages char­ acteristically defines subjects, the conclusion to be drawn from these facts is that passive is a grammatical relation changing rule in these languages. This conclusion is consistent with the theory of relational grammar. Third, in each language there is a specific morpheme for marking the passive verbs, and these (morphemes) can only be interpreted as such. In the above sentences these mark­ ers are: -w- for Swahili, -am-/-emfor Lingala and Dzamba. Finally, the agent is preceded by the preposition na in all three languages. (24)

2.2.2 Topicalization and Verbal Agreement. Common passives in Bantu languages do not present any problem for either the standard theory or the relational grammar analysis of VA, but this does not hold for topicalized constructions in the Dzamba-type languages. To see this, consider, first, the Dzamba sentences in (25): (25)

a.

oPetélo a-kpét-el-eki bâzi nz-eté wáàbo. spc-Peter' ag-cut-App-ed women trees here (Peter chopped down (some) trees here for the women.)

22

AFRICAN LINGUISTICS

b.

ba-bâzi ba-kpét-el-eki ÓPetélo nz-eté waàbo. (The women chopped down (some) trees here for Peter.) (For the women, Peter chopped down some trees here.)  ba-bâzi ngó ba-kpét-el~eki ÓPetélo nz-eté waàbo. (As for the women, they cut (some) trees here for Peter.) ((As) For the women, Peter cut (some) trees here.) d. í-nz-eté i-kpét-el-eki ÓPetélo bâzi waàbo. (The trees, Peter cut for some women here.) e. í-nz-eté ngó i-kpét-el-eki ÓPetélo bâzi wáàbo. ((As for) The trees, Peter cut for some women here.) f. *ba-bâzi (ngó) ÓPetélo a-kpét-el-eki nz-eté wáàbo. g. *í-nz-eté (ngó) ÓPetélo a-kpét-él-ékí bâzi wáàbo. These constructions are paralleled in Lingala, Likila, and several other Bantu zone  languages. As discussed in detail in Bokamba (1976a,b), these sentences occur as re­ sponse to follow-up questions in which the speaker in­ quires about two or more objects but the respondent an­ swers the question only to one of the objects. What is particularly interesting about these sen­ tences is that they involve inversion and changes in gram­ matical relations, as exemplified by the VA pattern in (25b-e). Specifically, given the structure underlying sen­ tence (25a), topicalization in Dzamba and related lan­ guages may be viewed as involving the promotion of an ob­ ject to a subject position and the demotion of a subject to a position immediately after the verb, as in (25b-e). This rule can promote any object NP and prepositional phrases, including instrumentais and locatives; and may be formulated to optionally insert the topic-comment mark­ er ngo in Dzamba and nga in Lingala and Likila. After the application of this rule, VA will apply between the verb and its derived subject. The inversion is obligatory, as attested in part by the ill-formedness of (25f,g); and the order of application stated here appears to be the on-

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

23

ly principled and simplest alternative to account for the facts under consideration. It should be pointed out here, however, that the interaction of these rules sometimes leads to ambiguities just in case the topicalized NP is animate and capable of performing the action of the verb. Sentences (25b-c) are such cases. The reader may recall at this point that the inver­ sion just described is similar to that exhibited in the formation of passives, and may be wondering if topicalization and passivization in the Dzamba-type languages are not in fact related. The resemblance is actually superfi­ cial, because the two rules differ in significant respects syntactically and semantically. For instance, while pas­ sive obligatorily introduces a na-phrase and a passive suffix, topicalization simply inserts, optionally, a ngo/ n-phrase without affecting the morphology of the verb stem. There is, for instance, no passive morphology in the sentences (25), nor is there a possible passive reading from any of these sentences. Admittedly, the two rules have the same effect on VA, but this is also a superficial or coincidental resemblance, as will be shown shortly. Topicalized sentences such as (25) do not present any problem to the relational grammar analysis of VA insofar as the prediction of the correct agreement forms is con­ cerned, but they raise serious questions regarding the as­ sumption, espoused by both this and the standard theory, that VA is a cyclic rule. In particular, if we assume with Ross (1967b) that topicalization is a transformation­ al rule that involves copying, Chomsky-adjunction, and de­ letion as stated in (26), rather than being a base genera­ ted structure as proposed in Chomsky (1977),10 the deriva­ tion of sentence (25b), e.g., would proceed as in (27): (26)

Topicalization SD:

X 1

SC: 2 # [ 1 (27)

a. b. c. d. e.

(optional) NP 2

Y 3

0

3]

→

( o P e t e l o l o - k p e t - e l - a + Impf b a z i nz-ete waabo) S2(ba-bazi #S 1 [oPetelo lo-kpet-el-a + Impf 0 nz-ete waabo]) S2(ba-bazi #S 1 [oPetelo a-kpet-el-a + Impf 0 nz-ete waabo]) S2(ba-bazi # S1[ 0 a-kpet-el-a + Impf oPetelo 0 nz-ete waabo]) *Ba-bazi a-kpet-eki oPetelo nzete waabo.

24

AFRICAN LINGUISTICS

In other words, assuming that (27a) underlies the senten­ ces (25) above, topicalization (26) will apply to it to produce (27b). This structure in turn will undergo Sub­ ject-Verb Agreement to yield (27c) where the verb agrees with the logical subject, oPetelo. At this point (27c) will have two NPs occurring before the verb, with one of them functioning syntactically as the subject (i.e., oPe­ telo) and the other (i.e., ba-bazi) having no clearly identifiable function either semantically or syntactically. Since structures of this type are disallowed, the logical subject must be postposed to a position after the finite verb via what we have termed Subject-Verb Inversion rule (cf. Bokamba (1976b). Subject Verb Inversion will yield (27d). The sentence resulting from the application of these rules in the order described here is the ungrammatical (27d). This sentence is ungrammatical, because VA is controlled by the logical, rather than the derived sub­ ject (ba-bazi). There are two possible solutions that immediately come to mind here in attempting to characterize the deri­ vation of the grammatical sentence (25b). First, if we wish to maintain the hypothesis that VA is a cyclic rule, the rule of Subject-Verb Agreement will have to reapply to the same verb after the application of Subject-Verb Inversion, i.e., on cycle S2,-in order for the derived subject to be able to control verbal agreement. The appli­ cation of VA on S 2 would be consistent with the notion of cyclicity were it "not due to the fact the rule reanalyzes the same verb from a cycle (S ) that has already been by­ passed and affected by it. The implication here is that the second application of VA nullifies or deletes the first one. Now, unless one assumes that topicalization is also a cyclic rule and that it has applicational preced­ ence over VA, this mode of application clearly constitutes an unnecessary violation and distortion of the notion of the cycle as generally understood in the standard theory (Chomsky 1965; Pullum 1979). Further, topicalization has generally been taken as a post-cycle rule (cf., e.g., Ross 1967b, Morgan and Green 1972). Clearly, the analysis sug­ gested in (27) cannot be entertained as a possible solu­ tion. An alternative analysis would be to assume that both VA and topicalization are postcyclic rules that are order­ ed with respect to each other and with respect to the third rule in the following manner:

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

(28)

Rule

25

Interaction

a. Topicalization b. Subject-Verb Inversion  Subject-Verb Agreement Under this analysis the derivation of sentence (25b) would proceed as follows: (29)

a. (oPetelo lo-kpέt-έl-a + Impf bazi nz-ete waabo) b. S2(ba-bazi #S 1 [oPetelo lo-kpέt-εl-a + Impf 0 nz-ete waabo]) c. S2(ba-bazi # S1[Ø lo-kpέt-εl-a + Impf oPetelo 0 nz-ete waabo]) d. S2(ba-bazi # S1[ ba-kpέt-εl-εki oPetelo 0 nz-ete waabo]) e. Ba-bazi ba-kpέt-εl-εki oPetelo nz-ete waabo.

That is, in this case Topicalization will apply to the object bazi "women" in (28a) to yield (28b) which will, in turn, meet the structural description of Subject-Verb Inversion. This structure will undergo the rule and pro­ duce (28c) to which VA will apply to yield (28d). After the removal of the brackets, the resulting sentence will be the grammatical (28e). This analysis will consistently predict the correct output, and will not encounter the difficulties discussed in connection with the first ap­ proach. Since the data considered thus far involve simple sentences, it is difficult to ascertain whether both Top­ icalization and VA are cyclic rules. It is entirely pos­ sible that Topicalization is a cyclic rule, and VA a postcyclic rule, or that both are cyclic in Dzamba and close­ ly related languages where parallel facts exist. 12 The latter possibility is suggested by the occurrence of data such as (30c-e) where the indirect object mw-έnzi "mes­ sage" governs verbal agreement in the lower and higher clause: (30)

Dzamba

a.

ózáki lo-wo-o + Impf kííbo oPcsc lo-ézaSpf-J to-say that Pcsc to give a + Impf bá-bana mw-έnzi. the-children a message

b.

ózáki a-wc-cki kííbo ÓPcsc a-éza-aki Spf.-Jack Ag-say-ed that Poso Ag-give-ed

26

AFRICAN LINGUISTICS

bá-bana mw-έnzi. the-child message (Jack said that P s gave the children a message.) ózáki a-wo-oki kííbo ímw-énzi(ngo)mwéza-aki ÓPoso bá-bana. (Jack said that the message Poso gave [to] the children.) ímw-έnzi mu-wo-oki (ngo) ózáki kííbo the-mess. Ag-say-ed Jack that ÓPoso a-ézaaki bá-bana. P. Ag-give-ed (The message Jack said that Poso gave to the children.) *(The message said to Jack that Poso gave to the children.) ?ímw-έnzi (ngo) mu-wo-cki ozáki kííbo mw-éza-aki oP. bá-bana. (The message, Jack said that Poso gave it to the children.) *ímw-έnzi a-wo-oki ózáki kííbo óPoso a-éza-aki bá-bana. a-ezaaki oPoso ba-bana. That is, given a complement clause structure like (30a) and the assumption that the rules involved are ordered as in (28), if Topicalization (optional) is not applied, Sub­ ject-Verb Inversion will not apply either as its struc­ tural description will not be met. Subject-Verb Agreement will, however, apply obligatorily and will yield (30b) where the logical subject of each clause controls verbal agreement. Now, if Topicalization is applied, the resulting sen­ tence may be either (30c), (30d), or (30e). In particular, if the rule applies in the lower cycle, the indirect ob­ ject mw-έnzi will be moved to the beginning of that clause and Subject-Verb Inversion will apply subsequently to produce an intermediate structure on which VA will oper­ ate. At the time that VA applies the topicalized NP will be adjacent to the verb in the lower clause. If no other rule intervenes in the higher clause, the application of VA to this structure will yield sentence (30c) where the derived subject governs agreement. The application of VA in this derivation can be viewed as either cyclic or postcyclic: it is indeterminate. If we assume that the former is the case, since there is no argument to the contrary,

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

27

then all three rules can apply again on the second cycle (S2) on the output of the first application to move ímw¿nzi to the beginning of the sentence as in (30e). This sentence, as will be noticed, is questionable: some speak­ ers accept it, but others do not. The main problem appears to be that there are too many foci. The other possibility would be to apply Topicalization only on the second cycle, and then VA on both cycles after Subject-Verb Inversion has had a chance to apply on S 2 . These applications will produce sentence (30d) where VÄ is governed by the logical subject in the lower clause, and by the derived subject in the higher clause. Sentences (30f-g) are ill-formed be­ cause VA is controlled by the logical subjects. The occurrence of data like (30) raise some fundamen­ tal questions about the postcyclicity of Topicalization and the actual formulation of the rule itself. These prob­ lems may be handled in at least three different ways : first, one might relax the requirements imposed on the cy­ clic rules so as to allow them to refer to materials in a lower cycle that has already been bypassed in a deriva­ tion. This proposal, while logically attractive, appears to have very little support from the facts under consider­ ation here and those from other languages. The relaxation of the requirements for cyclic rules would be only appli­ cable to VA in the Dzamba-type languages, since no other language known to us has an agreement rule that behaves in the manner described above. Hence this is not a viable alternative. Second, one might wish to reject Ross's (1967) analysis of Topicalization and reformulate it as a movement rule involving sister-adjunction. Such a pro­ posal would not only obviate some of the objections made above, especially in connection with the sentences in (25), but would also distinguish this rule further from Left Dislocation. In particular, Left Dislocation has been analyzed as involving copying with chomsky-adjunction and pronominalization (Ross 1967). If we assume this analysis, we can account nicely for facts such as the Dzamba and Swahili sentences in (31) and (32). 3 (31)

Dzamba

a.

b.

óPetelo a-kpέt-εl-εki bâzi nz-eté wáàbo. (Peter chopped/cut down [some] trees for the women) bá-bâzi, ÓPetélo a-bá-kpέt-έl-έki nz-eté waàbo. the-women, P. Ag-them-cut-App-ed trees here

28

AFRICAN LINGUISTICS

([As for] The women, Peter chopped down [some] trees here for them.) ínz-eté, ÓPetélo a-í-kpέt-εl-εki bâzi waàbo. the-trees, P. Ag-them-cut-App-ed ([As for] The trees, Peter chopped them down here for the women.)

c.

(32)

Swahili

a.

daktari a-li-leta ki-ti nyumba-ni. doctor Ag-Pst-bring chair house-loc. (The doctor brought a/the chair home.) b. ki-ti, daktari a-li-ici-leta nyumba-ni. chair doctor Ag-Pst-it-bring ([As for] The chair, the doctor brought it home.) c. nyumba-ni, daktari a-li-pa-leta ki-ti. house-loc doctor Ag-Pst-it bring (To a specific place in the house, the doc­ tor brought a/the chair.) 14 Assuming that there is a pronominalization rule and that its domain involves two cycles (see Wasow 19 79 for a dis­ cussion of some of the problems with this rule), the leftdislocated sentences above where the italicized clitics are correferential to the left-dislocated NPs would be possible only if Left-Dislocation is viewed as involving Chomsky adjunction. Chomsky-adjunction appears to be nec­ essary in this case but not in topicalization. The accep­ tance of this reformulation of topicalization, it appears to us, would remove the undesirable cycle to which VA ap­ plied for the second time to the same verb in the deriva­ tion of sentences (25b-e). A third alternative would be to reject the hypothesis that VA is a cyclic rule, at least in the Dzamba-type lan­ guages. To do this one would have to show that there are derivations in which VA applies either precyclically or postcyclically. Currently the existence of precyclic rules in syntax has yet to be established convincingly (Pullum 1979), and we are unaware of the occurrence of such rules in Bantu languages. There is no need to dwell on this al­ ternative. There are, however, several postcyclic rules that interact with VA; these include relative clause for­ mation, gapping, and conjunction reduction. Of these, the interaction of the latter with VA appears to be the strong­ est and most clear case for proving the non cyclicity of the latter (cf. Bokamba 1981). We turn our attention to an examination of these interactions.

29

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

2.2,3

Relativization

and

Verbal

Agreement.

There

are several strategies of relativization in Bantu lan­ guages which have been discussed in various recent studies (cf. Givôn 1972b, 1979; Bokamba 1976a, b, 1981; Keach 1980); but many of them are not immediately relevant to the present study. We shall, therefore, restrict ourselves to those that bear immediately on the discussion here. All three languages under consideration here have such strat­ egies, but the extent to which VA can be demonstrated un­ ambiguously to interact with relativization varies from one language to another. The data from Swahili are ambig­ uous, those from Lingala less so, and those from Dzamba unambiguous. Let us consider, first, the facts from Swa­ hili. Swahili has three types/strategies of relativization: (1) the General Relative; (2) the Amba Relative; and (3) the Tensed Relative (Wilson 1970, Givón 1979, Keach 1980, Bo­ kamba 1981). The first and second involve the suffixation of the relative marker (Rel) to a class of verbs and the element amba, respectively. The third type requires the prefixation of the relative marker to a tensed verb. As in other languages, Swahili distinguishes subject relativ­ ization from object relativization. The former construction can be viewed as involving two rules: Relative Pronominalization, and Deletion, as may be seen in the following: Swahili

(33)

a.

m-toto m-refu a-li-ye-anguka hapa child Ag-tall Ag-Pst-Rel-fall here a-me-vunja m-kono wake. Ag-Perf-break arm Ag-his (The tall child who fell [down] here broke his/her arm.)

b.

m-toto

m-refu

amba-ye a-li-anguka hapa say-Rel Ag-Pst-fall here m-kono wa-ke.

a-me-vunja Ag-Perf-break (The tall child who fell [down] here broke his/her arm.) There is no inversion in subject relativization in either the tensed or amba strategy. Further, VA is unaffected in the sense that it is triggered by the logical subject in each clause. Also, as the glosses indicate, there is no difference in meaning between the two sentences, even though different strategies are involved.

AFRICAN LINGUISTICS

30

Object relativization, in contrast, involves a third rule: Subject-Verb Inversion (SVI). In particular, when an object or any other VP dominated noun phrase is rela­ tivized, the tensed relative strategy requires SVI, where­ as the amha-relative does not: its occurrence is optional. This fact is illustrated in: Swahili

(34)

a.

Juma a-li-nunua vi-ti amba-vyo Maryamu Juma Ag-Pst-buy chairs say-Rel M a-li-taka. Ag-Pst-want. (Juma bought the chairs which Maryamu wanted.) b. Juma a-li-nunua vi-ti amba-vyo a-lisay-Rel Ag-Pst(vi-) taka Maryamu. them-want. (Juma bought the chairs which Maryamu wanted.) c. Juma a-li-nunua vi-ti a-li-vyo-fvi-) Ag-Pst-Rel-themtaka Maryamu. want (Juma bought the chairs which Maryamu wanted.) d. *Juma a-li-nunua vi-ti Maryamu a-livyo-(vi-) taka. Relativization involving instrumentais and locatives in this language also follow this pattern. The occurrence of SVI in (34b), in contrast to its non­ occurrence in (34a), may be regarded as a stylistic alter­ nation; but the inversion in (34c) cannot be so viewed, because the application of the rule here is obligatory. This fact is attested in the ungrammaticality of (34d). The occurrence of SVI in such cases appears to be moti­ vated by the surface constraint against the occurrence of double "subjects" before a verb as stated previously.15 It will be noted further that the occurrence of SVI does not change the behavior of VA: the rule is still triggered by the demoted subject, Maryamu, which is en chomage in (34c) . There are two possible ways that sentences like these can be accounted for: First, one might argue that (34c),

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

31

for example, is derived by applying VA cyclically and then relativization (including subject-verb inversion) postcyclically. This order of application would explain why VA is still controlled by the underlying subject, Maryamu. Second, one might maintain that (34c) is derived by ap­ plying VA postcyclically after relativization. In this case the rule of VA would scan the entire string and copy the relevant feature specifications of the subject NP on­ to the appropriate verb irrespective of the location of this subject relative to the verb. That is, the rule will copy either from the left or the right depending on the location of the subject NP at the time of the application. Both of these alternatives appear to be equally plausible, although the second might create a slight complication in the formulation of the rule; but there are no additional Swahili data known to us that would force a choice between these analyses. Similar, but unambiguous data exist in Lingala and Dzamba to force such a choice, however; and we turn now to these languages, first, Lingala. This language, like Swahili, has two types of relative clause constructions: those involving a free relative pronoun, and those with a (verb) bound relative pronoun. These are illustrated in (35) and (36): (35) a. mw-áná óyo a-sómb-ákí 1i-tckc awâ a-kέi. child Dem. Ag-buy-ed mat here Ag-left (The child who bought the mat here left/has gone.) b . mw-áná a - s ó m b - á k í l i - t c k c áwâ a k έ i . (The child [who] bought the mat here left ? has gone.) c. mw-áná -sómb-ákí li-toko áwa a-kέi. Rel-buy-ed The child who bought the mat here left/has gone.) (36)

a.

mo-paya

á-sómb-ákí

li-toko li-ye li-záláki áwâ. guest Ag-buy-ed mat Dem Ag-be-ed (The guest bought the mat which was here.) b. mo-paya á-sómb-ákí li-toko li-zál-áki áwa. (The guest bought the mat which was here.) c. mo-paya á-sómb-ákí li-toko lí-zál-ákí áwâ. (The guest bought the mat which was here.) These sentences illustrate subject relativization, and

32

AFRICAN LINGUISTICS

there is, as stated previously, no subject-verb inversion. Verbal agreement, therefore, remains largely unaffected. These constructions differ in at least two important re­ spects from the corresponding Swahili sentences: First, subject as well as object relative clauses in Lingala in­ volve the use of demonstrative pronouns as a common strat­ egy. Second, these demonstrative pronouns may be freely omitted without affecting either the morphology of the verb, as seen in the (b) sentences, or the meaning of the sentence. When there is no demonstrative pronoun underlyingly, the relative pronoun is prefixed onto the verb as the first morpheme and thereby replaces the agreement marker, as in the (c) sentences above. In most cases, ex­ cept for singular nouns of the human class, this marker is similar to the agreement prefix but bears a high tone, as in (36c) . Object relativization, as in the Swahili case, in­ volves subject-verb inversion obligatory whenever the rel­ ative pronoun is bound, but optionally when it is not: (37) a. mo-paya a-sómb-ákí li-toko li-ye pcso á-tóng-ákí lelo. Dem. P. Ag-weave-ed (The guest bought the mat which Pcsc weaved today.) b.

mo-paya á-sómb-ákí li-toko li-ye litóng-ákí Poso lelo. (The guest bought the mat which Poso weaved today.) c. mo-paya á-sómb-ákí li-toko p5so á-tongáki lεlc. (The guest bought the mat Poso weaved to­ day.) d. mo-paya á-sómb-ákí li-toko lí-tóng-áki P 5 S O lεl5. Rel-weave-ed P. (The guest bought the mat which Poso weaved today.) e. *mo-paya á-sómb-ákí li-toko (li-ye) Poso lí-tóng-áki... f. *mopaya á-sómb-ákí li-toko li-ye á-tóngáki Pcsc... These sentences illustrate several characteristics of ob­ ject relativization. First, they show that VA in the em­ bedded clause remains unaffected whenever the logical subject precedes its verb (37a, c ) . Second, sentences

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

33

(37b, d) show that subject-verb inversion in the embedded clause will occur whether or not the independent "relative pronoun" surfaces; and that once inversion has occurred, the underlying subject loses its ability to trigger VA (37b, d ) . Instead, the derived subject (i.e., the logical object) controls VA and thereby changes grammatical rela­ tions. This conclusion is supported by the ungrammatical­ l y of sentences like (37f) which is identical to (37d), except that it has the incorrect agreement prefix. (37e) is ill-formed because the embedded subject has not been postposed. Finally, the ungrammaticality of (37e, f) sug­ gests strongly that VA in this language is based on adja­ cency; that is, the rule is defined on the NP which is ad­ jacent to.the verb at the point in the derivation when it applies. The generalization to be made here is that VA applies after relativization, a postcyclic rule. This be­ ing the case, VA itself, at least in this language and related ones (see Dzamba below, e.g.), must be a postcy­ clic rule that is not sensitive to underlying grammatical relations. The facts of object relativization in both Swahili and Lingala exhibit obligatory and optional subject-verb inversion in the embedded clauses. Lingala shows further that the relative pronoun may be freely deleted without affecting either the agreement process or the grammaticality of the sentence. While the Lingala data are much clearer than the Swahili concerning the postcyclicity of VA, and this is true even if we assume topicalization to be involved in such derivations, the optional occurrence of subject-verb inversion and the free deletion of the relative pronouns may have obscured the ordering of the main rules: relativization and VA. It would be highly in­ structive and interesting at this point to find data that exhibit only obligatory subject-verb inversion. Such data would constitute evidence of the strongest kind in that subject-verb inversion would be viewed as a necessary subrule of relativization, and the relationship between this rule and VA would become clearer. Dzamba, to which we now turn, is such a language. We have shown elsewhere that relative pronouns in this lan­ guage are all bound to the verb, and that object but not subject relativization entails obligatory subject-verb inversion in the embedded clause (Bokamba 1976a, b, 1981). Subject relative clauses in Dzamba have the surface form illustrated in (38) and object relative clauses have that exemplified in (39):

34

AFRICAN LINGUISTICS

(38)

a.

b.

c. d.

(39)

a. b.

o-mama a-eza-aki i-zi-bata lc-so. the-mother Ag-give-ed the-duck rice (The mother gave the duck some rice.) ó-mama á-éza-aki í-zi-bata í-zi-yak-aki the duck Rel-Ag-come-ed waàbo lo-so. (The mother gave the duck which came here some rice.) i-zi-bata zi-komel-aki i-lc-so. (The duck pecked on the rice.) í-zi-bata i-zi-yak-aki waàbo zi-komelRel-Ag-come-ed Ag-peck-ed akx i-lo-so. on (The duck which came here pecked on the rice.) ó-mw-âna á-omb-aki i-lo-so loo me. (The child bought the rice today.) í-zí-bata i-zi-éza-aki ó-mama í-lo-so Rel-Ag-give-ed the-mother zi-kέi.

(The duck to which [the] mother gave the rice left/has gone.) c.

*í-zi-bata ó-mama a-éza-aki i-lo-so zikei. d. *í-zi-bata á-éza-aki ó-mama i-lo-so zi-kei. e. *í-zi-bata ó-mama í-zi-eza-aki i-lo-so zi-kέi. f. í-zi-bata zi-komel-aki i-loso í-lú-ombthe-rice Rel-Agaki omw-ana loo me. buy-ed the-child today (The duck pecked on the rice which the child bought today.) g. *i-zi-bata zi-komel-aki i-lo-so ómw-ana á-omb-aki loo me. h. *i-zi-bata zi-komel-aki i-lo-so a-ombaki ómw-âna loo me. i. *i-zi-bata zi-komel-aki í-lo-so ómw-ana í-lu-omb-aki loo me.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

35

These facts illustrate several properties of verbal agree­ ment in simple and relative clauses. First, the VA pat­ tern exhibited in the simple sentences (38a, c) and (39a) , and tfyat in the subject relative clauses (38b, d) appears to be consistent with the hypothesis that VA is a cyclic rule: the subject prefixes in .these sentences are con­ trolled by the logical subject NPs. Second, in the object relative clauses (39b, f ) , however, VA in the embedded clause "is controlled by the derived subject NPs: i - z i - b a t a and i-lc-sc. These NPs became the superficial subjects of their relative clauses only after relativization, which, as we have seen, involves subject-verb inversion, had ap­ plied. If VA had applied cyclically, as it appeared to have done in the derivation of sentences (38), the ex­ pected outputs would have been sentences (39c-h); but as evidenced here these are ungrammatical. Third, as in the cases of Swahili and Lingala, object relativization in Dzamba requires subject-verb inversion in the embedded clause; this property accounts for the ungrammaticality of (39e, i) which are otherwise well-formed except that the logical subjects have not been postposed. (39c) is partly ill-formed for the same reason. The generalization that emerges from these facts is that VA applies after relativization, a postcyclic rule, and is itself a postcyclic rule. Now, it is entirely possible that Dzamba-type lan­ guages (including Likila and Lingala) can only relativize on subjects in the Keenan-Comrie (1977) accessibility hi­ erarchy, and that relativization of other terms requires a strategy of topicalization.17 Even if this were the case, VA would still have to be considered a postcyclic rule as long as the hypothesis that relativization is a postcyclic rule is maintained. To see this, it is suffi­ cient to consider just one derivation under the cyclic analysis of VA. Consider, for example, the case of sen­ tence (39b) given in (40) below: (40) a. S2(izi-bata #(s1 -mama lo-éza-a + Impf ízi-bata ílc-sc) sl ≠ lo-kέε+ IP)S2 b.

S2(ízi-bata #(s1 ízi-bata -mama lo-éza-a + Impf 0 ílc-sc) sl lo-kée + IP)

c.

S2(ízi-bata #(s1 ízi-bata 0 lo-éza-a + Impf -mama 0 ílcsc) sl lo-kέε + IP)

d.

S2(ízi-bata #(s1ízi-bata Ø zí-éza-aki 6mama 0 ílc-sc) s1 # lo-kée + IP)

36

AFRICAN LINGUISTICS

e.

S2(ízi-bata #(s1ízi-bata Ø zí-éza-aki mama 0 í l c - s c ) S 1 # z í - k έ - í )

f.

S 2 ( í z i - b a t a # ( S 1 í - 0 z í - é z a - a k i 6-mama 0 í l c - s o ) s l # zí-kέ-í)

g.

ó-

* I z i - b a t a í - z í - é z a - a k i 6-mama í l c - s c z í k£-í. Assuming that both Topicalization and VA are cyclic rules and that Relative Clause Formation is a postcyclic rule, the interaction of these rules as illustrated in (40) will produce ungrammatical sentences. Specifically, if we assume a Keenan-Comrie analysis for Dzamba and re­ lated languages, the relativization of the indirect ob­ ject ízi-bata "the duck" will require first of all the preposing of this object to the beginning of the embedded clause via topicalization. This rule will convert, in oth­ er words, (40a) to (40b). Since this structure meets the structural description of SVI, it will apply on S 2 to pro­ duce (40c) which will, in turn, be scanned for the appli­ cation of VA. The application of this rule will produce (4Od) where verbal agreement is controlled by the derived subject, izi-bata, in the first cycle ( S 1 . Since no other rule can apply on this cycle, the second cycle will be scanned for possible application. At this point the struc­ tural description of both VA and relativization is met. The former will apply first to (40d) to yield (40e) where VA is governed by the logical subject, izi-bata, and then the latter rule will apply to this output to yield (40f) . Since the identical NP has already been fronted by Topi­ calization, the appropriate subrule of relativization to apply here will be Relative Prenominalization. After the removal of the brackets the resulting sentence will be the ungrammatical (40g). Sentence (40g) is ungrammatical because the subject prefix that occurs on the embedded verb bears a low tone, indicating that it is triggered by the derived topicalized subject. Verbal agreement controlled by topicalized sub­ jects always bear low tone, just like all other subjects, including passive (cf. Bokamba 1981) . Verbal agreement triggered by relative subjects, in contrast, always bears a high tone in this language and many other Bantu lan­ guages (cf. Bokamba 1981). If we assume, as is tradition­ ally done in generative transformational grammar, that one of the operations of a T-rule involves the scanning and identification/marking of elements that are to undergo a particular rule, the ungrammaticality of (4 0g) can be ex-

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

37

plained in terms of the nonmarking of the topicalized NP i zi-bata as a constituent to undergo the rule of Relative Clause Formation. This did not occur until the second cy­ cle at the point at which this rule applied. It appears to be clear from the above derivation that VA cannot be a cyclic rule in this language and other Dzamba-type languages where parallel facts exist (cf. Bokamba 1981) . If we assume that VA is a postcyclic rule and Topicalization is a cyclic rule that involves sister adjunc­ tion, rather than Chomsky-adjunetion, we can correctly predict the derivation of sentence (39b). That is, the interaction of the rules under this analysis will yield the correct results. This can be seen from the following derivation: (41)

a.

S2(izi-bata #(S1 o-mama lo-eza-a + Impf izi-bata ilc-sc)S1 # lo-kέε + 1 P ) S 2

b.

S2(izi-bata #(slizi-bata o-mama lo-éza-a + Impf 0 ílc-sc) sl # lo-kέε + IP)

c.

S2(izi-bata #(S1 izi-bata 0 lo-éza-a + Impf ó-mama 0 í l c - s c ) # lo-kée + IP)

d.

S2(ízi-bata #(S1 ízi-bata í-zí-éza-aki ómama 0 ílc-sc)S1 # zi-ké-í)

e.

S2(ízi-bata # ( . 0 0 í-zí-éza-aki δ-mama 0 ílc-sc)S1 # zi-kέ-í)

f.

ízi-bata í-zí-éza-aki 6-mama ílc-so zike-i. (The duck to which mother gave the rice has left.) This output will then undergo Relative Clause Formation on S 2 whereby the second occurrence of izi-bata will be marked (+Rel Pron). VA will apply to this structure, es­ sentially (41c), to yield (4Id) where the embedded verb contains both a relative marker and a relative subject prefix, and the matrix verb bears a regular subject pre­ fix controlled by the logical subject.18 Subsequently the second occurrence of i z i - b a t a will be deleted under identi­ ty, as in (41e) and the resulting sentence will be the gram­ matical (4lf). We have assumed in this discussion that the deriva­ tion of relative clauses in Dzamba involves four major rules: (1) Topicalization, (2) Subject-Verb Inversion, (3)

AFRICAN LINGUISTICS

38

Relative Clause Formation, and (4) Subject-Verb Agreement. These rules, in view of our analysis (41), are intrinsic­ ally ordered. The existence of the first rule is suggested by the preposing of the NP undergoing relativization, and may be theoretically justified in the Keenan and Comrie (1977) framework where it can be argued that Dzamba and closely related languages can relativize on nonsubject NPs only after they have undergone topicalization. Syntactic­ ally, the preposing exhibited in relative clauses in these languages is very much reminiscent of this rule (see sec­ tion 2.2.2), but semantically and morphologically the de­ rived structure do not correspond to or reflect topicalized sentences. In particular, there is no topic/focused reading in relative clauses, nor is there the type of agreement pattern found in topicalized sentences. This is unlike languages such as ChiShona (Bantu zone S.13) which uses a passive strategy for non-subject relativization. In this language there is syntactic, semantic, and mor­ phological evidence that passive is actually involved (cf. Bokamba 1981). In view of the weakness of the evidence in support of the involvement of topicalization in the derivation of relative clauses, it is reasonable to assume that the pre­ posing of object NPs and other VP dominated NP (i.e., in­ strumentais and locatives) is the result of a different rule: Rel-NP Preposing (cf. Bokamba 1981). This rule is analogous to the generally accepted rule of Relative Pro­ noun Fronting in languages like English and French; the only difference is that it preposes noun phrases rather than relative pronouns. If this proposal is correct, then Relative Clause Formation will be seen as entailing not only SVI, but also as having applicational precedence over VA. As a result, SVI can no longer be considered a cy­ clic rule, as suggested earlier: it must be viewed as a postcyclic rule since Relative Clause Formation is postcyclic. If this is correct, then the postcyclicity of Sub­ ject-Verb Agreement can no longer be doubted, at least for the Dzamba-type languages. 2.2.4

Coordinate

Deletions

and

Verbal

Agreement.

Further evidence in support of the hypothesis that Sub­ ject-Verb Agreement in Bantu languages is a postcyclic rule may be found in constructions involving various types of reduced coordinate structures. Rules such as Conjunction Reduction (involving reduced subject and ob­ ject NPs) and Gapping (reduced verbs) are generally as­ sumed to operate postcyclically (Koutsoudas 1971, Morgan and Green 1972). Now, in Bantu languages when these rules

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

39

interact with VA, the latter invariably applies last (Givon 1970, 1972, Bokamba 1981). Let us consider a few ex­ amples of these types of constructions in the languages under investigation here (cf. Bokamba 1981 for a detailed discussion). Subject-verb agreement in unreduced conjoined sen­ tences operates in the same manner as in simple sentences: the agreement is governed by the logical subject in each conjunct. This can be seen in examples such as the fol­ lowing : (42)

Lingala

a.

Mo-sálá (cl. 3) mu-síl-ákí na ntcfngo*, mpé work SP-finish-Impf in morning, and ba-kabínda (cl. 2) ba-kέí mbóka. carpenters SP-went home (The work was completed this morning, and the carpenters went home.)

b.

Li-báta (cl. 5) li-ko-lía kpánga, mpé duck SP-Fut-eat manioc and ntaba (cl. 9) e-ko-lía mpcndu. goat SP-Fut-eat manioc leaves. (The duck will eat the manioc/cassava, and the goat will eat the manioc leaves.)

c.

Ngáí na-ko-kεnda mpé yc o-ko-tíkala awâ. I SP-Fut-go and you-sg. SP-Fut-remain (I will go, and you [sg.] will stay/remain here.) In sentence (42a) the subject in each conjunct belongs to a different noun class: the noun mo-sal a belongs to class (3/4) singular, while ba-kabinda is a class (1/2) plural. The former triggers the subject prefix {mu-}, while the latter triggers {ba-}. Similarly, in (42) the subject of the first conjunct, li-bata, is a class (5/6) singular noun, and that of the second conjunct, ntaba, a class (9/10) singular noun. Each governs a different sub­ ject prefix: {li- } for li-báta, and {e-} for ntaba. How­ ever, in (42c), while both subjects belong to class (1/2) singular, they refer to different persons: ngáí refers to first person singular, whereas c refers to second person singular. As a result, they control different subject prefixes on the verb: {na-} for ngáí, and {o-} for c.

40

AFRICAN LINGUISTICS

Parallel facts can be cited from the other two languages (see examples 43 and 44 below). As seen earlier, there are no alternative agreement prefixes for the subject nouns and pronouns contained in (42). And as long as there is no other syntactic rule applying to such structures, no meaningful conclusion can be drawn from the behavior of VA in conjoined sentences. If we consider sentences involving VP-Reduction and in which the verb is not restricted to cooccur only with animate subjects, however, the behavior of Subject-Verb Agreement in conjoined sentences becomes apparent. Let us examine, for instance, sentences (43a, d) which have un­ dergone VP-Reduction to produce (43b) and (43e), respec­ tively: (43)

Lingala

a.

Eyenga a-yá-ákí áwâ lelc, mpe Nsengela Eyenga SP-came here today and Nsengela /

/

.

/

/

A

-

i

l

'

a-ya-aki awa lelo. SP-came here today. (Eyenga came here today, and Nsengela came here today.) b. c. d.

e.

Eyenga mpe/na Nsengela ba-yá-ákí áwa lelc. Eyenga and Nsengela SP-came here today. *Eyenga mpe/na Nsengela a-yá-ákí áwa lelo. Ngái'na-ko-zonga mbóka Iels te, mpe yo I SP-Fut-return home today not, and you o-ko-zonga mbóka lelc te. SP-Fut-return home today not (I will not return home today, and you [sg.] will not return home today.)

Ngái na yo to-ko-zónga mbóka lelo te. I and you SP-Fut-return home today not (You and I will not return home today.) f. *Ngáí na yó,na-ko- zonga .mbóka lelo te. o-ko-zonga It will be noted here that in both sentences (43a) and (43d) the verb in each conjunct agrees with its subject noun. In the former sentence, the verb ko-yá-a agrees with its subject which is a third person singular or class (1/2) singular. The subject prefix for each verb form in this case is {a-}, as expected. When VP occurs

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

41

and the subjects are regrouped, however, the agreement prefix exhibited on the remaining verb is plural, viz. {ba- } (class 1/2) . A singular agreement prefix, as attested in the ungrammaticality of (43c), is not permitted. Simi­ larly, in (43d) the verb ko-zonga agrees in person and number with the subject in each conjunct: in the first conjunct the subject prefix is {na-}, and in the second it is {o-}, as seen previously. As in (43c), a singular agreement with one of the subject nouns is not possible (43f); instead, the verb must bear the first person plural subject prefix, {to-}. The facts of Dzamba are analogous to those from Lingala. Further exemplification of the phenomenon is provided by the following KiSwahili data: KiSwahili

(44)

a.

Wewe u-ta-chukua m-zigo hu-u, na yeye you-sg SP-Fut-carry load Dem-Ag and he/she a-ta-chukua m-zigo hu-u SP-Fut-carry load Dem-Ag (You will carry this load, and he/she will carry this load.)

b.

Wewe na yeye m-ta-chukua mi-zigo hi-i. you and he SP-Fut-carry load Dem-Ag (You and he/she will carry these loads.) *Wewe na yeye{ u-ta-chukua } mi-zigo hi-i. a-ta-chukua Juma a-li-ondoka jana na Asha a-liJuma SP-Pst-go out yesterday and Asha ondoka jana. SP-Pst-go out yesterday (Juma went out/away yesterday, and Asha went out/away yesterday.)

c. d.

e.

Juma na Asha wa-li-ondoka jana. Juma and Asha SP-Pst-go out yesterday (Juma and Asha went out/away yesterday.) f. *Juma na Asha a-li-ondoka jana. It will be noticed here that in the unreduced coordinate sentences (44d) VA is governed by the logical subjects, which are all singular. In the corresponding reduced sen­ tences, however, VA must be controlled by the conjoined subjects; agreement with any of the underlying singular subject nouns leads to ungrammaticality (44c, f ) . As in

42

AFRICAN LINGUISTICS

the case of the Lingala data, the derivation of sentences like (44b, e) are possible only if the two rules apply in this order: VP-Reduction and Subject-Verb Agreement. The examples presented thus far on the interaction of VP-Reduction and Subject-Verb Agreement in coordinate sen­ tences have all consisted of subject nouns belonging to the same noun class, the human class (1/2). In cases like this and others where the subjects of the conjuncts be­ long to the same noun class the agreement in the reduced coordinates is almost always controlled by the plural of that class. That is, if two or more singular subject nouns occur in a conjoined sentence and VP-Reduction applies, the subject prefix on the remaining verb in that sentence will govern the plural agreement of that class (e.g., 2, 4, 6, 8, 10). If the conjoined subjects belong to differ­ ent noun classes, however, a class-gender conflict arises; and different strategies must be employed to resolve the conflict (cf. Givón 1970, 1972, Voeltz 1971, Bokamba 1981). The class-gender resolution strategies used and the class(es) into which the conflicts are resolved vary from one language group to another. In certain Bantu languages there is no possible gender conflict resolution strategy for many of the nonhuman noun class nouns; this appears to be the case in Zulu and Xhosa (cf. Voeltz 1971, for Xhosa). In the languages under consideration here there are two main strategies. The first involves the resolu­ tion of the class-gender conflict into the "things" class (7/8), and the second a nonresolution pattern based on the adjacency principle. Consider, first, the former strategy illustrated by the following Dzamba data: (45)

Dzamba

a.

i-zo-ombi (cl. 5) zi-bi-bunga, bo Spf-ax SP-Fut-lose and é:-ngbongo (cl. 7) e-bi-bunga. Spf-machete SP-Fut-lose. (The ax will be/get lost, and the machete will be/get lost.)

b.

i-zo-ombi (cl. 5) n'e:ngbongbo bi-bi-bunga. the-ax and the-machete SP (cl. 8)-Fut-lose (The ax and the machete will be/get lost.)

c.

*i-zo-ombi (cl. 5) ma-bi-unga. SP (cl. 6)

n'e:-ngbongbo (cl. 7)

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

(46)

43

d.

*i-zo-ombi (cl. 5) n'érngbongo (cl. 7) i-bi-unga. SP (cl. 10)

a.

í-Mu-ngate (cl. 3) mw-etez-í, bo í-bó-mpe the-manioc SP-arrived and the-manioc (cl. 14) v-etez-i. leaves SP-arrived. (The manioc [root] has arrived, and the manioc leaves have arrived.)

b.

i-Mu-ngate (cl. 3) n'í-bó-mpe (cl. 14) the-manioc and the manioc leaves bi-etez-i. SP (cl. 8)-arrived (The manioc and the manioc leaves have ar­ rived. )

c.

*i-Mu-ngate (cl. 3) n'í-bó-mpe (cl. 14) ma-etez-i. SP (cl. 6) d. *í-Mu-ngate (cl. 3) n'í-bó-mpe (cl. 14) mi-etez-i. SP (cl. 4) What these sentences show is this: first, given the unre­ duced examples in (45a) and (46a), the application of VA will result in the occurrence of subject prefixes that re­ fer to the subject noun in each sentence, viz. {zi-} (cl. 5) and {e-} (cl. 7) in (45a), and {mw-} (cl. 3) and{v-}19 (cl. 14) in (46) . Second, in the VP-reduced sentences (45b) and (46b) verbal agreement is governed by class (8); in none of these sentences can it be controlled by the plural class of one of the subject nouns, unless that noun belongs to class (7) as in the case of é-ngbongbo. This fact is illustrated in the ungrammaticality of the (c) and (d) sentences. The strategy exemplified in (45) and (46) is the most commonly used in class-gender conflicts in this lan­ guage, Likila, and Lingala, whenever the subject nouns belong to different classes and can be interpreted, to­ gether, as "objects/things." This is also true of LuGanda and ChiBemba (Givón 1970). The facts presented above can be duplicated not only in the Dzamba-type languages (Likila, Lingala, and Libinza), but also in KiSwahili. The main difference between

AFRICAN LINGUISTICS

44

the Dzamba-type and the latter language is that animals, regardless of their size, govern human class (1/2) agree­ ment; as a result, there is no class-gender conflict when­ ever nouns referring to animals occur as subjects in con­ joined sentences (see Chapter 2 ) . For nonhuman nouns, how­ ever, the pattern is precisely as in Dzamba, as illus­ trated below: (47)

Swahili

a.

Ki-ti (cl. 7) ki-me-vunjika, na m-guu chair SP-Perf-broken and leg (cl. 3) wa meza u-me-vunjika. of table SP-Perf-broken (The chair is broken, and the table's leg is broken.)

b.

Ki-ti- (cl. 7) na m-guu (cl. 3) wa meza chair and leg of table vi-me-vunjika (vyote). SP (cl. 8)-broke (all) (The chair and table's leg are [all] broken.)

*Ki-ti na m-guu wa meza zi-me-vunjika (vy-ote). SP (cl. 10) d. *Ki-ti na m-guu wa meza i-me-vunjika (y-ote). SP (cl. 4) Ag(cl. 4)-all As in the case of Dzamba, gender conflict resolution in class (10) is not permissible, nor is it in the plural class of one of the subject noun, unless it is class (8). It is evident from the data presented thus far that Sub­ ject-Verb Agreement must apply after Conjunction Reduc­ tion, i.e., VP-Reduction in this case, to have access to both subject NPs in order to govern the class (8) subject prefix. This hypothesis is given further support in the oc­ currence of the second verbal agreement strategy in re­ duced coordinate structures. In this strategy, VA is com­ monly governed by the subject noun that occurs adjacent to the verb, as long as the two conjoined nouns are of equal status in the natural and animacy hierarchies. Con­ sider, first, the case involving equal subjects illus­ trated in the following Dzamba and Swahili sentences: c.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

(48)

Dzamba

a.

i-Zo-ombi (cl. 5) n'é:ngbongbo (cl. 7) the-ax and the-machete e-bi-bunga. SP (cl. 7)-will-lose (The ax and the machete will be/get lost.)

b.

*i-Zo-ombi n'é:ngbongbo zi-bi-bunga. SP (cl. 5) é:-ngbongbo n'í-zo-ombi zi-bu-bunga, the-machete and the-ax SP (cl. 5) willlose (The machete and the ax will be/get lost.)

c.

(49)

45

Swahili

a.

b.

Ki-ti (cl. 7) na m-guu (cl. 3) wa meza chair and leg of table u-me-vunj ika. SP (cl. 3) broken (The chair and the table's leg are broken.)

*Ki-ti na m-guu wa meza ki-me-vunjika. chair and table's leg SP (cl. 7)-broken c. M-guu wa meza na ki-ti ki-me-vunjika. leg of table and chair SP (cl. 7)-broken (The table's leg and the chair are broken.) That is, the Dzamba sentence (45b) and the Swahili (47b) can be alternatively expressed as in (48) and (49), re­ spectively. What we have here is a case of avoidance of classgender conflict resolution: instead of having the differ­ ent class-genders collapsed into a neutral class, the rule simply applies to the subject that is immediately adjacent to the left of the verb. The rule apparently cannot skip the NP adjacent to the verb and analyze the first, i.e., that occurring at the beginning of the sentence, in order to copy its feature-specifications, because the resulting sentences (48b), and (49b) are ungrammatical. The first subject NP can, however, govern VA provided that it is moved to the position adjacent to the verb, as in (48c) and (49c) . Additional facts exemplifying the interaction of co­ ordinate deletions with VA can be cited, but they will not affect the analysis presented here. We have shown

46

AFRICAN LINGUISTICS

elsewhere (Bokamba 1981) that whenever VA interacts with known cyclic rules such as passivization, reflexivization, and postcyclic rules like Conjunction reduction and VPreduction, it always applies last (Bokamba 1981). Unless it can be shown that the last two rules are cyclic, the conclusion that follows is rather clear: VA is a postcy­ clic rule because it applies after postcyclic rules. 3.0

CONCLUSIONS AND THEORETICAL IMPLICATIONS 3.1 Summary, We have presented here data on verbal agreement in three Bantu languages: Dzamba, Lingala, and Swahili. The aim of the paper was two-fold: (1) to exam­ ine the behavior of the rule in various sentences, and (2) propose an analysis for the facts in question. It has been shown in this regard that VA, contrary to the pre­ vailing impression given in the recent literature, is a complex rule. The complexity is evident at all levels: se­ lection of the features of this NP that control agreement, and level at which rule applies. In particular, it was shown in section (2.1) that VA in simple sentences is invariably triggered by the logi­ cal subject NP, and that several features (including noun class, person, and number) of that NP governed the agree­ ment prefix attached on to the verb. In the following section (2.2) the first conclusion was modified by showing that in constructions involving inversions it is the de­ rived subject that governs verbal agreement. The relevant data examined were drawn from passive, topicalized, and relative clause sentences. This analysis was followed by a discussion of more rule interaction to determine the level at which VA applies. Data from relative clauses and reduced coordinate sentences demonstrated that VA in the languages examined here applies postcyclically. 3.2 Theoretical Implications. Whether or not the analysis proposed here is correct remains to be determined by future research. What is evident, however, is the fact that VA cannot be viewed as a cyclic rule in these lan­ guages without (a) complicating their grammars, and (b) without abandoning many of the basic assumptions associ­ ated with the principle of the cycle and rules like Rela­ tive Clause Formation, Conjunction Reduction, VP-Reduction, and other postcyclic rules. To my knowledge no one has yet presented any convinc­ ing evidence in support of the cyclicity of VA. The evi­ dence generally referred to in the literature as suppor-

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

47

tive of this hypothesis comes from constructions involving the rule of There Insertion. The argument is essentially that the occurrence of sentences like (50) below cannot be accounted for unless VA is assumed to operate cyclic­ ally: (50) a. There are crocodiles at the bottom of the Zaire River. b. There were three crocodiles caught by the fisherman tonight. c. *There is crocodiles at the bottom of the Zaire River. d. *There was three crocodiles caught by the fisherman tonight. It is argued with respect to such data that there becomes the derived subject of the sentence after the insertion has occurred, and it should, "normally" govern VA as in (50c-d). The fact that these sentences are ungrammatical, in contrast to (50a-b), shows that VA is cyclic rule. This argument, however, has been rejected as invalid in some recent studies (e.g., Pullum 1979), because there are sentences like (5lb-c) below in English that cannot be explained under this analysis (Pullum 1979: 175, 206): (51) a. There are likely to be storms in the North Sea. b. There {are}three smiling tigers and a dead *is keeper in the cage. c. There{?*are is 's) a dead keeper and three smiling tigers in the case Similar examples have been cited in Morgan (1972). The problem here is that in (51a) the plural noun storms can­ not be considered, as Pullum (197 9) has correctly ob­ served, to have been the subject of the verb at any point in the derivation, at least in any straightforward way. In (5lb-c), in contrast, the conjoined NPs ought to be analyzed by the rule of VA as a plural subject and, there­ fore, the agreement will be in the plural; but as evi­ denced here, this is not the case when the NP "a dead keeper" immediately follows the verb. The VA pattern ob­ taining here is based on the commitative strategy in co­ ordinate sentences where the NP adjacent to the verb con­ trols verbal agreement (Givón 1972, Bokamba 1981). The occurrence of sentences like (51b-c) clearly challenges the validity of the cyclicity hypothesis for verbal agree­ ment.

48

AFRICAN LINGUISTICS

While we do not have adequate data at this point from extensively studied languages such as English, French, German, and Italian to determine the extent to which the conclusions presented in this study apply to these languages, we suspect that future research will find VA to be a noncyclic rule. Evidence from conjoined structures (Morgan 1972) and inversions (Green 1977) sug­ gests that this may be the case. A cursory examination of errors made in sentences involving conjunction reduction and disjunction, both in standard and nonstandard varie­ ties of different languages suggests that VA is a. late rule. We have found in a preliminary examination of such constructions that speakers of different languages tend to apply VA late in the derivation; as a result, the rule is often triggered by the subject that is last mentioned, the one occurring adjacent to the verb. It appears from such facts and others presented here that VA is based on pragmatic considerations such as top­ ic/comment, and varying degrees of animacy/inanimacy in what has been characterized as "natural hierarchy" (Hawkinson and Hyman 1974, Givon 1976, Bokamba 1981). It is precisely for cases like these, which appear otherwise to be syntactically difficult to characterize, that Perlmutter and Postal (1974) proposed the Dummy Agreement Law. One question that arises in connection with facts like those considered here is whether Subject-Verb Agree­ ment ought to be analyzed as cyclic in some languages and postcyclic in others. This question cannot be answered adequately until further studies on the subject have been conducted. What is evident from the few available studies, including Vanek (1977) where the cyclicity of the rule is questioned but not rejected explicitly, is that VA cannot be a cyclic rule in general. With regard to the facts ex­ amined here, our conclusion is that it is a postcyclic rule.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

49

FOOTNOTES

* The research reported upon here was supported by the University of Illinois Graduate College Research Board grant ICR 40-32-52-320 dur­ ing the academic year 1979-80. Earlier versions of this paper ap­ peared in SLS 9,2 (1979) and CLS 16. I am indebted to the following colleagues for their comments on earlier versions of this paper: Nick Clements, Charles Kisseberth, Salikoko Mufwene, and Alexis Takizala M. The final responsibility is mine alone. 1. Alexis Manaster-Ramer informed me (in a discussion subsequent to a lecture at CLS in February 1980) that Lawler's data may not be ac­ curate. Whether or not this is correct will not affect the present analysis. The complexity of the rule is demonstrated here independ­ ently of Lawler's study. 2. The Lingala and Swahili data presented here are drawn from the standard dialects of these languages, because they illustrate better the processes under discussion than the non-standard dialects. For a discussion of the levelling off of grammatical agreement in these languages, see Bokamba (1977). 3. Luganda, which is known to have the largest number of noun classes to date, for example, has twenty-one; it does not have classes 19 and 21 (cf. Weimers 1973: 165 for a recent discussion). 4. One of the major differences in this respect between Dzamba/ Lingala and Swahili is that the latter has locative classes, while the former do not. 5. I am indebted to Mr. Hassan Marshad, a native speaker of Swahili from Kenya, for this observation. 6. This is true only of the standard dialect of the Lingala-type languages, because a different pattern which recognizes this dichot­ omy has emerged in the non-standard dialects. The change involves, in some cases, a drastic reduction of the agreement prefixes so as to lead to a general agreement system based on the [human] vs. [nonhuman] dichotomy; and in other cases on the [animate] vs. [inanimate] opposition (cf. Bokamba 1977, 1979, for a more detailed discussion of this).

50

AFRICAN LINGUISTICS

7. Givon has since abandoned this position for a pragmatic approach based on the notion of topic-comment (Givon 1976). 8. But uncommon passives such as those involving locatives in cer­ tain Bantu languages do present such problems to Relational Grammar. For some discussion of these constructions see Dalgish (1976) and Dalgish and Sheintuch (1977). 9. This hypothesis is based on the interaction of VA and There-In­ sertion in English. As discussed in section (3.0), however, the evi­ dence is unconvincing (see also Pullum 1979). 10. Chomsky (1977: 91) proposes to derive topicalized sentences from the following base rule under the X-bar theory: S

→

TOP

s

If we assume this rule, which is based on English syntax and subsumes certain behavior of comparative constructions in this language, the discussion here would be unnecessary. This rule, however, will con­ sistently generate ungrammatical structures in a number of Lingalatype languages, thereby making the application of Subject-Verb In­ version obligatory in a way that establishes an ad hoc correlation between a base and T-rule in these languages. Because of these ob.jections, I have preferred to retain the original formulation pro­ posed in Ross (1967b). I should point out, however, that the argu­ ments against the cyclicity of VA do not depend on the formulation of this rule. 11. I am assuming the following formulation of Subject-Verb Agreement (Bokamba 1981): NP

X

V

Y

Conditions: a) Term 2 cannot be an NP, but it may be zero or an S. b) F = all inherent and relevant derived semantic features of the NP at the point at which the rule applies. 12. I am indebted to Nick Clements for the suggestion that both rules may be cyclic. 13. Hirschbühler's (1975:164) reformulation of Left-Dislocation as a base-generated rule, as below, would also account nicely for the same facts: S

→

(NP) S

The facts of the languages discussed here, as well as those from other Bantu languages, do not force a choice between these analyses.

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

51

Hirschbühlerf s proposal may be preferred on general grounds, however, if it is assumed that Pronominalization is not a T-rule. 14. There are two alternative constructions. They are as follows: a. nyumbani, daktari a-li-Jcu-leta ki-ti. (To some place at/in the house, the doctor brought the chair.) b. nyumbani, daktari a-li-m-leta ki-ti. (To a specific place inside the house, the doctor brought the chair.) 15. The inversion appears to be dictated by both semantic and syntac­ tic considerations. Semantically the logical subject loses its func­ tion once another NP has been promoted to the subject position. And from the point of view of the syntax of Subject-Verb Agreement, the derived subject must control verbal agreement. It appears, then, that the application of SVI is in part a strategy to create the conditions under which Subject-Verb Agreement can apply. 16. I have shown elsewhere (Bokamba 1976b, 1981) that morphemes like {zi-} on the verb of the embedded relative clause in Dzamba has to be viewed as a verbal agreement prefix which is independent of the relative marker, because of facts like those below where the negative marker -ta- intervenes between the relative and the agreement mor­ phemes : a. i-zi-bata i-ta-zi-eza-aki o-mama l-b-so emba zi-ke-i the-duck Rm-Neg-Ag-give-Impf the-mother the-rice not Ag-go-IP (The duck to which (the) mother did not give any rice left.) b.

í-bí-eka i-ta-bi-tokí-ákí banga émbá bi-taban-í. the-food Rm-Neg-Ag-heat-Impf us not Ag-spoil-IP (The food that we did not heat/warm up is spoiled.)

What these facts indicate is that the relative marker for all nonhuman noun classes is identical, but the agreement prefix serves as the differentiating element for the various noun classes. The latter always bears a high tone if it occurs adjacent to the former, other­ wise the tone is copied onto the negative prefix, as above. 17. This hypothesis can be considered only if Topicalization is viewed as T-rule, rather than a base-generated PS rule. 18. I am assuming here that since Dzamba and other closely related languages do not exhibit both subject prefixes and relative markers, as for example in Swahili, Subject-Verb Agreement can be viewed as subsuming Relative Pronominalization. It will be -noted here that the verbal agreement prefix in a relative clause differs from that in a non-relative clause in that it bears a high tone; morphologically the two prefixes are identical. 19. This morpheme, {v-} , is underlyingly derived from {bu-} via wglide and fricativization. This is a common rule in Dzamba and re-

52

AFRICAN LINGUISTICS

lated languages whenever {bu-} occurs before another vowel. 20. See also Pullum and Roberts (1975) for a more detailed discussion of this question.

REFERENCES

Ashton, E.O. 1946 Swahili Grammar. London: Longmans. Bokamba , E.G. 1976a Relativization in Bantu languages revisited. In Peter A. Reich, ed., The Second LACUS Forum 1975, 38-50. Columbia, S.C.: Hornbeam Press. 1976b Question Formation in Some Bantu Languages. Ann Arbor: Uni­ versity Microfilm International. 1977 The impact of multilingualism on language structures: The case of Central Africa. Anthropological Linguistics 19: 181202. 1979 Noun class and agreement changes in Bantu languages. Paper read at the Colloquium on Grammatical Changes, University of Leiden, The Netherlands, September. 1981 Aspects

of Bantu Syntax.

Unpublished manuscript.

Chomsky, N. 1965

Aspects

of

the

Theory

of Syntax.

Cambridge, MA: The M.I.T.

Press. 1977 On Wh-Movement. In P.W. Culicover, T. Wasow, and A. Akmajian, eds., Formal Syntax. New York: Academic Press. Crockett, D.B. 1976

Agreement

in Contemporary

Standard

Russian.

Cambridge, MA:

Slavica. Dalgish, G.M. 1976 Passivizing locatives in OluTsootso. Studies guistics Sciences 6: 57-68.

in the

Lin­

Dalgish, G.M. and G. Sheintuch 1977 On the justification for language-specific subgrammatical rules. Studies

in African

Linguistics

8: 219-39.

53

VERBAL AGREEMENT AS A NONCYCLIC RULE IN BANTU

Givón, T. 1970 The resolution of gender conflicts in Bantu conjunction: when syntax and semantics clash. In Papers from the 6th Regional Meeting. Chicago: Chicago Linguistic Society, The University of Chicago. 1972a Studies in ChiBemba and Bantu Grammar. Supplement 3, Studies in African Linguistics. Los Angeles: African Studies Center, UCLA. 1972b Pronoun attraction and subject postposing in Bantu. In P.M. Peranteau, J.N. Levi, and G.C. Phares, eds., The Chicago Which Hunt: Papers from the Relative Clause Festival, 19097. Chicago: Chicago Linguistic Society, University of Chi­ cago. 1976

Topic, pronoun, and grammatical agreement. In C. Li, ed., Subject and Topic. New York: Academic Press.

1979

On Understanding

Grammar.

New York: Academic Press.

Green, G. 1977 Do inversions in English change grammatical relations ? Studies in the Linguistic Sciences 7: 157-81. Gregersen, E.A. 1967 Prefix and Pronoun in Bantu. IJAL Supplement, vol. 33, No.3. Bloomington, Indiana. Indiana University. Hawkinson, A.K. and L.M. Hyman 1974 Hierarchies of natural topics in Shona. Studies

Linguistics

in

African

8: 199-218.

Hirschbuhler, P. 1975 On the source of lefthand NPs in French. Linguistic 6, 1: 155-65.

Inquiry

Keach, B.N. 1980 The Syntax and Interpretation of the Relative Clause Con­ struction in Swahili. Ph.D. Dissertation, University of Massachusetts, Amherst. Keenan, E.L. and B. Comrie 1977 Noun phrase accessibility and universal grammar. Inquiry 8, 1: 63-99.

Linguistic

Koutsoudas, A. 1971 Gapping, conjunction reduction, and coordinate deletion. Foundations of Language 7: 337-86. Lawler, J. 1977 A agrees with  in Achenese: A problem for relational gram­ mar. In P. Cole and J.M. Sadock, eds., Syntax and Semantics 8: Relational Grammar, 219-48. New York: Academic Press.

54

AFRICAN LINGUISTICS

Morgan, J.M. 1972 Verb Agreement as a rule of English. In P.M. Peranteau, J.N. Levi and G.C. Phares, eds., Papers from the 8th Regional Meeting. 278-86. Chicago: Chicago Linguistic Society, The University of Chicago. Morgan, J.M. and G. Green 1972 A guide to the study of syntax. Unpublished manuscript, Uni­ versity of Illinois. Perlmutter, D. and P. Postal 1974 Relational grammar : LSA Linguistic notes. Mimeographed. Pullum, G.K. 1979 Rule Interaction and the Organization Garland Publishing, Inc.

of a Grammar. New York:

Pullum, G.K. and D. Roberts 1975 Verb agreement and there-insertion. The necessity of the Dummy Agreement Law. (Squibs on Relational Grammar.) Ms., University College, London. Ross, J.R. 1967a Gapping and the order of constituents. In M. Bierwisch and K.E. Heidolph, eds., Progress in Linguistics: A Collection of Papers. The Hague: Mouton. Ross, J.R. 1967b Constraints on Variables in Syntax. M.I.T. Doctoral Disser­ tation. Bloomington: Indiana University Linguistics Club. Vanek, A.L. 1977 Aspects of Subject-Verb guistic Research, Inc.

Agreement.

Edmonton, Canada: Lin­

Voeltz, E. 1971 Surface constraints and agreement resolution: some evidence from Xhosa. Studies in African Linguistics 2, 1: 37-60. Wasow, T. 1979 Problems with pronouns in transformational grammar. In F. Heny and H.S. Schnelle, eds., Syntax and Semantics 10: Se­ lections from the Third Groningen Round Table. 199-222. New York: Academic Press. Weimers, W.E. 1973 African Language Structures. California Press. Wilson, P.M. 1970 Simplified Bureau.

Swahili.

Berkeley, CA: University of

Nairobi, Kenya: East African Literature

AKAN VOWEL HARMONY: A NONLINEAR ANALYSIS George N. Clements (Cornell University) What is the structure of phonological representa­ tion ? Until recently, few linguists would have hesitated to reply that the matter was simple: phonological repre­ sentations consisted of strings of phonemes, perhaps themselves analyzable in terms of matrices of distinctive features. However, any serious study of so-called "pro­ sodic" processes in phonology quickly shows that such a view is untenable. Prosodic properties such as tone, stress, and intonation require a conception of phonolog­ ical structure which allows phonetic properties to be re­ lated to each other in rather more complex ways, involving nonlinear associations between features of different cat­ egories. Vowel harmony, a phenomenon located at midpoint be­ tween true prosodie characteristics such as stress and tone and purely local phenomena such as the assimila­ tion of one segment to a neighbor, provides a particular­ ly interesting field for the study of phonological repre­ sentation since it involves properties which elsewhere (and more frequently) define phonemes (vowels) in the traditional sense; in other words, it involves the "suprasegmentalization" (in a sense to be defined) of prop­ erties which are normally segmental in scope. According­ ly, the study of vowel harmony is of particular interest for our growing awareness (and knowledge) of nonlinear modes of phonological representation. Vowel harmony has recently received the attention of a number of linguists who have proposed several different models for the description of vowel harmony in nonlinear terms. It is of particular importance that such discus­ sion should be informed by detailed studies of particular vowel harmony systems exhibiting some typological variety;

AFRICAN LINGUISTICS

56

otherwise, we run the risk of basing fundamental proper­ ties of our models on preliminary observations that may prove to be misleading when the intricacies of given sys­ tems are examined more closely. The present study attempts to provide an account of one vowel harmony system of West Africa, that of Akan (Asante dialect). It will be shown that the specific features of Akan vowel harmony provide strong motivation for the theoretical framework of áutosegmental phonology, as I have argued in briefer treat­ ments elsewhere (Clements 1976, 1977). Section 2 of this study reviews the phonetic basis of Akan vowel harmony, and section 3 provides a descrip­ tive overview of the system. In section 4 and 5 two major earlier descriptions of Akan vowel harmony are examined, those of Schachter and Fromkin (1968) and Stewart (1967). Sections 6 and 7 present an autosegmental account, and section 8 examines confirming evidence from a process of Vowel Raising. Finally, section 9 offers some general the­ oretical considerations. 1. The account which follows is based primarily up­ on the speech of Aaron Poku-Appiah and George Mensah, un­ dergraduate students at Harvard who assisted me in this investigation during the period 1976-1978. Mr. Poku-Appiah and Mr. Mensah are both from the Kumasi area, and their speech is representative of the Asante dialect as it has been described by other writers. 2. The phonemically distinctive surface consonants of the Asante dialect of Akan are listed in the following chart. Here and throughout this study, I use a broad pho­ netic transcription based (with a few departures) on the I.P.A. system, and similar to the system used in Stewart (1966b). A fuller description of Asante consonants, sup­ ported with palatographic tracings, is given in Dolphyne (1965). (1) voiceless oral stops rounded voiced oral stops rounded

labial alveolar palatal velar glottal

AKAN VOWEL HARMONY

57

voiceless fricatives rounded voiced nasal stops rounded glides

In addition to these, some speakers of Asante are reported to have a phonemically distinctive glottal stop, restric­ ted to prepausal position: /to/ 'buy1', /to ? / 'die in bat­ tle' (Dolphyne 19 65). Furthermore, some speakers have a phonemic contrast between the front rounded glide /ẅ/ and the back rounded glide /w/ before the vowel /c/: /ẅc/ 'to do 1 , /wo/ 'to pound' (Dolphyne 1965). Finally, some speak­ ers, as noted in section 8 below, have a contrast between palatalized and unpalatalized / s Y , s/ before the vowel /a/: /syanl/ 'to descend', /sanl/ 'to untie'. / s w y / is a voice­ less, labial, palatalized alveolar fricative. Palataliza­ tion and rounding are simultaneous, not sequenced coarticulations. The vowels of Akan fall into two sets according to their harmonic category. High and low vowels have phonemically distinctive nasalized counterparts, which behave like oral vowels with respect to vowel harmony.

As was first pointed out by Stewart (1967), the phonetic basis of the distinction between the vowels of each set is the position of the tongue root: advanced for the vow­ els of set 1, neutral or retracted for the vowels of set 2. This distinction can be confirmed visually by ob­ serving an Akan speaker in profile as he produces tokens of such pairs of vowels as /i, I/ or /e, ε/: the advancing of the tongue root produces a noticeable protrusion at the angle of the throat and upper neck in each case. The role of tongue root advancing has been subsequently con­ firmed in cineradiographic tracings by Painter (1973) and Lindau (1975). Lindau elaborated upon Stewart's original findings by observing that not only tongue root advancing,

58

AFRICAN LINGUISTICS

but other gestures are involved in producing the set 1set 2 distinctions, operating in conjunction to increase the volume of the pharyngeal cavity at the expense of the volume of the front oral cavity. These gestures include the lowering of the larynx, the retraction of the rear pharyngeal wall, and the raising of the velum. In view of these findings Lindau has proposed the term "expanded pharynx" to describe this feature. As no issue of sub­ stance hangs upon our choice of terminology (the articulatory correlates not being at present a matter of con­ troversy) , I shall retain the more familiar term "ad­ vanced tongue root" for the purposes of the present dis­ cussion. The feature "advanced tongue root" is distinct from the feature "tense" as this has been characterized by Wood (1975). According to Wood, the principal articulatory correlate of the feature "tense" is the increased constriction of the vocal tract at the point of closest constriction of the vowel in question: the palate for front vowels, the velum for high back vowels, the uvula and upper pharynx for mid back vowels, and the lower phar­ ynx for low vowels. The cineradiographic tracings of Akan vowels published by Lindau (1975) show that the degree of primary constriction in the production of corresponding set 1 and set 2 vowels may differ only slightly, while the degree of constriction at the tongue root varies sub­ stantially (Figures 38-41); in particular, for some speakers the constriction at the velum is not higher, but lower for /u/ than for /o/ (Figures 12, 13, 14, 40; sim­ ilar results were obtained for Igbo, Fig. 23, and Ateso, Fig. 26). Similarly, as is commonly the case in vowel harmony systems of this type, the low vowel which falls into set 1 differs from the low vowel which falls into set 2 in involving less, rather than more, constriction in the lower pharynx (Fig. 16). Wood, who did not inves­ tigate African languages in his study, argues on theoret­ ical grounds that increasing the size of the pharynx would not contribute significantly to the lowering of the first formant, generally taken as the primary acoustic correlate of the feature "advanced tongue root". However, available X-rays suggest that the rear cavity in the pro­ duction of set 1 vowels expands to roughly twice the vol­ ume it normally attains in the production of set 2 vow­ els; in his experiments, Wood increased the size of the lower pharynx by much smaller amounts (Figures 5, 8, 9 ) . This in itself would explain the negligeable effect of tongue root advancing in his results.1 The contrast between the vowels [a] and [a] is not

AKAN VOWEL HARMONY

59

phonemic in Asante. , a raised and fronted low vowel (often transcribed which approaches [ε] in articula­ tion, occurs to the exclusion of [a] in the following en­ vironments: (3)

a.

preceding a [+advanced, +high] vowel in a following syllable: 'navel1

-furuma n pc w i-e

'cow'

y funu

'belly'

kaŋk b i

'millipede'

b -y i-e

'witchcraft'

p t ii

'to slip'

w

ŋ

b.

c.

