Current Approaches to African Linguistics: Vol 1 Vol. 1 [Reprint 2020 ed.] 9783112420065, 9783112420058

Citation preview

Current Approaches to African Linguistics (vol. 1)

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

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

Ivan R. Dihoff (ed.)

V 1983 FORIS PUBLICATIONS Dordrecht - Holland/Cinnaminson - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.S.A. and Canada: Foris Publications U.S.A. P.O. Box C-50 Cinnaminson N.J. 08077 U.S.A.

ISBN 90 70176 68 8 (Bound) ISBN 90 70176 57 2 (Paper) ® 1983 Foris Publications - Dordrecht. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in the Netherlands by ICG Printing, Dordrecht.

Preface

This volume contains a selection of papers read at the 11th Annual Conference on A f r i c a n Linguistics w h i c h took place o n April 10, 11, and 12, 1980 at Boston University. Thanks are due to George N. Clements, Mary M. Clark, Edward A l l e n and Harold Fleming for their assistance, to Vincent Kowal for help in editorial matters, to M a r k Dyer for his assistance in organizing the conference, and to B o s t o n University for a grant in support of it. Ivan R. Dihoff Y a l e University

Table of Contents

PREFACE I. A R E A L A N D HISTORICAL LINGUISTICS 1. Some A r e a l Characteristics of A f r i c a n Languages

1 3

Joseph H. Greenberg 2. Adamawa-Eastern: Problems and Prospects

23

Patrick R. Bennett 3. The Eastern Jebel Languages

49

M. Lionel Bender 4. N o u n Class Prefixes in Proto-Edoid

59

B e n Elugbe 5. Ijo's Closest Linguistic Relative

85

Charles E. W. Jenewari, Defaka II. PHONETICS A N D PHONOLOGY

113

6. The Phonetics of Zu/'hoasi, a K h o i s a n Dialect of South West Africa/ Namibia J. W. Snyman 7. Nasal Vowels and Nasalized Consonants in Gbe

127

Hounkpatin C. Capo 8. The Hierarchical Representation of Tone Features

145

G. N. Clements 9. Tone in the Verbal System of A n y i

177

W i l l i a m R. L e b e n 10. Tone Rules and Derivational History in Edo Phonology A i r e n Amayo

185

11. The Analysis of Tone in Acholi, Luo and Lango

197

David Dwyer 12. Ewe and the Theory of Tone Spreading

213

Mary M. Clark 13. Accent in Tonga: an Autosegmental Account

227

John Goldsmith 14. The Asante Twi Tone Shift John M.

235

Stewart

III. LEXICAL STRUCTURE

245

15. Manding Lexical Behavior in Sierra Leone Krio

247

Ian F. Hancock 16. The Ideophone as a Phonosemantic Class: the Case of Y o r u b a

. . . .

263

James F. Fordyce 17. A Lexical Treatment of Cairene A r a b i c Object Clitics

279

Ellen Broselow 18. A Prosodic Account of Arabic B r o k e n Plurals

289

John J. McCarthy IV, SYNTAX

321

19. Anaphora, Cataphora, and Topic Focusing: Functions of the Object Prefix in Swahili

322

Keith A l l a n 20. The Complex Structure Conspiracy and the Grammar of Mandingo Complementation

337

Mallafe Drame List of Contributors

359

I. Areal and Historical Linguistics

Chapter I

Some Areal Characteristics of African Languages Joseph H. Greenberg

1. General surveys of African languages often list a series of traits which are said to be characteristic of Subsaharan Africa as a linguistic area.' These enumerations tend to be fairly similar and to include: 1. Phonological items such as tone, click sounds, implosives, prenasalized stops, labiovelar stops and the prevalence of open syllables, 2. Morphological traits e.g., noun class systems and complex verb derivational systems, 3. Syntactic phenomena such as the use of a verb 'to surpass' to express comparison, serial verb constructions and what Welmers call semantic ranges. These include idioms such as 'fruit' = 'child of the tree', the use of body part terms as prepositions and the existence of a verb whose primary meaning is 'to eat1 but includes a wide range of further meanings such as 'to conquer', 'to win an adversary's piece in a game1 and 'to have sexual intercourse'. Of these characteristics, many are in fact found extensively in other major world areas. These include tone, a preference for open syllables and the use of body part terras as prepositions. A few, however, are confined to Africa but are not always of wide distribution within it, for example, 2 click consonants.

Others are common in Africa and are found elsewhere,

but only infrequently. An example is labiovelars, both stops and nasals. Ideally, if what is meant by an Africa areal characteristic is one which is found everywhere in Africa but nowhere else, then clearly none exists. For present purposes then, we will define areal properties in less stringent terms, as those which are either exclusive to Africa, though not found everywhere within it, or those which are especially common in Africa although not confined to that continent. The justification for these two definitions is that they give rise to similar questions which we may say are of an areal type. What is the actual distribution geographically within Africa of each such trait? How can these distributions be explained historically? Do diverse areal characteristics show at least roughly similar distributions so that an areal classification of African languages themselves becomes feasible? Are the traits which seem

4

Joseph H. Greenberg

most particularly African on a worldwide basis concentrated within certain areas within Africa itself? In an article published in 1959, the present writer conjectured that this was so. It was stated in the following terms. "It is noticeable that various sections of the continent differ in the intensity with which they partake of such common characteristics. There is a large central area in which all of these characteristics are found in most languages. This core consists mainly of the Niger-Congo languages, Songhai, the Central Sudanic subgroup of Macro-Sudanic, and, to a certain degree, the Chad subgroup of AfroAsiatic" (1959, p. 24). Although since that time there have been a number of further listings accompanied sometimes by brief discussions in general works, outside of Heine's work on word order in African languages, to my knowledge no attempt has been made to map areal traits systematically or account for their distribution . As a set of premliminary studies in this direction, I will consider here four traits in regard to their areal and genetic distributions in Africa. Since the number of traits is limited and the survey of the data is far from complete, any conclusions regarding them will necessarily be highly tentative and largely in the hope of stimulating further discussion and research in a rather neglected but important set of topics in African linguistic studies. In the theoretic discussion, I will be interested particularly in the resolution of what might be called the isolation-contact paradox. Given the existence of certain linguistic phenomena over vast and often quite continuous areas, one is tempted to explain them in terms of equally extensive linguistic contacts. If so, how can such explanations be reconciled with the enormous linguistic diversity and genetic differation which is also characteristic of Africa from the linguistic point of view? Of the four topics considered here, two are phonological, namely labiovelar stops and labiodental flaps. The third is the use of a verb meaning 'to surpass' to express comparison. The fourth is a problem of semantic range, the use of a single term meaning both 'meat' and 'animal'. 2. The labiovelar articulations with which we will be concerned are mainly coarticulated stops. Most commonly languages have either an unvoiced or a voiced stop or both. The corresponding nasal is far less common and is almost always found in languages which have at least one non-nasal stop. The labiovelar stop has occasionally been reported as implosive. Labiovelar fricatives also occur, though only rarely.

Areal Characteristics of African Languages The coarticulated stops and the corresponding nasals are sometimes said to be West African but are, in fact, more extensively distributed. The most detailed statement that I have found is that of Welmers: "There is an interesting geographical distribution of doubly articulated stops in Africa. They occur primarily in languages grouped in a strip from the Atlantic into the Central African Republic, across the West African bulge and somewhat farther east. A number of the coastal languages do not have double stops, and to the north they extend only I.e. irregularly beyond the forest into the grassland. Thus northern Bambara do not have either /kp/ or /gb/, but southern Bambara and Maninka have /gb/. Senari (a Southern Senufo language) has both, but Suppire (a very closely related language to the north) has neither; Senari /kp/ corresponds to /b/ or /bb/ in Suppire, while Senari /gb/ is of secondary origin". (1973, pp. 47-8) . There is one area outside of Africa in which these sounds are found. The Kate-Ono group of non-Austronesian (Indo-Pacific) languages in northeastern New Guinea and some Austronesian languages of Melanesia have these sounds.^ The following survey of the distribution of labiovelar stops in Africa is in certain respects incomplete or perhaps even inaccurate. My own review of the literature is far from exhaustive. In addition there are certain sources of inaccuracy or incompleteness which are inherent in the literature itself. Since, as was noted by Welmers, even dialects of the same language may differ on this point, it is not safe to conclude from their absence or presence in a description to a corresponding situation in dialects other than those on which the description is based. Indeed, it is often not possible to deduce which dialect or dialects are involved from the description itself. Moreover, for many languages, we still have only word lists. In the absence of an accompanying systematic phonology, the non-occurence of these sounds cannot with certainty be deduced from their absence in a specific list, only their existence from their occurence. Nevertheless certain general conclusions can be drawn. As noted by Welmers, these sounds are found in a largely continuous but at certain points fairly narrow east-west band starting with the Atlantic and ending roughly at the Nile-Congo divide (Map 1). Of the four major stocks in Africa, it is lacking entirely only in Khoisan. There are instances of continous distributions involving several major stocks or subgroups. An example is the eastern section in which there is a solid block embracing languages of the Adamawa-Eastern branch of Niger-Congo, northeastern Bantu languages. Central

5

6

Joseph H. Greenberg

Sudanic languages of the Nilo-Saharan family and finally a few languages of the Eastern Sudanic branch of the same stock (Alur and Kakwa). The chief pockets of non-occurrence within the large zone of almost continuous distribution, outside of numerous areas of Fula settlement, are Twi, but not the rest of Akan, most Chadic languages of the Jos Plateau and Bauchi in Nigeria and, finally, a very considerable area farther east in Nigeria and Cameroun consisting of some Bantu languages of zone A (e.g. Duala), a large number of Grasslands Wide Bantu languages (e.g. Ngwe, Bamun) and the Bantoid Mambila and Vute. Whether this corridor is itself continuous I cannot determine because of lack of information concerning certain Grassland languages. The opposite of such pockets of non-occurrence is the existence of islands of labiovelar speech outside of the main area. The most important of these from the historical point of view is Katla, the only Kordofanian language with labiovelars, at a distance hundreds of miles from the nearest occurrence elsewhere. If we consider the distribution just sketched in closer detail it is clear that in spite of.its appearance in three major stocks, it is a predominantly Niger-Kordofanian phenomenon both in regard to territorial extent and more importantly breadth of occurrence in relation to subgroupings. These sounds appear in every substock of Niger-Congo and also, as has been seen, one Kordofanian language. It has a substantial or even dominant distribution within each subgroup. These facts may be summarized as follows: In the West Atlantic group, the entire southern subgroup has labiovelars. It is likewise found in Bijago of the Bissao Islands which should probably be assigned separate genetic status within West Atlantic. In the northern subgroup of West Atlantic on the other hand, it is mostly lacking. It seems to be found only in Balante and Nalu which are closest geographically to the Southern subgroup. In the Mande languages, the entire southern subdivision of the western languages (Mende, Kpelle, etc.) has the sounds under discussion. The northern subdivision of the western languages likewise has labiovelars except for Soninke and the northern dialects of the Malinke-Bambara-Dyula complex. Further, all of the Eastern Mande languages (e.g. Mano, Busa) have these sounds except Bisa and Samo, both of which are well within the northern non-labiovelar zone. Most Voltaic languages have labiovelars so that it will be easier to

Areal Characteristics of African Languages list the exceptions. These include a few Gurunsi languages (Kassena, Lyele, Nunuma), Mossi, Northern Senufo and Dogon. Here once more it is the northern languages which lack labiovelars. The Kwa languages will be treated here separately although it is generally agreed that the boundary between Kwa and Benue-Congo is an artificial one. It appears likely that all Kwa languages except Twi and Igbira possess labiovelars. In Benue-Congo, all the Plateau languages except Kambari, all the CrossRiver languages, and all the Jukunoid languages possess this feature. Among non-Bantu Bantoid languages it is lacking only in Mambila and Vute. In Wide Bantu, it is found in Ekoi, Nyang, Mbe and some Mbam-Nkam languages. In Narrow Bantu it is, of course, generally absent. However, there are a substantial number of languages in Guthrie's zone A in the northwest which possess labiovelars e.g. Yaunde, Bafia. It is also found in some languages in zone C and many in zone D (e.g. Kibira) in northeastern Zaire. In Adamawa-Eastern, all Adamawa languages except the Nielim-Bua-Koke subgroup and possibly Tula along with Dadiya, which are in another subgroup, have labiovelar stops. These two subgroups are both north of the main group of Adamawa languages. It seems that all Eastern languages have these sounds without any exceptions. It has already been noted that one Kordofanian language, Katla, has labiovelars. In Nilo-Saharan the picture is very different. Labiovelars are dominant in only one subgroup. Central Sudanic, where it is found everywhere except 4 the Sara-Bagirmi languages in the west.

These latter are well within the

main non-labiovelar area. The only other occurrences within Nilo-Saharan are in two languages of the Eastern Sudanic subfamily. One of these is Alur, a Western Nilotic language situated between Logbara and Lendu, both Central Sudanic languages with labiovelars. The other is Kakwa, an Eastern Nilotic language of the Bari cluster, in contact with Logo and Madi, both Central Sudanic languages. In the Afroasiatic family, labiovelars are found only in a few Chadic languages. These include Gwandara, the closest relative to Hausa, the Ron language of the Jos Plateau, Gerka, the Basharawa dialects in the vicinity of Kanam, some languages of the Bata group, and Kotoko. We should also add that one Indo-European language, Krio, the Englishbased Creole of Freetown, has labiovelars. On the other hand, the Portuguese-based Creole, at least as spoken in the Casamance area of Senegal,

7

8

Joseph H. Greenberg

does not. In both of these instances we have conformity with the general distribution of these sounds in West Africa. The preceding review of the evidence should be sufficient to show that coarticulated labiovelars are basically a Niger-Congo feature. I propose that at least one labiovelar ought to be reconstructed for Proto-NigerKordofanian. Regarding Niger-Congo there are a number of widespread etymologies e.g. kpa 'kill', 'die', involving labiovelars. In a number of reconstructions of subgroups, one or more of these sounds have been posited as original for the intermediate proto-language, e.g. by Manessy for Voltaic and Wolf for Benue-Congo. With reference to Niger-Kordofanian there seem to be a number of etymologies in which various branches of Kordofanian generally show b corresponding to Niger-Congo labiovelars.^ The occurrence of a labiovelar sound in Katla would then be a lone survival within Kordofanian. Unfortunately there are no Katla cognates for the Niger-Kordofanian etymologies just referred to. Thus its appearance in a lone language in Kordofanian in geographical isolation from the main labiovelar area might be explained genetically as a survival. The sounds under discussion are surely not to be posited for proto-NiloSaharan although they may well be proto-Central-Sudanic. In Thayer's comparative study based on Bongo, Kresh and Sara, she reconstructs a single 7 labiovelar for the ancestral language. In that case their non-appearance in Sara-Bagirmi would be a secondary loss. For Chadic and Indo-European we are dealing with what are obviously instances of recent contact. We may conjecture, then, the basic course of events in something like the following terms. Labiovelars in Africa originated in the Niger-Kordofanian languages in which, however, they were lost almost everywhere in Kordofanian and in Niger-Congo across a broad band north of the forest belt in West Africa as well as sporadically elsewhere. Their absence in Proto-Bantu had as a consequence that they are largely lacking in the southern third of Africa. From Niger-Congo these sounds spread at an early date to Central Sudanic in the Nilo-Saharan family probably from Adamawa-Eastern. This resulted in a solid bloc of such languages from which a further diffusion from one or another or both led to the appearance of labiovelars in Bantu languages and a few Eastern Sudanic languages. We are referring here in the case of Bantu to zones C and D where they are probably secondary. In Zone A further investigation is required. Considering the separate genetic

Areal Characteristics of African Languages status of this zone within Bantu it is possible that it was inherited here. Elsewhere, in Chadic and in Krio, it is again a phenomenon based on contact with Niger-Congo languages. If the basic theory just outlined is accepted, then insofar as NigerKordofanian in concerned, the problem is in general to account not for its presence but its loss. The mystery is why it disappears over an essentially contiguous area in the northern part of the West African bulge. This would be less difficult to explain if it always changed to the same sounds. We are familiar with instances of the propagation of sound changes across dialect and even language boundaries e.g. the High German consonant shift. But in the present instance, the changes are various except, apparently, for the Kordofanian change to b. The coarticulated stops change most commonly to labialized velars, velars, or labials and often to more than one of these in the same language depending on adjacent, usually following vowels. In reconstructing at least one labiovelar stop for Proto-Niger-Kordofanian, a theory is proposed which is incompatible with that of Westermann, who assumed that labiovelar stops in his Western Sudanic languages were not original but arose everywhere from labialized velars, i.e., sounds of W 8 the K

type.

I believe, in this instance, like his parallel theory regard-

ing the independent origin of noun classes in various West Sudanic languages from originally classless languages, that we are dealing with an inherently implausible hypothesis, namely that a sound which is unusual on a worldwide basis should arise again and again independently in the same area. Particularly in view of its occurrence in Kordofanian, it becomes more plausible to look for a single genetic origin followed by widespread loss and occasional spread by contact. Nevertheless, that labiovelar stops can arise from labialized velars will be indicated in later discussion. One kind of evidence that the movement is rather from coarticulated stops to labialized velars than vice versa will be direct historical attestation, but of course in Africa this is usually lacking. Our one major source is Koelle's Polyglotta Afvicana, and there are a few instances in which he recorded labiovelar stops in Niger-Congo languages which have since been replaced by other sounds. There are, apparently, no contrary instances. One example is Bambara. Koelle states explicitly that his informant was from Segou, and he records gbe 'white1, gbasi 'flog', and gbalo 'skin'. Musa Travele from the same city, in his dictionary published in 1913, gives the corresponding words as g'e, gosi, and gold with a variant wolo.

9

10

Joseph H. Greenberg

This last form suggests a labialized velar as a possible intermediate stage, at least before back vowels. Igbira (now called Ebira in Dalby's atlas) is a language of the Nupe group. Koelle's Egbira-Hima has four words with gb. Ladefoged in his general work on the phonetics of West African languages records its sound sys9 tem as lacking labiovelars, as does Wolff. A word list recorded by the former in the Journal of West African languages gives two of the four words found in Koelle, zpa 'root' for Koelle's egba and dbz 'neck' for Koelle's ogbe.It

is of interest to note that Ebira is spoken at the confluence

of the Niger and Benue, completely surrounded by languages with labiovelar stops, so that this change cannot be attributed to language contact. A third instance from Koelle is Vute for which he records labiovelar stops, whereas Hofmeister in his work on that language does not record any and gives a table of sounds in which it is lacking. Unfortunately, none of the specific words recorded by Koelle as containing these sounds is found in Hofmeister.* * As we have seen, however, there is evidence for the acquisition of these sounds by contact with languages that have them, the most obvious mechanism being, of course, loan words. Much more remarkable are a few instances in which what may be called convergent sound change occurs. In all of these instances it is labialized velars which are replaced by stops in indigenous words. One instance is the Chadic language Gwandara, closely related to Hausa but spoken farther south and in contact with Gade and Koro, BenueCongo languages with labiovelar stops. An example is Hausa tagwaye 'twins' for which the Karshe and Kyankyan dialects of Gwandara both have tegbe. 12

Another dialect Nimbia, however, has tagusy. A second instance of this process is the Zugu dialect of Zaberma, of the Songhai subgroup of Nilo-Saharan. Zugu is in Benin and in close contact with Voltaic languages with labiovelar stops. Among the examples cited by Funke are kpe 'king' corresponding to koi, and the identical forms meaning 'to go' and kpili 'blood' compared to kuli and kpara 'city' to kwava. The latter forms of each pair are found in other dialects of Zaberma.^ A third example is mentioned by Westermann and Ward in Kakwa, a Western Nilotic language closely related to Bari. The instances 14cited are Kakwa kpen 'bird' = Bari kweni and irynan 'four' = Bari inwan. Except for some of the Zugu examples which simply have ku in related dialects, one can conjecture a substitution mechanism by bilingual speakers who might replace the labialized velars with stops and from whom it might then spread to

11

Areal Characteristics of African Languages monolingual speakers. Much more difficult to interpret is the presence of labiovelar stops in some Chadic languages of the Kotoko-Logone group. They occur frequently in Gulfei and Affade, infrequently in Makari, but not at all in the closely related Logone.^ As in the previously discussed instances they have developed through internal change from labialized velars or from ku or gu (cf. Zugu). Those languages are not in contact with any languages with labiovelars. 3. The second areal property of African languages which will be discussed is the labiodental flap. In contrast to the labiovelar stops that we have been discussing which, as has been noted, are also found in the Pacific, these sounds are, as far as present knowledge extends, found exclusively in Subsaharan Africa.^ Moreover it is a far more infrequent sound. In spite of its rarity, it has, like labiovelars, been reported as occurring in Niger-Congo, Nilo-Saharan and Afroasiatic languages. The largest concentration of examples appears to be in the Central Sudanic branch of Nilo-Saharan. (Map 2). Tucker and Bryan report its occurrence in Mangbetu, Asua, Mamvu and Lese, all within one branch of Central Sudanic. It is also stated, in the same source, to be very common in Kresh and to occur spora17 dically in Baka and Morokodo which are likewise Central Sudanic languages. In all of these languages it occurs in common words. It is not reported elsewhere in Nilo-Saharan. In Niget-Kordofanian almost all known examples are from Adamawa-Eastern. Once more Tucker and Bryan describe it as most common in the Ndogo-Sere group, examples being given from Sere, Mundu, Ndogo and Bai and once more in common words. Particularly interesting is its occurrence in Ndogo, Sere and Bai in the word of 'child' v i an obvious cognate of the widespread Niger-Congo root bi- Richardson gives an example from an Adamawa language Kapere, a southern dialect of Mbum in which it occurs in the second person plural pronoun v-i. He also notes its occurrence in Ngbaka Mabo, a form of Mbaka Limba, and in the Yangere dialect of Banda, both18of these once more languages of the Eastern division of Adamawa-Eastern. The one instance of this sound being reported from a Niger-Congo language which is not a member of the Adamawa-Eastern subgroup is indeed a startling one. This is its occurrence far to the south in Shona, a Bantu language group in Zimbabwe. According to Fortune it occurs only in ideophones and rarely even in those. In Hannan's Standard Shona Dictionary an example is cited

12

Joseph H. Greenberg

from the Zezuru dialect where it expresses the "idea of flicking an object".19 The only other occurrences are in Chadic languages. In my own fieldwork I encountered it in Gerka in the word for 'buffalo'. The same example is reported in print by Jungraithmayr as kavin, the reflex of a common Chadic 20 root.

Its existence is also attested in two further Chadic languages,

Tera (Newman) and Margi (Hoffmann). In both of these languages it only occurs in ideophones. As can be seen from the foregoing exposition, the labiodental flap bears in its distribution an important though only partial resemblance to that of labiovelar stops. In Nilo-Saharan it is once more concentrated, this time exclusively, in Central Sudanic. Once more it shares a trait, most probably through contact, with the Adamawa-Eastern branch of Niger-Congo. However, in this case it appears absent from the rest of Niger-Congo except for the lone instance of Shona. Once more, also, its occurrence in Afroasiatic is confined to the Chadic branch. Since it sometimes occurs only in ideophones which are likely to be omitted or poorly reported in grammars and dictionaries, one may suspect that fuller information in the future may fill in some of the gaps in its apparently sporadic distribution. If it does have a single origin, as seems likely, the choice would seem to be between Central Sudanic and Adamawa-Eastern as the ultimate source, without a convincing basis for choice. 4. For our third topic we turn to the expression of the comparative, and usually the superlative also, by means of a construction using a verb meaning 'to surpass'. In many of the languages for which it is described in Africa, it coexists with other constructions but in these cases it is usually the dominant method and often it is the only one mentioned. There cure two chief variants which may be literally translated as follows: 1. 'X is large, surpasses Y'; 2. 'X surpasses Y (in) largeness'. In the latter type we most often have an abstract noun or an infinitive. The superlative is usually expressed by a parallel construction in which Y is substituted by 'all' e.g. for the first variant 'X is large, surpasses all'. Sometimes the comparative occurs in this construction without the superlative or vice versa. Often our information is incomplete. Clearly this is a very widespread construction in Sub-Saharan Africa, more extensive in fact in its distribution than labiovelar stops. (Map 3). Although many languages of the world can express comparison by such locutions as 'she surpasses me in intelligence', it is clear that in English and

Areal Characteristics of African Languages indeed almost everywhere outside of Africa this in only an occasional method and not the basic way of expressing comparisons. The only other linguistic area in which I have encountered this as a usual construction is that of Native American languages and here it is not common. An example 21

is the Zamuco language of Bolivia. It is much more difficult to obtain an accurate notion of the distribution of this feature in Africa and to explain this distribution historically than in the case of the phonological features we have just been discussing. The first problem simply arises from the lack of relevant published data. Even a short word list will frequently indicate the presence of coarticulated labiovelars, even though its absence, without a more formal phonological description, is harder to establish. For the comparative construction we need grammars and lexicons. Even these are frequently of no assistance, particularly, it should be noted, if written in accordance with recent models. A traditional grammar organized according to parts of speech will almost always have a section on comparison, since Classical times viewed as an "accident" of the adjective. As for lexicons, even if there is an entry meaning 'to surpass1, unless it is a very full dictionary with examples, it will often be of no assistance. Even if the word is present and the examples do not express comparison, this is of course not decisive regarding the absence of the construction. In regard to historical interpretation of the data, there are certain difficulties inherent in the topic itself as compared with phonology. A construction of this kind can spread from language to language by loan translation, or caique as it is sometimes called. Here, unlike the situation in phonology, such as presence in loan words only, which can often tell us the direction of transmission, there is no corresponding method. Further, it is far more difficult to distinguish between genetic persistence and language contact. This is because it may represent the continuation of an inherited construction, but with lexical replacement of the key term 'to surpass', not a highly stable item. With all of these difficulties it is still possible to describe in broad terms its distribution in Africa and to attempt a historical explanation. The construction we are discussing is found in all four major stocks of Africa as well as in Krio. It is also apparently common in the various Creoles of the Caribbean where it presumably has an African origin. However, discussion of this aspect has been excluded from the present discus-

14

Joseph H. Greenberg

sion. In Niger-Congo it is found in some languages of each major branch. Unfortunately I could not discover any data for Kordofanian. We can make the following general observations regarding its distribution in Niger-Congo when compared to that of labiovelars. Like the latter it is found throughout the West African forest belt as shown by occurrence in most Mende, Kwa and Benue-Congo languages, and it likewise extends eastward into the Adamawa-Eastern group. It shows, however, two chief differences. It extends farther north, occurring for example in a number of languages of the northern branch of West Atlantic, such as Fula. It is likewise found in some Voltaic and Mande languages north of the labiovelar line. On the other hand it is by no means as continuous in its distribution. In every region in which it is found there are significant gaps. Thus it is apparently absent in Serer Sin in West Atlantic, Susu of Guine in Mande and Tem in Voltaic. This is probably to be attributed to a general tendency for it to be superseded independently and through internal factors by constructions which 22 are more common on a worldwide basis. A further instance of more widespread distribution than labiovelars is its prevalence in Bantu areas. I have recorded 23 ocurrences in almost all of Guthrie's zones. Bujeba was the only contrary instance which was encountered. At the same time we cannot reconstruct the construction as such for Proto-Bantu. Three different roots are the most commonly found: pit-, chit- and kid, all of which mean 'pass' in some languages and 'surpass' in others and have different regional distributions. One possibility is that this semantic change has occurred independently in a parallel fashion in different lexical items, but of course, this does not preclude its spread by loan translation through caiques of different words meaning 'to pass' in different areas. Another possibility is that it was Proto-Bantu and persisted through the change of lexical expression. Of course, and perhaps in more likely fashion, both factors were at work in different individual cases. The distribution of this construction in the Nilo-Saharan languages is much wider than that of labiovelars. It extends well beyond Central Sudanic, including for example Songhai and Kanuri in the west and Eastern Sudanic and the Coman languages in the East. It is not, however, found as far north as Barea, Nubian, Maba or Daza. As in Niger-Congo there are important gaps, e.g. its non-occurrence in Mangbetu in Zaire, a Central Sudanic language with labiovelars.

Areal Characteristics of African Languages As for Afroasiatic languages, it is found not only in some Chadic languages but in Cushitic and even some Semitic languages in Ethiopia. In regard to Chadic, its distribution is wider than that of labiovelars, occurring for example in Hausa, but it is again not found in the Chadic languages east of Lake Chad. Regarding Ethiopia, Bender in his outline sketch of Gumuz, a Coman language noted the prevalence of this construc23 tion "in most Ethiopian languages". Finally, in Khoisan it occurs in ¡Kung, a Northern Bushman language in which its presence might be reasonably interpreted as resulting from 24 contact with Bantu. If the comparative construction we have been discussing is a single historically connected phenomenon, as seems likely from its rarity in other parts of the world, it would seem that we could eliminate Khoisan and Afroasiatic as original sources because of its marginality in both cases. Although found in Chadic, Cushitic and Ethiopian Semitic, it is lacking in Chadic in precisely those eastern languages which seem most exempt from Niger-Congo contacts. Likewise, in the Ethiopian area it is marginal to its occurrence in Nilo-Saharan. Although found in Amharic it is here an alternative to the typically Semitic use of 'from', and is not found in Ge'ez or Tigrinya. It is also lacking in many Cushitic languages, e.g. Somali, Beja and Sidamo. The choice, then, seems to lie between Niger-Congo and Nilo-Saharan within both of which it occurs extensively and, as has been seen, well north of the labiovelar area in West Africa, and east and north of it in East Africa, as well as in almost all the Bantu area. For judging both the origin and age of this construction in Africa, it is regrettable that no evidence could be found regarding its presence in KGrdofanian. Still, though much more tentatively than for the labiovelars, the existence of this construction in all branches of Niger-Congo and far more extensively than for labiovelars, and its absence in Nilo-Saharan languages like Maba, Daza, Barea and Nubian which are remote from NigerCongo contacts, suggests an ultimate Niger-Congo origin. 5. The final question of this sort considered in this paper is a problem of semantic range, the existence of a single term meaning both

1

animal' and

'meat'. I do not have any solid evidence for this anywhere outside of Subsaharan Africa. Before considering the question of its distribution in Africa, however, it will be necessary to consider first what is meant