'to sneeze'

nsT

 ri

'to weigh'

wa-t u

'he has dug it'

preceding a [+advanced, -high, -low] vowel in a following noun syllable: plr ko

'pig'

kw

[proper name]



-

'parrot'

-go

'velvet'

preceding a root beginning with the sequence CYa, where Y is any member of the set /c , , s Y , s w y , ç w f ŋ/, as well as / s / from * / s y / in dialects no longer having / s y / (see section 8 below): 9 -j w a e-c

w

'assembly' w

-c ã

-ja

'scars' 'father'

w

m -c a

'I have cut (it)'

w -ŋanl

'he has awakened'

y

w -s anl Elsewhere we find [a] to the exclusion of

'he has descended' :

60

AFRICAN LINGUISTICS

(4)

a.

before any low vowel ( a-n ŋc î-e a-ta kasa wa-ŋ nsT

b.

before any other syllable except as specified in (3): 3 bapcI a-bεrεwa a-ko kc a-popcIrεnI-I jwarI yarI wa-to wa-çwe wa-ko-tu wa-be-tu a-tene wa-hojo mu finally: s i ka kosua m-moja e-c a a-mlna sa b i sa

or [a]): 'cows1 'female twin' 'to speak' 'he has sneezed'

'cassava' 'old woman' 'fowl' 'frog' 'to bathe' 'to be ill' 'he has thrown (it)' 'I have looked at (it)' 'he has gone and dug (it)' 'he has come and dug (it):' 'it (news) has spread' 'he has loosened it' 'money' 'egg' 'blood' 'scar* 'hole' 'to dance' 'to ask'

[a] is therefore to be considered a surface variant of /a/,4 and in the dialect under consideration here will not occur in underlying representations, /a/ has other, less striking variants. A very slightly fronted variant is found to occur after the palatal consonant in  a 'to cut'; data given in Painter (1973, Table I) show this vowel to be far more closely related along all articulatory para-

AKAN VOWEL HARMONY

61

meters to the low vowel of ba 'to come1 than to the low vowel of y a hũ 'we have seen'. A certain amount of contro­ versy surrounds the variant of the low vowel in words like bisa 'to ask'. Schachter and Fromkin (1968) record it as a tense central vowel []. However, I have been informed by Schachter that "while some speakers pronounce this /a/ differently from the /a/ of sa or kasa, others may not" (personal communication, 1974). In the speech of Mr. Poku-Appiah and Mr. Mensah, the low vowels in cwa 'to cut' and bisa 'to ask' are not perceptibly different from the low vowels in kasa 'to speak1. Spectrographic measure­ ments taken by the author are given below: Table I: Spectrographic Measurements of Akan Vowels [ai] Fi: 600 F 2 : 1270 tokens : 21 key:

[a2] 625 1355 27

[a3] 635 1330 9

520 1550 4

[a1] [a2] [a3]

 'to come' b î sa 'to ask' w ca 'to cut1 m -tu 'I have dug' While these measurements can only be taken as suggestive, they are consistent both with the auditory characteristics of these vowels as I hear them and with the transcriptions of Stewart (1967), who writes [3] for my and who does not distinguish [ax, a 2 , a 3 ] , writing them all as [a]. We may therefore assume that in at least some varieties of Asante, and [a] are surface reflexes of a single un­ derlying vowel phoneme, and we may introduce the fol­ lowing Segment Structure Condition stating that all low vowels are [-advanced] in underlying representation: → [-advanced]

3. In words containing no low vowels, all vowels belong either to set 1 or to set 2; that is, all vowels must be either [+advanced] or [-advanced]:

AFRICAN LINGUISTICS

62

(6)

'nest1 'stone 1 'rat1 'eagle' "'he pierced (it)' 'he showed (it)' 'he came and dug (it)' 'he came and threw (it)'

e-bu-o e-boo-kusi-e -k dI-c o-fiti-i -clre-I o-be-tu-i -bε-to-I cf.

f it¡ cIrε tu to

'to pierce' 'to show' »to dig' 'to throw'

Low vowels may occur in roots containing vowels of either the [+advanced] or the [-advanced] set, as the following examples show: (7)

yarI  ri b i sa pira baylre y funu Q-kruma wowa

'to be sick' 'to weigh' 'to ask' »to sweep' 'yam' 'belly' 'okro' 'bee'

In particular, the low vowel may cooccur in a single root with vowels of both categories, provided they occur to either side of it: (8)

plrakô fuηanl ka ηka b i

'pig1 'to search' 'millipede'

Such examples show that the low vowel of Akan does not have the status of the "neutral" vowels of such more fa­ miliar languages as Finnish and Mongolian. True "neutral" vowels occur exclusively in roots of one harmonic catego-

63

AKAN VOWEL HARMONY

ry or the other. The Akan low vowel, on the other hand, is "opaque" in the sense that it interrupts harmony do­ mains, and may (as we shall see below) initiate new do­ mains subject to its control. Roots of mixed category containing low vowels such as those listed in (7) and (8) are in no way exceptional in Akan and are in fact very common. Such roots will be termed mixed vowel roots. In contrast to these are roots containing vowels from different harmonic sets, but not dis­ containing /a/. Roots of this type, which I will term h a r m o n i c roots, are highly exceptional; only two (first noted by Dolphyne (1965)) have been recorded: (9)

piηe ηinsεη

'to come close' 'to

be pregnant'

Words formed from mixed vowel roots and disharmonic roots conform to the principle of root control described in Clements (1976): nonlow prefixes agree in harmonic category with the initial root vowel, and suffixes agree with the final root vowel; cf. the earlier examples, as well as: (10)

'he asked (it) 1

o-bisa-I -kari-i

'he weighed (it)' 'she became pregnant' 5

o-pinsenl-I

Prefixes containing low vowels are subject to the prin­ ciples determining the distribution of [a] and [a] (cf. (3) and (4) above). The occurrence of low vowels in pre­ fixes is highly restricted; in particular, a prefix con­ taining (or consisting of) a low vowel always occurs in the first syllable of a word in surface representation. Low vowels do not occur in suffixes, with the sole excep­ tion of the nominal derivational suffix found in such forms as akuraa /a-kura-a/ 'village'. Let us finally consider a set of forms which at first sight appears quite aberrant. Examining the behav­ ior of roots which begin with one of the set of conso­ nants symbolized by C Y in (3c) followed immediately by /a/, we find that a prefix vowel must be [+advanced], while any suffixes must be [-advanced]. For instance: (11)

o-c w a-I

'he cut it'

wu-be- w arl o-panl-I

'you will bathe' woke up'

'he

AFRICAN LINGUISTICS

64

It would be natural to try to explain these apparently de­ viant forms by attributing the feature [+advanced] to the root-initial consonant (these consonants would bear this feature only before / a / ) . However, no observer has yet found any evidence for the presence of such a feature as an intrinsic specification, nor have I been able to detect any externally visible indications of such a feature (such as might consist of a bulging at the angle of the front of the neck and the underside of the chin) in the produc­ tion of prefixless forms, such as the singular imperative. In fact, the explanation for these forms is historical; see further discussion in section 8. 4. In this section and the next we examine two pro­ posals for the description of Akan vowel harmony within generative frameworks. In A Phonology

of

Akan:

Akuapem,

Asante

and

Fante

(1968) , Schachter and Fromkin present a detailed analysis of major portions of Akan segmental and tonal phonology within the general framework of Chomsky and Halle (1968) . In their version of this framework, vowel harmony is treated as a consequence of the operation of languageparticular rules, in the following way. All morphemes are assigned uniform -underlying shapes in which each segment is fully specified for all features. Morpheme Structure Conditions (MSC's) function as conditions of well-formed­ ness on morphemes in the lexicon; they do not have the ability to change features, and they effect no changes in the phonological composition of underlying phonological matrices. Phonological rules (P-rules) apply to such ma­ trices by inserting, deleting or permuting segments and changing feature values. In Schachter and Fromkin's analysis, vowel harmony involves the features tense/lax, implemented in Akan by tongue root advancing.6 Underlyingly, all affixes are lax, and all vowels in a morpheme agree in tenseness or laxness. Two phonemically distinct low vowels are recog­ nized, /a/ and /Ә/ (a low central tense vowel, distinct in value from the front vowel transcribed here as [a] and transcribed by Schachter and Fromkin as []); //, however, is subject to the restriction that it is never the only vowel in a root, unless preceded by /g/. The phono­ logical rule accounting for vowel harmony in affixes is re­ stricted to the domain of the word, and causes affix vowels to become tense if the nearest root vowel is [-low, +tense]. Schachter and Fromkin state this rule as follows (p. 98, 261): 7

AKAN VOWEL HARMONY

65

(12) Vowel Tensing (Schachter and Fromkin)

cond.: (a) X, Y do not contain # # ; (b) if the affected segment pre­ cedes X, X does not include a [+vocalic, +root] segment; (c) if the affected segment fol­ lows Y, Y does not include a [+vocalic, +root] segment. Finally, a late rule fronts (derived or nonderived) [] to [a] before any tense vowel: (13) Low Vowel Fronting

(Schachter and Fromkin)

Much of Schachter and Fromkin's analysis follows from two fundamental assumptions: first, that all vowels are fully specified in underlying representation; and second, that all vowels in a morpheme agree in tenseness. First, it follows that Akan has an underlying ten-vowel system in which /a/ and // are phonemically contrastive (the underlying representation of 'ask1 cannot be /bisa/, since the two vowels would disagree in tenseness; it must therefore be /bisa/). Second, low root vowels cannot con­ dition tenseness harmony, since affix vowels are always lax when the nearest root vowel is low, whether it is /a/ or //. Thirdly, affix vowels must be specified as lax underlyingly. A fourth consequence, not mentioned by Schachter and Fromkin, is that a further rule is needed (following (13)) to map all remaining occurrences of [] into [a] in those dialects (such as the one described here, or those described in Stewart's studies) which have surface [a] wherever Schachter and Fromkin record surface []. Several considerations suggest that it would be de­ sirable to abandon the second of these fundamental as­ sumptions, and to regard Akan as having an underlying nine-vowel system in which /a/ has no tense counterpart. First of all, in Schachter and Fromkin's analysis forms

66

AFRICAN LINGUISTICS

such as those in (8) must be considered exceptional. If, on the other hand, we regard occurrences of low vowels as bounding harmony domains, these forms become regular. A more serious problem regards the analysis of roots of the following type: i w arI

(14)

'to bathe 1

j w anI

'to flee'

;jaI

'to abandon'

w

c arI ç w anI y

(mu)

'to cross' 'to peel; hatch'

s anI

'to descend'

panI

'to awaken'

Such roots, which are quite characteristic, always take tense prefixes and lax suffixes (cf. (11)). To account for this behavior, Schachter and Fromkin set up underly­ ing representations like the following, for 'flee': /guagT/ (p. 1 9 4 ) . 8 These forms then constitute another class of exceptions to the MSC which requires all vowels in a morpheme to agree in tenseness. However, they cannot be analyzed as compounds, nor are they in any respect un­ usual in regard to other phonological rules (most or all of these roots have cognates elsewhere in Akan dialects, and must be regarded as native). To mark these roots as exceptions to the MSC governing tenseness in single mor­ phemes would miss a generalization: with only the two ex­ ceptions noted in (9), all apparent exceptions to vowel harmony in roots involve low vowels. If we assume that vowel harmony constraints apply only to sequences of nonlow vowels in Akan, a number of advantages can be obtained. First, the forms of (8) and (14) are no longer exceptional. Second, we can eliminate the superfluous tenth vowel /a/ together with the pecu­ liar restrictions upon its occurrence, and the extra rule needed in the dialects described in this study to map [] into [a]; in those dialects in which [a] occurs in such forms as [bisa] or [ j w ә n Ĩ ] , we may postulate rules of phonetic specification similar to the ones that will ac­ count for the variants [a1, a 2 , a 3 ] in the dialects re­ ported on in Table I. Third, we may eliminate the stipu­ lation given in the statement of Vowel Tensing (12) that low vowels do not condition vowel harmony; suffixes fol­ lowing roots like /bisa/ are lax (i.e. nonadvanced) sim­ ply because /a/ is lax. Having clarified these points, we may turn to a con-

AKAN VOWEL HARMONY

67

sideration of the theoretical framework adopted by Schachter and Fromkin. As we have seen, they adopt a rule-based model of vowel harmony which accounts for vowel harmony in terms of two independent types of statements : MSC•s determining cooccurrence restrictions in roots, and P-rules determining the harmonic category of affixes. Within this framework, it is entirely accidental that the same set of restrictions on vowel cooccurrence should ap­ ply internally in roots and externally across morpheme boundaries. In formal terms, it would be just as simple to describe a system which had harmony of a different type (or no harmony at all) within roots. Thus, one can draw no conclusions from examining the formal properties of this model concerning the relationship between vowel harmony constraints in affixes and vowel harmony con­ straints in roots. Furthermore, if we consider the formal statement of the P-rule given in (12) we see that a number of condi­ tions must be explicitly stated: (15) i. vowel harmony involves the spreading of a phonetically definable property across cer­ tain domains: this follows from the fact that the feature [+tense] is mentioned twice, once to the right of the arrow and once in the environment; ii. vowel harmony is obligatory (this mode of application is the default case, applying to rules not explicitly designated as "op­ tional"); iii. vowel harmony is bidirectional: this, fol­ lows from the double dash of the environ­ ment; iv. vowel harmony applies in the domain of the word:9 see condition (a); v. vowel harmony obeys the principle of root control, according to which the harmonic category of affixes is determined by the harmonic category of the nearest root vow­ el: see conditions (b) and (c). Now in fact, these properties are not particular to Akan, but characterize all "symmetrical" vowel harmony systems presently known, as well as many "asymmetrical" systems10 (Clements 1977) . Thus, it is incumbent upon an explanato­ ry theory of vowel harmony that these properties should follow from the structure of the theory itself. In a the­ ory of vowel harmony in which these conditions must be separately restated each time they are encountered, one cannot derive the general properties of vowel harmony

68

AFRICAN LINGUISTICS

from an examination of the theory itself, since each lan­ guage may, in principle, vary in its implementation of vowel harmony processes in arbitrary and potentially un­ limited ways. Such a theory, then, does not provide any principled characterization of the notion "vowel harmony"; within such a framework one would be entitled, at most, to speak of a related "family" of phonological processes showing striking, but fundamentally fortuitous affinities. In a rather different point of view, it can be pro­ posed that language learners have available to them an alternative analysis of vowel harmony processes according to which such properties as those enumerated in (i)-(v) are determined by the regular structural characteristics of phonological representation. A strong argument can be made that this alternative analysis is "simpler" (and therefore, more likely to be selected) than rule-based analyses in terms of the sort of criteria that have usu­ ally been invoked in evaluating alternative analyses, roughly involving a calculais of the relative amount of specific information associated with the alternative anal­ yses. From such a point of view, the principles governing vowel harmony are not unique to it, but are motivated on independent grounds as components of general phonological theory. Before proceeding to a discussion of this alterna­ tive treatment, however, it would be useful and instruc­ tive to examine a further description of Akan vowel har­ mony, that of Stewart (1967). Stewart's treatment is of particular interest here in that it attempts to motivate a "prosodic" treatment of vowel harmony within the gener­ al framework of linear models of phonological representa­ tion. In subsequent sections we will examine a rather different "prosodic" model which incorporates some of Stewart's insights, while avoiding some of the problems inherent in strictly linear approaches. 5. The framework assumed in Stewart (1967) is dif­ ferent from that assumed in Schachter and Fromkin's work in one fundamental respect: it allows certain (prosodically-operating) phonetic features to be ordered among fully-specified phonemes in phonological representations. These features serve as "operators" which impose their value upon "spans" of neighboring phonemes, but are not overtly realized (e.g. as independent segments) them­ selves. Stewart's model is inspired by the notion of "su­ pralinear" or "suprasegmental" phoneme developed by Z. Harris and W.E. Weimers in a series of studies in the early and mid 1940's; see Harris 1942, Weimers and Harris

AKAN VOWEL HARMONY

69

1942 and Weimers 1946. Stewart proposes to treat Akan vowel harmony in terms of the prosodic feature H (identified as the feature "ad­ vanced tongue root"), which occurs in phonological repre­ sentations under certain conditions. Vowels are specified as [-advanced tongue root] in underlying representation (termed "hypophonological" or "deep" representation by Stewart). Thus, Asante Twi (the dialect of Akan described in Stewart's study) has only five underlying vowels, /I o ε c a/. When the prosodic feature H is present in a root or word, it establishes a domain or "span" of harmo­ ny across strings of neighboring segments. The extent of this domain is defined by occurrences of the low vowel /a/, which bounds domains, with the provision that an oc­ currence of /a/ on the left of a domain forms part of that domain, while an occurrence of /a/ to the right does not. Otherwise, harmony domains do not contain low vowels. This characterization of harmony domains can be summarized by the following schema, in which X, Y, Z, and W are variables: (16)

# (X a)=Y H Z (a W) #

where Y, Z do not contain # or a. Any vowel occurring in a string corresponding to the un­ derlined portion of the schema is assigned the feature [+advanced tongue root]. Any root may contain at most one occurrence of the prosodie feature H. The placement of H is determined by partly different principles for verb and noun roots. In verb roots, H occurs to the right of the rightmost high vowel, if there is one. In noun roots, H occurs to the right of the rightmost nonlow vowel, if there is one. In both verb and noun roots, H may additionally occur to the immediate left of a C Y a sequence. It follows from these conditions that a root may control H-harmony only if it contains a nonlow vowel (and in addition, in the case of the verb root, at least one high vowel) or a Cya sequence. No other positions are possible. H occurs only in roots. The restrictions may be summarized as follows: (.17X H meets conditions (i) or (ii) , and (iii) : i- Y[+syl -low] H Z (where Z ≠ ... [+syl -low] ...) ii. Y H Cy a Z iii. H is in a root In a verb root, H additionally meets the fol­ lowing condition (Asante dialect only):

AFRICAN LINGUISTICS

70

1 2

The following examples illustrate: (18) underlying surface ml-tIHe mi-tie 'I listen' -soHro o-suro 'he is afraid' -be-karlH -be-kari 'he is going to weigh (it)' I? -bε-Hj-warI oo-be- warl 'he is going to wash' -  t

- w

eH

koto

'parrot' w

e

'knee'

Finally, is qualified s I H k (16). a s i k a in the case ' m o n of e y ' verbs by the statement that /a/ only belongs to the same harmony do­ main as a following nonlow vowel if that vowel is high; thus we have: (19) wa-kc-Hj-warI ? wa-ko-jwarI ?he has gone and washed' This qualification need not be added in the case of those dialects which have [a] in the prefix in such forms (see note 3 ) . To summarize, Stewart's theory of Akan vowel harmony postulates only five underlying vowels, /Ι ο ε ο a/, and a prosodic feature H which is assigned to certain roots according to the conditions summarized in (17) . The no­ tion of domain is independently characterized in terms of a language-particular statement, (16), which determines the assignment of the feature [+advanced tongue root] to all vowels occurring within it. It will be appreciated that this theory provides an economical description of Akan vowel harmony, while accounting for lexical con­ straints upon the occurrence of the feature [+advanced tongue root]. Moreover, it provides an account of forms that were exceptional within Schachter and Fromkin's ac­ count. Forms like [plrako] 'pig' are not exceptions, due to (17i), which uniquely determines the underlying repre­ sentation /plrakoK/, and (16), which uniquely specifies the phonetic form. And forms like [jwanl] 'to flee' are not exceptions, since these are recognized as constitut­ ing a special, idiosyncratic class (cf. (17ii)). On the other hand, this account provides no straight­ forward treatment of disharmonic roots such as those il­ lustrated in (9) and (10). By (17iv) the underlying rep-

71

AKAN VOWEL HARMONY

resentation of 'to become pregnant' must be /ŋIHnssŋ/, and that of the affixed form is accordingly /o-ŋIHnseŋ-I/. Principle (16) would cause all vowels, incorrectly, to acquire the feature [+advanced tongue root]. The root can­ not be assigned a feature marking it as an exception to (16)., since (16) must apply at least to the first root vo­ wel. Nor can the final root vowel / ε / be assigned such a feature, since this would not prevent (16) from applying, incorrectly again, to the suffix vowel. There is, in fact, no obvious way to handle these forms within the context of independently-motivated theories of exceptions in pho­ nology. 3This problem might appear to be of slight impor­ tance in the case of Akan, since the forms cited in (9) are, as mentioned, the only disharmonic roots so far known. However, disharmonic roots are quite common in vo­ wel harmony systems, and if the description of Akan is to form part of a general framework for the description of prosodie processes (including vowel harmony), it must provide a well-motivated account of the behavior of such forms wherever they occur. Another problematical aspect of this account con­ cerns the placement of the feature Η in underlying repre­ sentation. Any theory making use of prosodie features of this sort within the general framework of linear repre­ sentation will be faced with a considerable amount of in­ determinacy with respect to the choice of where they are to be located in linear strings of phonemes. For instance, in the case of Akan the following set of representations would all be consistent with the phonetic shape and pho­ nological behavior of the noun root [kotOj-we] 'knee': (20)

a. b.

Hk t kHt

w

c. d.

 Ht ktH

W

w

e. f. g.

 t Hjwe  t iwHe  t jW£H

w

Given this fact, the rule requiring H to occur to the right of the rightmost nonlow vowel (17i), which uniquely selects (20g) as the underlying representation of 'knee', is a linguistically arbitrary one. The problem here is that the prosodie feature H characterizes the entire mor­ pheme as a unit, rather than a juncture between segments. Thus, once we have decided to abstract a prosodicallyoperating feature such as H, it would be desirable to find a way of expressing this property more directly in our theory of phonological representation, for instance by allowing this feature to be represented as coterminous with the entire phoneme sequence:

72

AFRICAN LINGUISTICS

(21)

In order to do this, of course, it would become necessary to abandon a strictly linear theory of phonological rep­ resentation, and allow certain types of nonlinear repre­ sentation, A proposal to this effect will be examined in more detail in the next section. To the extent that a phonological theory makes avail­ able a formally unconstrained class of language-particu­ lar rules for the description of phonological processes, which may moreover be subject to a variety of essentially arbitrary conditions, it is incapable of expressing crosslinguistic regularities of the sort given in (15). In the framework examined in this section, only the first of the properties of (15) can be made to follow from the theory (by requiring that all prosodic features have consistent, wel-defined properties). The remaining properties do not follow from the theory but follow (if at all) from stip­ ulated conditions on the operation of the rules. In par­ ticular, the principle of root control (15v) does not follow from Stewart's account, since as we have seen this account does not provide for the phonological and phonet­ ic properties of disharmonic roots. If a phonological theory is to succeed in characterizing the general prop­ erties of any process type in language-independent terms, it must do so by providing a set of principles from which all significant cross-linguistic regularities follow as a natural consequence . 6. Let us now consider the analysis of Akan vowel harmony within a framework of nonlinear phonology. The account given here develops an analysis proposed in Clements (1976, 1977) and falls within the general con­ text of what is generally known as autosegmental phonol­ ogy as this has been formulated in Goldsmith (1976, 1979) and further developed, with respect to vowel harmony and related processes, by McCarthy (1977, 1979), Chinchor (1979), Poser (1979), Nash (1980), and others. The autosegmental analysis of any phonological sys­ tem involves a specification of the following, partly in­ dependent parameters, among others:14 (22) i. the class of features (termed P-segments) which are represented on a separate auto­ segmental tier; ii. the class of P-bearing units, that is, u-

AKAN VOWEL HARMONY

73

nits which must be associated with P-segments under the provisions of the wellformedness conditions governing the class of autosegmentalized features in question (here, the class comprising tongue body, tongue root, and lip features); iii. the class (possibly null) of opaque seg­ ments, that is, those segments which are associated with an autosegmentally-represented feature or feature matrix in under­ lying representation. An autosegmental display consists minimally of two tiers, one containing the autosegmentally-represented features or feature matrices (P-segments) and the other containing the segments with which they are associated (P-bearing units). Associations between these two tiers are governed by a set of conventions which preserve well-formedness through phonological derivations. The elements of one tier may be related in a one-to-one or one-to-many fash­ ion with elements of the other tier.. Phonological pro­ cesses may apply to the elements of one tier independent­ ly of the elements of the other; their output, however, is always subject to the conventions governing well-form­ edness. A version of these conventions is given below which incorporates the notion of "priority" developed in Clements (1976).15 These conventions do not require all elements on the autosegmental tier to be associated; some (those that remain unassociated after the application of the conventions) may remain in representations as floating elements. A general interpretive convention is assumed ac­ cording to which "floating" elements in surface represen­ tation are not transferred to the articulatory component, thus remaining "unpronounced". Following the terminology introduced in Clements (197 6), let us call the tier on which P-segments are rep­ resented the P-level, and the tier on which P-bearing units are represented the P-base. Let us further call any maximal string of segments in the P-base containing at least one free (i.e. unassociated) P-bearing unit and no bound P-bearing unit an open string. We shall say that an open string is in the domain of a given P-segment if and only if an association line could be drawn from the Psegment in question to any P-bearing unit occurring in the open string without crossing another association line:16

74

AFRICAN LINGUISTICS

(231

P-level:

P-base: the open strings are bed and f; bcd is in the domain of P, Q, and R; f is in the domain of R, but not of Ρ or Q. We may then formulate the association conventions as fol­ lows: (24) Association Conventions i. Given a continuous string S consisting of one or more free P-segments and an open string Τ occurring in its domain, associ­ ate (free) P-segments in S to (free) Pbearing units in Τ in a one-to-one manner from left to right (such association is nonoverlapping, maximal, and exhaustive in the sense of Clements and Ford (1979)); ii. Given an open string Τ remaining after the operation of (24i), associate each (free) P-bearing unit in Τ with the P-segment in whose domain it falls (giving precedence to the P-segment associated with a P-bear­ ing unit occurring to the left of T ) . These conventions may be illustrated with schematic ex­ amples. Assume a P-level containing the free P-segments P, Q and a P-base containing the free P-bearing units a, b, c, d, as follows: (25)

Convention (24i) requires that Ρ be associated with a and Q with b, as shown below: (26)

AKAN VOWEL HARMONY

75

No further association can take place under the provisions of (24i), since there are no further free P-segments. (24 ii) is now applicable, and requires Q to be associated with  and d: (27)

We have not so far provided motivation for the parenthes­ ized clause in (24ii). This clause is required for the case of examples like the following: (28)

Here,  is an opaque segment, i.e., one containing an un­ derlying association with a P-segment. (24i) is applic­ able first and associates Ρ with a, giving: (29)

Now (24iil must apply. In this case, there are two "neigh­ boring" P-bearing units on either side of the free Pbearing unit. Without the parenthesized clause, (24ii) could be satisfied in either of two ways: either by as­ sociating Ρ and b or by associating Q and jb. The paren­ thesized clause requires the first of these: (30)

It should be noted that the Association Conventions stat­ ed in (24). are not proposed solely for the treatment of vowel harmony, but are independently motivated by the analysis of tone languages, as argued in Clements and Ford (1979). Thus it is a property of the present account of vow­ el harmony that it does not make use of formal mechanisms that are not independently justified in the description of other phonological subsystems.

76

AFRICAN LINGUISTICS

7. The formal specification of the Akan vowel har­ mony system involves three statements. First, the features [+advanced], [-advanced] are autosegmentalized (repre­ sented on the P-level as P-segments). Second, all [+syllabic ] segments are P-bearing units, and only these. Third, low vowels are opaque. Thus the Akan vowel harmony system is fully defined by the statement given in (31): (31) Akan Vowel Harmony i. P-segments: [+advanced], [-advanced] ii. P-bearing units: [+syllabic] iii. opaque segments: [+syllabic, +low] Further simplification is possible. Notice first that since both autosegmentalized features belong to the same feature category, the features named in (31i) can be re­ placed by the single category specification [advanced]. This is not an arbitrary simplification. If we assume that the complexity of any vowel harmony system is a func­ tion of the number of items that must be mentioned in its formal specification, it will follow that a system which selects two features from the same feature category as its P-segments will be simpler than one that selects two features from different feature categories (which cannot therefore be abbreviated, and which will count as two units in the evaluation of the system). This result is supported by the easily-made observation that the predom­ inant number of vowel harmony systems specifying two fea­ tures as their P-segments specify two features of the same category. Secondly, once we have identified the fea­ tures belonging to the category [advanced] as P-segments, it becomes unnecessary to identify the class of P-bearing units, since these are universally vowels for this class of features. More generally, if the class of P-segments is drawn from the primary vowel features (tongue body, tongue root, and lip features) in any autosegmental dis­ play system, the P-bearing units will always be vowels. Thus, the statement given in (31ii) is redundant, and can be eliminated. (This is not, of course, to deny that con­ sonants may exhibit variants determined by the harmonic class of the words in which they occur; the claim is that only vowels function as P-bearing units in the formal sense of this term, unless further specification is given. Furthermore, in some vowel harmony systems, certain vow­ els are neutral; such vowels must be explicitly excluded from the class of P-bearing units in the formal specifi­ cation of the system. The simplest case is therefore that in which all and only vowels function as P-bearing units, and this is precisely the case of Akan.) Given these ob-

AKAN VOWEL HARMONY

77

servations, we may revise the rule determing vowel har­ mony in Akan to the following statement: (32) Akan Vowel Harmony (revised) i. P-segments: [advanced] ii. opaque segments: [+syllabic, +low] (32), as a statement of the grammar, informs us that the features [+advanced], [-advanced] are not present in the phonological matrices defining vowel segments but are rep­ resented on a separate autosegmental tier, and that each occurrence of the low vowel is lexically associated with an occurrence of a member of the feature category [ad­ vanced], which must, moreover, be [-advanced] due to con­ dition (5). Given (32) and the Association Conventions (24),4the main features of the Akan vowel harmony system are determined. Let us consider, now, the lexical representation of Akan roots. It will be apparent that there is a certain amount of indeterminacy in the system. Thus, given a verb root like [fiti], there are several lexical repre­ sentations consistent with the phonetic and phonological behavior of this item. Four are given in (33): (33) a.

.

(capital letters are used here and throughout to repre­ sent vowels that are not specified for the feature cate­ gory [advanced] on the P-base). In (33), each successive display contains fewer elements (P-segments or associa­ tion lines) than the last. A clear criterion is therefore available for selecting among them. Following suggestions in Goldsmith (1976) and Clements (1976), it is proposed that given a set of alternative representations of a given lexical entry, all of which are consistent with its phonetic and phonological behavior, the simplest (in terms of the number of elements constituting it) is se­ lected. On the basis of this criterion, (33d) is uniquely selected as the lexical representation of [fiti].

78

AFRICAN LINGUISTICS

Accordingly, the full lexical representation of the verbal word [o-fiti-i] 'he pierced (it)' is the following: (34)

Here it is assumed that affixes hear no inherent specifi­ cation for the feature category [advanced], but acquire their specification as a result of association under the Association Conventions to the P-segment that governs them. In the present case, (24i) associates the P-segment +A with the leftmost vowel, (35)

and (24ii) associates +A with the remaining vowels: (361

This display provides a unique determination or tne pnonetic form of the word ofitii as far as its vowels are concerned. The derivation of words like [o-clrε-l] 'he showed (it)f proceeds analogously, the only difference being that the P-segment characterizing the root is in this case -A. It follows from these assumptions that words containing no low vowels will be uniquely of one category or the other, with the exception of words containing dis­ harmonic roots, to which we turn below. Let us consider next the treatment of words contain­ ing roots with low vowels. As a consequence of (5) and (32ii), all low vowels are lexically associated with an occurrence of the P-segment -A. The underlying represen­ tations of [c-kasa-I] 'he spoke', [o-bisa-I] 'he asked (it)', and [c-kari-i] 'he weighed (it)' are therefore as follows: (37) a.

AKAN VOWEL HARMONY

79

b.

.

Association Convention (24i) is applicable to (37b) and (37c), and accordingly we have: (38)

a.

(as before)

b.



Finally, (24ii) is applicable to all forms, and we have: (39) a.

b.

c.

The [+advanced] value of the low vowel in [c-kari-i] is assigned by a later rule, to which we return in section 9. Apart from this, the phonetic value of all vowels has been completely determined. Our assumptions guarantee the following two results:

80

AFRICAN LINGUISTICS

in mixed vowel roots such as bisa and kari, neither vowel will assimilate to the other; ii. root vowels control the harmonic category of the immediately flanking affix (the principle of root control). Neither of these results requires special conditions or stipulations; each is a natural consequence of the struc­ ture of the theory. Let us turn finally to the treatment of disharmonie roots /such as η i n s ε p 'to be pregnant'. The lexical spec­ ification of this item is the following:1 (40)

i.

(41)

The word [o-pinsεnl-I] 'she became pregnant' is derived as follows (for the epenthetic vowel, see note 5):

(42)

i.

(24ii) ii. The derivation of this form is a straightforward result of the principles proposed so far. 8. At this point we must return to a consideration of the C y a roots, which govern [+advanced] prefix harmony and [-advanced] suffix harmony. The minimal representation con­ sistent with the phonetic and phonological properties of a form like [iwanl] 'to flee1 is the following: (43)

The low vowel is lexically associated with an occurrence of the P-segment -A, as required by (5) and (32ii). This P-segment governs the harmonic category of suffix vowels. The free P-segment +A governs the harmonic category of prefix vowels, should there be any. If there are none, it will remain unassociated in surface representation, where by the convention mentioned earlier it will be unpronounced. The derivation of [o-jwanI-I] »he fled' is as follows:

AKAN VOWEL HARMONY

(44)

81

i. (UR)

ii. (24i) iii.

(24ii) Most writers on the subject have assumed that the Cya stems had their historical source in forms containing [+advanced] high vowels between the initial consonant and the low vowel, which dropped out after having conditioned the palatalization (occasionally accompanied by rounding) of the consonant. While no synchronic alternations remain to provide a strong source of motivation for such an analysis in the present-day language (though see Schachter and Fromkin 1968 for an attempt to support such an anal­ ysis with language-internal evidence), there is a certain amount of comparative evidence suggesting that it is not implausible as a diachronic analysis, at least for some forms. As an initial case, let us consider the verb root meaning ' to come down'. In some dialects, this form ap­ pears as [sYanI], while in others the initial sibilant has merged with its nonpalatalized congener to give [sanl]. In both dialects, this root contrasts with the root [sanl] meaning 'to untie', which governs [-advanced] prefixes: (45) f come down': dialects A: o-sYanI-I dialects B: o-sanl-I 'untie': dialects A, B: o-sanl-I These are Asante forms. The Akuapem form of the root 'to come down' as recorded by Christaller in 1881 is siao. where the breve indicates "extreme shortness", and Akua­ pem -Q is a frequent cognate of Asante -nI. It is high­ ly unlikely that Christaller wrote the sequence si for a unitary consonant with no internal sequencing, since Christaller was elsewhere careful to note the existence

82

AFRICAN LINGUISTICS

of unitary articulations where he found them. For in­ stance, in describing the sound here transcribed as [cw] as it occurs in Asante, Christaller wrote, "[the lips] are compressed from the outset, at the same time at which the tongue is applied to the palate, and are opened simul­ taneously with the withdrawal of the tongue" (1933, xix) . As it is more likely on phonetic grounds that [sy] devel­ oped from [si] rather than vice-versa, one must take the_ Akuapem form as more closely reflecting the historical an­ tecedent in relevant respects. Rather more substantial evidence is available for ί the forms beginning with labialized consonants. In all di­ alects but Fante, initial rounded velars and laryngeals were palatalized before palatal vowels (Christaller 1933, xiv). In Asante, these consonants became palatalized be­ fore a as well. Some indication as to how this might have taken place is provided by a set of Asante forms recorded by Koelie in 1854, reflecting Asante pronunciation as of the first quarter of the nineteenth century. In the fol­ lowing table, taken from Stewart (1966a), Akuapem forms are given for comparison: (46)

Akuapem egwa ogwao gwarl

Asante

(asI)

(Koelie)

egwia:se ugwia:ne gwia:re

Asante w

e a w

o a w

arl

(modern) (asI)

'market1 'sheep' 'bathe'

If we again take the Akuapem forms as representative of the historical antecedent, we see that Asante developed its rounded palatals by first introducing a front un­ rounded vowel before the low vowel, and then palatalizing the initial consonant while deleting the front vowel. 18 Comparison with Akuapem suggests a similar account of the origin of modern Asante forms in [cw] and [ç w ]. Now, the vowel harmony system of the Akuapem dialect is similar in essential structural respects to that of Asante, suggesting that a common system was inherited from the ancestral dialect. In particular, the low vowel is opaque in Akuapem as well as in Asante. Accordingly, we can attribute at least to the latest common ancestor of Akuapem and Asante forms that closely resemble modern Akuapem forms, which according to the analysis of vowel harmony given earlier must have had representations like the following: (47)

AKAN VOWEL HARMONY

83

Here, we have assumed a full vowel rather than a glide following the initial /g/ of *guarI; in fact, full vowels are recorded in this position in Christaller's transcrip­ tions of Akuapem and Fante forms, as well as in the al­ ternative transcriptions of 'market1 and 'bathe' given by Koelie: zguiaise and guia:re, respectively.19 To such forms the historical processes affecting Asante would have applied, as a result of which the roots became re­ structured without the etymological initial [+advanced] vowel. It will be noted in particular that while the in­ itial vowel of the root became historically irrecoverable, the same would not have been true of the initial P-segment, whose presence continued to be recoverable due to the fact that it continued to govern [+advanced] prefixes. The presence of a large number of roots of this type would have been mutually reinforcing. Thus it can plausibly be assumed that this segment survived the restructuring as a "floating" segment, as postulated in (43), which is asso­ ciated with prefixes in prefixed forms of the verb, but which is unassociated (and therefore unpronounced) in prefixless forms, such as the imperative singular. It should be noticed that the notion of "floating" P-segment, crucial to the above account, is a natural con­ sequence of the formalism of autosegmental phonology. It is strongly motivated in the study of tone languages by the existence of tonal morphemes (Goldsmith 1976) and downstep phenomena (Clements and Ford 1979). In spite of this, alternative accounts might be suggested which do not make use of this notion. Here we explore one such account. This alternative would be premised on the assumption that the initial consonant of a C y a root is opaque (that is, underlyingly associated with a P-segment, in this case +A). We would then have underlying representations like the following: (48)

This assumption would provide correctly for the case of the prefixed forms. In the case of prefixless forms, an additional rule would be needed to delete the initial Psegment (since the consonants in question are not observed to bear the feature [+advanced] in phonetic representa­ tion) . Let us consider how the phonological component of Akan would have to be revised in order to accommodate this proposal. First of all, all palatal consonants would be specified as opaque in the environment _a. This would require the replacement of (32) by the following:

AFRICAN LINGUISTICS

84

(49)

Akan Vowel Harmony (first alternative) i. P-segments: [advanced] ii. opaque segments: a. [+syllabic, +low] b. [+consonental, +high, -back] in the environment: _a.

The environmental condition is required by the fact that palatal consonants are not elsewhere opaque. Secondly, we would need the following MSC: (50) This insures that the opaque consonants specified by (49 iib) have the value [+advanced] in all cases. The problem with this approach, of course, is that (49iib) and (50), which have partly identical conditions, cannot be col­ lapsed into a single statement within the present frame­ work. As a consequence, the same information must be stated twice. In order to overcome this defect it might be proposed that the present framework should be extended in order to allow the rule that specifies opaque segments not only to identify the class of segments in question but also to as­ sign a certain value to such segments, in effect putting (50) in the place of (49iib). However, it should be clear that such a revision would involve a serious weakening of the notion "opaque segment". In the framework developed above, opaque segments are identified in terms of contextfree statements which do not assign them features other than those they bear intrinsically, as a result of the usual Segment Structure Conditions. In the revision under consideration here, virtually any class of segments could function as opaque segments, bearing any feature whatso­ ever, a result that would greatly expand the class of analyses consistent with any given set of data. Unless such a revision could be shown to draw independent justifica­ tion from cases that cannot naturally be handled in the present framework by floating elements, it cannot be maintained as a reasonable alternative. 9. We turn now to a final process of importance in a full account of Akan vowel harmony, that of Vowel Rais­ ing, The set 2 vowels /Ι ο ε ο a/ have the raised variants [Ι ο ε ο a] when the first syllable of the following word begins with a [+high, +advanced] vowel. In this context, set 2 high vowels merge completely with set 1 high vowels:

AKAN VOWEL HARMONY

85

that is, [I o] = [i u] , respectively. However, nonhigh vowels do not merge completely with set 1 vowels, but are acoustically intermediate between the usual norms for set 1 and set 2 vowels. Thus, the following sentences are dis­ tinct in normal speech: (51)

a.

mi-i-bu

b

bi

b.

mi-i-bu buo bi

'I'm breaking a stone' 'I'm breaking a nest'

(ε)bo

'stone'

(e)buo

'nest'

The realization of [a] varies perceptibly according to the nature of the syntactic juncture involved. When the fol­ lowing element is the indefinite article /bi/ or the nom­ inal postposition /mu/ 'in', this vowel is identical in quality to the vowel transcribed with this symbol in wordinternal position; that is, it approaches [ε] in quality. When this vowel occurs in a final verb syllable before a noun object, [a] only sporadically reaches the value it has elsewhere; usually it appears as a slightly raised, but not at all fronted variant of /a/. In the speech of Mr. Mensah, lew vowels are fully fronted in the final syl­ lable of noun subjects preceding verbs; in addition, by Vowel Raising, they are raised to the neighborhood of [e] if the first vowel of the following word is high and ad­ vanced. Some examples of Vowel Raising follow: (52)

a.

b

bay*rε bay!ε no baylr bi .

w w w

c.

' no bi

s

n

bap I bap I no bap I bi

'yam' 'the yam' 'a yam 1 a

k

e 1 'the snake' 'a snake' 'cassava' 'the cassava' 'a cassava'

d . s o no sono no son bi

'elephant' 'the elephant' 'an elephant'

e. ŋŋw a ŋŋw a no ŋŋw bi

'snail' 'the snail' 'a snail'

f.

kof i jI sIkap kof i j· : s i ka

g.

kof i bisa slkap kof i bis s i ka

'Kofi receives a knife' 'Kofi receives money' 'Kofi asks for a knife' 'Kofi asks for money'

86

AFRICAN LINGUISTICS

It should be apparent that the rule described in section 2 determining the occurrence of the low vowel [a] is noth­ ing but a special case of Vowel Raising. Notice that only /a/ is subject to the effect of Vowel Raising word-inter­ nally, since nonlow vowels will always undergo vowel har­ mony; on the other hand, all vowels are subject to Vowel Raising across word boundaries, where vowel harmony is in­ applicable. The correctness of this analysis is confirmed by the fact that all vowels - and not just /a/ - undergo Vowel Raising before a following C y a root: (53)

kwaamI warl sono warl ρεt ε w arl w w arl_ _ ama warI

*Kwame bathes' 'an elephant bathes' 'a vulture bathes' 'a snake bathes' 'Amma bathes'

Thus we need postulate no special rule to account for the raised variant of the low vowel. As first observed by Berry (1957), Vowel Raising is not local to the syllable immediately preceding the con­ ditioning syllable but influences the articulation of pre­ ceding syllables as well, causing them to acquire increas­ ingly raised variants in a gradual "crescendo" as the con­ ditioning syllable is approached. Phonetic "clines" of this sort are not unique to Akan; similar "crescendo" or "decrescendo" effects have been described by other writers and characteristically involve features that function prosodically in at least some languages. Thus, we find "clines" involving nasalization in Guaraní (Lunt 1973), pharyngeal constriction or emphasis in Arabic (Ali and Daniloff 1972), tone in Bukusu (Austen 1974), and stress in Nama Hottentot (Hagman 1977). Three features seem to be common to these systems. First, clines are assimilatory in nature; we do not find cases in which, for instance, vowels become increasingly nasalized before a specified class of oral segments. Second, assimilation is polarized in the direction of the marked value of the feature in question. Third, all clines are phonetically transparent in the sense that no rules of the phonology apply subse­ quently to remove the features that conditioned the cline.21 In view of these characteristics, it would be incor­ rect to describe Vowel Raising in terms of a phonological rule of the usual kind. In particular, we would be unable to explain why it is that Vowel Raising is noncategorical (i.e. assigns nonbinary feature specifications), why its

ÄKAN VOWEL HARMONY

87

domain includes several syllables, and why it is defined on the output of all other phonological rules. An alterna­ tive would be to treat it in terms of the model of speech production proposed in Perkell (1977). In this model, which is analogous to (and could be regarded as a specifi­ cation of) the "phonetic component" of Chomsky and Halle (1968) , the phonetic representations generated by the phonological component are transmitted to the speech motor control system, where they are converted into sensory goal "scores" which determine articulatory movements. The pri­ mary procedure involved in converting motor goals into mo­ tor commands makes "on line" or moment-to-moment adjust­ ments based on the current state of the system. In partic­ ular, anticipatory coarticulation is construed, following Henke (1967), as involving a "look-ahead" procedure de­ fined in terms of a "window" covering approximately four segments at any instant in time. The window scans the "score" from left-to-right along the horizontal (time) ax­ is, in such a way that the segment currently being pro­ duced is on the left. The procedure assigns an "urgency" to each sensory goal, according to its distance from the left of the frame, so that a goal nearer the left is as­ signed a stronger urgency. Such a model offers a natural account of Akan Vowel Raising, a process of anticipatory coarticulation whose domain may extend not just several segments, but several syllables to the left of the conditioning segment. In par­ ticular, it offers a straightforward explanation of all the characteristics noted earlier: the noncategorical, assimilatory nature of Vowel Raising; the fact that the as­ similation is polarized to the "marked" value; the cres­ cendo effect; its multisyllabic domain; and its phonetic transparency. The appropriateness of Perkell's articula-tory model could hardly receive stronger confirmation.22 Returning to the examples of (53), it should be clear that they present only apparent counterevidence to the generalizations offered above. The segment which conditions Vowel Raising in examples like these is not the initial vowel of Jw ari but rather the floating P-segment +A which was postulated on quite independent grounds to account for vowel harmony in prefixes. Under the present analysis, it is no accident that the same class of roots that conditions prefix vowels of set 1 also conditions Vowel Raising in preceding words. On the basis of representations such as (43), our analysis predicts that forms like jwarI will be­ have just like forms beginning with [+advanced] vowels with respect to segments occurring to their left, a pre­ diction that is borne out by two independent processes in Akan.

88

AFRICAN LINGUISTICS

10. In the conception assumed throughout this study, it is a primary task of phonological theory to place a principled limitation upon the number of alternative anal­ yses of any phenomenon available to the language learner, in a way that is consistent with the invariant, recurrent properties of the phonological systems in question; that is, in a way that makes precise general claims about such systems that are (as far as we know) correct. There are two strategies according to which such a limitation can be achieved: first, by drawing a categorical distinction between phonological analyses that are possible in prin­ ciple and those that are impossible in principle (that is, by distinguishing between "entertainable" and "nonentertainable" analyses); and second, by providing an account of the relative complexity of alternative possible anal­ yses, all of which are consistent with the available pri­ mary data, by means of an evaluation metric which in­ duces the selection of the "simplest" of the alternatives. The present account of vowel harmony, like recent nonlinear treatments of other phonological domains, makes available a new class of analyses for certain phenomena. It is important to see that in doing so, it does not in­ troduce undesirable indeterminacy into linguistic analy­ sis, leading to a loss in explanatory adequacy. It is to this question that this final section is devoted. Let us consider the analysis of Akan vowel harmony from the point of view of the language learner. The pri­ mary fact which the language learner must account for is the existence of a regular set of constraints upon vowel cooccurrence in roots coupled with a related and highly regular set of vowel alternations involving the inflec­ tional affixes. Suppose we allow that the choice of the nonlinear analysis is not imposed categorically by our theory, but that the learner may entertain a number of alternative hypotheses (including both linear and nonlin­ ear analyses), selecting the simplest of them in terms of an appraisal of the relative complexity of the various alternatives. A further elaboration of this approach might develop a notion of "weighting" according to which the learner would be predisposed toward one of the various alternatives prior to any appraisal of their relative complexity, but we will disregard this possibility in the present discussion.23 Considering first a linear analysis of the sort con­ sistent with Chomsky and Halle (1968), the initial task facing the learner will be that of assigning a unique un­ derlying representation to each alternating form. At the same time a phonological rule must be devised to specify the following information:

AKAN VOWEL HARMONY

(54)

i. ii. iii. iv.

v. vi. vii. viii. ix. x.

89

the feature(s) which alternate; the class of segments affected; the class of conditioning segments; the class of "irrelevant" segments, that is, those that may intervene between the condi­ tioning and affected segments; further environmental conditions (if any); the domain of the rule; the exceptions to the rule (if any); the directionality of the rule; the optional cr obligatory nature of the rule; the ordering of the rule with respect to all other rules of the phonology.

as well, perhaps, as such factors as the presence of any special diacritic features in the rule, the cyclic vs. the noncyclic nature of the rule, and so forth. For each dif­ ferent hypothesis concerning the nature of underlying re­ presentation, a new rule must be devised, and the re­ sulting solution compared with the others. Needless to say, given the possibility of variation along so many pa­ rameters it will not be a simple task to determine the optimal rule analysis short of much trial and error. Under the alternative, nonlinear analysis, the set of parameters along which choice is possible is substan­ tially reduced. First of all, the learner is not forced to an arbitrary hypothesis concerning underlying repre­ sentations of alternating vowels. Since the alternating vowels have no "basic" alternant, that is, one appearing in an "elsewhere" environment not subject to vowel harmo­ ny, affix vowels are simply not assigned P-segments in their underlying representations; that is, they are un­ specified for a vowel harmony category. Having determined this, the learner must then select along the following parameters: (55)

i. ii.

the feature(s) which are autosegmentally represented; the class of opaque segments.

together with any further rule(s) - in this case, Vowel Raising - which are additionally required to account for final phonetic shapes. The analysis of vowel harmony is then completed. It should be apparent that there are sev­ eral advantages in such an analysis« The question of as­ signing underlying forms to alternating forms is solved in a nonarbitrary way; the amount of further information required as given in (55) is substantially reduced; and

90

AFRICAN LINGUISTICS

finally, the underlying forms arrived at through such an analysis are closely related in derivational terms to the observed surface forms, and more importantly, are directly referrable to observable phonetic properties of utterances. Thus by virtually any metric of "simplicity" or "learnability" one might want to propose, the nonlinear analysis proves to be the more highly valued. There is every rea­ son to believe that even if phonological theory does not exclude

linear

analyses

of

unbounded

processes

like

vowel

harmony in principle, in cases like Akan the language learner will select the nonlinear analysis with little hesitation. If we suppose, then, that such analyses of unbounded phenomena are available to the learner and that they are relatively costless in terms of an evaluation metric, we are in a position to explain several of the properties of the Akan vowel harmony system as consequences of the structure of phonological theory; for example: (56)

i.

ii.

iii.

root harmony is formally identical to af­ fix harmony because it is defined by one and the same rule (32); we have not had to state separate MSC's for roots and P-rules for affixes; vowel harmony is bidirectional and obliga­ tory because it is determined not by Prules but by the Association Conventions (24); vowel harmony applies to the domain of the word because this is the only choice available in the case of "symmetrical" vow­ el harmony systems: the domain is predict­ able from the form of the rule, which "autosegmentalizes" a set of vowel features.

Other properties of Akan vowel harmony need not be stipu­ lated as rule conditions since they follow from the form of lexical representation. Root vowels govern affix V O W Ä els since only the former are assigned to harmonic cate­ gories in underlying representation. C y a roots govern set 1 (root-advanced) vowel prefixes due to the presence of "floating" P-segments in their lexical representations; these segments, although complicating the lexical entry in formal terms, are easily recoverable by the learner due to their influence on preceding vowels in two independent processes. From these observations we are led to a conclusion which has not, I think, been generally appreciated in re­ cent discussion. If nonlinear phonology has "opened the

AKAN VOWEL HARMONY

91

formal floodgates" (Goldsmith 1979) in allowing a rich and novel set of analyses involving new types of rules and representations, it has not thereby weakened the explana­ tory power of linguistic theory. The introduction of a nonlinear model need not lead to any increase in indeter­ minacy in linguistic analysis. Quite to the contrary: by virtue of the extreme simplicity of the newly-admitted class of analyses considered here, we have seen that in­ and linguistic'theory determinacy is effectively reduced, strengthened in proportion - a result that can only be welcomed.

APPENDIX

In a recent study, Stewart (1982) revises his earlier analysis of Akan vowel harmony, recasting it in a modified version of the Schachter-Fromkin framework. Stewart claims that vowel harmony in the Asante dialect is less general than previous descriptions have supposed, in that nonhigh vowels may freely initiate new harmony spans within the word: compare Asante o-tiε-I 'he listened', o-tie o-tie» The distinction 'he listens' with the cognate Akuapem o-tie-i, here is attributed to an Asante innovation. This evidence does not, however, force a revision of the account of Asante vowel harmony given here. There is no reason to suppose that the Asante root 'listen' has been restructured as /tie/; rather, we may suppose that the unsuffixed form /tie/ is still taken as basic, and that a phonological rule specific to the dialect (or subdialect) in question accounts for the disharmonic sequence of the suffixed form. To show that nonhigh vowels are not constrained to harmonize with preceding vowels in the word it would be necessary to show that there are lexical contrasts like /tie/ versus the hypothet­ ical /tie/. As far as I know, no evidence for such contrasts exists apart from the two forms piŋcε and ŋiŋsεŋ cited earlier; indeed, if these two forms are to be analyzed as compounds, as Stewart argues, then even these exceptions disappear. Finally, the nominal suffix -0/- always harmonizes to the preceding root vowel: beside e-bu-o 'nest' and ε-bo-c 'stone' we find no forms like the hypothetical *e-bu-ε. These facts seem difficult to reconcile with Stewart's new analysis, but are fully consistent with the analysis presented above.

92

AFRICAN LINGUISTICS

ACKNOWLEDGEMENT S

Earlier versions of portions of this paper were read at the LSA Slimmer Meeting, Amherst, 1974, at the CUNY Conference on Vowel Harmo­ ny, New York, 1977 and at the Workshop on Harmony Processes, M.I.T., 1981. I would particularly like to thank J.M. Stewart for reading and criticizing an earlier version of this manuscript. Research for this study was supported in part by a grant from the Harvard Graduate So­ ciety for Advanced Study and Research.

NOTES

1. The fact that the features "advanced tongue root" and "tense" are not known to function independently of each other in any known vowel system suggests that they may involve variant implementations of a single, more abstract phonological feature. 2. Note that /ç/ does not occur before /a/, /c/ occurs before /a/ only in the noun /a-caca/ 'hunchback', where it does not condition the [a] variant in either syllable, /p / occurs before /a/ only in the noun /ŋ-ŋ am/ 'fontanel', from which we can induce no conclusions about its influence on prefix vowels. 3. Some Asante dialects are reported to have [wa-be-tu], [wa-ko-tu]; apparently, such dialects have [a] before mid advanced vowels as well as high advanced vowels, in verbs as well as in nouns. Mr. Poku-Appiah and Mr. Mensah consistently used the nonadvanced vowel [a] in these and similar examples. 4. The complementary distribution of [a] and [a] was noted by Stewart (1967, 187), who points out that they constitute distinct phonemes "in the Bloomfieldian sense" on the basis of such minimal pairs as ntam 'oath' and ntam 'between' (fn.2); underlyingly, these forms are represented /η-tam/ and /n-ta mu/, where /mu/ in the latter is the postposition 'inside'. Stewart does not state explicitly whether he considers [a] and [a] as reflexes of a single phoneme in underlying representation, although on the basis of his discussion there is eve­ ry reason to do so.

AKAN VOWEL HARMONY

93

5. Consonant-final roots have vowel-final alternants when suffixed; the quality of this vowel is phonoloaically predictable (Schachter and Fromkin 1968, 184). I assume, following Schachter and Fromkin (but for another view see Stewart 1976, 100) that the final root vow­ el in such forms is epenthetic; this assumption is not, however, cru­ cial to the present analysis. 6. "For final specification, all [+tense] vowels can be interpreted as vowels in which the tongue root is advanced, in contrast with their [-tense] counterparts where the tongue root is not advanced" (58). 7. The environment of this rule is interpreted as a schema abbrevi­ ating the two contexts _X [+voc, etc.] and [+voc, etc.] Y . 8. Even so, they are unable to derive surface forms such as [obejw әnI] 'he will flee' (p. 194). Their rule system gives the fol­ lowing derivation: /c-bε-guagI/ Underlying representation c-bε-guaQÏ

Nasalization (P06)

c-bε-guiaI

I-insertion (P07)

W

c-bε-g uiaQI

Rounding (P09)

c-bε-gWiaI

u-deletion (P10)

c-bε-j W aŋ o-bε-;tWaŋ I

I

Palatalization (Pll) I-deletion (P13) Vowel Tensing (P69): n/a

[Ŋ] => [n] (P70) *c-be-jwanï First, no rule in their system will tense [a] to []. Presumably, this problem could be solved by replacing /a/ with // in the under­ lying representation. Second, the rule of I-deletion bleeds Vowel Tensing, for two independent reasons: (i) I-deletion is ordered be­ fore Vowel Tensing; (ii) rules are assumed to apply cyclically. Schachter and Fromkin take note of the second problem (p. 98) but not the first. 9. We return to a discussion of what Schachter and Fromkin term "Vowel Tensing Across Word Boundaries" below. 10. For those in be taken evidence

present purposes we may take "symmetrical" systems to be which one alternant of regularly alternating affixes cannot as more "basic" than another on the basis of distributional alone, as in Akan, Finnish, and Turkish.

11. The verb roots /tene/ 'to spread' and /hojo/ 'to loosen' given in (4b) are the only exceptions to this statement that I have come across. 12. Stewart's examples are cited in the transcription used elsewhere in this text, according to which the symbols ky, kwy, gy, gwy, hwy,

94

AFRICAN LINGUISTICS

ny, swy, and sy are replaced by their phonetic equivalents /,  , j jw Çw f ŋ swy s Y respectively. 13. For one such account see Zonneveld (1978). I know of no motiva­ tion for considering the forms cited in (9) to be compounds, at least in the contemporary language. 14. See Clements (1976) for further discussion. In a somewhat dif­ ferent approach, Halle and Vergnaud (1978) have introduced notions of metrical phonology into their account of vowel harmony. This ap­ proach differs from the present one primarily in allowing directional (left-branching and right-branching) trees as well as nondirectional trees in representations, in not recognizing unassociated or "float­ ing" P-segments either in underlying or derived representation, and in adopting an alternative version of the conventions governing wellformedness. As this work is currently undergoing amplification and revision, it would be premature to offer extensive discussion of it here; see Halle and Vergnaud (1982) for a more recent statement. 15. See Clements and Ford (1979) for a slightly different version of these conventions. The present version is somewhat simpler, while adequate to the description of the range of phenomena that motivated the earlier statement. 16. We may correspondingly say that a P-bearing unit occurring in an open string is governed by the P-segment in whose domain it oc­ curs, and which is the unique P-segment that can be associated with it under the Association Conventions. 17. This display could be simplified further by eliminating one or the other (but not both) of the association lines. 18. Compare the synchronic derivation offered by Schachter and Fromkin, summarized in note 8. 19. The reader should not be misled by the apparent violations of vowel harmony in Koelie's transcriptions; e was probably used to transcribe [I], to which it is acoustically similar. 20. In a stricter transcription than that used here, vowel nasaliza­ tion would be indicated after nasal stops where it is occasionally distinctive. In (52-3) the final vowels of 'snail' and 'Amma' are oral; elsewhere in the examples in this study, vowels are nasal after nasal consonants, except for mid vowels which never have per­ ceptually salient nasalization in this position. 21. The data described in Austen (1974) are problematical for these generalizations. In some examples (but not all), the domains across which low-toned syllables acquire increasingly high pitch are flanked on the right by low-toned syllables. "Updrift" of this sort, which is known in other tone languages, is nonassimilatory in nature and may constitute a special case of "clines" whose explanation is other than that suggested below.

95

AKAN VOWEL HARMONY

22. An alternative approach to the description of assimilatory coarticulation, based on descriptions of nasalization in Guaraní, is reported in Halle and Vergnaud (1978). While this system makes many of the same predictions as that proposed by Perkell, it is somewhat richer to the extent that it can be used to describe lines that are phonetically opaque - that is, clines that are conditioned by morpho­ logical features, or by phonological features that are eliminated from the representation by later rules. It needs to be specified, furthermore, how this approach would account for the phonetic dis­ tinction between the vowel sequence [c...u] in examples like [ kumasI] 'go to Kumase' and the phonetically distinct vowel sequence [ o — u ] in [ko-fupanI] 'go and search'. 23. For example, Goldsmith has suggested that segmentation into phonemes is a hypothesis that the learner arrives at only after passing through a series of intermediate stages in which the autosegmental index is larger that 1 (Goldsmith 1976, chapter 4). In this view, adult autosegmental systems arise from a language acquisition path in which the autosegmental index is never forced as low as 1. It would follow that the autosegmental analysis is accessible to the language learner long before a linear analysis is available.

REFERENCES

Ali, L.A. and R.G. Daniloff 1972 "A Cinefluorographic-Phonologic Investigation of Emphatic Sound Assimilation in Arabic," in A. Rigault and R. Charbonneau, eds., Proceedings of the 7th International Conference of Phonetic Sciences, Mouton, The Hague. 639-45. Austen, Cheryl L. 1974 "Anatomy of the Tonal System of a Bantu Language," in Papers from the Fifth Annual Conference on African Linguistics (Studies in African Linguistics, Supplement 5), 21-33. Berry, Jack 1957 "Vowel Harmony in Twi," Bulletin of the School of and African Studies 19(1), 124-30.

Oriental

Chinchor, Nancy 1979 "On the Treatment of Mongolian Vowel Harmony," in E. Battistella, ed., NELS 9 (CUNYFORUM Nos. 5-6, Fall 1978/Spring 1979),

96

AFRICAN LINGUISTICS 171-86.

Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English,

Harper and Row, N.Y.

Christaller, the Rev. J.G. 1933 Dictionary

of the Asante and Fante Language called

Tshi (Twi).

1st ed. 1881. 2nd. ed., revised and enlarged, 1933. Basel, Basel Evangelical Missionary Society. Clements, G.N. 1976 "Vowel Harmony in Nonlinear Generative Phonology," unpublished ms., Harvard University. 1977 "The Autosegmental Treatment of Vowel Harmony," in W.U. Dressier and O.E. Pfeiffer, eds., Phonologica 1976:'Akten der dritten Internationalen September 1976 ( Innsbrucker

Phonologie-g, Wien, 1-4 Beiträge zur Sprachwissenschaft,

19), 111-9. Clements, G.N. and K.C. Ford 1979 "Kikuyu Tone Shift and its Synchronic Consequences," tic Inquiry 10(2), 179-210.

Linguis­

Dolphyne, Florence 1965 The Phonetics

and Phonology of the Verbal Piece in the Asante

Dialect of Twi, unpublished Ph.D. dissertation, University of London. Goldsmith, John 1976 Autosegmental Phonology. by Garland Press, 1979. 1979

Ph.D. dissertation, M.I.T., published

"The Aims of Autosegmental Phonology," in D. Dinnsen, ed., Current Approaches to Phonological Theory, Bloomington, In­ diana University Press, 202-22.

Hagman, Roy S. 1977

Nama Hottentot

Grammar (Language Science Monographs, 15),

Bloomington, Indiana University Publications. Halle, Morris and J.-R. Vergnaud 1978 "Metric Structure in Phonology," unpublished ms., M.I.T. 1982

"Harmony Processes," in W. Klein and W. Levelt, eds., Crossing the Boundaries in Linguistics,

Reidel, Dordrecht,

1-23. Harris, Z.S. 1942 The Phonemes of Moroccan Arabic," JAOS 62(4), 309-18. Henke, William 1967 "Preliminaries to Speech Synthesis based on an Articulatory Model," Proceedings of the 1967 I.E.E.E. Boston Speech Con­ ference, 170-77.

AKAN VOWEL HARMONY

97

Koelie, S.W. 1854 Polyglotta

Africana.

London.

Lindau, Mona 1975

[Features]

for Vowels (U.C.L.Ά. Working Papers in

Phonetics,

30), December, 1975. Lunt, Horace 1973 "Remarks on Nasality: the Case of Guaraní," in S.R. Anderson and P. Kiparsky, eds., A Festschrift for Morris Halle, Holt, Rinehart, and Winston, Inc., New York, 131-9. McCarthy, John 1977 "An Autosegmental Account of Classical Arabic Vocalism," un­ published ms., M.I.T. 1979

Formal Problems in Semitic Phonology and Morphology, un­ published Ph.D. "dissertation, M.I.T.

Nash, David 1980 Topics in Warlpiri tation.

Grammar, unpublished M.I.T. Ph.D. disser­

Painter, Colin 1973 "Cineradiographic Data on the Feature 'Covered' in Twi Vowel Harmony," Phonetica 28, 97-120. Perkell, Joseph 1977 "Articulatory Modeling, Phonetic Features and Speech Produc­ tion Strategies," in G.N. Clements, ed., Harvard Studies in Phonology, vol. 1, 349-68. Poser, William 1979 "Nasal Contour Consonants and the Concept of Segment in Pho­ nological Theory," B.A. Honors Thesis, Harvard university. Schachter, Paul and Victoria Fromkin 1968

A Phonology of Akan: ~2ikuapern, Asante and Fante Working Papers in Phonetics, 9 ) , August 1968.

Stewart, John 1966a "Asante Twi in the Polyglotta Africana," in Sierra

(U.C.L.A.

Leone

Language Review 5, 111-5. 1966b Comparative African

and Krachi, Ghana.

Wordlists

No.l:

Awutu, Larteh,

Nkonya,

Institute of African Studies, University of

1967

"Tongue Root Position in Akan Vowel Harmony," Phonetica 16, 185-204.

1976

"The Final Light Syllables of Akan (Twi-Fante) and their Significance for Volta-Comoe Reconstruction," in H.M. Trutenau, ed. , Languages of the Akan Area: Papers on Western Kwa Lin­ guistics and on the Linguistic Geography of the Area of An-

AFRICAN LINGUISTICS

98

cient Begho. Transactions of the Linguistic Circle of Accra, 3: Mitteilungen der Basler Afrika Bibliographien 14, 93-156. Stewart, J.M. 1982 "Akan Vowel Harmony: the Word Structure Conditions," manu­ script, Rijksuniversiteit te Leiden. A Dutch translation of this paper ("Vokaälharmonie in het Akan") appeared in GLOT Vol. 5(3), 321-342. Weimers, W.E. 1946 Ά Descriptive Grammar of Fanti supplement to Language 22(3).

(Language Dissertation

no. 39),

Weimers, W.E. and Z.S. Harris 1942 "The Phonemes of Fanti," JAOS 62(4), 318-33. Wood, Sidney 1975 "Tense and Lax Vowels - Degree of Constriction or Pharyngeal Volume ?" Working Papers of the Lund University Phonetics Laboratory

11, 109-33.

Zonneveld, Wim 1978 A Formal Theory of Exceptions in Generative Phonology (Studies in Generative Grammarr 1 ) , Foris Publications, DordrechtHolland.

TONS ET ACCENTS EN KINYARWANDA R. Furere (UNAZA, Kinshasa) et A. Rialland (CNRS, ERA 433)

Nous partirons ici de données phonétiques1 fournies par des courbes mélodiques calibrées au quart de ton, pour déterminer d'abord les réalisations des tons en divers contextes et dans différentes positions dans l'énoncé (initiale ou finale). Ces principes de réalisation étant re­ connus, nous pourrons ensuite dégager les schemes tonals phonologiques des noms (non composés) et des verbes (en nous limitant à certains temps et à certains modes). A partir de là, nous verrons comment ceux-ci sont les mani­ festations d'un "accent tonal" et comment le passage par ces différents niveaux d'analyse était nécessaire pour le mettre en évidence. Ce faisant, nous donnerons progressivement une image du système tonal du kinyarwanda assez différente de celle à laquelle ont abouti les auteurs précédents et nous ferons apparaître un système combinant rythme et mélodie d'une façon inédite. 1. LES TONS ET LEUR REALISATION EN FONCTION DES CONTEXTES TONALS ET DE LEUR POSITION DANS L'ENONCE 2.2.

Deux tons

phonologiques

Nous décrirons l'ensemble du système du kinyarwanda avec deux tons: haut et bas et montrerons comment les ré­ alisations constatées sont celles de ces deux tons en dif-

AFRICAN LINGUISTICS

férents contextes tonals ou dans différentes positions dans l'énoncé. 1 .2.

Un ton

par

more

Nous verrons qu'il y a un ton par more, c'est à dire un ton par voyelle. Les voyelles longues seront décom­ posées en deux mores. La démonstration en sera faite au cours de notre étude des schemes tonals nominaux où nous verrons l'importance de la more en tant qu'unité rythmi­ que . 1.3.

Principales

variantes

de réalisation

des

tons

Les principales variantes de réalisation des tons ap­ paraissent dans certains contextes tonals ou dans certaines positions dans l'énoncé, début ou finale; elles consti­ tuent alors des marques de début ou de finale d'énoncé. Nos exemples seront souvent ceux des auteurs précé­ dents, en particulier ceux de Kimenyi (1976) . 1.3.3. La réalisation d'énoncé

de

tons

successifs

à

l'initiale

Deux tons hauts successifs sont réalisés en "crescen­ do" lorsqu'ils se trouvent à l'initiale d'énoncé, exemples :

TONS ET ACCENTS EN KINYARWANDA

101

Figure 1, nous voyons que les mores ma et a initiales portent une mélodie montante. Celle-ci commence à un niveau plus haut qu'elle ne le ferait pour un ton bas. Cette mélodie montante est la réalisation de deux tons hauts successifs et initiaux, comme permettra de le comprendre l'examen de la réalisa­ tion de ce même mot en position non initiale. Figure 2, ce même mot se trouve cette fois précédé d'une syllabe porteuse d'un ton haut. Le crescendo se reforme entre les deux mores initiales porteuses de ton haut que sont na et ma. Figure 3, c'est le même syntagme /ni / "avec ma mère" qui à son tour, est précédé de syllabes de tons hauts. Ses trois premiers tons hauts n'étant plus initiaux sont réalisés sur la même hauteur. Ces exemples permettent de montrer que les tons hauts initiaux connaissent une réalisation "en crescendo". Le premier ton haut réalisé un peu plus bas ne doit pas être confondu avec un ton bas. Phonologiquement, il reste haut et le rabaissement de sa réalisation constitue une marque de début d'énoncé. Diverses interprétations en termes tonals ont été pro­ posées pour ces réalisations. Ainsi, /maama/ a-t-il été noté: /maama/ par Coupez (1961), Overdulve (1975) et Meeussen (1959) en kirundi, le ton haut désignant pour ces auteurs un "ton haut antérieur", c'est à dire dont la réalisation est anticipée sur la voyelle précédente "qui porte de la hauteur même si elle a le ton distinctif bas" (Coupez et Meeussen, 1961) mamà par Kagame (1960) où désigne un ton haut dou­ ble^ [màama] par Kimenyi (1976) niveau phonétique et /maama/

2

AFRICAN LINGUISTICS

niveau phonologique, dans le sens générativiste /maamâ/ par Kayoboke (1980) Un seul auteur place un ton haut à l'initiale: Kagame. Tous les autres placent un ton bas, quel que soit le ni­ veau de leur analyse. Nous examinerons ultérieurement (en 1.3.5.) la ques­ tion du ton final de // et nous expliquerons pourquoi nous avons posé un ton haut phonologique. Le rabaissement du ton haut initial se rencontre dans d'autres langues bantoues: il est par exemple signalé en haya (Byarushenao, Hyman et Tenenbaum, 1976). 1.3.2. La réalisation ton bas

de

tons

hauts

successifs

après

un

Après un ton bas, on retrouve un crescendo comparable a celui de l'initiale, exemples:

TONS ET ACCENTS EN KINYARWANDA

103

Ces figures 4 et 5 montrent clairement le crescendo des tons hauts successifs après un ton bas. On voit d'ailleurs que, comme à l'initiale, le premier ton haut n'est pas identique à un ton bas mais qu'il est réalisé un peu plus haut que lui, sans atteindre toutefois, la hauteur du ton haut que le suit. 1.3.4.

Le rabaissement d'un ton bas en finale d'énoncé Celui-ci sera réalisé un peu plus bas que le ton bas qui le précède, exemple:

"jeter un coup d'oeil" "regarder rapidement" Cette figure montre bien que le ton bas en finale d'une séquence de tons bas est réalisé plus bas que les tons bas qui le précèdent. Il s'agit également d'une mar­ que de finale d'énoncé. 1.3.5. La réalisation descendante d'un ton haut en finale d ' énoncé Tout ton haut en finale de mot est réalisé descendant lorsque ce mot se trouve en finale d'énoncé, exemples :

104

AFRICAN LINGUISTICS

figure 8

Le ton haut de ce monosyllabe est réalisé descendant lorsqu'il est en finale d'énoncé (figure 7), modulation qu'il ne présente pas lorsqu'il est dans d'autres posi­ tions, par exemple dans le contexte où il apparaît, figure 8. Lorsque le ton haut est lui-même précédé d'un ton haut, la configuration de la descente mélodique est vari­ able. Elle peut commencer à un niveau haut comme dans // (figures 1, 2 et 3), le dernier ton de ce mot pouvant être posé comme haut par référence à sa réali­ sation haute en position non finale illustrée par la fi­ gure suivante :

TONS ET ACCENTS EN KINYARWANDA

105

figure 9

Elle peut commencer à un niveau un peu plus bas comme dans /ïgitl/ (figure 14) , exemple que nous verrons ulté­ rieurement mais, il n'y a cependant pas confusion avec la séquence tonale haut-bas dont le ton haut préfinal se ré­ alise descendant lorsque celle-ci se trouve en finale d'é­ noncé comme nous allons le voir immédiatement. Ces réalisations descendantes ou fortement rabaissées du ton haut final se rencontrent dans de nombreuses langues bantoues comme le signale Hyman (1978). 1.3.6. devant

La réalisation descendante du ton haut préfinal un ton bas en finale d'énoncé Le ton haut en position préfinale dans un mot, c'est à dire précédant un ton bas final est réalisé descendant lorsque le mot se trouve en finale d'énoncé, exemples:

1Q6

AFRICAN LINGUISTICS

f i g u r e 11

Ce mot: /umusore/ présente un ton haut en position préfinale devant un ton bas. En finale d'énoncé, le ton haut préfinal: celui de la syllabe so est réalisé descen­ dant (figure 10). A l'intérieur d'énoncé, il est de hau­ teur haute, sans modulation et se trouve être le point culminant du mot, du fait de la réalisation en crescendo de tons hauts suivant un ton bas vue précédemment· Ces réalisations ont donné lieu à diverses interpré­ tations et ce mot /umúsore/ a été notéî /umusore/ par Coupez (1961), Overdulve (1975), le ton haut désignant un ton haut antérieur; umusore par Kagame (1960) où "désigne un ton moyen [umúsore] par Kimenyi (1976) au niveau phonétique /umusóre/ au niveau phonologique /umusôre/ par Kayoboke (1980) Lés deux tons hauts successifs, celui de mû et celui de so n'ont pas été reconnus comme tels à travers la vari­ ation de leur réalisation en fonction des contextes. Il faut noter que le ton haut préfinal porte parfois une courbe mélodique descendante moins nette et est réa­ lisé sur une hauteur intermédiaire entre la syllabe de ton haut précédente et celle de ton bas terminale, ce qui peut expliquer le ton moyen noté par deux points, posé par Kagame sur la syllabe so. Il s'agit encore d'une marque de finale d'énoncé qu'on retrouve dans d'autres langues bantoues, en haya, par exemple (Byarushengo, Hyman et Tenenbaum, 1976). Dans notre analyse des schemes tonals nominaux et ver­ baux, nous présenterons des courbes mélodiques de réalisa­ tion des différents schemes et nous verrons s'appliquer ces principes de réalisation tonale. Nous verrons égale­ ment, ce qu'on peut déjà entrevoir, comment leur recon-

107

TONS ET ACCENTS EN KINYARWANDA

naissance nous a amenés à une transcription phonologique différente de celle des auteurs précédents.

SCHEMES TONALS NOMINAUX ET ACCENT Nous partirons des courbes mélodiques des différents schemes et montrerons selon quelle démarche nous passons du niveau phonétique au niveau phonologique. Nous verrons ensuite comment les schemes tonals dégagés sont les mani­ festations d'un accent. 2.1. 2.1.1,

Les schemes tonals Les

schèmes

tonals

des

noms à thème

d'une

more

2.1.1.1. Les schemes tonals des noms à thème d'une more de ton bas Ceux-ci ne comportent que des tons bas. exemples: ïfù "de la farine àgàfù "un peu de farine" thème: f  thème: fu ino "orteil" ùmùzï "racine" thème: no thème: zi ïbya "testicules" ùmùryà "corde d'un in­ thème: à strument de muìgìhù "nuage" sique" thème: hù thème: ryà ìkìvì "mesures de terre pour le labeur" thème: vï Dans ces noms, 1'augment comme le préfixe précédant le thème sont de ton bas. 2.1.1.2. Les schemes tonals des noms à thème d'une more de ton haut Nous distinguerons ici les noms qui comportent un préfixe mais pas d'augment, ceux qui comportent un pré­ fixe et un augment, enfin ceux qui comportent un augment et un préfixe nasal ou un augment mais pas de préfixe syllabique. -Noms avec préfixe mais sans augment exemples

108

AFRICAN LINGUISTICS

figure

12

f i g u r e 13

Le ton haut final de /kúbá/ est realise descendant en fi­ nale d'énoncé (figure 12). En position non finale,3 il perd cette modulation et lorsque le mot /kúbá/ se trouve à l'initiale, il devient le sommet du crescendo formé par les deux tons hauts successifs (figure 13). Nous posons donc deux tons hauts sur cette forme et non un seul comme Kimenyi (1976) ou Overdulve (1975). -Noms avec préfixe et augment exemples:

Le ton haut final est réalisé descendant en finale d'énoncé (figure 14). La descente mélodique ne commence pas ici au niveau haut de la syllabe précédente mais très nettement plus bas. Nous avons déjà signalé qu'il ne peut y avoir de confusion avec le scheme haut-bas dans la même position. Lorsque ce mot /ïgitì/ est en position non fi-

TONS ET ACCENTS EN KINYARWANDA

109

nale, nous retrouvons la réalisation en crescendo des tons hauts suivant un ton bas et la syllabe [ti] porte alors le sommet mélodique du mot. Ce mot est constitué du thème -ti, du préfixe gri­ et de 1'augment i-. Le préfixe prend donc le même ton que le thème mais 1'augment présente un ton bas. -Noms avec augment sans préfixe syllabique C'est à dire sans préfixe du tout ou avec un préfixe nasal qui ne forme pas syllabe Cet augment reçoit un ton identique à celui du thème, à savoir un ton haut. exemples: /ìfì/ "poisson" / ihó / "asticot" thème: fi thème: ho /iki/ "été" /it šó/ "tache" thème: i thème: ts¿ /ihwá/ "épine" /isi/ "la Terre" thème: hwá thème: si /imhû/ "peaux" thème: hú /agáfi/ "petit poisson" /ùrúhú/ "peau" thème: fi thème: hú /ï g i hwá/ "grande /àgátšá/ "épervier" épine" thème: hwá thème: tsâ Nous pouvons donc déjà noter que le préfixe et 1'aug­ ment (en l'absence de préfixe syllabique) ont un ton in­ duit du ton du thème. D'autre part, 1'augment précédant un préfixe syllabique porte toujours un ton bas. 2.1.2.

Les schemes tonals des noms à thème de deux mores 2.1.2.1. Les schemes tonals des noms dont le thème est de scheme bas-bas Dans ces formes, le préfixe est de ton bas ainsi que 1'augment. Elles ne comportent donc que des tons bas. exemples: /ïžosï/ "cou" /ïňònï/ "oiseau" thème: zbsï thème: ňo ni /umusoro/ "impôt" /îkïberò/ "cuisse" thème: sòr ò thème: ber ò /umùgabò/ "homme" /ùmùglna/ "butte" thème: gabb thème: glna 2.1.2.2. Les schèmes tonals des noms dont le thème est de schème haut-bas exemple :

110

AFRICAN LINGUISTICS

Le ton haut préfinal de /ukuboko/ est réalisé descen­ dant en finale d'énoncé (figure 16). Dans les autres posi­ tions, il est réalisé sans modulation et un crescendo se forme entre les syllabes [ku] et [bo], toutes les deux porteuses de ton haut et suivant un ton bas. En kirundi, Meeussen (1959) étudiant le même mot fait remarquer que celui-ci peut recevoir plusieurs réalisa­ tions phonétiques. Les mêmes variations nous ont semblé difficilement acceptables en kinyarwanda. Le mot /ukuboko/ est composé du thème -bókò, du pré­ fixe ku- et de 1'augment u. Comme dans les noms à thème monosyllabique de ton haut, le préfixe reçoit un ton haut et 1'augment qui le précède un ton bas. D'autre part, en l'absence de préfixe syllabique, 1'augment porte un ton haut, exemples :

TONS ET ACCENTS EN KINYARWANDA

111

/íňánà/ ^ "veau" thème: nana /ísúka/ "houe" thème: súka

/ísónl/ "honte" thème: sonï /inóni/ "claquement de doigt" thème: nóni /ïkîyége/ "sorte de /umútima/ "coeur" champignon blanc" thème: yége thème: tima. /umútsima/ "pâte" /àmá hórò/ "paix" (de sorgho ou millet) thème: tsima thème: hórò 2.1.2.3. Schèmes tonals des noms dont le thème est de schème haut-haut exemples :

En finale d'énoncé, le ton haut de la syllabe finale re est réalisé descendant (figure 16). En position non fi­ nale, il est réalisé haut et constitue le deuxième palier du crescendo formé par les deux tons hauts successifs des syllabes [go] et [re] après le ton bas de la syllabe [mu]. Le mot /umugore/ est composé du thème -gore, du pré­ fixe mu- et de 1'augment u. Nous constatons que cette fois-ci, le préfixe ne prend pas le ton haut. Nous verrons en étudiant les noms à thème de trois ou quatre mores qu'il en est toujours ainsi lorsque le thème com­ porte un nombre pair de tons hauts. L'augment, qu'il pré­ cède immédiatement le thème ou qu'i] soit placé devant un préfixe syllabique, est toujours de ton bas. exemples: /ibaba/ ^ "aile" /inkúbá/ "foudre" thème: baba thème: kubi /ìnkòkó/ "poule" /ïnkakà/ "sexe du tau­ reau"

112

AFRICAN LINGUISTICS

thème: koko thème: káká /îtémé/ "traverse, /umukéné/ "pauvre" pont" thème: terne thème: kene /ùmùseke/ "sorte de roseau" thème: séké /ìgìtemé/ "coupe" (terrain dont on a coupé les arbres en vue de le dé­ fricher) Il existe donc trois types de thèmes de deux mores, les thèmes de schème bas-bas, haut-bas et haut-haut. Le schème bas-haut est exclu. Nous comprendrons plus tard d'où vient cette limitation. 2.1.3.

Les schemes tonals des noms à thème de trois mores 2.1.3.1. Les schèmes tonals des noms dont le thème est de schème bas-bas-bas Ces formes ne comportent que des tons bas comme leurs homologues monosyllabiques ou disyllabiques exemples: /umîihararb/ "sympathie /ùmùybbbrb/ "canal" de courte durée" thème: haràrb thème: yòbòrò /umùhirïtsi/ "celui /ùmwààmbi/ "flèche" qui pousse" thème: hìrìtsì thème: ààmbì 2.1.3.2. Les schèmes tonals des noms dont le thème est de schème haut-bas-bas exemples :

TONS ET ACCENTS EN KINYARWANDA

113

La modulation qui affecte les mores zu et u et qui cor­ respond à la réalisation de la séquence tonale haut-bas est toujours présente quelle que soit la position du mot dans l'énoncé (figures 20, 21 et 22). Le ton initial est rabaissé en début d'énoncé (figure 20) mais est réalisé haut dans le contexte: /kubon izûuru/ où il peut s'opposer à un ton bas initial de mot (voir la réalisation de /itáábi/ dans le même contexte, figure 25). Le mot /izûùru/ se compose d'un thème: zúùrù et d'un augment i qui en l'absence de préfixe syllabique, reçoit un ton haut. Dans les noms formés sur des thèmes de ce scheme, le préfixe syllabique porte un ton haut et 1'aug­ ment placé devant ce préfixe un ton bas.

114

AFRICAN LINGUISTICS

exemples:"enfants" "blanc" "sorte "

/úmwáànà/ "enfant" thème: ààna. /umwààkà/ "année" thème: ààka. /íkámlro/ "endroit où on trait les vaches" thème: kimiro /àmizuùrîi/ "nez" (plur.)

thème: négàmb /iglkomeye/ "ce qui est dur" thème: komeye thème: zúùrù /àhamànukà/ "pente" /ùbutegetsï/ "pouvoir" thème: manuka thème: tégétsï /umûkenero/ "pagne porté à la taille" thème: k é ň e r ò 2.1.3.3. Les schèmes tonals des noms dont le thème est de schème haut-haut-bas exemples :

TONS ET ACCENTS EN KINYÄRWANDA

115

Le deuxième ton haut, celui de la deuxième voyelle étant prêfinal est réalisé descendant en finale d'énon­ cé, ce qui crée une modulation qu'on peut voir sur le tra­ cé (figure 23) . En position non finale, ce ton haut pré­ final est réalisé haut (figure 24) . Le premier ton, celui de la première voyelle est bas. Dans le contexte /kúbon i t á á b i / , (figure 25) il contraste plus nettement avec un ton haut qu'en début d'énoncé. Ce mot /itáábl/ est constitué du thème - t á à b i et de 1'augment i. L'augment qui précède le thème reste de ton bas, le thème táábi comportant deux tons hauts. Dans les noms formés sur des thèmes de ce scheme, le préfixe syllabique porte un ton bas, ainsi que 1'augment, exemples: /Imáári/ "de l'ar/imbwiîžà/ "sorte de légent" gume" thème: máári thème: bwiiza. /ùmwáámï/ "roi" /ïkiňogote/ "porc-épic" thème: áámï thème: óó te /àgàtábirè/ "dernier stade du labour" thème: t á b i r è /ùmùtémérl/ "couvercle en vannerie" thème: temeri /ùmùyóbóke/ "celui qui reconnaît une autorité" thème: yóbóke /ùmunoróró/ "prison /imigilire/ "façon d'a­ gir"

thème: nóróró thème: gilire 2.1.3.4. Les schèmes tonals des noms dont le thème est de schème haut-haut-haut exemples :

116

AFRICAN LINGUISTICS

En finale d'énoncé, le dernier ton haut est réalisé descendant (figure 26). En position non finale, il est ré­ alisé haut (figure 27). D'autre part, à l'initiale d'é­ noncé, on retrouve la réalisation en "crescendo" des tons hauts successifs, et le rabaissement du premier ton haut. Ce mot: /ítúúzé/ est constitué d'un thème: túúzé et d'un augment i. Celui-ci précédant un thème comportant trois tons hauts, donc un nombre impair de tons hauts, prend lui-même un ton haut. Dans les noms formés sur des thèmes de ce scheme, le préfixe syllabique portera un ton haut. Il en sera de même pour l'augment en l'absence de préfixe syllabique. L'augment devant préfixe syllabique portera un ton bas. exemples: /ítée&á/ "loi" /íneezá/ "bienfaisance" thème: teé&á thème: néézá /išiingé/ "sorte /lgišórÓ6wá/ "sorte de d'herbe" ver"

TONS ET ACCENTS EN KINYARWANDA

117

thème: siingé thème: sóróbwá /ïklbiriti/ "boîte /úmwéérá/ "le fait d'ad'alumettes" voir une peau sèche" thème: blritl thème: eer  Le nombre de schèmes possibles pour les thèmes de trois mores est donc limité: il s'agit des schèmes: bas-basbas, haut-bas-bas, haut-haut-bas, haut-haut-haut. Nous fournirons ultérieurement une explication à cette limita­ tion. En progressant dans l'étude des schèmes tonals, nous voyons s'affirmer de plus en plus, la relation entre le ton du suffixe ou de 1'augment et le caractère pair ou im­ pair du nombre de tons hauts du thème nominal. 2.1.4. trois

Les schemes mores

tonals

des

noms

dont

le

thème

dépasse

Pour vérifier ici la relation entre le ton du préfixe ou de 1'augment et le caractère pair ou impair du nombre de tons hauts dans le thème, nous présenterons les schèmes tonals des noms en fonction du nombre pair ou impair des tons hauts du thème. Il faut noter que les noms non composés dont le thème dépasse trois mores sont peu nombreux. 2.1.4.1. Schèmes des noms dont le thème ne comporte que des tons bas exemples: /ùmùgèèndèrèò/ "une ruelle" thème: geender

ero

2.1.4.2. Schemes des noms dont le thème comporte un nombre impair de tons hauts exemples: /àgâtsiïnsïlnb/ "talon" thème:

tsiïnsïïnb

/urumegemége/ thème: megemege /umúsúmiarl/ thème: sumaari /Ikibénobonb/

"pousse (de cheveu, d'herbe)" "clou" "ricin"

thème: bono bono

/agásígislgl/ thème:

"ce qui reste"

sigisigi

Dans ces exemples, le préfixe porte un ton haut pour satisfaire l'exigence de parité du nombre de tons hauts. 2.1.4.3. Schemes des noms dont le thème comporte un nombre pair de tons hauts /umukororbbmbya/ "arc-en-ciel" thème:

kororbbmbya

118

AFRICAN LINGUISTICS

/umusèrèbanya/ "sorte de lézard" thème: serebanya. Dans ces exemples, le préfixe porte un ton bas, l'exi­ gence de parité étant satisfaite par le nombre pair de tons hauts du thème. 2.2.

Schèmes tonals nominaux et accent

Les résultats établis précédemment pour chaque scheme, une fois rapprochés, conduisent à interpréter les faits par un accent tonal. 2.2.1.

Schèmes des thèmes et accent Nous avons v u q u e les thèmes présentaient les schemes suivants: thème thème thème d'une more d e deux mores de trois mores a aa aaa / / / a aa aaa aa aaa /// (en nous limitant aux thèmes d'une, deux o u trois mores) aaa Ces schemes traduisent u n accent placé comme suit: a aa aaa a a a a aa aa aa 'a aaa ' (en reprenant la notation d e McCawley (1968) pour le j a ­ ponais o ù le trait v e r t i c a l e s t p l a c é après la syllabe accentuée et indique le point d e chute d e la mélodie) D a n s ces t h è m e s , toutes les possibilités d e placement de l'accent ainsi q u e son absence sont r e p r é s e n t é e s . L'accent e s t m a r q u é au niveau tonal p a r u n plateau m é l o d i q u e partant d e la syllabe accentuée jusqu'au début du thème. L e s plateaux mélodiques peuvent être notés d e la f a ­ çon suivante: a a"

aa â"a

aaa aaa

a a"

â~â~a â~â~â~

119

TONS ET ACCENTS EN KINYARWANDA

(en reprenant ici encore la notation de McCawley (1968) pour le japonais) On peut voir que ces formes s'opposent entre elles par la présence ou 1f absence de plateau mélodique et la longueur de ce dernier. 2.2.2,

Schemes

tonals

des

noms

et

accent

2.2.2.1. Formes entièrement à tons bas et absence d'accent Dans ces formes, aucune syllabe n'est marquée par un accent. Il n'y a donc pas introduction d'un plateau mélo­ dique. Toutes les syllabes sont alors de ton bas. 2.2.2.2. Formes ne comportant que le thème seul Nous avons vu que dans les thèmes, toutes les mores précédant l'accent sont de ton haut. Il n'est donc pas surprenant que dans les noms constitués uniquement d'un thème, il en soit de même, indépendamment du nombre de tons hauts résultant. exemples: notation notation accentuelle tonale maama' daata'

maama daata

"mère" "père"

Dans cette catégorie, nous inclurons également les pronoms personnels. exemples : notation notation accentuelle tonale we' we' "lui" geewe tweebwe

'

gééwè "moi" twéébwé "nous"

2.2.2.1. Formes nominales comportant soit un pré­ fixe, soit un augment soit les deux Nous avons vu comment le ton du préfixe et de 1'aug­ ment dépendait du nombre pair ou impair de tons hauts du thème. Si le thème comporte un nombre impair de tons hauts, le premier élément préfixé, que ce soit un préfixe ou un augment prendra un ton haut. Dans le cas contraire, il prendra un ton bas. On voit donc apparaître ici une condi­ tion de parité portant sur le plateau mélodique. Le pla­ teau mélodique doit être pair et, soit il le restera s'il y a-un nombre de tons hauts pair dans le thème, soit il le deviendra par adjonction d'un ton haut sur l'élément préfixé. Le scheme tonal est donc entièrement prêdictible à partir

de l'accent

du thème,

sachant que toutes les

mores d-'-un thème nominal précédant l'accent portent un

120

AFRICAN LINGUISTICS

ton haut et que le plateau mélodique doit être pair. Une triple notation permet de le constater dans les exemples suivants: notation de l'accent l

f

i

l

g

i

i

t

ina

i

k

u

m

i

u u

z

u

'

g t

u

o u

ikibiriti i

Ikibonobo

no

2.3.

n

t

e e

i

h

á

m

n

k

m

k

o i

a

t

r

e a

r

ù

a

n

n

u

b

i

i

k

t

e

e i

l

a ï

é

z r

s

a

m g

u

b

b

a

n u

g

á á

a

ï t

i o

u

á

k b

a

w

ï

o a

u

t

a

o

g

w

I

k i

u z

a

i

l

b

b

m

i

z ï s

m I

u

'

agatsi



t n

u

i

u

b

i

k

u

u



m

ikino'

r

'

i a

ù

e

a

a

g

a

u 

a

umwaa ï

r

n



t

o

b

a

ahama

k

o

l

m

'

a

umwa i

o g

b

i

a

b

notation tonale

f

'

n

u

i

notation des plateaux mélodiques

n

t s

i l

ï

n

ikibonobono

Conclusion

Nous avons donc progressivement dégagé les schemes tonals des noms (non composés) à partir des réalisations phonétiques ainsi que le système accentuel dont ils sont la manifestation. C'est au moment du dégagement des schemes tonals que nous avons commencé à nous éloigner des auteurs précédents comme permettra de s'en rendre compte le tableau placé en appendice et regroupant les notations par divers auteurs des exemples que nous avons étudiés (ou d'analogues au niveau tonal). Partant de schemes tonals différents, nos raisonne­ ments pour en rendre compte ne pouvaient ensuite que di­ verger de ceux de ces auteurs. Ainsi sommes-nous arrivés à dégager un système à accent tonal et η'avons-nous pas

b

TONS ET ACCENTS EN KINYARWANDA

121

retrouvé les principes d'anticipation et de déplacement tonals de Kimenyi (1976). Ce système à accent présente des points communs avec celui du Ganda, autre langue bantoue de la zone J. Dans cette langue, il ressort, malgré les divergences entre les auteurs (Kagoma et Stevick, 1968; Kalema 1977) qu'on peut rendre compte des schemes tonals à partir d'un accent. Toutes les syllabes précédant cet accent, sauf certaines syllabes initiales spécifiées comme basses, sont hautes comme en kinyarwanda, selon notre analyse. Le désaccord entre les auteurs touche plus particulièrement la hauteur des dernières syllabes des mots. Il semble dû en grande partie à des intonations différentes. D'après Kalema (1977), en "intonation de phrase", toutes les syllabes suivant l'accent sont basses, donc comme en kinyarwanda. Le système que nous venons de dégager n'est pas non plus sans rappeler le système d'une langue non bantoue mais bien connue: le japonais. Dans cette langue, selon, entre autres McCawley (1968 et 1978), toutes les syllabes précédant l'accent sont hautes (hormis la première si elle-même n'est pas accentuée). Toutes celles qui le sui­ vent sont basses. Dans ces langues, l'accent se manifeste par des pla­ teaux mélodiques comme en kinyarwanda, mais la particula­ rité du kinyarwanda est de soumettre ces plateaux mélo­ diques à des exigences rythmiques.

2.

SCHEMES TONALS DES INFINITIFS ET ACCENT

L'étude des schemes tonals des infinitifs qui sont des noms formés sur des thèmes verbaux constituera une transition entre l'étude des schemes tonals des noms et ceux des verbes. Le mot "thème verbal" sera utilisé ici dans l'ac­ ception de Meeussen (19 59) et s'appliquera à l'ensemble constitué par le radical, les dérivatifs et la terminai­ son vocalique. Nous désignerons d'autre part par le terme de "noyau verbal", le radical et la voyelle qui le suit, unité dont le contour se dégagera au cours de notre étude, en particulier quand nous examinerons les schemes du subjonctif et des formes négatives. Nous suivrons la même démarche que pour l'étude des schemes tonals nominaux.

122 3.1. 3,1.1. ni

AFRICAN LINGUISTICS

Les schèmes tonals des infinitifs Les infixe

schemes

tonals

des

infinitifs

sans

dérivatif

3.1.1.1. Les schemes des infinitifs formés sur des thèmes sans ton haut Ces infinitifs constitués du thème et du préfixe de classe ku/gu  sont entièrement de ton bas. exemples: /gùsya/ "moudre" /kuva/ "suinter" thème: syà thème : va /gukina/ "jouer" /kurera/ "élever (un en­ fant) thème: kina thème: rera /gùseka/ "rire" /gutbta/ "être humide" thème: seka thème: tota. /kurura/ "être /gukena/ "être pauvre" thème: rùrà /guhiruta/

thème: kena "agoni*-

/guhirima/

's'écrouler'

ser" thème: hiruta /guhaanda/ thème: haanda

"piquer"

thème: hïrïmà /guklïnga/ "fermer" thème: kïïngk

3.1.1.2. Schemes des infinitifs dont le thème com­ porte un ton haut Les thèmes des infinitifs ne peuvent comporter qu'un seul ton haut, quel que soit le nombre de leurs syllabes. Le préfixe ku reçoit un ton haut comme dans les noms qui ne comportent qu'un seul ton haut. Un palier mélodique de deux mores hautes se trou­ ve ainsi formé au début de ces infinitifs.. exemples: /guta/ "jeter" /gúpfá/ "mourir" thème: ta thème: Pf¿ //curya/ "manger" /gúkyi/ "faire jour" thème: kg a thème: rya ^ /gûkémk/ "empêcher" // "voir" thème: kémk thème: bonà "traire" /gúsáñk/ "réparer ' ^ /gikimi/ thème : kama. thème: saà "border" /gútáta/ "espionner" /kûrôtà/ thème: rotà (un lit) thème: tata // "finir" /kurobth/ 'rever' thème: hob sa thème: roo ta /guharùra/ "désherber* /giîtutïrà/ "couper (à la houe) des roseaux" thème: tutira thème: hàrùra

123

TONS ET ACCENTS EN KINYARWANDA

Les

3.1.2. dérivatifs

schemes

tonals

des

infinitifs

contenant

des

Les dérivatifs s'ajoutent au radical verbal et n'intro­ duisent que des tons bas, quel que soit le scheme du radical verbal. exemples : infinit, infinit, infinit, infinit, sans avec un avec deux avec un dérivatif dérivatif dérivatifs dérivatif et plus /gutema/

/gutemera/

/gutemeesa//gut

"couper"

"couper pour"

"faire couper"

/kvtbonà

/kubonera/

/kuboneesa/

/

"voir"

"voir pour"

"faire voir"

fgukina/

/gukmira/

/gukmiisa

"jouer"

"jouer à"

"faire jouer"

/guta/

/guteera/

/guteesa/

"quitter, jeter"

"jeter pour"

"faire quitter"

^ \

sereza/

\ \ V \ /

/, / , /

/

eme e

"faire couper pour"

\ ^ v ^

v.\

\\v\

\

/ // /gukmiisiriza

/guhoondera/

"concasser"

"concasser pour"

3.1.3.

Les

schemes

tonals

des

infinitifs

»

^ v \

> . ^

/

"faire quitter'

es-

"faire con­ casser pour" avec

infixes

3.1.3.1. Les schemes des infinitifs avec infixes dont le thème ne comporte que des tons bas Tous les tons de ces formes sont bas: aucun infixe n'introduit de ton haut, exemples: /kùglsyà/ "le moudre" thème: syà /kumuhana/

thème: hana /kugïhoonda/ thème:

hoonda

"le punir"

"le concasser"

,

/guteeresa/

/guhoondeesa//guhoonde "faire ereza/

concasser"

/

"faire jouer à"

/gútáabwa/ "être jeté" /gùhôonda/

\

124

AFRICAN LINGUISTICS

3.1.3.2. Les schemes tonals des infinitifs avec in­ fixes dont le thème comporte un ton haut thème

d'une

more

Insertion d'un infixe exemples :

L'infinitif /kùmuha/ présente donc un scheme tonal: haut-haut-bas. Son ton haut préfinal est réalisé descen­ dant en fin d'énoncé (figure 28) mais simplement haut en position interne d'énoncé (figure 29). Cet infinitif est composé du thème ha, porteur d'un ton haut, de l'infixé mu et du préfixe ku· L'infixé mu et le préfixe ku reçoivent un ton haut. Mais trois tons hauts successifs se trouveraient formés si le thème gar­ dait son ton haut. Or, on peut constater que celui-ci de­ vient bas. L'exigence d'un nombre pair de tons hauts se

TONS ET ACCENTS EN KINYARWANDA se

trouve

donc

ici

satisfaite

par

un autre

125 biais.

Insertion de deux infixes exemples :

L'infinitif /kûhamuha/ présente donc un scheme tonal: haut-haut-haut, les tons hauts étant toujours soumis aux mêmes lois de réalisations en fonction de leur position dans l'énoncé: rabaissés en début d'énoncé et réalisés descendants en finale d'énoncé. Cet infinitif est composé du thème -ha, des infixes -mu- et -ha- et du préfixe ku-. Les préfixes et les in­ fixes reçoivent un ton haut et le ton haut du thème se trouvant en quatrième position, donc dans une position paire est maintenu. theme

de deux

mores

Les schemes tonals des infinitifs formés sur des thè­ mes de deux mores résultent des mêmes mécanismes que ceux

126

AFRICAN LINGUISTICS

des thèmes monosyllabiques puisqu'ils ne comportent comme eux qu'un seul ton haut, exemples :

TONS ET ACCENTS EN KINYARWANDA

127

Dans l'infinitif sans infixe: /gutema/, le thème porte son ton haut. Celui-ci est réalisé descendant en finale d'énoncé (figure 32). Par contre, dans l'infinitif avec un infixe qu'est /kúmúthma/, le ton haut de -téma qui arriverait en troisième position est supprimé. Et, dans le thème l'infinitif avec deux infixes: /kúmúhátéma/, garde son ton haut qui est placé en quatrième position. thème

de

trois

mores

Leurs schemes tonals résultent toujours des mêmes processus, exemples :

128

AFRICAN LINGUISTICS

Le thème conserve son ton haut, lorsqu'il est en po­ sition paire à partir du début du mot, c'est à dire dans /gukuunda/, et dans /kukimukuundlra/. 3.2.

Les schèmes tonals des infinitifs et l'accent

Comme les schemes tonals nominaux, les schemes tonals des infinitifs s'expliquent parfaitement par un accent to­ nal. 3.2.2.

Schèmes tonals des thèmes et accent Les thèmes des infinitifs comportent un ton haut sur la première more, à savoir: aa aaa a aa ou en termes accentuels a aa t

*

aaa aaa /

a a a a aa c'est à dire que l'accent ne peut frapper que la première more et que les formes des thèmes de l'infi­ nitif ne se distinguent entre elles que par la présence ou l'absence d'un accent. 3.2.3.

Schèmes des infinitifs et accent 3.2.3.1. Les formes à tons bas Leur scheme s'explique par le fait qu'elles sont sans accent, et que donc aucun plateau mélodique ne se trouve

TONS ET ACCENTS EN KINYARWANDA

129

introduit. 3.2.3.2. Le ton bas des dérivatifs Les dérivatifs se trouvant placés après la syllabe accentuée, s'il y en a une, sont automatiquement de ton bas, comme le sont toutes les syllabes venant après l'ac­ cent. 3.2.3.3. Les formes d'infinitifs comportant des tons hauts et accent Nous avons vu que le préfixe et les infixes rece­ vaient toujours un ton haut. Nous avons également remar­ qué que le plateau mélodique était toujours pair mais que sa parité était obtenue par un processus différent de ce­ lui rencontré dans les noms. Dans les infinitifs, le début du plateau mélodique est fixe: c'est le préfixe ku mais l'accent se déplacera éventuellement d'une syllabe, si besoin est, pour mainte­ nir la parité du plateau mélodique. A côté du concept d'accent, marqué par un plateau mélodique, concept déjà plusieurs fois utilisé dans l'a­ nalyse s'autres langues, nous sommes ammenés à introduire celui de "point d'ancrage", pour designer de début de plateau mélodique.

ce point

fixe

Celui-ci peut rentrer "en compétition" avec l'accent puisqu'il est susceptible d'entraîner son déplacement. Ce­ pendant, il ne peut être introducteur d'un plateau mélo­ dique, ce qui est la spécificité de l'accent; il n'est qu'un point imposé de début de plateau mélodique. Nous verrons ultérieurement en analysant les formes verbales, le parti que le kinyarwanda tire de l'existence d'un accent et d'un point d'ancrage pour opposer diffé­ rents temps et différents modes. Les séries suivantes nous permettront d'illustrer ce maintien des plateaux pairs à partir d'un point d'ancrage fixe, par déplacement de l'accent. Le point d'ancrage sera noté par un point souscrit. notation de l'accent thème: ná

thème: témà

notation des plateaux mélodiques

notation tonale

guha'



guha

kumu ha

Tiumûha

kumúha

kuhamuha



kúhámúhá

gute

ma

gûTêma

gitema

kugi

tema

kuhamute

ma

3EïïgTtema

kugitema



kúhámitema

AFRICAN LINGUISTICS

130

thème: kuùndà  uunda kumu kuunda gukimuku undira 3.3.

gukuunda kumukuunda gukJEukuundira

gukuunda kumukùùndà gukîmukuùndira

Conclusion

Nous avons donc pu donner une explication ac.centuelle aux schemes tonals phonologiques que nous avons dégagés. Mais par rapport à notre étude des schemes tonals des noms, nous avons vu apparaître un fait nouveau: le déplacement de l'accent selon un point d'ancrage fixe et caractéris­ tique de l'infinitif pour maintenir, si besoin est, la pa­ rité du plateau mélodique.

4.

SCHEMES TONALS VERBAUX ET ACCENT

Nous n'examinerons ici que les schemes tonals des verbes à quelques temps de l'indicatif et du subjonctif, à la fois, à l'affirmatif et au négatif. Notre but est de dégager les mécanismes accentuels déterminant leurs sche­ mes tonals. Nous verrons comment ceux-ci sont déterminés par la place de l'accent, la place du point d'ancrage, et la longueur du plateau mélodique. Mais, nous commencerons, comme précédemment, par établir les schemes tonals euxmêmes . 4.1. Schemes tonals des verbes à l'immédiat, au récent et au prétérit de l'indicatif affirmatif et accent tonal 4.1,1. récent

Les schèmes tonals des verbes a l'immédiat, au et au prétérit de l'indicatif affirmatif Nous nous limiterons ici à l'immédiat, au récent et au prétérit. Dans notre appellation des modes, des temps et des constituants du verbe, nous suivrons la terminolo­ gie de Coupez (1961) et Meeussen (1959), reprise par Overdulve (1975) 4.1.1.1. Schèmes tonals des verbes à l'indicatif im­ médiat imperfectif disjoint Ce temps est formé segmentalement de la marque -ra, préposée au thème et de la voyelle final -a. Les thèmes présentent, dans ce temps, leur scheme tonal lexical, c'est à dire celui qu'ils présentent dans les infinitifs.

TONS ET ACCENTS EN KINYARWANDA

131

- schemes tonals des verbes dont le thème ne comporte que des tons bas Comme nous pouvions nous y attendre, tous les tons de ces formes verbales sont bas, qu'elles comportent ou non des infixes: exemples: arasyà "il moud" kragisyà "il le moud" a-ra-sya

a-ra-gi-sya

prëf.-morph.-thème temps ararbrà

"il regarde"

a-ra-rora

préf.-morph. -inf.-thème temps

àràkïrbrà

"il le regarde" arahamùrbrà "il le regarde là"

a-ra-ki~rora

a-ra-ha-mu-rora

prêf.-morph.-thème temps aràhbbnda a-ra-

préf.-morph.-inf.-thème préf.-mor.temps inf.-inf. temps "il concasse" araglhbbndà "il le concasse"

hoonda

préf.-morph.-thème temps

a-ra-gi-hoonda

préf.-morph.-inf.-thème temps

- schemes tonals des verbes dont le thème comporte un ton haut Les thèmes ayant leur scheme lexical, ils ne peuvent comporter qu'un ton haut. Les schemes verbaux seront les suivants: - verbes à thème d'une more exemples:

132

AFRICAN LINGUISTICS

TONS ET ACCENTS EN KINYARWANDA

133

La forme /araha/ présente un scheme bas-haut-haut. Le dernier ton haut est réalisé descendant en finale d'énon­ cé (figure 38), mais simplement haut en position non fi­ nale (figure 39). La forme /arámúha/ présente un scheme bas-haut-hautbas. Son ton haut préfinal est réalisé descendant en fi­ nale d'énoncé (figure 40). La forme /arákímúhá/ présente, elle un scheme bashaut-haut-haut-haut. Son dernier ton haut est réalisé descendant en finale d'énoncé (figure 42), de façon très semblable à celui de /araha/ (figure 38). La forme /araha/ est constituée d'un préfixe person­ nel a, d'une marque préposée ra et du thème verbal ha. Le thème étant de ton haut, la marque préposée ra reçoit un ton haut. La forme /arámúha/ a une constitution parallèle à mais avec l'infixé mu en plus. celle de /araha/, Le préfixe personnel sujet reste toujours de ton bas. Le morphème temporel ra et l'infixé reçoivent un ton haut, mais le ton haut du thème est supprimé parce qu'étant de rang impair (troisième ton haut). Nous retrouvons le même processus que dans les infinitifs. La forme /arakimúha/ présente un infixe de plus que la forme précédente, c'est à dire deux infixes: ce sont ki et mu. Ces deux infixes portent un ton haut. Le ton du thème qui devient le quatrième ton haut est maintenu. - verbes à thèmes de deux mores exemples :

134

AFRICAN LINGUISTICS

TONS ET ACCENTS EN KINYARWANDA

135

La forme /ara terna/ est de scheme bas-haut-haut-bas. Le ton haut prêfinal de la syllabe te est réalisé descen­ dant lorsque^le mot est en finale d'énoncé (fiqures 4 3). La forme /aramutèmà/ est de scheme bas-haut-haut-bas-bas (figure 44) . La forme /aráklmútémera/ présente un scheme avec quatre tons hauts suivis de deux tons bas (figure 45). Ces formes ont la même structure que celles des exem­ ples vus précédemment. Elles comportent toutes la marque préposée ra, porteuse d'un ton haut. La première /àritémk/, ne comporte pas d'infixé, la deuxième /aramutema/ en com­ f porte un, l infixe^u porteur également d'un ton haut, la en comporte deux: ki et mu, troisième /aráklmútémera/ tous les deux avec des tons hauts. Le ton haut du thème tema cède la place à un ton bas lorsqu'il se trouve en po­ sition impaire, à partir de la syllabe ra. - verbes à thème de trois mores exemples

136

AFRICAN LINGUISTICS

Nous retrouvons toujours les mêmes processus: la marque ra préposée reçoit un ton haute Les infixes que sont mu dans la forme /aramukùùnda/ et ki et mu dans la forme /arakimukuundira/ reçoivent également un ton haut. Le ton haut du thème fait place à un ton bas lorsqu'il est en position impaire, comme dans /aramukùùndà/ (figure 47). autres exemples : /arabona/ "il voit" /arábyina/ "il danse" /aramùbbna/ "il le voit" /arahàbylnlra/ "il danse là" /arahamûbona/ "il le voit /arahayibyinirà/ "il le là" danse là" (infixes ha et mu) (infixes ha et yi) dérivatif ir /arkroota/ "il rêve" /arâkobsa/ "il dote" /àramurbbta/ "il rêve de /àramukobsà/ "il la dote" lui" 4.1.1.2. Les schèmes tonals des verbes à l'indica­ tif immédiat perfectif disjoint Les formes de ce temps sont semblables à celles de 1'imperfectif correspondant, si ce n'est qu'elles com­ portent un -ye (dont l'adjonction provoque souvent des mo­ difications segmentales dans les bases). Leurs schemes tonals sont donc semblables: /aràboonye/ "il vient de voir" /àrâmubbbnye/ "il vient de le voir" /arámúhábóbnye/ " il vient de le voir là"

/aràtèmye/ "il vient de couper" /àramutemye/ "il vient de le couper" /aràkimutemeye/ "il vient de le couper pour lui"

137

TONS ET ACCENTS EN KINYARWANDA

/arakùunze/

"il vient

d'aimer"

/aramukuunze/ de l f aimer"

/araroye/ regarder

"il vient

/àramùrbye/ regarder

"il vient de le (vite)"

"il vient de le (vite)"

4.1.1.3. Les schèmes tonals des verbes à l'indicatif immédiat imperfectif conjoint Ce temps est marqué segmentalement par la voyelle finale -a. et l'absence de marque temporelle. Les thèmes ne peuvent porter que des tons bas. Ils perdent donc leur scheme lexical pour adopter un scheme uniforme qui consti­ tue une des marques de ce temps. Comme toutes les formes construites à partir de formes à tons bas, celles de ce temps ne comportent que des tons bas qu'elles comprennent ou non des infixes, exemples /aha/ "il donne"... /atema/ "il coupe"... /amùha/ "il le donne"... /amutema/ "il le coupe" /akïmùha/ "il le lui /akimùtemera/ "il le coupe donne"... pour lui"... /akùùnda/ "il aime"... /arbra/ "il regarde (vite)"... /amukuunda/ "il 1'aime".../amùrbra/ "il le regarde (vite)"... /àkïmùkhundïra/ "il l'aime pour lui"... /areeba/ "il regarde"... /* ■ % χ  _ / „.-, 1 Ä v.Ä„avv4Äii /amureeba/ il le regarde'...

N.B. Les points... indiquent que la forme ^ ^ necessaire_ ment suivie d'autre chose

4.1.1.4. Les schèmes tonals des verbes à l'indica­ tif immédiat perfectif conjoint Les formes de ce temps ne se différencient de leurs homologues imperfectives que par leur terminaison -ye (et les modifications morphophonologiques associées à sa pré­ sence) . exemples: /ahaaye/ "il vient de /àtemye/ "il vient de coudonner" per" /amuhaaye/ "il vient de /amutemye/ "il vient de le le donner" couper" /areebye/ "il vient de /akuunze/ "il vient d'airegarder" mer" Les formes de ce temps ne comportent que des tons bas.

138

AFRICAN LINGUISTICS

4.1.1.5. Les schèmes tonals des verbes à l'indicatif récent imperfectif conjoint ou disjoint Ce temps est marqué segmentalement par la marque pré­ posée a, et une marque aspectuelle - postposée. Les thèmes ne peuvent porter que des tons bas. Leur scheme est donc une des marques du temps, exemples: /yahaga/ "il donnait" / yasaga/ "il ressemblait" (récemment) /wahaga/ "tu donnais" /twaasaga/ "nous ressem­ blions" /yatemaga/ "il coupait" /y aro rag a/ "il regardait vite" /yakuundaga/ "il aimait" / mwaarbbtaga/ "vous rêviez" Les thèmes étant de ton bas, toutes les formes de ce temps, quelle que soit leur composition ne peuvent porter que des tons bas. 4.1.1.6. Les schèmes tonals des verbes à l'indicatif récent perfectif conjoint. Il ne se différencie de son homologue imperfectif qu'au niveau segmentai, c'est à dire par la terminaison -ye qui vient remplacer la terminaison , exemples: /yahaaye/ "il a donné /yatemye/ "il a coupé" (récemment)" /yarbye/ "il a regardé /yakuunze/ "il a aimé" vite" /yktîïtse/ "il a insulté" /twaarutse/ "nous avons vo­ mi" /ykrbbsh/ "il a rêvé" 4.1.1.7. Les schèmes tonals des verbes à l'indicatif prétérit imperfectif disjoint Segmentalement, ce temps est marqué par les marques ra et a préposées et par la marque  postposée. Un seul type de scheme est admis pour les thèmes, le scheme avec un seul ton haut, quel que soit le scheme lexical du thème. Ce scheme avec un ton haut constitue donc une des marques du temps. Comme dans les formes dont le thème comporte un ton haut, vues précédemment, la marque ra reçoit un ton haut ainsi que les infixes. Le ton haut du thème n'est mainte­ nu que s'il est en position paire, exemples : /yarahaga/ "il donnait" /yàrâthmkgk/ "il coupait" (il y a longtemps) /yaramuhaga/ "il le don- /yâràmutêmagà/ "il le counait" pait"

TONS ET ACCENTS EN KINYARWANDA

139

/yáráróraga/ "il regardait (vite)" /ydrarèèbàgà/ "il regardait" /ydrdmiiréebàgà/ "il le regardait" /yarakùùndàgk/ "il aimait" /yárámdkúhndaga/ "il l'aimait" 4.1.1.8. Les schèmes tonals des verbes à l'indicatif prétérit perfectif disjoint Il ne se différencie de son homologue perfectif qu'au niveau segmentai, c'est à dire par une terminaison -ye qui prend la place de la terminaison - et les transformations morphologiques qui en découlent, examples : /yáráhaayh/ "il a donné" /yárátemye/ "il a coupé" (il y a longtemps) /yàràmuhàkyh/ "il l'a /yárámdtámyh/ "il l'a coudonné" pé" /yáráreebye/ "il a regar- /yárárbye/ "il a regardé dé" (vite)" /yárámúréhbyh/ "il l'a /yaràmûroyh/ "il l'a re­ regardé" gardé (vite)" 4.1.1.9. Les schèmes tonals des verbes à l'indicatif prétérit imperfectif conjoint Segmentalement, ce temps est marqué par la marque a préposée au thème et par la terminaison -. Un seul type de scheme est possible pour les thèmes, le scheme avec ton haut. La marque temporelle a reçoit un ton haut ainsi que les infixes et seulement deux tons hauts sont admis: les deux premiers tons hauts à partir de la marque temporelle, exemples: /yáhágk/ "il donnait" /yàtémàgh/ "il coupait" (il y a longtemps) /yamuhkgk/ "il le don/yâmutemkgk/ "il le counait" pait" /yarorkgk/ "il regardait /yàkimùtemeragk/ "il le (vite)" coupait pour lui" /yámúrorkgk/ "il le re­ gardait (vite)" 4.1.1.10 Les schèmes tonals des verbes à l'indicatif prétérit perfectif conjoint Ce temps ne se différencie de son homologue imper­ fectif qu'au niveau segmentai, par sa terminaison ye. exemples : /yáhákyh/ "il a donné" /yârSyh/ "il a vu" (il y a longtemps) /yàgihkkyh/ "il l'a /yâréhbyh/ "il a regardé" donné"

140

AFRICAN LINGUISTICS

/yakimuhaaye/ "il le lui a donné" /yatemye/ "il a coupé" /yákúunze/ /twáátemye/ "nous avons coupé" /yagLtemye/ "il l'a coupé"

"il a aimé"

4.1.2.

Schèmes tonals et accent 4.1.2.1. Schèmes tonals des thèmes et accent Dans l'indicatif affirmatif, limité aux temps que nous avons étudiés, les thèmes comportent un ton haut sur la more initiale, comme dans les infinitifs, à savoir: a aa aaa /

a ou en termes accentuels: a f

/

P

aa

aaa

aa

aaa

/



a a a a aa Autrement dit, dans toutes les formes que nous venons de voir et qui comportent des tons hauts, le point de dé­ part de l'accent est la première syllabe du thème. L'ac­ cent peut être ensuite déplacé selon des modalités que nous allons maintenant examiner. 4.1.2.2. Schèmes tonals des formes en ra et accent cent - les schemes tonals des formes en ra et accent Ceux-ci peuvent être interprétés accentuellement de la même façon que les schemes tonals des infinitifs : -les schemes tonals entièrement à tons bas traduisent l'absence d'accent. -dans les schemes à tons hauts, nous avions remarqué que le nombre de tons hauts devait être pair à partir de la première marque temporelle qui est soit ra soit a lorsque celle-ci apparaît devant la marque ra et que le ton haut excédentaire, en position impaire était supprimé. Il s'agit là des effets du déplacement de l'accent pour maintenir un plateau mélodique pair par rapport au point d'ancrage que constitue la première marque tempo­ relle. Le préfixe sujet précédant le point d'ancrage sera exclu du plateau mélodique et portera un ton bas. Le déplacement de l'accent, le réglage des plateaux mélodiques par rapport au point d'ancrage peuvent être illustrés par la triple notation des exemples suivants.

141

TONS ET ACCENTS EN KINYARWANDA

notation de l'accent et du point d'ancrage

notation des plateaux mélodiques

araha

araha aramu

notation tonale

aramuha

ha

arakimuha

arakimuha

arate

aratema

ma

aragi

tema

arakimute araku

aragitema mera

aramu

kuunda

arakimuku

ar akimutemer

a

arakuunda

unda undira

aramukùunda. arakimukuundira

Toutes les syllabes précédant l'accent jusqu'au point d'ancrage portent un ton haut, toutes celles qui le suivent, un ton bas. - Les schemes tonals des formes sans ra et accent Ces formes se différencient des formes en ra, par la longueur du plateau mélodique qui ne peut recouvrir que deux mores mais non par la position du point d'ancrage qui reste la marque temporelle. La place de l'accent dépendra donc de ce point a*an­ crage et de la nécessité de former un plateau de deux syl­ labes . notation des notation notation plateaux tonale de l'accent mélodiques et du point d'ancrage yagi

haaye

yagihaaye

yaki

muhaaye

yakimuhaaye

yate



yatemye

yagi

temye

yagitemye

twaa

temye

twaatemye

yagi

temye

yàgitemye

yaro

raga

yâroràgà

yamu yate

roraga maga

/ / \ \ \ yamuroraga yàtemaga

142

yamu'temaga yaki ' mutemeraga

AFRICAN LINGUISTICS

"^amutemaga TjakJmutemeraga

yámutemaga yakimutemeraga

Le préfixe personnel ne forme pas ici une syllabe: la syllabe ya résulte de la contraction du préfixe a et de la marque temporelle a. Elle porte le ton que reçoit la marque temporelle, à savoir le ton haut. Toutes les mores précédant l'accent portent un ton haut, toutes celles qui le suivent un ton bas. 4.1.3. Conclusion Comme pour les schemes des noms et ceux des infinitifs, nous avons pu montrer aisément que les schemes tonals que nous avions dégagés s'expliquaient par des phénomènes accentuels. Nous avons pu d'autre part, constater qu'à côté de la présence ou de l'absence d'accent, la longueur du plateau mélodique constituait une marque temporelle. 4.2. Schèmes tonals des verbes au subjonctif (immédiat) et accent 4.2.1. Les schèmes tonals des verbes au subjonctif (im­ médiat) Segmentalement, ce mode est marqué par la finale e. Les noyaux (radical + voyelle suivante) présentent un scheme tonal constitué uniquement de tons hauts. Nous n'avions encore rencontré ce scheme ni dans les infinitifs ni dans les indicatifs affirmatifs. Il constitue la marque tonale du mode subjonctif. 4.2.1.1. Schèmes tonals des verbes à thème d'une more ex enroles :

TONS ET ACCENTS EN KINYARWANDA

143

144

AFRICAN LINGUISTICS

La forme /áhé/ comporte deux tons hauts, le dernier se réalisant descendant en finale d'énoncé (figure 49) mais simplement haut en position non finale (figure 50). La forme /amúhé/ est de scheme bas-haut-haut. Comme dans l1exemple précédent, le dernier ton haut se réalise descendant en fin d'énoncé (figure 51) et haut à l'inté­ rieur d'un énoncé (figure 52). La forme /akúmúhe/ comporte un ton bas final. Le ton préfinal est réalisé descendant en finale d'énoncé (fi­ gure 53) mais haut, en position interne (figure 54). La forme /áhé/ est constituée du préfixe sujet a et du thème verbal du subjonctif he. Le préfixe sujet reçoit ici un ton haut. La forme /amúhé/, elle, est constituée du même pré­ fixe a mais qui cette fois reste de ton bas et ne reçoit donc pas un ton induit du thème, de l'infixé mu qui, lui, reçoit un ton haut et enfin du thème hé, avec son ton haut. La forme /akimuhe/ comporte toujours le même préfixe a porteur d'un ton bas, deux infixes ki et mu ayant reçu un ton haut et le thème hé dont le ton haut n'a pas été maintenu. Toutes ces formes comportent donc deux tons hauts. Un des tons hauts est porté par le thème lui-même dans /áhé/ et /amúhé/, le deuxième ayant pour support la syllabe pré­ cédente que ce soit celle du préfixe, comme dans /áhé/ ou celle de l'infixé comme dans /amúhé/. Dans /akimúhe/, les deux tons hauts du subjonctif sont portés par les deux in­ fixes et le ton haut du thème placé en troisième position se trouve supprimé. Donc, bien que le thème soit monosyllabique, le sub­ jonctif est marqué par deux tons hauts et non un seul ton haut conformément au nombre de syllabes du thème 4.2.1.2. Schemes tonals des verbes à thème de deux mores (avec noyau de deux mores) exemples :

TONS ET ACCENTS EN KINYARWANDA

figure 56

figure 57

figure 58

l45

146

AFRICAN LINGUISTICS

La forme /ateme/ présente un scheme tonal: bas-hauthaut, le dernier ton étant réalisé descendant en finale d'énoncé (figure 55) et haut en position non finale (fi­ gure 56) . La forme /agíteme/ a un scheme bas-bas-haut-haut· Son dernier ton haut est également réalisé descendant en fin d'énoncé (figure 57). La forme /akimùtemere/ comporte un scheme bas-hauthaut-bas-bas-bas (figure 58). Les deux tons hauts du subjonctif sont ici portés par le thème disyllabique dans /ateme/ et /agíteme/, .forme dans laquelle l'infixé reste de ton bas. Par contre, ils sont portés par les deux infixes kî et mu dans /àkimutemere/. Ces exemples montrent que dans les formes du sub­ jonctif, les infixes ne peuvent recevoir des tons que s'ils sont deux. D'autre part, deux tons hauts seulement sont admis dans les formes avec infixes. 4.2.1.3. Les schemes tonals des verbes à thèmes de trois mores (avec noyau de trois mores)

TONS ET ACCENTS EN KINYARWANDA

147

figure 61

La forme laréébél comporte après un ton bas trois tons hauts consécutifs. Le dernier est réalisé descendant en finale d'énoncé (figure 59). La forme /amùréébe/ présente un scheme bas-bas-hauthaut-bas. Son ton haut préfinal est réalisé descendant en fin d'énoncé, d'où la modulation qu'on peut observer sur la figure 61. La forme /amùréébe/ est constituée d'un préfixe su­ jet a de ton bas, d'un infixe mu qui ne reçoit pas de ton haut et du thème réébé qui ne porte plus dans cette forme que deux tons hauts. Dans la forme /akimureebere/ les deux infixes ki et mu portent des tons hauts. 4.2.1.4. Les tons des thèmes et les tons des dériva­ tifs Dans ces formes du subjonctif, on voit nettement se dégager le noyau, puisque toutes les syllabes du noyau sont de ton haut par contraste avec les dérivatifs qui le suivent et qui sont porteurs de tons bas. exemples: /akoré/ "qu'il travaille" /àkoréesereze/ "qu'il fasse travailler pour lui" /a témé/ "qu'il coupe" /atémére/ "qu'il coupe pour" /àtémuùlire/ "qu'il joigne pour" (réversif) /kkoosé/ "qu'il dote" /akéoséese/ "qu'il fasse doter" Le noyau korê porte deux tons hauts aussi bien dans la forme /akoré/ sans dérivatif que dans la forme /àkoréesereze/ avec deux dérivatifs. Toutes les syllabes qui suivent le noyau sont de ton bas. De même, le noyau /témé/ est porteur de tons hauts dans la forme sans déri­ vatif /atémé/ aussi bien que dans les formes avec dériva-

148

AFRICAN LINGUISTICS

tifs que sont /atemére/ ou /atémûulïre/ '(avec, dans cette dernière forme, une voyelle u due au type de dérivatif). Le noyau trisyllabique // présente ses trois tons hauts dans la forme // sans dérivatif comme dans la forme /akoéséese/ avec un dérivatif. Ces noyaux, marqués au niveau tonal par ces tons hauts dans les formes du subjonctif, sont composés du ra­ dical et de la voyelle suivante. Ils ne peuvent com­ porter qu'une, deux ou trois mores. Ils n'ont trois mores que lorsque le radical est de structure C W C , comme le radical reeb"regarder" du thème réêbé du verbe /areebé/ "qu'il regarde". Les "élargissements" des radicaux et les dérivatifs, à supposer qu'il soit possible de les distinguer les uns des autres, ont le même compor­ tement tonal et introduisent les uns comme les autres des tons bas après le noyau verbal, exemple: /bàzirikane/ "qu'ils se souviennent" Le noyau est ziri. Le  avec la voyelle qui le suit n'est pas intégré au noyau autres exemples: /akiné/ "qu'il joue" /akinire/ "qu'il joue pour" /akînîïsïrlze/ "qu'il fasse jouer pour" /akûundé/ "qu'il aime" /akuúndlise/ "qu'il soit aimé" /àhârure/ "qu'il désherbe" /khirirïre/ "qu'il désherbe pour" /khinuure/ "qu'il déplie" /aháandúure/ "que'il enlève" /aroondogore/ "qu'il bavarde" /agêendérere/ "qu'il rende visite" Dans chacune de ces formes, on reconnaît le noyau aux tons hauts. 4.2.2.

Les

schèmes

tonals

et

l'accent

4.2.2.1. Les schemes tonals des noyaux et l'accent Les schemes tonals sont les suivants: /

/ /

f / /

a aa aaa et l'accent dont dérivent ces schemes est placé de la façon suivante: / / * a aa aaa Les noyaux du subjonctif (immédiat) portent donc tous un accent sur la dernière more. Comme dans let.

149

TONS ET ACCENTS EN KINYARWANDA

thèmes nominaux, toutes les syllabes précédant cet accent portent un ton haut. 4.2.2.2. Le ton bas des dérivatifs Les dérivatifs sont en eux-mêmes atones. Placés après la voyelle accentuée, ils reçoivent un ton bas. 4.2.2.3. Le noyau: unité accentuelle Nous avons vu que le noyau constitué du radical et de la voyelle suivante se trouvait marqué par les tons hauts dans les formes de subjonctif sans infixe. Ces tons hauts sont la manifestation d'un accent porté par la der­ nière syllabe de cette unité. Cette unité est une unité accentuelle,

c'est l'unité

de placement

de l'accent

du

verbe, c'est à dire l'ensemble de syllabes susceptibles d'être frappées par l'accent du verbe. Elle est encadrée par des éléments sans ton propre (marque temporelle, in­ fixes, dérivatifs), ces éléments pouvant entraîner des déplacements de 1-accent. 4.2.2.4. Les schemes tonals des formes au subjonctif et accent Traitons d'abord des schemes des formes du subjonctif comportant seulement un préfixe sujet et un noyau. Nous avons observé que pour les thèmes monosylla­ biques, le plateau mélodique devait être pair et pouvait s'étendre au préfixe sujet pour remplir cette condition de parité minimale. Pour les noyaux de deux mores ou de trois mores, le plateau recouvre l'ensemble du noyau mais sans empiéter sur le préfixe sujet. Dans ce mode, le pla­ teau mélodique doit donc s'étendre sur l'ensemble du noyau et être au minimum de deus mores. Avec l'insertion des infixes se trouve introduite une nouvelle condition réglant la longueur du plateau mélo­ dique, â savoir qu'il doit être de deux syllabes. S'il y a un seul infixe, soit l'accent restera sur la dernière syllabe (comme dans /amuhe'/ ou /amuteme'/ soit il se dé­ placera d'une syllabe (comme dans /amuree'be/) de façon à ce que le plateau mélodique ait deux syllabes. Le pla­ teau mélodique ne se trouve donc pas ici soumis à l'at­ traction d'un point d'ancrage situé en dehors du thème. Lorsqu'il y a deux infixes, le premier infixe attire le plateau mélodique et devient le point d'ancrage du pla­ teau mélodique. L'accent se trouve alors déplacé et sort du thème (comme dans /akimu'reebere/). L'apparition d'un point d'ancrage uniquement dans ces formes et non dans toutes les formes de ce mode nous paraît due à une exi­ gence rythmique générale dans la langue, à savoir l'im­ possibilité de trouver trois syllabes de ton bas avant un

150

AFRICAN LINGUISTICS

plateau mélodique dans un mot. Les triples notations suivantes nous permettront d'il­ lustrer ces différents phénomènes: notation notation des notation de l'accent et plateaux tonale du point d'ancrage mélodiques ahe

ahe

ateme

ateme

areebe

areebe amuhe

amuhe' akimu

he

akimuhe amuteme

amuteme akimu

temere

akimûtemere \

\

ρ \ j *

amuree be akimu * reebere

amureebe

amurore

amurore

4.2.2.

akimureebere \

'

\

f

/

Conclusion

Le subjonctif immédiat est marqué au niveau tonal par le scheme entièrement haut des noyaux, lequel est la mani­ festation d'un accent placé sur le dernière more du noyau. Ces schemes nous ont également permis de cerner l'unité accentuelle et tonale qu'est le thème tel que nous l'avons défini dès le début de l'étude des schemes tonals des infinitifs. 4.3. Schemes tonals des verbes à l'indicatif immédiat, récent, prétérit négatif et accent Nous suivrons toujours la même démarche: nous déga­ gerons d'abord les schemes tonals puis nous montrerons qu'ils sont interprétables comme la manifestation d'un accent. 4.3.1. diat,

Les récent

schèmes prétérit

tonals

des négatif

verbes

à 1'indicatif

immé­

4.3.1.1. Les schemes tonals des verbes à l'indicatif immédiat imperfectif négatif Ce temps est marqué segmentalement par la présence

TONS ET ACCENTS EN KINYARWANDA

151

de la marque de négation nti/si. La voyelle finale de ces formes est le a, comme dans les formes affirmatives cor­ respondantes. Au niveau tonal, les schemes des noyaux sont entièrement de tons hauts comme ceux des verbes au sub­ jonctif que nous venons d'étudier. - Les schemes tonals des verbes à noyau d'une more exemples: figure 62

figure 63

152

AFRICAN LINGUISTICS

figure 64

figure 65

figure 66

TONS ET ACCENTS EN KINYARWANDA

153

La forme /ntlbáha/ présente un scheme bas-haut-haut. Son dernier ton haut est réalisé descendant lorsqu'il se trouve en position finale d'énoncé (figure 62). La forme /ntaamûha/ est, elle, de scheme bas-bashaut-haut. Son dernier ton haut est réalisé descendant en position finale d'énoncé (figure 63) mais haut en position non finale (figure 64). La forme /ntaakîmuha/ présente en fin de scheme la succession tonale haut-bas. Son ton haut préfinal est ré­ alisé descendant en fin d'énoncé (figure 65) mais haut à l'intérieur de l'énoncé (figure 66). La forme /ntibahá/ est constituée de la marque de né­ gation nti du préfixe sujet ba et du thème ha. Comme les formes du subjonctif vues précédemment, les formes de ce temps négatif comportent au moins deux tons hauts. Lorsque le thème est monosyllabique, le deuxième ton haut se trou­ ve porté par le préfixe sujet lorsqu'il précède immédiate­ ment ce thème, d'où le ton haut porté par le préfixe ba. La marque de négation nti/si porte un ton bas. Nous ver­ rons qu'elle garde toujours un ton bas dans ces formes né­ gatives de l'indicatif immédiat. La forme /ntaamûha/ est constituée de la marque de négation nti du préfixe sujet à, les deux se contractant en ntaa, de l'infixé mu et du thème ha. Le préfixe mu por­ te ici le deuxième ton haut. La forme /ntaakimúha/ est constituée également de la marque de négation nti, du préfixe sujet a (contractés en ntaa) de deux infixes ki et mu. La marque de négation ntï et le préfixe sujet sont tous les deux de ton bas. Les in­ fixes ki et mu reçoivent un ton haut. Le thème n'a pas gardé son ton haut placé en position impaire. - les schemes tonals des verbes à noyau de deux mores exemples :

.154

AFRICAN LINGUISTICS

TONS ET ACCENTS EN KINYARWANDA

155

La forme /ntaatema/ comporte deux tons hauts en fi­ nale. Son dernier ton haut est réalisé descendant puis­ qu'il est en finale d'énoncé (figure 67). La forme /ntaàmutémk/ comporte un scheme haut-bas en finale. Le ton prêfinal est réalisé descendant en finale d'énoncé (figure 68) mais haut en position interne (fi­ gure 69) . La forme /ntaakimùtemera/ comporte deux syllabes à ton haut, les syllabes ki et mu, les autres étant à ton bas. La forme /ntaatema/ comporte la marque de négation ntï, le préfixe sujet a, contractés en ntaa, et le thème tema. La marque négative et le préfixe sujet présentent tous les deux le ton bas qui leur est propre. Le thème témà comporte deux tons hauts. La forme /ntaamutema/ est constituée de la marque de négation nti, du préfixe sujet a de l'infixé mu et du thème tema. La marque négative et le préfixe sujet y sont toujours de ton bas. L'infixé mu reçoit un ton haut. Le dernier ton haut du thème tema étant en position impaire se trouve supprimé. La forme /ntaakimùtemera/ comporte un infixe de plus que la forme précédente. Ce deuxième infixe, prend, comme le premier infixe un ton haut. Deux tons hauts se trouvent donc ainsi portés par les infixes. Ce sont les deux pre­ miers tons hauts de la forme verbale et les deux seuls maintenus, les autres, c'est à dire ceux du thème ayant été supprimés. - les schemes tonals des verbes à noyau de trois mores. exemples: /ntaaréeba/ "il ne regarde pas" /ntaamureeba/ "il ne le regarde pas" /ntaakîmureebera/ "il ne regarde pas pour lui" La forme /ntaaréeba/ est constituée de la marque de négation ntï du préfixe sujet a et du thème -rèébâ. La marque de négation et le préfixe sujet présentent des tons bas, comme dans les autres formes vues précédemment. Le thème -rééba comporte autant de tons hauts qu'il a de syl­ labes comme le thème du subjonctif. La forme /ntaamureeba/ comporte un infixe. Cet in­ fixe reçoit un ton haut, à la différence de ce que nous avions vu dans les schemes du subjonctif où un infixe seul, ne recevait pas de ton haut. Seulement deux tons hauts sont conservés: les deux premiers, ceux des syllabes mu et re. La forme /ntaakîmureebera/ comporte deux infixes, tous les deux porteurs de ton haut. Seuls deux tons hauts étant admis dans la forme verbale, les deux tons hauts

156

AFRICAN LINGUISTICS

suivants, à savoir ceux du thème sont supprimés, autres exemples: ntïbâs "ils ne se ressemblent pas" inf.: gùsh ntibakorà "ils ne travaillent pas" inf. :

gùkora.

ntibawukora. "ils ne le travaillent pas" ntaamená "il ne jette pas" (un liquide) inf.:

kumena

ntaakimèna ntibaboha inf.:

"il ne le jette pas" "ils ne lient pas"

kúbóha

ntlbàmibohà ntibakuúndá inf.:

"ils ne le lient pas" "ils n'aiment pas"

gukuunda

ntibamúkúunda "ils ne l'aiment pas" ntlbahúungá "il ne vient pas d'aimer" inf. :

guhîiùnga

ntïbamuhuùnga "ils ne le fuient pas" 4.3.1.2. Les schemes tonals des verbes à l'indicatif immédiat perfectif Ce temps ne se différencie de son homologue imperfectif que par sa terminaison ye et les modifications morpho­ logiques associées au ye. /ntaatemyé/ "il ne vient pas de couper" /ntaamùtemue/ "il ne vient pas de le couper" /ntaaroyè/ "il ne vient pas de regarder (vite)" /ntaamùroye/ "il ne vient pas de le regarder" (vite)" /ntaakúúnze/ "il ne vient pas d'aimer" /ntaamukiùnze/ "il ne vient pas de l'aimer" 4.3.1.3. Les schemes tonals des verbes à l'indicatif récent imperfectif négatif Ce temps est marqué segmentalement par la marque de négation nti , la marque préposée a et la terminaison · Au niveau tonal, il est marqué par le scheme entièrement à tons bas des thèmes. Ces formes verbales ne comportent donc que des tons bas. exemples:

TONS ET ACCENTS EN KINYARWANDA

/ntïyatemaga/ /ntïyamùtemaga/ /ntïyareebaga/ /ntïyamùreebaga/

157

"il ne coupait pas (il y a un in­ stant)" "il ne le coupait pas" "il ne regardait pas" "il ne le regardait pas"

4.3.1.4. Les schemes tonals des verbes à l'indicatif récent perfectif négatif Ce temps ne se différencie de son homologue perfec­ tif que par la terminaison ye et les modifications mor­ phologiques associées au ye. exemples /ntïyatemye/ "il ne vient pas de couper" (action achevée) /ntïyamùtemye/ "il ne vient pas de le couper" /ntïyareebyh/ "il ne vient pas de regarder" /ntïyamùreebye/ "il ne vient pas de le regarder" 4.3.1.4. Les schemes tonals des verbes à l'indicatif prétérit imperfectif négatif Ce temps a segmentalement les mêmes formes que 1'in­ dicatif récent imperfectif négatif. Par contre, les noyaux ont des schemes entièrement à tons hauts et non à tons bas. exemples : /ntïyatemaga/ "il ne coupait pas" /ntïyamùtemaga/ "il ne le coupait pas" /ntlyareebaga/ "il ne regardait pas" /ntïyamùreebaga/ "il ne le regardait pas" Dans ces formes, la marque de négation nti reste tou­ jours de ton bas. La syllabe ya qui suit et qui résulte de la contraction de la marque a et du préfixe a reçoit un ton haut. D'autre part, on observe une limitation à deux du nombre de tons hauts. Nous avons donc posé que le noyau était de scheme totalement haut mais sans que celui-ci ne puisse jamais se manifester comme tel. 4.3.1.5. Les schemes tonals des verbes à l'indicatif prétérit perfectif négatif Ce temps a segmentalement les mêmes formes que l'in­ dicatif récent perfectif négatif et tonalement les mêmes schemes que son homologue imperfectif. exemples: /ntïyatemye/ "il n'a pas coupé"

AFRICAN LINGUISTICS

158

/ntïyâmutemye/ /ntïyâréebye/ /ntlyámireebye/

"il ne l'a pas coupé" "il n'a pas regardé" "il ne l'a pas regardé"

4.3.2. Les schemes tonals des verbes à 1'indicatif immé­ diat, récent et prétérit négatif et accent 4.3.2.1. Les schemes tonals des thèmes et l'accent Les noyaux peuvent présenter deux types de schemes to­ nals, à savoir (a) a aa aaa (a) a aa aaa ou en termes accentuels: (a) a aa aaa (a) a aa aaa Ces schemes tonals et accentuels ne sont pas des schemes lexicaux; ce sont des marques des différents temps négatifs. 4.3.2.2. Schemes tonals de ces formes verbales néga­ tives et accent Lorsque le thème ne comporte que des tons bas et est donc sans accent, la forme verbale demeure sans accent et donc entièrement de ton bas. Considérons maintenant le cas où le verbe comporte des tons hauts. Comme dans les formes du subjonctif, le scheme des formes sans infixe est entièrement prédictible à partir de l'accent du thème et de l'exigence d'un minimum de deux tons hauts, exemples notation notation notation de l'accent des plateaux tonale mélodiques ntibaha' ntibâhâ ntibáhá ntaatema ' ntaatëma" ntààtéma ntibareeba' ntibarëeES ntibàrééba Dans les formes avec infixe ou avec marque temporelle apparaît un point d'ancrage qui est la syllabe après le préfixe sujet, que celle-ci soit le premier infixe ou la marque temporelle. C'est le point de départ du plateau mé­ lodique. D'autre part, avec la marque temporelle ou les in­ fixes, se trouve introduite la condition d'un plateau mé-

TONS ET ACCENTS EN KINYARWANDA

159

lodique de deux syllabes. L'accent se déplace pour maintenir un plateau mélo­ dique de deux syllabes par rapport au point d'ancrage, comme l'illustrent les exemples suivants exemples : notation de l'accent et du point d'ancrage ntibamuha ntibamuko

ntibamuku ntiyamu

notation tonale

ntibamuha ra

ntibaboha ntibamubo

notation des plateaux mélodiques

ntibamukora ntibaboha

ha

ntibamuboha

unda

ntibamukuunda

temye

ntiyamutemye

ntiyamu

reebye

ntiyamureebye

ntiyamu

temaga

ntiyamutemaga

Dans les trois dernières formes, le préfixe sujet et la marque temporelle a sont contractés et portent le point d'ancrage que reçoit la marque temporelle. 4.3.3,

Conclusion Les schemes tonals de ces formes restent toujours prédictibles à partir d'un accent, de conditions de lon­ gueur du plateau mélodique et d'un point d'ancrage qui n'apparaissent qu'avec l'introduction d'une marque tem­ porelle ou d'un infixe. 4.4. Schemes tonals des verbes au subjonctif négatif et accent Ce temps est marqué segmentalement par la marque de négation nti/si. La voyelle finale de ces formes est le e, comme dans les formes du subjonctif affirmatif corres­ pondantes. Au niveau tonal, les noyaux ont les mêmes schemes de base que le subjonctif et l'indicatif négatif, à savoir des schemes entièrement hauts. Nous verrons ce­ pendant que ceux-ci ne peuvent jamais se manifester comme tels.

160 4.4.1.

AFRICAN LINGUISTICS Les

schèmes

4.3.1.1. more exemples :

tonals

des

verbes

au subjonctif

négatif

Les schèmes tonals des verbes à thème d'une

figure 71

figure 72

"qu'il ne lui donne pas" La forme /ntààhe/ (figure 71) est de scheme haut-haut -bas. Le ton préfinal porté par la deuxième partie de la voyelle longue de ntaa est réalisé descendant, cette forme étant ici en fin d'énoncé. La forme /ntaakïmuhe/ comporte deux tons hauts ini­ tiaux suivis de tons bas. La forme verbale /ntââhe/ est constituée de la marque de négation nti , du préfixe a et du thème he.

161

TONS ET ACCENTS EN KINYARWANDA

Le ton haut du thème se trouvant en position impaire est supprimé. La forme verbale /ntaakïmùhe/ est constituée de la marque de négation nti, du préfixe sujet a, de deux in­ fixes ki et mu et du thème hé. Seuls sont maintenus les deux tons hauts initiaux d'une succession de cinq tons hauts qui comprendrait le ton haut du thème ainsi que les tons hauts reçus par la marque de négation,- le préfixe sujet et les deux infixes. 4.3.1.2. Les schèmes tonals des verbes à noyau de deux ou trois mores exemples : /ntaateme/ "qu'il ne coupe pas" /ntaamuteme/ "qu'il ne le coupe pas" /ntaaklmutemère/ "qu'il ne le coupe pas pour lui" f /ntààreebè/ "qu il ne regarde pas" /ntáamure ehe/ "qu'il ne le regarde pas" /ntááklmureebere/ "qu'il ne le regarde pas pour lui" Dans toutes ces formes, la marque de négation et le préfixe sujet portent un ton haut, alors qu'ils portaient un ton bas dans les formes négatives de l'indicatif. Deux tons hauts seulement sont présents dans ces for­ mes. 4.4.2.

Schèmes

tonals

et

accent

Les schemes tonals de ces formes négatives peuvent s'expliquer à partir de noyaux entièrement de tons hauts, c'est à dire avec un accent sur la dernière syllabe. Ce­ lui-ci se trouvera déplacé de façon à constituer un pla­ teau mélodique de deux mores par rapport à un point d'ancrage qui est la marque négative, exemples: notation de l'accent et du point d'ancrage ntaa'ne ntaa'muhe ntaa'kimuhe ntaa'terne ntaa'muteme ntaa'kimutemere

notation des plateaux mélodiques

notation tonale

ntââhe ntââmuhe ntaâkimuhe ntaaterne ntlamuterne ntalkimutemere

ntááhe ntaamùhè ntaaklmùhè ntaatèmè ntaamùtèmè ntaakimùtèmèrè

AFRICAN LINGUISTICS

162

ntaa'reebe ntâareebe ntaarèèbè ntaa'mureebe ntââmureebe ntaamùrèèbè ntaa'kimureebere ntaakimureebere ntaakimùrèèbèrè Au niveau accentuel, ces formes se différencient des formes négatives de l'indicatif uniquement par leur point d'ancrage sur la marque négative. Ce point d'ancrage con­ stitue donc une marque de ce mode négatif. 4.4.3.

Conclusion

L'étude des schemes de ce mode négatif nous a fourni un bon exemple pour illustrer le rôle du point d'ancrage comme marque spécifique d'un temps donné.

5. CONCLUSION SUR L'ETUDE DES SCHEMES TONALS VERBAUX Pour rendre compte des schemes tonals de chaque forme verbale, nous n'avons eu qu'à spécifier: -l'accent -le point d'ancrage et les conditions de longueur du plateau mélodique (pair ou double) Ce système explicatif est particulièrement économique puisque qu'il ne requiert pas plus d'un accent par forme verbale. Il n'y a donc au maximum une seule syllabe marquée introductrice d'un plateau mélodique.

6.

CONCLUSION GENERALE

Nous avons donc proposé une réanalyse du système du kinyarwanda du niveau phonétique jusqu'au niveau accentuel en passant par le niveau tonal. Et, nous avons vu ap­ paraître une structure rythmique et mélodique originale. Il s'agit d'une langue où les mots peuvent être avec ou sans accent. Son accent peut être dit libre dans le mesure où il peut toucher n'importe quelle syllabe du thème mais le rythme est introduit au niveau de la mani­ festation de cet accent tonal ou les tons se regroup­ ent dans des mesures à temps pairs.5

TONS ET ACCENTS EN KINYARWANDA

163

FOOTNOTES

1. Ce travail a été fait de décembre 1980 à septembre 1981 à Paris, principalement au laboratoire de phonétique de Paris V où ont été ob­ tenus tous les tracés présentés ici. Nous avons pu disposer une jour­ née par semaine de décembre 1980 à juillet 1981 des installations de ce laboratoire et visualiser autant de courbes mélodiques qu'il nous a semblé necessaire sur l'écran d'un oscilloscope (Gould), la détec­ tion de mélodie étant assurée par un mélographe conçu par le C.N.E.T. (Centre National d'Etudes des Télécommunications). 2. Les courbes ont été décalquées sur celles obtenues grâce au mé­ lographe et transcrites sur papier par 1'oscillomink. La segmentation a été f ai te à partir de l'os ci llogr arome non représenté ici. Les symboles placés sous les courbes indiquent à quel segment du mot correspond chaque partie de la courbe. Ce ne sont pas des symboles phonétiques: nous avons conservé à ce niveau la transcription phonologique pour des raisons de simplicité d'exposé. Ainsi, avons-nous gardé par exem­ ple, le symbole b alors que le /b/ se réalise [β], d'où d'ailleurs l'absence du "creux" qu'introduirait une occlusive sonore dans la courbe mélodique. 3. Nous opposons ici position finale à position non finale pour mettre en évidence l'effet de la position dans l'énoncé sur la réa­ lisation des tons. Il est évident que les tons des mots changent dans certains contextes mais nous pensons pouvoir expliquer ces changements à partir des schemes tonals que nous dégageons ici. En effet, ceux-ci sont pour nous principalement dus à des exigences rythmiques et à la formation d'unités accentuelles et tonales regrou­ pant plusieurs mots; ils feront l'objet d'une étude ultérieure. 4. La répartition ku/gu suit la loi de Dahl. La sonore apparaît lorsque la consonne initiale de la syllabe suivante commence par une consonne sourde et inversement. 5. Dans un travail en préparation, nous introduisons la notion de pied pour rendre compte de cette rythmicité.

AFRICAN LINGUISTICS

164

REFERENCES

Byarushengo, E.R. et L.M. Hyman 1981 Tonal accent in haya (ms). Byarushengo, E.R., L.M. Hyman et S. Tenenbaum 1976 Tone, accent and assertion in haya dans Studies Tonology, Hyman cd. 183-205.

in

Bantu

Coupez, A. et Th. Kamanzi 1961

Grammaire

rwanda

simplifiée,

Usumbura.

Coupez, A. et A.E. Meeussen 1961 Notation pratique de la quantité vocalique et de la tonalité en rwanda. Orbis 10, 428-433. Hyman, L.M. (éd.) 1976 Studies in Bantu Tonology, of Southern California.

SCOPIL 3, Los Angeles: University

Hyman, L.M. 1978 Historical Tonology, in Tone: a linguistic ed., Academic Press.

Survey,

V. Fromkin

Kagame, A. 1960

La langue

du rwanda

et

du burundi

expliquée

aux

autochtones,

Kabgayi. Kagoma, 1968 Kalema, 1977

F.. et E.W. Stevick Luganda basic course, U.S. Government Printing Office. J. Accent modification rules in luganda, Studies in African Linguistics 8, 127-141.

Kayoboke, F. 1980 La tonologie du kinyarbranda, de Paris III.

thèse de 3ème cycle, Université

Kimenyi, Α. 1976 Tone anticipation in kinyarwanda, dans Studies nology, Hyman ed., 167-181.

in Bantu

To­

McCawley, J.D. 1968

The phonological

component

of a grammar of Japanese

The Hague:

Mouton. 1978 What is a Tone Language ? dans Tone: a linguistic V.A. Fromkin ed., Academic Press.

Survey,

TONS ET ACCENTS EN KINYARWANDA

165

Meeussen, Α.Ε. 1959 Essai de grammaire rundi, Annales du musée royal du Congo belge, Tervuren, Belgique. Overdulve, C M . 1975 Apprendre la langue rwanda, The Hague-Paris: Mouton.

REMARQUES SUR LES SYSTEMES DE TRANSCRIPTION DE CES AUTEURS

1. Chez Overdulve, le ton haut sur voyelle brève désigne un ton haut antérieur, c'est à dire un ton dont la hauteur est "anticipée" sur la voyelle précédente, celle-ci étant porteuse d'un ton bas distinctif. Sur voyelle longue, le ton haut désigne un ton haut antérieur s'il est porté par la première partie de cette voyelle il désigne un ton haut postérieur lorsqu'il est placé sur la deuxième partie de la voyelle. Sa hauteur est alors "anticipée" sur la première partie de la voyelle. La définition du ton haut que donne cet auteur l'amène à noter un ton haut pour deux de nos tons hauts phonologiques. Le der­ nier de ses tons hauts se trouvera situé à l'emplacement de notre :·. accent puisque celui-ci coincide toujours avec le dernier ton haut. 2. Kimenyi (1976) distingue seulement deux tons au niveau phonolo­ gique mais trois au niveau phonétique où il pose un ton descendant. Sa notation phonologique est phonologique dans le sens générativiste et ne correspond donc pas à notre niveau phonologique. Malgré ces différences de niveau d'analyse, on peut remarquer que les divergen­ ces sont nombreuses avec notre notation et moins systématiques que celles d'Overdulve. 3. Chez Kagame, " désigne un ton moyen bref, "~ un ton moyen long, un ton haut bref et * un ton haut long. La notation de Kagame est donc phonétique, elle est assez précise; elle néglige cependant cer­ taines modulations qu'elles soient portées par des voyelles brèves ou longues. Les formes entre parenthèses correspondent aux formes en contexte. 4. Chez Kayoboke, désigne un ton montant et  un ton descendant, qui sont posés au niveau phonologique à côté d'un ton haut et d'un ton bas. Pour nous, il s'agit de réalisations phonétiques contextu­ elles.

AFRICAN LINGUISTICS 166

la nôtre

APPENDICE: Overdulve (1975)

Kimenyi (1976)

TRANSCRIPTION TONALE

Kagame (1960)

Kayoboke

(1980)

N.B. Pour rendre la comparaison plus complète, nous avons souvent eu recours à des mots de scheme sembla-, ble, à défaut d'exemples identiques.

ON TONE IN SUKUMA John Goldsmith (Indiana University and University of Chicago) 1.

INTRODUCTION

The tonal system of Sukuma, a Bantu language of Tan­ zania, has been insightfully described and analyzed by Richardson (1959). This note attempts to refine Richard­ son's analysis, placing it within the context of autosegmental phonology, and the general study of Bantu accent that is discussed1 in Goldsmith (1982) and Clements and Goldsmith (1983). Several issues arise in analyzing Sukuma from this perspective. For example, there is the thorny problem of the treatment of a High tone which displaces one, two, and sometimes three syllables to the right of the position which it underlyingly belongs to, and, secondly, the cor­ rect formulation of the Well-formedness Condition in the light of data from Sukuma, which, like many eastern Bantu languages, does not permit High tones to spread simulta­ neously over more than one syllable. Despite this apparent violation of the Well-formedness Condition, the rightward movement of this displacing High tone is sensitive to as­ sociation lines; the H tone will not displace if the syl­ lable to its right is already associated. Finally, the hybrid nature of the Sukuma system - a seeming hybrid of tone and accent - is of considerable interest; this char­ acter is strikingly illustrated by the fact that we find an underlying four-way contrast for the initial vowel of a noun stem. The first vowel can underlyingly be High toned, Low toned, have no tone at all, or be assigned a tone melody of the form Low+High tone, where the High is a rightward displacing High tone, of the sort which con­ stitutes the focus of this paper.

168

AFRICAN LINGUISTICS

Another interesting aspect of the Sukuma data con­ cerns the historical origin of the High tone that dis­ places so strikingly far to its right. Although tonal dis­ placement, both to the right and to the left, is very com­ mon in Bantu languages (see, for example, Goldsmith 1982), Sukuma has carried this process, it seems, to extremes. If we begin with the simple observation that Eastern Bantu introduced early on a rule shifting a High tone one syl­ lable to the right, we may be led to view Sukuma's drift towards a double displacement not merely as an increase in the number of syllables that the rule in question shifts the High tone, but rather as the result of a "cloning" or recreation of the High tone shift rule. Furthermore, the observed shift of a High tone by three syllables (rather than the more familiar two) ONLY occurs across word bound­ ary. This fact seems worthy of special attention, and in Section 7 below, I discuss Kiparsky's theory of lexical phonology, and its consequences for such systems such as Sukuma's.

2.

A FIRST LOOK AT THE SUKUMA TONAL SYSTEM

When we look at the tonal properties of verb stems in Sukuma, we quickly discover, as Richardson notes, that certain syllables contribute a High tone to a later syl­ lable. The syllable that eventually bears the High tone may be one, two, or even three syllables away. Further­ more, verb stems divide quite readily into those that are tonally inert (corresponding roughly to Proto-Bantu Low tone), and thus do not contribute one of these displacing High tones, and those that do contribute a displacing High tone (corresponding just as roughly to Proto-Bantu High tone)!.. See (1) . (1) a. Neutral Class sol 'choose' ku-sol-a kaab 'divide' ku-kaab-a lagal 'drop' ku-lagal-a gaagaan 'bustel about' ku-gaagaan-a s¡ 'grind' ku-s[-a ¡b 'forget' ku-i b-a a lu I 'dress up' ku-alul-a b. Displacing Class bon 'see' ku-bon-a + H laai 'sleep' ku-laal-á

TONE IN SUKUMA

tono I baabaat sy ¡b ogoh

'pluck1 grope one's way' 'spit' 'steal' 'fear'

f

169

ku-tonol-á ku-baabáat-a ku-sg-a + H ku-ib-a + H ku-ogoh-a*

The verbs in (1) are given first as just a stem, and then in infinitival form following the gloss. Any verb in (lb) , except ku-baabáat-a (which has already landed its own High tone), will contribute a High tone to a following vowel, circumstances permitting. The stems with one vowel contribute a tone two vowels (moras) later; this will be onto an extension vowel, if there are enough vowels with­ in the verb itself. This can be observed by adding the suffix -anij- 'simultaneously' to stems of the sort found in (1), as seen in (2). (2) Infinitives with -anij- suffix a. Neutral class ku-soI-an i j-a ku-tonol-anij-a ku-îb-anîj-a ku-alul-anij-a b.