16

Joseph H. Gieenberg

semantically by this equation. What is most common in languages of the world is that there should be a word meaning 'edible flesh' and that this same word should mean 'human flesh' as is the case, unlike English, with German Fleisch which is, of course, cognate with English 'flesh'. There is often a separate word for 'animal' which in its unmarked or central meaning is 'quadruped', so that for example, there is either a separate word for 'fish' or a phrase such as 'animal of the water' is used when 'fish' is meant specifically. It is also very common that the same word should be used for 'hunted animal' and food from such animals, e.g. French gibiev or English game, or that the same word should be used for a particular animal and the meat from that same animal whether wild or domesticated, e.g. French mouton which means both 'sheep' and 'mutton'. What is found very commonly in Africa, and apparently never or at least not commonly elsewhere, is that the same word should be used for 'animal' in general, whether wild or domesticated, and that the same word should be used of animal food in general. Of the two meanings 'meat' and 'animal', 'meat' seems to be in some sense primary or unmarked and 'animal' secondary and marked. The reasons for asserting this are the following. There is a widespread Niger-Congo root nam, nyam which often has this double meaning and may well be reconstructible as having both meanings. When, however, there are two separate terms in Niger-Congo languages, if, as is often the case, one is the nam root, then the presumed innovation is almost invariably the word for 'animal'. The most conspicuous exception is the Grasslands languages where the form bep or the like is found in many languages with the meaning 'meat' while the noma root survives as 'animal'. The other indication is that sometimes the two terms are separate but involve the same root. In Niger-Congo languages which have class affixes it sometimes happens that the earlier form remains in the meaning 'meat' but the same root with an additional affix of class 1, 'human', 'animal' has the meaning 'animal'. An example is the Bantu language Tonga in which 'meat' is nyama in class 9 while 'animal' is munyama in class 1. In instances of this kind I have considered the language as having the equivalence 'meat' = 'animal'. As with the other African areal properties we have been considering, there are instances in Niger-Congo (I have no information in regard to Kordofanian), Nilo-Saharan and Afroasiatic languages. Regarding Khoisan the evidence is sparse and somewhat confusing. I tend to draw a negative conclusion here. I have not investigated the question in regard to Indo-Euro-

Areal Characteristics of African Languages pean based pidgins and Creoles in Africa and the Caribbean in detail, but believe it to be generally present. In regard to Niger-Congo, the distribution of the trait we are considering, with the exception of its prevalence in Bantu, is very much like that of labiovelars, in that it tends to be concentrated in the forest zone of West Africa rather than in the savannah to the north. (Map 4.) Thus in West Atlantic it is found in the languages of the southern branch that I have investigated but in none of the northern branch. Again in Mande it is reported from the southwestern languages except for Susu while it is found exceptionally in Soninke farther north. As in some instances of the comparative construction which is influenced by the Arabic construction employing a preposition or postposition 'from', there are a number of examples of Arabic influence in Islamic areas in which the marked term 'animal' is borrowed from Arabic hayawan

while the older word is retained

in the meaning 'meat'. It may therefore have once extended farther north. The semantic equation we are discussing is present in many Voltaic languages, being on the whole less frequent in the northern part of the area. It is found in most Kwa languages but with a few conspicuous exceptions, e.g. the whole Togo group, Twi, and Adangme. Again it is found in most Benue-Congo languages including many Bantu languages. The facts regarding Adamawa-Eastern are interesting. All the Adamawa languages for which I have information possess it, but most of the Eastern languages do not. Those which do not frequently retain the nyam root in the meaning 'meat' with various other terms for 'animal'. The few that do, e.g. Ngbandi, generally have a term ga or za which seems to be a loan word from Central Sudanic, e.g. Logo, Madi za< Lendu eza,

za.

It was noted earlier that there was a widespread Niger-Congo root nyam or nam meaning 'meat', 'animal'. However this root does not apparently occur in West Atlantic or Mande, even though many of these languages have the semantic equation 'animal' = 'meat'. In Nilo-Saharan the distribution is much more restricted than in NigerCongo. Thus Didinga is the only Eastern Sudanic language in which I have found it. In Nubian, the word for 'animal' is borrowed from Arabic. This is also the situation in Maba. For the rest it is present in a fair number of Central Sudanic languages though also absent in quite a few. In Central Saharan it is found only Kanuri and it is not found in the Songhai of Gao. In Afroasiatic it is once more found in the Chadic languages of Nigeria but is absent in the languages spoken east of Lake Chad. It is mostly

18

Joseph H. Greenberg

absent in Cushitic except for Elmolo and Boni and it seems not to occur in Ethiopian Semitic. Once more the basic facts of distribution, and in this case the existence of a widespread inherited root, suggests an ultimate Niger-Congo if not Niger-Kordofanian origin. We see that in this case as in the others there is significant evidence for linguistic contact between languages of the Eastern branch of Niger-Congo and the Central Sudanic subgroup of NiloSaharan. 6. The present study then tends to confirm the notion of a sort of nuclear area in Africa in which areal characteristics are most intense. It also suggests that some or most of these originated in Niger-Congo languages and that their present distribution is an outcome of processes both of genetic survival and spread by contact. The incompleteness of these studies and the tentative nature of the conclusions cannot be too strongly stressed. It is hoped, however, that it will serve to stimulate further work on areal phenomena both on the traits discussed here and on others which have been mentioned by various investigators.

NOTES 1. Examples include Welmers (1974) and Gregersen (1977). 2. By Africa is meant here and in subsequent mention Subsaharan Africa, unless otherwise specified. 3. For labiovelars in non-Austronesian languages in northeastern New Guinea see especially Pilhofer (1927). Codrington in his classic work on the Austronesian languages of Melanesia noted (1885, p. 211) that labiovelars are "common in Melanesia", though apparently only the unvoiced non-nasalized stop. He noted that in many languages they lose either the velar or the labial element, a development as we shall see also common in Africa. 4. Stimulated by an observation of Goodman (1970), I reviewed the evidence for subgrouping in Nilo-Saharan and am now of the opinion that ChariNile is not a valid grouping. Central Sudanic and Eastern Sudanic are no closer to each other than either is, for example, to Saharan. However, Kunama and Berta are clearly closer to Eastern Sudanic than they are to Central Sudanic and either or both are perhaps simply Eastern Sudanic languages. 5. See Manessy (1979) and Wolf (1971). 6. Greenberg (1963). See especially items 21, 26, 31, 35 on the etymological list p. 453ff. 7. Thayer (1976).

Areal Characteristics of African Languages 8. Westermann (1927), passim. 9. Ladefoged (1964a), and Wolff (1954). 10. Ladefoged (1964b). 11. Hofmeister (1918). 12. Matsushita (1947). 13. Funke (1915). 14. Westermann and Ward (1957). 15. Paul Newman, personal communication. 16. This is confirmed by the Phonological Archive at Stanford. 17. Tucker and Bryan (1966) especially pp. 29, 63 and 86. 18. Richardson (1956), p. 80. 19. Fortune (1967), Hannan (1961). 20. Jungraithmayr (1965). 21. Chôme and Lussagnet (1958). 22. Thus in Zamuco as described in Chôme's Arte which was written about 1750, the construction based on 'to surpass' is the only one mentioned. A modern study (Keim, 1964, p. 501) shows that this construction has been replaced by the use of a preposition meaning 'in relation to 1 . 23. Bender (1979). 24. J.W. Snyman, personal communication. BIBLIOGRAPHY Bender, M. 1979 Gumuz: A sketch of grammar and lexicon. Afrika und Uebersee, 72, 1, 38-69. Chôme, I., and S. Lussagnet 1958 Arte de la lengua Zarnuca. Journal de la Société des Americanistes 47:121-78 Codrington, R.H. 1885 The Melanesian languages. Oxford: Clarendon Press. Dalby, D. 1977 Language map of Africa and the Adjacent Islands, London. Fortune, G. 1967 Elements of Shona: Zezuru dialect. Salisbury: Longmans. Funke, E. 1915 Die Sprachverhaeltnisse in Sugu, Dahomey, Franz. Westafrika. Zeitschrift fuer Kolonialsprachen, 5: 257-69. Goodman, Morris, 1970 Some questions on the classification of African languages. ISAL 36.2: 117-122. Greenberg, Joseph H. 1959 Africa as a linguistic area. Continuity and change

20

Joseph H. Greenberg in African cultures, ed. W.R. Bascom and M.J. Herskovits, 15-27. Chicago: U. of Chicago Press. 1963 The languages of Africa. The Hague: Mouton.

Gregersen, E.A. 1977 Language in Africa: An introductory survey. New York: Gordon and Breach. Hannan, M. 1961 Standard Shona dictionary. London: Macmillan. Heine, B. 1976 A Typology of African languages, based on the order of meaningful elements. Berlin: Reimer. Hofmeister, J. 1918 Kurzgefasste Wute-Grammatik. Zeitschrift fuer Kolonialsprachen , 9: 1-19. Jungraithmayr, H. 1965 Materialen zur Kenntnis des Chip, Montol, Gerka und Burrum (Suedplateau, Nordnigerien). Afrika und Uebersee, 48: 161-82. Keim, H. 1964 Das Zamuco - eine lebende Sprache. Anthropos, 95: 457-516. Koelle, S.W. 1854 Polyglotta Africana. London: Missionary Society. Ladefoged, P. 1964a A Phonetic study of West African languages. Cambridge: University Press. 1964b Igbirra notes and word-list. JWAL 1: 27-38. Manessy, G. 1979 Contribution à la classification généalogique des langues Voltafques. Paris: SELAF. Matsushita, Shuji 1947 A comparative Vocabulary of Gwandara dialects. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa. Pilhofer, G. 1927 Formenlehre von zehn Mundarten und Nachbarsprachen des Kate. ZfKS 18: 196-231, 298-315. Richardson, J. 1956 Linguistic survey of the northern Bantu borderland. vol. 2. London, New York: Oxford U. Press. Thayer, L.J. 1976 A comparative-historical phonology of the Chari languages (Nilo-Saharan languages of Central Africa). Naples: Instituto Orientale di Napoli. Travelé, M. 1923 Petit dictionnaire Français-Bambara et Bambara-Français. 2nd ed. Paris: Guethner. Tucker, A.N., and M.A. Bryan 1966 Linguistic analyses. The non-Bantu languages of north-eastern Africa. London, New York, Capetown: Oxford U. Press. Weimers, W.E. 1974 African language structures. Berkeley: u. of California Press. Westermann, D. 1927 Die westlichen Sudansprachen und ihre Beziehungen zum Bantu. Berlin: de Gruyter. Westermann, D., and i.e. Ward 1957 Practical phonetics for students of African languages. London, New York: Oxford U. Press.

MAPS

3

Chapter 2

Adamawa-Eastern: Problems and Prospects Patrick R. Bennett

The languages classified by Greenberg as Adamawa-Eastern' are probably the most poorly documented of all the major divisions of Niger-Congo. Though they cover an area extending from Nigeria to the Sudan, and include, in Gbaya, Sango, and Zande, some of the most widely spoken languages of Africa, they have, perhaps because of their position relatively far from the coast, received little attention. Today, some of the gaps in our knowledge of these languages are being filled by linguistic researchers, 2 and some comparative work has begun as well. I do not propose in this paper to discuss in any great detail the progress that has been made in linguistic research on Adamawa-Eastern. For general surveys of the field the reader is referred to Samarin (1971), Boyd (1978), and Barreteau and Monino (1978).^ In the area of bibliography even the latest of these is outdated, due to the increasing flow of new data. For example, in the time between the first version of this paper and its revision I received vocabularies of four languages of the group kindly supplied by Philip Noss. This paper will concentrate instead on problems of comparison, subgrouping, and classification of Adamawa-Eastern languages. It is the second step in an on-going re-assessment of relation4 ships within Niger-Congo. The first step focussed on the postulation and sub-grouping of South Central Niger-Congo. I had originally intended to look next at a suggested parallel group. North Central Niger-Congo, to which the Kru languages and Greenberg's Gur and Adamawa-Eastern were tentatively assigned. The relatively great linguistic complexity and poor documentation of these groups, however, have made it advisable to restrict this study to the one division. The questions we must ask in comparative linguistic evaluation of Adamawa-Eastern are three. First, are the languages properly assigned to this rather than to other branches of Niger-Congo? The currently accepted classifications rest on a basis of typology and Greenbergian 'mass-comparison' rather than one of linguistic reconstruction and subgrouping by innovation. Many of the assignments are based on the evidence of regrettably poor vocabularies (which in the case of too many languages remain our sole

24

Patrick R. Bennett

evidence). Especially in an area linguistically so fragmented as the NigeriaCameroun-Chad region (where the Adamawa-Eastern languages are concentrated), it is almost inevitable that individual languages should be mis-assigned. Second, what in fact are the interrelationships of the subdivisions? Most of African linguistic classification so far is reliable enough at the level of the super-group. Classifications of the level of Niger-Congo or Afro-Asiatic, or even the level of Chadic or Nilotic, while by no means indisputable, are clear enough that most comparativists will probably agree on their general outlines. Again, at the level of the dialect cluster or smaller language group relationships are sufficiently obvious that few would seriously doubt the unity of, say, the Berber languages or the Temne-Landuma group. The network of intermediate relationships, however, remains largely unresearched; even within Bantu, perhaps the most carefully studied corner of Niger-Congo, a convincing subgrouping remains to be carried out. Thirdly, we must of course ask, is there an Adamawa-Eastern? Do the various groups of languages assigned to Adamawa-Eastern constitute a viable unit within Niger-Congo? Are they properly grouped together, and, if so, what is their relationship to the rest of Niger-Congo? In the case of South Central Niger-Congo, it would seem that the accepted grouping was decidedly deficient at this level, since the data fail to support either a viable distinction between Kwa and Benue-Congo or the inclusion of Kru. Clearly, a final answer to these points will not be possible. Even now the level of documentation is insufficient to permit adequate analysis in many areas. It does, however, seem appropriate to make a start. As a first step in the subgrouping of Adamawa-Eastern, a lexicostatistical grouping was carried out with the aid of the computer program already used in the study of South Central Niger-Congo. The list used consisted of 102 glosses; vocabularies were obtained for 50 languages."* The languages were chosen to give the greatest possible spread of documentation; where possible, subgroups were represented by three or more members. Unfortunately, but necessarily, the languages of the westernmost divisions were slighted; available vocabularies were simply too few and too unreliable. As in the preceding study, the items were coded to reflect cognacy judgements at three levels of probability. Where phonologic correspondences were obvious from the wordlist data, regular correspondence was a

Adamawa-Eastern factor in cognacy decisions; many, however, inevitably had to be made impressionistically. Some of these judgements, however, were later born out by the evidence of linguistic comparison. The output of the program was interpreted to give a preliminary - and tentative - subgrouping, shown in 1) below. 1) Lexicostatistically Based Subgrouping Ma(B7) Mba(B8) Mondunga(B8) Ndogo(B6) Tagbu(B6) Bviri(B6) Indri(B6) Togoyo (B6) Feroge(B6) Mundu(B5) Yogo (B5) Ngbaka-Ma'bo (B5) Gbanziri (B5) Monzombo(B5) Golo(B2) Gabu(B2) Banda(B2)

Note: Numbers are those of Greenberg's subgroups. Distances in the lines marking subgrouping are based roughly on the statistical differences, but for several reasons

Ngala(B4) Zande(B4) Barambo(B4) Pambia(B4)

are not to scale.