High-tone class ku-bon-an f j-a ku-tonol-anif-a ku-jb-anfj-a ku-ogoh-ánij-a ku-laal-anij-a ku-sy-anfj-a

However, if the Η-tone should try to fall on the first syllable of a following noun by this principle, it will generally go one syllable further, as we see in (3). (3)

a.

a -ka -bon-a ba-l'm-a he-TENSE-see-Final Morpheme cultivators

b:

a -ka -sol -a ba-lim-a he-TENSE-choose-Final Morpheme cultivators

c.

ba-lim-a

'cultivators'

We return below to the proper treatment of the shift of the High tone off of the first syllable of a noun or ad­ jective. One central point of interest in this note involves the nature of such tonal displacement. A crucial property of this effect is that it is blocked by any "inherent" tone of a morpheme. That is, as we shall see, a vowel can be underlyingly associated with either a High or a Low I

AFRICAN LINGUISTICS

170

tone, "singleton" tones which do NOT displace; when this happens, the H tone which would normally shift onto or across this vowel will be blocked from displacing. This blockage, which occurs equally in the nominal and verbal systems, is familiar from other tonal systems, and is al­ so to some extent reminiscent of prosodie properties of opaque segments in vowel harmony systems. The data from Sukuma forces us to address clearly with the way in which our theoretical model handles the case of rules whose ap­ plication is sensitive to unassociated or "floating" vow­ els. 3.

VERBAL INFLECTION PATTERN.

Consider, by way of introduction, an example from the verbal inflection system. The Class 1 (animate singular) subject prefix a- has no inherent tone, while the Class 2 (animate plural) subject marker has a displacing High tone, as we would expect etymologically. The tense marker -ku~ of the Future has no tone, and we find, therefore, the following pattern. (Here and throughout, I shall mark only High tone, unless otherwise indicated.) (4)

a. Toneless verbs stem -sol-: (i) a ku sol a (ii) ba ku sol a LH b. Displacing verb stem ■-bon-: (i) a  u bon a^ (ii) ba ku bon a 1  I / / ^ L H L H L h

Let us assume that a a certain point in the deriva­ tion, a point corresponding to what I have elsewhere called the Underlying Tone Level (see Goldsmith 1982, e.g.), a form like in (4aii) is represented as in (5). (5)

ba-ku-soI-a L H

In (5), only the L is associated; comparing (5) with (4a), we would furthermore like the L to associate, ultimately, with the syllable sol, not ku. One might conceive ex­ pressing this characteristic by means of a basic tonal pattern "LLH", or by means of a new mechanism insuring the double association of the L toneme, or finally with a phonological rule that would presume first the automat­ ic association of H to ku in (5), and which would simply

171

TONE IN SUKUMA

shift the H one syllable to the right. For present pur­ poses, I shall adopt the third solution. We will shortly see that the first proposal cannot work; the second solu­ tion has quite a bit in its favor, but I will not discuss it here. We will adopt, then, the rule of Η-shift, as in (6). I will assume, then, that universal principles ac­ count for the association of the H tone to the vowel fol­ lowing the one bearing the Low tone.

As I indicated above, (6) High Tone Shift only ap­ plies when the receiving vowel is itself unassociated. I have indicated this "floating", or unassociated, status with the superscript in (6) . The relevant data from the verbal inflectional system is presented in (7). (7) Subj neutral displ'g

Neutral verb stem (sol) No Obj Marker a ku sol a ba ku sol a

b. neutral displ'g

Neutral Obj Marker Displ Obj Mark a ku ba sol  a ku ku sol a ba ku ba sol a ba ku ku sol a

Displacing verb stem

(bon)

i ku ba bon a + H a ku ku bon a+H ba ku ku bon a+H ba ku ba bon a+H In each case, "+H" is used to indicate that a High tone will fall on the next word. Examples like ba ku ku sol a show that the tense marker ku is inherently toneless, rather than, say inher­ ently Low in tone. As we observe, the displacing High tone from the preceding subject marker displaces over ku with no interference. If it were associated with a tone, that would not be possible. Two important exemples*to observe in (7) are given here as (8) and (9). a ku bon a+H ba ku bon a+H

(8)

(9) The examples in (8) and (9) illustrate the fact that the second H of the basic LH tone melody will stay on its vowel when the following vowel is already associated (in this case, the H tone stays on the tense marker -ku-,

172

AFRICAN LINGUISTICS

since that is followed by the Object Marker -ba-). I shall assume, following a suggestion made by Halle and Vergnaud (1982), that the grammar of a language may contain the specification that vowels may not be doubly associated with tones. This specification may override the algorithm that adds association lines so as to maintain a Wellformed representation. In any event, this specification will pre­ vent the H tone from associating with any vowel in exam­ ples such as (8) . If this approach is correct, then it is clear from examples such as those in (9) that the melody involved in the case of a displacing H tone is in fact LH, as suggested above, and not LLH. If the basic melody were as in (10), then the H would be left unassociated here as well; but in fact the H does associate, as we saw in examples in (8) and in (4b). (10)

4.

LINKED TONES IN THE VERBAL MORPHOLOGY: ΚΑ-PAST.

In the ku-future tense, is made between singular and where the singular (class 1) plural (class 2) subject has ody. (11)

as we have seen, a contrast plural subjects as in (11), subject has no tone, and the a Low + High basic tone mel­

i. a

In the ka-past tense, however, the plural animate subjects all appear to have a fixed Low tone: (12) na-ka-sol-a 'I chose1 u-ka-sol-a 'you chose' a-ka-sol-a '(s)he chose1 tu-ka-sol-a 'we chose' mu-ka-soI-a 'you pi. chose' ba-ka-sol-a 'they chose' So far, then, singular and plural subjects appear to be alike in this sense; in particular, the plurals have no displacing H. But the plurals do have an inherent Low tone, unlike the singular subject markers, as illustrated in the data in (13). There we see that when the preceding word has a displacing H tone, the plural subjects refuse

TONE IN SUKUMA

173

to accept that displacing H, unlike the singular subject markers. (Parenthetically, a displacing High tone that shifts onto a verb from a preceding word can move no fur­ ther into the word than the first syllable; in this way, verbs differ strikingly from nouns. This is a very impor­ tant point, to which we will return below.) (13)

Form of the -ka- past when preceding word has a displacing High tone ná ka sol a ύ ka sol a á ka so I a

'tu ka sol a 'mu ka sol a 'ba ka sol a

(High tone on final syllable of preceding word.)

Thus the contrast between sungular and plural subject markers in this tense appears now to be as in (14). (14)

a.

a

However, a more revealing account is available by maintaining the same underlying form for the plural sub­ ject markers (those of the form in (14b)) as we had pos­ ited for the future (-ku-) tense, as in (8) or (9), that is, a linked Low with a displacing High tone. This dis­ placing High tone is, however, blocked from any tonal re­ alization by a basic Low tone on the tense marker -ka-, as in (15). (15)

-ka-

I

underlying form

L There are, however, two subject prefixes which are Low-toned without a displacing Η-tone, those of class 16 and 18 (ha, m u ) , subject markers for locative subjects. Consider the following examples, in which the verb is preceded by an element which is attempting to displace a High tone onto the verb (compare this with (13) above):

5.

TONOLOGY OF NOUNS.

As I noted at the beginning of this paper, my obser­ vations are heavily dependent on the work of Richardson. When faced with the tonal behavior of the noun stems, Ri­ chardson declared, "...such is the nature of Sukuma tonal

174

AFRICAN LINGUISTICS

structure that the analysis of the tonal contours of some of its independent nomináis has produced not so much a manageable system of tone classes, as an enumeration of the different combinations into which elements with vari­ ous kinds of tonal distinctiveness may enter. Under these circumstances all that can be done by way of analysis is to assign tonal distinctiveness to the constituents marking them in such a way that the symbols may serve the same purpose as the letter of number which indicates a tone class in other languages." (p. 24). Richardson is too modest; with some minor adjustments, the analysis he proposes can be recast in autosegmental terms, and can be used to shed light on more general questions of the treat­ ment of tone and accent in Bantu. The first characteristic of the nominal system is that it can be factored into two parts: in addition to a large set of syllables in which proto-Bantu H corresponds to a tone melody of the form LH, where the H displaces, and in which proto-Bantu L corresponds to no tone or ac­ cent (much as we have already seen in the verb stems), there are also additional fixed or linked L and H tones in the nominal stems. The following is an exhaustive list of the tonal types of noun stems to be found in Richard­ son's work: (17).

a.

b.

'chief

(P-B *tem])

See also -sumbi 'stool' (from P-B *tymbf 'owl', in this same class, and ginhi which takes no overt prefix, 'relative (family)' (P-B *dugu)

c.

'leper' see also H (this class differs from  only by being vowel initial, which af­ fects the tonal behavior)

d.

- I i ma

e . ' l a m p ' f.

(cf. class a, whose H dis­ places)

TONE IN SUKUMA

175

Surprisingly, what we find here in the nominal system is essentially a potpourri of tone patterns representative of a true tone system (and Richardson indicates that if one looked into the polysyllabic stems, one would find an even richer set of tonal combinations) - LH melody, HL melody, etc. - plus additionally the displacing H pattern described by the "L-H" melody. The sum total of the gener­ alizations that can be drawn seem to be these, not entire­ ly unfamiliar, ones: if there is a tone melody (other than the L-H melody) in the underlying specification of a word, it is associated with the vowels of the word in a one-toone fashion, starting from the left. In addition, a syl­ lable that is not assigned a tone in this tone-language­ like fashion may be associated with the basic accentual melody L H, which has the general behavior described in earlier sections. As I indicated above, there is not enough data in Richardson's description to hazard a propos­ al concerning the historical origin of the distinction be­ tween those original High tones which became displacing High tones and those which remained fixed High tones. It would, of course, be of extraordinary interest if the second class turned out to have a different historical or­ igin (late borrowings, for example); certainly this re­ mains to be investigated. Noun stems with an L H melody send a H tone to the next word, as in (18). Stems like 'tala1, however, with one L tone and one H tone do NOT send an H tone to the next word. (18)

Other things being equal, the displacing High tone from the first noun would appear on the first syllable (the prefix ba) in (18a) . There is, however, a rule of some generality that shifts a displaced High tone off the first syllable of a noun, shifting it to the right. In most cases, this syllable from which the High tone is shifted is a class prefix; this is not always the case,

176

AFRICAN LINGUISTICS

however. It may be the class prefix plus something else, if the two together form a single syllable; and the pre­ fix itself may be phonologically null. (29) Noun and Adiective Hiqh tone shift

A linked tone, regardless of whether it is the L part of an L H melody or a "singleton" tone, will block the spreading of a High tone displacing from an earlier word. Thus, while a displacing High tone from the stem -bon- ap­ pears on the second syllable of a word like (17b) or (17d), the displacing High cannot move onto the stem tone past the prefix syllable with a word of the (17a), (17c), (17e) category, etc. This contrast is illustrated in (20). (20)

The phrase-final tonal characteristics of these nouns are interesting. Words of the type in (17a), that is, those with non-final LH melodies, are described as ending with a Low tone when phrase final (that is, [batemi]). Those of the form (17b), or (17c), with final LH melody, are said to end on an extra Low tone when phrase-final (that is, [badugu]). This observation can be directly ex­ pressed within the analysis presented here if the phenom­ enon is taken to be intonational. That is, phrase-finally, there is still only a two-way contrast between High and Low, but the High is at the pitch-level of an earlier Low, and the Low is at Extra Low level. Non-associating Hs do not associate, of course, and are thus not realized. This is illustrated in (21). (21)

TONE IN SUKUMA

6.

177

TENSE MARKERS AND FINAL MORPHEMES

The most interesting part of a Bantu tonal system from a Bantu comparativist's point of view is in the tense markers and the Final Morpheme. The Final Morpheme in the verb, along with the Tense Marker following the subject concord marker, indicates grammatical information pertain­ ing to tense, mood, and aspect. We have already noted that some Final Morphemes have no tone at all, as, for example, in the -ka- past tense and the -ku- future tense discussed above. There are oth­ er tenses, however, whose Final Morpheme has a LH melody (of which the H will shift onto a following word, if there is one). One such tense is the Near Future conditional. Here the Tense marker is a fixed Low tone; see (22). (22) he-TNS-it-choose-FM Because of the fixed tone on the Final Morpheme, the H from the tense marker cannot undergo rule (6) High Tone shift. However, in all tenses with a tone on the Final Morpheme, the Final Morpheme tone is not, curiously, strictly limited to the Final Morpheme. When the Final Morpheme does have a tone or a Tone Melody, it spreads leftward "as far as possible", in a sense which we shall now discuss. If we add extensions to the example in (22), we would expect (23a). In fact, we find (23b). (23) a. a

This back-spreading is an instance of the wide-spread phenomenon first noted by A.E. Meeussen, whereby the Fi­ nal Vowel's tone spreads over post-radical vowels. Let us consider the Sukuma formulation of this rule as in (24). (24)

It is worth noting that one natural, alternative hypothe­ sis which does not involve such a rule at all will not work. This alternative would be to say that the tones which come from the object marker, the verb stem, and the Final Morpheme are all mapped onto their counterparts in

178

AFRICAN LINGUISTICS

a one-to-one fashion, from left to right. This does, in fact, work for a large class of cases, but fails to work in one tense, a case where the approach describe in (24) has no difficulty. This case is where the tense marker has no tone, but the Final Vowel has a L+H melody, as in Tense 19, the past habitual Motional. We find there forms such as u-aka-sol-aga + displacing H, with the neutral verb stem sol, rather than u-aka-sol-aga. Returning to the data in (23), then, rule (24) spreads the tone back over the verbal unit before High Shift (6) can apply, thereby effectively blocking it from applying. As the pair of forms in (25) illustrates, rule (24) can be seen to work simply and straightforwardly if we take "verbal unit" in the specification of (24) to mean the unit beginning with the Object Marker, not beginning with the verb stem. We must also take the intitial tone association to associate the H as well as the L, and then rule (24) associates the L of the Final Morpheme as far as possible to the left to the first unassociated syl­ lable, but not the first syllable of the verbal unit. (Re­ call that this last clause is necessary because of the Tense 19 forms mentioned in the preceding paragraph). (25)

Habitual Negative tense

Although the details are somewhat surprising in the degree of their complexity, the point of rule (24) is clear. With it in hand, we can turn to the interesting be­ havior of the Tense Marker -a- of Tenses 3 and 4 (Immedi­ ate Past and Today Past, respectively). These Tense Mark­ ers pose important questions from a comparative stand­ point, and their synchronic treatment in related languages is a matter of some dispute currently; see especially the discussion of the cogate -a- in Tonga in Goldsmith (1982, 1983) and Halle and Vergnaud (1982). Consider the data in (26) from the Immediate Past.

TONE IN SUKUMA

(26) Subject singular

a.

Neutral stem

179

-sol-

plural

singular

plural

As noted in (26) , there is an indeterminacy in the data presented here (which is all the data Richardson gives) concerning whether the Final Morpheme has its own LH melody in (26b), the case of the accented stems. The issue could be settled in an instant; if these verbs have an extra-Low tone on the last vowel when phrase-final, then their final vowels have a LH melody; otherwise they do not. It is furthermore unclear whether the forms in (26a) do have a final displacing H tone, as I have indi­ cated in the examples. The otherwise unexpected mainte­ nance of the High tone on ku in u a ku sol a strongly suggests that there must be a Low FV tone; in other cases, this is accompanied by a displacing H tone. More observa­ tions will be required here. The main problem in this data comes from the first two forms. All things considered, it is clear that the tense marker -a- has a tone melody LH with a displacing High. As noted in Goldsmith (1983), this puts it squarely in line with languages such as Tonga, Luganda, Ruri, and Digo in having remnants of a past tense marker -a- which puts a High tone on a following syllable. However, when the subject marker consists solely of a vowel (i.e., with classes 1, 4 and 9; 3rd singular is class 1 ) , two unusual things occur. First - and this is true across the lan­ guage - when the following syllable begins with a vowel, itself associated with a tone, the two vowels together form a unit which will not accept a displacing tone from a preceding word (i.e., (6) High Tone shift will not shift a High tone onto such a word). This suggests that a de­ rived syllable is formed by the union of these two vowels,

180

AFRICAN LINGUISTICS

and this syllable, containing a tonally associated vowel, will not undergo rule (6). Second, in this tense, and the very similar Today Past (Tense 4), a secondary regularity is observed. If the subject marker is of class 1, 4, or 9 (that is, consists of a single, vocalic segment), and there is no object marker, the High tone of the Tense Marker is irregularly displaced one syllable further to the right. While a number of similar phenomena have been noted in related languages (cf. especially the displacing effects of monosyllabic subject markers on the tone of the Far Past Tense Marker in KiRundi; on this, see Meeussen (1959)), and again while the derived syllable seems to play a central role in this progress, I have no ac­ count for it, and will leave the question pending, await­ ing further data.

7. SPECULATIONS ON THE HISTORICAL EVOLUTION OF SUKUMA, DISJUNCTIVITY, AND LEXICAL PHONOLOGY. In this final section, I would like to discuss the possibility that lexical phonology, as discussed by Kiparsky (1982) and others cited there, may give us some insight into how Sukuma got to be the way it is. From a historical point of view, the main puzzle that Sukuma poses is certainly why the High tone displaces "doubly" within a word, and triply across word boundary. How could Sukuma have arrived at such a synchronic state? We know that rightward displacement by one syllable from the etymological (and synchronically) appropriate syllable was a wide-spread innovation in Eastern Bantu (cf., inter alia, Goldsmith 1982, and references cited there). Unfortunately, in the matter of historical change, most explanations must begin with an unexplained exogenous impulse, and this one is no different; I shall simply be­ gin this scenario at a point when Sukuma looked (as it must have) like the "simpler" displacing languages, such as Ci-Ruri (see Massamba (1983)) or Rimi (Schadeberg (1979)), where High tone could displace one syllable to the right. Thus at this point, some underlying High tones were realized one syllable to their right; as in Rimi, but un­ like Ruri, there were lexical contrasts, apparently, re­ garding which High tones displaced to the right (since, as we saw in the nominal system, not all High tones dis­ place to the right). Any synchronic account of Sukuma, then, must have a

TONE IN SUKUMA

181

way of indicating which H tones displace in the manner discussed in this paper, and which ones, primarily in the nominal system, like the H of tali, in (17e), do not. The framework of autosegmental phonology implicitly suggests that the contrast be accounted for by having the non-dis­ placing Hs underlyingly associated in the lexicon, and have the displacing H morphemes be underlyingly unassociated, or floating, and furthermore only appear as part of a Basic Tone Melody LH, of which the L tone is associated. We will return momentarily to just how far in the derivation the H tone must be kept floating; suffice it to say for the moment that at the point where (6) High Tone Shift applies, there must be a natural way to prevent it from applying to underlyingly associated High tones. The simultaneous presence of (6) High Tone Shift in Sukuma in addition to the reanalysis of the etymologically High tones as (synchronic) LH melodies is an unusual case strongly resembling what has been called rule replication, or rule mitosis. In such cases, a single original rule has split, or replicated, within the grammar so that its ef­ fects are felt at two distinct points in the derivation. Something similar has occurred in Sukuma. Here, however, the reanalysis of original High tone as a Low-High melody peacefully coexists with the High Tone Shift rule. The reanalysis of *H as a LH melody should have sufficed to eliminate the Eastern Bantu rule of High Tone Shift out of the grammar Sukuma (though perhaps some remnant of the original tone shift rule would still have been motivated across word-boundaries, depending on one's assumptions about the automatic association of the H tone in the LH melody). Apparently, however, High Tone Shift did not get lost along the way. It remained in place, now operating on a form that was already a lexicalization of its earlier ef­ fects. That is puzzling; but a still more puzzling matter remains. Present-day Sukuma contains, in addition to (6) High Tone Shift, the rule of (19) Noun and Adjective High Tone Shift, both of which I repeat here. (6)

High Tone Shift

182

AFRICAN LINGUISTICS

(19)

Noun and Adjective High Tone Shift

These rules have the same effect; that is, they dis­ place a High tone one syllable to the right. In addition, both rules are constrained so as not to displace a High tone onto a vowel that is already associated with a tone. Nonetheless, we can be quite sure that they are two sep­ arate and distinct rules. (6) clearly can feed (19), but it cannot feed itself; hence the two rules are distinct. That High Tone Shift, when properly formulated, will not reapply, or feed itself, shifting a High tone twice, can be easily seen from the data in (2), for example, within the verbal infinitive system. The displacement of the High tone off of the Final Vowel and onto the first syl­ lable of the object in (3), through (repeated here), is a typical example of how (6) High Tone Shift can feed (19) Noun and Adjective High Tone Shift. This

feeding is very common; whenever the Low of a Low-High melody occurs on the penultiate syllable of a verb, for example, High Tone shift will reassociate the High from the final vowel of the verb onto the first syllable of the following object. If the second syllable of the ob­ ject is tonally free, rule (19) will push the tone over one syllable further. A very curious fact about these rules, furthermore, was noted in the text just above (13) . Unlike the case where a displaced High tone has been first shifted onto the first syllable of a noun or an adjective, High Tone Shift (6) will not apply to move off a High tone that is associated with the first syllable of a verb. The reader should recall that this shift of a High tone onto the first syllable of the verb may occur automatically, with­ out the intervention of any rule, if the final vowel of the preceding word bears a Low-High melody. As Richardson (p.35) notes, a displacing High from a previous word will fall onto the first syllable of a verb in the -ku- tense, for example, as in (27), and will displace no further.

TONE IN SUKUMA

183

(27)

As Richardson notes, "In those examples, the infix -kuin no way prevents the realisation two syllables later of H tone due to class 2 [subject prefixes], but can never bear H tone associable with a preceding external element. This is a feature of all N [i.e., toneless - JG]D.V.P.s [subject prefixes] rather than of individual tense in­ fixes . " Let us revise High Tone Shift, then, as in (28), so the rule respects this condition. (28) High Tone Shift (revised)

Condition: where a is not present Granted, this addition to High Tone Shift is formally awkward; but the reality of the phenomenon seems to re­ quire it. However, the formal awkwardness of this modifi­ cation in (28) is compounded, I would suggest, by the further observation that the curious spot where High Tone Shift (if formulated as originally done in (6)) is con­ strained NOT to apply is the environment where (19) comes into play. That is, the place where the otherwise general rule of High Tone Shift does not correctly operate is where it would shift a High Tone off of the first sylla­ ble of a verb; but precisely there another rule, (19), (the reader will recall that we explicitly argued that it was a distinct rule) exists in the language that is writ­ ten to shift a High tone off the first syllable of a Noun or Adjective. Let us summarize these observations. The general rule of High Tone Shift applies first, and can in general feed (19), which shifts a High tone even further into a following noun or adjective. High Tone Shift itself, though, is inherently blocked from shifting a High tone from the first syllable of -a word, if that is where its structural description happens to be met (as in (27) above)). We are finding, then, that the later rule is some­ how exerting a blocking influence on the earlier, that is, constraining the domain of application of the more general rule, High Tone shift. It was, in fact, just this sort of disjunctive rule application that was proposed in some

AFRICAN LINGUISTICS

184

detail in Goldsmith (1983) for two accent shift rules in Tonga. More generally, it appears that the presence of two (potentially conflicting) commands in the grammar is interpreted by the "operating system" of the grammar, to employ a computer analogy, as restricting the range of ap­ plication of the more general command so as not to include the range of the more specific. This more general view of the function of the Elsewhere Condition, as it has come to be known since Kiparsky (1973), is also adopted in large measure in Kiparsky's theory of lexical phonology (Kipars­ ky (1982)). There Kiparsky suggests that cyclic phonological rules, all neutralization rules, are in an "elsewhere" relationship with lexical entries themselves. Kiparsky metaphorically suggests that the phonological content of a lexical item can be thought of as a phonological "rule", in order to preserve the letter of the earlier statement of the Elsewhere Condition, which governed the interaction of rules per se, rather than governing the interaction of two different sorts of entities, cyclic rules, on the one hand, and static lexical entries, on the other. But this ,strategem has the unfortunate effect of covering up the more general, regulative use being made of the Elsewhere Condition in Kiparsky1s general program. To return to the problem at hand. Let us simplify the rules of High Tone Shift (28) and Noun and Adjective High Tone shift by eliminating the Low tone in their de­ scription. It is true that all displacing High tones are preceded by an associated Low tone; but that is not a sufficient condition for being a displacing High tone (consider the case of tala, discussed above). In fact, as we shall see below, the rules of High tone shift will on­ ly apply to tones which have been associated post-cyclic­ ally, in Kiparsky's sense. The rules then become (29) and (30), where the familiar convention has been used that a dotted line indicates part of the structural change of the rule, and a cross association indicates a line that is deleted by the rule. (29)

High Tone Shift Condition: there may not be a word boundarv here. . ..

TONE IN SUKUMA

(30)

185

Noun, Adjective High Tone Shift

In sum, then, the odd negative condition in rule (30) will be automatically imposed by the Elsewhere condition in the version I have suggested here and in Goldsmith (1983), with the additional proviso that the blocking effect of (30) is extended from nouns and adjectives to all lexical items, for a reason that is not presently clear. Kiparsky has also suggested, within the context of Lexical Phonology, that neutralization rules only apply to derived forms in the first component in which they have an opportunity to apply (see Kiparsky (1982) for a de­ tailed exposition of this proposal). Derived forms, ac­ cording to this proposal, are forms which have, in rele­ vant ways, either been affected by a preceding phonolog­ ical rule in the component, or which contain material from separate morphemes. This proposal allows us to account for the non-ap­ plication of High Tone Shift ((6) or (29)) to H tones which are underlyingly associated, as in tala (17e). As­ sociation of floating H tones is clearly a non-lexical, or post-cyclic rule, since it can associate a floating H tone from one word with the initial syllable of a follow­ ing word. High Tone Shift is also clearly post-cyclic, since it can push a High tone from a final syllable to a word-initial syllable. Lastly, High Tone Shift only ap­ plies to tones associated in a derived way, that is, it fails to apply to underlyingly associated High tones. The Elsewhere Condition thus seems to be a highly useful devise in exploring the interaction of similar rules in Sukuma.

8.

CONCLUSION

In this note I have explored some of the theoretical consequences of Richardson's account of Sukuma tone. Many questions remain unanswered at this point, in part due to a lack of data, but autosegmental accounts of tone and of accent appear to be providing new levels of insight into the nature of the phonological systems of the Bantu lan­ guages.

186

AFRICAN LINGUISTICS

FOOTNOTE

1. A rather different approach may be found in Batibo (1976). I am grateful to Larry M. Hyman for useful discussion.

REFERENCES

Batibo, Herman 1976 "A New Approach to Sukuma Tone", In Hyman 1976. Clements, G.N. and John Goldsmith, eds. 1983 Autosegmental Studies in Bantu Tone. Dordrecht: Foris Press. Goldsmith, John 1982 Accent Systems. In van der Hulst and N. Smith 1982. 1983 Tone and Accent in Tonga. In Clements and Goldsmith 1983. Halle, Morris and J.R. Vergnaud On Autosegmental Phonology. In van der Hulst and Smith. Hyman, Larry, ed. 1976 Studies in Bantu Tonology. Los Angeles: Southern California Occasional Papers in Linguistics. Kiparsky, Paul 1973 Elsewhere in Phonology. In A Festschrift for Morris Halle, edited by P. Kiparsky and S. Anderson. New York: Holt, Rinehart and Winston. 1982 Lexical Phonology. In van der Hulst and Smith. Massamba, David 1983 Tone and Accent in Ci-Ruri. In Clements and Goldsmith. Meeussen, A.E. 1959 Essaie de grammaire rundi. Tervuren. Ri chardson, Irvine 1959 The Role of Tone in the Structure of Sukuma. London: School of African and Oriental Studies.

TONE IN SUKÜMA

Schadeberg, Thilo 1979 "Ueber die Tone der verbalen Formen im Rimi". Afrika und Uebersee. Band 57. van der Hulst, Harry and Norval Smith, eds. 1982 The structure of Phonological Representations. Dordrecht: Foris Press.

187

L'EXPRESSION DU PASSIF EN MANINKA

Cl. Grégoire (Musée royal de l'Afrique Centrale)

1.1. A notre connaissance, les constructions passives propres aux langues mande du groupe nord n±ont guère été décrites jusqu'ici, quoique leurs caractéristiques princi­ pales soient connues de ceux qui travaillent sur ce groupe linguistique. Il est curieux de constater, par exemple, que ni l'Etude descriptive de la langue susu (M. Houis, 1963), ni le Lehrbuch des Susu (M. Friedländer, 1974), ni la Grammar

of

Gambian

Mandinka

(E.C. Rowlands, 1959), ni

la Grammar of Vai (W.E. Weimers, 1976) ne traitent de cette question1. Quelques lignes y sont consacrées dans The structure of Faranah Maninka (R. Spears, 1965) OU dans les fascicules polycopiés que Kouloubali et Bailleul ont publiés sur la grammaire du bambara (s.d.). Ces deux au­ teurs notent du reste qu'en bambara "l'emploi du passif est plus fréquent qu'en français" et comme les données que nous avons recueillies sur le maninka de Kankan con­ firment cette observation, il nous a paru intéressant d'exposer de manière aussi complète que possible les pro­ cédures que cette langue utilise pour produire les énon­ cés passifs. Dans la mesure où notre documentation le permet, nous tenterons également de déterminer jusqu'à quel point les structures observées en maninka de Kankan sont représentatives de ce que l'on peut observer dans d'autres langues mande du groupe nord.

1.2. La majorité des énoncés maninka présentent une structure S Asp. (0) V. S'il est présent, l'objet nominal ou pronominal précède le lexeme verbal. La plupart des morphèmes de conjugaison se placent entre le NP sujet et

190

AFRICAN LINGUISTICS

le NP objet ou le lexeme verbal. Pourtant, certaines mar­ ques d'aspect sont suffixées (ou postposées) au lexeme verbal et quelques temps (comme le présent actuel affirmatif ou négatif2) utilisent à la fois ces deux types de morphèmes verbaux. Quant au complément de type ciconstanciel, il se place après le lexeme verbal et est accompa­ gné d'une postposition. Notons qu'il n'existe aucune mar­ que tonale ou autre qui permette de distinguer le sujet et l'objet, ceux-ci n'étant identifiables que par la place qu'ils occupent dans l'énoncé. On a par exemple3:

1.3. Comme on vient de le voir, il existe en maninka deux types d'énoncés qui se distinguent par la présence ou l'absence d'un constituant objet. Dans ces deux types d'énoncés, le verbe se conjugue de manière identique sauf sur quelques points. En effet, aux formes perfectives de l'affirmatif, s'il n'y a pas d'objet, le morphème verbal est -ri (suffixe au lexeme verbal) alors que, s'il y a un objet, le morphème verbal est kà (placé entre le sujet et l'objet). On a ainsi:

L'EXPRESSION DU PASSIF EN MANINKA

191

D'autre part, les verbes employés sans objet connaissent des formes st'atives et immédiates dont l'équivalent n'est pas attesté pour les verbes qui ont un objet, du moins dans le corpus recueilli1*. On a par exemple:

On pourrat conclure, de ce qui précède, qu'il existe en maninka des verbes intransitifs et des verbes transitifs s'oppo­ sant en deux catégories lexicales nettement distinctes. Les faits sont pourtant plus complexes car il est très fréquent qu'un même verbe puisse être utilisé avec ou sans objet dans des énoncés actifs du langage courant. Un tel état de chose pourrait justifier que l'on ne parle pas de catégories lexicales transitive et intransitive, mais qu'on parle uniquement d'emplois transitifs ou in­ transitifs, les verbes n'ayant aucun trait de catégorisa­ tion à cet égard. Dans les pages qui suivent, nous adop­ terons tout d'abord cette attitude et nous en discuterons par la suite, afin de voir si l'étude des constructions passives permet d'éclairer les problèmes que pose l'éven­ tuelle catégorisation des verbes. Nous nous limiterons pourtant à quelques suggestions dans ce domaine, le but de notre étude n'étant pas de résoudre une question dont la complexité nécessiterait un traitement détaillé.

AFRICAN LINGUISTICS

192

1.4. Les verbes derives du maninka se forment tantôt par la suffixation tantôt par la préfixation d'un morphème dérivatif au radical verbal simple. Nous n'envisagerons ici qu'un seul type de verbes dérivés: il s'agit de ceux qui sont obtenus par la préfixation du dérivatif la auquel les auteurs attribuent unanimement un sens causatif (ou factitif)5. Notons que l'adjonction de ce dérivatif entraîne la production d'un verbe qui est normalement utilisé avec un complément d'ob­ jet dans les énoncés actifs, quelle que soit la construc­ tion dans laquelle le verbe simple figure de préférence. On a ainsi: lo

const, intrans. ou réfl. "se lever" constr. trans, "dresser, construire" vs. la lo constr. trans, "mettre debout" wá constr. intrans. "aller" vs. láwa constr. trans. "envoyer" ye I £ constr. intrans. "monter" vs. layéis constr. trans, "soulever" sali constr. intrans. "prier" vs. la sali constr. trans. "faire prier, conduire la prière de" se constr. intrans. "parvenir" vs. lase constr. trans. "envoyer, faire parvenir"

1.5 Enfin, il est peut-être utile de rappeler qu'il n'existe en maninka aucune procédure d'accord verbal avec le sujet, le verbe ne portant ni marque de nombre ni mar­ que de personne.

EI.l. R, Spears (1965, p.92) reconnaît, en maninka de paranah, l'existence d'une transformation optionnelle de Êype6: f

NP

Asp.

NP'

Vtrans

*(->)*

NP' Asp.

VtraM ^

ilauteur ne parle à aucun moment de transformation passive; 11 commente la règle établie en signalant que "any string of the above type may delete the first NP and substitute for it NP' which is the object of the V  . Il illustre ses propos au moyen de plusieurs exemples parmi lesquels hous choisirons: ψ ψ

an + bara+rnsi+fà ψ nïsi+bara+fà ?

"we have killed the cow" "the cow has been killed"

L'EXPRESSION DU PASSIF EN MANINKA

193

Toutefois, p. 96, l'auteur cite, à propos d'autre chose, les exemples que voici: "you cut wood for me yesterday" "wood was cut by your hand for me yesterday" Ces énoncés montrent clairement que l'application de la transformation dont il vient d'être question n'implique pas la suppression du premier NP: celui-ci peut être postposê au lexeme verbal et il est alors accompagné de la postposition bolo. Des faits identiques peuvent être observés en maninka de Kankan et en maninka de Sigiri: par l'application de la transformation, le NP objet de­ vient sxijet et le NP sujet devient un complément qui pa­ raît bien être un complément d'agent. La transformation présente donc des caractéristiques telles que l'on peut, à première vue du moins, la considérer comme une trans­ formation passive. Nous la nommerons ainsi jusqu'à nouvel ordre et nous discuterons plus loin de la manière dont il convient de l'envisager, compte tenu des caractéristiques générales du système. L'existence de cette transformation en maninka de Kankan permet d'opposer les énoncés suivants:

femme a préparé le riz

le riz a été préparé par la femme

mouillé

été mouillé par la femme il ne le fait pas cela n'est pas fait par lui7 II·2. Un certain nombre d'observations peuvent être faites d'une part sur l'occurrence du complément d'agent et d'autre part, sur le traitement que subit la forme ver­ bale lors de l'application de la transformation passive.

194

AFRICAN LINGUISTICS

II.2.1. Tout comme la plupart des langues, le maninka at­ teste de nombreux énoncés passifs dans lesquels aucun com­ plément d'agent n'est exprimé. Il faut donc prévoir une procédure d'effacement de ce complément. On peut noter également que l'informateur admet beaucoup plus facilement que le complément d'agent soit exprimé si celui-ci est un substantif [ + animé] ou un pronom qui réfère à un substan­ tif de ce type. Si le substantif est [- animé], le complé­ ment d'agent tend à être refusé et il est très rare dans les énoncés produits spontanément. Nous en avons pourtant relevé quelques exemples, comme:

il a été tourmenté par sa jalousie qui est opposé, dans le même texte, à: sa jalousie l'a tourmenté ou comme:

, les animaux mouraient de faim en brousse, litt.: les animaux étaient tués en brousse par la faim Notons que bolo est également un lexeme nominal signifiant "bras, main" et qu'en tant que postposition, il figure soit avec le complément d'agent soit avec le substantif désignant le possesseur dans les constructions qui expri­ ment 1'équivalent d'"avoir" (cf. "la femme a un canari" litt.: "un ca­ nari est la femme à"). Or, le substantif qui désigne le possesseur est très généralement [+ animé] lui aussi. La restriction décrite ne paraît pas dépendre du sens propre au lexeme bolo puisque de nombreux lexemes désignant des parties du corps sont utilisés comme postpositions et peuvent alors accompagner des substantifs [- animé] (cf. kono "ventre" et /da: , "dans le pot", |sovkono| "dans le village"). Vraisemblable­ ment, la rareté des compléments d'agent [- animé] s'ex­ plique en maninka par des limitations d'ordre lexical analogues à celles qui fonctionnent dans de nombreuses langues avec des conséquences comparables. Enfin, il semble que, dans certains énoncés passifs, la présence du complément d'agent soit obligatoire ou, du moins, vivement souhaitée et que cette exigence soit due

L'EXPRESSION DU PASSIF EN MANINKA

195

à la nécessité d'éviter la confusion entre des sens dis­ tincts. Ainsi, notre informateur n'accepte pas la traduc­ tion "la vache a été tuée" pour l'énoncé ni si bàra fà ι cité par Spears; cet énoncé signifie selon lui "la vache est morte", alors que nïsi bàra fàz kamberem bolo signi­ fiera "la vache a été tuée par le jeune homme". Il se peut qu'il y ait là des différences dialectales mais, comme le comportement linguistique de notre informateur a été cohérent et constant sur ce point, il est nécessaire de chercher à l'expliquer: nous y reviendrons plus loin. Notons que, d'après Kouloubali et Bailleul, le bambara utilise la postposition fï pour marquer le complément d'a­ gent (cf. a bugora à fa ft "il a été battu par son père"). Cette postposition a, fondamentalement, un sens locatif, ce qui est intéressant dans la mesure où plusieurs langues bantoues utilisent une forme locative de classe 17 pour exprimer le complément d'agent. D'autre part, en bambara, fï accompagne le substantif qui désigne le possesseur (cf.  b'1 f  ? "que possèdes-tu ?") et a donc des em­ plois comparables à ceux de la postposition bolo du maninka. Comme les deux langues possèdent les deux postposi­ tions, il semble bien qu'elles aient utilisé, pour expri­ mer le complément d'agent, une postposition dont les em­ plois habituels se prêtaient à une telle spécialisation. Enfin, il est important de signaler que le mandinka n'ex­ prime pas le complément d'agent8 et qu'il en est de même en soso. Dans cette dernière langue, nous avons pu relever des énoncés comme : la pluie a abîmé le champ le champ a été abîmé mais nous n'avons pas obtenu de phrase comportant un complément d'agent. L'informateur traduisait une phrase comme "le chien a été attaché par mon père" en utilisant l'énoncé actif correspondant m baba nam bàré xlrlxi "c'est mon père qui a attaché le chien"; il traduisait également le complément d'agent français en recourant à une subor­ donnée temporelle, comme par exemple dans: le champ a été abîmé quand la pluie est venue Les textes recueillis ont confirmé ce qui avait été ob­ servé au travers des questionnaires linguistiques. Il est sans doute difficile de conclure à partir d'une informa­ tion qui demeure fragmentaire et il faudrait diversifier les enquêtes relatives à l'existence de procédés permet-

196

AFRICAN LINGUISTICS

tant d1exprimer le complément d'agent, mais il n'est pas exclu que l'on puisse montrer que, dans les langues mande nord où ils sont attestés, ces procédés sont dus à une évolution relativement récente. C'est aussi ce que semble indiquer le fait que le bambara et le maninka, qui sont des systèmes étroitement apparentés, utilisent des post­ positions différentes pour marquer ce complément, alors que le mandinka emploie les mêmes postpositions dans le même contexte avec un sens très différent (cf. kodoo domota η fee, kodoo domo-ta η bulu "l'argent s'est trouvé dé­ pensé malgré moi, j'ai dépensé l'argent sans le faire ex­ près" ) 9 . II.2.2. Les problêmes que pose le traitement de la forme verbale lors de l'application de la transformation passive sont apparemment très semblables dans les différentes langues mande nord pour lesquelles une certaine documen­ tation peut être réunie. Les faits observables en maninka de Kankan paraissent donc représentatifs d'une situation beaucoup plus générale. Dans cette langue, puisqu'il n'exis­ te aucune espèce d'accord verbal, la transformation pas­ sive n'apporte pas de modification au verbe de ce point de vue. Pour le reste, la seule incidence que l'on puisse ob­ server est la suivante: le verbe qui, dans l'énoncé ac­ tif, a les morphèmes caractéristiques du verbe utilisé avec un objet prend, dans l'énoncé passif, les morphèmes qui caractérisent le verbe figurant dans une construction sans objet. Pourtant, étant donné qu'il existe très peu de différences entre les deux types de conjugaison, le verbe de l'énoncé actif et celui de l'énoncé passif cor­ respondant sont le plus souvent identiques. Ils ne dif­ fèrent qu'aux formes du perfectif affirmatif et par le fait que les formes statives et immédiates peuvent figu­ rer dans les énoncés passifs. On a ainsi: a), perfectif affirmatif:

la femme a préparé le poulet le poulet a été préparé par la femme elle a acheté du riz du riz a été acheté par elle

L'EXPRESSION DU PASSIF EN MANINKA

b)

197

immédiat:

quel mouton tueras-tu ? /sa: jumam fa:t'i: bolo/ Isaa juman faa-to ι bo Ι ο Ι id. (litt, quel mouton va être tué par toi ?) c)

statif:

Le couteau que j'ai acheté au marché est cassé

Le couteau qui a été acheté par moi au marché est cassé. A tous les autres temps ou aspects de la conjugaison, la forme verbale demeure inchangée et lfon a, par exemple, outre les énoncés qui ont été cités plus haut: elle préparera le poulet le poulet sera préparé par elle le poulet sera préparé ne mets pas d'eau dans la marmite id. (litt, que l'eau ne soit pas mise par toi...) Ainsi, dans des couples de phrases comme: /sèmba bàr'a: dàmun/ Isenbá4bárá à damuni l'éléphant l'a mangé et: /sèmba bàra damun/ Isenbá*bárá dámúnl l'éléphant a été mangé /à d'à: kè bi/ là df à k¿ bi'i il le fera aujourd'hui et: /à d f k£ bi/ là df ké bi Ί cela sera fait aujourd'hui /à ma jî rrun/ là má jPrrnn'l il n'a pas bu d'eau et: /a ma min/ la ma min l cela n'a pas été bu si le second énoncé peut être identifié comme résultant d'une trans-

198

AFRICAN LINGUISTICS

formation passive avec effacement du complément d1agent, c'est parce que le verbe qu'il contient est utilisé dans une construction intransitive, alors qu'il est normalement utilisé dans une construction transitive. Il semble donc que ce soit par rapport à ce que l'on sait du comportement normal de verbes comme dàmun, ki, min, etc.. que l'absence d'objet fonctionne comme une marque de passif. La chose est bien connue et a été décrite par plu­ sieurs auteurs qui lfont justement mise en rapport avec le fait que les langues mande nord maintiennent l'ordre  V des énoncés actifs contenant ce type de verbes, quel­ les que soient les conditions syntaxiques ou sémantiques. Des énoncés français comme "il mange", "il prépare" se­ ront traduits soit en ajoutant l'objet sur lequel l'action s'exerce le plus fréquemment (cf. I ye kinin damunna "tu manges du riz", i ye suman tïbila "tu prépares le repas", etc.), soit en utilisant une périphrase dans laquelle le verbe ki "faire" a pour objet un substantif dérivé signi­ fiant "l'action de.." (cf. î ye damunni kh "tu fais l'action de manger", i'ye tibili kl "tu fais l'action de préparer"). De même, si l'objet du verbe est une subor­ donnée qui est normalement postposée à la principale, une contrainte absolue impose l'emploi d'un pronom personnel explétif qui anticipe sur la subordonnée complément et se place devant le verbe de la principale. On a ainsi:

Les hommes ont dit qu'ils ne l'avaient pas vu (litt, les hommes l'ont dit qu'ils ne l'avaient pas vu)

Dis-leur de venir (litt.: dis-le leur qu'ils viennent) Notons qu'un petit nombre de verbes ont un comportement spécifique dans la mesure où, s'ils ne sont pas accompa­ gnés d'un objet nominal (ou pronominal) explicite, ils n'entrent pas dans la construction périphrastique qui comporte une forme nominale dérivée désignant l'action et le verbe ki "faire". Ils figurent, dans ce cas, à la forme réfléchie. On aura ainsi: tu as bu de l'eau tu as bu Il est difficile d'expliquer cette particularité, mais

199

L'EXPRESSION DU PASSIF EN MANINKA

une hypothèse sera formulée à ce sujet en II.4. Pour l'instant, on peut constater de façon générale que des verbes comme dámun et tibi paraissent bien être des verbes fondamentalement transitifs, puisque leur ap­ parition dans des constructions intransitives est liée à l'application de la transformation passive. II.3. D'autres verbes, comme kúnun "se réveiller", wá "aller", se "parvenir", etc.. peuvent, par contre, figurer sans objet dans des énoncés actifs. On a par exemple: il s'est réveillé il est allé.. il est arrivé, il est par­ venu C'est le dérivatif causatif de ces verbes qui apparaît dans les constructions actives, de sorte que l'on a: il l'a réveillé il a envoyé de l'argent / à ka k a w a n d i l i l à s e / parvenir le message,

là kà k à w à n d f - 1 f 4 le conseil

lá.sél

il

a

fait

Comme il est normal, les énoncés transitifs de ce type peuvent subir la transformation passive. Le verbe dérivé au moyen du morphème la figurera alors dans une construc­ tion intransitive. On pourra donc opposer les énoncés sui­ vants : A.l. 2. 3. 4.

/à kununna/ là kúnún-rál il s'est réveillé /à k'a: lákuniin/ là kà à lá.kunúnl il l'a réveillé /à lákununn'a: bolo/ là lá.kúnún-rá à bólól il a été réveillé par lui /à lákununna/ là là.kúnún-rál il a été réveillé

où, si l'on choisit un temps où il n'y a pas de différen­ ces dans les morphèmes selon que le verbe a ou non un ob­ jet: B.l.

/à

bára

kuniin/

2.

/à b a r ' a : veillé

3.

/à

bára

là

làkunun/ lakunun

bárá là

à bolo/

kúnúnl

bárá là

à

il

s'est

lá.kunúnl

bárá

lá.kúnún

réveillé il

l'a

ré­

à bo Ι ο Ι

200

4.

AFRICAN LINGUISTICS

il a été réveillé par lui /à bára lakuniin/ là bárá lá.kunúnl veillé

il a été ré­

On voit que A.l à kununnk et A.4. à lakununnà d'une part, B.l. a bara kunun et B.4. à bira lakunùn d'autre part, ne se différencient que par la présence du dérivatif la qui fonctionne, dans A.4. et B.4., comme une marque de passif. S'il assume ce rôle, c'est parce que la forme dérivée causative est normalement utilisée dans une construction transitive: son utilisation dans une construction intran­ sitive suppose que la transformation passive ait été ap­ pliquée. Le mécanisme décrit paraît montrer que des ver­ bes comme kunun "se réveiller" sont fondamentalement des verbes intransitifs. C'est l'adjonction du causatif la qui permet de les introduire dans des énoncés transitifs et la transformation passive ne peut donc s'appliquer qu'à des énoncés contenant cette forme verbale dérivée. II.4. Si l'examen des constructions passives semble mon­ trer que certains verbes maninka peuvent être considérés comme lexicalement transitifs ou intransitifs, la langue atteste un groupe de verbes que l'on ne peut à première vue classer ni dans la première catégorie ni dans la se­ conde. Si l'on considère, par exemple, une opposition comme: il est mort là-bas il a tué la vache on peut constater que fa : ne se comporte pas comme les intransitifs wa ou kunun puisqu'il apparaît dans une con­ struction transitive sous sa forme radicale. On pourrait penser que fa : est transitif et que les énoncés ou il est employé intransitivement correspondent au passif des énon­ cés où il est employé transitivement. Dans le premier exem­ ple cité, le sens de "mourir" devrait donc être inter­ prété comme "être tué". On a vu pourtant que cette inter­ prétation n'était pas admise par l'informateur, du moins dans les cas où l'énoncé ne comporte aucun complément d'a­ gent. De plus, avec d'autres verbes dont le comportement est identique à celui de fa : "mourir/tuer", une telle in­ terprétation ne paraît pas se justifier du point de vue sémantique. Par exemple, le verbe b¿ qui signifie "sortir" lorsqu'il est employé intransitivement (cf. à bora so kbno "il est sorti du village") peut être utilisé transi­ tivement dans des énoncés comme:

L'EXPRESSION DU PASSIF EN MANINKA

a il /à il /à il

201

ka fanim b a kan na/ la ka fanin b a kan lal a ôté le vêtement de sur lui, il s'est déshabillé ka bfm b / I à kà bin* b a enlevé les herbes, il a désherbé ka ku l um b; / là kà kulun4 b a fabriqué une pirogue

La signification "sortir" ne se laisse pas expliquer à partir du passif de "enlever" ou de "fabriquer". Ces deux derniers sens, par contre, peuvent être interprétés comme des causatifs dont la signification littérale serait "faire sortir" et cette explication convient d'autant mieux que bo est utilisé transitivement avec le sens de "fabriquer" lorsqu'il s'agit d'objets de bois que l'on façonne en sculptant dans la masse ou lorsqu'il s'agit de médicaments que l'on extrait de plantes ou d'autres sub­ stances . Notons que le sens "tuer" pour faz trans, peut également être interprété comme "faire mourir", c'est-àdire comme le causatif de faz intrans. "mourir". Il sem­ ble donc que les verbes de ce type soient des verbes in­ transitifs, mais qu'ils aient la particularité d'avoir une forme dérivée causative sans marque formelle. Cette hypothèse est confirmée par le fait que, dans des constructions transitives, ces verbes admettent la commutation entre la forme sans dérivatif et la forme dé­ rivée par adjonction de la. Le sens demeure inchangé mais, comme c'est le cas dans de nombreuses autres langues, il semble que le causatif sans marque formelle soit employé de préférence quand l'objet est considéré comme inactif (cf. maninka: à ka bîm jbo "il a enlevé les herbes"; fran­ çais: il a sorti son mouchoir) alors que le causatif à marque formelle est utilisé de préférence quand l'objet% est considéré comme actif (cf. maninka: alu 7c ' à : saman k a z libo bon na "ils l'ont tiré pour le faire sortir de la maison"; français: il a fait sortir son fils). Si l'utilisation du causatif est déterminée de la sorte, on peut expliquer que faz "tuer" s'oppose à lafa ζ "faire mourir", puisque tuer est conçu comme une action qui ne suppose aucune participation de l'objet, tandis que faire mourir consis­ te à exercer une action dont le résultat est que l'objet meurt. De même, on ne doit pas s'étonner que le maninka dise indifféremment:

et:

il a conduit le camion, litt. "il a fait courir le camion"

AFRICAN LINGUISTICS

202

puisque le camion peut être considéré comme un objet iner­ te ou au contraire comme un objet doté de la faculté de se mouvoir. Il est intéressant de noter qu'en dehors de ces em­ plois, bori ne s'utilise en maninka que sous une forme ré­ fléchie, avec le sens de "courir, s'enfuir" (cf. a k'ï: bori "il a couru, il s'est enfui"). Or, selon D. Creissels (1982), le mandinka bori est soit intransitif (cf. a borita "il s'est "il a couru") soit réfléchi (cf. a yé i'bori enfui") et, selon Bailleul (1981), le bambara Jboli est in­ transitif dans le sens de "courir, s'enfuir, couler" mais transitif dans le sens de "conduire, piloter". Il semble donc que ce verbe soit originellement un verbe intransitif signifiant "courir, bouger rapidement" et que ses emplois comme verbe transitif puissent être interprétés comme des causatifs sans marque formelle signifiant "faire courir, (faire) bouger rapidement". Une telle explication convient pour rendre compte des faits observés en maninka même si cette langue n'utilise plus le verbe intransitif originel. De plus il semble que d'autres verbes attestent une évolu­ tion de même type puisqu'on a notamment: dyula1°

pan

bambara11 pan maninka g ban

intrans. "sauter, voler" cf. wùlu pinna "le chien a sauté"; kono pinna "l'oiseau a vo­ lé" intrans. ou réfléchi "sauter, voler" réfléchi "sauter, voler" cf. a jbir'i: gbkn "il a sauté" ou "il a volé"

On a vu plus haut qu'il existait une différence de com­ portement entre des verbes comme dimun ou tibi et un ver­ be comme min "boire" qui, exceptionnellement devient ré­ fléchi s'il n'y a pas d'objet explicitement mentionné. Comme Bailleul (1981) signale qu'en bambara mïn est tran­ sitif avec les sens "boire, faire boire, sucer, fumer" et intransitif ou réfléchi avec les sens "boire, se désalté­ rer", il n'est pas exclu qu'à l'origine ce verbe ait ap­ partenu à la catégorie des intransitifs, même si sémantiquement, il est plus difficile d'admettre que l'utili­ sation transitive de min doit être interprétée comme une forme causative sans marque formelle. En maninka, on au­ rait donc, d'une part, des verbes comme 2 "être, (se) mettre debout" pour lesquels tous les emplois sont attes­ tés puisqu'on a: /à I rá m : Iù tέ ma/ il s'est levé au milieu des gens /à ka b o n i l a construit la maison /à 'Τ : il s'est levé

L'EXPRESSION DU PASSIF EN MANINKA

/à ka dénnîn tá k'à: kába kan/ pour le mettre debout sur le rocher

203

il a pris l'enfant

et des verbes comme bbrî, gbán, min qui auraient perdu leur emploi intransitif au profit de l'emploi réfléchi possible avec le causatif sans marque formelle. Il est probablement significatif à cet égard que, selon notre informateur, les énoncés: /à I  m : I ù tέ ma/ /à k'Y: I m : lu te ma/ sont des énoncés équivalents du point de vue du sens. Il va de soi que le problème posé ne pourra être ré­ solu que par une étude systématique des différentes clas­ ses de verbes. Ce qui précède est tout au plus une hypo­ thèse de travail dont les divers aspects devraient être vérifiés et testés à partir d'une étude portant sur un nombre de verbes beaucoup moins restreint. Pourtant, il semble qu'une telle hypothèse soit utile pour expliquer certaines caractéristiques du passif. Elle permet en effet de supposer que la transformation passive s'applique dans certains cas à des énoncés comprenant un verbe dérivé sans marque formelle dont le correspondant intransitif existe encore dans la langue, ce qui expli­ querait l'opposition entre: /à fa:ra yen/ il est mort là-bas (intrans. actif) /à ka nisi fa:/ il a tué la vache (causatif trans, act.) /ni s f fà:r'a: boio/ la vache a été tuée par lui (causa­ tif trans, passif) où il y a neutralisation formelle entre le verbe intran­ sitif et l'utilisation passive du causatif sans marque formelle, la présence du complément d'agent étant le seul critère qui permette de distinguer les deux types d'em­ plois. Il semble que ce soit précisément dans des cas de ce genre que l'effacement du complément d'agent n'est guère toléré. La chose paraît normale, comme il paraît normal aussi que l'effacement du complément d'agent soit admis quand le verbe intransitif n'est plus employé dans la langue: en effet, si min intransitif n'est pas utilisé, un énoncé comme à ma mïn ne peut être interprété que comme un énoncé passif.

204

III.

AFRICAN LINGUISTICS

OBSERVATIONS GENERALES

On peut certainement concevoir une analyse du verbe maninka dans laquelle les bases verbales ne seraient pas caractérisées lexicalement par rapport au trait [±transitif] mais seraient considérées comme capables de figurer aussi bien dans des constructions transitives que dans des constructions intransitives. Une telle analyse con­ duit à admettre qu'il n'y a pas de passif réel dans la langue et à accorder un statut identique à toutes les constructions intransitives, même si certaines d'entre elles se traduisent par des passifs en français. Cette interprétation peut être justifiée par le fait qu' il n'exis­ te aucune marque formelle du passif et il va de soi qu'elle est plus facile à défendre dans les langues mande qui n'expriment pas le complément d'agent. Pour ce qui est du maninka, nous pensons toutefois que, même dans une telle perspective, il est très diffi­ cile d'échapper à la nécessité d'établir une sous-catégo­ risation au sein de la masse des verbes, à un moment ou à un autre de la description. On a vu en effet, dans les pages qui précèdent, que tous les verbes ne se comportent pas de la même manière par rapport à l'opposition exis­ tant entre les constructions intransitives et les con­ structions transitives. En particulier, les oppositions de sens qui sont liées à ces deux types d'emplois varient considérablement selon les verbes qui s'y trouvent impli­ qués et ces variations sont systématisables. Ainsi, des verbes comme dámun "manger" ou tibi "préparer" opposent une signification active au transitif à une signification passive ä l-'-intransitif; d'autres verbes peuvent avoir, à 1'intransitif, un sens actif neutre ou un sens passif causatif, alors qu' ils ont un sens actif causatif lorsqu' ils figurent dans une construction transitive. Enfin, il existe également des verbes dont l'emploi dans une con­ struction transitive s'accompagne presque toujours de l'adjonction du morphème dérivatif là, même si théorique­ ment l'omission de ce dérivatif peut être imaginable dans des énoncés marqués stylistiquement12. De tels verbes sont également accompagnés du dérivatif la lorsqu'ils figurent dans une construction intransitive avec un sens passif (cf. à sera yen "il est arrivé la-bas"; a ka kawandill lase Ala ma "il a fait parvenir le message à Dieu"; à hara lasè "cela a été envoyé"). Il semble donc que l'on soit confronté à des méca­ nismes qui supposent effectivement l'existence de diver­ ses catégories de verbes et il paraît plus avantageux,

L'EXPRESSION DU PASSIF EN MANINKA

205

dès lors, de considérer que ces catégories sont lexicales. De plus, les rapports syntaxiques qui existent, en maninka, entre les énoncés transitifs actifs et certains énon­ cés intransitifs peuvent être décrits au moyen d'une transformation qui présente toutes les caractéristiques de ce que l'on considère habituellement comme une trans­ formation passive. Certes l'application de cette trans­ formation ne détermine pas l'apparition d'une marque spé­ cifique au sein de la forme verbale. Mais si l'on consi­ dère l'extrême diversité que de telles marques présentent dans les langues du monde, puisqu'il s'agit de marques de conjugaison dans les langues européennes mais d'un suffixe de dérivation analogue à celui qui exprime le causatif ou l'action réciproque dans les langues bantoues, la situa­ tion qu'on observe en maninka paraît moins étonnante et elle ne nous semble pas de nature à remettre encausel 'existence d'une transformation passive véritable. De plus, en ad­ mettant que cette transformation existe, on résout les problèmes que pose la coexistence d'énoncés intransitifs qui, tout en étant formellement identiques pour ce qui est du verbe et du sujet grammatical, n'en ont pas moins des sens distincts. L'opposition entre ces énoncés est expliquée par le fait qu'ils n'ont pas la même histoire transformationnelle et qu'ils sont engendrés par des pro­ cessus différents. L'établissement des catégories lexicales qui organi­ sent l'ensemble des verbes n'a été envisagé ici qu'inci­ demment et de manière incomplète. Une enquête menée sur l'ensemble des verbes recensés serait indispensable pour aboutir à une description exhaustive de ces catégories, sans parler des cas particuliers qui nécessiteraient sans doute, en maninka comme dans de très nombreuses langues, un traitement spécifique. Les verbes qui ont été étudiés ici paraissent montrer qu'il serait opportun de distin­ guer des bases verbales ayant un trait lexical [ + trans.] et qui ne figurent dans des constructions intransitives qufà la suite de l'application de la transformation pas­ sive. D'autres bases verbales seraient notées [- trans.] (cf. ¿ "sortir", fa: "mourir", etc...); elles ne se­ raient utilisées dans des constructions transitives que lorsqu'on leur adjoint un trait [+ causatif] se manifes­ tant formellement par le dérivatif là antéposé ou par une marque 0, dans des conditions que nous avons tenté de dé­ crire partiellement. Enfin, des bases verbales comme  "sauter, voler", borl "courir" ou min "boire", qui ne connaissent aucune utilisation intransitive active mais se comportent pourtant à certains égards comme des verbes intransitifs, pourraient être caractérisés comme à la fois [ + trans.] et [+ causatif], le couplage des deux

206

AFRICAN LINGUISTICS

traits amenant obligatoirement leur emploi dans des con­ structions transitives. Il faudrait préciser les règles qui permettent d'engendrer le nom verbal qui s'obtient à l'aide du suffixe -li ou de prévoir l'utilisation réflé­ chie de la forme verbale, mais il semble que de telles règles pourraient être formulées de manière satisfaisante. Nous pensons que l'utilisation de procédures descriptives de ce type pourrait être féconde. Nous pensons surtout qu'il y a avantage à reconnaître l'existence d'une trans­ formation passive en maninka, parce qu'il est alors pos­ sible de distinguer différentes utilisations de la con­ struction intransitive et d'expliquer ces distinctions au sein d'un modèle cohérent13.

NOTES

1. Nous regrettons de ne pas avoir pu consulter Ch. Bird, Aspects of Bamhara syntax, Ph.D. Dissertation, Univ. of California, Los An­ geles, 1966. 2. Nous utilisons, pour désigner les divers aspects ou temps de la conjugaison, des dénominations qui décrivent leur valeur approxima­ tive mais doivent être considérées comme provisoires. 3. Afin de faciliter l'analyse des exemples, nous en donnons une notation courante et une notation qui restitue la forme de base que nous leur attribuons. Dans la graphie courante placée entre barres obliques, les tons sont notés tels qu'ils sont réalisés mais, par souci d'économie, nous n'indiquons que les modifications de niveau, sauf dans la séquence ..^/. où le signe ' marque, au contraire, le maintien du niveau abaissé. 4. Vu le rôle qu'elles peuvent jouer dans l'énoncé maninka, nous préférons considérer les formes statives et immédiates comme des formes conjuguées à part entière et expliquer leurs autres emplois par une procédure de dérivation. Le fait qu'elles sont propres aux verbes utilisés sans objet paraît confirmé par la description qui en est donnée dans la grammaire maninka diffusée par le Service d'Alphabétisation de la République de Guinée (Conakry, Anonyme, 1968,

L'EXPRESSION DU PASSIF EN MANINKA

207

polycopié). 5. Cf. notamment G. Dumestre, 1981, p.49-67 et R. Spears, 1965, p. 63. 6. La règle établie par R. Spears est beaucoup plus complexe et elle nécessiterait une explication qui n'est pas utile à notre propos. Nous en donnons ici une version simplifiée qui permet d'observer l'essentiel du mécanisme décrit. 7. Ces deux énoncés ont été proposés par l'informateur comme équi­ valents du français "il ne le fait pas". 8. Selon D. Creissels (communication personnelle) 9. Selon D. Creissels (communication personnelle) à qui nous devons également les exemples cités. 10.

Braconnier et Derive (1978).

11.

Bailleul (1981).

12. Cf. par exemple l'emploi transitif causatif de na "venir" dans la langue de certains griots bambara. Cet emploi est attesté dans: bl ma Bakari nà Segu "ce n'est pas aujourd'hui que Bakari est venu à Ségou (litt.: aujourd'hui n'a pas fait venir Bakari à Ségou)" ou dans 1 tulo nà yàn "amène ton oreille ici". D. Creissels a attiré notre attention sur ces exemples dont le second provient des oeuvres transcrites et traduites par G. Dumestre dans La geste de Sègou, Classiques africains, A. Colin, 1979. 13. Nous remercions très vivement Th. Schadeberg, M. Dominicy et D. Creissels dont les critiques et les suggestions nous ont apporté une aide considérable dans la rédaction du présent article. D. Creissels nous a de plus communiqué des données intéressantes en ce qui concerne le mandinka et le bambara.

NOTE BIBLIOGRAPHIQUE

Bailleul, Ch. 1981 Petit dictionnaire bambara-français; français-bambara, Avebury Publishing Company, England. Braconnier, C. et M.J. Derive 1978 Petit dictionnaire dioula, ILA, Abidjan.

AFRICAN LINGUISTICS

208

Creissels, D., S. Jatta et . Jobarteh 1982

Lexique

mandinka-français.

Mandenkan, III, Paris.

Dumestre, G. 1981

La morphologie

verbale

Friedländer, M. 1974 Lehrbuch des Susu, Houis, M. 1963 ' Etude

descriptive

en bambara,

Mandenkan II, 49-67, Paris.

VEB Verlag Enzyklopädie, Leipzig. de la

langue

susu,

Bull. IFAN 67, Dakar.

Rowlands, E.C. 1959

Ά Grammar of Gambian Mandlnka,

SOAS, London.

Service d1alphabétisation de la République de Guinée 1968 Grammaire maninkaf Conakry (polycopié) Spears, R.A. 1965

The structure

of

Faranah

Maninka,

Ph.D. Dissertation, Indiana

Univ. r- U.M.Inc. Ann Arbor, Michigan. Weimers, W.E. 1976 Ά grammar of Vair Angeles, London.

Univ. of California Press - Berkeley, Los

THE AUTOSEGMENTAL AND METRICAL NATURE OF TONE TERRACING

C.T. James Huang (National Tsing Hua University)

1.

INTRODUCTION

Tone terracing is a term commonly used in reference to the set pf interrelated phenomena known as downdrift, downstep, upstep and upsweep, which have been widely ob­ served among many African languages and some American In­ dian languages.* The most commonly reported of these phe­ nomena are downdrift and downstep. Downdrift is generally defined as an automatic progressive overall lowering of pitch throughout a phonological phrase. Thus in a sequence of alternating high (H) and low (L) tones, for example, each occurrence of H (or L) is slightly lower in pitch than the preceding occurrence of the same tone, so that the overall pitch pattern of the phrase is a downward "terraced" one. Generally, this type of lowering does not occur between adjacent occurrences of the same tone. A well known example of downdrift is the following sentence from Akan (Schachter 1961:235), which has the underlying tone sequence as indicated in (1), but a surface repre­ sentation with six phonetically distinct levels of pitch, as indicated in (2): (1)

(2)

me-de-me-n-na-m-fo-no-ba-a-me-fi L L H H L L L H H L H L 'I brought my friends to the house.'

210

AFRICAN LINGUISTICS

As a result of downdrift, an underlying H at the end of a long phonological phrase may be realized on a lower pitch than a L occurring in an earlier part of the phrase. The process of downdrift is automatic in that given any sequence of tones, every H after L is always one step lower than a preceding H and every L after H is always one step lower than a preceding L. Occasionally, a drop in pitch is observed to occur on a H immediately after an­ other H. The position of such a drop is not directly pre­ dictable from a given surface sequence of tones. This drop is referred to as downstep. The phenomenon "upstep" has the characteristic of raising the pitch of a tone one step higher than a pre­ ceding occurrence of the same tone. In the position where the tone is expected to undergo downdrift, upstep brings the tone back to a pitch level as high as the preceding occurrence, so that it appears as if no downdrift has oc­ curred in this position. The occurrence of "upsweep" is less frequently observed. It involves the successive overall raising of pitch over a sequence of some sort. The processes of tone terracing have been described and analyzed in various ways in the literature. In this paper I will examine the essential properties of tone ter­ racing, and address the problem of what an optimal theory of formal representation may look like that best accounts for these properties and gives a principled explanation for their clustering. In section 2, I will outline the major features of terracing. It is then proposed in sec­ tion 3 that a correct understanding of the types of phe­ nomena under consideration requires (a) that we draw a clear distinction between matters of tone and matters of pitch; . (b) that tone and pitch should be represented as two autonomous levels such that rules may apply on one level without regard to the internal structure of the oth­ er level; and (c) that, like stress, tone and pitch should be given a metrical treatment. In section 4 I will compare the theory with earlier proposals and show that these ear­ lier attempts are inadequate either in that they do not make correct empirical predictions or in that they do not capture real significant generalizations, or both. In the last section I will return to the proposed theory and make a number of extensions and refinements to accommodate more complicated cases of terracing.

TONE TERRACING

2.

211

IMPORTANT PROPERTIES OF TONE TERRACING

Among the features of tone terracing that have been reported in the literature, the following are generally held to be true: (a) Since the occurrence of downdrift is automatic, it does not necessitate the recognition of additional tö­ nernes. For example, in a two-tone language in which downdrift occurs, for each phonemic tone recognized, there is at most one different kind of tone following or preceding it. The status of downstep, on the other hand, may be re­ garded as partially phonemic, in that its position of oc­ currence is not directly predictable from the surface though, once its position is known, the way it influences the pitch realization of a phrase is predictable. Each downstep H establishes a ceiling for all following H tones, so that following a downstep H there is again at most one contrasting tone. This fact crucially distin­ guishes a downstep H in a two-tone language from what might be recognized as a mid tone in a three-tone lan­ guage, which would be fully phonemic in that there could be two contrasting tones found after and before it. (b) The interval in terms of pitch values between a H and a following L, or between a L and a following H, remains relatively constant throughout a phonological phrase. This fact can be easily seen from the representa­ tion given in (2) above. The interval between the second H and the immediately following L, for example, is the same as that between the fourth H and the immediately fol­ lowing L, and that between the fifth H and the following L, This fact indicates that underlying tonal contrasts are somehow "preserved" despite the extent to which ter­ racing has occurred to influence the realization of tones. (c) The distance in pitch height between two occur­ rences of the same tone in a terraced sequence is often not equal to (and is usually less than) the distance be­ tween two phonemic tones. For example, for a given speaker a H may be realized at a pitch level higher than a fol­ lowing L by 80 Hz on the fundamental frequency, while a downstepped or downdrifted H may be lower than its pre­ ceding H only by 20 Hz or so. This fact can again be seen directly from the representation given in (2). The drop that occurs on the third H, for example, is smaller in size than what happens between a H and a following L, since this H is still higher than the preceding L. 1 (d) A related point is that the ratio of the amount of drop involved in terracing in relation to the distance between phonemic tones need not be constant in a given

212

AFRICAN LINGUISTICS

language, or even for a given speaker. There is enough indication that this is the case. For example, in Clifton's (1976) study of Twi, the sentence in (3) below is given two distinct pitch representations, as (3b) (p. 182) and (3c) (p. 187): (3) a. me sekän foforo kesee no nye H !H!H H!H H L HH H L H 'My other big knife is no good.' b. 1 23 344 6 55 5 7 6 c. 1 23 344 7 55 5 8 6 In (3a), a H immediately after ! represents a downstep H. In (3b) and (3c) integers are used to represent pitch height, with the higher numbers indicating lower pitch values. In both (3b) and (3c), each instance of downstep or downdrift is represented by a drop of one pitch in­ crement from the preceding occurrence of the same tone. The contrast between a H and a following L, on the other hand, is indicated to have the value of two pitch incre­ ments in (3b) (the contrast between 4 and  or between 5 and 7), twice the amount of each instance of terracing, but in (3c) this contrast is indicated to be three times the amount of each instance of terracing (three pitch in­ crements between 4 and a following 7 and between 5 and a following 8). Whether or not these are precise reflections of the actual variations, the fact that both transcrip­ tions are used indicates the possibility of variation. Another indication of the variation is provided by the phenomenon of "speech planning" (Weimers 1973:88). It happens that speakers must vary the amount of pitch drop that occurs in terracing based upon the length of an ut­ terance, the longer the utterance, the less pitch drop at one time. This is understandable in that one has a limited pitch range within which to realize an indefinite number of levels resulting from terracing in utterances of vary­ ing lengths. Still another indication of the variation is the fact that speakers not only differ with respect to their pitch range (as observed between male and female speakers), but also vary their pitch range depending upon speech situa­ tions. A Chinese radio announcer, for example, uses a much wider pitch range at work, so that the tones are more clearly distinct, than when he or she speaks on informal occasions. Thus the amount of pitch drop per instance of terracing may vary from one situation to another, depend­ ing upon the pitch range used in each situation. (e) Downdrift of a sequence of the form H L H L H ... is often perceived in a different way than downdrift of a

TONE TERRACING

213

sequence of the form L H L H L ... In particular, in the former case, downdrift is perceived as involving a lower­ ing of a non-initial H (among others), while in the latter case, downdrift may be perceived as involving a "raised" initial L. (f) Tone terracing does not occur phrase-initially. In Armstrong (1968:51), for example, this fact is simply built into the definition of downdrift: Downdrift is de­ fined as "the tendency of non-initial low tones to pull succeeding high and mid tones downwards in pitch." The lack of phrase-initial downstep has also been indicated by various writers (e.g. Peters 1973). 2 (g) The amount of pitch drop that occurs at each in­ stance of phonemic downstep is exactly the same as that which occurs at each instance of automatic phonetic downdrift within a given phonological phrase. In the Twi sen­ tence above, for example, each instance of downstep IH in­ volves the drop of one pitch increment, as seen between the values of the first and second tones in (3); and each instance of downdrift also involves exactly the same amount of drop, as seen between any two Hs separated by a L. This fact shows that downdrift and downstep have some­ thing in common, and should be somehow related. (h) In a sequence involving both downdrift and downstep, a downstep H always occurs after another H; there is never a downstep H immediately after a L. One can again see this from the Twi example above. The eight properties that we have outlined above are more or less well known and generally held to be true. Let us now consider the following question. Assuming that these are true properties of tone terracing, what are the generalizations, if any, that they express, and how should they be treated in an optimal theory of mental representa­ tion? Any theory that aims to adequately account for the phenomena of tone terracing must take into consideration all of these properties as well as whatever generaliza­ tions they express. We now turn to the task of construct­ ing such a theory.

3.

AN AUTOSEGMENTAL-METRICAL ACCOUNT

Let us first consider what possible generalizations may be drawn from the properties outlined above. A look through these properties reveals certain interesting com­ mon features among them. On the basis of their common fea­ tures they seem to fall naturally into three groups. In

214

AFRICAN LINGUISTICS

particular, the first three properties, indicated in (a), (b), and (c), jointly indicate that tone terracing does not affect underlying tonal contrasts, or that matters of tonal contrasts are quite distinct from matters of tone terracing. The next three properties, indicated in (d), (e), and (f), suggest the generalization that there is nothing "absolute" in value, either in tonal contrasts or in matters of tone terracing. Rather, what is relevant is the relative value of tones or pitch. A H is relatively high in relation to a L. A tone having undergone downstep is lower in relation to a preceding occurrence of the same tone. The actual pitch values of these entities are large­ ly irrelevant for linguistic purposes. Finally, the last two properties, (g), and (h), both suggest that downdrift and downstep are closely related phenomena. The generalizations indicated may not be entirely obvious at the moment, but I will assume them to be cor­ rect and go on to consider possible ways to account for them. They should become clearer as we continue. Consider the first generalization. One way to capture it is to characterize tone terracing as a process that operates on an extra-tonal level without regard to the internal structure of the tonal level. To see how this may be done, it is useful to draw a distinction between the notions of pitch range, or register, and tone. Tones are linguistically abstract entities expres­ sing, in the languages in which they occur, an opposition between any two tone-bearing units (rimes or vowel-nuclei) within a given range of pitch. Pitch, on the other hand, corresponds more closely to the values of fundamental frequency. A register, or pitch range, sets the range within which contrasting tones are defined. Given this distinction, let us now look at tone ter­ racing in the following way. Since terracing preserves tonal contrasts, we may say that it incolves the shifting of the register only, but not of the tones. More specifi­ cally, I propose that (a) tones and pitch registers be represented as two autonomous levels, and (b) terracing be represented as a process operating in the registral level alone, making no reference to the internal structure of the tonal level. This proposal thus incorporates the theory of autosegmental phonology of Goldsmith (1976) and further extends it, claiming that, just as tones consti­ tute an autonomous level somewhat independent of segments (as Goldsmith argues), registers also constitute an au­ tonomous level somewhat independent of tones (and of seg­ ments) . As for the second set of properties, I propose that a natural extension of the "metrical" theory of stress of

TONE TERRACING

215

Liberman and Prince (1977) will serve our purpose of cap­ turing the generalization that seems to underly them. Li­ berman and Prince argue that stress values are entirely relative, and that an adequate system of representation should not make crucial use of integers, such as [+ 3 stress], etc., as in the system developed in Chomsky and Halle (1968). Instead, they provide a system whereby stress-bearing units are organized under a tree each of whose nodes is labeled as either relatively strong or weak in stress in relation to its sister node. The notion of relative degree inherent in their approach may be easi­ ly extended to our treatment of tones and of tone ter­ racing. Finally, in order to capture the generalization that downdrift and downstep are closely related phenomena, a plausible idea is to analyze them in such a way that they involve some common rule or principle in the system we propose. These are the ideas that may lead us to the right kind of theory required. I will now spell out one mode of execution that will put these ideas to work. I propose that every sequence of tones is organized into a sequence of formal objects called "tonal feet,1" in much the same way that a sequence of segments is organized into a se­ quence of formal objects called syllables. These tonal feet are then regarded as the basic pitch-bearing units upon which the assignment of pitch register values will occur, in much the same way that the syllables (or rimes) are the basic tone-bearing units. The process of foot-for­ mation is carried out in accordance with the rules in (4) below: (4)

Foot Formation:

The first two rules provide that, scanning from left to right, the initial tone of a phrase (as in (4a)), as well as every H immediately after a L (as in (4b)), is as­ signed an independent foot node, indicated here by the

216

AFRICAN LINGUISTICS

symbol Φ. Every tone not assigned a foot node either way is then associated by the provisions of (4c), which may apply iteratively. The feet are the basic pitch-bearing units (PBU's). Since tones are defined within pitch reg­ isters and must be realized in terms of pitch, the rule (4c) may be regarded as a way to satisfy the following R(ealization)-Criterion: (5)

R-Criterion: Every tone is assigned to one and only one pitch-bearing unit.

The sequence of tonal feet that results from the rules in (4) will be the input to the rules of pitch as­ signment that account for the phenomena of tone terracing. For downdrift, pitch assignment is carried out by (6). (6)

Pitch Assignment: a.

Assign a right branching tree to a sequence of tonal feet.

b.

For each pair of two nodes Ni and N 2 / label Ni as L iff it branches. (If a node is not labelled L then it is H.)

The result of applying (6a) to any sequence of tonal feet will be a tree of the form (7a), which is then la­ belled as (7b) by (6b):

(7)

a.

b. The rule (6b) is an adapted version of one of the universal labelling conventions provided in Halle and Vergnaud (1979), which contains a number of extensions of Liberman and Prince's metrical theory of stress. The re­ sult (7b) is achieved by requiring that the rule disre­ gard, or be "blind" to, the internal structure of any foot (whether it branches or n o t ) , taking the sequence of feet as an autonomous level, with each foot in the

TONE TERRACING

217

level being considered an autonomous non-branching whole. To illustrate, the Akan sentence considered above in (1) has the following form after foot-formation (4a-c) ap­ plies: (8)

The pitch assigment rules (6a,b) will turn this into (9): (9)

Level of Pitch Register

Tonal Level This is the final output of our rules. It is not, how­ ever, a direct phonetic representation of the utterance. It gives only a relational structure showing the relative jbitch height of the tones, but says nothing about the ab­ solute pitch values at any point in the tree. It says that the tones that belong to the first foot are realized in a higher register than the tones belonging to the sec­ ond foot, and those of the second foot are realized at a higher pitch than those of the third, etc. The progres­ sive increase in the depth of embedding under a L node in the tree of the registral level corresponds directly to the progressive pitch drop involved in downdrift. Thus, there are three L nodes above the last foot, two above the third foot, and one above the second foot, but none above the first foot, in the representation above. This directly corresponds to the fact that tones in the sec­ ond, third, and fourth feet have been lowered by one, two, and three steps, respectively, though not the tones assigned to the first foot. The amount of drop per in­ stance of pitch lowering represented at the autonomous registral level is not related to the distance in pitch between a H and a L within the tonal level. The ratio of the amount of pitch lowering in relation to the amount of pitch difference in tonal contrasts cannot be determined by any general system of representation, but varies.

218

AFRICAN LINGUISTICS

from language to language, speaker to speaker, and even from utterance to utterance. Up to now we have considered only cases of downdrift. Let us turn now to downstep. It is generally agreed that downstep does not differ significantly from downdrift, ex­ cept in that its position is unpredictable from the sur­ face. The similarity of these two processes is readily seen from one of the last two properties discussed in sec­ tion 2'. One plausible hypothesis on the status of downstep is that it is simply downdrift except that its position of occurrence must be lexically marked. Thus, the status of downstep is like that of the pitch accent in, say, Jap­ anese, where the position of the accent must be lexically marked, though everything else can be derived by rule. On the other hand, it has been suggested that many cases of downstep are in fact cases of downdrift made opaque due to vowel deletion. Thus, it has been suggested that the downstep in the well known example below is the result of a simple application of downdrift preceding a rule of vow­ el deletion:3 (10)

'my stone1 underlying representation by downdrift by vowel vowel which deletes both the vowel and the tone output

Neither the phonemic nor the purely phonetic approach are entirely satisfactory, however. The phonemic approach is defective in that there are many cases of predictable downstep (such as (10)).1* Furthermore, as Clements (1979) has shown, the purely phonetic approach is untenable not only because the ordering relation required in the hypoth­ esis above is problematic, due to the fact that downdrift is a lower-level rule than vowel deletion, the latter of which often subject to morphological conditioning, but also because there are languages or sentences where downstep alone, but not downdrift, occurs. For example, in Hyman and Tadadjew (1976), it is indicated that DschangBamileke has downstep, but no downdrift. Another such lan­ guage is Kikuyu, as reported in Clements (1979) and Clements and Ford (1979). Furthermore, in languages where downdrift does occur, it occurs only in certain sentence types, e.g. declaratives. In other sentence types, no downdrift occurs, though one may also find occurrences of downstep here. In all of these cases, downstep cannot be treated as a special case of downdrift.

TONE TERRACING

219

In Clements and Ford (1979) , it is argued that downstep has the status of a floating L tone which triggers pitch lowering on all following tones. The existence of a floating tone is easily handled within an autosegmental approach, according to which the deletion of an element in the segmental level need not entail the deletion of an element in the tonal level. Their approach solves the or­ dering problem associated with (10) , allowing vowel dele­ tion to occur before pitch lowering. There are also a number of other facts which they cite in support of this analysis. I will adopt it, and analyze downstep as in­ volving the following rule of foot formation: (11)

Φ L Η | 1 2 =* 1 2 Condition: 1 is unassociated with any segment. To account for the relationship between downdrift and downstep, I assume that they differ minimally only in that while downdrift involves the foot formation rule (4b), downstep involves the rule (11). All the other rules pro­ posed above, namely (4a), (4c) and the pitch assignment rules in (6), are involved in both downdrift and downstep. 5 For illustration, consider the following sentence from Efik (Winston 1960:186): (12)

L Η !H Η !H Η !H H !H HÎH H L Ekpeny  emen inuen nyn edi uf  'Ekpenyong picked it up and came home.1 Each ! stands for a floating L tone. By the downstep rule (11), each Η immediately following this floating tone will be assigned to an independent foot node. By the downdrift rule (4b), the first Η tone will also be assigned to an independent foot node. Furthermore, the phrase-initial L tone is assigned to a foot by (4a). Finally, all remaining tones will be associated according to (4c). These proces­ ses of foot formation thus organize the tonal sequence in (12) into the following form: (13)

Finally, the pitch assignment rules in (6) apply to give the final output:

220

AFRICAN LINGUISTICS

(14)

Since the floating tones are unassociated with segments, they are simply phonetically "empty categories," and may be ignored by convention at the final level of phonetic representation. All relevant information concerning the pitch pattern of the sentence is contained in (14). Sim­ ilarly, the Twi sentence cited above has the following representation after (4), (11) and (6) have applied: (15)

This metrical tree can be converted easily into an integer representation if one wants to, after one determines what value to assign to each drop in pitch in the register lev­ el and what value to assign to the distance between H and L in the tonal level. Thus, either one of the integer representations given in Clifton (1976) , as indicated in (3), can be obtained from (15), depending upon actual speech situations.

4.

HOW THE PROPERTIES FALL OUT

We have proposed an autosegmental and metrical system of representation of tone terracing and illustrated the way it works. Let us see how it may account for all the properties indicated in section 2 and capture the general-

TONE TERRACING

221

izations that underly them. Consider the first three properties again. Since we have adopted an autosegmental approach to both tones and pitch registers, it is natural to expect that processes that operate on one autonomous level need not affect the content of the other level. Thus tone terracing (or rather register-lowering) does not necessitate the recognition of additional tones (in the tonal level). Preservation of tonal contrasts by tone terracing is also a natural con­ sequence of this system, because register lowering applies to each tonal foot or pitch-bearing unit as a whole (rath­ er than to tones), without affecting the internal rela­ tionship of contrasting tones within each foot. Further­ more, it is also a natural consequence that the amount of pitch drop occurring in terracing need not be equal to the distance between two contrasting tones. The next three properties may be taken as natural consequences of our system because of the metrical view we have adopted. Ttms, the variation in the amount of pitch f per instance of terracing, as well as in the ratio of this drop in relation to the difference between contrasting tones, indicates that the actual values of pitch to be assigned to a sequence of tones are subject to factors that are largely pragmatic and that cannot be, and should not be, fully specified in a formal grammar. As is already obvious, these factors are completely ab­ stracted away from our system. The metrical view also explains the difference in perception (and realization) that may happen between a sequence of the sort H L H L H and one of the sort L H L H L. This is because all independent phonological phrases may be pronounced at as high a pitch level as each other. More significantly, according to our system, the two se­ quences have the following representations: (16) a. b.

In the tonal level of (16a), the first H tone does not undergo downdrift, but the second H does, as it is assigned to the second tonal foot. Since the first L in this sequence belongs to the first foot, the distance be-

222

AFRICAN LINGUISTICS

tween the L and the second H is shortened. Therefore, the second H is realized and perceived as being lowered from the height of the preceding H by one step. On the other hand, in the tonal level of (16b), the first occurrence of H belongs to the second foot, with the first foot con­ taining only one tone, L. There is also a shortened dis­ tance between the initial L and the first H. But since the amount of tonal contrast is usually greater than the amount of contrast resulting from terracing, the H tone that occurs in the second foot is still higher than the L occurring in the first foot. Therefore, within this se­ quence, the tone with the highest pitch is the second tone, namely the first H. Now, since (16a) and (16b) are independent phonological phrases, the tone of the highest pitch in (16b) can be as high as the tone of the highest pitch in (16a). Therefore, while the shortened distance in (16a) between the first and the second foot is realized by a drop in pitch on tones of the second foot, the short­ ened distance in (16b) can be realized by a rise in pitch on the tone of the first foot, i.e. a "raised" initial L. The lack of phrase-initial tone terracing follows from the metrical account in a similar way. Since the pitch values of tones are relative to each other within the domain of an independent phonological phrase, every tone in phrase-initial position always belongs to the first foot, which receives the highest pitch within this phrase. The pitch of a phrase-initial tone always serves as the reference point to define terracing, in other words. A phrase-initial downdrift would simply make no sense. Can a downstep (i.e. a floating L) occur phraseinitially? Such a situation is possible, given no special reason why it cannot. However, if the floating L is fol­ lowed by a H, the entire sequence will be of the form L H L H L ... just discussed. Suppose that such a sequence does exist, then the H following the initial L will be a phrase-initial downstep H on the surface. But such a case of terracing is often realized as a "raised" initial L. Since in this case the raised initial L is phonetically empty, no effect of a phrase-initial downstep H is ob­ served. The occurrence of a floating tone or downstep is allowed everywhere, in other words, but its effects are neutralized in phrase-initial position, given our metri­ cal system.6 As for the last two properties of tone terracing, they fall out as consequences of our assumption that downdrift and downstep share essentially the same set of rules: (4a), (4c), (6a), and (6b). The only difference is that while downdrift involves (4b), downstep involves (11), with each of their specific conditions on applica-

TONE TERRACING

223

tion. Thus, the amount of drop per instance of downdrift is exactly the same as that per instance of downstep, be­ cause the output of the foot formation rule (4b) and that of (11) are assumed to be both tonal feet, which are indistinguishably subject to the rules of pitch assignment. Finally, the non-occurrence of downstep H after L in a sequence that also undergoes downdrift follows from our assumption that both downstep and downdrift involve the creation of a tonal foot on a H tone right after a L, though downstep involves the further condition that the L is a floating tone. (Only floating Ls trigger downstep, but all Ls may trigger downdrift.) Note that there is no special reason why a floating L tone cannot occur immedi­ ately before a H and immediately after a non-floating L. In such a situation we have a sequence of the form ... L I H ..., where ! stands for a floating L. The existence of the 1 would trigger the rule (11) , which would assign the following H to an independent foot. But note that even if the ! were not there, the H would still be assigned to an independent foot, i.e. by the downdrift rule (4b), since it would still be immediately preceded by a L (in this case the non-floating L ) . Thus, again, the occurrence of downstep H after L is in principle allowed, but its surface effects are always neutralized.7 Given that the effect of downdrift is transparent (observable from con­ text) , but the effect of downstep opaque, it is usually claimed that downstep H never occurs after L. The explanation offered for the superficial non-oc­ currence of downstep H receives considerable support from the fact that this observation holds true only in sequen­ ces which also undergo downdrift. If a sequence does not undergo downdrift, then the effect of a downstep H after a non-floating L will not be neutralized, and its occur­ rence should be observable. This is indeed the case. For example, Hyman and Tadadjeu (1976) have shown that, in Dschang-Bamileke, only downstep but no downdrift occurs. It is precisely in this language that they also observe the occurrence of downstep H after L. The second of the following examples shows a downstepped H after L (Hyman 1979:11): (17)

a. [ L H ] apa 'lid' b. [ L ÎH ] àpa 'taro1 That the second syllable of 'taro' is associated with a downstepped H and not a mid tone is evidenced by the ceil­ ing it establishes, limiting a following underlying H to an equal pitch height:8

AFRICAN LINGUISTICS

224

[ L H # H] àpa sa'o 'the lid of the bird1 [ L !H # H] apa sárj »the taro of the bird1 Other examples supporting this account of the non-occur­ rence of downstep H after L are also available. Since downdrift occurs generally only in declaratives but downstep may occur in all sentence types, an otherwise neu­ tralized downstep H after L may often be observed in a non-declarative sentence of a downdrift language, as for example in Igbo (John Goldsmith, personal communication). (18)

5.

a.

COMPARISON WITH ALTERNATIVE TREATMENTS

I have outlined the basics of an autosegmental and metrical system of representation of downdrift and downstep, and shown how the well known properties of these processes may be naturally derived within such a system. It is appropriate to consider now some previous attempts and see how the present theory may be said to represent an improvement over them. One type of representation that has been proposed in the literature is exemplified in such works as Fromkin (1972), Peters (1973), Clifton (1976). Fromkin suggests that the surface values of terraced tone patterns may be determined by rules which assign integers to lexically specified binary features. Specifically, she proposes to assign the pitch values 1 and 3 respectively to all H and L tones by the rule (19), and account for downdrift and downstep by (20), applied iteratively from left to right after (19): (19) a. [+H] =» pi [-H] => ρ 3 (20)

[αΗ]

=>

[ p < + l > a

]/ [ ^

d >

]b U+Mid> aJ

(where the feature matrix [+H, +Mid] designates a downstep H, and ρ abbreviates "pitch"). Rule (20) expands into the six in (21) :

TONE TERRACING

225

(21)

As an example, t h e same Akan s e n t e n c e t h e f o l l o w i n g manner:

(1) i s d e r i v e d

in

/me-de-me-n-ma-m-fo-no-ba-a-me-fi/ L L H H L L L H H L H L 11 1 1 1 by (19a) 3 3 3 3 3 3 3 (19b) 4 (21e) 4 4 (21f) 2 (21b) 2 (21c) 5 (21e) 3 (21b) 6 (21e) 3 3 1 1 4 4 4 2 2 5 3 6 Output The rules (21c) and (21f) ensure that no pitch lowering occurs between identical tones. (21b) and (2le) jointly account for downdrift: the former lowers the pitch of a H by one step from the pitch of a preceding H when these two Hs are intervened by an arbitrary sequence of Ls; and the latter lowers the pitch of a L by one step from the pitch of a preceding L when these two Ls are intervened by an arbitrary sequence of Hs. The rule (21a) accounts for downstep: it says that any a [+H, +Mid] (a H marked for downstep), when occurring immediately after another H, lowers its pitch by one step from the pitch of the preceding H. The rule (21d) says that a [-H, +Mid] lowers its pitch by one step from the preceding L when immediate­ ly preceded by another L. This last rule is never realized, however, since the system does not have any [-H, +Mid] or downstep Ls. It is important to note that, in this type of repre­ sentation, no distinction is made between the level of tone and that of the pitch register. As such, the Fromkintype theory is mixed-level, non-autosegmental system. Fur­ thermore, as the system makes crucial use of integers, it is not a metrical system. Let us first concentrate on the

226

AFRICAN LINGUISTICS

non-autosegmental nature of the system. There are a number of drawbacks to any theory having this nature. First of all, the properties that, as we observed, seem to suggest that matters of tone and matters of pitch register are distinct, do not receive a natural interpre­ tation. For example, the preservation of tonal contrasts by terracing is accounted for in this system as the result of the stipulated application of two different rules. In particular, just in case a H lowers its pitch by the rule (21b), a following L must also lower its pitch, this time by a different rule, (21e). This result is ensured by the requirement that both rules exist and one of them apply if and only if the other does also. Such a stipulation misses a generalization. Furthermore, if the system fails to include a rule lowering the pitch value of a L accord­ ing to each rule lowering the pitch value of a preceding H, the system will be empirically inadequate. There is in­ deed an empirical failure of precisely this type in From­ kin* s system, as observed by Peters (1973). More specif­ ically, although Fromkin provides the rule (21a) to ac­ count for the pitch value for a downstep H, she does not have a rule that lowers a L following the downstep. The rule (21d), as just mentioned, is never realized. While this empirical failure can be corrected, as is done in Peters (1973) for example, the important point is that the system remains essentially stipulative in nature. In the autosegmental theory, preservation of tonal contrasts is a natural consequence of the postulation that terracing affects pitch-bearing units as a whole, without altering the internal relationship between tones within a given foot. Another indication of this conceptual problem in the mixed^· level system is that the amount of pitch drop per instance of downdrift or downstep is ensured to be less than the distance between contrasting tones by requiring the initial rules (19) to assign 1 and 3 respectively to H and L, and the rules in (21) to each assign at most 1 pitch increment at one time, again by stipulation. There is no clear indication why the (19) and (21) should assign different values of pitch. Within the autosegmental ap­ proach, the non-equality between contrasts in the tonal level and those in the registral level are natural conse­ quences of the fact that these two levels are autonomous. A second defect of the type of system proposed by Fromkin is that a sequence of the form H L H undergoes a different rule than a sequence of the form L H L, and the phenomenon of downdrift that happens in them seems to re­ ceive two different interpretations. In the former case, the second H lowers its pitch after a preceding L. This

TONE TERRACING

227

might be interpreted as a process of assimilation (verti­ cal assimilation in the sense of Hyman 1973). In the lat­ ter case, the second L lowers its pitch after a preceding H. This process certainly is not one of assimilation, and it is not entirely clear what it is (perhaps a process of disimilation?) Within our system, both instances receive a uniform interpretation. The two sequences have the fol­ lowing forms : (22)

a.

These two sequences are organized in different ways in the tonal level, but look identical in the registral lev­ el. Since downdrift affects only the registral level, the two sequences above receive a uniform interpretation: reg­ ister lowering occurs on both their second pitch-bearing units. A third problem with the mixed-level system is its obvious complexity. In our theory, we "modularize" the grammar into two autonomous components, and obtain the desired results of terracing by relatively simple rules, such as the two rules of (6). This result seems to be conceptually more desirable, and is in accordance with recent advances in syntactic theory, represented in, say, the modular framework of "Government and Binding" (Choms­ ky 1981) . Other objections to the type of system represented by Fromkin (1972) have been separately raised elsewhere, for example in Clements (19 79), who also points out, among other things, the strange character of the iterative rules proposed, as well as the non-universal character of the system. In a system of the sort proposed by Clements (1979) and partially followed in Hyman (1979), there is a formalism that characterizes terracing somewhat inde­ pendently of tonal patterns. Clements proposes that the tones of a given language are organized into a set of formal objects that he called "tone level frames." Each tone-level frame consists of "a set of n tone levels or­ dered under the relation '>' ('higher than 1 )/ and a set of n~l intervals," which are "relations holding between neighboring tone levels, that is, tone levels which are consecutive in the ordering established by the relation 'higher than' (p. 547-48)." For example, in a sequence of

228

AFRICAN LINGUISTICS

the form H L H L H L H, each tone is projected onto either the level H or L of the tone level frame. The intonational pattern of the entire sequence is then characterized in terms of these tone-level frames. Each tone-level frame "does not constitute a set of absolute acoustic parame­ ters. Rather it is subject to modification as a result of the intonational processes that apply to it. The identity of the frame itself, however, is not affected by these modifications (p. 549)." For each individual language, the internal structure of each tone level frame, e.g. the interval between two tone levels within each frame must be specified. For a two-tone terraced-level language, for example, Clements suggests that the interval may be arbitrarily set at 2 (pitch increments). By convention, an initial H is desig­ nated as 1 and an initial L as 3. To derive the correct pitch representation of a sentence involving terracing, Clements further suggests marking downstep by a ! symbol preceding the downstepped tone, and incorporates downdrift with downstep by a rule that introduces a 1 in the envi­ ronment H L:

(23)

0

=*

! / H

L

The mark ! (either underlying or introduced by (23)) is treated as an abstracting rule-triggering segment on the tonal level, whose function is to induce pitch lowering: (24)

Τ

=>

+1 pitch / Η !

which assigns one pitch increment to every maximal string of tones (T ) following any ! immediately preceded by H. The final * output is obtained by adding up all incre­ ments assigned to each tone. A sample derivation is given below: 9 (25)

a bera nt L L H 1HH L L H !HH 3 3 1 11

3 3 1

b

-L Η L H Η IL H 1 3 1 1 1 1 1 1

11

Η Η 1 1

22

22

53

(lexical representation) (23) (initial assignment) (24) (24)

Output

It should be apparent that Clements 1 tone level frames are intended to have the effect of maintaining the autonomy of a tonal level - in a way similar to the ef­ fect of our pitch-bearing unit or tonal foot. Our repre-

TONE TERRACING

229

sentation appears to be clearer, however, in that it di­ rectly reflects the autonomy of the tonal level by having tones of the same foot dominated by a node whose internal structure is inaccessible to rules of a higher level.10 A more important difference between Clements1 system and the system proposed here, is that the former makes crucial use of integers, and is thus non-metrical in na­ ture. In this respect, Clements1 system is more like Fromkin's. There are also a number of drawbacks to the nonmetrical theory. First, as Liberman and Prince (1977) have argued, stress values have meaning only in relation to each other. Likewise, the assignment of an initial L as +3 is meaning­ less when taken alone. Furthermore, an integer system in­ volving multi-valued features like [+7 pitch] etc. is quite complex and theoretically undesirable, especially in view of the fact that other aspects of phonology can be captured within a theoretically much more simple frame­ work employing only binary features. In other words, in­ stead of a complex multi-valued system, it is theoretic­ ally desirable to modularize it into an indefinite number of levels of representation, within which every node is binary branching and describable in terms of a simple [+H] vs. [-H] feature system. Secondly, a non-metrical system makes its own scope of application too narrow to provide a general descrip­ tion of the terracing phenomena across variations of the sort noted above. For example, the stipulation that an initial H is assigned 1 and an initial L assigned 3 pitch increments is only suitable for languages or sitations in which each instance of terracing involves the drop of pitch half the size of the contrast between underlying contrasting tones. Such a system cannot be extended to languages involving "tonal downstep" (cf. footnote 1 ) . It is probably also too narrow to be applicable across dif­ ferent speech situations within a speaker's idiolect, given that speakers vary the amount of pitch lowering de­ pending upon utterance length, as noted above. Thirdly, the lack of phrase-initial terracing, which is naturally attributed to the relational nature of tone and pitch within our metrical system, is not derivable without some sort of stipulation or other in Clements' system. Suppose there is a phrase-initial '. in some se­ quence. The fact that this initial I will not have a sur­ face pitch-lowering effect is ensured by the stipulation in the rule (24) that a I induces pitch lowering only when it immediately follows a H. Similarly, in Fromkin's system, this fact is ensured by a stipulation that every case of downstep H (her [+H, +Mid]) induces pitch lowering

230

AFRICAN LINGUISTICS

only when there is an immediately preceding H, as is in­ dicated in her rule (21a). The stipulative nature of both accounts is obvious. Finally, the difference in realization and perception between sequences of the form H L H L H ... and those of the form L H L H L ... is again not derivable without some stipulation. As we shall see shortly, Clements' rule (23) should be revised as follows, in order to derive the lack of downstep H after L in a sequence involving downdrift: (26) 0 => ! / L H (This is, in fact, the rule adopted in Hyman 1979.) If this is the case, then there will be a I immediately be­ fore the first H tone of the sequence of (25) above, or of any sequence of the form L H L H L ... If each 1 in­ duces one drop in pitch, then the tone with the highest pitch in (25) or in any sequence of the form L H L H L ... will be assigned the pitch value 2. But this will imply that an independent phonological phrase of this form must always be pronounced at a pitch level one step lower than any phonological phrase of the form H L H L H ..., where the first H is assigned the pitch value 1. This does not seem to be correct, as one can tell from the integer transcription that Clements has given for (25), where the first H is assigned the value 1, and from the transcrip­ tion Fromkin has assigned to the Akan example, where the first H also has the value I. (But cf. footnote 6. Recall also the definition of downdrift given by Armstrong (1968: 51), which says that only non-initial low tones have the effect of inducing pitch lowering on following tones.) In order not to obtain the value 2 for the first H within a system that incorporates (26) , it is necessary to stipu­ late somehow that the leftmost H in any sequence is not subject ot pitch lowering regardless of whether there is a preceding !. There is one more point to be mentioned in comparing our system of representation to the ones represented by Fromkin (1972) and Clements (1979). This has to do with the lack of downstep H after L in a sequence that also undergoes downdrift. This fact is easily derived as a case of neutralization in our system, but not so in either Clements' or Fromkin's. In the latter two systems, the result is obtained again through a stipulation. According to Fromkin, this is obtained by stipulating that the rule (21a) apply only when there is a H immediately preceding the downstep H. In Clements' system incorporating (23), the effect of an underlying downstep H immediately fol­ lowing a L also cannot be neutralized without stipulation.

TONE TERRACING

231

Such a sequence would have the form H L ! H ... After the rule (23) applies, the sequence will be turned into the form H î L I H . .., with two Is, one preceding and one following the L. The second I thus cannot be neutralized by the first I. In order to obtain the fact that there is no downstep H after L, it is therefore necessary to stip­ ulate, again, that the second ! here does not induce pitch lowering. Precisely this stipulation is contained in the rule (24), which requires that the tone preceding ! must be H in order for the rule to apply. The only way to avoid this stipulation in this system would be to as­ sume the rule (26) instead of (23) , as Hyman (1979) has done. In this case, a downstep H after L will occur in a sequence having the underlying form H L ! H ... Since this is also the form one will get by applying (26) even in the absence of the underlying I, the surface effect of this underlying I is neutralized in this position. This, how­ ever, brings one back to the problem just mentioned in the immediately preceding paragraph.

6.

REFINEMENTS AND EXTENSIONS.

We have seen sufficient reason, I think, for the be­ lief that a modular system of the sort developed in this paper qualifies as a more nearly adequate system of men­ tal representation. We have been concerned throughout, however, only with fairly well known examples of downdrift and downstep. In order to qualify as an adequate general theory of terracing, the system must also be ca­ pable of accounting for examples of upstep and upsweep, as well as other less well known, more complicated cases of downdrift and downstep. In this concluding section I will outline some of the possible extensions and refine­ ments of this system which may be made to accommodate these latter cases. First of all, the occurrence of upstep and upsweep cannot be accommodated readily within the present system without some refinements. A close look at the nature of our rules reveals that what we need to do is replace the rules in (6) with their mirror image: (27) a. Construct a left-branching tree over a se­ quence of tonal feet. b. Given two sister nodes Ni and N 2 , label N 2 as L iff it branches.

232

AFRICAN LINGUISTICS

All the other rules suggested earlier may remain applica­ ble, A further modification that has to be made is that the tonal feet formed for upstep or upsweep must be for­ mally distinguishable from those formed for downdrift and downstep, so that one may apply (6) or (27) properly. Fur­ thermore, since it often happens- that a sequence may con­ tain instances of both downstep and upstep, it will be necessary to represent the two types of terracing by two trees. The following example is taken from Hyman (1979:12): (28)

The last H is marked as being upstepped. Since it is pre­ ceded by a downstepped H, the effect of the upstep is to bring the pitch of the following H back to the height of the first H. In the following representation, the upper tree accounts for downstep, and the lower tree for upstep: (29)

If the amount of pitch drop per instance of upstep is equal to that per instance of downstep, then an overall evaluation of the two trees in (29) may be interpreted as representing all the relevant information contained in (28). Whether or not this is entirely correct remains yet to be seen, against a wider range of data on upward ter­ racing, comparatively little of which has been reported in the literature.11 Another area which may call for some refinement of the system concerns languages in which downdrift occurs only to Hs in a sequence of alternating Hs and Ls, while the Ls remain constant at a rock-bottom low level of pitch. For example, ignoring the effect of an upstep el­ ement, the ShiTswa sentence (30) is reported in Weimers

233

TONE TERRACING

(1973:92) as having the surface form (31), where all the Ls are realized on a constant low level: (30) H L H H L H L H!H H vamuwona mufana wahosi (31)

Suppose that Weimers' transcription is correct.12 The generalization here, it seems, is that Hs and Ls may be­ have differently in some languages with respect to pitch assignment. For such languages, I suggest a bifurcation of the pitch-register in the following manner. First (4b) ap­ plies to assign the second, fourth, and fifth Hs each to an independent foot node, and (11) assigns the sixth H to another foot. Then all Hs are projected to one register and all Ls are projected to a different non-overlapping register. These will be called the upper and the lower register, borrowing the terms from traditional discussions of tonogenesis. In this way, all tones that are assigned to the beginning of a foot are in the upper register. Af­ ter (4a) and (4c) apply, we obtain the following represen­ tation, corresponding to (31): (32) Upper Register

Lower Register The postulation of two separate registers in the way suggested makes certain interesting claims. It also offers

234

AFRICAN LINGUISTICS

an interesting analysis of the tone patterns of three-tone languages such as Ikom^Yala (Armstrong 1968) and Ga'anda (Newman 1971), which exhibit both downdrift and downstep. I will illustrate with Ikom. In Armstrong (1968), it is reported that Ikom has three underlying tones H, M, and L; and further that eve­ ry H downsteps after a M or a L, and every M downsteps af­ ter a L. Such a situation would seem easy to handle by generalizing the rules in (4b) so that every tone occur­ ring after a lower tone gets assigned to a new foot. How­ ever , the situation is more complex. Consider a sequence like (33)., where two Hs are intervened by a sequence of Ms and Ls that also meet the requirement for downdrift/ downstep: (33); H H  IM  LH   lago lotu 'He put stones in holes at night.' A theory using only one register will predict that the last H will step down twice, once due to its occurrence after a L and once due to the downstep of a preceding M. However, as has been brought to my attention by Barry Schein (personal communication), the transcription given by Armstrong shows that the H steps down only once. This suggests that Hs and Ms behave quite independently of each other with respect to terracing. The problem may be settled in the following way. After (4b) applies to assign the last H to an independent foot, project all Hs to the upper register and all the non-Hs to the lower register. In the upper register, then, the last H will step down on­ ly once, in accordance with Armstrong's report. If the analysis suggested in this final case is cor­ rect, note that there is at most a two-way contrast with­ in each register, so that in the lower register M may be set in contrast with L by the single parameter [aH]. By modularizing pitch registers into two components, a threeway tonal contrast may be reduced to a more simple binary system of tonal features. This result again seems to be desirable, and is consistent with the already highly mod­ ular character of the grammar we assume.13

TONE TERRACING

235

FOOTNOTES

* This paper is a revised and expanded version of a paper delivered at the 10th Annual Conference of the North Eastern Linguistic Socie­ ty. An abbreviated version appeared in the proceedings of the con­ ference. I am grateful to Morris Halle, Nick Clements, Larry Hyman, John Goldsmith, Ken Hale, Jay Keyser, Paul Kiparsky, and Moira Yip for insightful comments on various earlier versions of the paper. 1. In some rare cases, the drop occurring in terracing may equal the distance between contrasting tones. Such cases are termed "total downsteps" in Clements and Ford (1979).. 2. The correctness of this claim is to some extent controversial, but I think it is essentially correct. See footnote 6 for additional comments. 3. See, for example, Peters (1973:141). Hyman (1979:16), on the other hand, proposes that the downstep in the example is developed from a process of tonal simplification, by which a falling tone fol­ lowed by H is simplified to a sequence of two level tones: H fol­ lowed by downstepped H. 4. Thus, it is not correct to say that the position of occurrence of downstep is unpredictable. It is only unpredictable from the sur­ face observable sequence of tones. 5. Although the rule (11) looks very similar to (4b), they must be kept apart, since the two rules are different in scope of applica­ tion, as will be indicated below. 6. Hyman (personal communication) has indicated that some cases of phrase^initial downstep may acually be observed on the surface, and not necessarily neutralized, though he also mentions that this is on­ ly optional. He also indicates that in a sequence of the sort L H ..., the H can be heard as distinctively lower than a non-downstepped H uttered in isolation. These remarks contradict the observations that I have held true, based upon remarks made by other writers (e.g. Armstrong 1968, Peters 1973, etc.) However, I prefer to inter­ pret his facts as indicating that speakers may utter such sequences in comparison with some basic reference point established in other phonological phrases. In such situations, these phrases may be re-

236

AFRICAN LINGUISTICS

garded as non-independent phonological phrases in some sense. Judging from the conflicting reports, it seems that at least it is possible (though not necessary) to pronounce any such initial Hs at a level that is normally used for non-downstepped Hs, especially when such phrases are pronounced in absolute isolation (i.e. without any pre­ ceding or following utterances pronounced within the same speech sit­ uation) . The same situation happens also with stress systems. For ex­ ample, one may pronounce the English article the with or without stress in isolation. Furthermore, the stress of long in Long Island may be perceived to be distinctively weaker in stress (tertiary) than the long in a long island (secondary), even when both are uttered in isolation. These facts, it seems, should not be taken to invalidate the claim that stress, pitch, and tone are essentially relational in na­ ture. 7. The interpretation of this fact as a case of neutralization has been proposed quite early, in Schachter (1961), though no previous system of formal representation, to my knowledge, has accounted for it without some degree of stipulation. 8. The downstepped H in the second example may be perceived as lower than the non-downstepped H in the first example, if the initial L is kept at the same pitch height in both examples. If the downstepped H in the second example is as high as the H in the first example, there will be a "raised"initial L in the second example. Cf. footnote 6 above. 9. The penultimate vowel in this example will have to be deleted be­ fore the final output is obtained. 10. It should be noted that Clements1 (1979) paper does not reflect his current thinking in its entirety. Therefore the discussion that follows in this section does not apply to his current view. In his more recent paper (Clements 1981), he has independently developed a metrical theory that is similar to ours in many ways. A discussion in this section of his earlier theory is worthwhile, however, and through this discussion I believe that an argument can be made for the view that we now share. 11.

For an alternative treatment of upsteps, see Clements (1981).

12. Hyman has informed me that there is some disagreement over whether Weimers' transcription is entirely correct. The controversy is either over (a) whether it is possible for Hs to downdrift with­ out Ls also doing so; or (b) whether there is not some systematic difference between the two kinds of downdrifting. We tentatively as­ sume that the answer to these two questions is yes. 13. This result is also consistent with the theory of tone features developed in Yip (1980). Yip provides the two parameters [aH (tone)] and [&Upper (register)] to characterize all tones of languages. A conception about the independent nature of each of the parameters is

237

TONE TERRACING

shown to play an important role in the analysis of tones in several Chinese dialects.

REFERENCES

Armstrong, R.G. 1968 "Yala (Ikom), a Terraced-Level Language with three tones," Journal of West African Languages 5.1.41-50. Chomsky, N. 1981 Lectures on Government and Binding, drecht.

Foris Publications, Dor­

Chomsky, N. and M. Halle -1968 Sound Pattern of English, Harper and Row, New York. Clements, G.N. 1979 "The Description of Terraced-Level Tone Languages," Language 55.536-558. 1981

"The Hierarchical Representation of Tone Features," Harvard Studies in Phonology 2.50-107, Indiana University Linguistics Club.

Clements, G.N. and K.C. Ford 1979 "Kikuyu Tone Shift and Its Synchronic Consequences," Lin­ guistic Inquiry 10.179-210. Clifton, J.N. 1976 "Downdrift and Rule Ordering," Studies 7.175-194. Fromkin, V. 1972 "Tone Features and Tone Rules," Studies tics 3.47-76.

in African

in African

Linguistics

Linguis­

Goldsmith, J. 1976 Autosegmental Phonology, M.I.T. Ph.D. Dissertation. Halle, M. and J.-R. Vergnaud 1979 "Metrical Structures in Phonology," mimeographed, M.I.T.

238

AFRICAN LINGUISTICS

Hombert, J.-M. 1974 "Universals on Downdrift: Their Phonetic Basis and Signifi­ cance for a Theory of Tone," Studies in African Languages. Hyman, L. 1973 "The Role of Consonant Types in Natural Tonal Assimilation," in L. Hyman ed., Consonant Types and Tones, Southern Cali­ fornia Occasional Papers in Linguistics, University of South­ ern California. 1979

"A Reanalysis of Tonal Downstep," Journal guages and Linguistics 1.9-29.

of African

Lan­

Hyman, L. and M. Tadadjeu 1976 "Floating Tones in Mbam-Nkam," Studies in Bantu Tonology, Southern California Occasional Papers in Linguistics, Univer­ sity of Southern California. Liberman, M. and A. Prince 1977 "On Stress and Linguistic Rhythm," Linguistic 336. Newman, 1971 Peters, 1973

Inquiry

8.249-

R.M. "Downstep in Ga'anda," Journal of African Languages 10.15-27. A. "A New Formalization of Downdrift," Studies in African Lin­ guistics 4.139-154.

Schachter, Ρ. 1961 "Phonetic Similarity in Tonemic Analysis," Language 37.231238. Weimers, W. 1973 African Languages Structures, Berkeley.

University of California Press,

Winston, F.D.D. 1960 "The 'Mid' Tone in Efik," African Language Studies 1.185-192. Yip, M. 1980 The Tonal Phonology of Chinese, M.I.T. Ph.D. Dissertation.

GLOBALITY IN THE KINANDE TONE SYSTEM

Larry M. Hyman (University of Southern California) and Nzama Valinande (Institut Pédagogique National, Kinshasa & Georgetown university)

0.

INTRODUCTION1

Kinande, a Bantu language spoken in Eastern Zaire, has not been the subject of a detailed tonological analy­ sis. The existing descriptive works often note that the language distinguishes high (H) and Low (L) tones, but there has been, to our knowledge, no attempt to provide unique underlying forms and a system of rules to derive the surface tones.2 In this study we shall present the to­ nal structure of the Kinande verb system and consider its properties in some detail. We shall argue that only H tones are present in underlying forms, and that whenever a H tone is deleted or shifted, a L is left in its place. In this way we are able to recognize intermediate stages in tonal derivations where there are three values: H, L and 0 (no tonal specification). After the application of all the phonological tone rules, a late rule specifies any remaining toneless vowels as L tone. In the following sections we shall first establish the underlying tone of verb radicals (R's) and then the underlying tone of the final vowel (FV) of the agglutinative verbal unit (VU). We shall examine closely the process of

AFRICAN LINGUISTICS

240

H tone anticipation (HTA) and other tone rules, demon­ strating that rule ordering paradoxes require a global reference to -underlying tones. After all of the rules have been formalized and sample derivations provided, we conclude with a brief consideration of the tonal proper­ ties of nouns in Kinande.

1.

THE VERB RADICAL (R)

As seen in the following infinitive forms, two tonal patterns are distinguishable: 3 (1)

a.

e- i-hu:m-a

'to hit'

b.

e- f-tu:m-a

»to send 1

The infinitive belongs to class 5 in Kinande and thus ex­ hibits the initial vowel (IV) e- and the noun prefix ri(and vowel harmony alternants). As observed, the differ­ ence in tone between the two patterns is realized on the noun prefix. Since the same prefix is involved in both patterns and must therefore have the same underlying to­ nal representation, it is clear that the tonal difference in (la,b) is attributable to the difference in the radi­ cals 'hit' vs. 'send'. Notice also in (2) that the length­ ening of the penultimate vowel as well as its H tone are not present when there is a following object: (2)

a.

e- i-hum-a valina:nde

'to hit Valinande'

b.

e- í-tum-a valiná:nde

'to send Valinande'

We therefore conclude that the penultimate H is an intonational feature associated with the final pause boundary. Since we attribute the tonal differences in (1) and (2) to the verb radicals, we are left with the underlying repre­ sentations in (3) . (3)

a.

/e- i-hum-a/

b.

/e- i-tum-a/ H

In (3a) we observe that 'to hit' has no underlying tonal features: if appearing before pause, its penultimate vow­ el will be lengthened and assigned H tone; all other vow­ els will receive L tone by a late tule. In (3b) we observe that 'to send' has an underlying H tone on its verb radi­ cal which, however, must be anticipated onto the preceding vowel - in this case, onto the class 5 prefix ri-. The forms in (4), which involve an object marker (OM) -mu-

THE KINANDE TONE SYSTEM

241

'him/her', provide further evidence for this analysis: (4). a. e-r i -mu-hú:m-a 'to hit him! b. e-rí-mú-tú:m-a 'to send him' The penultimate intonational H is seen on both forms. In (4b) the anticipated H of -turn- is realized on the OM H -mu-, which directly precedes the verb radical. In addition, the infinitive is one of the verb forms which takes a H tone OM if the radical has a H tone. Thus, 'to send him' is underlyingly /e-ri-mu-tum-a/, which undergoes two applications of tone anticipation. Since we shall have more to say about the exact nature of this process, we shall postpone formalizing H tone anticipation (HTA) until a later section. 2.

THE FINAL VOWEL (FV) Having established that verb radicals may or may not have an underlying H tone associated with their radical vowel (RV), we turn now to the question of the final vowel (FV) of the verbal unit (VU). In the infinitive forms in (3) the FV was -a, which did not carry an underlying H tone. It may seem curious at this point in our exposition that we have not chosen to represent it with a H tone, which could be anticipated onto the penultimate vowel when occurring before pause. The forms in (2) would, in this analysis, represent a deletion of the H of the FV when fol­ lowed by an object. This analysis was attempted but led to great complications in the statement of tonal alternations within the verb paradigm. The crucial fact about verb forms in Kinande is that in addition to the H vs. 0 opposition on verb radicals seen in (3), there are only two basic tonal patterns, as far as the FV and any extensions (suffixes) are concerned. The first pattern has already been seen in the infinitive forms: its properties include that it is realized with H-L on its final two vowels when preceding pause, but as L-L when preceding an object. In (5) we illustrate this same pattern via the Perfect I tense, whose markers are [-káßi-] (5) a. tu-ká-ßi-mu-hú:m-a 'we have just hit him' b. tu-ka-ßι-mu-tu:m-a 'we have just sent him' In (5b) the H of the radical -turn- is anticipated onto the OM, as it was in (4b). In order to show that the lengthen­ ing and H tone assignment are properties of the penultimate vowel before pause, consider the related examples in (6).

242

AFRICAN LINGUISTICS

(6)

a. tu-ká-βi -mu-hum- ί ; r-a b. tu-ká-ßi-mú -tum -í : r-a

'we have just hit for him' 'we have just sent to him'

In these examples the applied extension -ir- requires that the OM -mu- be interpreted as a benefactive or re­ cipient, as seen in the glosses. This extension, as also observed, undergoes lengthening and receives the intonational Η tone, since its vowel is in penultimate position before pause. In (71, however, neither lengthening nor penultimate Η tone is found when the VU is followed by an object: (71

a.

tu~ká-ßi -mu-hum-ir-a valiná:nde 'we have just hit V. for him'

b.

tu -ká-ßi-mú-tum-ir-a valiná:nde 'we have just sent V. to him'

Verb forms which thus alternate between (lengthened) H-L and (non-lengthened) L-L on their final two vowels are analyzed with a toneless FV, as originally claimed for infinitives in (3). Before considering the second pattern found on verb forms, it is necessary to point out that some verb tenses are exempt from penultimate lengthening. Examples from the zero tense, whose marker is -ká-, are seen in (8). (8)

a.

tu-ká-mu-hum-ir-â : 'we hit for him' b. tu-ká -mú-tum- ir-â : 'we send to him'

A HL falling tone occurs on the long FV which, however, is lacking when the verb forms are followed by an object: (9)

a.

tu-ká-mu-hum-ir-a valiná:nde 'we hit V. for him'

b.

tu-ká-mú-tum-ir-a valiná:nde 'we send V. to him'

We propose the underlying representations in (10).

THE KINANDE TONE SYSTEM

(10)

a.

243

/tu-ka-mu-hum-ir-a-a/ H b. /tu-ka-mu-turn-ir-a-a/ H H There is a double FV which both blocks penultimate length­ ening and which results in the penultimate H being as­ signed to the first FV -a, yielding the HL falling tone in (8) . The lengthening rule, which requires a pause boundary, will somehow allow the two FV's in (8) to be realized as a long vowel. In the absence of a pause bound­ ary, as in (9), the two vowels will coalesce as a single [a]. The curious non-anticipation of the underlying H of the tense marker -ka- onto the subiect marker (SM) in (10) H. will be addressed below. Having established therefore that VU's differ ac­ cording to whether they permit penultimate lengthening before pause or not, we are ready to turn to the second tonal pattern. We first exemplify this pattern with verb forms in the P 2 past tense whose forms are the pre-initial (PI)„ mó-, -a-, and the ending -ir-e: (11) a. mó-tw-a-mu-húm-i :r-e 'we hit him' b. mó-tw-a-mu-turn-f :r-e 'we sent him' We predict on the basis of what has been established thus far that the H on the OM -mu- in (lib) is the result of tonal anticipation from the verb radical -turn-. We observe H that a double H tone is required in this second pattern, but the rule(s) conditioning its placement can only be as­ certained by first examining additional forms. In (12) we replace the OM's with the noun object 'Valinande' we have seen in earlier sentences: (12) a. mó-tw-á-húm-ir-e valiná:nde 'we hit Valinande' b. mó-tw-a-túm-ír-e valiná:nde 'we sent Valinande' Two interesting tonal effects are observed. First, in (12a), the H that was on the OM in (11a) is now on the tense marker -a-. We hypothesize that the two H's of this tone pattern go on the radical vowel (RV) and the pre-radical vowel (PRVL when the radical is underlyingly tone­ less, as is -hutn-. In (12b) we note that the two H's have remained on the same vowels as in (11b), i.e. on the RV of

AFRICAN LINGUISTICS

244

and the tense suffix -ir-. On the other hand, the H or has not been anticipated onto the tense prefix -a*-, something to which we return below. It is impor­ tant to observe, whatever the correct statement of this pattern, that the two H tones do not delete when the VU is followed by an object. Thus, neither of the two H's is attributable to intonation the way the penultimate H of the first pattern was. This is our primary argument for regarding the first pattern as having a toneless FV. We shall now argue that the FV of the second pattern has a H tone. To complete the data needed to determine the proper­ ties of the double-Η pattern, consider the forms in (13), which have the applied extension -ir-, an OM, and a noun object as well: (13) a. mó-tw-a-mú-húm-i-ir-e valiná:nde 'we hit V. for him' b. mó-tw-a-mú-tum- í- ír-e valiná:nde 'we sent V. to him' In (13a). the two H's are on the RV and PRV; in (13b) they are on the penultimate and antepenultimate V's.4 The FV in (13) is thus -e-, which somehow yields a doubled H pat­ tern. H Other tenses with doubled H tones will also have an underlying H FV. As pointed out already with respect to the toneless FV in (8)-(10), this second pattern with a H FV may also involve a double FV, preventing penultimate lengthening: (14) a. tw-a-mú-húm-a-a [-a:] 'we hit him' b. tw-a-mú-túm-á-a [-a:] 'we sent him' In (14) the examples are from an alternate of the Ρ2 tense which has the tense marker -a-, but the FV -a-a, realized, as indicated, as a long vowel. As seen clearly in (14b), the penultimate Η is on the first of the two FV's. The radical -tum- receives the antepenultimate Η and, as beH fore, the OM -mu- adquires Η tone by Η tone anticipation from the verb radical. We can summarize our findings in the following way: (15)

a.

Underlying

H RV+HFV

Surface

=

Η on penultimate and ante­ penultimate V's; Η of RV anticipated onto preceding V, if possible (see below).

THE KINANDE TONE SYSTEM

b. c.

0 RV + H FV H RV + 0 FV

= =

245

H on RV and PRV. H of RV anticipated onto preceding V, if possible; H on penultimate V before pause. d. 0 RV + 0 FV = H on penultimate V before pause. It should be clear that all V's between the RV and the FV are toneless (0). The SM's and OM's are usually toneless, (see note 7) , while the tense markers may, according to the case, be either H or 0. In the next section we shall make more precise the notion of H tone anticipation. Before addressing this is­ sue, it is important to point out two additional facts about the H FV pattern. The first concerns VU's such as in (16), still from the Ρ 2 tense: (16) a. mó-tw-a-mú-hú:m-a 'we hit him' b. mó-tw-á-mú-tu:m-a 'we sent him' The tones in (16a) indicate unambiguously that the FV must have an underlying Η in this tense: thus we obtain H's on the RV and PRV, since -hum-is a toneless radical. In (16b), however, we obtain a pattern we have not seen up to now. The Η on the OM -mu- is clearly from anticipation from -tum-. In addition, the H on the tense marker -a- is Η from the OM, which receives a phonological Η because it precedes an underlyingly Η tone radical (cf. note 7 ) . But why are the last two V's L tone ? If the FV had been tone­ less, we would have gotten a penultimate intonational H in (16b), exactly as was observed in (lb) (cf. (3b), (5b), (6b) and (8b)). Thus, it is necessary to say that the un­ derlying tone of the FV in (16b) is H, but that this H is deleted. Care must be taken, however, that in deleting the final H in the context of an immediately preceding H verb radical, the resulting forms do not merge with H-0, i.e. with forms such as (3b) which receive the penulti­ mate intonational H. The second note concerns a few verb radicals such as -hek- 'carry', which are toneless, but which have the re­ alization in (17a) when occurring with a H FV: (17) a. mó-tw-á-hek-i :r-e 'we carried' (= irreg­ ular verb) b. mó-tw-á-hí-i :- 'we arrived' (= regu­ lar verb)

AFRICAN LINGUISTICS

246

The normal pattern of RV and PRV H is exemplified in (17b), obtained from the toneless radical -hik- 'arrive1 and the H FV -e (cf. (lla) and (12a) In (17a), however, we obH serve the PRV H without the RV H. The toneless verb radi­ cal -hek- 'carry' and a few other radicals (e.g. -sek' laugh', -sat- 'play') exhibit the expected PRV H tone, as in (17a), but will presumably have to be marked with a special rule feature stating that they cannot take the RV H of this pattern. The regular verb radical -hik- suggests that the exceptionality of the other radicals is not phonologically predictable.5

3.

H TONE ANTICIPATION

In our discussion so far we have assumed that some­ how an underlying H is anticipated onto a preceding V. We noted one curious problem for tone anticipation in (10). There it was observed that the H of the tense marker -kawas not anticipated onto the SM tu-. Similarly we H note that the H of a PI such as mo- is not anticipated off H of the VU (leaving a L tone in its place). We therefore propose that the VU has the structure in (18). (18)

[

[ ] ] VU X F Y The VU consists of a foot (F) with contents Y preceded by a pre-foot with contents X. X includes any pre-initial morphemes preceding the SM (e.g. mo-), the SM, and some H tense markers (e.g. -ka- and -a-). Y includes some tense/ H aspect markers, the OM's, the verb radical and extensions and FV. We now can say that H tone anticipation is re­ stricted to applying within the foot. H tones in the prefoot thus remain on their underlying V's. A second fact falls out from this pre-foot/foot di­ vision of the VU. In (12b) we noted a curious problem for tone anticipation. There it was observed that the H of the radical -tum- was not anticipated onto the tense markH er -a-. An additional example of this failure to obtain H anticipation is seen in (19b). (19) a. u-hum-e-e [-e:] 'you hit !' b. u-tum-é-e [-ê:j 'you send !'

THE KINANDE TONE SYSTEM

247

In these imperative forms the FV's are underlyingly -e-e. With the toneless radical -hum- in (19a) we thereH fore obtain H on both the PRV and the RV, and with the H tone radical -tum- we obtain H on both the penultimate H and antepenultimate vowels in (19b). However, in (19b) the H of the radical is not itself anticipated onto the SM u- 'you'. A special rule is thus required that will delete the underlying H of a verb radical if it is initial within the foot. Care must be taken that this deletion does not cause H radicals to merge with 0 tone radicals after application of this rule. In (20b) we provide a form in the P3 tense which has undergone both the dele­ tion of FV H and RV H: (20) a. /mo-tu-a-hum-a/ =» mó-tw-á-hú :m-a H H 'we hit' b.

/mo-tu-a-tum-a/ =» mó-tw-a-tu :m-a H H H 'we sent' The complications that a derivation such as in (20b) cre­ ates for a rule ordering solution will be seen in the next section. Before turning to this and related prob­ lems, though, consider the forms in (21). (21) a. tú-|ya-hum-a-a 'we hit' b. tu-ná-|ya-hum-a-a 'we hit indeed' In (21a) the SM tu- 'we' is H tone on the surface, al­ though it is known that SM's are underlyingly toneless. The hypothesis is that this H is from the following P1 tense marker -lya- which has an underlying H tone that H can be anticipated. In (21b) we see the confirming evi­ dence: the underlying H of -lya- is anticipated onto the H toneless marker -na- 'indeed' and the SM is therefore re­ alized with a L tone. It is clear that the foot boundary occurs between the SM and the tense marker -lya- in (21a). H Thus, anticipation is possible across the foot boundary. The specific rule which deletes the H of the RV when it is initial within the foot does not affect -lya-, since H the latter is not a verb radical, but rather a tense mark­ er. Its H is therefore free to be anticipated outside of the foot onto the SM in (21a).

248

AFRICAN LINGUISTICS

The exact determination of the foot boundary and the tone of each tense, aspect or mood morpheme is a difficult task, although we believe the foot to be the only way the complicated VU and its tonology can be successfully char­ acterized.

4.

RULE ORDERING PARADOX

We have pointed out a couple of cases where rule or­ dering would appear to be crucial. The purpose of this section is to demonstrate that rule ordering cannot ac­ count for the Kinande data. It will subsequently be con­ cluded that some other mechanism is called for if the Ki­ nande verb forms are to be correctly derived. We shall limit ourselves to a discussion of the fol­ lowing rules : (22) a. H tone shift (HTS) b. RV H tone deletion (RHD)  H tone anticipation (HTA) d. Penultimate H tone assignment (PHA) It will be shown that the above rules interact in such a way that PHA must both follow and precede HTS. Other par­ adoxes can be demonstrated with the remaining rules (H tone doubling and FV H tone deletion) or with slightly different formulations of one or another rule. In other words, we believe there is no escape from some rule order­ ing paradox unless a global "trace" is kept of H tones which are shifted, deleted or anticipated. Consider the four rules in (22). HTS refers to the process by which the H of a FV is shifted to the postradical V (abbreviated, V 2 ) if the radical is toneless. As we shall see in the following section, we assume that this H is first shifted to the V 2 , then is anticipated on­ to the RV, and finally is doubled onto the PRV. Other con­ ceptions were considered, but had to be abandoned because of rule ordering paradoxes. It is clear that the HTS rule must precede the RHD rule exemplified in (12b). and (19b) . Recall that this rule deletes the H of a H tone radical if this radical occurs initially within its foot. If RHD were to apply before HTS, H tone radicals would merge with toneless radicals. The RHD rule must in turn precede HTA or else the H of the foot-initial radical in (12b) and (19b) will be in­ correctly anticipated across the foot boundary.

THE KINANDE TONE SYSTEM

249

Finally, HTA must precede PHA, an intonational bound­ ary tone assignment rule. If the penultimate intonational H were assigned prior to HTA it would incorrectly be an­ ticipated onto the antepenultimate V in verb forms. The order we have established, therefore, is as listed in (22): HTS precedes RHD which precedes HTA which precedes PHA. The paradox now involves the relationship between HTS and PHA. Consider the underlying VU in (13a), given in (23) without an OM: (23) mo-tu-a-[ -hum-ir-ir-e H H (underlying tones) H H (by HTS) H H (by HTA) H H H (by HTD) * H H H H (by PHA) As seen, the VU in (23) has within its foot only a FV H tone. This H is first shifted to the V 2 by HTS. Then it is anticipated onto the RV by HTA. (It is doubled by H tone doubling (HTD), although this is not important for the argument here.) After HTD the desired form mó-tw-áhúm-ir-i:r-e is obtained by the low-level gliding and pe­ nultimate lengthening rules. However, as indicated in the last line of (23), the penultimate intonational H will be assigned incorrectly by PHA. In order to prevent this from happening, we must order PHA prior to HTS, i.e. the first rule in the series of ordering relations in (22). But we have already seen that PHA must follow HTA which must follow HTS. Thus PHA must apply both before and af­ ter HTS, hence an ordering paradox. Our intuition is that the penultimate H tone assign­ ment rule should be late, since it is intonational in na­ ture. However, the shift rule must occur early. The nec­ essary conclusion to draw is that when HTS shifts the H of the FV to the V 2 , a trace must be left behind. We con­ sider what this trace might look like in the following section.

5.

CAPTURING GLOBALITY

The problem discovered in the last section is that PHA, an intonational process, must not apply if at an earlier stage in the derivation the FV had a H tone. This covers also the case where the H of the FV is deleted be-

250

AFRICAN LINGUISTICS

cause it is directly preceded by a H tone radical (e.g. as in (16b) and (20b).).. In other words, the tone system requires global reference to underlying H tones which are no longer there. The three basic approaches that are available to us are as discussed with respect to Luganda by Hyman (1983) and enumerated below: (a) We could assume, contrary to what has been pro­ posed here thus far, that underlying forms in Kinande are fully tonified with H's and L's. Any time a H tone is shifted, deleted or anticipated, a mid tone (M) is left in its place so that it can be distinguished from under­ lying L tone. A late rule will change all M tones to L. This system starts with a H v s . L opposition, develops a H vs. M vs. L intermediate opposition, and then ends up with the H vs. L surface opposition which is required. The M tone is thus no more than a diacritic use of a phonolog­ ical feature (in the sense of Kiparsky 1968/73). Note in this regard that our choice has been in part arbitrary. We could have begun with H vs. M underlying tones, leaving a L tone in the place of a shifted, anticipated or deleted H. A late rule would still convert all M's to L's. Other ad hoc feature solutions are not hard to imagine. This so­ lution is to be dispreferred because of its arbitrariness. (b) A second solution employs the asterisk notation introduced for Tonga by Goldsmith (1982) and for Luganda by Hyman (1983). Instead of underlying * H vs. L, we be­ gin with some V's having an asterisk (V), others lacking this asterisk . An early rule (perhaps the earliest rule) assigns H tone to each ν (and perhaps L tone to each although this can be put off until later, as in the third solution below). When the Η of the FV is deleted or shifted, the * remains, so that a rule such as PHA can refer to a final vowel , as opposed to ■- .. In other words, some of the work would be done by the * vs. ° dis­ tinction. This solution is more straightforwardly diacrit­ ic in nature than the solution in (a). It would be hard to claim that each * represents an "accent", as has usual­ ly been the case for solutions employing asterisk nota­ tion. The sole reason for using *'s in Kinande would be to have a double representation of the underlying tonal opposition: * vs. ° quickly becomes Η vs. L. After HTS, FHD etc. a third value, L, is created. Obviously, a dou­ ble representation of any phonological opposition will successfully combat problems of globality. The result, however, will be much more power and redundancy than we actually need to describe sound systems. (c)_ The globality problem encountered in Kinande (and

THE KINANDE TONE SYSTEM

251

several other Bantu languages) is restricted to tone. In this regard, solution (a) is to be dispreferred to solu­ tion (b), since a third phonological value is a strategy that could be employed more generally in phonology. Spec­ ifically, solution (a) makes the prediction that there will be segmental cases of globality parallel to the tonal ones found in Bantu languages. Since one does not want to open up the doors for all kinds of globality, a principled rejection of solution (a) for its ad hoc and arbitrary character would be welcome. Solution (b), note, could conceivably be extended to create false oppositions such as vs. , unless the */° opposition is restricted only to vowels. But even in this case it could be used, say, as a diacritic distinguishing vowel qualities if aster­ isks are not restricted to cases of tone. Pending a more satisfactory definition of * which does not treat it as an anomalous entity in phonology, we shall propose what we believe to be the best approach to problems in global tonology. As stated at the outset of this paper, we propose that Kinande has only H tones in underlying representations. We thus begin with an op­ position between H and 0 or, in Stevick's (1969) termi­ nology, between "tonic" and "non-tonic".6 When a H tone is shifted, deleted or anticipated, a L is left in its place. This yields the intermediate H vs. L vs. 0 oppo­ sition we require. Finally, as a last rule, any V which has not received a H or L from a tone rule (or from its lexical representation) will be assigned a surface L tone. The advantages of this kind of analysis are dis­ cussed by Hyman (1983) and Hyman and Byarushengo (1984). By starting with "toned" vs. "untoned" V's, we are able to capture the "marked" status of H tone in these Bantu languages and to incorporate exactly the kind of globality encountered. We need not postulate ad hoc feature specifications (e.g. M in solution (a)) nor use ad hoc diacritics such as * (solution (b)). In fact, the claim is inherent in the H vs. 0 underlying system that only languages that have this kind of opposition (as opposed to completely tonified underlying forms) will have globality.

6.

THE RULES

Having established the underlying H vs. Ø tonal op­ position for Kinande, we now formalize the rules re­ ferred to in earlier sections. These rules apply in the

252

AFRICAN LINGUISTICS

order listed. Rule 1. Final H tone deletion (FHD). If the final two V's of the VU both have H tones, change H-Η to H-L: (24)

Rule 2. Radical H tone deletion (RHD). If the first V of the foot is the radical vowel and is H, change this H to L. (25)

•Rule 3. H tone shift (HTS). If the FV is H and the RV is toneless, shift the H of the FV to the post-radical (V2) vowel: (26)

Note that this rule crucially refers to the presence of a third vowel between the RV and the FV. Since the vari­ able X in this rule can also include a V, there may be more than three vowels within the structural description of this rule. What is critical is that no change will take place if there are only two vowels, i.e. if there is a toneless RV followed directly by a H FV. Rule 4. H tone anticipation (HTA). Change any H tone occurring within the foot to L and place a H on the preceding vowel: ( 2 7 ) { c o n d i t i o n : V must be within H foot)

The restriction of HTA to occurring within the foot must be stipulated by a condition rather than by placing a [F boundary, since the V which anticipates the foot-in­ ternal H may be either inside or outside the foot. Note that HTA applies from left-to-right, so that a sequence

THE KINANDE TONE SYSTEM

253

such as in (28a) will become (28b) rather than (28c): (28) (a) V  V  V X (b) V  V  V X H H H L H H H L (c) V  V  V X H L *H L L In (b) we see also the reason why there is a parentheti­ cal (L) included in the formulation of rule 4. Rule 5. H tone doubling (HTD). If the last V of the VU has a L tone, and there is no other H tone intervening between it and the last H of the VU, the latter is dou­ bled onto its preceding vowel: ( 2 9 ) 

V H

o

X

+

V L

VU]

(where X does n o t t a i n V) H

con-

This rule is formulated as introducing a second H tone, although one could equally well associate the single out­ put H tone with the two vowels, i.e. with its own V and the preceding V. Rule 6. Penultimate H tone assignment (PHA). If the FV of a word (here, a verbal unit) is toneless, and if this word appears before pause, assign a H to its penul­ timate vowel: (30)

V

C0

V

//

0 This boundary tone may be assigned either to a toneless or a L tone penultimate V. Rule 7. L tone assignment (LTA). At this point, any V not having a tone receives a L. It is possible that a single L is assigned to any sequence of untoned syllables or that a separate L is assigned to each V. Nothing seems to depend on this distinction.

AFRICAN LINGUISTICS

254

7.

SAMPLE DERIVATIONS

The above seven rules will derive all of the verb forms we have considered. Additional low-level rules will be required, for example, to assign penultimate length­ ening, create glides and vowel coalescences etc. These rules go beyond the scope of this paper however. We now provide three sample derivations illustrating the opera­ tion of each of the seven rules we have established. (31) /mo-tu-a-[mu-hum-ir-ir-e/ 'we hit for him' (P2) H H UNDERLYING TONES H H L HTS (rule 3) H H L L HTA (rule 4) H H H L L HTD (rule 5) H L H H L L L LTA (rule 7) [mo-tw-a-mu-rfm-ir-i: r-e]

8.

(32).

/mo-tu-a-[t um-a/ H H H H H L H L L H L L L [mo-t w-a-tu:m-a]

'we sent' (P3) UNDERLYING TONES FHD (rule 1) RHD (rule 2) LTA (rule 7)

(33)

/tu-ka-[mu-tum-a-a/ H H H H L H H L H L H H L H L [t u-ká-mu-t um-a-a]

'we send him' (0) UNDERLYING TONES HTA (rule 4) PHA (rule 6) LTA (rule 7)

CONCLUSION

In the preceding sections we have seen the motiva­ tion for a tonal analysis of Kinande involving only under­ lying H tones. Seven rules were formalized and exemplified in different verb tenses. In this final section we consid­ er briefly some of the tonal properties of nouns. We shall limit the discussion to nouns having four syllables: an initial vowel, a noun class prefix (CV-), and two stem syllables. Because nouns are potentially subject to PHA (rule

255

THE KINANDE TONE SYSTEM

6)., it is necessary to compare their realization both be­ fore pause and in some neutral environment. The neutral environments that have been found include adjectives such as -l¡to 'heavy' and the numerals ' 7 ' , '8' and'9'. The following table gives the nine different patterns found on nouns having the structure V-CV-CVCV: (34)

a. b. . d. e. f. g. h. i.

before - |i to before ,// L-L-L-L L-L-H-L L-L-L-L

L-L-L-L

L-H-L-L

L-H-H-L

L-H-L-L

L-H-L-L

L-H-H-L

L-H-H-L

L-L-H-L

L-L-H-L

L-L-H-L

L-L-H-HL

L-H-H-L

L-L-H-HL

H-L-H-L

H-L-H-HL

## 59 11 52 50 21 24 9 1 5

example -ku-γ ú: lu

'leg'

ε-k i-sa :nga 'island1 ' arm' o-mú-ka:|¡

'woman'

a-ká-hú: ka

' insect'

o-ßu-ho:t i

'beans' 'girl'

e-k í-n í: ya : 'anger' ó-|u-sú:sû:

'giddi­ ness'

The examples are cited as they would be pronounced before pause. It can be noticed in these forms that all nouns undergo penultimate lengthening, even those that have long final vowels, e.g. (34g-i). On the other hand, not all nouns undergo PHA (rule 6) . Some of the above patterns may be exempt because their FV is underlyingly H. At least one pattern, (34b), poses an interesting problem, however. Our sample includes 11 nouns which are pronounced L-L-L-L before 'heavy' and also before pause, as opposed to 59 nouns which are pronounced L-L-L-L before 'heavy' but which undergo PHA to be pronounced L-L-H-L before pause. The nouns in (34a) are underlyingly toneless; if those represented by (34b) are also toneless, they will have to be marked [-rule 6 ] . This is a strange set of af­ fairs, however, since PHA is an intonational rule and we do not expect lexical items to have rule exception fea­ tures preventing the assignment of intonation to them. An alternative, which is still troublesome, is to start with either a floating H after these nouns or, equivalently, a lexically associated L tone on their FV. Either solution would provide a phonological account of why PHA is blocked. Even if we dismiss pattern (34b) as exceptional in nature, we still have at least one regularly occurring pattern more than the logical combinations of underlying tones would predict. (We shall assume that patterns (34gi) are also exceptional. Many of these nouns are borrow-

AFRICAN LINGUISTICS

256

ings or reduplications.) The IV and noun class prefix are toneless. The possible tones on stems should therefore be 0-0, 0-H, H-0 and H-Η. We identify (34a) with 0-0 and (34c) with H-0. Thus, the H of > 'arm' is anticipated H onto the noun class prefix and may, if occurring before pause, receive a new penultimate H. We can also identify (34d) with H-Η, which undergoes FHD(rule 1) and HTA onto the noun class prefix. (There appears to be no foot struc­ ture in nouns.) This leaves (34e) as the realization of 0-H, undergoing HTA and HTD (rules 4 and 5). Among the commonly occurring patterns, (34f) remains unaccounted for. The establishment of clear underlying tonal repre­ sentations for nouns is difficult and a subject requiring further attention. What we wish to point out as possibly relevant for our analysis of verb tones is the fact that nouns show (a) evidence of H tone anticipation from an initial H on a following word, and (b) "junctural" H tones between them and certain following elements. The following summarizes the tonal alternations we have observed on nouns: (35) a. Nouns of all tonal patterns will acquire a final H when followed by any of the fol­ lowing : 1. adjectives in the class of -nene 'big'8 2. the possessive pronoun 'their' 3. the demonstrative 'this/these' 4. the preposition/conjunction na 'with, and' 5. the referential marker -o 6. the numerals '1 * and '6' 7. preceding a new clause beginning with 'after' and 'if' b.

Nouns which have only L tones and which al­ low PHA will receive a final H when fol­ lowed by any of the following: 1. 'which' 2. 'my', 'your', 'his', 'our', 'your pl.' 3. associative marker 4. the numerals '2-5' 5. 'all'

c.

Nouns of any tonal pattern allowing PHA will receive a final H when occurring just before the VU, i.e. at the end of the sub­ ject NP.

THE KINANDE TONE SYSTEM

257

d.

Nouns will fail to acquire a final H no matter what their tonal pattern when fol­ lowed by: 1. adjectives in the class of-|ito • heavy'8 2. the numerals '7-9' We assume that (35a) represents cases where the word fol­ lowing the noun begins with an underlying H. This wordinitial H can be anticipated onto a preceding noun of any tonal pattern by HTA (rule 4 ) . We have no account of the pattern represented by (35b). (35c), however, seems to represent a special case of what we have termed PHA. Ex­ actly the class of nouns that can receive PHA before pause can receive final H tone when directly preceding the VU. This has suggested the possibility to us that there is a phrase boundary (%) between the subject and the predicate as in Luganda (see Hyman (1983)) and that this boundary has a H boundary tone. Since the final pause boundary // includes a % boundary there also is a H boundary tone, but also a following L boundary tone. In other words, the prepausal boundary tones might be as in (36). (36)

X

% // H L This would require a slight restatement of PHA, but would allow us to tie together the phrasal H tone with the pre­ pausal HL found whenever the vowel before a phrase or pause boundary is toneless. Here, as throughout the Kinande tone system, more research is required to work out the most general statement.

258

AFRICAN

LINGUISTICS

FOOTNOTES

1. This paper is the result of collaborative the two authors in Los Angeles during 1982-3. on the speech of the second author whose work the subject of a dissertation in progress for

research conducted by The analysis is based on Kinande grammar is Georgetown University.

2. Previous work mentioning tone in Kinande includes Baudet (1947), Fraas (1961), Furere (1967), Kahindo (1981) and Kavutirwaki (1978). While none of these works treats tone in depth, Kavutirwaki at least marks tone on dictionary entries, and Kahindo, who treats our penul­ timate H tone as a "tonal accent", sorts out some of the same issues we deal with in this paper. In our study the vowels i,e,u,o,a are [-tense], while the vowels are [+tense]. 3. In our transcriptions, only H tone is marked (by an acute ac­ cent) . Surface H tones are indicated also by the acute accent; under­ lying H tones are indicated beneath the segmentals, following the practice of autosegmental phonology. 4. Although it has not been explicitly stated, it should be clear that penultimate lengthening is a late rule creating an extra vowel length which is not counted in establishing which V is penultimate or antepenultimate. 5. A single verb radical, -yend- 'go', which is underlyingly tone­ less, has only a H RV when its FV is underlyingly H. Thus, compare mó-tw-a'-Yénd-i :r-e 'we went', with the examples in (17). 6. Another analysis which was considered, but rejected, was to have underlying L vs. 0 in Kinande. Tonal anticipation would have thus assigned a H to any V preceding a L tone V. While this did not work out well in Kinande (where we would have had underlying L tones with­ in the foot, but underlying H tones preceding the foot), it is pos­ sible that some other Bantu languages, e.g. Tonga or Sukuma, might be profitably analyzed in this way (cf. Hyman and Byarushengo 1984). 7. The tone of OM's depends crucially on the verb form. Although we have not exhaustively studied this question, it appears from preliminary investigations that there are three possibilities: (a) some tenses treat the OM's as toneless; (b) some tenses treat the OM's as H tone; and (c) some tenses assign a H tone to an OM only if the verb radical is itself H tone, underlyingly. 8. The adjectives which do not assign a H tone to a preceding noun include the following (as pronounced in class 1 before pause): 'heavy', mú:-|i 'tall', mu-lé:re 'young', mú-hya mú :-hya 'new', mu:sâ: 'only', mu-ßu:ya 'good, generous', and mu-tswa:tswâ: 'mean'. Those which do assign a H to the FV of the preceding noun

THE KINANDE TONE SYSTEM

259

inlude: mú-né:ne 'big', múr-βΐ: 'bad', mú:-kê: 'small', mú-ky:hi 'short', mu-ßi:si 'raw' and mw-é:ru 'white'. We analyze the former adjectives with a toneless concord prefix mu-, and the latter with a H tone concord prefix

REFERENCES

Baudet, Guilbert 1947 Elements de Grammaire Kinande. Bruxelles: Establissements Généraux d'Imprimerie, Soc. An. Fraas, Pauline A. 1961

Ά Nande--English

and English-Nande

Dictionary.

Washington,

D.C.: Laubach Literacy Fund, Inc. Furere, Medard 1967 "Esquisse grammaticale de la langue nande". Mémoire de Li­ cence, Université Lovanium de Kinshasa. Goldsmith, John 1982 "Accent systems". In N. Smith and H. V. Hulst (eds.), The of Phonological

Structure

Representations,

47-64. Dordrecht:

Foris. Hyman, Larry M. 1983 "Globality and the accentual analysis of Luganda tone". Journal

of

Linguistic

Research,

2.1-40.

Hyman, Larry M. and Ernest Rugwa Byarushengo 1984 "A model of Haya tonology". In George N. Clements and John Goldsmith (eds.), Autosegmental Studies in Bantu Tone, Dor­ drecht: Foris. Kahindo, Lufungula 1981 "Description syntaxique du nande". Thèse pour le Doctorat du 3e Cycle, Université René Descartes, Paris V. Kavutirwaki, Kambale 1978

Lexigue Nande-Français,

Français-Nande.

Kinshasa/Gombe:

Editions du Secrétariat Général de l'Episcopat du Zaire. Kiparsky, Paul 1968/73 "How abstract is phonology ?" In O. Fujimura (ed.), Three Dimensions

of Linguistic

Theory,

5-56. Tokyo: TEC.

260

AFRICAN LINGUISTICS

Stevick, E.W.

1969

"Tone in Bantu". International tics, 35.330-341.

Journal of American

Linguis­

LE MOT CHOMAGE ET SES IMPLICATIONS SOCIO-POLITIQUES DANS LA VOIX DU CONGOLAIS (1955-1959)

B, Kadima-Tshimanga (I.S.P., Bukavu)

En 1955, un Congo encore affaibli par la guerre et devenu incapable d'apporter une pierre édificatrice soli­ de à la subsistance de l'économie de la Métropole, forte­ ment touchée par la guerre, et au développement de sa propre économie, avec son industrie débutante ébranlée dans ses fondaments mêmes, doit faire allégeance avec les conditions économiques mondiales. "L'effort de guerre" ordonné dès 1947 par le Gouverneur Général P. Ryckmans, s'il peut venir à la rescousse de toute l'économie congo­ laise, constitue juste un appoint à une industrie secon­ daire qui, quelques années plus tôt, "a pu se développer au Congo grâce au fait que la Belgique, n'ayant aucun monopole en matière d'importation, n'avait qu'un intérêt mineur à freiner la création sur place d'industries manu­ facturières. Ainsi le Congo satisfaisait-il lui même une grande partie de ses besoins que les nations voisines"1. La lecture lexicologique que nous faisons du texte de la Voix du Congolais se veut "lecture d'une societé"2· D'un point de vue synchronique, nous posons la question de savoir comment "des phénomènes linguistiques, éventu­ ellement des domaines partiels d'une langue, constituent un système en fonctionnement en un point donné du temps, ou plutôt pendant un certain laps de temps"3. Les consi­ dérations suivantes ne prétendent pas à l'exhaustivité ni à 1'elucidation complète du domaine en question, à savoir les relations sémantiques du mot chômage et leurs inci­ dences sur la société conqolaise de 1955 à 1959. Nous vi­ sons plutôt à l'explication progressive de là relation lexicale entre les classes socio-économiques au Congo

262

AFRICAN LINGUISTICS

belge et la trichotomie sociale par le biais de l'inté­ gration de l'infrastructure. La revue la Voix du Congo­ lais, pris comme motif, peut être considérée comme fai­ sant partie d'une pragmatique en tant que support du lan­ gage d'une catégorie sociale, celle des évolués. Elle permet la référence aux rôles sociaux des auteurs car, en fait, "c'est au niveau de la pragmatique que les interre­ lations entre la structure de la société et de la culture d'une part, et les formes linguistiques de l'autre, jou­ ent un rôle essentiel". Nous nous rendrons, certes, à cette évidence que des actes de parole déterminés institutionellement se décrivent à l'intérieur de la même com­ posante pragmatique "d'un système idiosyncratique" puis­ qu'ils dépendent des normes socio-culturelles valables dans une communauté linguistique déterminée. Dans la Voix du Congolais, le vocabulaire semble suggérer une solution un tant soit peu humanitaire au problème du chômage. Cette solution est envisagée, nous semble-t-il, moins par élan philanthropique pur que pour préserver les acquits de la colonisation. Les expressions suivantes qui, dans la revue, entretiennent des relations paradigmatiques avec chômage nous paraissent sans éguivoque: un danger pour la société, de lourdes répercussions Le souci primordial semble être sociales et politiques. celui de sauver l'image du Congo "oasis de paix". La me­ nace que contient le mot spectre dans le syntagme le spectre du chômage est ainsi celle des luttes des classes que pourraient entraîner le développement du chômage et ses retombées politiques. C'est, à notre avis, dans cette seule mesure que la Voix du Congolais propose d'écourter le temps du chômage. La revue, en dépit de ses opinions politiques, accuse l'organisation sociale belge au Congo, qui n'offre pas les mêmes chances d'embauche à la popula­ tion indigène. L'état de chômeur semble être une situa­ tion de fait infligée à un peuple avide de travail. L'ex­ pression le chomage involontaire est porteuse de ce grief que la Voix du Congolais retient contre la politique so­ ciale du gouvernement colonial. Sont présentés comme cau­ "le se du chômage, et justifient l'adjectif involontaire, recours aux machines" et "l'engagement des employés étrangers". Les responsabilités sont ainsi déterminées en ce qui concerne le manque de travail dont souffrent bon nombre de Congolais. Mais par-delà l'apparente lutte pour la résorption du chômage, la Voix du Congolais craint plutôt des bouleversements socio-politiques auxquels sem­ ble exposé le Congo en ces cinq dernières années de la colonisation belge. Elle met, dans un même champ séman­ tique que chômage, virus et plaie-moderne.