Sango(B3) Ngbandi(B3) Gbaya(Bl) Ngbaka(Bl) Kali(A6) • Galke(A6) Tuboro(A6) Mbum (A6) Yasing(A6) Mundang(A6)

if

Vere(A4) . Sewe(A4) Kutin(A4 Woko(A4) Pape(A4) • Duru(A4) Chamba (A2) Kolbila(A2I) Mumuye(A5) Fali(All)

26

Patrick R. Bennett Lexicostatistically Based Subgrouping (Cont.)

I I

Longuda(A10) Tula(A1) Yungur(A7) Jen(A9) Boa(A13) Kulaal(A13) Daka(A3)

Two large units were clearly supported by the statistical study. One consisted of Greenberg's groups 2, 5, 6 and 8 of the Eastern branch. The second included groups 2, 4, 6 of the Adamawa branch. Three smaller groups, Greenberg's groups 3, 4, and 7 of the Eastern branch, were affiliated with the first unit, but at a much lower level of relationship. In the Adamawa branch, only Greenberg's group 5 seemed to affiliate similarly with the larger cluster. Greenberg's groups 1 and 10 in Adamawa, though not showing a high level of relationship, seemed to pattern together. Curiously, the Gbaya group, which Greenberg classed with the Eastern branch though with reservations based on similarities to Adamawa languages, appeared to pattern with neither branch. Nor was the unit status of Adamawa-Eastem confirmed by the statistical results. The next step, as with the previous study of South Central Niger-Congo, involved the expansion of data for the groupings statistically established. The resulting expanded vocabularies were then examined with an eye to linguistic reconstruction. Here I concentrated on the former of the two large groups mentioned above, that consisting of Greenberg's Eastern 2, 5, 6, and 8, which will be referred to in this study as Ka.^ One reason for this choice was the existence of Boyd's comparative study of 7 part of the other major group. This greatly facilitated the process of comparison in that branch. In the case of the KS languages, however, it was necessary to collate data from very fragmentary sources to arrive at a satisfactory corpus for comparison. Sets of reconstructions were compiled for the Ndogo-Bviri cluster, Feroge-Manga, Mba-Mondunga, Mundu-Gbanziri, and the Banda group. Although most researchers - and the indications of the statistical program - group Ndogo-Bviri, Feroge-Manga, Indri, and Togoyo and treat them as a unit, this did not prove profitable here. It is true that these languages are closely related, especially in lexicon. But there is no evidence of any significant lexical or phonologic innovation setting

Adamawa-Eastern

27

them off from the remainder of the group. Indeed, Ndogo-Bviri very often patterns separately from the other three. It might have been possible to establish regular correspondences and reconstructions for Feroge, Indri and Togoyo, without Ndogo-Bviri, but the correspondences are quite complex and it was decided not to do so. In the case of Mba-Mondunga, major problems resulted from the relatively poor documentation, especially for verbs, and the degree of separation. Though the two languages are clearly closely related, sharing a number of innovative features as well as their retention of the original concord system, their separation has resulted in lexical divergence. With this and the documentation, it is more often than not possible to find a cognate in only one of them for a given Proto-Ka item, and the "regularities" of correspondences established under such conditions are less than reliable. This was somewhat surprisingly and regrettably the case with the Banda group as well. Though Banda is an important dialect cluster, and has received a comparatively large proportion of the linguistic attention given the K3 group, very little reliable material exists. Tisserant's dictionary, though useful in many ways, gives us neither the phonetic detail nor the information on dialectal variation we need. As a result, my reconstructions for the other groups, even Mba-Mondunga, may be called reconstructions, shaky though sane of the ground they rest on may be. Those for Banda, however, are at best pseudo-reconstructions, especially dubious in the areas of vowel-quality and tone. For Mundu-Gbanziri, the situation is different. With Santandrea's material on Mundu in the east, Thomas' for Ngbaka-Ma'bo in the west, and the Bako dictionary of Brisson and Boursier, reconstruction of a relatively high degree of reliability is possible, especially given the volume of lexical data available (though admittedly mostly of poor quality) for Yogo, Bangba, Monzombo, and Gbanziri. There remain problems, of course; often a crucial language is not represented, tone is frequently probleg matic , and I can only consider my reconstruction fully reliable when at least two of the apparent subgroups are represented, and even then only if one of the reliable sources is included. Here, however, I have been 9 able to establish a relatively large list of reliable reconstructions. All these Ka reconstructions were then collated to yield approximately 300 sets of possible cognates. In seme cases all groups were represented and the probability of cognacy high. An example is given as 2a) below.

28 2a)

Patrick R. Bennett Togoyo

"WATER"

: ngu

Mundu-Gbanziri

: ngbcj

Feroge-Manga : nkeBa

Banda

: nku

Indri

Mba-Mondunga

: ngio-m

Ndogo-Bviri

: ngo : ngua

Such items, however, constitute a minority. Out of about three hundred such sets, only thirty are attestable or reconstructable from all seven subdivisions. And of these, several involve such complex -indeed, dubious resemblances as are seen in 2b). 2b)

"BONE"

Togoyo

baag

Ndogo-Bviri

Mli

Mundu-Gbanziri

blk

Feroge-Manga

babegi

Banda

gbabi (?)

Indri

fcika

Mba-Mondunga

bebe-w

Any reconstruction - or even judgement as to cognacy - must be highly tentative under such circumstances. Yet, entries from at least half the subdivisions are available for the majority of the cognate sets. And if many of the identifications are dubious, much reconstruction is nonetheless possible. A number of repeated patterns of correspondence appear in the data. Such items as 2a), for example, contrast with others to attest to an original - and reconstructable - contrast between voiced and voiceless prenasalized consonants. Such items as 2c) contrast with sets like 2d) to attest to an original a/a contrast.^ 2c)

2d)

Togoyo

"THREE" Ndogo-Bviri

ta-o

Mundu-Gbanziri

Feroge-Manga

taa

Banda

Indri

ba-ta

Mba-Mondung a

taa-ta va-ta fcia-la O )

Togoyo

ngag

Ndogo-Bviri

ki

Mundu-Gbanziri

k3

Feroge-Manga

ka

Banda

nko

Mba-Mondunga

kS-1

"BREAST"

Indri

ngarja

Even certain of the most complex sets - like 2b) - can in fact be reconstructed. Though only Mba-Mondunga retains an operative concord system, Feroge-Manga, Indri, Togoyo, and Mundu-Gbanziri frequently show final consonants or syllable which appear to be reflexes of the class-marking suffixes. Ndogo-Bviri shows such fossilised suffixes as well, though rather less frequently. If the finals of the Ndogo-Bviri, Feroge-Manga, Indri,

Adamawa-Eastern Togoyo, and Mundu-Gbanziri are ignored, and the reduplication of FerogeManga, Mba-Mondunga, and perhaps Banda disregarded, we can reconstruct for "BONE" a form*il -, probably assigned to a class with a velar consonant in its suffix. The relationships between Ka and the remainder of Ubangian, and relationships among the remaining groups of Ubangian, are unclear. The Ngbandi group shows a slightly greater degree of cognacy with the KS group than either does with the Zande group. There is also greater typological similarity. It is difficult to determine, however, to what extent this reflects contact. The Ngbandi group has had great influence on - and been influenced by - various members of the KS group, especially Banda. Zande and its relatives have influenced other portions of Ka, especially the eastern members, but have received less influence in return. Ma shows remarkably low cognacies with all other Ubangian languages. On the basis of such diagnostic items as *k3 "breast", Ma does not affiliate with any member of the Ka group: Ma agrees with Ngbandi and the Zande group in most such cases. A slightly greater degree of cognacy with the Zande group is probably due to contact. The situation is complicated by the extremely poor documentation of Ma. It should be noted here that Tucker and Bryan's grouping of Ma with Mba-Mondunga, based on the presence in both of a working class system absent in other Ubangian languages, is not supported by lexicostatistics or the evidence of lexical and phonologic innovation. Though the class systems are essentially identical, they represent chance retentions of a feature lost elsewhere, and cannot serve as a basis for classification. At present, I must consider Ka, the Ngbandi group, the Zande group, and Ma 11 to be parallel branches of the Ubangian languages, which are set off from the remainder of Adamawa-Eastern by a number of isoglosses. In the items "breast", "man", and ind "leaf" - 3a) through 3c) an initial*k lost elsewhere. 13

12

Ubangian shows

3a) "BREAST" Ka

ka

Ma Zande Ngbandi

:

Boa-Kula

Gbaya Chamba-Namshi

vat

Mangbei-Mbum

waa

Fali

Tula-Longuda Yungur Burak-Jen

30 3b)

3c)

Patrick R. Bennett "MAN" Kà

: koa

Gbaya

wérì

Boa-Kula

bar

Ma

: kofo

Chamba-Namshi

va

Tula-Longuda

bwà

Zande

: kwói

Mangbei-Mbum

wà

Yungur

ksre

Ngbandi

: kó

Fali

Burak-Jen

béé

Kà

: kwa

Gbaya

Ma

: kpe

Chamba-Namshi :

Zande

: kpé

M angb e i-Mbum

: ywa

Yungur

: (k)wàxma

Ngbandi

: kué

Fali

: wasi

Burak-Jen

: yaQ

"LEAF" : wàà Ma

Boa-Kula

: wa

Tula-Longuda :: wà

In items 3d) through 3f) the contrast between Ubangian and the remainder of Adamawa-Eastern is seen. 3d)

"FAT" Kà

meo

Ma Zande

3f)

: no

Chamba-Namshi : rxicj

Boa-Kula

: nu

Tula-Longuda : nl

Tula-Longuda

Zande

zo

Mangbei-Mbum

Yungur

Ngbandi

du

Fali

Burak-Jen

"STEAL" Kà

: zi

Gbaya

Ma

: di

Chamba-Namshi :: lu

Zande

: di

Mangbei-Mbum

Ngbandi : nzi 4f )

nyuwa

"TONGUE"

"EAT"

4d)

: mu

: zu : ri

Fali

Boa-Kula Tula-Longuda Yungur Burak-Jen

"NECK" Kà

miri, gò Gbaya

g£r

Boa-Kula

Ma

me

Chamba-Namshi

gaa, du Tula-Longuda : mil, gwa vò

Zande

góro

Mangbei-Mbum

goa, sò Yungur

: kwe

Ngbandi

gò

Fali

gono

: dwi

Burak-Jen

32

Patrick R. Bennett

In 4a) , the proto-Ubang ian form for "mouth" can be reconstructed as *mcj on the basis of Ma and the majority of Ka. This contacts nicely with the *T)a

stem reconstructable for most of Adamawa-Eastern. Both are rather

normal reflexes of the common Niger-Congo item, which may have been some such form as *wa. However, the *ngba which seems to be reconstructable for the Zande group and the Ngbandi *yo would indicate that the innovation is not general in Ubangian. One might then consider it a Ka innovation, or possible evidence for a grouping of Ka and Ma distinct from the rest of Ubangian. Unfortunately for this hypothesis, Ndogo-Bviri shows a form *njwi. We may be forced either to reconsider the Ka grouping or to assume that the form *ymu arose after the subdivision of Ubangian and 14 of Ka either through parallel evolution or by contact between branches. The items in 4b) illustrate a similar problem. The general Niger-Congo form for "tongue" is approximately "(de)dem; it is frequently subject to metathesis. Much of Adamawa-Eastern displays such metathesized forms. In Ubangian, further, most languages seem to have reinterpreted the final of assumed proto-Ubangian *mli> as a class marker.*^ We might with some reason attribute the metathesis at least to a rather early point in the history of Adamawa-Eastern, certainly before the separation of Ubangian. Yet we see in Togoyo a form lomi. How are we to interpret this? A cowardly way out would be to assume that it is either a loan from seme form of Bantu (though there is no valid evidence to support such an assumption) or a compound with some (unidentifiable) item prefixed to the *mi which other Ubangian forms would lead us to expect. To consider it a re-metathesized form is equally unjustified, though more daring. The identical problem arises in the reconstruction of Chamba-Namshi, where most branches show *mti-l, but the Mumuye group has a form *cfem-. Perhaps one should at this point revise views on subgrouping, or drastically revise assumptions as to the data of the metathesis. In this case, however, this is one of a number of lexical features which link Mumuye with Mangbei-Mbum rather than Chamba-Namshi, most of which seem to involve retentions rather than innovations in Mumuye and Mangbei-Mbum. As there are several innovations linking Mumuye with ChambaNamshi, it seems simpler to assume that Mumuye separated from ChambaNamshi before these changes took place. Such an interpretation is not apparently available in the case of Togoyo. In 4c) through 4e) we see another type of problem. The items "eat", "bear", and "steal" are part of a series of correspondences between a

Adamawa-Eastern Ubangian *z and a

*d or *r elsewhere in Adamawa-Eastern or Niger-Congo

generally. The type of sound shift is not uncommon in Niger-Congo, and is well-attested in Gur, for instance. In each of these cases, however, a cognate form occurs in one or another branch of Ubangian (though not normally in Ka) with d or P. There is no consistency in which language group does not show 2, nor is there obvious conditioning in any of the groups showing such correspondences. How, then, is the shift to 3 to be interpreted in subgrouping these languages? Somewhat less of a problem is posed by the items like 4f). Niger-Congo contained - and present Niger-Congo languages contain - several sets of near synonyms. Sane languages retain these sets, others replace one or more items, often with other members of the set. Such cases do not present any serious obstacles to subclassification. Such problems are, of course, not unique to Adamawa-Eastern comparison. However, in a language group so little studied they render decisions as to subgrouping so much more difficult. At a later stage, after presentation of the original version of this paper, and after acquiring further data on certain key languages, I compiled similar sets of reconstructions for the Mangbei-Mbum and ChambaNamshi groups. For these two groups, as had already appeared probable from the lexicostatistical study, it was possible to reconstruct inventories of around 160 proto-items each, with many items common to both lists. The status of the Mangbei-Mbum group was very clear, given a set of shared phonologic innovations, particularly a series of spirantizations, devoicings, and a characteristic replacement of nasals by prenasalized stops. These, added to some lexical features, clearly define the group. It was more difficult to demonstrate the validity of the Chamba-Namshi branch. This is a rather conservative group phonologically. It also shows much more internal divergence than Mangbei-Mbum. Even in the lexicon there are not very many innovative items shared by the entire group. One may, however, cite *ga- "horn", *du- "fish", and *bik- "snake". The two groups are clearly separated by phonologic shifts and lexical innovations, such as Chamba-Namshi *zii-l 'head', "irregularly corresponding to Mangbei-Mbum *tu-l, from North-Central Niger-Congo *du~. However, they are linked lexicostatistically, typologically, and by such lexical items as *ya-k "mouth" and *to—l "sorghum". In both groups the same problems arose as with the reconstruction of