LE MOT CHOMAGE

263

Les deux images, virus et plaie, se situent dans un même champ conceptuel. En effet, si le "virus" constitue la cause d'une maladie, "la plaie", elle, est une vérita­ ble maladie par la douleur physique qu'elle provoque. Placés dans une même orbite sémantique que chômage, virus et plaie, empruntés au registre médical, expriment res­ pectivement deux réalités différentes. Le chômage, en tant que "virus", provoquerait la chute de l'organisation socio-politique en place au Congo belge de 1955-1959. Par contre, pris comme "plaie", le chômage représente la gan­ grène qui mine, lentement peut-être mais sûrement, l'os­ sature sociale du Congo. C'est pourquoi la Voix du Congo­ lais met les autorités en garde contre ce double "danger" que constitue le chômage. En outre, si plaie moderne et virus entrent dans le champ sémantique de chômage, ils se rapprochent également du syntagme exode rural et dénoncent de la sorte l'aban­ don des activités agricoles traditionelles des villages au profit d'une prétendue organisation du travail en ville. L'adjectif moderne est sans équivoque à ce sujet. Le chômage ne peut être que l'apanage de l'organisation du travail salarié en ville. Le phénomène est inconnu dans les villages où chaque paysan adulte est dans l'o­ bligation de disposer d'au moins un champ afin de subve­ nir à sa subsistance. Dans certaines tribus du reste, la possession des plantations était -est encore- considérée comme une condition de passage à l'âge adulte, et un gage de la stabilité de l'individu et de sa famille. Elle est même parfois la condition essentielle au marriage. Il apparaît dans la Voix du Congolais, sur le même étymon que chômage, le vocable les chômeurs. La configu­ ration sémantique de ce mot est marquée par un fort désé­ quilibre sur le plan de ses équivalences. Celles-ci re­ couvrent tantôt des sens péjoratifs, tantôt des sens net­ tement méliorés. Dans le même domaine sémantique que chô­ meurs, la revue présente des expressions assez émouvantes: les victimes d'une situation sociale, les hommes et les foyers sans ressources, tous ceux qui manquent de travail, ceux qui ne peuvent plus se procurer du travail, ceux qui se sont écartes des villageois depuis de longues années, des déracinés, les chômeurs involontaires; les sans-tra­ vail involontaires, les chômeurs noirs, les hommes sans travail.

Le soubassement linguistique de ce classement pré­ sente le chômeur comme un homme se trouvant dans une si­ tuation sans choix. Le chômage ne peut donc pas être rat­ taché à une attitude oisive ou paresseuse. Elle est l'in­ éluctable conséquence d'une situation creée par "l'exode rural". Les Congolais émigrés en ville ne peuvent, faute

AFRICAN LINGUISTICS

264

de terrain et à cause du "déracinement", se livrer aux travaux de champ. Le vocabulaire utilisé justifie l'exode rural par le désir de trouver de l'emploi dans les grands centres. L'échec rencontré et l'état de fait créé par le chômage sont une dénonciation, sur le plan du discours, de la précarité de la politique sociale au Congo. La pro­ motion sociale reste une promotion fragmentaire. A l'in­ suffisance des écoles s'ajoute celle de l'emploi. Les Congolais en sont d'inévitables victimes, d'autant que, dans certains secteurs, les employés préfèrent recourir aux services des "employés étrangers". Ce fait explique tout le sens de l'adjectif involontaire dont sont coiffés chômage,

chômeurs

et

sans-travail.

Le manque de travail a causé un tel désenchantement que la revue utilise l'expression chômeurs professionnels. Toutes les voies restent donc bouchées. Les chômeurs vi­ vent dans le désespoir de trouver un emploi quelconque. En même temps, l'adjectif étranger dans le syntagme des employés étrangers recouvre un sens univoque. Il se si­ tue, lexicalement, dans la série oppositionnelle des ad­ jectifs noir et congolais, et rétrécit, par conséquent, son sens aux seuls étrangers Blancs. Le mot chômeur entre dans la formation du syntagme nominal la misère des chômeurs. Ce syntagme s'insère dans le même réseau discursif que hommes et foyers sans res­ sources. Toute une catégorie d'individus est ainsi ca­ ractérisée, selon la Voix du Congolais, par la faim et la "misère". Le programme colonial en matière sociale de­ meure, vraisemblablement, un leurre. Toutefois, au cru du vocabulaire qui exprime la situation du chômeur congolais s'oppose un vocabulaire qui, loin de compatir à la misère des sans-emploi, échaffaude un tissu lexical fait de ter­ mes assez inflexibles. A chômeur sont associés alors les syntagmes un homme aigri et dangereux, classe des mécontents, ces hommes qui

les parasites, ont connu la

la ville,

des aigris. Une seule qualification couronne cette liste: le vol. La société congolaise de 1955-1959 offre ainsi une structure ternaire: employeurs

VS. travailleurs

vs.

sans-travail.

Le Congo semble vivre, sur le plan social, dans une at­ mosphère tendue de luttes et de conflits sociaux. Le pre­ mier conflit oppose travailleurs et employeurs. Et subsidiairement à ce démêlé employeurs vs travailleurs naît, sur le plan lexical, l'opposition main-d'oeuvre autoch­ tones vs main-d'oeuvre étrangère. L'existence de la main d'oeuvre étrangère entraîne la constitution de la cohue de sans-travail. La Voix du Congolais présente la situa­ tion de ces derniers sous deux optiques lexicales contra-

265

LE MOT CHOMAGE dictoires. Néanmoins, elle renchérit sur la situation en stigmatisant "l'exode rural". Selon toute apparence, le début de l'exode coïncide avec ce que la revue appelle "l'occupation européenne en Afrique". Le terme même occu­ pation dénonce la substitution de l'autorité coloniale à l'autorité indigène et cela, par la force. Il entretient dans son réseau d'associations les syntagmes dépeuples par

les

recrutements

et manque

d'institutions

sociales.

Ces deux dernières expressions évoquent respectivement le système de travaux forcés auxquels la colonisation as­ treint les Noirs, et l'insuffisance de l'encadrement des Congolais dans les centres urbains ainsi creés. L'inévi­ table fâcheuse conséquence en est la montée du chômage et l'abandon du travail de la terre. En rapport avec le mot chômage,

la

Voix

du

Congolais

atteste une suite verbale qui désigne les actions des chômeurs. En tout, quatre syntagmes verbaux déterminent toute l'action de cette catégorie de personnes: souffrir du chômage, devoir être renvoyés au village, jouer un et politique du pays, payer rôle dans l ' é v o l u t i o n sociale une certaine contribution aux anciens employés pour être embauchés.

Par ces syntagmes verbaux, la misère des sans-travail semble provoquer chez les auteurs de la Voix du Congolais un apparent élan de pitié. La revue propose alors que les chômeurs retournent alors au travail des champs. Nous avons néanmoins de sentiment, compte tenu de l'évolution lexicale du discours, qu'il s'agit plus de la peur que d'une conviction humanitaire. Les auteurs de la Voix du Congolais craignent de constituer dans leurs environs im­ médiats une "classe d'aigris", souvent incontrôlables, qui risque de se réveiller, de déferler sur la ville et d'ébrécher l'organisation sociale et politique en place. Le mot classe lui-même présente les chômeurs comme une force socio-économique et même politique certaine. On ne peut, au Congo de 1955-1959, compter sans eux. Le dernier syntagme verbal (payer une certaine con­

tribution

aux anciens

employés pour être

embauchés) dé­

nonce un fléau: celui de la corruption et du népostisme qui président au recrutement de nouvelles unités dans divers services. Les grands ennemis des chômeurs sont des employés. Le syntagme verbal considéré les présente comme des personnes véreuses et vénales. Avec les em­ ployeurs, ils forment, à des niveaux différents sans doute, la gangrène qui mine la société congolaise de ces dernières années de la colonisation. Le vocabulaire, à ce sujet, est sans équivoque. Employeurs et employés dé­ signent, en tant qu'entrées du lexique, ceux qui freinent l'accroissement de la main-d'oeuvre salariée et favorisent

AFRICAN LINGUISTICS

266

ipso facto la montee du chômage. Il y a là lieu de croire que tous les chômeurs du Congo belge de 1955-1959 sont, des "chômeurs ainsi que le suggère la Voix du Congolais, involontaires".

NOTES

1.

Bouvier, P. L'Accession

d'analyse 39. 2.

sociologique,

Matore, G. La Méthode

du Congo belge

à l'indépendance.

Essai

Bruxelles, Ed. de Sociologie ULB, 1965, p. en lexicologie.

Domaine

français,

Paris,

Didier, 1953, p.XV. 3. Brekle, H.E. Sémantique, 4.

Idem, p.102.

Paris, Armand Colin, 1974, p.90.

A NON-LINEAR ACCOUNT OF THE SYLLABLE IN LUGANDA

Francis Katamba (University of Lancaster)

1.1.

INTRODUCTION

In this paper I outline a non-linear or autosegmental analysis of the syllable in Luganda, a Bantu language of Uganda. The place of the syllable in Luganda phonology has been previously discussed by Ashton (1962) and Cole (1967). Both authors suggest a prosodic treatment of the syllable in this language. I shall attempt to develop this insight and use it to show that processes of vowel lengthening and shorten­ ing, vowel deletion and glide formation which at first sight seem to be unrelated all involve rules which are sensitive to syllable structure. Before I do that, however, I shall present a brief sketch of the consonant and vowel system of Luganda. (1)

Vowels i

(2)

e



 u

Consonants ρ

β

m

f

ν

t

d

n | s z



j

Ρ





g



w

268

AFRICAN LINGUISTICS

The structure of the syllable is generally simple as (3) below shows (3) (C) (G) (:) (where N= nasal, C= consonant, V= vowel, G= glide (y,w), N= syllable nasal, ' : length) examples V CV NCV N

e 'at' mu 'in' lu-ga:-nda 'Luganda' m-pa: 'I give' (only initial, pref ' consonantal nasals are sy1CGV: mwe:zi •month' labic) GV wa-la 'far' CCV ddu-ka 'run1 CCV: ddu:-ka: 'shop' Note that nasals are always homorganic with a following consonant and that both vowels and consonants may be long or short. 1.2.

STRONG CONSONANTS

There have been two approaches used for showing length in Luganda (and in general phonological theory) one involving double segments, which is the most widely used approach, and the other involving the feature [long] (Katamba 1974). In this paper I suggest that neither of these two approaches is correct and argue that length is a prosodic feature of the syllable and not a property of individual segments. I shall develop this point presently. For the moment let us first note that all vowels contrast for length but of the consonants only the fol­ lowing can be long: (4) Chart of strong/long/geminate consonants. PP bb

tt

cc

kk

dd

jj

gg

mm

nn

00

THE SYLLABLE IN LUGANDA

ff

269

SS

vv zz The convention of using double letters to show length is used here without prejudicing the theoretical argument. Phonetically strong consonants tend to be approxi­ mately one and a half times longer than their weak coun­ terparts. In addition they are generally more fortis ar­ ticulations (Katamba 1974) . Strong seems to me a more suitable label for these segments than long. There are interesting gaps in the pattern in (4): w 1 y/ are never strong; a voiced continuants /β corresponding voiced stop does duty for them in environ­ ments where strengthening is required. (5) bb => β = » w ggw => dd | => jj  Meeussen (1955) showed that diachronically most Luganda strong consonants arose when a Proto-Bantu syllable containing * i was lost and a following consonant was made both more fortis and longer. This process happened par­ ticularly to the initial consonant of noun class 5 roots where Proto-Bantu prefix was *d¡ but it was not a morpho­ logically conditioned change. As (6) shows, strengthening occurred virtually anywhere where a consonant was pre­ ceded by * i . (6) Proto Bantu Luganda ku - y¡ ba

'steal'

ku- bba

ku - y i ta

'kill' 'branch' 'ten'

tta

*d i - tabi * di - umi

ttaßi kkum i

Strong consonants also o c c u r w h e n a nasal prefix is a t ­ tached to a nasal root initial c o n s o n a n t . Such a nasal prefix is normally derived from a P r o t o - B a n t u *ni w h i c h has lost its v o w e l as a r e s u l t of s y n c o p e . T h i s nasal r e ­ presents either t h e class 9/10 prefix or the first p e r s o n singular p r o n o u n : (7)

η

-

ηo

m - ma i a

(n - ηjo [+ long] / V Assuming that this rule has applied, a conspiracy need not be invoked in order tΦ account for the length adjustment processes we observed. All that is required is a modification of the association lines linking segments to the middle tier of the hierarchy. In the case of a vowel being deleted, the adjustment in the association lines is shown in (21) where an asso­ ciation line is shifted from the vowel in the coda to the lengthened nucleus vowel which, after 20 has applied now extends into the coda position: in brief, (20) feeds (21). (21a) UR

THE SYLLABLE IN LUGANDA

(21b)

277

UR

PR

In the case of glide formation, an association line is shifted to the onset from the nucleus to which it is underlyingly attached but this doesn't affect the sylla­ ble weight as (20) turns the second vowel into a long vowel before (22) applies. (22) UR PR

As both (21) and (22) indicate, syllable weight is normally underlyingly determined: a syllable is heavy if both the coda and nucleus positions in the core are filled· Syllable weight tends to be stable. If rules mod­ ify the segmental tier, syllable weight may be unaffected because it is a property of the syllable as a whole and

278

AFRICAN LINGUISTICS

not a feature of individual segments. You will also have observed in (21) and (22) that I have used the convention that segments which are normally referred to as syllabic are associated with the nucleus of the syllable- When vowels are non-syllabic they are not linked to the syllable nucleus by association lines; they are instead linked to the onset. This convention which has been suggested by others in the literature might make the use of the feature syllabic unnecessary. One requirement which a theory of the syllable is expected to fulfil is to provide a principled way of di­ viding words into syllables. Most studies of Bantu pho­ nology assume that except of syllabic nasals, which are tautosyllabic, all syllables end in a vowel. A syllable boundary is therefore automatically placed after a vowel together with a consonant to its left. This obviously as­ sumes that except for syllabic nasals, the syllable in Bantu is always open. In this paper I have rejected the closed vs open syllable dichotomy, preferring instead the light vs heavy dichotomy. The motivation for this is what I shall call the rhythmic law in Luganda. This is the constraint which states that a strong consonant may not be preceded by a long vowel. It is a constraint which overrides any rule which would otherwise require the lengthening of a vowel. (23) V => [- long] / - CC The effects of (23) are shown in (24a) where lengthening occurs and in (24b) where it is inhibited. (24a) UR PR nti a - kola ntya:kola 'that (s)he that (s)he - work works' tu - a - ku - kußa twa:kukußa 'we hit you' we past you hit (24b) UR nti a - bba ntyabba 'that (s)he steals' that (s)he steal tu - a - ku-kußa twakkußa 'we hit you' we past you hit (casual speech) Forms like nt yabba should have a long first vowel just like nt ya: ko 1 a as a result of the application of rule (20) but they do not. It seems plausible to suggest that the rhythmic law blocks (20) because there are only two posi­ tions in the core of a heavy syllable and in the case of Luganda in these two positions there is only room for

THE SYLLABLE IN LUGANDA

279

either a long vowel or a short vowel and a strong (long) consonant. For ntyabba I assume the derivation in (24c) (24c) UR

PR

One effect of the fact that there are only two posi­ tions in the core which can be maximally occupied either by a long vowel or by a short vowel and a strong conso­ nant is that lengthening rules cannot apply cumulatively where that would lead to the number of segments in the core exceeding the permitted maximum. Thus, for example, an underlying vowel which satisfies the structural de­ scription of the rule which lengthens the second of two adjacent vowels and also meets the structural description of the rule which lengthens vowels before NC sequences does not undergo cumulative lengthening. (25) UR PR ßa - Ø - goßa ßagoßa 'they chase' they pres chase ßa - a - goßa ßa:goßa 'they chased' they past chase ßa - a - η - goßa ßa:ngoßa 'they chased me' they past me chase *(ßa::ηgo a) A related point that needs to be raised is the con-

280

AFRICAN LINGUISTICS

trast between the behaviour of strong consonants and that of nasals in medial NC sequences. Whereas strong conso­ nants are ambisyllabic, the nasals belong solely to the same syllable as the vowel to their right: the vowel be­ fore an NC sequence can occupy the two positions in the core when it is lengthened by (20). (26) UR PR

•if1 Luganda is a tone language with two level tones High (H) (') and Low (L) (') and a Falling tone (F) (^) which is a sequence of HL within the same syllable. Tone is generally used as a criterial factor in de­ ciding whether a given unit is a syllable. Every syllable underlyingly bears a tone. The tone bearing capacity of a syllable depends on the weight of that syllable. A light syllable i.e. V, CV. or N bears one tone. (27) UR PR 'woman1 ' leopard' A nasal in an NC sequence word initially is also tone bearing and is therefore regarded as being tautosyllabic while word medial nasals in NC sequences are not. A heavy syllable can bear a maximum of two tones but a light syllable cannot. (28) UR PR 'child' 'children'2 The examples in (29b) show contour tones on short vowels. They would represent serious counter - examples to my analysis if I had stated that only long vowels can bear contour tones in Luganda.

THE SYLLABLE IN LUGANDA

(29a)

281

UR PR ò-|í -gula ò | íg ù|á 'you will buy' You fut. buy ò -- | í - sòmá ò | í sòmá 'you will read' You fut. read

(29b) ò - |í -'tta' olîtta 'you will kill' You fut. kill ò - |í -'dduka ò|îddùká 'you will run1 You fut. run Diachronically strong consonants arose from a syn­ copation process: VC> CC as we saw in the first section 12 22 of this paper. The syllable tier was lost together with the consonants and vowels but the tone hierarchy survived. A change along these lines from Proto-Bantu can be as­ sumed for Luganda *yìtá > 'ttá 'kill'. It is the tone left floating that combines with the preceding tone. Since word medial strong consonants are ambisyllabic, the second syllable in the form 'o|îtta is heavy and hence capable of bearing two tones.3

3.

CONCLUSION.

If length is regarded as a suprasegmental feature whose realisation depends on the hierarchical structure of the syllable, a simple unified account of a number of phonological rules in Luganda can be formulated without resorting to the concept of conspiracies and risking the introduction of questionable teleological explanations in­ to phonology. Using a multi-tiered approach to phonology and the notion of syllable weight I have attempted to avoid these danglers.

282

AFRICAN LINGUISTICS

FOOTNOTES

1. Protor-Bantu and were extremely high vowels. In several Bantu languages their reflexes induced fortis articulations of some sort. In Tswana, for instance, [d] and [1] are in complementary dis­ tribution, with [d] occurring before / i / and /u/ which are reflexes of Proto-Bantu and 2. LH contour tones are not allowed in Luganda. Where a rising tone would arise from a combination of L and H there are rules which change LH to HL or some other acceptable configuration. This is ex­ emplified by the following derivation. UR

a) b) c) d) e)

mù - yávù mù - ávù mù a: vù mwă:vù mwa:vu

'poor' (cf. , "(s)he becomes poor") (by a rule deleting /y/ after a CV prefix) (by lengthening rule 20) (by glide formation) (by tone adjustment rule LH =>H) in a syllable

3. Some indirect support for the heavy vs. light syllable dichot­ omy is provided by studies of Kiganda music like that of Cooke (1970), Cooke observed that songs often accompanied by a clap-pulse which falls with metronomic regularity every six "syllable units" or "morae". What I have described as heavy syllables contain two such units and light syllables have only one such unit.

REFERENCES

Allen, W.S. 1965 Vox Latina, 1968

Vox Graeca,

London: Cambridge University Press. London: Cambridge University Press.

Ashton, A.N. 1962 "The syllable in Luganda: a prosodic approach" Journal African Languages, 1, 122-66. Clements, G.N. ed. 1981 Harvard Studies in. Phonology University Linguistic Club.

of

Vol. 2. Reproduced by Indiana

283

THE SYLLABLE IN LUGANDA

Clements, G.N. and K.C. Ford 1979 "The description of terraced level tone languages" Language 55, No. 3. Cole, D.T. 1967 Some features of Ganda Linguistic Witwatersrand University Press.

Structure

Johannesburg:

Cooke, Ρ. 1970 "Ganda xylophone music: another approach" African ciety Journal Johannesburg 4(3):62-80. Dinnsen, D.A. 1979 Current Approaches to Phonological Indiana University Press.

Analysis.

Music

So­

Bloomington:

Goldsmith, J. 1979 "The aims of autosegmental phonology" in Dinnsen (1979). 1976a "An overview of autosegmental analysis" Linguistic 2: 23-·68.

Analysis

1976b Autosegmental Phonology. Doctoral dissertation, M.I.T. Re­ produced by Indiana University Linguistic Club. Kahn, D. 1976 Syllable-based Generalisations in English Phonology. Doctor­ al dissertation, M.I.T. Reproduced by Indiana University Linguistic Club. Katamba, F. 1974 Aspects of the Grammar of Luganda. University of Edinburgh.

Doctoral dissertation.

Leben, W.R. 1977 "Length and syllable structure in Hausa" Studies Linguistics, Supplement 7.

in

Newman, Ρ. 1972 "Syllable weight as a phonological variable" Studies frican Linguistics 3. 201-323.

African

in A-

ON THE SYLLABLE STRUCTURE OF CERTAIN WEST AFRICAN LANGUAGES* Jonathan Derek Kaye (Université du Québec à Montréal)

0. Sequences of the form Consonant-Liquid-Vowel (hence­ forth CLV sequences) pose a number of problems for the description and analysis of many West African Languages. In addition, they constitute apparent counter-examples to the theory of syllable markedness expounded in Kaye and Lowenstamm, 1981a. In this article, an analysis of such sequences will be presented which, on the one hand, ex­ plains a number of mysterious properties of these sequen­ ces, and on the other, provides a striking confirmation to the syllable markedness theory. One of the morals re­ sulting from this study is that it is imprudent to make simplistic assumptions regarding the syllable structure of these (and other) segmental sequences. Weimers, 19 73:26-7 noted the difficulty in treating CLV sequences. He cites several examples where Words su­ perficially of the form CLV are to be analyzed as bisyllabic, i.e. CVLV where V represents an extra-short (bare­ ly audible) vowel. Weimers' arguments are based on evi­ dence from tonal patterns. In the cases cited, languages distinguish monosyllables from bisyllables in terms of certain tonal patterns. It is shown that CLV sequences behave like bisyllabics in terms of their tonal behaviour and hence, should be so considered. A particularly revealing example of this sort comes from Agni, an· Akan language spoken in the eastern Ivory Coast (Pilote, 1982). Contour tones (sequences of two tones falling on the same syllable) do not occur in Agni verbs. Agni has an intransitive verb marker consisting of a high tone which associates with the verb occurring in phrase final position. In the case of monosyllabic verb

286

AFRICAN LINGUISTICS

stems, a high tone appears on the verb (Agni verb stems are all low-toned) as in (1) . (1)

he cries he leaves he eats

In bisyllabic forms the resulting tonal sequence on the verb is cvcv, as seen in (2). (2)

he sleeps he chats he pinches

Finally, CLV verbs inevitably show verbs (of which more later). (3)

as do G(lide)V he laughs

....

he writes he looks he studies

The most straightforward analysis seems to be that a low toned verb is followed by a floating high tone, the intransitive marker. This high tone associates to the last syllable of the verb stem. In the case of true mono­ syllables this association involves the dissociation of the original low tone, given that contour tones are ex­ cluded from verbs. Thus, we derive monosyllabic verbs with a high tone as in (4). (4) =*·

=>

[ suú

In bisyllables the low tone is retained on the initial syllable with the high tone falling on the final syllable. (5)

a. H =*·

=»

[dàf é]

SYLLABLE STRUCTURE

287

Assuming that CLV and CGV sequences are equally bisyllabic we obtain the following derivations: (5) b.

H Η =>

=>

Having arrived at this point in the derivation, the sur­ face forms may be obtained by assuming either (a) the forms are in fact bisyllabic phonetically, or (b) there is a late process of syllable reduction which involves the retention of both tonal and segmental melodies. The point is that the tonal patterns of Agni verbs show clear­ ly the bisyllabic nature of CLV and CGV sequences. Examples of this sort lead Weimers to conclude: "In g e n e r a l such c l u s t e r s [CLV clusters/JDK] should not be a s ­ sumed u n t i l i t i s c l e a r t h a t they w i l l not confuse t h e t o n a l a n a l y s i s ; a t b e s t they are r a r e , and an a n a l y s i s of any l a n ­ guage which recognizes them should be considered suspect un­ l e s s t h a t a n a l y s i s i s e x p l i c i t l y defended a g a i n s t o t h e r a l t e r ­ n a t i v e s . " 1973:27. In the cases which I would like to consider here, the CLV sequences are indeed monosyllabic, however they do not involve, as I will show, CL "clusters" if "cluster" i s interpreted in the appropriate way. I will begin the d i s ­ cussion with a brief summary of the theoretical framework which I will employ in this a r t i c l e . 1. The syllable structure which I assume here is given in (6) below.1

288

AFRICAN LINGUISTICS

(6)

The syllable is a branching hierarchical structure consisting of an obligatory onset and rime. The rime is expanded into a nucleus and an optional coda (not present in every language). The terminal nodes of the syllable structure are not segments but rather a series of points called the skeleton. Segments are linked to syllable structure via these points. It should be evident that the feature [syllabic] may now be dispensed with. The posi­ tion in syllabic structure of a given segment determines its status as syllabic (occupying a neclear position) or non-syllabic (occupying some other position). The addition of the skeleton allows one-to-many re­ lations between segments and terminal nodes of syllable structure. Thus, long vowels and geminate consonants can be represented as single segments associated with more than one skeletal point. (7) [bo:t]

[fatto]

SYLLABLE STRUCTURE

289

In like manner more than one element of the segmental tier may be associated with a single element of the skeleton. This, in fact, is the representation of affricates (in­ cluding prenasalized consonants) and light diphthongs.2 (8)

[ baet ƒ ]

[trwa]

[la:mba]

By universal convention the less sonorous of the two el­ ements associated to the same point is produced first in the speech chain. A provision must be made to account for the position of the nasal in prenalized consonants ([ b] not [b ]). Affricates are, of course, stop-initial [t∫] and not [∫t]. Light diphthongs invariably have the less sonorous element to the left [wa] and not [aw].3 Such a convention is required since I assume that the relative order of. segments is determined by the order of the skel­ etal elements to which they are attached. This idea can be expressed by -the following equivalences: (9)

The convention referred to above excludes, say [Jt] as a possible affricate, [bm] as a possible (pre)nasalized consonant, and [aw] as a possible light diphtong. Let us consider light diphtongs in detail. A light diphtong is simply a non-branching nucleus associated with two segments. It follows that in terms of metrical theory light diphtongs should behave exactly like syllables of

290

AFRICAN LINGUISTICS

the form (C)V. In fact they are syllables of that form. If one assumes that any [+sonorant] may potentially ap­ pear in the nucleus position of a syllable,4 there is ev­ ery reason to believe that all such elements may form a part of a light diphtong. The familiar examples of light diphthongs involve GV sequences. We shall see below that this does not exhaust all the possibilities. To place this last point in the proper context I will now turn to the theory of syllable markedness (Kaye and Lowenstamm, 1981ai. It is possible to define the class of possible syl­ lable inventories by considering just two parameters: the most highly marked onset and the most highly marked rime. Once these two parameters have been fixed, the syllable inventory of a given language is defined in the following way: (10) The presence of a syllabic constituent (onset, rime) of markedness Ρ implies the presence of syllable constituents of markedness P-l, P-2, ..., P-Q (Q=P). Thus, specifying the most highly marked rime and onset of a language defines an inventory containing these constit­ uents and every other constituent lower in markedness. Markedness can be calculated simply by counting the num­ ber of skeletal points dominated by the constituents in question: the greater the number of points, the greater the degree of markedness. In other words, an onset con­ sisting of three consonants, CCC, is more highly marked than one having two consonants, and so on. By principle (101 if a language has an onset CCC, it will also have CC and C.5 The onsets and the rimes of a language are in depend­ ent in the following sense: let M be the set of all on­ sets, and N, the set of all rimes, then the syllable in­ ventory of a language is Μ χ Ν, i.e. the cartesian prod­ uct of the sets of onsets and rimes. What this means is that for a given language, every possible onset (defined in terms of skeletal points not in terms of segments) cooccurs with every possible rime. This principle is clear­ ly a fundamental property of syllable structure and will play a crucial role later in the discussion. I will refer to this property as the Principle of Free Cooccurrence. Notwithstanding the existence of the principle of free cooccurrence, there is one relationship that holds between inventories of rimes and inventories of onsets: the most highly marked rime is at least as high (possibly higher) on the markedness scale as the most highly marked onset. Let us call this relationship the Rime-dominant

SYLLABLE STRUCTURE

291

principle. It may be expressed as follows: (11) Let ρ be the markedness index of the most high­ ly marked onset and r, the markedness index of the most highly marked rime for a given lan­ guage, then ρ < r. For example, the rime-dominant principle excludes lan­ guages with only open syllables (r = 1) which have bran­ ching onsets (p = 2 ) . The reader should be aware that I am not 'referring to individual syllables but rather to syllabic inventories. A language like French having both branching onsets (plume) and branching rimes (sac) may certainly have individual words with branching onsets and non-branching rimes (clef [kle], trop [tro]). The rimedominant principle means that if a language has branching onsets, it must also have branching rimes. With the two principles given above and the notion of markedness, we now have a comprehensive theory of possible syllabic sys­ tems for all languages. Having presented the major ele­ ments of the theoretical framework involved in this study, I will now discuss certain problematic cases.

2. Eastern Kru languages are typically described as having the following syllable types:6 (12) a. CV 'arm' 'villages' 'manioc* b. V 'bananas' àdáβ|á 'market' _ 'he/she' c. CLV 'house' 'sing' 'ask/want' d. C W 'study' 'snail' 'children* Beginning with the last example, it can be shown that C W sequences are to be analyzed as light diphthongs, i.e. they behave metrically like short open syllables. Their representation is as in (13) . (13)

That these forms can only be monosyllabic, is clearly in­ dicated by tonal evidence and their behaviour in harmony processes.7 Hence, pattern (12d) reduces to an instance

292

AFRICAN LINGUISTICS

of (12a). The examples of (12c) appear to pose severe problems for the theory outlined in section 1, and in particular, for the rime-dominant principle (11) . Since Vata has only open syllables, it contains no branching rimes. According­ ly the theory states it should have no branching onsets. Yet, sequences of the form CLV (12c) abound. The problem stems from the assumption that a word like s|^ has the syllabic structure of (14). (14)

This argument is not only unwarranted, it is wrong. The remainder of this section will be devoted to arguments showing that (14) is not the appropriate syllable struc­ ture for examples like (12c). When one considers languages which truly have bran­ ching onsets such as English, French, Ojibwa, etc., one notes that there are invariably constraints concerning what consonants can occur in what position of the onset. ht the very least consonants found in branching onsets conform to the sonority hierarchy (less sonorous precedes more sonorous in the onset) within reasonable limits. In English for example, liquids can occur after (some) ob­ struents and fricatives but not after nasals, liquids or glides. Ojibwa permits only glides as the second member of a branching onset. A reasonable conclusion is that a language with branching onsets invariably has constraints as to what elements may form these constituents. From this point of view CLV sequences in Vata (and the other Kru languages studied to date) are indeed odd. That is, one may pose the question of what consonants may occupy the  position in a CLV sequence. The answer is: any con­ sonant in the language. The consonantal inventory is given in (15a). In (15b) each consonant is exemplified in pre-liquid position. (15)

a.

p b f

t d s

 j

 g

ν

ζ

m 6

η

ρ

η

Ι

y

γ

kp gb

kw gw

W

SYLLABLE STRUCTURE

b.

293

drum to buy to thread house elf to hunt ν Iè clan kill to retain ζ le fish 8 kplïï to shave m la" to swallow kwla forest animal 1* to detach chicken saliva to bite jI à lion  to sing hedge to flow g bIú black mamba sun w gl to tear wIT fingers If CL sequences were analyzed as branching onsets, then Vata and many other Kru languages would have the unique distinction of having no constraints on CL sequences with­ in an onset. Notice that the sonority hierarchy is vio­ lated by these putative onsets. Sequences such as wl-, y I- and γ I- occur freely in Vata. Evidence such as that in (15) indicate that CL sequences are not to be analyzed as branching onsets. We now consider further evidence leading to the same conclusion. In (13) above light diphthongs were discussed. These diphthongs consist of a high vowel followed by a non-high vowel. Any such combination is permitted in Vata. 10 If CL sequences were indeed instances of branching onsets, we would expect to find these types of onsets in combina­ tion with the light diphtongs. This expectation stems di­ rectly from the principle of free cooccurrence discussed above. Thus, syllable structures of the type shown in (16a) yielding phonetic sequences as in (16b), should be present. (16)

pie

a.

[klla]

AFRICAN LINGUISTICS

294

In fact, such sequences are completely impossible in Vata as well as many other Kru languages. Once again if CL se­ quences are analyzed as branching onsets this distribu­ tional gap is unexplained.

3. Given the theoretical framework presented in section 1 above, we have all the ingredients for a successful analysis of CLV sequences in these languages. I propose that the LV component of this sequence is to be analyzed as a liquid-initial light diphthong as in (17) . (17)

[plá]

In other words, LV sequences in these languages have syl­ lable structures identical to the high vowel - non-high vowel sequences. Metrically speaking they are simply short open syllables. One may look askance at the sugges­ tion that a sequence like [la] is a diphthong, but in fact the possibility of such an analysis follows directly from our theory. The simplest conceivable theory of diphthongs would permit any segment that may occur in the nucleus to form part of a light diphthong. Now it is well known that syllabic liquids occur in a variety of languages. Thus, there is no principled basis for excluding them from par­ ticipating in the class of possible diphtongs. To exclude such diphtongs would necessitate complicating the theory in a clearly undesirable way. Having established that LV diphthongs exist as a theoretical possibility, I must now show that they are indeed the appropriate way to analyze the Kru facts. It was observed above that any consonant in the lan­ guage can occupy the Oposition in CL sequences, a fact difficult to comprehend if the sequence is in the onset. Under the diphthongal analysis this is the expected state of affairs, given the free cooccurrence principle. The absence of syllables of the form C L W is easily explained by the diphthongal analysis. Suppose we stipu­ late11 the following constraint for Kru languages: (18) A maximum of two elements may be associated to a single skeletal point.

SYLLABLE STRUCTURE

295

A putative syllable  I la, would have the structure in (19) contra (18). (19)

Thus, the present analysis with a very banal constraint (18) offers an explanation for the impossibility of C L W sequences in Kru languages. Where onsets are truly bran­ ching sequences of C L W are quite possible. French which has both branching onsets and light diphtongs has the ex­ pected combination of the former with the latter.12 (20) pluie trois [trwa] truite emploi No violation of (18) is involved here since these examples have the syllable structure of (21). (21)

French differs from Kru in having the possibility of CL sequences analyzable as branching onsets. This possibili­ ty is a direct consequence of the fact that French has closed syllables, since the branching onset parameter is available only if rimes branch. The theory of syllable markedness also furnishes an answer to the learnability problem. If we assume that there is no significant phonetic difference between syl­ lables of the type (22a) and (22b),

(22)

a.

b.

296

AFRICAN LINGUISTICS

the question arises as to exactly how a child succeeds in assigning the appropriate syllable structure when con­ fronted with a CLV sequence. In the case of the Kru child it must be remembered that having been exposed to no branching rimes, analysis (22a) is not available to her. Accordingly, she must choose analysis (22b). 13 The problem is now to explain why the French child chooses (22a) over (22b) since, in principle, either analysis is available. In order to exclude (22b) a learning strategy (markedness convention) could be added to the effect that in the unmarked case, there is a skeletal point for each segment and a segment for each skeletal point. Other things being equal, an analysis with a oneto-one relation between skeletal points and segments (22a)„ will be entertained before an analysis where such a relation does not hold. Thus, (22a) represents the unmarked case and will be choosen if syllable theory makes this choice possible. The diphthongal analysis of CLV sequences offers an explanation for some otherwise mysterious facts about Kru languages. In addition, the theory of syllable markedness offers an explanation as to how such structures may be acquired. Note further that far from being a counter-example to the theory of syllable markedness sketched above, Kru provides a striking con­ firmation to the claim that languages with no branching rimes cannot have branching onsets. I turn now to a fur­ ther example from Kru illustrating once again the diphtongal nature of LV sequences.

4. Lakota (Dida) has an interesting process of syllable loss related to tone.11* In a sequence of nuclear dominated skeletal points, those with no associated tone are de­ leted. In addition, a rule of the segmental tier deletes the first of two successive vowels. High vowels associated to a skeletal element do not undergo this truncation for reasons discussed in Kaye, 1983. Finally, a mid-high tone is deleted following a high tone. The representation of tone may involve one of the following tonal autosegments: L (low), Μ (mid), MH (midhigh) and H (high). Mid tones are ambiguous in that they may be represented as the linkage of the M autosegment to the skeletal, or else as the absence of any tonal linkage. The mid-tone is the unmarked tone; any syllable with no associated tone is realized phonetically as mid. 15 The following table shows how all these factors fit together in the context of verb plus object-clitic constructions. The (third person object clitics are a, ε, o, i, o and oa.

SYLLABLE STRUCTURE

297

The mid-high tone is lexically associated to these clit­ ics. 16 (23) Verb stem Verb stem - clitic phonetic form a

fc

In (23) the derivations of two high-toned verb stems plus clitic combinations are presented. In (23a) the vowel of the verb is high (referring to tongue height) and associ­ ated to a skeletal point. Accordingly this vowel is not subject to deletion. The mid-high tone is deleted fol­ lowing a high tone and the final nuclear point now no longer anchored by a lexical tone is deleted. The re­ maining vowel associates to the first nuclear point cre­ ating a light diphthong with a high tone. In (23b) the mid-high tone of the clitic is again deleted following the high tone of the verb stem. The stem vowel is non-high making it subject to deletion. The final skeletal point, unprotected by an associated tone deletes as well. The remaining vowel ε associates to the remaining skeletal point and the phonetic form is derived.

298

AFRICAN L I N G U I S T I C S

(24)

a.

fil

'plug [imperf]'

b.

 'throw [imperf]'

i fil - έ ' h e p l u g s i t 1

i

file

 pa - ε 'he throws it' ο ρεε

--(vowel loss)=»

Examples (24) include two imperfective verb forms with midhigh tones. In (24a) the high vowel is -not subject to de­ letion for reasons discussed above. Both nuclear points have lexically associated tones and hence, neither can be deleted. Aside from the ATR harmony, the derivation of [fi!i¿] involves no change. The form is bisyllabic con­ trasting with [fu^é], which is monosyllabic. In (24b) we observe a similar derivation, the difference being that since the stem vowel is non-high it is deleted. The re­ maining vowel spreads to the now vacant nuclear slot yielding the observed form. (25) a,

b.

SYLLABLE STRUCTURE

299

Verb stems with mid tones are illustrated in (25). The mid tone is the unmarked tone. Thus, the stem vowels of these verbs are lexically toneless. Crucially, this means that the first of the two consecutive nuclear points is unanchored by a lexical tone and it is this point which is deleted. Normally high vowels are not subject to dele­ tion but this constraint holds only if they are associated to a skeletal point. When the skeletal point is deleted as in (25), high vowels as well as non-high vowels under­ go deletion. In the case of high toned verb stems there is also the loss of a syllable (23) but it is the second nuclear point which drops. High stem vowels remain asso­ ciated to the skeleton and escape deletion. Light diphtongs are formed by the addition of the clitic vowel to the remaining skeletal point. In (25) the first nuclear point is deleted and with the stem vowel, regardless of its height. For our purposes it is the high-toned stems that are of interest. A high-toned stem with a high vowel (23a) results in the formation of a light diphtong. What hap­ pens in the case of a verb stem of the form CLv, , where v. is a high vowel with a high tone ? If CL sequences were truly branching onsets, one would expect behaviour parallel to that in (23a). We have seen, however, that there is good reason to believe that lv sequences are light diphthongs. If this is true, something special should happen when a clitic is added to a verb stem O f the form CLv.. To see why, consider (26). (26)

— (MH loss)-— loss) —

In the above derivation the MH tone is deleted following a high tone, unanchoring the final nuclear point which should then delete. The high vowel (v.) being attached to a skeletal point does not delete. The question is: what do we do with the clitic vowel, ε. It cannot associ­ ate to the remaining nuclear point since to do so would surpass the limit of two segments per point imposed by (18). The diphthongal analysis predicts that something un­ usual should happen. Before divulging what does happen, let me consider an example of a CLv verb stem with a non-high vowel.

300

AFRICAN LINGUISTICS

(27)

(27), follows the normal derivation of high-toned verb stems. The mid-high tone deletes as well as the final skeletal point and the stem vowel. The loss of the stem vowel is normal since it is non-high. The remaining clit­ ic vowel is free to associate to the nuclear point forming another LV light diphthong. Since only two segments are in­ volved, (18) is not violated. Consider now a verb stem with a final high vowel. (28) 61 f 'dig' ßlf - έ »dig it1 [6 ι lé] As predicted by the theory something special has in­ deed happened, of lé is bisyllabic unlike other high-toned verb stems (23, 27), and a "metathesis" has taken place with i and 1 changing places. To understand what has taken place we must consider the structure of this form in de­ tail. In the following representation I add the complete syllable structure.

It has already been noted that if the unanchored point deletes, the orphaned final vowel will have no point to which it can attach itself. The result of the application of the syllable deletion process is the creation of a structure which is ill-formed syllabicly. It has long been noted by many researchers that if the result of ap­ plying an otherwise general phonological process is an ill-formed structure, the process fails to apply. The principle accounts for the bisyllabic nature of ¿fie. It remains to account for the apparent metathesis. Let us continue the derivation where we left off in (29). 18

SYLLABLE STRUCTURE

301

(30) a.

In (30a) we have taken up the derivation after the de­ letion of the mid-high tone of the clitic. In (30b) a re­ presentation corresponding to [ f Ié] is given. We need a reasonable story explaining how to get from (30a) to (30b). One notes a number of cases of apparent line-cros­ sings which are excluded by all known versions of non­ linear theory. In fact these crossing association lines are, as I will show, simply a result of trying to repre­ sent three dimensional phonology on a two dimensional sheet of paper. Tones will clearly occupy a separate tier, hence, they will not be involved in any line crossings. I will not consider them further.19 In (30b) the I segment has dissociated from the nuclear point and reassociated to a newly created onset point. The i remains associated to its original nuclear point. We now have a bisyllabic form consisting of two CV type syllables, 6 f and le. Two ques­ tions remain: where does the onset point come from ? and, how can the I associate to that point without crossing an association line ? Vergnaud, 1982 proposes a convention to the effect that a segment attached directly to a syl­ labic constituent (O, N, C) produces an intervening skel­ etal point. This convention is given in (31). (31) (where Ρ is a syllabic constitu­ ent and S, a segment) Incorporating this convention into our Lakota analysis we have only to assume that the 1 of the light diphthong as­ sociates directly to the following onset (0). Convention (31) supplies the skeletal point between them. The second question involves the treatment of two segments associated to the same point in the skeleton. I have already suggested above that two segments associated to the same point are unordered with respect to eachother. It is clear that phonetically the liquid precedes the vowel in the light diphthongs but this does not reflect a phonological ordering (concatenation) of segments. This

302

AFRICAN LINGUISTICS

effect is accomplished by applying an interpretative con­ vention of the type suggested above (less sonorous pre­ cedes] . Such an enrichment of phonetic theory is inde­ pendently required to account for the absence of light diphthongs of the form [ai], [ei], [au], etc. Thus, con­ catenation need not hold between all elements of the seg­ mental tier, but only between elements associated to sep­ arate skeletal points. Segments associated to the same point are unordered with respect to eachother. The asso­ ciations in (32a) result in the concatenated string in (32b). (32) a.

b. Applying this model to the example at hand yields the string (33) b - e Thus, the 1 can associate to the following onset without crossing the association line binding the i to the nucle­ ar point. One can wonder why the I must associate to the available onset position and disassociate from its origi­ nal nuclear position, i.e. why bile and not bife ? A reasonable position would be to consider that the unmarked case is a one-to-one relation between the seg­ mental tier and the skeleton (one segment for each point, one point for each segment). If such a relation is pos­ sible for a given skeleton and a given segmental tier, it is obligatory. There is a clear relationship between this point and the question of resyllabification discussed in Kaye and Lowenstamm, 1981b. This explanation is tantamount to saying that when­ ever a light diphthong is followed by a syllable with a null onset, the light diphthong should break apart with the least sonorous element reassociating to the following onset. This is precisely what happens in the case of mid-high verb stems. Recall that mid-high verb stems plus clitics result in bisyllabic forms. In all cases CLV-V verb clitic combinations should yield CVLV sequences,

SYLLABLE STRUCTURE

303

which they do. (34) a.

In the examples of (34) no syllable loss takes place since both nuclear points are anchored by lexical tones. In (34a) the final stem-vowel is high and not subject to vowel loss. The empty onset of the second syllable is filled by the association of the segment 2 directly to the constituent O(nset). A syllabic point is added as per convention (31) . The procedure whereby the empty onset is filled is undoubtedly related to the process of resyllabification discussed in Kaye and Lowenstamm, 1981b. The idea presented there was that resyllabification is effected in such a way that null constituents are eliminated. This is, of course, precisely what is going on in the case un­ der discussion. The available onset is now occupied by the liquid. No line-crossing is involved since 2,1 are not in a relationship of concatenation. (34b) presents more of a challenge to explain. The mid-high verb stem ends in a non-high vowel which is de-

304

AFRICAN LINGUISTICS

leted in the course of the derivation. The remaining vow­ el, ε, winds up occupying both nuclear positions of the form, with the I in the intervening onset position. The problem again is to avoid line-crossings. Suppose that following the deletion of a, the ε propagates normally to occupy the vacated slot as shown above. Now 1 and ε are associated to the same point and accordingly, they cannot form a concatenated sub-string. The "deconcatenation" is shown on the third step in (34b). At this point 1 can as­ sume its normal position in onset. It is simultaneously preceded and followed by ε. Non-concatenated elements cannot be involved in line-crossings and so the correct form is derived. It should be obvious to the reader that a reasonable account of the facts considered in this sec­ tion can only begin once the diphthongal analysis is ac­ cepted. Indeed, one possibly by-product of this analysis would be the development of a thoery of metathesis, or rather, a theory eliminating the permutation of elements as a possible phonological operation. In the above exam­ ple the illusion of permutation was created by the lack of concatenation between the elements of a light diph­ thong. I.do not suggest that all apparent cases of meta­ thesis may be so treated, but it is surely worth noting that a high percentage of metathesis involve sonorant (i.e. potentially nuclear) segments. In this article I have shown that LV sequences can be successfully treated as light diphthongs: two segments associated to the same nuclear point. The advantages of this analysis are quite apparent and have been documented above. In addition Kru languages can now be eliminated as possible counter-examples to the Kaye-Lowenstamm syllable markedness theory. Indeed the theory provides an explana­ tion as to how such an analysis of CLV sequences may be learned. There are two other morals to be drawn from this discussion: the modern phonologist should be wary of making hasty assumptions concerning syllable structure. Syllable structure along with all other aspects of phono­ logical representation must be justified by theoretically sound analyses. Rashly analyzing Vata kle as having a branching onset is no more justified than assuming that German bund [bunt] ends in a phonologically voiceless segment. The second point I wish to emphasize is that facts crucial to even the most fundamental aspects of a given language will not necessarily be obvious to the linguist who has not a fairly detailed knowledge of that language. Reliance on secondary sources is becoming an increasingly dangerous enterprise. Phonological theory has developed considerably in the last ten years, but so has the nature

SYLLABLE STRUCTURE

305

of the evidence needed to justify a particular analysis.

FOOTNOTES

* Research on African Linguistics is supported in part by grants from the Conseil de recherches en sciences humaines du Canada (44410 -81-0503) and the Fonds institutionnel de recherche de l'Université du Québec à Montréal. I have benefitted greatly from discussions with Monik Charette, Hilda Koopman, Jean Lowenstamm, Dominique Sportiche and Jean-Roger Vergnaud. The analysis of Dida metathesis was first presented in a paper by Jean Lowenstamm and me delivered at the 13th NELS Conference, Montréal, 1982. 1. See Kaye and Lowenstamm, 1981a, 1981b, 1982, 1983, and forth­ coming, for details. 2. That is, non-branching nuclei consisting of more than one seg­ ment. On the difference between light and heavy diphthongs, see Kaye and Lowenstamm, forthcoming. 3. The sequence [aw] exists, but either as a (metrically) heavy diphthong or as a closed syllable with w occupying the coda position. Note that light diphthongs are identical in metrical structure to short open syllables; heavy diphthongs have the same metrical struc­ ture as long vowels. 4. This is subject, of course, to certain markedness conditions. That is, not every language which has a phoneme 1, permits its oc­ currence in the nucleus (syllabic I). 5. I will not discuss the question of null constituents here. See Kaye and Lowenstamm, 1981b and Lowenstamm and Kaye, 1983. 6. The examples here are from Vata, a member of the Dida-f group (Kaye, 1982). 7.

The evidence is discussed in detail in Kaye, 1983.

8. I is realized phonetically as η after a nasal, /mla~/ [mnâ"] . 9. 1 is realized as d before I. (/1 I¿/ = [dl¿]. d < 1 is transpar­ ent with respect to tone rules unlike the "real" d. 10.

See Kaye, 1981a for discussion.

306

AFRICAN LINGUISTICS

11. This "stipulation" may indeed be a property of human language and not at all a fact about Kru. Reports of single syllables bearing three tones appear to be the only obstacle to rendering (18) a uni­ versal constraint on phonological representation. 12. See the discussion of French diphthongs in Kaye and Lowenstamm, 1981b, forthcoming. 13.

In principle, a third possible analysis is available, viz.

where the CL sequence is treated as an affricate. Notice that this solution is descriptively inadequate, failing to account for the distributional properties discussed above. There are doubtless, uni­ versal constraints concerning what two segments may be associated with a single onset or coda point (i.e. what is a possible affricate). Such constraints surely involve agreement of point of articulation (pf and ts but not *tf or *px) . Furthermore, CL sequences in Kru do not phonetically resemble lateral africates (tl, dl) found in other languages. 14. See Kaye and Charette, 1981 for a very preliminary analysis of the correlation of tone and syllable loss. 15.

In certain cases a mid tone must be lexically marked.

16. In this and the following representations syllable structure is not shown aside from the terminal points. Other simplifications in­ volving the representation of high (tongue height) vowels have also been imposed due to the two-dimensional nature of the printed page. 17. The clitic vowel has undergone a dominant ATR harmony not rel­ evant to the discussion. 18. The following analysis is a result of my collaboration with Jean Lowenstamm. It owes much to our discussion with Jean-Roger Vergnaud. 19. In fact the propagation of the high tone onto the undeleted syllable is a problem. One would expect propagation only of floating tones (Halle and Vergnaud, 1982). The final syllable should be re­ alized with an unmarked (mid), tone. This is not the case which sug­ gests that the derivation does not proceed as indicated in (30). Other stories come to mind involving changes in the point at which the derivation blocks. I will not pursue this intriguing but complex question here. None of the possible reanalyses attenuate the force of the argument under discussion. For a similar case occurring in French see Kaye and Lowenstamm, forthcoming.

307

SYLLABLE STRUCTURE

REFERENCES

Halle, Morris and Jean-Roger Vergnaud 1982 On the framework of autosegmental phonology, in H. van der Hulst and N. Smith, eds., The Structure

presentations

(Part I),

of

Phonological

Re­

Foris, Dordrecht, 65-82.

Kaye, Jonathan 1981a Les diphtongues cachées du vata, Studies in African Linguis­ tics, 12,225-244. 1981b Comments on the role of the evaluation metric in the acqui­ sition of phonology, in C.L. Baker and J.J. McCarghy, eds., The logical problem of language acquisition, M.I.T. Press, Cambridge, Mass., 249-256. 1982 Les dialectes du dida, in J. Kaye, H. Koopman and D. Sportiche, eds., Projet sur les langues kru: Premier UQAM Montréal, 233-246 1983

rapport,

Harmony processes in Vata, in H. van der Hulst and N. Smith, eds., The Structure of Phonological Representations (Part II), Foris, Dordrecht.

Kaye, Jonathan and Monik Charette 1981 Tone sensitive rules in Dida, Studies supplement 8,82-85.

in African

Linguistics,

Kaye, Jonathan and Jean Lowenstamm 1981a Syllable structure and markedness theory, in A. Belletti, L. Brandi and L. Rizzi, eds., Theory of Markedness in Generative Grammar, Scuola Normale Superiore, Pisa, 287-316. 1981b De la syllabicate, ms., UQUAM. 1982 On the internal structure of the rime, paper read at the seventh G.L.O.W. colloquium, Paris. forthcoming On the notion of concatenation in phonology. (Written version of the paper read at the thirteenth NELS conference, Montréal, 1982). Lowenstamm, Jean and Jonathan Kaye 1983 Compensatory lengthening in Tiberian Hebrew, ms., UQAM. Pilote, Doris 1982 Quelques processus tonals en agni, ms., UQAM. Vergnaud, Jean-Roger 1982 Foundations of generative phonology, paper read at the seventh G.L.O.W. colloquium, Paris.

308

AFRICAN LINGUISTICS

Weimers, William

1973 African

Language Structures,

California, Berkeley.

ON JUNCTION IN TWO KISIU NASAL CLASSES J. Kelly (University of York)

KiSiu is a Bantu tongue spoken in Siu town on Pate Island, off the coast of north-eastern Kenya. The data that forms the basis for this paper was collected in Ke­ nya from two KiSiu speakers, both male, one in his fif­ ties, the other in his early twenties.* The transcrip­ tions given below are those for the older speaker. The younger speaker's corpus is, as far as the relevant mate­ rial is concerned, slightly less extensive, partly be­ cause he did not know some of the more recondite flora terms, partly because he used, in general, more terms from the prestigious Mombasan Swahili of the coast fur­ ther south. But in matters of phonetics his forms are as like the older man 1 s as to allow the latter to stand as representative of both. On the basis of various patterns of agreement, KiSiu can be said to have a system of noun classification. There are some fifteen or so noun-classes, each being characterised by a prefix or set of prefixes. Morpholog­ ical analysis of nominais needs, then, to recognise in them a primary two-part structure of prefix and stem. Prefixes, as the name implies, are ordered before stems in morphological statement. The phonetic material that relates to them is not always, though, separate from, or other than, that relating to stems, nor does this materi­ al always precede that relating to stems. Phonological statement, as viewed here, has as one of its tasks the mediation between such bodies of relevant phonetic mate­ rial, as subjectively identified in utterance, and de­ scriptive categories such as are established for other levels such as morphology. This it does by means of gen­ eralisations and abstractions of various kinds and de-

AFRICAN LINGUISTICS

310

grees. In this paper we focus our attention on the phon­ ology of the prefix-stem junction in two classes of KiSiu nomináis. As in other Bantu tongues, there are complex rela­ tionships between the noun-classes of KiSiu and various grammatical and semantic categories, one relevant one here being the association of pairs of classes with sin­ gularity and plurality. The two classes we are considering are such a pair. This pair of classes is, in addition, as­ sociated with trees, shrubs and plants, and, it has some­ times been thought, with long, narrow objects in general. The following is a set of nomináis in the two classes un­ der consideration. The transcriptions are as I have them in my field-notes with the omission of a certain amount of detail deemed immaterial to the issues under discus­ sion. For the sake of completeness of coverage the list contains the majority of the words in these classes that I have in the corpus. I call these two classes Nasal Classes 2 and 3: they are known to Bantuists as Classes 3 and 4. The Nasal Class 3 list on the right provides plurals for the Nasal Class 2 list. The transcriptions are simplified in that, for ex­ ample, [ƒ] is used for what is t-J] in the original notes: and the resonance markers for the initial nasals in Type la have been omitted, since they are various, and the variability is of phonological significance, as explained below. Symbols have IPA values, with the addition that is used over [m] to indicate dark (velarised) resonance, and < is used over the same nasal symbol to indicate clear (palatalised) resonance. * is used in the glosses to mean f some kind of'. It should be borne in mind that  and V, as estab­ lished in the text below, are abstract phonological cat­ egories, and will not necessarily correspond to what might be 'consonantal1 or 'vocalic1 elements in the data: this observation applies particularly to Types lb and lib. Nasal

l

a

Class

a

r

2

N

a

m

s

a

l

Class

III bone vein pillow river day tree

etc.

3

KISIU NASAL CLASSES

: I ¡ma ¡iarjga

311

root mountain sand door room coconut-palm tree* tree* tail

T

m-vuke m : pepe m : 6uj u Q:koma

m : f u ko

soup millet quarter load canoe cannon cuttlefish tower heat lip tree* baobab tree* tree* cashew-nut tree tree* tree* rice banana-palm bag fish* canoe tree* tree*

All plurals are in [mT-]

312

AFRICAN LINGUISTICS

eel lb

mango-tree tree*

III

lliu

cassava IIa

_

__

lib

moon heart body year fire thorn smoke reef spoon village tree*

IVa

town light dwarf-palm

IVb

All the items in the above list can be seen to be­ long to their respective classes on the basis of the concordial patterns they contract with attendant modifiers and verbs. But they display a wide range of phonetic phe­ nomena in the section of their make-up that relates, broadly, to the prefix-stem junction on the morphological record. The phonologist will interpret these phenomena in terms of units and categories in such a way as to show up what are constants of structure, The interpretation of such features as articulatory postures and the various resonances, both buccal and na­ sal, has to be carried out against the background of more primary prosodie characteristics such as those of sylla­ ble structure and accent, since these are the basic prop­ erties of utterance, along with its rhythmic or pulsing properties. The items we are dealing with here are, on

KISIU NASAL CLASSES

313

another level of analysis, word-units. They are delivered in a particular, isolate, style, and have certain phonet­ ic features that attach to this status, such as those re­ lating to the accentual system. The rhythmic and intonational characteristics of these words are relatively easy to state: and these themselves interlock with other phon­ etic characteristics to provide an amalgam that relates to the phonological property of accent. This amalgam in­ cludes within its scope, as well as rhythmic and intonational ingredients, certain articulatory and phonatory ones, such as the central quality of vowels that accom­ panies shortness and low-falling pitch in final position as marks of non-accent; or implosion that typifies b, d and g as initial  of accented syllables. In the phono­ logical formulae that will be set up to relate to and generalise from our data in the remainder of this paper, no account will be taken of such accentual features, since they are, generally speaking, not relevant to what is under discussion. The formulae that constitute phonological statements for this data relate to two types of junction, close and open. Close is further subdivided into obstruent and nonobstruent, and open into zero and non-zero. Within the formulae we recognise a two-term resonance system, writteny and w , a one-term articulatory system q , and a one-term nasality system ~ . In more general statements for a number of KiSiu nominal classes, both the nasality and the articulatory terms would be offset against other terms in their respective systems, so non-nasal against nasal, for example: but we are looking here at two classes which both have nasality as properties of their prefix-stem junction, and non-nasality does not arise. c These, together with and , used less formally, make up the set of prosodie categories necessary for the statement, that is to say, categories whose implica­ tions for the data are other than punctual. They operate, at least potentially, over structures of one or more syl­ lables. Units whose implications are punctual only, and for identifiable places in structures as set up, are called, in the Firthian approach to phonology adopted here, phonematic. Phonematic units here are C, V and . Since  is a sub-class of V, differing degrees of ab­ straction are involved between the units, both here and elsewhere in the statement, as we move from one point to another in structure. This is because, to deal with this particular aspect of KiSiu phonology, certain units can stand maximally generalised without detriment to the statement. There is, for instance, a lot to be said about the syllable structure of KiSiu stems with respect to re-

AFRICAN LINGUISTICS

314

lationships between successive Vs or between Cs and Vs in sequence. But none of this is relevant to the junction phe­ nomena, so  in stems need not be more closely specified, except in the case of close-type junction, where the prop­ erty of occlusion or non-occlusion has to be given for Ci in stems. The phonological statements appropriate to the junc­ tion types for Nasal Class 2 are: I

Close a b

o b s t r u e n t C 0 f c b ν  V non-obstruent C_^ V  V

II

Open

a

z

b

non-zero

e

r

o



V V  V

and for Nasal Class 3: III Close IV

Open

 V  V

a zero b non-zero Type III items serve as plurals for all Type I items: Type IV, a or b, depending on the quality of the stem vowel, serves as plural for Type IIa, and Type IVb for Type lib. The phonetic realisations of these units and cate­ gories are:  syllabicity q articulatory closure preceding  (all cases ex­ cept la) or concomitant with  (la): c have their appropriate realisation along with q in this latter case w lip-rounding and protrusion, together with raising of the back of the tongue into a posture of open approximation with the velum  lip-spreading, together with the raising of the blade of the tongue into a posture of open ap­ proximation with the front of the hard palate lowering of the velum from the beginning of ut­ terance until the first obstruent It will be noted that q is intermediate between a unit and a category, as presented here. In syllables not characterised by w in Nasal Class 2, q acts as a prosodie element. This kind of indeterminacy is not uncommon when

KISIU NASAL CLASSES

315

setting statements out in this way. There is a problem of ordering there that remains to be solved. In addition to the symbols above, π, , , and are used informally for bilabial, obstruent, non-obstruent, and open, nonopen (vowels). The open vowels in question are [e   ] , the non-open ones [i] and [u]. Again informally used are f,  and b for front, central and back oral articulations. Although these notations have informal status in this pa­ per they would almost certainly be those used formally in a treatment of the phonology of KiSiu nominal stems. The implications for utterance of some of these for­ mulaic elements are considerable. w and Y, for example, have implications for the pronunciation of Vj in stems in Type IIa: the first stem vowel of mwiIi is a good deal backer than that of miili, nyamba has a fronter first vowel than mwamba, and so on. But the detailed treatment of these has to take into account the prosodie character­ istics of such stem syllables, and this is not presented here. The nasal prosody ~ has, likewise, the nasalisation of the whole of such words as as its realisation. In other cases there is a great deal of freedom as to realisation. la, it will be seen, has no resonance category w or  : realisations here of  may be of any resonance type, as well as having rhythmic length and firmness of articulatory contact. The discrepancy in the resonance of , together with this rhythmic length, is quite striking in many cases, and a contributor to the auditory impression of syllabicity which accompanies these elements. In this data nkia 'tail' has a markedly different resonance in the velar nasal, set off against the fronted k: mbuyu 'baobab1 has a central res­ onance in the nasal as against a dark one in []. In this respect these items are quite different, auditorily, from items with nasal prefixes in what I call the Resonance Classes (Bantu Classes 9/10). Nasal prefixes there are short, non-syllabic, and tend to be integrated in reso­ nance. For this and other reasons their phonology is to be stated quite differently from Nasal Classes 2 and 3, although the rather insensitive 'Standard' orthography now in use treats them the same, writing mburn for both 'ostrich' (Resonance Class 1) and 'coffee-plant' (Nasal Class 2 ) . A number of benefits are to be gained from the use of such an approach within phonology. Firstly, it takes into account all the phonetic matter which is relevant to structural statement. It takes this into account in both senses of this rather ambivalent expression, by both

316

AFRICAN LANGUAGES

being drawn from it, and by giving it precise value. Secondly, it uses, or attempts to use, only natural relationships and classes: none of the realisations are other thamaccessible to instrumental investigation. Thirdly, the classes and categories usecT are general enough to be applicable over a wide range of data other than that for which they are established. In some cases the formulaic expression may have to be modified in order to do this, but the modification itself then has linguis­ tic significance. The formulae here were set up expressly and exclusively for KiSiu: they can stand, however, with certain modifications, for KiMvita, the Bantu tongue of Mombasa island, to which KiSiu is held by many authorities to be closely related. KiMvita differs from KiSiu in that i) some stems are reclassified: 'mango-tree' is in lb rather than lib, 'town1 is in la rather than lib ii) there is only q  V  V for IVa iii) in la w is present at  along with q and ~, so there is no realisation throughout  of the f,  or b prosodie characteristics of C 1 , as there would be in KiSiu. Now, none of these modifications involves anything gross, such as the introduction of a new element into a formula, or, say, the reordering of the elements: and the judgements of writers that KiSiu is 'like1 KiMvita is an informal and subjective expression of their appreciation of such facts as those above. The formulae almost certain­ ly would have to be modified in more radical ways to en­ compass tongues more remote from the KiSiu/KiMvita com­ plex, such as those which have stem-initial  in all cases. And the degree of modification is, again, a formal measure of degree of relationship. Fourthly, and finally, the approach used here allows the linguist to express, relatively simply, what might appear to be irregularities as regularities and/or to point out similarities which evade the linguistic con­ sciousness when presented in other terms. KiSiu might be thought to be 'odd1, for example, in that the plurals of items such as mwamba, mwaka, moshi, are nyamba, nyaka, nyoshi. But, as the formulae show, this is the most sim-, pie case of singular-plural relationship in the data. In articulatory terms, [mw-] is to [p-] as [u] is to Γι], and the relationship, whilst analogous to [mu-/mT]/is less complex than it. Descriptions in other phonological terms can rarely put such relationships as neatly as they are put here, even when they are captured at all; and it is relationships such as these that can be of particular

KISIU NASAL CLASSES

317

importance for historical and comparative Bantu studies, more particularly as being possible prefigurations of future developments. There are other aspects in which KiSiu is of inter­ est to the Bantuist and the phonologist: and, indeed, the totality of Bantu languages, still largely unexplored, offers a great challenge to phonological theory, as so many phenomena in them run counter to alleged funiver­ sale '. Firthian phonology, in its turn, offers a subtle and sensitive methodology for dealing with the interpre­ tation of the matter of spoken language. I have brought them together in this paper each as a foil to the other in the hope that linguists might be encouraged to give due and serious consideration to the claims of Firthian phonology as an approach to the conduct of phonological analysis.

* financial support for this work was provided by the So­ cial Science Research Council (Project No. HR 4411) to whom my thanks are due.

THE PHONOLOGY OF STATE IN KABYLE BERBER* Michael Kenstowicz, Yousef Bader, and Rachid Benkeddache (University of Illinois)

As in other Berber languages, Kabyle nouns surface in two shapes which, following traditional practice, we shall refer to as the "bound" and the "free" state. The basic generalization regulating the distribution of the two shapes is that a noun appears in the bound state when it is immediately preceded by the head of its phrase; elsewhere it appears in the free state. Kabyle is a VSO language. Hence, a noun is in the bound state when in subject position, but in the free state in direct object position. Similarly, a noun is bound by a preceding prep­ osition (with a few exceptions) and by a preceding numer­ al in a quantifier phrase. In addition to the direct ob­ ject position nouns are free in subject position when preposed to the verb as well as in citation. There are some rather complex phonological alterna­ tions associated with the bound state in Kabyle. Our goal in this paper is to discover and formulate the rules un­ derlying these alternations, relying on the approach to syllable structure sketched in work by Halle and Vergnaud (1980).. On this view phonological segments are repre­ sented in terms of at least two separate tiers: the seg­ mental tier and a CV tier representing the gross syllabic profile of the segmental string. The CV tier serves as the terminal point for a number of different patterns of phonological organization. One of the most important of these is organization into Onsets and Rimes, the two im­ mediate constituents of the syllable. As we shall see, some of the phonological rules associated with state in Kabyle are more easily expressed if formulated in terms of Onset and Rime.

320

AFRICAN LINGUISTICS

In Kabyle the majority of native nouns as well as nativized loanwords are characterized by an initial vowel a- in the singular and i- in the plural. The precise mor­ phological status of these initial vowels is a thorny problem of Berber linguistics that we shall eschew here. We assume that they can be treated as separable morpho­ logical elements of albeit obscure semiotic exponence. The data in (1) show that these initial vowels partici­ pate in the characteristic alternations associated with state in Kabyle. The first entry in each cell shows the noun preposed to the verb yet tes 'slept' and hence in the free state, while the second entry shows the noun in its underlying postverbal position, whence it surfaces in the bound state. Many Kabyle nouns inflect overtly for number and gender, θ marks feminine nouns as a prefix (and for many as a suffix as well, but only in the singular). Plu­ ral is marked by the suffix - for masculine and -in for feminine nouns.

(1)

For feminine nouns we see that the bound state consists in the replacement of initial-vowel a- or i- with schwa or zero. The choice between schwa and zero is rule-gov­ erned. Schwa appears when the stem begins with two con­ sonants, while zero appears when the stem begins with a CV sequence. Other examples appear in (2). 2 (2).

free

state

θam uθ θafunasθ

bound

state

country' 'mare

KABYLE BERBER

321

'Berber woman1 •hat' 'mule1 fern. 'girl' 'cat' fern. 'donkey' fern. 'street' 'room' The appearance of the schwa in the feminine bound-state forms is a reflex of the general Kabyle scheme for im­ posing a syllable structure on a string of consonants. Ignoring certain complications, 3 the Kabyle syllablebuilding rules can be phrased as follows: i) assign a vowel to a Rime; ii) assign a prevocalic consonant to an Onset; iii) assign a string of one or more unassigned consonants to alternating Rime and Onset positions start­ ing from the right end of the string; iv) if a Rime node fails to dominate a vowel slot, adjoin a V slot as a left daughter to the Rime; v) spell out a V slot that is un­ linked to a segment in the phonemic core as schwa. With the help of these syllable-building rules we can thus say that for feminine nouns the bound state mor­ phology consists simply of the deletion of the initial vowel. The form receives the following derivation. (31

The syllable building rules organizing a string of successive consonants regulate the distribution of schwa in forms such as 'work !' imper. versus 'I worked'. (4).

322

AFRICAN LINGUISTICS

The masculine bound state forms of (1) are a bit more complex. The gross generalization is that we find the a- initial vowel replaced by  or and the i- ini­ tial vowel by i - or . Once again the choice between the two realizations of the bound state alternants is controlled by the initial CV structure of the following stem. The and - alternants appear if the stem be­ gins with a consonant cluster, while  and i appear if the stem begins with CV. (5)

free

state

bound

state

' hand' 1 fingers' 'leg1 1 chair1 'automobile' 'man1 'cat1 'birds' 'pen' 'cats' The schwa vowel of the we- and - alternants can be treated as a reflex of the syllabification process men­ tioned earlier, suggesting that the underlying represen­ tation for the préfixai element in the bound state mas­ culine nouns is a glide w or y. The vocalic alternants u and i can be derived with the help of the following vocalization rule. (6)

This rule is to be understood as one that changes   slot into a V slot in the rime; as a result the semivow­ els y/w will automatically be reinterpreted as the cor­ responding vowels i/u. The lefthand bracket expresses the

KABYLE BERBER

323

restriction on the rule that it only applies when the un­ derlying y/w is initial in the rime; i.e. not when it is part of an ay or aw diphthong. The vocalization rule is ordered between the syllable-building rule that organizes a string of  slots into alternating rime and onsets and rules (iv, v) that insert Schwa. The bound state forms urumi, irumyen, and W9mli£ are derived as follows. (7)

inapplic.

inappl.

vocalization (6)

epenthesis

In addition to the bound state prefixes, exactly the same vowel/semivowel alternation is exhibited by the 3 sing.mase, verbal prefix, which we take to be underlying /y-/: 'he gave', 'he slept', 'he ran', but i-ruh 'he left'. In Kabyle the schwa vowel is subject to a great deal of allophonic variation. In the environment of the glides y and w it is raised so that and . sound like yi and wu and in normal rates of speech are often difficult to distinguish from i and u respectively. However, it is clear that there is a phonological contrast. One piece of support for this contention derives from the behavior of the bound state forms in sandhi. Kabyle has a general sandhi rule truncating the first of two successive vow­ els: 'he found a lamb', 'he will kill the dog', and , 'he will kill  the lamb' are normally pronounced as 1, , and , respectively. The ualternant of the bound state prefix triggers truncation of a preceding vowel, but the we- alternant never does;

324

AFRICAN LINGUISTICS

thus, while _, _ 'the European gave* can be pro­ nounced as [yefka warg'az] 'the man gave' can never be pronounced with the final a of absent: Another sandhi rule sensitive to this contrast be­ tween and _u_- is one that permits a word-final conso­ nant to occupy the onset position when the following word begins with a vowel. This rule is at the root of the schwa-zero alternation exhibited by the verb in (8).. (8). west i lu 'the pen broke'

yarz ukursi 'the chair broke'

In 'the chair broke' the final consonant of the verb may form an onset with the following rime-initial word, but in 'the pen broke' this is not possible since the word begins with an onset position. Although we have not had the opportunity to study the matter in detail, the syllabification across word boundaries in Kabyle can apparently lead to wholesale reorganization of the syllable structure of the prece­ ding word, as evidenced by the following partial para­ digm. (9) 'he fixed the house' The latter example is particularly interesting in that it appears as if the vocalization rule has reapplied. These data, if indicative of phrasal syllabification in general, appear to force the conclusion that either syl­ lable structure is assigned at the level of the phrase, or that it is assigned at the word level but completely remade at the level of the phrase. The question obvious­ ly deserves further study. To briefly summarize, we have seen that the bound state in Kabyle differs from the free state in the dele­ tion of the initial vowel and, for masculine nouns, prefixation of w- to words with initial vowel a- and prefixation of y- to words with initial vowel i-. The alter­ nations associated with the bound state follow from in­ dependently needed rules in the grammar.

KABYLE BERBER

325

So far we have given examples of nouns with initial vowel a- in the singular and i- in the plural. This is by far the most common pattern. A brief sample of other types of initial vowels that delete in the bound state are given below. state bound state (10) free 'lamb'

' lambs' ' fiance' 'fiances' ' sister' 'daughter' ' women ' These forms permit us to sharpen our analysis of the glide that appears in the bound state, showing that the choice between w- versus y^ depends on the quality of the initial vowel rather than on the number (singular versus plural) of the noun stem. Thus, in the forms for 1 fiancé1 we find a bound state form y- in both the singu­ lar and the plural, reflecting the fact that the initial vowel is i- in both the singular and plural free state forms. The words for 'lamb' exhibit an initial vowel pat­ tern that is the opposite of the usual pattern. We have initial vowel i- in the singular and a- in the plural free state forms. In the corresponding bound state forms we find the glides y- in the singular (i.e. içerri from by vocalization) and w- in the plural. The words for 'sister' and 'daughter' represent a small class of semantically feminine nouns which lack the prefix Θ-. In their bound state forms not only is the initial vowel missing, but they also show a glide like the masculine nouns, the glide being w- if the initial vowel is u-. The initial glides of the bound state are thus apparently disjunctive with the feminine prefix θ-. Finally, there are quite a few noun stems whose ini­ tial vowel does not delete in the bound state form. When such nouns lack the θ- prefix, we find that the glide ap­ pearing in the bound state form is w- if the following vowel is back (i.e. a or u) and y- if the following vowel is front (i.e. i ) . (11),

free agur as i f

state

bound wagur was if

state 'moon' 'river'

326

AFRICAN LINGUISTICS

'small river' ' mouth' 'day' ' days' 'bush' 'widower' 'widowers' 'widow' 'widows' Nondeletion of the initial vowel appears to be regular for nouns which begin with a geminate consonant (a notable ex­ ception being axxam, wexxam 'house'). It is tempting to try to analyze the nondeleting initial vowels as part of the root and restrict deletion to préfixai vowels (so that arumi would have the analysis /a+rumi/, while 'moon' etc· of (11) would be /agur/). This approach finds support from the fact that when the initial vowel a- is clearly derived morphologically, it does delete in the bound state form. However, in cases where there is no clue as to the deriva­ tional source of a noun, there is no noncircular way of deciding whether to treat a given initial vowel as a pre­ fix or as a radical. We shall assume that those (relative­ ly small number of) nouns whose initial vowels fail to de­ lete in the bound state are to be marked as exceptions in the lexicon. (The generalization that pregeminate initial vowels normally fail to delete can be recorded by a lexi­ cal redundancy rule). The data from (10) show that the rule inserting the glide in the bound form must apply before the deletion of the initial vowel, since the latter determines the quality of the former. We can state these two rules of glide in­ sertion and initial vowel deletion as follows. (12) (where X gov­ erns N)

KABYLE BERBER

(13) V -> 0 / [ X [N [+segment]

327

(where X governs N)

An interesting feature of our analysis is that al­ though bound state içerri and free state 'lamb1 have the same phonetic representation and the same under­ lying representation (namely they do not have the same derivation. In the bound state form rule (12) assigns an initial glide yielding ; rule (13) deletes the initial vowel to give whence by vocalization. In Kabyle nouns cannot assign case to a complement NP. Instead, a "preposition"  is inserted, which in turn assigns case. There is a certain amount of phonology as­ sociated with this preposition. It surfaces most clearly before names, which systematically lack an initial vowel. (14) ' grave of Ahmed' 'grave of Xira' ' grave of Fatima' 'grave of Baya' 'grave of father' The bilabial fricative ν obligatorily hardens to a stop after the nasal and η assimilates the point of articula­ tion of jb (both underlying and that derived from ν by hardening). Before sonorant consonants, η may optionally undergo complete assimilation to the sonorant. (15) 'grave of Muhend' 'grave of Nura' 'grave of Lusif' 'grave of Rashid' 'grave of Yusef' 'grave of Wa91i' We assume that complete assimilation is not the result of a feature changing rule but rather is the reflex of a rule deleting η from the phonemic core, leaving an empty  slot, which is then linked with the following consonant.

328

AFRICAN LINGUISTICS

(16)

(17)

(18)

The forms with initial glides η wa 9 I i/bbwa9 Ι ϊ, *n bwa 91 i) reflect the application of a general rule of Kabyle morphophonemics that hardens a consonant into a stop by adding the feature [-contin] to certain consonantal segments when they dominate two con­ secutive  slots. (19)

When the complement of the η preposition is a noun with an initial vowel, we find that the η is obligatorily absent. Instead the stops JbJbw and gg' appear if the bound state form of the noun begins with a glide w- or y-, re­ spectively. If the bound state form of the noun begins with a vowel the preposition η is missing, the noun sim­ ply surfacing in the bound state form. (20) 'grave

of

N

bound

state

free

state

'man' 'men1 'sister1 'European' 'Europeans1 'widower' These rather complex alternations follow from the rules we have already established. All we need assume is that deletion of the η preposition (16) is ordered before

KABYLE BERBER

329

glide vocalization (6) and that the rule linking the va­ cated  slot with a following consonant (17) is ordered after the vocalization rule. The forms 'az and u rum i are derived as follows· Underlying [n ~ ^ ~~ and [n [arumy]] are converted to [n [wrg'az]] and [n [wrumy]] by insertion of glides (12) and the de­ letion of initial vowels (13) in the bound state. The derivation continues as follows (see 21). First (i) , the rule (16) deleting  before a sonorant consonant applies, leaving an empty  slot. Then (ii) vocalization (6) ap­ plies , converting a glide in rime position to the corre­ sponding high vowel. The η-deletion rule (16) must pre­ cede glide vocalization, since η is not deleted before an underlying vowel (cf. ahme 'grave of Ahmad'). Next (iii) the rule (17) linking an empty  slot with a following consonant applies. It will fail to apply in the derivation of [urumi] because the underlying glide has been vocalized to a vowel. In step (iv) the rule supply­ ing a vowel to a rime lacking one applies. Finally, in (v), the rule hardeninq a continuant spanning two  slots (19). yields (21)

It is to be noted that the unlinked  slot that arises in the derivation of urumi from the deletion of the preposition η must be deleted, since the regular Kabyle

330

AFRICAN LINGUISTICS

sandhi rule truncating a vowel before a vowel is not in­ hibited: urumi is pronounced urumi. (22)

Our Berber Kabyle example thus differs from the behavior of h-aspiré in French, where an empty  slot blocks liai­ son (Clements and Keyser 1980). Finally the data in (23) show examples of the femi­ nine noun θamŝiŝθ 'cat' fem, as complement to the η prep­ osition in possessive phrases with θa sotta 'tail' and 'foot1 as heads. (23)

'foot of the cat' These data evidence the general Kabyle rule hardening θ to a stop after nasals as well as the optional deletion of the preposition n, leaving an empty  slot that is filled by the following consonant of the phonemic tier. In addition, in the phrase 'foot of cat1 the vowelless rime initiating the word 'cat' has been supplied with a schwa. In 'tail of cat', on the other hand, the rime node dominating the  slot of the preposition η has merged with the rime node dominating the final vowel of the preceding word. (24)

In (25) we have examples of 'woman' as com­ plement to the preposition  in possessive phrases headed by 'grave' and amiŝiŝ 'cat'.

KABYLE BERBER

331

(25)

'cat of the woman' In these examples we see (somewhat surprisingly) that  can separate a cluster of consonants that have undergone the local rule hardening θ after a nasal. But if the op­ tional rule deleing  applies giving rise to a geminate, we see that the geminate is not separated by schwa (the expected outcome). If it is assumed that the inserted vowel occupies the phonemic tier, then we have a formal explanation for the strong tendency not to separate gem­ inate consonants. In order for geminates to be separated, we would have to assume the existence of an intermediate rule splitting apart the segmental phoneme to, in effect, provide a space for the schwa, as suggested below. (26)

The formal complexity of this operation would rank the resultant grammar high in markedness, accounting for the strong tendency for geminate consonants to remain unseparated. Note that if it is assumed that the schwa occu­ pies a different tier from the phonemic tier, then an al­ ternative explanation must be provided for why geminates resist separation. In order to account for the fact that a schwa is in­ serted before the geminate in we assume that Kabyle has the following rule permitting a geminate consonant to be assigned to the rime. This rule will of course be ordered before the rule adjoining a V slot to the left of a vowelless rime. (27).

Given rule (27) the phrase 'cat of the woman' receives either of the derivations in (28), depending on whether the n-deletion rule has applied.

332

AFRICAN LINGUISTICS

(28)

An analysis of the schwa and geminate consonants similar to the one developed here was proposed by Halle (1978) in unpublished work based on secondary sources for other dialects of Berber. Our analysis differs from Hal­ le's primarily in the use it makes of subsequent develop­ ments that have occurred in the theory of syllable struc­ ture - in particular the role of the CV tier and explicit statement of the rules in terms of onset and rime cate­ gories. In sum, we have shown that the complex phonological alternations associated with nominal state in Kabyle fol­ low from the relatively simple rules of glide insertion and vowel deletion plus rules that are independently needed in the grammar. We have seen that many of these rules are most easily expressed when formulated in a framework that incorporates the notions of a CV tier and the syllable categories onset and rime. We hope to return to other aspects of Kabyle phonology in subsequent stud­ ies (see Bader to appear).

KABYLE BERBER

333

FOOTNOTES

* In this paper we describe the speech of the third author, a na­ tive of Ain el Hammam (ex-Michelet), a town located about 50 km southeast of Tizi Ouzzou, Algeria. This study was supported by funds from the University of Illinois Research Board, 1982. 1. In our transcriptions the raised schwa appearing between words is no different in phonetic or phonological value from other schwas in Kabyle. We have transcribed it thus in order to better indicate where the word boundaries fall. 2. θ is realized as a stop t after nasals and 2 by a general rule of Kabyle phonology. 3. Syllables ending in two consonants are possible when the second consonant is a dental fricative Θ, s, ζ, , We assume that these consonants can be treated as an "appendix" to the syllable (cf. Halle and Vergnaud 1980 for discussion of this notion). Also, rules (iv) and (v) of the text may not insert a schwa within a geminate cluster, the inserted schwa instead appearing before the geminate consonant. We return to this point at the end of the paper. 4. The nouns asif (b.s. wasif) 'river' and the diminutive θasifG (b.s. GasifG) have regular plurals (b.s. isaffen) and 9isaffin (b.s. θsaffin), respectively. An interesting question for future research will be the distribution of exceptions to the bound state phonology (i.e. rules 12 and 13) vis a vis the morphological relationship among the words. We would hazard the guess that if stems differ solely in terms of "external" inflection, such as asif and θ-asif-θ do, then they will agree in value with respect to the rule features for the bound state; while if the stems are related via "internal" inflection, as singular asi f but plural isaff- are, then there will be no strong tendency for them to share the same value for the rule features. If this distinction between internal versus external inflection turns out to be valid it could be ex­ plained on the assumption that words related by external inflection are built from the same lexical entry by morpholexical rules, while words related by internal inflection have separate lexical entries.

334

AFRICAN L I N G U I S T I C S

REFERENCES

Bader, Y. To appear Topics in Kabyle Berber Phonology and Morphology. (Uni­ versity of Illinois Ph.D. thesis). Clements, G.N. and S.J. Keyser 1981 A Three-tiered theory of the syllable. Occasional Paper # 19, Center for Cognitive Science, M.I.T., Cambridge, Mass. Halle, M. 1978 Metrical structure in phonology. Unpublished ms. Halle, M. and J.-R. Vergnaud 1980 Three dimensional phonology. Journal of Linguistic Research 1,1.83-105.

ON THE CORRESPONDENCE BETWEEN LINGUISTIC TONE AND MUSICAL MELODY William R. Leben (Stanford University)

1.

INTRODUCTION.

In this paper I offer a reinterpretation of a study by Richards 1972. Richards looked at a song in Hausa, a tone language, and attempted to determine the degree to which the musical melody resembles the linguistic tones of the lyrics. His result was unambiguous but unimpres­ sive; the correspondence was statistically significant but admitted many deviations. This contrasts with what we know from other studies of the correspondence between tunes and tones, where in general the result is that the correspon­ dence is either extremely close or practically nonexistent. My reinterpretation of Richards' data will suggest that for this particular Hausa song, a comparison should be made not between the musical melody and the phonemic tone levels but rather between the musical melody and a normal­ ized intonational realization of the lyrics. I will try to conclude something from this about the nature of phonetic representations.

2.

A STUDY OF WAKAR INDEFENDAN, "THE SONG OF INDEPENDENCE".

The song Richards examined was performed and recorded in 1964, at the height of its popularity in northern Nige­ ria. Richards compared the pitch displacements in the song with those that would be predicted in the spoken language on the basis of the phonemic tone differences in Hausa.

AFRICAN LINGUISTICS

336

Hausa has two underlying tones, H and L, which can com­ bine on some heavy syllables to give a sequence HL, or falling. Falling tones have rather low frequency (in the present song, there are only 22 falls, as compared to 488 highs and 337 lows), and so to simplify his study Richards counted falling tones as low. He looked at the first 52 lines, about one quarter of the song in question, and the first two lines appear in (1): (1)

The way that Richards tabulates the tune-tone cor­ respondence is the following. To normalize the shift from one musical note to the next with the concurrent move from the phonemic tone level of one syllable to the next, Richards translates musical note sequences and phonemic tone sequences into a different notation, as follows: for a transition between a relatively H pitch and a relative­ ly L one, Richards writes F; for a transition between a relatively L pitch and a relatively H one, Richards writes R, for rising; for level movements from syllable to syllable, Richards writes L. Note that in Richards' system L stands for level, not for Low tone. A sample transcription using these new symbols is given in (2). This is a transcription of the first line in (1). Where there is a sequence of notes on a single syllable of lyr­ ics, I count only the first note. I assume that Richards does the same, but he does not mention this explicitly. One other simplification Richards makes is to count pho­ nemic falling tones as if they were low tones; this sim­ plification is arbitrary, but it does not throw off the count severely, since falling tones make up only 22 of the 847 tones Richards counts in the complete text. (2)

Sample transcription: Tune: R L L L L L L F F Tone: R L L L F R L L F Richards now counts the number of tune-to-tone cor­ respondences in the actual song and compares these to the number of correspondences that would occur by chance,

LINGUISTIC TONE AND MUSICAL THEORY

337

that is, if the poet did not aim for a fit between the linguistic tones and the musical ones. Richards' results are statistically significant but woefully unimpressive. The attested cases of perfect matches between tune and tone are a mere 53.4% of the total. Although the percent­ age of matches to be expected by chance is lower, it is only somewhat lower, in the range of 35% to 40%, de­ pending on how one calculates. The difference is large enough to be statistically significant but nonetheless not really very large at all. Only slightly more than half of the musical notes actually coincide with the pho­ nemic ones. This is a puzzling finding, for at least two reasons. First, we wonder why a composer would seek to make the melody conform to the phonemic tone pattern to such a limited extent. Poets are entitled to nod, but not 46.6% of the time. Second, in other cases in the litera­ ture, it seems that composers do not behave like this at all. Either they pay rather close attention to the tones or they ignore them altogether. In the next section I summarize the literature I have encountered on this topic in African languages.

3.

SURVEY OF OTHER SOURCES.

Greenberg 1949 reports on one Hausa song, a popular song two lines of which he transcribes. He notes that there is no correspondence between the musical and lin­ guistic tones. Within each foot (of two or three sylla­ bles!, the syllables are all on the same musical note, regardless of the linguistic tones of the lyrics. Howev­ er, Greenberg refers to Maloney, who has found correspon­ dences between linguistic tones and music in a number of West African languages, including Hausa. Unfortunately, I have not yet been able to consult this source. Jones 1959 and Schneider 1943-44, 1961 concur that there is a dramatic correspondence between tune and text in Ewe, a Kwa language of Togo and Ghana. Jones, in fact, notes an agreement of slightly more than 90% between the linguistic tones and the musical pitches of the one song for which he calculates the correspondences. This is all the more impressive because Ewe has three phonemic tones, High, Mid, and Low. Jones and Schneider disagree, some­ what bitterly, over the reasons behind the small minority of cases in which tune and tone don't match. Schneider presents a number of hypotheses, the most interesting of

AFRICAN LINGUISTICS

338

which has to do with tone being overridden by meter: H may be L in song on a metrically important syllable, and L may be H in song on a metrical point of secondary value. Finally, Simmons 1980 takes the close correspondence between tune and tone for granted in Efik, a Benue-Congo language of Nigeria, presumably because the correspondence is so exact, though Simmons offers no actual tabulations. And Simmons raises the further question of whether it is the tones of the lyrics that set the pattern for the mel­ ody or the tones of the melody that narrow down the choice of words for the lyrics to tonally matching ones. Thus, with the exception of Richards, all of the sources quoted find either a clear tune-tone correspon­ dence or a clear lack of it. In the remainder of this pa­ per I will argue that the Hausa song which Richards ana­ lyzed does in fact exhibit a remarkable correspondence between musical melody and linguistic pitch, but that this correspondence is obscured by paying attention to the phonemic tone levels H and L. Rather, these tone lev­ els must be adjusted by rules which are in fact related to the rules of Hausa intonation. The basic point can be seen by observing that most of the violations of the tune -tone correspondence in the following text, taken from four more lines of Wakar indefendan, are of one of two kinds. First, the third syllable of the proper noun Nigeria is marked as L in the phonemic transcription, but it is on the same note as the preceding high-toned syl­ lable in the song. This is simply a consistent transcrip­ tion error on Richards' part: phonemically, the third syllable of Nigeria is actually high, not low. The second violation involves the fact that in practically every se­ quence HLH, L in the song is realized on the same level as preceding H, rather than being relatively low. The de­ viant L tones are encircled. (3!

LINGUISTIC TONE AND MUSICAL THEORY

339

Thus, what appears to be the case is that the poet pat­ terned the melody not after the phonemic tone levels per se, but after these levels modified by a rule that makes a L equivalent in fundamental frequency to a preceding H. Interestingly, as I learned from a study reported in Leben 1983, the L in HLH is quite variable in where it is plotted with respect to the preceding and following tones; in actual Hausa intonation curves it is not unheard of for L to be not only as high as the following H but even as high as the preceding H. This being the case, it is possible that the poet is modeling the song on the basis of a version of Hausa intonation that is indeed stylized, but that is clearly adopted from one of the options taken in normal speech. However, before we can consider this topic any further, we must first take a closer look at Hausa intonation in general.

4.

INTONATION IN HAUSA.

The standard view of Hausa intonation is that, to a first approximation, a low tone lowers the register for a following high tone. As a result, the H on the right in HLH is, on the average, lower in fundamental frequency than the rightmost H in HLH. This is true, but in addi­ tion, many investigators have either assumed or claimed that the reason behind this downdrifting pattern is that in a sequence of tones, a low tone is plotted, say, two steps below a preceding H, while a H is plotted only one step above a preceding L. This point of view emerges from work focused on Hausa in particular, such as Hodge and Hause 1944, Meyers 1976, Miller and Tench 1980, 1981, as well as from general works concerned with the formaliza­ tion of the downdrift phenomenon, such as Schachter and Fromkin 1968, Clements 1979, and Hyman 1979. However, there are signs that the lowering effect of a L on a fol­ lowing H is independent of where that L is plotted on the actual intonational curve. The results are reported in greater detail in Leben 1983, but the basic phenomenon is the following. In fluent speech, L tones are extremely variable in their realization. Quite often they are no lower than the following H, as Meyers 1976 has observed. And in quite a few cases L is actually higher than the following H. Thus, the following sort of picture, traced from an oscillo­ graphic print-out using the pitch extraction system of

340

AFRICAN LINGUISTICS

the Phonology Laboratory at the university of California, Berkeley, is not atypical: (41

'he went on to say...1

Note that the encircled syllables are approximately as high as the preceding high-toned syllables, even though they themselves are phonemically low (as can be judged by eliciting them in isolation, or in very slow speech). As I propose in Leben 1983, I believe that this phe­ nomenon shows that the mapping of low tones onto intona­ tion contours ought to be regarded as separate from the effect that low tones have in many languages of lowering the register for the following H. That is, first we map high tones onto a register, and subsequently, we add the values for L, as depicted in (5): (5) a. H plotting H L H L H Ρ max Ρ min b. L plotting (arrow shows possible range of variation) H L H L H

In light of this, one interpretation of the song is that the composer is aware of the fact that L in HLH is more variable in its realization than other tones. What he is consistently doing in the present song, then, is assigning L to one of the extremes that it occupies in real discourse· On this view, the mapping of the encir­ cled L's is not exceptional at all, and the tune-tone

LINGUISTIC TONE AND MUSICAL THEORY

341

correspondence for the song is near 100%.

5.

DISCUSSION.

This seems to tell us something about this piece of Hausa music, namely that its tune reflects Hausa intona­ tion and not just Hausa tone. But a reasonable question to ask is whether this has any wider significance. I think that it possibly does. Bloomfield held the view that between physical phonetics and the level he called phonemics there was no linguistically significant level. He reasoned that only phonemics provided a non-arbitrary criterion for factoring certain properties out of speech signal and leaving others in. The criterion was based on whether the property in question was distinctive or not, in the sense of distinctive that Bloomfield defined. Hence, if you tried to represent a level somewhere be­ tween the phonemic level and the level of physical pho­ netics, there was no apparent non-arbitrary criterion that u could use in order to sort out the properties that you left in the phonetic transcription from the properties that you chose to omit from the transcription. Nowadays, we have more or less abandoned Bloomfield's assumptions on this matter, and we speak of a so-called level of "systematic phonetic" representation. Unfortunate­ ly, the properties of this level are not defined in any language-independent way, and so really Bloomfield's point still ought to hold. I think that the musical struc­ ture of this Hausa song may help to provide a non-arbi­ trary basis for the notion "systematic phonetic represen­ tation", for at the very least this shows that the compos­ er was not relying solely on phonemic tones. In particu­ lar the composer was making use of certain characteris­ tics of Hausa intonation contours. But the song melody is obviously not identical with the actual physical funda­ mental frequency patterns of spoken utterances, nor can the melody be directly derived from these fundamental frequency patterns. Thus, the composer seems to be draw­ ing representations from a level that is more superficial than the phonemic level yet more abstract than the physi­ cal signal. I suggest that the study of songs like these may help to show which characteristics of an intonation contour are salient to Hausa speakers, so that they can draw on these even though they are not contrastive in any way.

342

AFRICAN LINGUISTICS

We can gather from work on metrics by Kiparsky and others that poets are able to draw on rather abstract properties of phonological representations in putting to­ gether their poems. I think it might be possible to in­ terpret the present study in somewhat the same vein, only here we have an art form that taps linguistic representa­ tions which are abstract, but which are nonetheless closer to the surface than the level of phonemic or un­ derlying representation.

REFERENCES

Clements, G.N. 1979 "The description of terraced-level tone languages." Language 55.536-558. Greenberg, J.H. 1949 "Hausa verse prosody." Journal of the American Oriental ciety 69.125-135. Hodge, C. and H. Hause 1944 "Hausa tone." Journal Hyman, L.M. 1979

of the American Oriental

"A reanalysis of tonal downstep." Journal

guages and Linguistics

So­

Society 64.2.

of African

Lan­

1.9-29.

Jones, A.M. 1959 Studies in African Music. London: Oxford University Press. Meyers, L.F. 1976 Aspects of Hausa tone. UCLA Working Papers in Phonetics 32. Leben, W.R. 1983 "Intonation in a tone language." Paper presented at Four­ teenth Conference on African Linguistics, University of Wisconsin, Madison. Miller, J. and J. Tench 1980

"Aspects of Hausa intonation, 1." Journal

tional 1980

Phonetics

Association

"Aspects of Hausa intonation, 2." Journal

al Phonetics

Association

of

the

of

the

Interna­

10.45-63.

12.78-93.

Internation­

343

LINGUISTIC TONE AND MUSICAL THEORY Richards, Ρ. 1972 "A quantitative analysis of the relationship between lan­ guage tone and melody in a Hausa song." African Language Studies XIII, 137-161. Schachter, P. and V. Fromkin 1968 A Phonology of Akan. UCLA Working Papers in Phonetics 9.

Schneider, M. 1943-44 "Phonetische under metrische Korrelationen bei gesproch­ enen und gesungenen Ewe-Texten." Archiv fur Vergleichende Phonetik 7,1/2.1-6. 1961 "Tone and tune in West African music." Ethnomusicology 5,3. 204-215. Simmons, D.C. 1980 Extralinguistic

uses of tonality

sity of Alabama Press.

in Efik

folklore.

Univer­

AN ACCENTUAL APPROACH TO TONE IN KIMATUUMBI David Odden (Yale University)

There has been recent debate on the nature of accent in phonological systems, especially as it relates to tone in an autosegmental model. Accent might, for example, be a feature, analogous to [+ back]. 1 It may or may not be related to stress, and it"may or may not be an alternative to tone (so that tone and accent are mutually exclusive). The view set forth in Goldsmith (1981) is that accent is a formal entity, an asterisk, which links the tonal tier and the syllable tier. I will argue here that Goldsmith's general view is correct, by showing that in the grammar of Kimatuumbi, tone and accent have different formal char­ acteristics. I show that tone and accent must coexist in certain derivations, so that some rules refer to both tone and accent. This then makes it impossible to claim that accent is changed into tone, or that accent is simply a way of saying that tone is restricted in its underlying occurrence. I first discuss the tonal pattern of lexical nouns and adjectives, where I argue that an accentual analysis is necessary to explain what constitutes a possible lexi­ cal tone pattern for nouns and adjectives. I also argue that accent is a lexical property of the syllable in Kima­ tuumbi, and not of the vowel. But a prerequisite to showing that accent is necessary in the language is showing that it is possible. That an accentual analysis is even possi­ ble is not obvious on the surface. There are a number of complications which obscure the underlying accentual sys­ tem, and rules are developed to account for these compli­ cations. Some of these rules are conditioned by accents, and some of these rules are conditioned by tones. For ex­ ample, Tone Copy is only conditioned by accents, whereas Retraction applies to tones, whether derived from morpho-

AFRICAN LINGUISTICS

346

logically conditioned tone-assignment rules or from lexi­ cal accent. I then discuss the tonal pattern of verbs and dever­ bal nouns and adjectives, and argue that accent does not characterise verbal tone patterns. The tonal rules for verbs refer to the number of vowels in a stem and not to syllables, and the tonal rule Retraction also applies to H tones in verbs. The accentual nominal system and the tonal verbal system are then contrasted. The possible surface tone pat­ terns for nouns are significantly different from the pos­ sible surface patterns for verbs. Tone Copy does not ap­ ply to tonally derived verbs, but does apply to accentual­ ly marked nouns. The accentual analysis of nouns also simplifies a tone sandhi rule, which would otherwise be complicated with adhoc reference to morphological fea­ tures. In the last section, I compare the Kimatuumbi accen­ tual and tonal system with other accentual systems, with an eye to evaluating the purported connection between ac­ cent and rules such as the leftward movement of H tone, or the existence of a rule lowering H tone after H tone. I also discuss the change within Bantu from being a tonal language to being an accentual language.

1.

PREVIOUS WORKS ON ACCENT

Earlier works on accentual languages, such as Japan­ ese, have started with the viewpoint that a language is accentual, rather than tonal, if it suffices to mention a single place in the word where pitch rises or falls. In Tokyo Japanese, one need only learn which is the last H toned syllable in a word, and McCawley (1978) analyses the language as a pitch accent language. 2 (

1

)

(

m

a

k

u

r

a

ga)

"pillow"

(kokoro ga) * (atama ga)

"heart"

(sakana ga)

"fish"

"head"

A similar analysis is suggested for Luganda (McCawley (1968). Thus, the first often-cited characteristic of a pitch-accent language is that there is only one accent in the word. 3 In an accent language, morphemes of  tonal units can be grouped into no more than n+1 accentual clas-

TONE IN KIMATUUMBI

347

ses. But in a typical two-tone languaqe, the number of tonal possibilities comes closer to 2n. For example, in Shona, a Bantu language with L and H tones, bisyllabic noun stems may bear any of the four logically possible tonal patterns LL, LH, HL and HH (Odden (1981a)). (2)

bhu'kù

"book"

bàdzá

"hoe"

bángá

"knife"

sàdzà

"porridge"

Shona would thus be classified as a tone language, since no analysis could be maintained whereby a single tonal unit in a word is identified as bearing the mark of prom­ inence. In general, a tonal language (with two tones) will allow 2 n tone contrasts in words of η prosodic units, and Shona comes reasonably close to fully exploiting this possibility. To see the merit of the requirement on accentual sys­ tems that only one accent is allowed per morpheme, consid­ er the potential problem of allowing multiple accents in a words. Assume a language such as that in (3). (3)

nènè

"fly"

ηéne

"big"

nénè

"cat"

nèné

"three"

In an accentual account of this language, the basic tone melody is H, and a stem may bear multiple accents in this language. Accordingly, the surface forms in (3) would have the underlying accentual forms in (4) . (4)

nene * * nene

"fly"

nene * nene

"cat"

"big" "three"

In addition, a purely tonal analysis would be possible, as in (5). (5)

L L nene

"fly"

H L nene H H nene

"cat" "big"

348

AFRICAN LINGUISTICS

L H "three" nene It is clear that, in this language, accent merely serves as a diacritic mark for H tone; every H tone comes from an accent, and every accent is realised as a H tone. There is, then, no principled reason to select an accentual analysis over a tonal analysis, if there is no principle to the effect that morphemes bear only a single H tone. The approach that is generally taken is that, in or­ der to avoid arbitrariness in selecting an analysis for a language, accentual analyses will not be freely available, and will require certain extra motivation. It is there­ fore reasonable to begin with the assumption that mor­ phemes bear only a single accent, in order to cut down on arbitrariness in deciding if a system is tonal or accen­ tual. But I will later argue that an absolute version of the "one-per-morpheme" restriction in untenable, when I show that two mono-morphemic stems, "Muslim rosary" and "circle" in Kimatuumbi have two underlying accents. Earlier studies completely separated "tone" and "ac­ cent", so that "tone" is apparently not ever assigned to vowels in McCawley's analysis of Japanese. In his dis­ cussion of accent in Somali, Hyman (1981) conjectures that tone is never assigned to vowels, and he devises rules to directly assign pitch-integers to accent-marked vowels. In Japanese, the separation of tone and accent may reflect the phonetic fact that accent merely "marks the drop in pitch". However, there is no convincing evidence that tone languages are so radically different, and in fact we may simply see the accent-versus-tone question as being reduceable to a discussion of the conditions needed to analyse a language in terms of stars versus lexical tonal autosegments, and what the relationship between a star and a tone is. I assume that all accent languages have tones at some stage of the derivation, and that these tones are what is subject to physical interpretation. It is with an eye to such questions that I shall discuss Kimatuumbi tonology here. Another question which may be raised in studies of accent languages is what the domain of accent is. McCawley (1978) claims that the domain of accent in Japanese is the syllable, based on the fact that long syllables may not have a contrast between vv and vv. I shall show here that Kimatuumbi is like Tokyo Japanese in this respect. In fact, I argue that representing accent at the level of the syllable rather than the mora not only explains on prin­ cipled grounds the restrictions on accentual contrasts,

TONE IN KIMATUUMBI

349

but also yields a simpler grammar. On the other hand, in other languages, accent may be contrastively assigned to either the first or second mora of a long syllable. For example, in Beja, Somali, Lithuanian, Ancien Greek,+and Osaka Japanese, there is a contrast between vv and vv. As McCawley (1978) has argued, languages may be char­ acterised partially as pitch-accent languages, and par­ tially as tone languages. For example, for the description of nominal tone patterns in Luganda, McCawley suggests a purely accentual approach. However, he observes that an accentual approach for verbs is extremely complex and adhoc, and Goldsmith (1976) in fact proposes a tonal anal­ ysis for verbal tone in Luganda.1* McCawley thus suggests that a language may start out as accentual, and will be­ come tonal in the process of a derivation. I shall show here that, although tone and accent can be distinguished in Kimatuumbi in many ways, there are many cases where tone and accent are indistinguishable, and many cases where accent must be subsumed under tone. The approach taken in Goldsmith (1981), which shall be followed here, is that accent is a formal entity, sym­ bolised with a star (*), which may appear in a phonolog­ ical representation. This entity serves as the link be­ tween the segmental and tonal tiers in his analysis of Tonga tone and accent. He argues that Tonga has the basic tonal melody HL, and that, for example, the noun 'buji "smoke" bears accent on the final vowel, viz. i-bu-si. One copy of the basic tonal melody is inserted for each accent, and the accented tone is associated with the ac­ cented vowel. One of the predictions of Goldsmith's approach is that accent should in principle be allowed to remain in a derivation at least past the staqe where accents are deleted or inserted and the tonal melody is mapped onto the word. I shall argue here that this prediction is verified in Kimatuumbi, since some rules which apply af­ ter the mapping of the accented H tone onto accented vow­ els must refer to the presence or absence of accent. It will in fact be shown that some rules refer simultaneous­ ly to accent and tone. And, finally, it will be shown that Kimatuumbi has a strong claim to being analysed accentually, since a pure­ ly tonal analysis of the language is unworkable. This re­ sult is significant since Hyman and Byarushengo (1982) and Hyman (1982b) have suggested that "accent" verses "tone" is a specious distinction, and "accent is best reanalysed as restricted tone. But a restricted tonal anal­ ysis is incorrect for Kimatuumbi.

350

2.

2.1 .

AFRICAN L I N G U I S T I C S

ACCENT IN NOUNS

Ά first

approximation

As suggested earlier, if we find in a language that the number of surface tonal patterns in a word of η vow­ els is n+i or less, we may begin with the assumption that the language is an accentual language. Kimatuumbi nouns 'are, by this criterion, characterisable by accent, since there are fewer than n+1 tonal patterns in nouns of η vowels. In fact, it turns out that there are only n+1 to­ nal patterns possible for nouns of η syllables (rather than vowels). However, it is not immediately ovbious on the surface that Kimatuumbi words are characterised by underlying accent, especially with non-contrastive dis­ tribution of accent within the syllable, and not under­ lying tone. The accentual system of Kimatuumbi is ob­ scured on the surface by numerous tonal processes which add and move tones and delete accents. For example, there is a rule retracting a H tone to a preceding long vowel. Or, there is a rule assigning a H tone to the last vowel of penultimately accented nouns. Moreover, verbs and de­ rived nouns and adjectives follow different tonal prin­ ciples from lexical nouns and adjectives obscuring the overall tonal system of the language. The forms in (6) are representative of the surface phonetic contrasts pos­ sible in Kimatuumbi words. "rice" "bread" "cut up" "lipring" "drip" "mosquitoes" "one who blunts" "rabbit" "you should blunt" "marble" "watermelon" "hartebeests"

351

TONE IN KIMATUUMBI

Kimatmimbi is thus only an accent language at the under­ lying level, but is not radically different from other "pure" tone languages at the surface level. The task of giving an accentual analysis of Kimatuumbi therefore requires the unraveling of a number of rules which map accentually marked lexical entries onto tonal surface forms. Below I give forms typifying the different rules applicable to lexical nouns and adjectives. Table 1 Form

Gloss

Underlying form

Rules

mv[r i ingo

circle

ITAR (19)

 i ρf i ηg i I i

piece of wood

mv i  i ¡ ngo *  i ρ i i ng i 1 i  i w i  i 1 yo ngo1ogo

HTA

k¡w í  i 1 y o

cover

ngo1ogo

beer brewing place

ma koongon f

large hartbeeste

mboopo

machete

¡t é1elko

cooking pot

nt óopé

handle

m ó

bag

ηi eéniemá

mosquitoes

n c h í m b i 1 f i|í drip m a n g a | f ba

circumcisers

ngoóngono

chicken drumsticks

ñama

meat

nd fm i

tongues

* ma koongon[ * mboopo

*  it e 1eeko

* ntoope *   ηjeenjema * nch i m b i 1 i i 1 i * ma * nga 1 \ba

ITAR Η Τ Α (13)

ITAR, PPHTA (23) ITAR, R (28) ITAR, R, PPHTA ITAR, TC(46) ITAR, TC ITAR, TC, R ITAR, TC, PPHTA L S (51)

ngoongono

D (55)

ñama *

HTA, NR(61)

nd i m i

TC, D, NR

I will therefore begin by discussing the cases which are as free as possible from the complicating factors, and then go on to explain the more complicated cases. In nouns which are not complicated by other rules, a noun may bear only one H tone, and that H tone may appear on any syllab­ le of the word. Any of the other logically possible combi­ nations having two H tones are not found.5 "canoe" (71 "recruit'

352

AFRICAN LINGUISTICS

"wood" "net" "rabbit" "circle" "gravel" "Christian" "bird" "cat"

Since there are a limited number of tonal patterns pos­ sible, we have a prima facia case for an accentual analsysis of Kimatuumbi nouns. The noun  ¡-t ú kut u ku bears an accent on the first stem syllable, the noun n-k¡lfs¡ty bears an accent on the second stem syllable, and the noun changaláwe bears accent on the penultimate stem syllable (that accent actually originates on the final syllable and is retracted by a rule). I assume that the basic tone melody is H, and that there is a tone-mapping rule which maps the melodic H tone onto the vowel of any accented syllable. This rule will be refined later to handle assignment of accent to long syllables. Finally, I assume that any vowel which does not bear a H tone is assigned a L tone.6 (8)

Initial

Tone-Accent

Association

Rule

(ITAR)

The nouns in (7) are assigned their accent by application of the Initial Tone-Accent rule to the accented forms in (9). (9) Thus, at this point, the relationship between accent and surface tone is very simple - each accented syllable has a H tone.

TONE IN KIMATUUMBI

2.2. H tone

assignment

and unaccented

353

nouns

The first complication of the accentual picture to be discussed here is the problem of unaccented nouns. Ev­ ery word in Kimatuumbi has least one H tone when the word stands at the end of an utterance. However, there are nu­ merous nouns, all having a H tone on the second vowel of the word, which have no H tone when non-final in an utter­ ance. As argued in Kisseberth and Odden (1980), these nouns have no H tone or accent in their underlying repre­ sentations, and undergo a sandhi rule assigning a H tone to the second vowel of the word in isolation. "cover" "my cover" "rice" "I put rice in a bucket" "fog" "that fog"

(10)

These stems may be contrasted with the stems in (7), which retain their stem H tone in phrase-medial position in (11). ( 1 1 ) " m y canoe" "my wood" "big bird" The noun ki-wikilyo is therefore unaccented, and the H tone found in isolation derives by application of H Tone Assignment. On the basis of the forms above, the tone assignment rule could have been stated to assign H tone to the steminitial vowel of unaccented nouns. However, when nouns in Class 9 are considered, it becomes clear that the correct generalisation for tone assignment is that tone is as­ signed to the second vowel of the word, and that the noun class prefixes k¡- et al. count as the first vowel. Since the Class 9 prefix has no vowel, as shown below, the iso­ lation H tone is assigned to the second vowel of the stem, which is also the second vowel of the word. (

1

2

)

"

b

e

d

b

u

g s " "many bedbugs" "place for initiates"

354

AFRICAN LINGUISTICS

"my place for initiates" "steinbuck" "I didn't see the stein­ buck" "hunting net" "he didn't take the hunting net" The following rule is therefore proposed to account for the phrase-final assignment of H tone unaccented nouns. (13) [-accent] pause H Tone Assignment (HTA) It should also be noted that there is no evidence in the isolation form to show that the noun "fog" is unaccented, but "bird" is accented. The evi­ dence for this accentual distinction is found only in the nonfinal forms of these nouns.7 The evidence where H Tone Assignment applies to nouns with a nonsyllabic prefix gives a different picture of the formulation of the H Tone Assignment rule. Still, one might modify the approach which assigns accent (rather than tone) to these nouns by simply modifying the rule, so that in isolation a rule places a star on the second vowel of an unaccented noun. This approach is incorrect for two reasons. First, whereas assigned H tone is as­ signed on the basis of vowels, accent is never assigned to the individual vowel, but is rather assigned to the syllable. Second, as I shall show later, for shorter noun stems, i.e. CVCV and C W C V stems, initial-accented nouns have a distinct phonetic tone pattern from unaccented nouns in isolation (which receive a H tone by HTA), so that initial-accented ntoope is phonetically realised as whereas underlying unaccented is realised, by application of H tone Assignment, as A similar argument is made by Haraguchi (1977) for the Osaka dialect of Japanese, where he shows that final-accented nouns have a different phonetic pattern from unaccented nouns; a fi­ nal accent is realised as a falling tone, whereas an un­ accented noun takes a level H tone on the final vowel. In contrast, unaccented nouns in Tonga are assigned an accent on the prefix vowel. With these notes, we can see that, looking at the nonfinal forms, the following nouns bear no underlying ac­ cent, and take their H tone by application of H Tone As­ signment.

TONE I N KIMATUUMBI

(14)

n-kóta

ngoóngowe nguIú mbopó ngulu yílyu 2.3.

Accented

long

355

"cane" 11 lots of cane" "watermelon" "he didn't take water­ melon" "marble" "that isn't marble" "large" "that large knife"

vowels

I have discussed only nouns with short accented syl­ lables, or else having only unaccented syllables. Long syllables may be accented as well, so in the following ex­ amples, the initial long vowel of the stem bears a falling tone, which I claim is the realization of an accented long syllable. ( 1 5 ) " p i e c e of wood" "steinbuck" "chicken drumstick" "bedbug" Since these nouns retain the H tone on the initial vowel in all positions, these nouns must be accented. (16) "I didn't see a bed­ bug" "the chicken drumstick is rotten" matándolo ganaanchímá "many steinbucks" Some nouns with initial H tone on a falling vowel must be analysed as being unaccented, and take their H tone in isolation by application of HTA (13) . Thus, there is in isolation a surface neutralisation of the contrast between accented and unaccented stems,since in the following examples, there is no way to distinguish accented and un­ accented stems in isolation. (17) ma-túunguI a "pi. large green tumeric plants" "hen house" "woods"

356

AFRICAN LINGUISTICS

"watermelon" unaccented nouns have the same superficial pattern as in­ itial-accented vowels in isolation, due to the application of H Tone Assignment. The underlying distinction is brought out below, since in phrase-medial position, un­ accented nouns bear no H tones, whereas accented nouns retain their accent in phrase-medial position. (18) ma-túngulu galyú "those large green tume­ ric plants" "I don't see the hen house" "those woods" "that watermelon" We can thus be certa in that the nouns in (17) have an underlying initial accent. The assignment of H tone to the first vowel of a long vowel in (17) leads us to revise the H tone mapping rule, so that the rule maps a H tone on­ to the leftmost V node of the accented syllable. (19)

Initial

Tone-Accent Association

Rule (revised)

This then covers the essential points concerning the in­ itial assignment of tone to accented syllables, and fur­ thermore distinguishes between accented and unaccented stems. I have assumed that the lexical accent is assigned to the syllable node, not the individual vowels of the syllable, because it provides an explanation for the fact that there is no syllable-internal contrast in position of accent. So, while position of accent is unrestricted at the syllable level, in that any syllable of the stem may bear the accent, within long svllables there can be no contrast in Kimatuumbi between vv and vv. Representing accent at the syllabic leve.1 explains this restriction, since the unattested vv -vv contrast would be impossible in principle to maintain. Syllable representation of ac­ cent allows for a simpler grammar elsewhere. 2.4.

Post

Préfixai

H Tone Assignment

The next complication is a rule assigning a H tone to the initial vowel of a noun stem when the stem is preceded

TONE IN KIMATUUMBI

357

by a noun class prefix vowel and is followed by a L tone. By hypothesis, stems in Kimatuumbi bear no more than a single accent, and since each accent is associated with a single H tone, multiple H tones should not occur in stems. Nevertheless, numerous stems do in fact have more than one H tone. ( 2 0 ) " c o o k i n g pot" "large canoes" "drivers" "stubbornness" "cassava" "bishops" "black pulse plants" However, an independently motivated rule is neces­ sary in the grammar which assigns a H tone to the first stem vowel after a noun-class prefix vowel, providing the stem is accented, and the first two vowels of the stem are L toned. Nouns which have a form in Class 9-10 (which does not have a vocalic class prefix) will take an ini­ tial H tone after a noun class prefix such as ma-, pro­ viding that the H tone is not assigned to a vowel which stands immediately before a H toned vowel. The following alternations illustrate application of the Post-Prefixal H Tone rule (discussed in Kisseberth and Odden (1980)). ( 2 1 ) " n a i l " "large nails" "hartbeest" "large hartbeests" "canoe" "large canoes" "boss" "bosses" "gravel" "large gravel pieces" "hospital" "hospitals" If the second stem vowel bears a H tone, the steminitial vowel cannot take a H tone by Post Prefixal H Tone Assignment.

AFRICAN LINGUISTICS

358

(22)

mbaká "cat" ma-paká "large cats" *ma-pá ká mboópo "machete" ma-mboópo "large machetes" *ma-mbóópo This alternation therefore motivates the following rule. (23)

Post-Prefix

High Tone Assignment (PPHTA)

The rule in (23) has been stated in such a way that it is sensitive to the presence of an accent (or H tone). There are two reasons for this restriction. First, the initial vowel must be separated from the H tone by at least one vowel (to account for failure of PPHTA to apply in ma-). Secondly, direct reference must be made to the presence of an accent in the stem, to keep (23) from applying to unaccented stems. The need for this restric­ tion is shown by the fact that unaccented stems in (24) do not have a surface H tone on the initial vowel by ap­ plication of PPHTA. ( 2 4 ) " m y hunks of gravel" "those aren't large nails" "my cover" "that isn't watermelon"

PPHTA does apply in phrase medial position, as shown by the example ma-chángaIáwe gaángu. This rule must therefore be prevented from applying to any noun which is unaccented, to prevent the ungrammatical form * ¡-w Í  i|yo chaángu, where the noun k¡wíkilyo is unaccented. This restriction is built into the rule by requiring the presence of an accent in the stem. This rule could also be formulated to require that a H tone be present somewhere in the stem; there is no compelling evidence which shows that this rule must refer to accent versus tone. There is also a categorial restriction on the rule, namely that the rule only applies to nouns. An adjective with two initial L tones does not take a H tone after a prefix, as shown by the form mweepeés¡ "light (Class 3 ) " ,

TONE IN KIMATUUMBI

359

from (and ultimately from underlying . The contrast mweepeés¡ "light" versus "nails" shows that PPHTA must be prevented from applying to adjec­ tives. With the Post-Prefixal High Tone rule, we can account for a much larger range of forms. We are now able to ac­ count for the fact that the nouns in (21) have two phonet­ ic H tones, without abandoning the claim that Kimatuumbi nouns generally only allow a single accent. The existence of this rule is of considerable importance is maintaining the accentual analysis of Kimatuumbi nouns, since a con­ siderable portion of the nouns of the language have two H tones, the first deriving from application of PPHTA, and the second representing the underlying accent. In (25) the first H tone will be derived from PPHTA, even though there is no direct evidence as there is in (21) that the initial stem vowel is basically L toned. ( 2 5 ) " c a s s a v a root" "plastered storage hut" "box of matches" "traveler" As long as the first two stem vowels are L toned, PPHTA will apply no matter how far away the accent is placed. 2.5.

Retraction

The next complication of the accentual system of Kimatuumbi is a rule retracting a H tone from a vowel left­ ward to the preceding long vowel. In (26), a penultimate long vowel bears a rising tone; this tone originates on the final, accented syllable. That these nouns are ac­ cented is shown both by the fact that the H tone could not be assigned by HTA (13) in a form such as where the H tone is on the third vowel of the word, and by the utterance-medial forms which retain the H tone seen in isolation. ( 2 6 ) " m a c h e t e " "that isn't a machete" "nail" "I don't want a nail" "loaf" "cooking pot"

AFRICAN LINGUISTICS

360

"road" "paper" These forms superficially pose a problem for the hypothe­ sis that accent can only contrast at the syllable level, but not at the vowel level, since, due to the general principles assigning tone to accented syllables, a rising tone should never occur on an accented syllable. However, assuming that these nouns bear a final accent (which takes the H tone) and undergo a later tone-retraction rule, then the anomaly of the position of the H tone is explained.9 Apart from maintaining the basic hypothesis that Kimatuumbi does not have a contrast in position of accent below the level of the syllable, the assumption that there is a tone-retraction rule in the language explains other facts. There is a gap in the tonal distribution of the language; there are no nouns with a H tone preceded by a long low-toned vowel (except initial-accented C W C V Class 9 nouns like ndooñá, which are explainable by rule order­ ing). Thus, nouns with the canonical shapes in (27) are impossible. (27) This then leads us to the following Retraction rule. Retraction

(28)

The phonetic forms mboopo and kitéleéko are derived in (29) (29) m b o o p o u n d e r l y i n g H * m

b

o

o

p

o

I

T

A

R

m b o o p o R e t r a c t i o n One piece of evidence that the nouns PPHTA in (26) bear final accent is that, in case the long vowels of the stem are shortened, the Η tone appears phonetically on the fi­ nal vowel. As argued in Odden (1981c) and Kisseberth and

TONE IN KIMATUUMBI

361

Odden (1980), there is a rule shortening the long vowels of the head of a phrase. Shortening thus prevents Retrac­ tion from applying, so the accented final vowel retains its Η tone. ( 3 0 ) " l a r g e machete" "many nails" "my loaf" "my cooking pot" This can be explained by ordering Shortening before Re­ traction, so that the underlying long vowel is bled away. * (31) mboopo ... underlying Η * mboopo ... ITAR mboρο ... Shortening An alternative but incorrect analysis would be that Retraction does not apply in certain phrasal domains. That analysis is untenable, since the phrasal conditions on Shortening are relatively complex (the rule must be stated to apply to the head of a phrase), and the Retrac­ tion rule applies exactly in those phrasal environments where Shortening does not. Moreover, when a noun fails to undergo Shortening (by being an exception to Shortening), then it will also undergo Retraction in all environments. ( 3 2 ) " w i n e " "my wine" "box" "my box" In the discussion of verbs, it will be shown that Re­ traction applies to Η tones assigned in verbs as well. Since verbs do not involve accent, Retraction cannot be formulated as an accent retraction rule. Of course, treating Retraction as an accentual retraction rather than a tonal retraction would be difficult even without con­ sidering verbs, since the Η tone must end up on the second vowel of the penultimate vowel. Initial-accented C W C V nouns are phonetically distinct from final-accented C W C V nouns, which undergo Retraction. The former set of nouns surface with the pattern CvVCv", due to the application of a Tone Copy rule, whereas the latter set of nouns surface with the pattern CVVCV. Thus, the following incorrect der-

AFRICAN LINGUISTICS

362

ivation would occur under the assumption that Retraction retracts accent to the preceding long syllable. * (33) n-kaate underlying * n-kaate Retraction (of accent) H n-kaate

Initial Tone-Accent As­ sociation

H H *n-kaate Tone Copy The Retraction rule then explains why certain accented nouns have a rising tone, without admitting syllable-in­ ternal contrasts in the position of accent. Moreover, the Retraction rule explains why the final accented syllable does not bear a phonetic H tone in the form mboopo. 2.6.

Tone

copy

Bringing additional nouns with long vowels into con­ sideration, there are more nouns with two phonetic H tones, which cannot be accounted for by PPHTA (23) . For example, the following C W C V nouns bear H tone on the first and final vowel. ( 3 4 ) " b e d " /ki-ndaanda/ * "handle" /mu-toope/ * "baobab" /mu-loonji/ Similarly, the following (CV)CVCVVCV nouns bear H tone on the second and last vowels. * ( 3 5 ) " a m a r a n t h a s " /ndebeele/ * "punishment" /asaaby/ * "pineapple" /|i-kynaai/ "belch" /k¡-kuikwi ikwi/ "drip" /mu-chimbi Mili/ These nouns bear accent on their penultimate long vowel, and are subject to a rule assigning a H tone to the final vowel. In all of the examples of accented long vowels discussed previously, the accented syllable was at least two syllables from the end of the word. But whenever the accent falls on a long penultimate syllable, the final syllable also takes a H tone, by application of the fol­ lowing Tone Copy rule.

TONE IN KIMATUUMBI

(36)

363

Tone Copy

ITAR (19) assigns a H tone to the first mora of the accented syllable in ntoope, and Tone Copy then assigns a H tone to the final syllable.10 ( 3 7 ) u n d e r l y i n g ITAR Tone-Copy Accented and unaccented C W C V nouns differ in their isolation forms as. well as in their utterance-medial forms; accented C W C V nouns undergo Tone Copy and have the isolation pattern CvVCv", whereas unaccented nouns, which undergo H tone assignment, have the surface pattern CvVCV. (38) ntoope "handle" η tope waángu "my handle" mpyynga "rice" mpynga waángu "my rice" Tone Copy gives us evidence that Retraction (28) must retract a H tone, not an accent. If Retraction were formu­ lated as accent retraction, we would incorrectly predict that final accented mboopo, which does not undergo Tone Copy, should have the same tone pattern as initial-ac­ cented ntoope, which undergoes Tone Copy. Tone Copy accounts for two facts simultaneously. First, the rule accounts for the fact that no accented nouns have the surface shape (CV)CvVCV (thus correctly ex­ cluding a logically possible but unattested noun such as *k{nfin¡ / kjníni chaángu, from /k¡-niini/), and also ac­ counts for virtually all of the remaining stems in the language with two phonetic H tones. Thus, if a stem has two H tones, either the first one is stem initial and de­ rives by applying PPHTA (23) or else the second one is fi­ nal, and derives by applying Tone Copy. Numerous other logically conceivable combinations of two H tones are cor­ rectly excluded, since they do not fall into the range of forms derivable by either PPHTA or Tone Copy. For example, nouns with tone patterns such as those in (39) are unat­ tested in Kimatuumbi. (39) *chángaIáwe

AFRICAN LINGUISTICS

364

There is at least one noun in the language which has three surface H tones; yet, given the rules PPHTA and Tone Copy, the noun is analysable as bearing a single un­ derlying accent. The noun _ may be assumed to have the underlying form " ; Tone Copy (36) will apply to derive the final H tone, and PPHTA (23) will apply to derive the initial H tone. 2.7.

Tone

Copy

and

Retraction

More data show that Tone Copy applies to penultimately accented nouns with a short penultima, leading us to a formal revision of the Tone Copy rule. Below are given examples of (CV)CVVCVCV nouns with penultimate accent. The penultimate syllable of these stems is almost the only stem syllable in the word with no H tone. The H tone as­ signed to the accented syllable is retracted to the pre­ ceding long vowel by Retraction, and a H tone is assigned to the final syllable by Tone Copy. ( 4 0 ) " m o s q u i t o e s " "spleen" "bean" "cowpea" "chicken predator" Tone Copy must therefore be modified so that a short accented penultimate syllable may, under some conditions, trigger the rule (it will be shown that there are certain apparent problems with allowing Tone Copy to apply after every accented penultimate syllable). In these examples, both Tone Copy and Retraction apply, as illustrated below. ( 4 1 ) u n d e r l y i n g ITAR Retraction Tone Copy

TONE IN KIMATUUMBE

365

As predicted, when the long vowel is shortened Re­ traction cannot apply, so the penultimate syllable retains its H tone. The lack of Tone Copy will be fully explained later when I show that Tone Copy only applies in utter­ ance-final position. ( 4 2 ) " m y mosquitoes" "long bean" "that chicken predator" There is yet another environment where Tone Copy ap­ plies, namely when both the penultimate and antepenulti­ mate syllables are long. The noun from under­ lying kipiikiiti, bears a penultimate accent, yielding a H tone, which is retracted to the preceding long vowel. In addition, the accent must trigger Tone Copy. ( 4 3 ) u n d e r l y i n g ITAR Retraction Tone Copy Given the interaction of Tone Copy and Retraction dis­ cussed here, we can also explain why there are no nouns of the shape *njeenjema in the language; any noun of the syllable shape CWCVCV with penultimate accent will un­ dergo Tone Copy (so the final vowel will bear a H tone), and it will also undergo Retraction, so the penultimate H tone will be retracted to the preceding long vowel. So far, I have only discussed examples of Tone Copy in penultimately accented nouns with a penultimate or an­ tepenultimate long vowel. However, this rule also applies to stems with no long vowels, in CVCV nouns with initial accent. Thus, m-pyko is underlying m-pyko. ( 4 4 ) " b a g " /my-pyko/ "voices" /ma-lobe/ "ground rat" Since Tone Copy also applies to these forms, modification of Tone Copy is necessary, so that the accented penulti­ mate short vowel in CVCV stems will condition Tone Copy.

AFRICAN LINGUISTICS

366 2,8.

Leftward

Shift

Up to this point, it would appear that the simplest version of Tone Copy merely states that any noun with an accented penultimate syllable takes a final H tone. This statement of Tone Copy is ultimately correct, but there are surface forms which appear to militate against this statement of Tone Copy, On the surface, Tone Copy does not apply to penultimate H-toned nouns of the shape CVCVCV or CVCVCVCV (nouns of the shape CWCVCVCV do not exist in the language). ( 4 5 ) " r e c r u i t " "gazelles" "musket" "gravel" There does not seem to be any concise way to formu­ late Tone Copy so that penultimate-accented stems will in general trigger Tone Copy, and still prevent the rule from applying to nouns of the stem shape (CV)CVCVCV. The solution which I therefore propose is that Tone Copy is maximally simple, as in (46). Apparent failure of Tone Copy in (45) will then be explained by some means other than a direct restriction on Tone Copy. (46)

Tone Copy

(revised).

Regarding the decision to represent accent at the sylla­ ble level in Kimatuumbi, the formulation of Tone Copy provides support for that decision. Besides explaining why thgre can be no contrasts within long syllables be­ tween vv and vv, accent at the syllable level simplifies the statement of Tone Copy. If accent were represented at the vowel level, Tone Copy would have to be complicated with irrelevant Co's and parenthetic vowels. (47) Having selected a final formulation of Tone Copy, we may return to the problem of nouns like changaláwe. Noun stems with the shape (CV)CVCVCV and accent in noninitial position are rare; the majority of the CVCVCV stems are initial accented or unaccented. The distribution of tones in (CV).CVCVCV nouns is not free. Although there are numer­ ous Class 9 nouns with the tonal shape CVCVCv", there are none of the shape CVCvOV (except unaccented stems, which are clearly identifiable by their contextual behavior). Thus, the stem ngalawá does not contrast accentually with

TONE IN KIMATUUMBE

367

any hypothetical stem such as *mbitíti. And, although there are many nouns in Class 9 with the shape CVCVCv"CV, there is only one with the shape CVCVCVCv" (bit¡yumá "Binti Juma cassava"), which may be treated synchronically as a compound, viz. [ [ bjt¡ ] [ yumá ] ]. Thus, in Class 9 nouns, final andnpenultimate nt for (CV)CVCVCV stems are nearly in complementary distribution (with only one exception). Moreover, in all other noun classes, CVCV"CV stems are quite common, while there are only four CVCVCv" stems with final H tone (the initial tone comes from PPHTA) . ( 4 8 ) " t o o t h l e s s person" "cunningness" "angel" "lazy person" In the case of the stem there is independent evi­ dence that the initial H tone derives from PPHTA, since the form "lazy person" is also possible, with no prefix, and no initial H tone. There may well be related forms for the stems et.al. which have no syllabic prefix, to provide evidence that all of these stems ulti­ mately derive from Class 9 stems. So final and penultimate accent for (CV)CVCVCV stems are essentially in complementary distribution, with a small handful of exceptions. The next question is how this complementarity is expressed, and how it relates to the problem of Tone Copy and the penultimate-accented stems in (44). The solution to this problem begins to take further shape when we motivate a rule retracting tone in CVCVCV noun stems with final accent, which have allomorphs in Class 9. Whenever a noun of the shape CVCVCV has a final accent, and does not have a syllabic noun class prefix in its basic allomorph, the addition of a syllabic prefix (in the plural or in the diminutive, for example) triggers an optional Leftward Shift rule, which shifts the H tone of the final accented vowel to the penultimate vowel. ( 4 9 ) " c i r c u m c i s e r for females" "that circumciser" p1. "those circumcisers" "canoe" "little canoe" "antilope"

368

AFRICAN LINGUISTICS

maputúka pl, maputúka ganaanchfmá "many antilopes" This retraction is optional, and if it does not apply, then PPHTA (23) will obligatorily apply to the initial L toned vowel. ( 5 0 ) " c i r c u m c i z e r s " "little canoes" "antilopes" Since PPHTA is elsewhere obligatorily, never option­ al, it would seem reasonable to assume that Leftward Shift is optional, and that if it applies, PPHTA is blocked. I assume that the Leftward Shift rule is conditioned by the configuration given in (51), so that reference must be made to the presence of a vowel in the nounclass prefix. (51) V + V V (optional) Leftward Shift Given Leftward Shift (51), we may now consider how an extension of this rule can account for anomalous penulti­ mate accent in (45), where Tone Copy does not apply. I as­ sume that changa |áwe has an underlying final accent, and undergoes a modified version of Leftward Shift, such as (52). (52) V

If Leftward Shift is partially optional and partially obligatory, nouns such as changa Iáwe and maputúka will be given an underlying final accent; Tone Copy cannot apply to nouns with final accent, and subsequent application of Leftward Shift will account for the surface violation of Tone Copy. The anomaly is then not that Tone Copy fails to apply, but rather that there are no nouns of the form (CV)CVCVCV in the language. The formal statement of the optional-obligatory con­ ditions for applying the Leftward Shift rule given in (52) is unfortunately insufficient. The spirit behind that mod­ ification is that the rule will apply in case its struc­ tural description will always be met (due to whatever mor­ phological or phonological circumstances may prevail for the word). Thus, changa |áwe must undergo the rule because there are always three preceding vowels, but application

TONE IN KIMATUUMBI

369

of Leftward Shift to maputuká is variable because some of the allomorphs of this stem do not satisfy the structural description of the rule. However, the formulation offered in (52) does not accurately express this generalisation; in stems like "recruit", the morphological structure is the same as in ma+putuka "antilopes" (viz. CV+CVCVCV), but Leftward Shift is not optional in the former case, because the structural description of the rule is always met for this stem, since all of its allo­ morphs require a syllabic prefix. So, while a statement with angled brackets may account for the Class 9 contrast changa Iáwe - ma + putuka, that statement cannot account for the contrast between ma+putuká and 2.9

Deaccentuation

Nouns with allomorphs in Class 9 have other peculiar accentual properties, besides those associated with Left­ ward Shift (52). There is a Deaccentuation rule which ap­ plies to the initial syllable of a word. In (53) , the in­ itial H tone of the noun derives from an accent, as shown by the fact that the H tone is retained in phrase medial position. ( 5 3 ) " c h i c k e n drumstick" "my chicken drumstick" "piece of marble" "I didn't see the piece of marble" "waterbucks" "many waterbucks" Looking at the corresponding nouns in Class 9, we see that the isolation H tone appears to have been shifted off of the initial vowel, and appears on the second vowel of the stem. This H tone cannot be an accent, since in phrase medial position, the H tone is lacking. ( 5 4 ) " c h i c k e n drumsticks" "he didn't take the chicken drumsticks" "pieces of marble" "my pieces of marble" "waterbuck" "I didn't kill a waterbuck"

370

AFRICAN LINGUISTICS

The crucial difference between the Class 9 prefix and all other prefixes in the language is that the former is nonsyllabic, so that the stem-initial syllable of a Class 9 noun is also word initial. This motivates an ac­ cent-reduction rule which deletes an accent from a wordinitial syllable. (55) The isolation H tone on the second vowel in (54) is ex­ plained by HTA (13); that rule applies to the second vow­ el of the word (which, for many of the examples previous­ ly discussed was also the first vowel of the stem). The Deaccentuation rule also applies after the Class 5 prefix li-, which is anomalous in "many areas of the phonology, behaving as though it were separated from its stem by a strong boundary. Nouns with an initial ac­ cent after the Class 6 prefix in (56) are deaccented after the class 5 prefix | ¡ - . The surface manifestations of deaccentuation are that the isolation H tone of the Class 5 allomorph is missing in phrase medial position (whereas the H tone of the Class 6 accented form is retained in phrase medial position) and the stem tone pattern shifts from CvCv' (by Tone Copy) to CvCV (by H Tone Assignment, without Tone Copy).11 (

s " "word" "holes" "hole" "big bags" "big bag" Since the Class 5 prefix is phonologically more distant from the stem than the other noun class prefixes, I as­ sume that it is separated from the stem by a stronger boundary, and application of Deaccentuation in Class 5 follows from the presence of that boundary. 2.10

5

6

Nominal

)

"

w

o

r

d

Retraction

Accented forms such as undergo De­ accentuation in their Class 9 allomorphs, giving an under­ lying form such as unaccented ndimi, which should then ap­ pear on the surface as (by application of H in isolation). Instead, the attested form is . The form would have been expected on the grounds that HTA

TONE IN KIMATUUMBI

(13) assigns H tone this case the final nouns which undergo tone. ( 5 7 ) "

371

to the second vowel of the word, in vowel. Instead, initial-accented CVCV Deaccentuation show stem-initial H t

o

n

g

u e " pl. "small bog" "bog" "pole" "poles" "cover" pl. In fact, while one would expect there to be nouns with alternations such as (where the final H tone derives from ΗΤΑ), no such nouns are found in the language. In their place are nouns like those in (57). The anomalous initial (assigned) H tone of CVCV stems in these nouns shows the need for an additional rule which retracts the expected final tone in CVCV nouns. In fact, both initial-accented and unaccented stems undergo this rule. When an unaccented CVCV stem has allopmorphs both with and without a syllabic prefix, the assigned H tone appears on the stem initial vowel (which, in the Class 9 form is anomalously the word-initial vowel). ( 5 8 ) " m e a t " "my meat" "sorghum hut" "my sorghum hut" "little sorghum hut" "my little sorghum hut" "ribs" "my ribs" "rib" "my rib" Again, even in the case of underlyingly unaccented nouns, there are no nouns in the language having the alternation even though Deaccentuation and H Tone As­ signment could in principle work in concert to bring about such an alternation.

AFRICAN LINGUISTICS

372

One might begin with the assumption that HTA (13) needs revision, to assign the isolation H tone to the stem initial vowel of a CVCV unaccented noun. But there is no way to revise the rule directly so that the assigned H tone isn't assigned to the final vowel of CVCV stems. Moreover, there is evidence that HTA should not be com­ plicated with syllable conditions and that the nonexis­ tence of CVCv unaccented surface forms is due to a Retrac­ tion rule. If an initials-accented CVCV adjective is pre­ ceded by the Class 9 prefix, Deaccentuation applies as ex­ pected, and the isolation H tone appears on the final stem vowel. "large

(Cl.7)"

"small

(Cl.7)"

"that isn't small" "large" "small" "that isn't large" "that isn't small"

So while the expected alternation is not found in nouns, it is found in adjectives. For that mat­ ter , there is no adjectival alternation as there is for nouns. This shows that there is a rule which retracts a H tone from final position to the preceding vowel in CVCV nouns, whether that H tone be the H tone from Tone Copy 12 or the H tone from HTA. This retraction must be limited to noun stems, since as observed above, adjective stems with the syllable shape CVCV in Class 9 undergo Deaccentuation and H Tone Assignment, but retain their phonetic H tone on the final stem vowel. Finally, there is one additional restriction on this rule, a restriction of considerable importance. The Nomi­ nal Retraction rule may only apply to a final unaccented H tone. The accented final H tones of Class 9 CVCV nouns seen below do not undergo Nominal Retraction. (60) mbaká "cat" mbaká

yaángu

"my cat"

nda ká

"dirt"

ndaká yaángu

"my dirt"

The grammar must distinguish between phonetic Η tones which derive from application of the ITAR to accented syl-

373

TONE IN KIMATUUMBI

lables and H tones which do not derive from accents, but from application of H Tone Assignment or Tone Copy. Unless the retraction rule is to be stated globally to refer to the origin of the final H tone in nouns, there must be a distinction between unaccented H tone and accented H tone when Nominal Retraction applies· This necessary distinc­ tion between accented and unaccented H tones 13 then shows that it would be incorrect to maintain that accent becomes tone; rather, the two must coexist. And, the fact that Nominal Retraction applies only to unaccented H tones also shows that an accentual analysis must be given to Kimatuumbi, since the language distinguishes not just be­ tween H and L, or accent and no accent, but between L, ac­ cented H, and unaccented H (the fourth possibility, ac­ cented L, cannot be argued for here, although it may ex­ ist as the output of Deaccentuation). This all suggest the following Nominal Retraction rule. (61)

Nominal

Retraction

[-] The derivation of the form ndími from underlying ndimi is given in (62). ( 6 2 ) u n d e r l y i n g Tone Copy Deaccentuation Nominal Retraction  * oThe fact that Kimatuumbi distinguishes between H, H and L is the strongest possible evidence for an accentual analysis of the language. Without this evidence, it would be possible (albeit undesirable) to analyse the language tonally, with only one H tone allowed lexically per word. But accent and H tone are not mutually interchangeable in Kimatuumbi, even in lexical stems, as they are in lan­ guages such as Haya. So, not only can Kimatuumbi be ac­ centual, it must be.

374

2.11

AFRICAN L I N G U I S T I C S

CVVCV

stems

The discussion of the tone pattern of nouns which have forms in Class 9 is incomplete since nothing has been said of such stems with the syllable shape C W C V . It turns out that there are four tonal classes for nouns of this shape, whereas the accentual analysis would only predict three possible contrasts (assuming, as has been necessary up to now, that only one accent appears in a morpheme and that the domain of accent in Kimatuumbi is the syllable, rather than the vowel). I shall therefore concern myself here with the task of analysing C W C V stems with Class 9 allomorphs. The final accented class poses no problem; the accent appears on the final vowel of the stem, and is retracted unless the long vowel is shortened. / * (63) mboopo "machete" /mboopo/ m baá i

"grasshopper"

/mbaa[/

ndoondwa "stars" /ndoondwa/ When the long vowel is shortened the final vowel has a H tone, but otherwise, the penultimate vowel has a rising tone in utterance-medial position. (64) mbopó yaángu "my machete" mba[ yulyu

"that grasshopper"

aatwétj mboópo if

"he didn't take the ma­ chete"

When these nouns have allomorphs with a syllabic prefix, the stem's tonal behavior remains the same (this fact is of importance because, as I shall show later, the surface CVv"CV is accentually three ways ambiguous) . (65).

mapaái

"grasshoppers"

Iutoondwa lutondwá

"star" lulyú

"that star"

naagabwénj mapaáj |í "I didn't see the grass­ hoppers" There are also unaccented nouns which take a H tone on the second vowel of the word in isolation by H Tone A s ­ signment. Comparison of the isolation forms of the Class 9 forms with the related Class 6 forms indicates that these nouns are lexically unaccented, since the Class 6 form has a single H tone on the first stem vowel (and if the initial vowel were accented, we would expect Tone Copy

TONE IN KIMATUUMBI

375

tc apply). The phrase-medial forms also indicate that these nouns are unaccented in all allomorphs. ( 6 6 ) " p r e g n a n c y " pl. augmentative "that isn't a pregnancy" "those aren't pregnancies" "pig" "huge pigs" "that isn't a pig" "those aren't huge pigs" But, there are two additional tonal classes, where only one additional tonal class would be expected in the pres­ ent analysis. Additional analysis is necessary to uncover the underlying account of these stems. One set of nouns has final H tone in Class 9. ( 6 7 ) " l i p r i n g " "wires" "hammer" "gazelle" Another set of nouns has penultimate rising tone in Class 9. ( 6 8 ) " f l y i n g termite" "cow" "elephant" "civet" Interestingly, for both tonal classes, the related form with a vocalic prefix has H tone on the first and last vowels. ( 6 9 ) " l i t t l e hammer" "wires" "lipring" "gazelles" "flying termites" "cows" "civets" Elsewhere in the nominal system the tone pattern CvVCv" de-

AFRICAN LINGUISTICS

376

rives from the application of Tone Copy to an initial-ac­ cented stem (n-toopé), so we may reasonably assume that these nouns are initial-accented. Moreover, since we al­ ready have accounted for the unaccented and the final ac­ cented class of nouns, only a single additional class (the initial-accented class) could be automatically incorpo­ rated into the accentual approach being advocated here. Assuming that both tonal classes of nouns in (67) and (68) have initial accent, and if the general restrictions on accent are to be maintained, some auxiliary hypothesis must be added. First, it is important that the set with penultimate rising tone in Class 9 ( (68) - the class that includes ndeémbo) is a rather small class; the following nouns ex­ haust the set of such nouns with penultimate rising tone, to the best of my knowledge. ( 7 0 ) " e l e p h a n t " "cow" "civet" "warthog" "flying termite" "small black ant" Moreover, this class is unstable, in that some nouns, such as n'oombe, have two surface variants, one with penulti­ mate rising tone (n'oombe) and the other with final H tone (n'oombe). Since this class of nouns is anomalous, in that it represents the unstable and far smaller class, one would expect the anomalous case to be given the the­ oretically anomalous analysis. Relegating the ndeémbo class to later portions of the analysis, we may consideran analysis to account for the ndoofñá class of nouns. There is a Deaccentuation rule to account for the fact that nouns like -togolo undergo a shift in tone in Class 9. This same Deaccentuation rule can account for the fact that the initial vowel of ndooñá bears no H tone; that vowel has undergone Deaccentuation. It is assumed that Tone Copy applies before Deaccentua­ tion, to account for the fact that the final vowel has a H tone. * (71) ndooña underlying * |H ndooña

Tone Copy

H ndooña

Deaccentuation

TONE IN KIMATUUMBI

377

If Deaacentuation applies before Tone Copy, the assigned H tone will icorrectly appear on the second stem vowel, giving the incorrect form *ndoona. * underlying (72) ndooñá * ndoona Deaccentuation *ndoona

H Tone Assignment

Some ordering of rules is also necessary between Re­ traction and Tone Copy of Deaccentuation, to prevent the final H tone of ndooñá from undergoing Retraction. There are two ways to prevent this H from being retracted. One is to order Tone Copy after Retraction, s that when Re­ traction applies, the underlying form ndooña does not have a final H tone. This solution is the simplest. A more complex solution is to impose a restriction on Re­ traction, that a H tone cannot be retracted when the pre­ ceding vowel has a H tone; in this case, the form ndoona cannot undergo Retraction, and later undergoes Deaccentua­ tion, and loses its initial H tone. With this restriction, Tone Copy could be allowed to apply before Retraction. There is no empirical evidence to show which hypothesis is correct. + The form ndooñá, from ndooña, is derived by independ­ ently motivated rules, assuming that Deaccentuation ap­ plies after Tone Copy. The tonal behavior found in the ndooñá accentual class of nouns is also found in Class 5, the other morphological class where Deaccentuation ap­ plies. The Class 5 form of the nouns in (73) has the shape CVVCv", while the plural in Class 6 has the shape CvVCv". (73) | i tuutá

"planting ridge"

matyutá Ii t matyyndá |jpeemba mapéembá

pl. uundá

"fruit" pl. "grain of sorghum" pl.

T h e s e n o u n s m u s t a l s o b e initially a c c e n t e d ; t h e C l a s s 5 a l l o m o r p h u n d e r g o e s T o n e C o p y and D e a c c e n t u a t i o n in a f a s h i o n e n t i r e l y a n a l o g o u s t o t h e ndoona c l a s s of C l a s s 9 nouns. An important difference between Class 5 C W C V stems and Class 9 C W C V stems is that there are no parallels in Class 5 to the ndeémbo class (as would be manifested by hypothetical alternations such as * I i t a á I a ~ matáalá). Having accounted for the regular case of initial ac-

AFRICAN LINGUISTICS

378

cent we may turn to forms like ηdeémbo ~ matéembó . Stems like ndeembo are different from the ndooñá class only in that their expected final Η tone is retracted to the pre­ vious vowel. The simplest way to account for the ndeembo class of nouns is to assume that these nouns are initial­ ly accented, undergo Deaccentuation and Tone Copy like the ndooñá class, but also are diacritically marked to undergo a secondary retraction rule in final position, a rule such as (74) . (74) [+DJ Special

Retraction

This rule would be restricted in that only nouns in Class 9 can undergo the Special Retraction rule - no nouns in Class 5 undergo the rule. Furthermore, as observed before, this class has a very limited and unstable membership, a trait which I believe is indicative of a diacritically constrained rule. 2.12

A Reconsideration

of

phrase-medial

position

So far, the tonal alternation H final ~ 0 medial has been accounted for by assuming that stems with that al­ ternation are unaccented and the H tone in isolation de­ rives from application of HTA. However, this does not hold true in all cases. Certain conclusions must be reiterated here. I argued in the preceding section that Tone Copy applies before Deaccentuation, in order to account for the fact that the final H tone in ndooñá is on the final vowel. It is not obvious how to explain the fact that in phrase medial po­ sition, the final H tone (from Tone Copy) is missing. This anomaly holds both in the ndooñá tonal class and in the ndeémbo tonal class. (

7

5

)

.

"

I

didn't see the giraffe" "I didn't see the lipring" "We'll take the horns for Kiwiiyo" "I didn't see an elephant"

There is at present no mechanism in the analysis which predicts that the isolation form ndooñá has the phrasemedial form ndooñá. Another apparent anomaly is the fact that in initialaccented CVCV stems which undergo Deaccentuation, the fi-

TONE IN KIMATUUMBI

379

nal H tone (from Tone Copy) which is seen in isolation is lost in phrase-medial context. ( 7 6 ) " l a r g e (CI.7)" "large (Cl.9)" "my large one" "small (Cl.7)" "small (Cl.9)" "my little one" Finally, in other stems where Tone Copy applies, whenever the stem stands in utterance medial position, the expected final H tone from Tone Copy is missing. ( 7 7 ) " m o s q u i t o e s " "piece of wood" "handles" "I didn't kill mos­ quitoes" If "I didn't take a piece of wood" "I didn't take handles" A unified explanation for the apparent loss of H tones in the preceding examples is simply that the Tone Copy rule applies only in utterance-final position. With this revision, we are able to account for all cases of "disappearing" H tone; either the H tone derives from HTA, or from Tone Copy. Notice also that a final accented H tone is not lost when it is preceded bv the H tone derived from PPHTA. (78) "I didn't see the hartbeests" "two bosses" So, it is not simply a question of never having final H tone after H tone in utterance-medial position (and in­ deed, such a rule would also fail to account for the loss of final H in ndooná). 2.12

Doubly-accented

stems

Up to this point, I have argued that Kimatuumbi nomi­ nal tonology is best explained by giving an accentual analysis to nouns, where stems bear at most one accent,

380

AFRICAN LINGUISTICS

which is then mapped onto a tonal form by application of mapping, shifting, copying and reduction rules. The goal of accounting for multiple H tones in a stem is important for the accentual analysis, since as discussed earlier, without beginning with the assumption that accents are limited "one-per-morpheme", an accentual analysis could be arbitrarily given for any tonal system. Although the enterprise of reducing multiple tones in a stem has been succesful up to this point, there are in fact a few nouns which bear two H tones which cannot be accounted for by any independently motivated rules, and which must bear two accents. Below, I give a list of all of the nouns which I have found in the language with two underlying accents. ( 7 9 ) " c i r c l e " "Muslim rosary" "a dance" "bats" "bug" "black army ant" "leftover grain" "hiccup" "crab" Most of these nouns are reduplicated, viz. or maput'put í. In the case of the noun kibatgleba"a dance", the unreduplicated form also ex­ ists (with the same meaninq). The two unreduplicated nouns and are anomalous in numer­ ous ways. First, nouns almost never have two long vowels in the stem, although the two, out of the nouns in the language, do. Second, long vowels are elsewhere never found in the fourth syllable from the end, although they are found in these stems. Third, the initial long vowels of these stems are immune to Shortening (although the pe­ nultimate long vowel undergoes Shortening). ( 8 0 ) " m y circle" "a long rosary" It seems likely that these stems can best be ac­ counted for by assuming that the stem contains two pro­ sodic spans, and that within a span, multiple accents are disallowed. Moreover, I assume that the failure of Short­ ening and the existence of long vowels in the fourth syl-

381

TONE IN KIMATUUMBI

lable in (80) is also explained by assuming that these nouns are in a sense "phonologically compound". The nouns in (79) would have structures like those in (81). (81)

In the case of the reduplicated nouns, the bracketings have morphological justification as well. 2.13

Summary This then covers the relevant complexities in the accentual and tonal system of Kimatuumbi nouns. I have shown here that the tonal pattern of nouns is best char­ acterised by assigning accent to one (or none) of the syl­ lables of a noun in underlying representation, that this accent is then associated with a H tone, and that the com­ bined accentual and tonal representation is acted on by various later rules, which insert and delete tones, erase accents, and retract tones in various environments. I have shown that the vast majority of nouns and adjectives in Kimatuumbi need only to be characterised lexically for the position of a single accent mark, although there are a small number of nouns with two accents. Below, I summarise the relation between underlying accent, syllable structure and tone, with examples showing the surface phonetic form for each accent and syllable structure type. m-pa ka

"boundary"

ι *

mpa ka ntind i

η-tínd í

"buttermilk"

η-kot a

"cane"

n-kaát e

"loaf"

η-tóope

"handle"

nt oope

m-py u nga

"rice"

muunga

syyngy lá

"rabbit"

ηjeénjema

"mosquito"

ma-t óondoI 

"steinbucks"

syyngy la * njeenjema * matoondol o

1i-tiindiii

"watermelon"

symaáli

"nail"

η kot a * η kaat e

l it i ¡ndi 1 i * symaal[

382

AFRICAN LINGUISTICS

"amaranthas" "bird cage" "guard hut" There are numerous logically conceivable combinations of tone which are unattested in Kimatuumbi nouns, even ob­ serving a restriction that no more than one H tone is pos­ sible in a word (a restriction which is clearly falsified at the phonetic level). For example, there are no nouns of the shape CVVCVCV, unless they are Class 9 unaccented nouns (which can be verified by inspection of the nonfinal form of the noun - moreover, any unaccented CWCVCV Class 9 noun must have the tonal shape CVv"CVCV) . It can­ not be reasonably maintained that Kimatuumbi surface forms are characterised by a single H tone per morpheme, since, systematically, penultimately accented nouns bear two surface H tones, and all nouns with a vocalic noun class prefix bear a H tone on one of the first two vowels. I have argued here that Kimatuumbi nominal tonology should be stated in terms of an underlying distinction in the position of accent. I have, however, also argued for the existence of a small number of stems which bear two accents. Some of these stems, viz. , are analysed as being composed of two morphological domains, although some, such as are not. These nouns, albeit quite rare in the language, show that the "one-permorpheme" restriction on accent cannot be maintained as an absolute and inviolable restriction. Rather, the "oneper-morpheme" restriction should be seen as a principle which strongly guides the linguist in deciding between accentual and tonal analyses, and which helps the child learning to select a tonal grammar for his language on principled grounds. The claim is, then, that if a language develops a considerable number of morphemes which must be analysed as bearing two accent marks, the child learning the language will reject the accentual analysis, and will instead develop a tonal analysis for the language.