34

Patrick R. Bennett

Ka or Ubangian. When we turn to Adamawa-Eastern as a whole from these subgroups, we find much the same types of inconsistencies and difficulties. At the higher level, however, it is much less clear that we are dealing with a valid linguistic unit. The lexicostatistical study linked Tula with Longuda and Boa with Kulaal, groupings which can be backed by non-statistical evidence. Though the evidence for the groupings is statistically rather less than that supporting Ubangian, Chamba-Namshi, or Mangbei-Mbum, it is nonetheless adequate. The lexicostatistics also supported a grouping of Chamba-Namshi with Mangbei-Mbum. Their unity is also attested by Boyd's reconstructions. Though he does not deal with the entirety of either group, the proportion of reconstructable items and regularity of correspondence is considerable. The lexicostatistics, however, did not support any other linkages between subgroups of Adamawa-Eastern. Cognacies between, say, Longuda and Mbum are at a level which might be matched by almost any two typologically similar Central Niger-Congo languages. Can, then, the classification Adamawa-Eastern stand as a valid unit? What would be needed to verify the unity of the group, especially given the lack of any lexicostatistical indication of unity, is evidence of internal lexical or phonologic innovation shared by the members of the group, and seme indication that no member of the group is more closely related to any other branch of Niger-Congo. The search for a general Adamawa-Eastern phonologic innovation seems fruitless. If any such innovation ever took place, it has been parallelled elsewhere in Niger-Congo, or obliterated by subsequent changes within Adamawa-Eastern subgroups. Like the case of the Ka word for "mouth", in 4a), the variation is too great for any defining innovation to be identified. In the area of lexical innovation, I have only the seven items in 5). 5) "beer"

: *pi

"fire"

: *wua

"breast"

: *xa

"neck"

: *gO

"elephant"

: *Pal

"tooth"

: *t£

"wind"

: *ya

The two items *xa "breast" and "pi "beer" appear to be restricted to Ubangian, Chamba-Namshi, and Mangbei-Mbum, while the others are more general. Some, like *Pal "elephant", probably owe their general spread

Adamawa-Eastern to borrowing. There are a number of other items, of course, which are general in Adamawa-Eastern. While some of these are general Niger-Congo vocabulary, the remainder might be used to define a characteristically AdamawaEastern set of isoglosses. However, far too large a proportion of these characteristically Adamawa-Eastern items and forms are shared with Gur; while they may strengthen the case for the unity of Adamawa-Eastern, they do not argue for its exclusivity. The quality and quantity of data available weaken what evidence there is. The general Adamawa-Eastern word *ya

"wind" listed in 5) is included

because it has not been found outside Adamawa-Eastern - so far. Should a clear cognate appear elsewhere as our knowledge of Niger-Congo expands, this "innovation" would have to be reevaluated. Again, some items not included as Adamawa-Eastern innovations, whether because the evidence for non-chance correspondence was inadequate or because lexical lists for some groups did not include them, might be recognized in future. Though the evidence of seven possibly innovative items, already weak, is thus rendered weaker by levels of documentation, we might still tentatively conclude that Adamawa-Eastern forms a unit - IF it could be shown that no subgroup patterns as closely, or more closely, with other branches of Niger-Congo. Except in the case of Daka, which, for reasons to be discussed later, is misplaced in Greenberg's classification, the border between Adamawa-Eastern and the Eastern sections of South Central Niger-Congo is quite well defined. While there are some resemblances between Adamawa-Eastern and Western South-Central Niger-Congo, the Gur languages are a more logical place to look for such linkages. Not only is Gur somewhat closer physically and typologically, but also there are several important lexical sharings between Gur and Adamawa-Eastern. In order to test this aspect of Adamawa-Eastern unity, I focussed on Tula-Longuda. The languages of this group are among the better documented of the westernmost divisions of Adamawa-Eastern, and are typologically closes to Gur. From the data available I took a list of items which could be reconstructed for Tula-Longuda. Though the Tula group and the Longuda cluster are linked by a number of lexical and phonologic isoglosses, the two, like Mba and Mondunga in Ubangian, have undergone considerable divergent evolution. Since I accepted only items attested in both subdivisions of Tula-Longuda, and because of the poor volume and quantity of available lexical data, only 32 such items could

36

Patrick R. Bennett

be included. This list was then compared with reconstructions from Chamba-Namshi, Mangbei-Mbum, Ubangian, Gbaya, Boa-Kula, and three subgroups of Gur, the Kassena-Lyele branch of Gurunsi and the Moore and Gurma divisions of Manessy's Oti-Volta group.^ Three of the original items on the list had to be omitted, since I could not find appropriate items in all of the groups compared. Cognates occurring in these lists were then counted, 18 first on a simple cognate/unrelated basis, then on a weighted scale which assigned two points to close cognates, one point to possible but significantly skewed cognates, and no point for forms assumed unrelated. The cognacy judgements attempted to be generous without being outrageous. Care was taken to consider what patterns of regular correspondence between groups could be seen in the more extensive data from which these 29 items were extracted. With some exceptions, the resulting pattern was that of a typical dialect chain. The smallest number of cognates was 11, shared by Gbaya and the Moore group. The greatest number shared was 25, between Ubangian and Mangbei-Mbum. On the weighted scale, the figures ranged from 14, between Gbaya and the Moore group and Gbaya and the Gurma languages, to 41, shared by the M6ore and Gurma subdivisions of Oti-Volta. The ratio between the weighted and simple counts, which may be taken as an index of shared innovation and/or shared retention, ranged from 1.12 (indicating little shared innovation/retention) between Mangbei-Mbum and the M6ore group to 1.78 (indicating strong similarity in direction and degree of innovation) for the two branches of Oti-Volta. By and large the lexical and innovation-based figures confirm one another. The diagram in 6) may be taken as reflecting approximately the relationships indicated by this count. 6)

Gu Mo

/»•Gr

TL *de-dsm as well. 16. The status of the nasal in "eat" is unclear. In Kâ, Chamba-Namshi, and Mangbei-Mbum it occurs sporadically throughout the subgroups - one language will have a nasal vowel, the next not. A nasal in this item is unusual in Niger-Congo generally. If these language groups were more consistent, one could identify this as a distinguishing innovation of the group Cameroun-Ubangian established below. Given such inconsistency, however, it is hard to know what to make of it. 17. For the Gur groups, see Manessy (1969) and Manessy (1975). 18. For a discussion of the role of weighting in lexicostatistlcal counts, see Bennett (1976). 19. Greenberg (1955), p. 12. BIBLIOGRAPHY A complete listing of sources consulted in the preparation of this paper would be excessively lengthy, and redundant as well, given the extensive bibliographies that exist. The linguistic materials on which my conclusions are based are drawn from most of the available grammars, wordlists, and in some cases texts. The works listed here include classifications and bibliographies which must form a background for any such work as this, and those works directly referred to in the text. Barreteau, Daniel and Yves Monino. 1978. "Les Langues Oubanguiennes", in Inventaire

des Etudes

d'Expression

Française

Linguistiques

sur

les

et sur Madagascar.

Pays d'Afrique

Noire

Edited by Daniel Barreteau.

Paris: Conseil International de la Langue Française. Bennett, Patrick R. 1976. "Some Problems of Bantu Lexicostatistics", in Lexicostatistics

in Genetio

and Guy Jucquois. Cahiers

Linguistics de l'Institut

II.

Edited by Isidore Dyen

de Linguistique

de

Louvain

3(5-6) . Bennett, Patrick R. and Jan P. Sterk. 1977. "South Central Niger-Congo: A Reclassification", in Studies Boyd, R. 1974. Etude

Comparative

in African

Linguistics,

8(3).

dans le Groupe Adamaua. Paris: Biblio-

Adamawa-Eastern

thèque de l a

SELAF.

Boyd, Raymond. 1978.

Linguistiques et

sur

" L e s Langues Adamawa", i n Inventaire

sur les Pays d'Afrique

Madagascar.

International

des

Noire d'Expression

E d i t e d by D a n i e l B a r r e t e a u .

Paris:

Etudes

Française Conseil

de la Langue Française.

Greenberg, Joseph H. 1955. Studies

Linguistic

in African

Classification.

New Haven: The Compass P u b l i s h i n g Co. Greenberg,

Joseph H. 1963. The Languages

of

The Hague: Mouton and

Africa.

Co.

Manessy, G a b r i e l . 1969. Les langues garunsij méthode comparative

à un groupe

d'application

de langues voltaïques.

de la

P a r i s : SELAF.

de langues voltaîques.

Manessy, G a b r i e l . 1975. Les langues Oti-Volta: d'un groupe

essai

classification

généologique

P a r i s : SELAF.

Samarin, William J. 1971. "Adamawa-Eastern", in Curèrent Trends in tics

7,

e d i t e d by T .

Tucker, A.N. and M.A. Bryan. 1956. The Non-Bantu Languages of Africa.

London: O x f o r d U n i v e r s i t y

of

North-Eastern

Africa.

North-Eastern

Press.

Tucker, A.N. and M.A. Bryan. 1966. Linguistic Languages

Linguis-

Sebeok. The Hague: Mouton and Co.

Analyses:

The Non-Bantu

London: O x f o r d U n i v e r s i t y

Press.

48

Patrick

R.

Bennett

APPENDIX: The Glosses in the L e x i c o s t a t i s t i c a l Study.

animal

chicken

fire

house

nose

spear

arm

child

fish

kill

oil

stone

arrow

cloud

five

knife

one

sun

ashes

come

fly

laugh

person

tail

bad

cry

four

leaf

rain

ten

bear

day

frog

leg

red

three

bee

die

give

leopard

river

tie

belly

dog

goat

liver

road

tongue

big

drink

good

long

saliva

tooth

bird

ear

gun

man

salt

tree

bite

eat

hair

moon

see

two

black

egg

head

mouth

sit

urine

blood

elephant

heart

nail

sky

war

bone

excrement

hill

name

sleep

water

bow

eye

hit

navel

small

white

breast

fat

honey

neck

smoke

wind

buy

finger

horn

night

snake

woman

Chapter 3

The Eastern Jebel Languages M. Lionel Bender

1. INTRODUCTION The East Sudanic family is the largest and most complex family in Nilo-Saharan, as established by Greenberg (see Greenberg 1963). The inclusion of East Sudanic in a larger Chari-Nile unit has been questioned (Goodman 1970, Bender 1976), and I now see East Sudanic as one of the coordinate families of Nilo-Saharan. My present thinking on the subject of Nilo-Saharan internal classification is as follows: A

Songay

E

East Sudanic

F

Central Sudanic

B

Saharan

El

Nubian

G

Berta

C

Maba

E2

Surma

H

Kunama

D

Fur

I

Komuz

E3

Nera

E4

Eastern Jebel

E5

Nyimang

E6

Temein

E7

Tama

E8

Daju

E9

Nilotic

E10

Kuliak

(Koman and Gumuz)

Note some changes in spelling and in terminology to bring the classification into line with recent findings: the disappearance of "Chari-Nile" (formerly combined families E-H), the term Surma for the former "DidingaMurle", Nera for "Barya", Eastern Jebel for "Tabi" or "Ingessana", Kuliak for "Teuso" or "Ik", and the new term Komuz (a hybrid of Koman and Gumuz, representing their coordination into a single family). Evidence is accumulating that families E1-E4 can be grouped together as "Eastern", and families E5-E8 as "Western" branches of East Sudanic, but this will not be pursued here.

50

M. Lionel Bender

One of the poorest-documented East Sudanic sub-families has been E4, until now considered as consisting of a single language. My 1978-79 fieldwork in Sudan, however, has established that this sub-family consists of at least four languages. Three of these were known only through minimal data and a misclassification in Evans-Pritchard's important article reporting his fieldwork of more than half a century ago (Evans-Pritchard 1932). This paper will present an outline of the Eastern Jebel sub-family languages and peoples, a consideration of how Evans-Pritchard came to classify some of these languages as Berta varieties, and an exposition of the major grammemes (grammatical morphemes) which establish the correct classification. In connection with the observations on Evans-Pritchard's data, mention will be made of extended lexical comparison which supports fully the classification based on grammemes. 2 . THE EASTERN JEBEL SUB-FAMILY

This name was chosen because the languages are spoken by communities associated with a number of hills (Ar. sg. jzbsl, pi. jnbaaZ) in Blue Nile Province, eastern Sudan. The name continues Greenberg's suggestion to use fairly neutral geographical names (e.g. Afroasiatic, East Sudanic) as the least misleading terminology for African language classification (e.g. 1963:50). 1. Gaam.

The former Tabi (name of a hill) or Ingessana (Arabic name of a

ring of hills). The people call themselves jog gaam (people (of) hill) and their language kov e gaam (speech of hill). Dialect variation among the ring of hills and adjacent plain is relatively minor. The three most important dialects are: Kukur (govr) to the northwest, Soda (tau) to the southwest, and Bau (bau) to the east. The dialect described herein is Bau 2. Aka.

The people call themselves fa-a-aka (people-of-Aka), the language

is called ulu-ka-fa-caka, and they are found at a jebel usually known as Silak (Sillok) south of the Ingessana Hills, west of Keli. 3. Kelo.

A small village near Keli, on the Demazin-Kurmuk road south of

the Ingessana Hills, and overlooked by the impressive Jebel Tornasi, is inhabited by a couple of hundred Tornasi people (self-name: faa-keeli),

Eastern Jebel Languages Their language is close to that of the similar small village under Jebel Beni Sheko, about 25 miles to the north on the same road. Kelo is described herein l.\mdu-fda-kSeld3 language-people-Kelo) . 4. Molo.

Another jebel south of the Ingessana Hills is Malkan; here are

found a few hundred people speaking tvP3-kB-molo (language-of-Molo). The total of the Aka-Kelo-Molo (Evans-Pritchard's Silak-Tornasi-Malkan) people (including Beni Sheko) probably does not exceed 1000. All these people are known as "Hamej" or "Funj" in the local Sudanese folk taxonomy, which implies that they are remnants of the former (16th-19th century) Funj Kingdom in eastern Sudan. Linguistically, these terms lump together people speaking Eastern Jebel languages (in the present case), Nilotic (e.g., Ulu, Burun), Koman (e.g., at Jebel Gule), Gumuz (in Ethiopia; see Bender 1975: 70), and probably others. In view of the distribution, the most likely hypotheses about the Aka-Kelo-Molo is that they represent remnants from a once wider distribution of the Gaam (who subsequently retreated to the Ingessana Hills) or on the contrary an expansion of the Gaam language to nearby jebels. The possibility of finding other Eastern Jebel varieties at other locations cannot be ruled out. For example, I believe that my verification that the Beni Sheko people speak a dialect of Kelo is the first such report in the literature. It should be recalled that in past times the Jebel people actually lived on the jebels, not in the plains below (e.g., The Beni Sheko in the saddle of the tricornate jebel of the same name). All three of AKM, especially Aka, are more heavily influenced by Arabic than Gaam is. How much this is a function of the particular informants remains to be seen.

3. EVANS-PRITCHARD'S OBSERVATIONS The only linguistic data on Aka-Kelo-Molo that I have seen is that of EvansPritchard 1932, collected in 1926. It consists of about ninety words, and does not include pronouns or some other key items which would have led to a correct classification of AKM with Gaam long ago. Evans-Pritchard was misled by a striking incidence of Berta loan words among basic lexicon into classifying these languages as "the second group of Berta languages" (alongside his and earlier observers' Berta of the Ethiopia-Sudan border area). Rather than look at Evans-Pritchard's complete list, it will be sufficient to look at the ten-word list used by Herman Bell in Sudan language

52

M. Lionel Bender

Survey Work, since the results are very nearly the same.