3.

THE VERBAL TONE SYSTEM

I will now take up the description of the tonal sys­ tem of verbs, with the goal of showing that verbs are not given their tones derivatively by rules assigning accent. Rather, I argue that rules assign tone to verbs directly. I shall also bring into the discussion nouns and adjec-

TONE IN KIMATUUMBI

383

tives productively derived from verbs, and show that they have the same basic tonal characteristics as verbs. The conclusion I will end with is that accent is the only de­ vice used in Kimatuumbi to mark lexical prosodic contrasts, so the nonaccentual nature of verbs and deverbal nouns and verbs follows from the fact that there are no lexical pro­ sodic contrasts in verbs stems. In verb stems in Kimatuumbi, as in numerous other Tanzanian Bantu languages such as Kihehe, there are no lexical tonal classes for verbs. Tone rules assign tone to all verbs on the basis of syllable structure and mor­ phological construction. For example, in the perfective, every verb stem bears H tone on the initial vowel. ( 8 3 ) " w e are asleep" "I'm growing" "I cut" "it is shining" Similarly, in the future tense and in the infinitive, the H tone is assigned to the initial stem vowel. ( 8 4 ) " t o cook" "to cut" "to shine" "I'll cook" "we'll whitewash" In noun focal tenses, (when an object prefix is pres­ ent) , the tone of the verb is on the initial syllable. ( 8 5 ) " I ' m cooking sweet potato for him" "He was cooking sweet po­ tato for me" Finally, in verb focal tenses with the prefix -na"even", the initial stem vowel has a H tone. ( 8 6 ) " H e even cooked for him" "He's even cooking" "they were even cooking for me" On the other hand, in the verb-focal tenses without the prefix -na- a single H tone is assigned to the last vowel of the verb.

AFRICAN LINGUISTICS

384

fell" "I'm cooking" "I was cooking" The verbs just discussed here form the class of as­ sertive verbs, viz. main-clause declarative, a morpholog­ ical class which typical has the "unmarked" tone pattern in Bantu languages. In Kimatuumbi, tone can be assigned by two simple rules. First, there is a general rule as­ signing the stem-initial H tone. (

8

7

)

"

I

(88) [+assertive]14 Then there will be another rule shifting the tone of the "unprefixed" verb-focal verbs seen in (84) to the final vowel. (89)

The more interesting cases of tone assignment are those which are sensitive to the number of vowels in a stem. For example, the H tone of a subjunctive verb is as­ signed to the third vowel after the subject prefix. In the following examples, this is also the third vowel of the stem. ( 9 0 ) " y o u should eat" "you should chop" "you should cook" "you should guard" "it should shine" "you should blunt" "you should scare" "you should dig tubers" Addition of an object prefixes, or one of the tenseaspect prefixes -a- "remote" or -ka- "go and" has the de­ rived effect of moving the tone to the second stem vowel, which is the third vowel after the subject prefix. ( 9 1 ) " y o u should cook it" "you should go cook" "you should scare me" "you should go scare"

TONE IN KIMATUUMBI

385

"you should guard me" "you should guard (fut.)" "you should blunt (fut.)" Finally, the tense-aspect prefixes -a- and -ka- can be combined with an object prefix, with the result that the H tone appears in the first stem vowel; again, the third vowel after the subject prefix. ( 9 2 ) " y o u should go cook it" "you should go scare me" "you should guard me (fut.)" This same rule applies in the subordinate progressive tense, formed with the prefix ka- before the subject pre­ fix, where the H tone of the verb appears on the third vowel after the subject prefix. ( 9 3 ) " w h i l e I was eating" "while I was chopping" "while I was cooking" "while I was guarding" "while I was blunting" "while I was cooking it" "while I was scaring" "while I was blunting it" Therefore, the following rule is assumed to assign H tone to verbs in the subjunctive and the subordinate present. (94) Sub.Pref + ( (V) V) Although the H tone would be expected to appear on the final vowel of the word in (95), the H tone instead appears on the second vowel; this happens just in case the final H tone is preceded by a long vowel. ( 9 5 ) " y o u should cut" "you should take" "while I was cutting" Retraction (28) applies in nouns to move a H tone off of a vowel when the preceding vowel is long. Applying this

386

AFRICAN LINGUISTICS

rule to the expected form u-kaaté yields the phonetically correct form y-kaáte. So Retraction applies in verbs as well as in nouns, and we may also conclude that Retraction applies after the rule assigning tone in the Subjunctive. This then provides additional motivation for the earlier analysis of Retraction as a tone retraction rule, rather than an accent retraction rule. 15 The Retraction rule does not apply in verbs across the board; thus, the final H tone of the following verbfocus verbs is not retracted, despite the fact that the structural description of Retraction is met at the sur­ face. ( 9 6 ) " I cut" "I was talking" "he is sleeping" The facts above can be accounted for in one of two ways. Either Retraction is restricted so that it does not apply to a focused verb- or else the tone rule shifting that tone from stem-initial position is ordered after Retrac­ tion. Other verb tenses have other tone assignment rules, and we shall see here that to a certain extent the tone assignment rules cut across morphological categories, so the various tone-assignment rules must in fact be broken up into two parts. In most subordinate nonpast forms, a H tone is assigned to the second vowel of the stem (in­ cluding the object prefix. The object prefix in Kimatuumbi, as in many other Bantu languages, is partially incor­ porated into the stem) . Thus, in the present tense pro­ gressive relative clause form, a H tone appears on the first stem vowel after an object prefix, and on the sec­ ond stem vowel when no object prefix is present. ( 9 7 ) *

'

"when "when "when "when "when "when "when "when "when "when

he he he he he he he he he he

eats" eats it" chops" chops it" cuts" cuts it" guards" guards it" scares" scares me"

TONE IN KIMATUUMBI

387

A similar tone assignment pattern is found in the tenseless "having Y'd" subordinate verb form (AG-angaROOT) as well as in the past conditional; the second vow­ el of the root is assigned a H tone if no object prefix is present, and the initial vowel of the stem is assigned a H tone if an object prefix is present. ( 9 8 ) " w i t h o u t cleaning" "without cooking" "without cutting" "without chopping" "without chopping it" "without cooking it" "if I had fallen" "if I had cooked" "if I had cooked it" I assume that the following tone assignment rule is at work here. (99)

stem (V).

The subordinate forms of the recent perfective are assigned two H tones, one of which appears on the second stem vowel (the first, after an object prefix), and the other of which appears on the final vowel. ( 1 0 0 ) " w h e n he cut" "when he cut it" "when he chopped" "when he chopped it" "when he cooked" "when he cooked it" "when he scared" "when he scared me" "when he uprooted" "when he uprooted them" A similar tone assignment pattern is found in the subordinate past tense progressive, and is also found in

388

AFRICAN LINGUISTICS

deverbal Stative adjectives. ( 1 0 1 ) " w h e

n he was cooking1 "when he was cutting" "when he was whitewashing" "when he was whitewashing it" "in a cut up condition (C1.3)" "in a cooked condition (C1.3)" "in a scared condition (Cl.l)" "in an uprooted condition (C1.3)" It thus appears that tone in subordinate verb forms can be assigned by rules referring to tense; any non-fu­ ture verb will be assigned a H tone on the second vowel of the stem, and any past tense subordinate nonconditional verb will also be assigned a H tone on the final vowel as well. Thus, I assume the following pairs of rules. (102) a. [V

b. Verbs in the far past perfective subordinate are as­ signed a H tone on the final vowel, but do not take a tone on the second stem vowel, as is predicted by the pre­ sent analysis. I assume that there is, in addition to the rules above, a rule which deletes the first H tone in the stem in the subordinate far past perfective. Examples of the far past perfective are given below ( 1 0 3 ) " w h e n I chopped" "when I took" "when I cut" 16 "when I cooked" "when I whitewashed" ■ "when I whitewashed it" These verbs differ from the recent past perfective subordinate verbs only in that this class of verbs has

389

TONE IN KIMATUUMBI

the tense prefix -a-, whereas the recent past forms do not. In both cases, the expected tone pattern would be I therefore assume the following tone-erasure rule. (

1

0

4

)

P

e

r

f

e

c

t

i

v e Tone Erasure

The possibility arises that this tone erasure rule applies either before or after assignment of the final H tone. There is evidence showing that tone-erasure applies before the rule assigning final H tone, but, to uncover this, we need to look into assignment of tone in redupli­ cated verbs. The pattern of reduplicated verb stems in the perfec­ tive indicates that unreduplicated stems and reduplicated stems are not tonologically identical. The problem is that there are certain options available with reduplicated verbs which are not available with unreduplicated verbs. The tone rules can apply regularly to reduplicated verb stems, assigning the first H tone to the second stem vow­ el of the entire stem (composed of the first and second copies of the stem), and assigning the last H tone to the word-final vowel. ( 1 0 5 ) " w h e n I frequently cut (recent)" "when I frequently chopped (recent)" "when I frequent­ ly plastered (recent)" "when I frequently chopped" "when I frequently cooked" "when I frequent­ ly plastered" For the recent past perfective, it is obligatory for the tone assignment rule to place the first H tone on the second vowel of the leftmost stem and the final H tone on the word final vowel. If the tone pattern is distributed over only the first stem copy, or over the last stem copy, an ungrammatical form results. (106)

However, for far past perfective stems (where tone-

390

AFRICAN LINGUISTICS

erasure applies), it is also possible to assign the final H tone to the final vowel of the first copy of the stem,17 (107)

Assuming that tone erasure has applied before the rule mapping H tone onto the final vowel, it is therefore possible to explain the distinction between the far past and recent past perfectives on the basis of application versus nonapplication of tone erasure. I therefore pro­ pose that the rule assigning the final H tone is sensitive (in a rather complex manner) to reduplication and the pre­ ceding tones, so that H tone is assigned optionally to the leftmost "final vowel" just in case there is no H tone in that stem, where otherwise the H tone is assigned to the word-final "final vowel". This analysis then depends on the prior application of tone erasure, in order for the dichotomy between recent and far past perfectives in fi­ nal tone assignment to be explained. As in certain other Bantu languages (Shona, Odden (1981a)), the reflexive prefix in Kimatuumbi requires the insertion of H tones in the stem; in the reflexive, a H tone is assigned to the second stem vowel and the last stem vowel. ( 1 0 8 ) " t o cook for oneself" "to cut oneself" ! "to plaster for oneself" "I just cooked for myself" "you should cook for yourself" "when I cooked for myself" "while I am cutting myself" Interestingly, although the Object Prefix in subordinate verb forms counts as part of the stem for tone assignment (so that a H tone appears on the initial stem vowel after an Object Prefix), the reflexive object prefix seems not to be absorbed into the stem, since it does not count as the first stem vowel. Rather, assignment of tone after the reflexive starts after that prefix. Furthermore, the as­ signment of the reflexive tone pattern takes priority over a tone pattern inherently required in the verb tense (for example the second-and-last pattern of the reflexive ap­ pears in kwif-kaátá instead of, and not in addition to, the first-vowel pattern of the assertive).

TONE IN KIMATUUMBI

391

Agent nominalisations of verbs are assigned two H tones,, the first of which is assigned to the initial vow­ el, and the second of which is assigned to the third vow­ el of the stem, in case the stem has only two vowels, the second tone is assigned to the second vowel, as guaranteed in (111).. ( 1 0 9 ) " c h o p p e r " "guard" "cook" "scarer" "cutter" "one who plasters for" The same tone pattern is encountered in abstract nom­ inalisations ("-ness"), formed with the suffix vowel i and the Class 14 prefix y-. ( 1 1 0 ) " t a l l n e s s " "fatness" "bluntness" The following rule will therefore assign the two H tones to the deverbal nominalisations. ( 1 1 1 ) N o m i n a l i s a t i o n [stem Finally, adjectives formed with the suffix -y have a H tone assigned to the second stem vowel. ( 1 1 2 ) " h e a v y " (topa "be heavy") "wet" (tépeengala "be wet") "rotten" (bóla "to rot") To summarise the productive rules for assignment of tone to verb stems, one or two tones will be assigned to the verb stem, depending on morphological construction. The first H tone may be assigned to the third stem vowel, in the case of the subjunctive and progressive subordi­ nate, to the second stem vowel in other subordinate con­ structions, or to the first stem vowel in assertive verbs. A second H tone may also be assigned to the final stem vowel in past tense subordinate forms. In addition, H tones are assigned to the first and third vowels of the stem in agent nominalisations and in abstract nominalisa­ tions.

392

4.

AFRICAN LINGUISTICS

COMPARISON OF TONE AND ACCENT IN KIMATUUMBI

One of the first things that becomes obvious in the comparison of verbal and nominal tone is that what con­ stitutes a possible verbal tone pattern is vastly differ­ ent from what constitutes a possible nominal tonal pat­ tern. For example, although in nouns, there is no pattern in agent nominalisations, the pattern can occur easily, for example "one who blunts". Here I discuss the verbal tone system, showing why it cannot be characterised with accent, but must instead be characterised with tone rules, which map tone onto stems without ever assigning accent to verbs. First, con­ sider the initial-vowel H tone pattern of the infinitive. If the H tone of the infinitive were derived by assigning accent to the initial syllable and mapping H tone onto that accented syllable by ITAR, then we would also incor­ rectly predict that Tone Copy should apply in the form "to chop for me" since that rule assigns a H tone to the final vowel of a penultimate accented noun. This would give the incorrect form Therefore, the pe­ nultimate accented noun ma-lóbé differs from the infini­ tive " since the former (but not the later) under­ goes Tone Copy.18 Similarly, although tone is assigned in the verbal form uniteléke to the penultimate syllable (and vowel), Tone Copy does not apply here either. If tone assignment in verbs were to assign accent marks, rather than tone, Tone Copy would incorrectly apply to penultimately accented verbs and deverbal nouns. ( 1 1 3 ) " s c a r e r " "you should cook for me" "while you were cooking for me" "when he cuts" "chop" Since Tone Copy applies after an accented penult and de­ rived forms do not have accent, the failure of Tone Copy

TONE IN KIMATUUMBI

393

in (113) is predictable. The alternative to the accentual/ tonal dichotomy is to encode the lexical/derived dichotomy directly in Tone Copy, so that it only applies in. nouns and adjectives which are derived. This move is at best du­ bious, since in no case must the global distinction "de­ rived" versus "lexical" be directly incorporated into the formal statement of any phonological rule in any language. Another difference between nouns and verbs is that, whereas nouns have accent on a single syllable (but not vowel), verbs are assigned tone on the basis of vowels, so that, unlike nouns, there may be a fundamental contrast between vv and vv (as opposed to the case in lexical stems, where the surface contrast derives by independent retrac­ tion rules). Thus, the basic unit for tone-assignment in verbs is the vowel, whereas in nouns, the basic unit for accentual distinctions is the syllable. If tone assignment in verbs were based on the assignment of accent, radically different accent-restriction principles would be necessary, to allow accent in verbs to be assigned to the vowel. In addition, verbs may freely be assigned two H basic tones, whereas, as observed in the discussion of nouns, only a single accent is generally assigned to nouns. The proposed distinction between tone and accent in Kimatuumbi also helps simplify the formulation of a sandhi rule. There is a rule in the grammar, Initial Tone Inser­ tion (discussed in Odden (1981c)), which assigns a H tone to a set of morphemes (diacritically marked, for present purposes), when they are preceded by an unaccented noun. ( 1 1 4 ) " w h a t type cover ?" "what type meat ?" "what type big one ?" "that large one" "what type bird cage ?" "what type handle ?" "that mosquito" "that long one" "that long one" When there is an accent somewhere in the noun, the fol­ lowing D morpheme bears L tone, and, furthermore, when a D morpheme is initial in an utterance, it also bears L tone. After an unaccented noun, a D morpheme takes a H tone. Finally, note that, although the adjective stem -kulu is basically accented (n-kúlú), it undergoes Deaccentuation

AFRICAN LINGUISTICS

394

in the Class 9 form ngulu above, and therefore does trig­ ger the sandhi rule. Now let us consider what happens when a verb precedes a D morpheme. In the forms below, where a H tone appears on the stem on a noninitial vowel, the Initial Tone Inser­ tion rule is blocked, and the D morpheme has a L tone on the surface. (

1

1

5

)

"

w

h

a

t type should I take ?" "what type did he fry ?" "what type should I go gaηi guard ?" "what type should I plaster for you ?" "what type will you eat ?" "what type should I eat ?" However, when the verb bears no Η tone, the D mor­ pheme takes a H tone by Initial Tone Insertion. ( 1 1 6 ) " w h a t type is he taking ?" "what type were they taking ?" "what type did you just take ?" The crucial distinction between verbs and nouns is that, if a verb bears a H tone on the initial vowel of the stem, Initial Tone Insertion will apply. ( 1 1 7 ) " w h a t type have you taken ?" "what type will you take ?" "what type should I go and take for you ?" This then leaves us with the task of capturing the generalisation regarding the categorial conditions on the rule. If the rule only refers to phonetic tone and its position, it will be necessary to complicate the rule to prevent nouns with initial Η from triggering Initial Tone Insertion, but also allow verbs with initial Η to trigger the rule. (118) ø → Η [+D]

A simpler solution not requiring this noun-verb distinc­ tion to be included in the rule is to phrase the rule in

TONE IN KIMATUUMBI

395

terms of tone and accent. The rule, as stated in (119), applies after a word with no accent, and no H tone after the stem-initial vowel. (119) ø -> H [ (H) L1 ] [-ace]0

[+D]

Initial Tone Insertion

If nouns are assigned their tone based on underlying ac­ cent, but verbs are assigned tone without an accentual stage, then the noun-verb dichotomy seen in the Initial Tone Insertion rule is explained. So, although the Initial Tone Insertion rule is complex in some respects, some of the complexity can be eliminated by an accentual analysis. I have shown here that various tonal phenomena of Kimatuumbi require nouns and adjectives to be lexically marked with accent, and that verbs and deverbal nouns and adjectives have their tones assigned by tone assignment rules. The principles for verbal tone assignment are based on vowels, not syllables, but that accent is assigned lex­ ically to the syllable, not the vowel. Accent must remain on the noun after tone is assigned to accented nouns, and some rules make crucial reference to. accent. This then shows that accent cannot be replaced with H tone. There has been a debate over whether accent should be viewed as a feature, i.e. [+ accent], or as an entity, generally a star (*).There is evidence in Kimatuumbi for both positions. In favor of the "entity status" of accent is the fact that in Kimatuumbi (and Tokyo Japanese), it is a property of the syllable, not the vowel.19 While other entities like  and V associate with the syllable node, no evidence is available that any features associate directly with the syllable. Rather, features are associated with the V and  nodes which are associated with the syllable. In this way, accent in Kimatuumbi acts like an entity. But accent in Kimatuumbi acts like a feature in having plus-and-minus values to which rules refer, reca­ pitulated below. (120) Tone Copy: [+accent] Initial Tone Insertion: [-accent]0 [-accent] Nominal Retraction: If accent were an entity, which is either there or not, then one would expect rules to refer to its presence, but not its absence, just as one expects rule to be condi­ tioned by boundaries, but not their lack. Still, since the literature abounds with analyses which include con­ ditions such as "does not apply across ≠ ", even this evi-

AFRICAN LINGUISTICS

396

dence may not ultimately reveal the true nature of accent.

5.

ACCENT AND TONE IN BANTU LANGUAGES

Having argued in the preceding sections that Kimatuumbi is in fact an accentual language (and moreover must be an accentual language), I shall take into consideration the general question of what properties are necessary for a language to be called "accentual" versus "tonal", con­ sidering in more general terms the problem of distinguish­ ing between an accentual and a tonal language. I will look at, and to some extent reject, suggestions made in Hyman and Byarushengo (1980) and Hyman (1982a,b), where it is suggested that certain types of rules are characteristic of accent languages. I then look at the historical change from tonal language to accentual language in some Bantu languages, concentrating on the transition from tonal analyses to accentual analyses. 5.1

Distinguishing

tone

and

accent

Although it is clear that Kimatuumbi is an accent language (in the lexicon), a more general question may be raised - how can one know, in lieu of rules which refer directly to accent, whether a language is accentual or not ? The term "accent" has been used to refer to phenomena related to pitch and intensity in languages like Iroquoian (Chafe (1977)) or Creek (Haas (1977)). The word is some­ times used to refer to stress (and, interestingly, many "tonally" accentual languages like Kimatuumbi, Haya or Tonga do not have independent "stress".) In fact, there is little agreement about what "accent" means, and it is difficult to extract a coherent generalisation from the literature (which encompasses generative and nongenerative traditions). It is clear that there is no consistent phys­ ical or acoustic correlate which separates "accent" from stress or tone. So, if there is no "theory independent" agreed on characterisation of accent, we must look for a "theory dependent" characterisation of accent, which will be most likely based on what accent does, not what it sounds like. It is within an autosegmental framework that it is most profitable to talk about what accent "does". In the general theory of accent outlined in Goldsmith (1981), the distinction between an accentual language and a tone language is that the former, but not the latter, uses a

TONE IN KIMATUUMBI

397

star to mark the position of the prosodiC contrast (lexi­ cal in Kimatuumbi, or morphologically derived in the case of verbs in Tonga). As previously discussed, it is also a working assumption that morphemes may bear only a single star, at least in lexical representations. We may thus rephrase the question more formally as asking when a lan­ guage may use a star. One position which an analyst might take is that a language should be analysed accentually just in case it can be, that is, just in case there are n+1 (or fewer) tonal contrasts for morphemes of η prosodiC units. In this view, it would be possible (and necessary) for a con­ siderable number of Bantu languages to be treated as ac­ centual, including languages such as HiBena (M. Odden (in progress)), Safwa (Vorhooeve (1971)), Kinga (Schadeberg (1971)) or Digo (Kisseberth (1981)). I shall refer to ·. this view as the free-accent hypothesis. An alternative view on the accent question would be that additional evidence is necessary in the grammar to force the conclusion that a language is accentual (beyond the constraints on multiple accent). This hypothesis I shall refer to as the accent-restriction hypothesis. Of course, in Kimatuumbi, additional evidence is available for example, the lexical - derived contrast in accentua­ tion, as well as rules which refer to accent distinct from tone, all provide additional evidence for an accentual analysis of the language. There is no a priori reason to expect that extra ev­ idence is necessary or unnecessary in order to maintain an accentual analysis of a language. It would rather seem to be a question of taste and not logic - whether one be­ lieves that accentual analyses require extraordinary mo­ tivation. No more analyses should be possible under the accent-restriction hypothesis than under the free-accent hypothesis. The former hypothesis sets certain minimal standards which must be met for a language to be accentual, but once those standards are met, the language will be analysed accentually. The later hypothesis states that, once the "one-per-morpheme" restriction is satisfied, a language will be analysed accentually. So, while the two hypotheses lead to different predictions about which lan­ guages will be accentual, neither is indeterminate (ex­ cept, perhaps, insofar as the decision-making criteria of the "restricted accent" hypothesis may be falsified or vague ). . The free accent hypothesis can be falsified in at least two ways. First, if a language is like Kimatuumbi in distinguishing accent versus Η tones (or L tones) but by the free accent hypothesis is not accentual, then the

398

AFRICAN LINGUISTICS

hypothesis is false. Second, if a language is organised lexically so that H tone can be associated with multiple vowels (where the associations do not derive from a wellmotivated spreading rule), but should be accentual by the free accent hypothesis, then the hypothesis is false. This is because accent, unlike tone, is associated with one syllable, not a sequence of syllables. So, it is not necessary, for some a priori reason, to go to great lengths to prove that a language is accen­ tual (unless, of course, one can show that the "free-ac­ cent" hypothesis is false, by showing that there is a lan­ guage which "could" be accentual given the "one-per-morpheme" constraint, but isn't). Still it is instructive to discuss possible principles which may be brought to bear in deciding whether a language is accentual versus tonal. I therefore propose to briefly review some of the argu­ ments for an accentual analysis of Haya proposed in Hyman and Byarushengo (1980) (henceforth H/B), in order to see what criteria have been brought to bear in arguing that Haya is an accentual language. Similar typological argu­ ments are found in Hyman (1978). It should be noted that Hyman and Byarushengo (1982) have rejected the use of the star for accent in a revised analysis of Haya, claiming that accent has no theoretical status. This, it seems to me, underscores the elusive nature of accent. Haya restricts the number of H tones possible in a stem - there are fewer than n+1 accentual possibilities for words of η vowels. So, given the "free accent" hypoth­ esis, Haya an accentual language. The question then ar­ rises, must it be an accentual language ? The entire gram­ mar is treated accentually in the account of H/B, so that, unlike Kimatuumbi, it is impossible to find a contrast be­ tween accented versus unaccented Η tones in, say, verbs versus nouns. Any, unlike Kimatuumbi, Η and accent (and L and no accent) are interchangeable, so where something could be said about an accent, it could also be stated in terms of tone, (as is the tack taken later in Hyman and Byarushengo (1982).) H/B rely on other criteria to argue that Haya is an accentual language. They argue from the nature of things - that is, they argue that certain tonal phenomena in Haya can only be explained if Haya is accentual. I will evaluate some of these claims here by com­ paring Haya and Kimatuumbi (or other Bantu languages, when the relevant problem is not found in Kimatuumbi). I shall argue here that while the H/B criteria may be indicative of accent, they are not probative. I shall show that the same rules H/B cite as evidence of accent in Haya also ap­ pear in Kimatuumbi in the verbal system, where accent does

TONE IN KIMATUUMBI

399

not enter into the picture. It can therefore be concluded that the mere existence of certain rules does not mean that the phenomena must be handled accentually, since they would also be handled tonally in Kimatuumbi verbs. One of the criteria is the "one-per-morpheme" crite­ rion; I have maintained this principle as the minimum (and maximum) condition necessary for an accentual analysis. Their second criterion is the existence of a deaccentuation rule where one accent is deleted after another. Thus, H/B observe that there is a rule in Haya such as (121). (121) * → ø / * This rule could be formulated tonally as in (122). (122) H → L/H Such a rule, formulable either as accent deletion or tone dissimilation, appears in a number of Bantu languages, often in rather disguised manifestations (see Odden (1981 a) for examples in Shona), typically as a rule lowering a H tone after a H tone. From a cross-linguistic perspective, a dissimilation rule lowering H tone after H tone is not as well attested as spreading rules such as the change of L to H after H. Moreover, there is an asymmetry in tonal dissimilation, namely that the change of L to H in the environment of L is very rare, if attested at all. Thus, when one takes into consideration what types of tonal rules are "typical" and what types are rare, a rule lowering a H after a H is indeed peculiar. H/B argue that this asymmetry is explainable with an accentual analysis. An accentual dissimilation such as (121) is motivated; for example, there are numerous places in Japanese where two accents should occur, but only one does in fact. The existence of a deaccentuation rule is, in fact, nearly predicted by the logic of accentual sys­ tems. If accentual analyses are available only when there is strong evidence for accent, subject to the one-permorpheme restriction, a deaccentuation rule would neatly account for anomalous cases where two accents are acciden­ tally assigned in a word. Without such a reduction proc­ ess, it could become difficult to maintain that the lan­ guage is accentual. An accent reduction rule like (121) then helps to maintain the possibility of an accentual analysis. There is, of course, no logical reason why a tonedissimilation process could not be stated in terms of a lowering rule such as (122); the validity of connecting accent and tone lowering rests on the claim that the dis­ similation rule (122) is impossible. Whether or not it is

400

AFRICAN LINGUISTICS

rare, dissimilatory lowering rules like (122) do in fact exist in cases where it cannot be maintained that accent is at work. There are two sandhi rules in Kimatuumbi (discussed in Odden (1981c) and Kisseberth and Odden (1980)) which lower a H tone in word-final position after another stem H tone, when a word is utterance-medial I will discuss one of these rules here. This rule will low­ er the final H of a word when preceded by another H tone, separated by any number of vowels. The rule is illustrated below with verbs which have no accent. ( 1 2 3 ) " I plastered for myself" "I plastered a house for myself" "when I plastered" "when I plastered a house for him" "when I am frying it" "when I am frying it in the fire"20 The presence of a preceding H tone is crucial in the application of this tone-dissimilation rule, since, as shown in (124), a final H tone in a verb not preceded by another stem H tone is not lowered in utterance medial position. ( 1 2 4 ) " w h e n I chopped" "when I chopped trees" "you should cook" "you should cook sweet po­ tato" Therefore, the grammar must contain the following rule. (125) H → L / [Χ Η Υ ]Ζ Final Deletion But, since verbs are directly assigned tone, Final Deletion is not an accent reduction rule. So, we must con­ clude that Η tone dissimilation rules are at least pos­ sible, if not common. There are other Bantu languages which cannot be treated as accentual (since they have four tonal contrasts for two-syllable stems, as Shona does), which nevertheless have one form or another of (122). I will show in a later section that Southern Kipare is such a language. H/B state that there is a connection between pragmat­ ic focus and accent, and observe that in Haya, a verb or

TONE IN KIMATUUMBI

401

noun which is not focused will not receive a H tone unless it is in phrase-final position. H/B suggests that this correlation follows from the hierarchical nature of ac­ cent. They claim (p.28) that "...the kind of semantic con­ ditioning on Haya tonal accents has not, to our knowledge, been pointed out for any such pure tone language". Seman­ tic and pragmatic conditions on tone rules are not unat­ tested; see Odden (1981a) for an example of semantic and pragmatic conditions on a tonological Question-Formation rule in Shona, or Stucky (1979). for a tone erasure rule is Makua which eliminates the first H tone of a focused noun in certain constructions. Rather than argue with the reasoning behind this connection, I will show that it is not a necessary connection. As shown in Odden (1982b), Kimatuumbi verb tenses belong to one of three groups; some tenses focus the verb, some focus the noun, and some are neutral, and have no focus. As was seen in the preceding sections, verbs gener­ ally have their H tone on the root-initial vowel. But in the noun-focus tenses (i.e. those where the verb cannot be the focus) unless the verb has an object prefix, the expected root-initial H tone is missing. (

1

2

6

)

"

I

am cooking sweet patato" "I am cooking the sweet po­ tato" "I was cooking sweet potato" "I was cooking sweet po­ tato for him" These forms then suggest a Η-tone deletion rule such as (127). 21 (127) H → ø / SP+ X [ +out-of -focus ] H/B argue that, in an analogous case in Haya, when a verb is not focused, it loses its accent. But the facts of Haya are nearly the same as in Kimatuumbi, although there are some differences, possibly important. In Kima­ tuumbi, the verbs seen in (126) can never be focused, so they can never be at the end of a clause. Moreover, the presence versus absence of an object prefix makes a dif­ ference in the applicability of the tone deletion rule, whereas a parallel restriction is not found in Haya. But, despite these superficial differences, the rules of H tone deletion in Kimatuumbi and the accent-deletion rule of Haya must be essentially the same phonological phenom­ enon. However, since tones are assigned to verbs in Kima-

402

AFRICAN LINGUISTICS

tuumbi directly, loss of H tone correlated with non-fo­ cused position need not indicate loss of accent. Hyman (1982a:43) presents the claim that there is some connection between accentual languages and a rule of Leftward Spreading of H tone. He states; "...leftward an­ property, since Hyman ticipation of H tone is an accentual and Schuh (1974). argued that 'pure' tones tend to spread rightwards..." Like the claimed connection between lower­ ing of H tone after H, the connection between accent and leftward spread of H tone is not shown to be explainable by some inherent property of accent or leftward spread of H tone. Rather, the connection lies in the putative fact that all languages with rules spreading H tone leftward are accent language. To test this connection, we need to see either if there are cases where leftward movement of H tone is not connected to accent in an accent language, or, more strongly, if leftward movement of H tone occurs in any language which could not be analysed accentually. For the former case, Kimatuumbi serves as an example. There are three rules in the grammar which move a H tone leftward; Retraction applying after long vowels, Nominal Retraction applying to unaccented nouns, and Perfective Retraction applying in certain verbs. In each of these rules, there is no connection between accent and leftward movement of H tone. Retraction applies both to H tones derived from accent and to those derived by direct mapping of tone to vowels in verbs. Perfective Retraction only ap­ plies to verbs, which have no accent. Lastly, Nominal Re­ traction applies to unaccented nouns. So, the connection between accent and leftward movement of H tone in Kima­ tuumbi is rather weak. In Makua (Ching and Kisseberth (1981)), there is a rule which retracts a stem-initial H tone in word-final position. Thus, where the H tone of the verb should appear on the first stem vowel, it appears on the prefix vowel. Underlying /u-lyá/ undergoes High Tone Retraction and ap­ pears on the surface as ύIya "to eat", or /ki-ho-lya/ ap­ pears on the surface as kihólya "I have eaten". So, Makua has the leftward shift of Η tone. But can the languages be independently called accentual ? The verbal tone system is not conclusive, since there are no lexical tonal or accen­ tual contrasts of any sort in verb stems. On the surface, verbs may have multiple Η tones, but they are all derived predictably; multiple accents in separate morphemes are found in the accentual analyses of verbs in Tonga and Luganda, so multiple derived Η tones in verbs tells us noth­ ing. Within the Makua nominal system, however, that there are both contrasts in the position of phonetic tone and the number of phonetic tones (Kisseberth (1982)).

TONE IN KIMATUUMBI

403

(128) "cassava flour ugali" "gum" "crows" Makua is therefore not an accent language, but it has the leftward movement of H tone (it also has a rule lowering H after H, which is claimed to be a characteris­ tic of accent languages). The question then might be asked whether Makua could be called an "accent-like" lan­ guage; perhaps this is plausible, since the language has undergone considerable historical reduction in the tonal system, even so, the formal connection between accent and leftward shift is untenable. We must therefore turn to languages which unambigu­ ously preserve the proto-Bantu fourway tonal contrast in noun stems. Carter (1973) has claimed that tonal antici­ pation (the leftward movement of H tone) seems to occur only in languages which have reduced the proto-Bantu to­ nal system. Two languages which have preserved the fourway tonal contrast are Kipare and Kishambaa. In Kishambaa, as discussed in Odden (1982a), a rule shifts the fi­ nal H tone off of noun-focus verbs, so that underlying ni-já nkhándé becomes ní -ja nkhándé "I'm eating food". But, the language cannot be accentual since, as I shall show later, it contrasts both position and number of un­ derlying H tones in stems. In Northern Kipare, there are two rules spreading H tone leftward (and none spreading H tone rightward). There is a rule which lowers a H tone to a downstepped H in utterance final position (Odden (1982c)), so that un­ derlying nkhúkú becomes nkhúlkú "chickens". In utterancemedial position, the H tones of this noun are at the same level, viz. nkhuku jángl ú "my chickens". The following rule is needed for utterance-final flattening in Northern Kipare. (129)

Within a metrical analysis of tone-registers, this rule is interpreted as "downstepping" an utterance-final H tone. After this rule, a rule raises a sequence of L tones before a flattened H tones to (downstepped) H. (130) !mbáhá "big" /mbahla'/ mbáhá jángiú "my big ones'

404

AFRICAN LINGUISTICS

"I bought a cow" "he bought a cow" "I bought large cows" "he bought large cows" Thus, the following rule is required in the grammar. (131)

So, Northern Kipare has a rule spreading H tone in­ definitely far to the left, but the language cannot be accentual, since it has four tone classes for disyllabic nouns, just like Shona. There is another rule spreading H to the left. Whenever the sequence HLH is encountered between words, that sequence is optionally (though most often) realised phonetically as HIHH. ( 1 3 2 ) " t h e big dog died" id. "the big dogs died" "it is a cat" id. "the cat died" id. The most straightforward account of this alternation is the following leftward Spread rule. (133)

Kishambaa and Northern Pare have rules spreading H tone leftward, yet neither is an accent language. There­ fore, it must be possible to have the leftward spread of H tone without a language being accentual. Hyman (1982b) (and essentially the same argument is given in Hyman (1978)) discusses an example of the left­ ward spread of H in Igbo, where, as in Northern Kipare, HLH becomes H!HH on the surface; thus, underlying becomes "three goats" by a rule spreading the

TONE IN KIMATUUMBI

405

second H tone leftward. The question is then why is this not a counterexample to the claim that the leftward spread of H tone is an accentual property (given that Igbo is a tonal language, rather than an accentual lan­ guage) . He states (p.13) that "there has been no discus­ sion of 'leftward spreading* of H tone in Luganda - only of leftward association of H onto toneless vowels ... Leftward spreading would have to involve a change from L to H, not an anticipation of a H tone by a neutral ø tone vowel." Further, he states (p.15): What is at issue is the following: (a) rightward tone spreading is the result of ex­ tending a H association to V's on the right; (b) leftward tone spreading is the result either of (i) associating a H onto a toneless V to the left, or (ii) 'floating1 a L followed by the association of a H onto the resulting toneless vowels to the left. So, Hyman would allow for the existence of a "local" rule which disassociates L tone from a vowel, with a concomi­ tant association of a following H tone to the unassociated vowels, and thus he can accomodate the case of Igbo or the local rule in Northern Kipare where HLH becomes H!HH. Moreover, he allows the possibility that the L tone might be associated with a number of vowels, so that, on the surface, HL*H could also undergo leftward spreading of H tone parallel to the change of HLH to H!HH. And, it shouldn't be necessary for there to be a surface downstep separating the H tones, so that HLH could also surface as HHH, where the "floated" L tone is simply deleted. So, it should also be possible, for example, for underlying HLLL LH to appear as HHHHH, assuming that the multiple surface L tones were represented as a single L tone associated with many vowels. Hyman's revised typology of leftward spread of H tone accounts for the cases of Kipare HLH becoming H!HH, as well as the case in Kishambaa where a final H tone moves leftward. The typology can account for the leftward movements in Makua and Kimatuumbi, since in these lan­ guages, it can convincingly be argued that surface L toned vowels really bear no underlying tone; these lan­ guages thus fall under his case (i). The typology could only be falsified by showing that there is a language which has unlimited leftward spread of H tone (not Igbo, where spread is limited to one vowel), and where the L toned vowels indeed do bear L tones when the Spreading rule applies. Northern Kipare provides an example of this type of language; the rule spreading "flattened" H tone infinitely far back across a series of L tones (including L tones in different morphemes and words) is not "local",

AFRICAN LINGUISTICS

406

and there is no reason to believe that the phonetic L tones of the language are not also underlying L tones (and in fact, there are lexical floating L tones in the language which are hard to account for if L tones are (2f tones),. We may then reconsider the enterprise of deciding just how much proof is needed to claim that a language is accentual. Extra evidence such as deaccentuation, focus­ ing or leftward shift are not of necessity indications of accent. One may then wonder whether there will be any rule-type22 which provides the absolutely necessary evi­ dence that a language is accentual (versus tonal). To be sure, there are cases such as Kimatuumbi or Luganda, where stronger evidence exists for the existence of accent in­ dependent of tone (in Kimatuumbi, certain H tones are un­ accented, and in Luganda, certain L tones are accented). But what of languages like HiBena (M. Odden (in prog­ ress)), H tones are severely restricted, so that only one H tone is possible in a word (but may appear in a number of positions in the word, excluding final position). There are no independent language-internal criteria which absolutely require HiBena to be analysed accentually (and as far as I am aware, neither is it absolutely necessary to analyse Tonga or Haya as accentual). An accentual analysis is possible in HiBena, since the language obeys the one-per-morpheme constraint - I would claim that an accentual analysis is therefore necessary. 5.2

Historical

change

The last question which I shall take up here is the question of how certain Bantu languages became accentual. The same question is asked in Goldsmith (1981), where he suggests two possibilities for explaining the change from a proto-Bantu tonal system to an accentual system in Ton­ ga. One possibility, not discussed in detail by Goldsmith, is that the addition of a tone spreading rule like (134) conditioned the change from tone to accent. (134).

I shall argue here that this spreading rule is the reason why some languages became accentual. Goldsmith, however, suggests another rule. The second possibility is that the addition of Meeussen's rule (122), which lowers a H tone after a H

TONE IN KIMATUUMBI

407

tone, provided the impetus for the tone-to-accent change. Goldsmith states that Meeussen's rule is the one rule which is shared by all Bantu accentual languages. For ex­ ample, he observes that in CVCV stems, proto-Bantu had a four-way tonal distinction, but Tonga has through histor­ ical change reduced that distinction to a three-way ac­ centual distinction.23 A similar reduction from four to­ nal classes to three is found in Kimatuumbi. (135) Proto-Bantu Tonga cvcv cvcv cvcv * cvcv cvcv As Goldsmith observes, Meeussen's rule is also found in tonal languages like Shona (see Odden (1981a)). The question he then asks is what prevented Shona from be­ coming an accentual language. Now it is clear that Shona is synchronically a tonal language - it has four tone patterns for bisyllabic nouns (whereas the maximum for an accentual language would be three), and the language is rich with tonal melodies, another characteristic of tonal (nonaccentual) languages. But the historical question is still open. While we cannot say that Shona is an accentu­ al language, can we say that it was historically an ac­ cent language ? There is some indication that Shona was an accent language, the most important being that it merges the four historical tone patterns of Proto-Bantu into three, just as Tonga and Kimatuumbi do (van Spaandonck (1967)). Through borrowings and other sources, Shona now has a class of HL nouns (bhuku "book", chikóro "school", chapupu "witness"). Factoring out those complications, at least in accounting for possible tonal patterns in lexical stems, an accentual analysis could have been possible in the history of Shona. Later changes then brought about the present nonaccentual grammar. Another indication of the former accentual nature of Shona is the asymmetrical treatment of H and L tones in Shona. As argued in Odden (1981a), the phonetic sequence HH in a noun (hóvé "fish") must be treated as a single H tone associated with two vowels, whereas the sequence LL (sadza "porridge") cannot be so treated, and most likely derives from an underlying form with no tone. While it makes sense in an accentual system to speak of the pres­ ence or absence of accent, it makes marginal sense to speak of the presence or absence of tone (rather than a

AFRICAN LINGUISTICS

408

particular tone). Yet it also makes more sense in the grammar of Shona to speak of a contrast between H (tone) and no tone· And, finally, there is a "post-accenting" morpheme in Shona, viz. the frequentative prefix -sí -. Again, "post-accenting" makes sense only with reference to an accentual system. So, in fact, there may well not have been anything that prevented Shona from becoming an accentual language. The language probably was accentual, and changed to a tonal system later in its development. To test the purported connection between Meeussen's rule and accentual languages, we may therefore look to see if any languages have Meeussen's rule in some form but also retain the four-way historical tonal distinction in noun stems (and in general do not "act" semi-accentual, as Shona does). The southern dialect of Kipare is such a language. This language does maintain the four-way con­ trast in possible tone patterns for nouns, as shown be­ low. Prote-Bantu "arm" "elephant" "wound" "man" However, Southern Kipare also has a rule which low­ ers a H tone after a H tone across morpheme boundaries, as shown in (137). ( 1 3 7 ) " t o lean" "I leaned" "I bathed" "he bathed" "best (Cl.l)" "best (Cl 2)" Southern Kipare cannot be (or have been) an accen­ tual language, since the language has four tone patterns for bisyllabic stems. Yet the language has the accentual­ ly-connected H tone dissimilation rule. It may well turn out that the addition of Meeussen's rule could have "put pressure" on other Bantu languages so that they decided to become accentual, but this pressure could not have been overwhelming. There are I think two considerations which may lead toward an understanding of the change of tone to accent (136).

Southern

Kipare

TONE IN KIMATUUMBI

409

in Bantu. First, it should be observed that the over­ whelming majority of the noun stems in Proto-Bantu are CVCV in shape - longer stems are infinitely rarer, and their tonal reconstructions are more problematic and var­ iable. In verbs, roots only contrast H versus L tone (ac­ cent versus no accent) for the whole root; roots cannot be constructed with tone patterns such as HL versus HH. Since, for the most part, the determination of the accen­ tual status of a language will be based on the treatment of CVCV nouns, we may consider what change in the grammar would be necessary for these nouns to undergo a change to an accentual system. As observed by Goldsmith, Tonga and other accentual languages have undergone a historical change whereby noun stems of the tonal pattern CvOv" and CvOV are merged into a single pattern. This merger in itself yields a system which, given the "free-accent" hypothesis advanced here, in itself must be accentual. Thus, it seems to me that the first question to answer in determining how certain Bantu languages became accentual is how the merger of the Proto-Bantu tone patterns Cv'Cv' and Cv"CV came about. This then brings us to the other solution, which Goldsmith does not discuss, namely a H tone spreading rule. There is a tone-spreading process found in a number of Bantu languages which spreads a H tone from the left to the right, subject to certain language specific con­ ditions (for example, in Southern Pare, the rule will not spread   tone to a word-final vowel). Such a rule is found historically or synchronically in languages such as Shona, Southern Kipare, and probably is the ancestor of the Tone Copy rule of Kimatuumbi. The restriction on Spreading in Southern Kipare that the rule cannot apply to a final vowel would have prevented the merger of proto-Bantu HH and HL, thus preventing accentual reanalysis in that language. An interesting example of Tone Spreading can be seen in KiShambaa, which has the rule spreading a H tone onto a final L tone, but does not phonetically neutralise the contrast Cv"CV and Cv"Cv". As can be seen in the data below, the historical pattern Cv"CV is realised as CvOv", but the historical pattern Cv"Cv" is realised as ctflCv". (138) nyama "meat" nyumbá "house" nyóká "snake" mamphémblá "corn" A general principle in KiShambaa is that whenever two

*nyama *nyumba *nyoka *mpemba

410

AFRICAN LINGUISTICS

separate H tone autosegments come together, they are sep­ arated by a downstep (so that synchronically, any combina­ tion of H+H always yields H'.H; HH can only result on the surface from application of the Tone Spreading rule). We therefore have some evidence how Tone Spreading applied in Bantu; it must have been a rule to associate a H tone with vowels to the right, and was not a rule changing a L tone to a H tone after a H tone. If Tone Spreading were to change a L tone feature to a H tone feature, the his­ torical and underlying forms Cv"CV and Cv"Cv" would incor­ rectly merge into the pattern *Cv"!Cv\ So, the shift from a tonal analysis (as in Pare) to an accentual analysis (as in Kimatuumbi) can be more likely attributed to the Tone Spreading rule; once that rule is added to the grammar (and, assuming that the lan­ guage is not like Shambaa in maintaining a phonetic dis­ tinction between spread versus separate H tones), for the vast majority of cases in the language, there is no ar­ gument against an accentual analysis. It then seems that the first stage towards accentual reanalysis is making such an analysis possible by eliminating the HH/HL con­ trast, through the application of Spreading. It is of course also possible that an accentual analysis might be brought about in a language by means other than Tone Spreading; indeed, the addition of Meeussen's rule to the grammar could also provide the necessary neutralisation of Proto-Bantu CvXZtf and CvOV (or, perhaps CVCv" and CvOv") , so that there may be Bantu languages which could not have had Tone Spreading but which became accentual by addition of Meeussen's rule. In both cases, the crucial considera­ tion would be whether the languages could by the freeaccent hypothesis be analysed accentually. 5.3

Accent

and

Stress

Little has been said here about the relationship be­ tween accent and stress. Yet it is commonly assumed that there is some connection between the two. Of course, there is a physiological and acoustic connection between the two, in that stressed vowels and "accented" vowels21* involve higher values for F 0 . In addition, stressed vow­ els and (based on my instrumental work on Shona) H toned vowels may also involve greater amplitude and longer du­ ration. So there is indeed, at the physical level, an in­ timate connection between stress and accent; they are physically identical. It is in the area of linguistic functions that the two can be differentiated.

TONE IN KIMATUUMBI

411

One difference between accent and stress is that ac­ cent is generally unpredictable (or at least not trans­ parently predictable). For example, placement of accent in Kimatuumbi is totally unpredictable, whereas assign­ ment of stress in Kipare is regularly on the penultimate vowel. Presumably, if one were to encounter a language with higher pitch or extra length on a certain vowel or syllable, and that pitch or length were predictable by some (simple) rule, then one would call the phenomenon "stress". But if there simply were no way to predict which vowel or syllable takes the higher pitch or extra length, one might be tempted to call the phenomenon "ac­ cent" . Another parallel between "stress" and "accent" is that both have been viewed as the stronger member of an opposition (+ stress, + accent), in that "weakening" rules tend to apply only to weaker members of the oppo­ sition, and "strengthening" rules tend to apply only to the stronger members of the opposition. Thus, a stressed vowel may resist some reduction rule, and an accented tone may resist replacement by an adjacent tone. Inter­ estingly, it appears that accent generally is relevant only for tone rules, but not vowel-reduction rules, while stress is relevant for vowel-quality rules, but not for tone rules. The general problem in relating accent and stress is that it is hard to find both in one language. As noted before, the Bantu accentual languages Kimatuumbi, Haya and Tonga are the languages with no low-level penultimate stress. So, the interaction of "accent" and "stress" will be hard to test. One way to relate stress and accent would be to show in some language that stress "functions" like accent, in that a stressed (H) toned vowel resists some tone rule. There are such seeds of an accentual analysis in KiShambaa. A rule combines multiple adjacent H tones inside a verb stem (which like many Bantu languages, includes the object prefix) into a single H tone (associated with many vowels). This rule is manifested phonetically by the fact that underlying distinct H tones are exceptionally not separated by a downstep within a verb stem, as they are in all other parts of the grammar. However, there is a restriction on the rule merging the stem H tones. The rule does not merge a stressed (penultimate) H tone with a following H tone.

412

AFRICAN LINGUISTICS

(139) "I didn't kill"

"I didn't eat"

"I didn't kill it"

"I didn't eat it" The following rule is necessary to account for the merger of H tones in (139b (140)

So, stress in KisShambaa has taken on partially accent­ like characteristics. We may then suspect that KiShambaa is on the way to becoming an accentual language; in fact, given the free-accent hypothesis, all that remains for that change to be completed is for the stem-internal con­ trast HH ^HIH to be eliminated, and, indeed, that contrast has a rather low functional load in the grammar.

6.

SUMMARY

I have demonstrated that Kimatuumbi must be given an accentual analysis, at least for lexical tonal contrasts. It has been shown in the course of the discussion that the grammar is not purely accentual, but divides the accent  tone distinction along lexical  derived lines. It has,

TONE IN KIMATUUMBI

413

moreover, been shown that Kimatuumbi must be an accentual language, since it actually contrasts accented and unac­ cented H tones. Using the accentual analysis of Kimatuumbi along with data from other Bantu languages, I have also discussed various principles which could be employed to determine whether, for any arbitrary language, a language is "accen­ tual" versus tonal. It has been shown that the only cri­ terion which has not been falsified in the discussion is the "one-perimorpheme" criterion; I have thus proposed the "free^accent" hypothesis, which states that if a lan­ guage obeys the one-per-morpheme constraint, then the lan­ guage will be accentual.

FOOTNOTES

* I would like to thank Larry Hyman for reading and commenting on an earlier draft of this paper; he does not necessarily agree with the claims made here. Earlier versions of this paper were presented at the University of California at Berkeley, the West Coast Confer­ ence on Formal Linguistics, and the 13th African Linguistics Confer­ ence. Data for the paper were collected during 1978-81 from Emmanuel Manday, and was supported financially by the Research Board and the African Studies Center at the University of Illinois, NSF Grant BNS7924523, and Dean Lois DeFleur at Washington State University. 1. This position is taken in Haraguchi (1977), where he suggests that the accent mark * is an abbreviation of the feature [accent]). 2. McCawley's analysis in fact is framed in terms of identifying the location of pitch-drop in a Japanese word, which is essentially iden­ tical to identifying the position of the last H tone. The phonetic realisation of tone in a pitch-accent language should not be confused with the abstract analysis of tonal restrictions in a language, stated in terms of the location of an accent mark. 3. In Japanese, only one accent may occur per word. But in many Ban­ tu languages, such as Tonga, more than one accent can appear in a word, especially in verbs, where two accented morphemes may appear in a word. So, while some languages also have word-level restrictions on

414

AFRICAN LINGUISTICS

multiple accent, others have restrictions only on morpheme-internal accents. 4. Hyman (1982a) presents an accentual analysis of Luganda, in­ cluding the verbal system. However, in Hyman (1982b) it is suggested that an accentual analysis is not necessary, and can be replaced with an analysis using only underlying H tones (and no L tones). Re­ gardless of the accent-versus-tone problems discussed in these works, technical details of the analysis of Luganda verbs still need to be ironed out, if only for the sake of completeness. So, for example, I see no way for Hyman's account to predict the final nonfailing H tone in the Luganda form gukweeka, since as I understand Hyman's analysis, every final H tone must fall (this would be expected since in Hyman's analysis, the L tone of the melody HL is associated with the last accent). Similarly, the forms bágúléetá and bágukweeká seem to cause problems for Hyman's analysis, since L tones between H tones inside a word should be impossible. The only further mechanism available, which might account for the tone pattern HL*H inside of words is a "list-" intonational rule raising a final sequence of L tones. This rule might be the source of the final sequence of H tones, although there is no evidence from Stevick (1969) that this is true. 5. The hypothetical forms with two H tones cannot even be derived by applying later rules such as Tone Copy or Post Prefix High Tone Assignment. 6. There are other alternatives, which have no advantages or dis­ advantages ,  far ar I can see. For example, one could assume the tone melody LHL, map the starred H to the starred vowel, and asso­ ciate L tones with the unstarred vowels. 7. Interestingly, for some nouns, the decision is inconsistent. Thus, some tokens of the utterance k¡sfg¡no chaángu "my elbow" have stem-initial H tone, and others have no H tone, viz. k¡s¡gino chaangu, 8. The final H tone derives by a sandhi rule, discussed in Odden (1981c).. 9. In the analysis proposed in Kisseberth and Odden (1980), the al­ ternation between penultimate rising tone and final H tone is ac­ counted for with a series of tone shift rules. As can be seen here, that analysis is not necessary if tone is treated accentually, and if Retraction is maintained in the grammar. The Retraction analysis was originally not applied to nouns such as mboopo, due to the exis­ tence of nouns which violate Retraction on the surface, such as ndoofía "lipring". It turns out, however, that such nouns belong to the initial accented class, so that the final H tone is not present when Retraction applies. 10. There is an alternative to this rule. One could assign the me­ lodic H tone and then associate that tone with the final vowel,

TONE IN KIMATUUMBI

415

giving one tone on two vowels. However, this approach would violate an otherwise well-supported convention of autosegmental phonology (Clements and Ford (1979) that associations between vowels and tones must be exhaustive, that is, they cannot skip over a vowel. Since the H tone of long accented penultimata is on the first vowel of that syllable, Tone Copy conceived as a rule drawing an association line would violate that convention. 11. This prefix has the underlying shape l i-, and undergoes a later rule lengthening that vowel, as discussed in Odden (1981c). 12. This will be the case in ndím¡ "tongues". I shall later demon­ strate that Tone Copy precedes Deaccentuation, which means that the derived final tone in this stem does in fact derive from application of Tone Copy. *  13. "Hyman (1982a,b) predicts that a language could contrast H, L 0 and L, but excludes this fourth possibility found in Kimatuumbi, H. 14. "Assertive" here is meant to characterise the class of tenses which are syntactically independent (not subordinate) but excludes the subjunctive, which can be independent. See Odden (1981a) for similar problems in Shona. 15. There is one problem with the application of Retraction in verbs, namely that it does not apply to a H tone after a long vowel when the H tone is in a nonfinal syllable, as in ybyyndáye "you should blunt". Retraction could be separated into two distinct rules, one applying to nouns, and the other to verbs, in a more restricted phonological environment. Or, the Retraction rule itself could be modified. One way to modify the Retraction rule (and thus retain a single rule for the single generalisation) would be to have the rule refer to both tone and accent, so that a H tone is retracted to a long vowel, except that is there is a syllable after the H tone, then that H tone must be accented. Thus, Retraction could be modified as:

16. The H tone is retracted from the final vowel to the penultimate by a rule discussed in Odden (1981b). 17. Interestingly, in Shona (Odden (1981a), subordinate tone mapping is assigned to the leftmost stem-final vowel possible, so that in a reduplicated stem, the stem melody appears entirely in the leftmost stem, and is assigned to the rightmost stem only when the leftmost stem is "filled up". In contrast, as argued in M. Odden (in prog­ ress) , based on tonal evidence, the leftmost stem in HiBena must be treated as an "infix", and the rightmost stem is the "basic" stem,

416

AFRICAN LINGUISTICS

since tones are assigned entirely to the final stem, not the initial stem. It is therefore apparent that reduplicated stems in Bantu lan­ guages cannot always be treated as tonoloaicallv eauivalent to longer nonreduplicated stems. 18. And nitémya cannot be treated like an unaccented noun, i.e. taking its H tone by HTA, since in medial position, the H tone of the verb is kept, viz. njtemya m}koongo "to chop trees for me". 19. Hyman (1982a:43) asserts that accent is a property of the mora, claiming that "The easiest way is to say that *, like tones, must be a feature on moras. (If it were a feature on rimes or syllables, we would have a stress language, not a tonal accent one, and presumably we wou,ld represent the prosodie units in terms of s's and wl s rather than 's and 0 , s ) . " The question whether we use s's and w1s versus accents seems to me to be an independent proposition; I cannot see why representing accent at the syllable level means that a language has "stress". 20. One might approach these data by restricting assignment of the second H tone to final position. Recall the Initial Tone Insertion rule (119) which puts a H tone on certain morphemes, including pa-, if the preceding word has H tone only on the first stem mora. If H tone were not first assigned to the final vowel here, incorrect *([langa pámwoótó would result, cf. η¡[ka langa pámwooto "I'm cooking at the fire" where underlying η\k\kálaanga is assigned only one tone, by (88). Note also in (124) that the same subordinate Η tone is retained medially when the first Η is deleted by Perfective Tone Erasure (104). 21. I use the feature [+out-of-focus], in order to distinguish be­ tween these verbs, and verb-focused verbs or focus-neutral verbs, which do not undergo this rule. 22. Goldsmith (1982) suggests an ordering relationship between the rules mapping the tone melody onto stems and the application of vow­ el merger rules. He states (p.57) "Note that Vowel Merger applies before Tone Assignment, as we expect in an accent language. In a true tone language, such as Kikuyu, Tone Assignment occurs before rules like Vowel Merger". No reason is given why we should expect there to be such an ordering relationship between Tone Assignment and Vowel Merger rules correlated with accentual versus tonal sys­ tems. And, in Kimatuumbi, an accentual language, the Vowel Merger and Glide Formation rules (Odden (1981b)) must be applied after as­ sociation of Η tone with accent; thus, underlying ly-umo "story" becomes Iwuumo , while ly-aanjy "piece of firewood" becomes Iwáanjy. If Glide Formation preceded Tone Assignment, ly-umo would become *lwuumó. 23. A similar position is taken in Hyman (1982a), where he claims that the addition of a rule "like" Meeussen's rule is sufficient for a language to become accentual. 24. Accent is usually realised as "H tone", but cf. Goldsmith's analysis of Tonga, where the accented vowel takes a L tone.

417

TONE IN KIMATUUMBI

REFERENCES

Carter, Η. 1973 "Tonal Data in Comparative Bantu," African 14:36-52.

Language

Studies

Chafe, W. "Accent and related phenomena in the Five Nations Iroquois Languages," in L. Hyman, ed., Studies in Stress and Accent. SCOPIL vol. 4:169-182. Cheng, C.C. and C. Kisseberth 1981 "Ikorovere Makua Tonology (part 3)," Studies tic Sciences vol.11, no.1:181-202.

in the

Linguis­

Clements, G.N. and K. Ford 1979 "Kikuyu Tone Shift and its Synchronic Consequences," Lin­ guistic Inquiry 10:179-210. Goldsmith, J. 1976 Autosegmental Phonology. Ph.D. dissertation, M.I.T. Distrib­ uted by Indiana University Linguistics Club. 1981

"Towards an Autosegmental Theory of Accent: The Case of Ton­ ga" Distributed by Indiana University Linguistics Club.

1982

"Accent Systems," in H. van der Hulst and N. Smith, eds., The Structure of Phonological Representations (part 1) 47-64. Foris: Dordrecht.

Haraguchi, S. 1977 The Tone Pattern of Japanese: Tone. Tokyo: Kaitakusha.

An Autosegmental

Theory of

Haas, M. 1977 "Tonal Accent in Creek," in L. Hyman, ed., Studies and Accent, SCOPIL vol.4:195-208.

in

Stress

Hyman, L. 1978 "Tone and/or Accent," in D.J. Napoli, ed., Elements of Tone, Stress and Intonation. Washington D.C.: Georgetown Univer­ sity Press. 1981 "Tonal Accent in Somali," Studies in African Linguistics vol.12:169-203. 1982a "Globality and the Accentual Analysis of Luganda Tone" ms.,

use. f

1982b 'Against Asterisks in Luganda and Bantu Tonology: a reply to Hyman (1982a)". Ms., USC. Hyman, L. and E. Byarushengo

AFRICAN LINGUISTICS

418 1980

"Tonal Accent in Haya: An Autosegmental Approach" ms., Uni­ versity of Southern California.

1982

"A Model of Haya Tonology," to appear in G.N. Clements and J. Goldsmith, eds., Autosegmental Studies in Bantu Tonology. For i s : Dordrecht.

Kisseberth, C. 1981 "Displaced Tones in Digo,1' Studies in the Linguistic Sci­ ences vol.11, no.1:73-120. 1982 "Aspects of Makua Nominal Tonology," presented at the 13th Annual African Linguistics Conference, Université du Québec à Montréal. Kisseberth, C. and D. Odden 1980 "Aspects of Tone Assignment in Kimatuumbi," in Studies the

Linguistic

Sciences

in

vol.10, no.1:125-140.

McCawley, J. 1968 "Some Tonal Systems That Come Close To Being Pitch Accent But Don't Quite Make It" Chicago Linguistics Society 6:526532. 1978 "What is a Tone Language ?" in Tone: A Linguistic Survey, V. Fromkin, ed., p.113-132. New York: Academic Press. Odden, D. 1981a Problems

of

Tone Assignment

in Shona.

Ph.D. dissertation,

University of Illinois. 1981b "A nonlinear approach to vowel length in Kimatuumbi," pre­ sented at the 12th African Linguistics Conference. 1981c "Syntactic conditions on phonological rules in Kimatuumbi," read at the Annual Meeting of the LSA, New York. 1982a "Tonal Phenomena in Kishambaa," Studies in African Linguistics 13,2: 177-208. 1982b "Focusing constructions in Kimatuumbi," ms. 1982c "A metrical account of pitch levels" ms. Odden, M. in progress

Phonology

and Morphology

of

Hehe-Bena.

Schadeberg, T. 1973 "Kinga: A restricted tone system," Studies guistics, vol.4:23-48.

in African

Lin­

Spaandonck, van M. 1967

Morfotonologische

Analyse

in Bantutalen.

Leiden: Brill.

Stevick, E. 1969

"Tone in Bantu," International tics. 35:330-341.

Journal

of American

Stucky, S. 1979 "The interaction of tone and focus in Makua," Journal

Linguis­

of

419

TONE IN KIMATUUMBI

African

Languages and

Linguistics.

Voorhoeve, J. 1973 "Safwa as a restricted tone system," Studies guistics, vol.4:1-22.

in African

Lin­

A LEXICAL TREATMENT OF TONE IN TIV* D. Pulleyblank (M.I.T.)

0.

LEXICAL PHONOLOGY

That phonology and morphology are interdependent is beyond question - and this interdependence has been ex­ amined by theories of phonology, both generative and nongenerative. Many of the proposals that have been made, however, have been little more than ad hoc devices adopted as required by the facts, but not following in a principled way from the theory. For example, Chomsky and Halle (1968) proposed a set of segmentlike boundaries that were characterized by fea­ tures such as (¿segment) and (¿formative boundary). While these boundary features were inserted by convention in general morphologically-defined contexts, it was also possible to insert or delete such boundary features by language-specific rules just as rules may insert or delete oth­ er phonological features. The result of this approach was a powerful device that did not need to bear any strict relation to independently motivated morphological struc­ tures . But let us assume that any theory of morphology must be able to provide a bracketing of words into component morphemes. Pesetsky (1979) then proposes that the inter­ dependence of phonology and morphology can be explained in the following way: The morphological bracketing is relevant to phonology because there is a class of phono­ logical rules that apply after the addition of each of a wordf s component morphemes. To represent this graphically, I have shown in (1) that a stem enters the phonology, therefore undergoing any relevant phonological rules; the word either leaves

422

AFRICAN LINGUISTICS

the lexicon at that point1, or proceeds into the wordformation component to undergo affixation. The (stem, af­ fix) constituent is returned to the phonology where any relevant rules are applied; the derived word either leaves the lexicon, or returns to the word-formation component to underqo further affixation; etc. (11 LEXICON

As Pesetsky (1979) and others have pointed out, the model in (1) automatically encodes the phonological cy­ cle, since a rule will apply first to the innermost mor­ phological constituent, and then successively to each constituent created by affixation. When a word leaves the lexicon, it is inserted into a syntactic phrase. At this point phonological rules may again apply but illustrating quite a different set of properties. Following Mohanan (1982), I will refer to the operations applying in the lexicon as 'lexical' rules and to the operations applying at the sentence level as 'post ^lexical' rules.2 It has been proposed by Pesetsky and Mohanan that when a lexical item leaves the lexicon, its internal bracketing is unavailable to post-lexical rules.3 This means that a rule that crucially refers to notions such as 'stem', 'affix', 'compound', etc. must be a lexical rule, since the relevant structure will not be available after a word leaves the lexicon. On the other hand, a rule that applies across word boundaries must be postlexical since words are only concatenated at the point where post^lexically they are inserted into syntactic phrases. It has also been proposed within the approach out­ lined above that a rule muat be lexical it if has lexical exceptions, or is triggered by a morphological feature. Moreover it has been shown (Kiparsky 19 82) that rules in the lexicon may be subject to a number of con­ straints that do not hold of post-lexical rules. For ex­ ample, if a given language's syllabic inventory does not include syllables with a branching rime, then a lexical rule of vowel deletion in such a language will not be able to create a branching rime. On the other hand, postlexical rules may create a variety of syllables unat-

TONE IN TIV

423

tested in lexical representations. Following earlier work in cyclic phonology (e.g. Kiparsky 1974, Mascaro 1976, Halle 1978, Rubach 1981), it is assumed that non-structure-building lexical rules can apply only in environments 'derived' on the relevant cy­ cle. For example, a lexical rule of vowel deletion such as shown in (2), would not apply to the hypothetical form given in (3a) since the structural description of the rule is met in the underlying representation of the mor­ pheme. Rule (2) would apply to (3b), however, since the structural description of vowel deletion is derived on the second cycle. (See Kiparsky 1982 for discussion) 2. Vowel deletion: V -»· 0 / V 3. a. [kooba] -»» kooba [[bako] oba]->bakoba Note that rules that 'build' structure - such as rules of syllabification, foot-assignment and tone-as­ sociation - apply even in non-derived contexts (Kiparsky 1982, Harris 1982). A consequence of the above is that a rule such as (2) above must be post-lexical if it applies 'across-theboard', i.e. in both derived and non^-derived contexts. A further consequence of the lexical approach out­ lined above is that all rules determined to be lexical must precede all rules determined to be post-lexical. So, for example, a rule applying across words could never be ordered before a rule that had a morphologically-condi­ tioned environment. Below is a summary of the properties discussed above that distinguish between rulesfy applying lexically and rules applying post-lexically: (4) Lexical rules Post-lexical rules a. may refer to worda. may not refer to wordinternal structure internal structure b. may not apply ab. may apply across cross words words c. cyclic c. non-cyclic d. subject to lexical d. not subject to lexical constraints constraints e. apply in derived e. apply 'across-theenvironments only board' f. may have lexical f. may not have lexical exceptions exceptions

AFRICAN LINGUISTICS

424

g. must precede all g. must follow all lexical post-lexical rules rules The theory of lexical phonology sketched above makes the strong claim that phonological rules do not exhibit a random selection of the properties given in (4). On the contrary, these properties cluster together such that a rule applying lexically manifests those properties given in the first column, while a rule applying post-lexically manifests those properties given in the second column. This clustering of properties has a number of consequen­ ces. To cite a single example, consider the following im­ plication for language acquisition. A learner observes that a particular rule exhibits a property that identifies it as being lexical. The learn­ er immediately knows that the rule in question will ex­ hibit all the properties associated with lexical proces­ ses - and the learner knows this without requiring direct evidence for each individual property. In the following sections I will examine some of the consequences of the above theory for an account of tone, looking in particular at tonal phenomena in Tiv, a BenueCongo language spoken in Nigeria. 1.0

The

cycle

In a language such as Tiv, tone can constitute part or all of any given lexical entry. So consider what will happen when an underived lexical entry enters the lexical phonology (cf. diagram 1 above): If the association con­ ventions apply automatically at all stages of a deriva­ tion (Goldsmith 1976), then they will apply as soon as a lexical entry enters the phonological component of the lexicon. Moreover, given this assumption, the association conventions will re-apply on every subsequent cycle. In other words, by adopting the suggestion that the associa­ tion conventions apply automatically whenever they can, it follows that within lexical phonology lexical tone as­ sociation must be cyclic.5 Consider therefore a configuration such as the fol­ lowing : 6 (5) Assuming that the association conventions assign tones to tone-bearing units one-to-one from left to right, then a theory with non-cyclic tone association would pre­ dict the tonal pattern given in (6a) while a cyclic ap-

TONE IN TIV

425

proach would give the pattern in (6b). (6) a. b.

Notice that following Williams (1971) and Clements and Ford (1979) , I assume that more than one tone will not be assigned automatically to a single tone^bearing unit. Moreover I will ignore until section 2 of this paper the question of whether the last tone in (6) spreads to the final toneless vowels. Consider now the following forms from the General Past tense in Tiv, where again I postpone discussion of the tones on the non-initial stem vowels until the next section. (7). General past H^stem L-stem 1 syllable: 'va dzà came went 2 syllable: "ungwa vende heard refused 3 syllable: "yevese ngohoro7 fled accepted These forms are exactly what we would expect given: a. a lexical representation for Tiv verbs as in (8) b. a General Past prefix: [Lc. a representation of downstep/downdrift that can be informally characterized as: 0 -* 1 / L H d. cyclic tone association (8)/"' H- s tem verbs b. L-stem verbs

Given the lexical representations of (8) and a L- pre­ fix , we have exactly the configuration shown above as (5). Consequently, the predicted form for a stem such as yevese using a non-cyclic approach would be *yèvése, correspond­ ing to (6a). If association is cyclic, however, then we obtain the correct results for both  and L stems :

426

AFRICAN LINGUISTICS

(9),

a.

b.

Cycle 1: Stem; Assoc. Conv. Cycle 2 : Gen. Past

Let us review this approach. The lexical representation of verb stems is not an issue: It is a fact of Tiv that verb stems may be mono­ syllabic, disyllabic or trisyllabic and such stems are either  or L. Some might not think it as obvious however that the downstep characteristic of such a tense should be ana­ lysed as a floating L-tone; and yet is is essentially this assumption that forces tone association to be cy­ clic. I will therefore present in the following sub-sec­ tion arguments for viewing downstep in Tiv as a floating L.

1.1.1. Downstep To begin with, Clements and Ford (1979) correctly point out that a theory of tone that provides a single mechanism for describing downstep is more restrictive than a theory of tone allowing several alternatives for the analysis of downstep. Moreover, they argue strongly that downstep in Kikuyu should be analysed as a floating tone. They pro­ pose therefore that all examples of downstep entities be analysed as floating tones. This turns out, moreover, to be a position for which there is empirical evidence in Tiv. As a general rule in Tiv, Η-tones undergo 'register' lowering (downdrift) when preceded by L-toned vowels. Consider for example the pitch-level of gá 'not' in the following examples: (10)a. á vé gá 'he did not come (recently)' b. á d z a g á 'he did not go' In (10a) gá is on the same pitch-level as á, but in (10b), gá is lower than á because of the intervening verb dzà . Looking next at the interval between  and a fol­ lowing !, we observe the same interval between the two Η-tones as in a HLH sequence where the intervening L is linked. Compare for example (11a) and (lib), where (lib) represents the General Past form observed in (7).

TONE IN TIV

(11)

427

a·

á v é ' h e came (recently)' á*va 'he came' Analysing downstep as a floating L-tone correctly predicts that the extent of register lowering will be the same whether the trigger is a downstep entity or a linked L-tone. Another argument comes from Arnott (1964, 1968) where it is shown that downsteps in Tiv alternate with Ltones under certain circumstances. Consider the following examples : (12) a. ùnyïnyà mbâ 'there are horses' horses copula b. mbá* van 'they are coming' cop. coming (13) a. b.

kásév women kásév women

mbâ copula mbá* gá cop. NEG

'there are women' 'there are not any women'

(14) a.

íwá ngî 'there are dogs' dogs copula b. íwá ngf* yevèsè 'the dogs are fleeing' dogs cop. fleeing The copulas in (12-14) have a falling tone on a short vowel pre-pausally but occur as  I in non-pausal position. And this tonal behaviour is not restricted to any particular class of lexical items; Arnott gives ex­ amples with copulas, verbs (Habitual 2) and nouns. If floating L-tones trigger downstep then this al­ ternation is easily explained by positing a rule that creates contours by linking a final floating tone to the last vowel of a word in pre-pausal position. (15) T-attachment:

On the other hand, if downstep is not analysed as being a floating L-tone, then there is no obvious expla­ nation for alternations such as those seen in (12-14). In order to make two final arguments for repre­ senting downstep as a floating L in Tiv, I will side­ track to motivate a general rule of H-spread.

428

AFRICAN LINGUISTICS

1.1.1. H-spread Consider pairs of nouns such as those given in (16): (16) stem singular plural a. daka dakà ùdakà 'type of gun1 b. dan darf ùdàr' 'half-penny' We observe that in the singular such nouns do not have a noun class prefix, while in the plural there is a L-tone u-prefix. These examples are included merely to illustrate the rather bland observation that the addition of a L-tone prefix to a noun has no effect on the tonal pattern of a stem. In contrast, consider below what happens when it is a Η-tone prefix that is added to a noun stem:8 (17) stem singular plural L

\

'

a.

bagu

bàgù

íbá'gij

b. c. d. e. 4= f.

kagh kegn gbise kaande uH mbor

Ikàgn Ikegñ* )g bisé )kaande * u '* imbor

ígá'gfi íke*gfi* ágbí'sá ákáande ' u'' ambor

'red

monkey'

'bundle' 'chicken' 'type of tuber' 'type of shellfish' . . ,'spring of water * We observe that the initial L of a stem such as bagu or gbise is displaced by the  of the class prefix. The L that has been so displaced will trigger