Gloss

Aka1

Silak1

Kelo1

Tornasi1

Molo1

1.

one

ligldl

ligldl

lSodi

lodi

leédì

2.

two

wäasf

wasi

wáadi

wat i

wäadi

3.

three

éedé

ede

aádá

ede

3odS

4.

fire

máa

ma

má

ma

má

5.

water

cíi

ci

kí?i

siki

si

6.

sun

bíidí

bidi

blidi

bit!

biidi

7.

moon

ágiSwá

agwa

á?uwa

aguwa

aguwa

8.

tongue

kälä

kúla

kälä

unküla

k31à

9.

tooth

jiíicé

~

Jlisu

10. blood

Jliìdl

—

gyefa,,

gyefa

kefe

gyéebá

gyèewà

jéefá MaTkan1

Berta'

Dui1

Gaam1

1.

ledi

m3n8k'o

duguni

t3m3n

2.

wadi

moholajl

koleni

daag

3.

odo

nriSuSè

zitigini

i>ti>

4.

ma

moo

mo

moo

5.

si

firí

feri

f Eg

6.

bldi

möndzu

munzu

t£l

7.

agwa

sT.gì

zlgi

turi

8.

kula

hàlé?

küla

kSlád

9.

—

ñdüfüdi2

—

10.

giewa

k'afa

of ut

Jiiid afad

First, it will be noted that my Aka-Kelo-Molo are in full agreement with Evans-Pritchard's Silak-Tornasi-Malkan on the nine items which are comparable (except possibly Kelo-Tornasi on "water"). My Berta and Evans-Pritchard's Dul agree on all but "blood" and the numerals. Other Berta varieties I have collected always have something like k'afa for "blood", while Evans-Pritchard's word looks like Gaam. The numerals show great variety in Berta (in most areas today, all Berta numerals have been lost in favor of Arabic), but none of the Berta or Dul forms given show any convincing resemblances to the other languages anyhow.

Eastern Jebel Languages

53

This leaves three items in AKM close to Berta ("fire, tongue, blood") and four close to Gaam: "three, fire, tongue, tooth". This is using EvansPritchard's Dul form of "tongue" and my Berta form of "blood". Two items ("fire" and "tongue") are about equally close to Berta and Gaam. It is easy to see how Evans-Pritchard was led to classify AKM with Berta on the basis of three or four items out of ten being almost identical or very similar to Berta words. He did not include Gaam in his comparative lists, however (Evans-Pritchard 1932/9-12) and thus did not see that the case for Gaam was just about as strong as for Berta (although he mentions on p. 53 that vocabularies of Ingessana are available). On larger samples (one hundred interns) I found the results to be about as follows: AKM among themselves

63% in common

AKM against Gaam

34%

AKM against Berta

17%

Gaam against Berta

14%.

Evans-Pritchard was misled by using a list consisting of 79 nouns and the first ten numerals. Loan words from Berta have entered this part of the lexicon to an extent to skew completely the distribution in a larger and more representative lexical sample. In the larger sample, AKM has twice as much in common with Gaam as with Berta (and Gaam and AKM are about equally distant from Berta), while in the smaller restricted sample, AKM seems to be about as close to Berta as to Gaam. 4. PHONOLOGY No attempt to examine AKM phonology has yet been made. It seems that all three languages are very close to Gaam phonologically. However, some interesting correspondences are found; these will be examined in Bender (forthcoming), where phonology will be taken up in some detail. On Gaam, see also Crewe 1975. A brief outline of Gaam phonology follows. 4.1. Vowels There are six distinctive vowels in Gaam: i, u, e, o, 3, a. There seem to

54

M. Lionel Bender

be two degrees of length: short (as above) and long (shown by ii, uu, etc.). For details, see Bender and Malik 1980. Some vowel alternations that occur are: /B/ with /a/, /o/ with /u/, /i/ with /e/. Part of the alternation seems to be a matter of vowel harmony between affixes and stem vowels. Non-distinctive variants [i, u, E, o] are found (corresponding to [ i, u, e, o]). It seems that the lax-tense distinction in the low vowel has led to the phonemicizing of /9/, but this has not spread to non-low vowels. Vowel sequences which occur are: ai, 3i, au, 9u, eu, iu as diphthongs, ei, ua, and ui occur as sequences. 4.2. Consonants InterLabial Dental Vs. stops Vd. stops

P b

Fricatives

f

Nasals

m

Liquids

Dental

Retroflex

Palatal

Velar

t d «

s n l,r

Glides The slightly retroflexed alveopalatal stops are doubtfully phonemic. Glottal stop [ ?] is marginal and not distinctive ; [h, s, x ] occur only in expletives and Arabic loans. Intervocalic [g, vl are heard for /p, b, f/. /s/ shows much variation, especially in the past-tense morpheme, to [s, z, 4, 9]. Final stops are devoiced: /b, d, d, j, g/

/p, t, t, c, k/.

The phoneme /*/, an interdental lateral fricative, is rare among world languages. However, it is unstable in Gaam, alternating with [s, z, 6] and often disappearing intervocalically. It does not appear initially. Some speakers do not have it at all (since the tongue tip protrudes in the articulation of this phone, it is easy to see, if not to hear). 4.3. Suprasegmentals As noted above, vowel length is problematical. Consonant length occurs

Eastern Jebel Languages at least morphophonemically. Stress carries a minor functional load. Three distinctive tone levels exist, plus failing tone (which can be interpreted as a high-low sequence). No rising tone has been encountered. Low-tone vowels have a tendency to be breathy-voiced. 5. A FEW GRAMMATICAL COMPARISONS A summary of the best-documented "grammemes" (Bloomfield1s term for "grammatical morphemes") of Aka-Kelo-Molo (AKM), Gaam, and Berta follows. These show the undeniable cohesion of AKM with Gaam as against Berta. Of course some common East Sudanic and Nilo-Saharan patterns also show up. Most striking are the pronouns. Another East Sudanic language (Mursi, of the Surma group) is adjoined to show that the pattern is East Sudanic. Subject Mursi

Pronouns

Gaam

Aka

Kelo

Molo

Berta

1 sg.

api

aan

ee

95

ar)

ali

2

ijle

oon

in

110

in

ggo

3

nor|

een

Ena

een (?)

rjine

eene

1 pi.

age

a'gan

Ege

ay

Oy

ha9aT|

2

ige

o'gon

Egu

uu

uu

ha9u

3

yok

'eg en

Ege

igeega (?)

?

mere

Object pronouns are similar to the above. Possessive suffixes are similar, as are verb-affixed pronouns. In addition, Gaam and all AKM languages have a singular possessed/plural possessed distinction which Berta does not. In Gaam, plural has a -gS extension to the suffix, in AKM there is more variability, but it is basically -n- singular and -0- or -gplural. Mursi has the same pattern (Turton and Bender 1976:541). Interrogatives are also fairly striking:

M.

56

Lionel

Bender

Interrogatives Mursi who?

what?

Gaam

Aka

Kelo

Molo

Berta

r)a

(na)r)ai

inu

ndolo

/rjaaga

/(ka)gai /iku

ini

/jliigin /jligE

Jli /jlani

d3i

uru

w£yu

uru

da

-tu

inda

inde

whence:

bidi,

noi/gyoi nai

?or)

when?

mi nag

where?

uuri

Jliin

/sini (?)

widi Q (Particle

-be

-ra

?i (?)

-

'arm/hand 1 a-

£-166

•eye*

o-vbt

'leg'

E-kSS

'tooth'

x-

a-lo a a-ki

Williamson (1969) demonstrates that the noun prefix system of Degema, a Delta Edoid language, passes the same tests as a number of the Benue-Congo languages. Elugbe (1976) amplifies the Degema noun prefix system and shows convincingly that noun prefixes are determined not by any phonological rule but by gender based largely on semantic classification. It can be argued that there is nothing particularly non-Kwa about the Degema prefix system because it lacks a matching concord and because Degema is spoken in the eastern Niger Delta where it is surrounded by Ijo and some Benue-Congo languages.

Noun Class Prefixes The case of Oloma, described by Elugbe and Schubert (1978) , is not subject to such an argument. As noted above, Oloma is one of the five Edoid languages on which Koelle has data in his 1854 work. In Koelle's data, some nouns are recorded with CV- prefixes. Thomas noted this in his 1910 work (though he apparently mistook Oloma for Ghotuo). Westermann (1926) drew attention to the gV- and rV- prefixes in Koelle's data and suggested that they were probably related to the li and ki prefixes of other class languages. Oloma is a North-western Edoid language spoken in the hills in the Northernmost parts of Bendel State. Elugbe and Schubert (1978) demonstrate convincingly that it has an elaborate noun class system: lexically determined classes, each marked with its own prefix, and concord. (4) Oloma: a.

b.

c.

d.

i-kphò

'river'

ii. ó-kphó

i. ó-kphò

3-nò

'this river'

iii. I-kphó

E-ni>

'these rivers1

i. ó-gbhò

pi.

I-gbhà

'wizard'

ii. ó-gbhS

à-nò

'this wizard'

iii. 1-gbhó

É-nS

'these wizards

lS-fE

'house'

i. à-fE

pi.

pi.

ii . à-f E

à-nò

'this house'

iii. 15-fE

lÉ-nS

'these houses'

à-kà

'tooth'

i. lE-kS

Pi-

ii . l£-k5 iii. a-kó e.

i. lE-cà ii . lE-cà iii . à-cà

pi.

ló-nS

'this tooth'

à-ni>

'these teeth'

à-cà

'egg'

lS-nS

'this egg'

E-nS

'these eggs'

In (4) we see that the choice of demonstrative prefix is determined by the head noun. We are therefore dealing with a clear case of concord. What is more, the concord elements are neither identical with nor predictable from the noun prefixes. Thus 'river' and 'wizard' both have an o- singular prefix paired with an i- plural prefix; but they attract different demonstrative concord markers in the singular. (Cf. in addition the case of (4) diii and (4) eiii.) Even so, it is obvious that concord is already very much reduced in

62

Ben Elugbe

Oloma: Elugbe and Schubert point out that although there are over 14 singular/plural noun prefix pairings, the concord markers identified are only those in (5): (5)

sg.

o

lo

gho

pi.

e

le

ghe

sg./pl-

a

More importantly, each class in Oloma attracts the same concord marker in all constructions: once we identify a concord element for a given class in a given construction, we have identified the phonological shape of that concord element in all other constructions. Adetuyi (1979) and Oyale (1979) report that Emhalhe (Somorika), spoken on a ridge adjacent to that on which Oloma is spoken, has noun classes and concord. Obiomah (1979) reports an equally interesting, if a little less elaborate, noun class and concord system for Ibilo, an Okpamheri dialect. Ibilo is spoken at the northern foot of the hills on which Oloma and Emhalhe (Somorika) are spoken. These hills used to be called the 'Kukuruku hills' then later 'Afenmai hills'. It does appear that the seclusion of these languages in the hills has made them to be less innovative in their noun class and concord systems. In point of fact, some vestige of concord is to be found in the average Edoid language. Amayo (1975) discusses the subject concord marker in Edo (Bini). In the other Edoid languages on which I have data, the presence of a subject concord marker is easily determined as is the evidence of concord in the following demonstrative construction (6) in Ghotuo, a Northcentral Edoid language: (6) Ghotuo (NCE): u-bele

'calabash'

i-

o-ghumhi

1

'slave

i-

o-ka

'maize'

e-

ee-o

'eye'

a-

ubele

khi S-nS

'this is a calabash'

lbele

khi e-ni

'these are calabashes

oghumhi

khi O-nS

'this is a slave'

ighumhi

khi e-n£>

'these are slaves'

oka

khi S-nS

'this is maize'

eka

khi e-ni

'this is maize (pi.)'

EEo

khi S-nS

'this is an eye ball'

¿o

khi e-n£>

'these are eyes'

Noun Class Prefixes Until the discovery of Oloma and the Ibilo and Emhalhe (Somorika) dialects of Okpamheri, all North-western Edoid languages, it was thought that concord in Edoid was restricted to subject concord markers and to number agreement between nouns and their determiners. 3. THE NOUN PREFIX IN PE There has been very little attempt before now to compare noun prefixes across languages in the Edoid group. Westermann's 1926 attempt goes only as far as putting Koelle's 1954 and Thomas' 1910 items side by side. He was obviously intrigued by the rV- and gV- prefixes which, as we noted above, he compared to the li- and ki- of other class languages. Westermann's flirtation with this matter was hampered by the fact that he had no data of his own. Elugbe (1973) reconstructs noun prefixes for PE but makes no attempt to relate these to anything outside Edoid. He reconstructs only three plural prefixes: i-, I- (predictable variants of the same plural prefix) and a-, totally unpredictable and restricted to a few parts of the body. In his study of the Ekpheli dialect of Yekhee (Etsako), Elimelech (1976) notes the phenomenon of pluralization by vowel prefix change. He notes that some are phonologically predictable while a few are unpredictable. A number of the unpredictable plural forms are in a- and involve body parts. Elimelech concludes that this bit of his data might be explained in historical terms as evidence of a remnant of the Proto-Niger-Congo ku-/a- pair. Unfortunately, he also implies that the other cases which are phonologically predictable are not evidence of an earlier noun class system. In (7) below, the reconstructed noun prefixes in Proto-Edoid are presented. The reconstruction is based on the assumption that the prefixes observable in present-day Edoid languages reflect an earlier more elaborate noun prefix system, and that the situation now obtaining in the modern Edoid languages is the result of processes of assimilation, vowel harmony, class shifting and levelling, etc. (7)

sg. -prefixes

pi.

CV-: ghU-

V-:

dhl-

-prefixes AI-

single-class prefixes V-:

AIU-

64

Ben Elugbe V-:

UIEOA-

Each capital letter in (7) is a summary of two alternants determined by vowel harmony. Thus U- represents an u - to alternation; I = i - I ; E = e 0=o

~E;

- O; and A = a - a. No Edoid language on which data are available has

shown any evidence of the involvement of tone in the process of pluralization by noun prefix alternation. The PE noun prefixes listed in (7) are paired as in (8). The numbering follows no clear order: 1.

ghU-/A-

2.

U-/A-

3.

U-/I-

4.

dhI-/A-

5.

I-/A-

6.

0-/A-

7.

O-/I-

8.

E-/I-

9.

A-/I-

10.

A-

11.

I-

12.

U-

Since the concord has decayed a great deal more than the prefix in present-day Edoid languages, it will not be easy to reconstruct the concord elements of PE. In any case, no attempt has been made to reconstruct concord as part of the present effort. We shall therefore refer simply to PE noun prefixes and noun prefix pairings rather than to noun classes or genders which are set up on the basis of criteria that normally include concord. 4. PE NOUN PREFIX PAIRINGS COMPARED WITH PB AND PCB Williamson (1969) applied to Degema, a DE language with an elaborate noun prefix system, the set of proposals which were devised by the BenueCongo Working Group for comparing noun class systems of Benue-Congo languages. The result appeared in that group's 1969 volume, Benue-Congo noun

Noun Class Prefixes

65

class systems, published by the West African Linguistic Society and the Afrika Studiecentrum. Degema employs a noun prefix system more elaborate than those of a number of Benue-Congo languages. Hence, in the manuscript version of Williamson's Degema paper, she asks: "is it possible to devise criteria whereby the noun-class systems of Kwa can be distinguished from those of Benue-Congo?". De Wolf (1971) compares Heine's 1968 reconstructed Proto-Togo Remnant prefix pairs with his own reconstructed pairs for Proto-Benue-Congo. He concludes, on the basis of striking resemblances between the two, that "... Togo Remnant is definitely part of Benue-Congo ..." It would be interesting to know what he would have said had he known what the reconstructed prefix pairing in Proto-Edoid is like. For, as it turns out, not only are there Edoid languages that are clearly noun class languages, the reconstructed PE noun prefix system also shows striking semantic as well as phonological resemblances to the reconstructed PB and PBC noun prefixes. In what follows, an attempt will be made to show which PE prefixes correspond to which in PBC as reconstructed by De Wolf (1971) and in PB. In order to make the exercise easier and more convincing we shall compare the reconstructed noun prefixes in their usual pairings. 4.1. ghV-/A- (of. PB 15/6 and PBC (ku/a)) This is where we find 'arm/hand' and 'ear', giving the impression that it is a kind of 'remnant body parts' pair. Individual Edoid languages include some other nouns such as 'leg/foot', 'hat/cap', etc. here. 4.2.U-/A- (of. PB 15/6 and PBC (ku/a)) In my earlier reconstruction, I made no distinction between this and the ghU-/A- pair. However, work on Emhalhe and Oloma has forced me to re-examine the evidence and it would appear that a distinction existed between these two at the PE level even if most Edoid languages do not now retain that distinction. The nouns here are: blood, ground, (possibly 'hat/cap', see 4.1. above), head, horn, house, liver/heart, marriage/husband, oil palm, rope, tail, tongue, year, and vein. 4.3. V-/I- (of. PB 3/4 and PBC (bu/i)) Although some Edoid languages showed a consonant gh- or w- here in items such as

1

boat/canoe', 'hair', and

1

grey hair1, I have not used that as a

66

Ben Elugbe

reason to split this pair into two —

possibly ghU-/I- and U-/I- —

as we

did in separating 4.1. and 4.2., because there is no clear comparative advantage to be had from such a splitting and because there has been some class shifting between this and 1 and 2. It is possible that nouns now reconstructable with ghU-/l- prefixes were original ghU-/A- nouns which have changed an A-plural prefix to I- which has become the most widespread and generalized plural prefix. The nouns to be found here include: back, boat, bone, cloth, grey hair, hair, leaf, moon, navel, skin, song, stone, tree. 4.4.dhI-/A- (of. PB

and PBC (H/aJ)

No Edoid language has a stop reflex of *dh as a prefix consonant. In a prefix position, *dh normally has lateral or r- type reflexes. However, evidence for the reconstruction of a consonant in this prefix is often quite straightforward (see section 5 below and appendix C at the end of this paper). The nouns in this pair include: belly/stomach/intestine, egg, eye, name, tooth, war. 4.5.1-/A- (of. again PB

and PBC Hi/a))

The relationship between this and the dhI-/A- pairing is not clear. I include here those nouns which show no clear evidence of a consonant in the prefix. Some of them may yet be found to belong to the dhI-/A- pairing. Until then, the following items are better entered under the I-/A- pairing: cloth, navel(?), neck, nose. 4.6. 0-/A- (of. PB 1/2 and PBC (u/ba)) I have only a few items here at the moment: corpse, (traditional) doctor, person/woman, thief. It looks like a persons class. 4. 7. 0-/IThis is a difficult pairing to associate with any of the well-known PB genders. It contains the nouns chicken, child, cock, cotton, father, hunter, king, and man. There is a possibility that 'hunter' and 'king' belong elsewhere, possibly the A-/I- pairing. 4.8.E-/I- (of. PB 9/10 and PBC (i/i))

Noun

Class

Prefixes

Here we have animal/meat, antelope, bag, body, elephant, fish, goat, leopard, monkey, river and tortoise. As the list shows, this is basically an 'animals' pairing. 4.9.

A-/IThis is a sizable prefix pairing in my reconstructions. Unfortunately, it

is not a pairing that one easily associates with the better-known PB genders. I- is the most common plural prefix in Edoid, and as the 3- alternant of A- has many varied reflexes, it is possible that some of the items here belong elsewhere: axe, bird(?), breast(female), dog(?), earth(soil) hat/cap (?), he-goat, housefly, hunter (?), iron, king(?), louse, market, monkey(?), mouth, night, oil, penis, rain, saliva, water, wind. Two nouns, 'bird' and 'dog' suggest that there may have been a ghA-/dhUprefix pairing in PE. One would require more evidence to recognize this suggested pairing. The possibilities of comparison are with PB 12 for the singular and PB 5 or 10 for the plural, giving us comparison with a rare PB 12^5 or an unusual PB 12^10. Its connections with De Wolf's reconstructed PBC genders are even more elusive. 4.10. A- (of.

PB 6 and PBC a and ma)

This unpaired prefix characterizes nouns like ashes, drink, sunshine

—

and, possibly, saliva. It looks like a mass noun prefix. 4.11.

I-

Here I have "ashes' as a possibility and 'faeces'. Comparison with PB and PBC must be based on more data. 4.12. U'Sleep' and 'fear' are the two nouns I have identified here. Comparison with PB 14 is possible, so it is with PBC bu. The evidence available suggests that there were more prefixes in PE. However, one would require more than the data now available on nouns (about two hundred of them) to be able to reconstruct more prefixes. Even so, what is available and displayed above is interesting not so much for its size as for its consistent comparability with reconstructed PB and PBC class pairings both in phonological shape and in semantic content.

68

Ben Elugbe

5. DID PE EMPLOY A SUFFIX? 5.1. The typical PE stem was of the shape C^V((C^) V). There were, however, restrictions on the -C^- position. Only the weak consonants -bh-, -dh-, -G- and -mh-, -nh-, and -N- could occur there. In addition, -N- may not have been fully realized except as nasalization of the stem vowels around it. In Yekhee (Etsako) and Uneme, a group of nouns are found with a final -li syllable. In cases where we can reconstruct any of these nouns to the PE level, we find that: (a)

they belong to our PE dhl- or I- prefix class of nouns ,-

(b)

the -li syllable in them cannot be linked or traced to the reconstructed

PE stem. In (9) below we compare some reconstructed PE noun stems from the dhl-/ A- pairing with their reflexes in the Ekpheli and Uzairue dialects of Yekhee (Etsako) and in Uneme: (9)

English

PE

Ekpheli

Uzairue

egg

dhI-kiN9

É-kèlì

É-k££

Uneme É-kEnhi ukp-è-kò

tooth

dhl-kwn

&-kòli

£-kòò

eye

dhl-dhu

ukp-è-lò

ukp-E-lòò

ukp-E-lòlì

à-lò

à-lò

à-lò

PE

Ekpeli

Uzairue

Uneme

eggs

A-kiN9

é-kElì

é-kEE

é-kEnhì

teeth

A-kcjN

à-k£>

a-kS

à-kO

eyes

A-dhco

ikpà-lò

lkp-à-lò

lkp-à-lò

à-lò

a-lò

à-lò

face The plural form of (9) is (10); (10) English

faces

From (9) and (10) we observe that, at least in the singular forms, there is a final syllable, basically -li, which is not a reflex of anything in our reconstructed PE stems for these items. This final syllable remains even where the lateral has been lost and the final -i is assimilated to the preceding stem vowel - thus: (11)

(a)

£-k6li

>

(b)

E-kEi3

>

(c)

£-k£E

We also observe that, in Uneme, in stems with a -C^- nasal -N

this extra

-li syllable is realized as -nhi, where nh is the nasal counterpart of 1. (See the appendix for the comparative series on which the reconstruction of some of the nouns in this class is based.

69

Noun Class Prefixes In view of (9) and (10), two questions arise: is the -li in these examples a class marker? If not, is it some kind of suffix acquired through a particular construction, especially one involving a post-nominal determiner? We may begin to try answering the first question by tackling the second. There is in Yekhee a demonstrative stem -li from which we have 3-11 'that one' and e-li 'those ones'. It is tempting to suggest that the suffixing of this demonstrative stem to a noun with its subsequent regularization is the source of this -li in the nouns under consideration. When we actually consider the demonstrative construction using this particular demonstrative in Yekhee and Uneme, we have (12): (12)

English

I. egg

Ekpheli

Uzairue

Uneme

e-keiì

e-kee

C-kEnhì

the egg

3n£-k£li

Sn£-k£E

3n£-k Ènhì

that egg

5n£-k Eli

sne-keeiì

ón£-k£nhinhi

those eggs

éné-k£lì

éné-k££li

éné-k£nhlnh£

II. tooth

è-k51ì

è-kàà

ukpà-kò

the tooth

5n£-kSli

5n£-kSS

Ónukpà-ki)

that tooth

3n£-kàlì

3n£-k351ì

3nukpà-k5nhi

these teeth

énà-k51ì

éna-kólì

énikpà-kSnhl

III. eye

ukpè-lò

ukpè-lòò

tìkp£-lòlì

the eye

5nukp£-lò

5nukpÈ-lòò

ónukp£-lòlì

that eye

ónukp£-lólì

3nukp£-lóólì

ónukp£-lólinhl

those eyes

énikpà-lólì

énikpà-lólì

énikpS-lónhl

the face

5nà-lò

énà-lò

5nà-lò

that face

ónà-lólì

énS-lólì

3nà-16nhi

ó-bà

telk

bone white

one

puskni

'one white bone'

Welmers (1973: 209): the ancestor of these languages employed both prefixes and suffixes. Some languages have lost the suffixes while others have lost the prefixes. However, a small number of languages reflect the original situation in that they still employ both prefixes and suffixes.

Noun Class Prefixes (c)

Greenberg (1977) : there was neither a prefix nor a class marker at the beginning. A class marker developed from a demonstrative through two stages of an article. This class marker, originally without order relative to the noun, became sometimes a prefix and at other times a suffix.

5.2. Professor Hoffmann's hypothesis is plausible and possibly applicable here. Oloma and Emhalhe demonstrate that even demonstrative stems attracted concord (The examples are from Oloma): (14)

gha-wa

ghS-nhi

'that dog'

l£-k3

15-n3

'this tooth'

lo~f£

le-nhi

'those houses'

a-ko

5-nhi

'those teeth'

o-kpho

3-n£>

'this river'

We can apply Hoffmann's hypothesis here if we assume that the dhl-prefix class in Yekhee and Uneme attracted a dhl-concord element which was already realized as li- in these languages as in (15), ignoring the prefixed 4

definite article OnV in this case: (15)

Early Yekhee/Uneme

Later Yekhee

Later Uneme

English

le-ke li-li

e-ke li-li

e-ke nhi-nhi

that egg

lE-lo li-li

e-lo li-li

e-lo li-nhi

that eye

lE-ka li-na

e-ko li-na

e-ko nhi-na

this tooth

From the developments in (15) we then have to-day's (16)"': (16)

English

Yekhee

Uneme

egg

£-kEli

E-kEnhi

eye

E-loli

E-ldli

tooth

E-koli

E-kS

Assuming that this is the explanation, our conclusion would be that the apparent suffix in these nouns is a reflex of a concord element. 5.3. Welmers' hypothesis derives its strength from the fact that each and every one of to-day's noun class systems, be they prefixing, suffixing or both, can be explained in terms of it. However, it is also founded on the premise that every suffix, every prefix in the noun class systems of NigerCongo (and indeed Niger-Kordofanian) is a class marker. Yet what the Hoffmand and the Greenberg hypotheses suggest is that this need not be.

72

Ben Elugbe If we were to accept that the -li in Yekhee and Uneme is a remnant of a

class marking suffix, we would have to conclude that only the dhl- prefix in PE had a matching suffix -dhl which has now been lost in other areas of Edoid. But in point of fact, it is impossible to determine whether in the case under review we are dealing with a class marking suffix or a postnominal concord element re-interpreted as a nominal suffix. 5.4. Greenberg's position is that by the very nature of its origin, the class marker in Niger-Kordofanian could not have been both a prefix and a suffix; that, for a start, it had no order relative to the noun; but that it gradually became fixed in order through being used as an article. The obvious conclusion is that none of to-day's prefix plus suffix noun class systems (the best examples are to be found in the Gur branch of Niger-Congo) reflects the state of affairs at the Niger-Kordofanian level. I find it easier to discuss this in relation to the other two hypotheses and the data in hand. Greenberg's hypothesis, I suspect, will be found to be compatible with many other hypotheses. Indeed, it would be compatible with the Welmers one if it did not imply that the original class marker could not have been both a prefix and a suffix. In relation to the Hoffmann hypothesis, I would like to draw attention to another aspect of the Greenberg hypothesis. He says that a basic prefix system may be renewed by suffixing and vice versa. He does not say in this short 1977 work what the various ways might be of renewing (or re-inforcing) one set of prefixes, say, with suffixes, but he does suggest it might be by determination. One would like to add that the Hoffmann hypothesis relates to this renewal stage. Hoffmann's hypothesis suggests that the ancestor of these languages may have been a prefix language which picked up suffixes via concord elements attached to post nominal determiners. We would conclude in the case under review that we are most likely dealing not with a suffix class marker but with a suffix picked up along the lines discussed in section 5.2. above. We would further suggest that this neither invalidates nor is invalidated by the Greenberg hypothesis. In fact, they complement each other. 6. CONCLUSION This discussion has been in three parts. In the first part we have shown that there are noun classes in Edoid. In the second part we have shown that the reconstructed noun prefix pairings in PE - which we would have called

Noun Class Prefixes genders but for the fact that concord has not been reconstructed so far compare favourably with PB and PBC pairings in both phonological shape and semantic content. Finally, we have shown that there are suffixes in a part of Edoid, but that these may have been acquired through concord elements attached to demonstratives. If then the Edoid languages, which are considered typical Kwa, have all this, how 'Kwa' are they? The discussion has been with us for a long time now and every new piece of evidence as in this case, calls for the merging or, at any rate, a re-classification, of Niger-Congo as it relates to Benue-Congo and Kwa (Elugbe and Williamson:1977) or as it relates to Benue-Congo, Kwa, and Gur (Stewart: 1976). NOTES 1. Throughout this paper the addition of an h to a symbol indicates a weak variety of the sound represented by that symbol. In Oloma, Uneme, and Yekhee, there is no need to add this -h to the lateral 1 as the only lateral is a weak one. In the case of weak stops, they are frequently in free variation with their fricative counterparts. 2. The vowel system of PE involved two sets of vowels based on vowel harmony of the West African Pharynx-width type: Set 1 expanded pharynx i u e o s

Set 2 non-expanded pharynx I to e o a

It is possible that our reconstructed.-N- in -C2- position in stems was realized only as nasalization of the surrounding vowels. 3. Elimelech (1976) gives (b)-stage forms for Uzairue and says that, in Ekhpeli, our (a)- and (b)-stages are in free variation. 4. We ignore any stages and minor details not relevant to the discussion. 5. It is probably after our (15) stage that Yekhee and Uneme introduced the ukpV- (pre-?) prefix. 6. Greenberg does make a reference to his 1978 paper "How do languages acquire gender markers?" I have not seen it. REFERENCES Adetuyi, A. 1979. The phonology and noun classes of Emhalhe. Long Essay for the B.A. (Linguistics) degree. University of Ibadan, Ibadan. Amayo, A. 1975. "The structure of verbal constructions in Edo (Bini)" Journal of West African Languages 10(1) : 5-27.

74

Ben Elugbe

De Wolf, P. 1971. The noun class system of Proto-Benue-Congo. The Hague: Mouton and Co. Elimelech, B. 1976. A tonal grammar of Etsako. Working Papers in Phonetics, 35. University of California, Los Angeles. Elugbe, B.O. 1973. A comparative Edo phonology. Doctoral thesis, University of Xbadan, Ibadan. Elugbe, B.O. 1976. "Noun class vestiges in Degema". Afrika und Ubersee, 59 (3): 224-33. Elugbe, B.O. and K. Schubert 1978. "Noun classes and concord in Oloma". Paper presented at the 13th West African Languages Congress, Freetown, Sierra Leone, April, 1978. Elugbe, B.O. and K. Williamson 1977 "Reconstructing nasals in Proto-BenueKwa". In Juilland (Ed.) Linguistic studies offered to Joseph Greenberg, pp. 333-63. Saratoga: Anma Libri and Co. Greenberg, J.H. 1955. Studies in African Linguistic Classification. Branford: Compass Publishing Company. Greenberg, J.H. 1966. Language families of Africa. 2nd revised edition. The Hague: Mouton and Co. Greenberg, J.H. 1977. "Niger-Congo noun class markers: prefixes, suffixes, both or neither?" Studies in African Linguistics, Supplement 7, 97-104. Heine, B. 1968. Die Verbreitung und Gliederung der Togorestsprachen. Berlin: D. Reimer. Hoffmann, C.F. 1967. "An outline of the Dakarkari noun class system and the relation between prefix and suffix noun class systems". In Colloques internationaux du centre national de la recherche scientifique, la classification nominale dans les langues negro-africaines, Aix-en-Provence (Paris). Koelle, S.W. 1854. Polyglotta Africana. London: C.M.S. (Reprinted in 1963). Melzian, H. 1942. "Zum Konsonantismus in den Dialekten der Beningruppe". Archivum für vergleichende Phonetik (6)2: 49-59. Obiomah, E.O. 1979. The structure of the Ibilo noun. Long Essay for the B.A. (Linguistics) degree, University of Ibadan, Ibadan. Oyale, M.A. 1979. The phonology and noun class system of Emhälhe. Long Essay for the B.A. (Linguistics) degree, University of Ibadan, Ibadan. Stewart, J.M. 1976. Towards Volta-Congo reconstruction. Leiden: Universitaire Pers. Thomas, N.W. 1910. Anthropological report on the Edo-speaking peoples of Nigeria. Part II: Linguistics. London: Harrison and Sons.

Noun Class Prefixes Weimers, W.E. 1973. African language structures. Los Angeles: University of California Press. Westermann, D. 1926. "Das Edo in Nigérien". Mitteilungen des Seminars für Orientalisohe Sprachen Volume 29, Part III: 32-58. Williamson K. 1969. "Degema". In Benue-Congo noun class systems, pp. 118Leiden: West African Linguistic Society and the Afrika Studiecentrum. (Taken from the unpublished 1969 manuscript version: "Degema: a test of the new proposals for comparing noun-class systems").

76

Ben Elugbe Appendix A

Proto-Edoid, Proto-Bantu and Proto-Benue-Congo class pairing at a glance.

PE

PB

PBC

1.

ghU-/A-

15/6

ku/a

2.

U-/A-

15/6

3.

U-/I-

3/4

bu/I, u/ti

4.

dhI-/A-

5/6

li/a

5.

I-/A-

5/6

li/a

6.

0-/A-

1/2

ù/ba

7.

0-/I-

3/4?

bu/i, u/ti? i/i 7

8.

E-/I-

9/10

9.

A-/I-

12/5?

10.

A-

11.

I-

6 ?

12.

U-

14?

ku/a

ma, a? p bi, bu?

77

Noun Class Prefixes Appendix B PE prefix pairings and their contents. 1.

2.

3.

ghU-/A-

dhI-/A-

-ßo

arm/hand

-chOGI

ear

-phaNI

belly/stomach/ intestine

-kiN8

egg

-dhu>

eye

-tI - (-3-la?)

blood

-ni

name

-to

ground, earth

-ko)N

tooth

(soil)

-khojmhl

war

U-/A-

-thu

hat/cap

-chismhi

head

-kwhanhI

horn

-kvraN

cloth

-bhaGrl-baGI

house

-kh8N-

navel

-tfis

liver/heart

-chsNi

nose

-cfomhl

marriage/husband

-di

oil palm

-dhuNi, -dhiNi

rope

-dhimhi

-thismhi

tail

-&CD

-dhamhl

tongue

pe rs on/woman

-kwe

year

-kwochi, -kwsihi

-nla, -nhla

vein

-ghi, -j-hi

thief

5.

6.

7.

U-/I-

I-/A-

0-/Acorpse (native)doctor

0-/I-

-ks

back

-khokho

chicken

-ko

boat/canoe, mortar

-mo

child

-kwhuNs

bone

-kwa

cock

-kw3N

cloth?

-dhudhu

cotton

-daNI

grey hair

-si-?

father

-txoN

hair

-chu3

hunter

-61

leaf

-vis,I-vie

king, priest

-ki

moon

-chl-

man

-khsN-

navel?

-phlNa

skin

-yodho, viQNi

song

-doGi

stone

-thaNI

tree

78 8.

9.

Ben Elugbe 10. A-

E-/I-nhamhl

animal/meat

-mhuNS

ashes

-thuSNi

antelope

-yoN

drink

-kwa

bag

-vuN8

sunshine

-gwINl?

body

-ci8Ni

saliva?

-ni

elephant

-chiNénhi, -ch(i)3nhi

fish

-bhoil

goat

-kweN

leopard

11. I-mhuNa

ashes?

-C3N9

faeces

-mhunha

monkey

-da

river

-bhlchlN(a)

sleep

-ghuNi

tortoise

-pIN

fear

A-/I- ÍU3NÍ

axe

-pl-

bird?

-VÍN3NS, -frhiNaNa?

breast (female)

-bhoja

dog?

-to -thu -ki3 -khINa -chu9 -khunhi -vis, -vie -guGi? -ki -mhunha -nua -owNa -bhi-chu3 -chi9 -ciaNi -mIN -fofo

earth/soil? hat/cap? he-goat housefly hunter? iron king, priest? louse market monkey? mouth night oil penis rain saliva water wind

12. U-

Noun

Class

Prefixes Appendix C

Comparative series on which the reconstruction of the dhI-/A- prefix pairing is based are presented here. The six nouns in this pairing are important for our arguments in section 5. Tone is phonetically marked. In rare cases, my data suggest that the tones for a particular item be re-checked. In such cases tone is not indicated. Items in brackets are considered non-cognate. Items with a following question mark are doubtfully

a- 'belly, stomach, intestine 1

1. P r o t o - E d o i d * d h l - p h a N I DE:

SWE:

NCE:

Degema

E-fal

Egene

E~fai

Eruwa

i-va

Isoko

i-va

Okpe

S-pha

Urhobo

(e-vu)

Uvbie

(e-vu)

cognate.

a-

([v]

Edo

(e-ko)

Aoma

e-ha

Auchi

Ile-fai

Uzairue

S-faa

is a breathy-voiced v)

1

intestine 1

(item taken from Elimelech 1976, no plural given)

NWE:

(pi.)

Ekpheli

e-fali

Uneme

e-f&

'intestine 1

Ghotuo

e-fa

'intestine'

Oloma

5- fa

Emhalhe

(we-gua)

Ibilo

(u-gu)

Uhami

e-fa

Ehueun

e-fa

Ukue

e-'fa

£'stomach'

(pi.)

Ben

Elugbe

Proto-Edoid DE :

SWE :

NCE:

NWE :

Degema

u-kle

Egene

i-kìè

Eruwa

£-kè

Isoko

é-ké

Okpe

i-yé

Urhobo

u-k£

Uvbie

u-ké

Edo

è-k£

Aoma

£-kè

SWE:

NCE:

3-

•egg'

Auchi

£-k££

Uzairue

£-k££

Ekpheli

£-k£lì é-

Uneme

6-kenhi é-

Ghotuo

É£-ke

e-

Oloma

l£-cà

a-

Emhalhe

rà-cà

t-

Ibilo

ilhà-cà t-

Uhami

É-kàjlÈ

Ehueun

ì-hÌ£

Ukue

ì-k£

Proto-Edoid DE :

*dhi-kìNS

*dhl-dhaj

Degema

à-cldj

Egene

à-ctò

Eruwa

a-reo

Isoko

È-là)

Okpe

E-rò

Urhobo

à-lhò

Uvbie

à-J?ù

a-

Edo

à-lò

Aoma

È-ò

Auchi

(ukp-)£-lò

• eye ' 'face '

à-

Noun Class Prefixes NCE:

NWE:

Uzairue

(ukp-)E-lòò

(ikp-)à-lò

Ekpheli

(ukp-)£-lò

(ikp-)à-lò

Uneme

(ukp-)£-lòlì

(Ikp-)à-lò

Ghotuo

È£-ò

à-

Oloma

(u-nògò)

Emhalhe

u-lhò

Ibilo

u-lhò

Uhami

à-rù

Ehueun

(à-dù)

Ukue

( è-dù)

4. Proto-Edoid DE :

SWE:

NCE:

NWE :

i-ni

9-

'name'

Degema

ì-nl

Egene

ì-nì

Eruwa

(io-vàrà)

Isoko

(

à-

Oloma

lÈ-kÒ

à-

Emhalhe

rà-kò

àà-

Ibilo

à-kò

Uhami

à-kò

Ehueun

à-kù

Ukue

à-kù

6. Proto-Edold DE:

*dhl- kON

*dhl-khtjnhl

Degema

I-qcjm

Egene

£-wùjncò

Eruwa

not available

Isoko

È-mtj?

Okpe

not available

Urhobo

o-fovhl?

Uvbie

not available

Noun Class Prefixes NCE:

NWE:

Edo

o-kuoo?

Aoma

(

Uzairue

o-khoo

Ekpheli

o-kholi

Uneme

6-khonhI

Ghotuo

o-ho

Oloma

16-humhu

Emhalhe

ra-humh~

Ibilo

ilha-hCB

Uhami

£r £-humu

Ehueun

Ere-hii

Ukue

£-h3mi

Chapter 5

Defaka, Ijo's Closest Linguistic Relative Charles E.W. Jenewari

INTRODUCTION Ijo is notable among Niger-Congo languages for its relative isolation within the language family. Westermann (1927) tentatively classified it as belonging to the Kwa group; and Greenberg, after placing it in a group of its own in his 1955 classification, also placed it among the Kwa languages in 1963 (Greenberg, 1963). Greenberg (1963) uses the term 'Kwa' to refer to one of six coordinate branches of the Niger-Congo family of languages, the other branches being West Atlantic, Mande, Gur, Benue-Congo, and Adamawa-Eastern. But he expresses doubts as to the validity of treating Kwa and Benue-Congo as coordinate branches of Niger-Congo, for the two language groups 'are particularly close to each other' (p.39). And quite recently, Elugbe and Williamson (1977) have reached the conclusion that 'Benue-Congo and Kwa form a single subfamily of Niger-Congo', which they label 'Benue-Kwa'. Williamson (1965:4, 1971:281, 1972:1-2) doubts the classification of Ijo as a Kwa language, for 'Ijo differs in many respects from both typical Kwa languages and typical Benue-Congo languages', and tentatively suggests that 'if Kwa and Benue-Congo are to be regarded not as separate, co-ordinate branches, but as one large branch, then Ijo will fit easily into this large branch as a special sub-group, and it will not be necessary to assign it arbitrarily to either Kwa or Benue-Congo' (1972:2). In a recent reclassification of Niger-Kordofanian languages Bennett and Sterk (1977) have independently proposed a classification of Ijo that is consonant with Williamson's suggestion: Ijo is classified as constituting one of the three branches of South Central Niger-Congo (henceforth SCNC), which subsumes the former Kwa and Benue-Congo branches; the other two branches are Western SCNC (equivalent to Greenberg's Western Kwa, minus Kru) and Eastern SCNC, comprising the rest of Kwa and all of Benue-Congo. Thus, according to Bennett and Sterk, Ijo is less closely related to languages such as Yoruba, Edo, and Igbo than these are to Greenberg's Benue-Congo languages. Bennett and Sterk's subgrouping is however only tentative, as is made clear in the following statement (1977:251):

86

Charles E.W. Jenewari 'The position of Ijo like that of Kru is ambiguous .... Some evidence links it with East SCNC, other data indicate that it should be classed as a coordinate branch, and a position as an independent branch of Central Niger-Congo or even Niger-Congo is not inconceivable. Many of the possible cognates are much changed, thereby distinguishing Ijo sharply from the remainder of SCNC'. It is not clear whether or not the ambiguity in the position of Ijo can

ever be satisfactorily resolved. However, it is encouraging to report that a new Niger-Congo language which appears to be the closest linguistic relative of Ijo has recently been dicovered in the Niger Delta. The language in question is Defaka, better known to the outside world by the Ijo appellation Afakani. It is expected that the discovery of this language will lead to some refinement of the Bennett and Sterk classification. This paper reports for the first time data from Defaka, and is divided into two parts. The first part provides information on the social background of the Defaka people, including language use in the community. The second part, which is the main thrust of the paper, deals with the nature of the affinity between Defaka and Ijo.

Part 1 Until very recently, the Defaka were not recognized as a separate group of people. Alagoa (1972:165) is probably the first scholar to report about them in the historical literature: "The town of Nkoroo has within it a ward, Afakani, whose members speak a completely different language. The Afakani language and traditions suggest a connection with groups like the Abuloma of Okrika or Udekama (Degema) of Kalahari'. His impression about their cultural-linguistic relationship is, however, in error, if the above statement is taken literally. For, even a cursory inspection of lexical items in Defaka, Abuloma, and Udekama (Degema) clearly reveals that Defaka is not specially related to either of the other two languages. Nor do the Defaka people ever refer to any historical (ethnic) connections with either of these two peoples. But this was not really the interpretation the author intended (Alagoa, personal communication). What he wanted to point out is the common settlement pattern in the present-day Eastern Niger Delta in which certain nonIjo-speaking communities exist as cultural-linguistic enclaves within the political domain of their respective numerically superior I jo-speaking neighbours.

Defaka The Defaka are a small community whose territory is fragmented geographically. One portion of them live, as Alagoa correctly noted, in the Afakani ward of Nkoroo town. The dominant population in this town is that of the Nkoroo, an

I jo people whose closest linguistic relatives are the Kalahari,

Okrika, and Ibani (Bonny). To the South of Nkoroo town lie the Opobo and the Andoni; to its North lie the Ogoni and the Ndoki; to its North-West lie the Okrika; and to its West lie the Bonny people. The other portion of the Defaka-speaking community live on the isolated island of Iwoma Nkoroo (founded in 1884 as an offshoot of the Afakani ward in Nkoroo) near the Ogoni town of Kono. The history of migration of the Defaka is a long narrative of the adventures of a small people constantly harassed by their numerically superior neighbours. According to the tradition narrated to the author, the original home of the Defaka was in the Iselema area, that is, in the Warri region of the present-day Bendel State of Nigeria. From there, they moved into the Central Delta and thence into the Eastern Delta, where they lived close to the Abuloma people in the Okrika territory. Later they abandoned this Okrika settlement and lived close to the Udekama (Degema) people in the Engenni area. Then they left for Abalama Olotombia in the Bonny territory, and later settled near Bodo in Ogoni. From there, they moved to Iyoba in the Andoni country, and later left this settlement to establish Olomama Nkoroo (Old town). From Olomama Nkoroo, they finally moved to the present-day Nkoroo town. The Nkoroo relate a similar tradition of migration. From Gbelegbeleala in the Central Delta, they moved into the Okrika territory, and later left for Iyoba in the Andoni country. From there, they proceeded to found Olom i 1 Nkoroo, which is geographically contiguous to Olomama Nkoroo town. Alagoa (1972:169) tentatively dates the establishment of the Nkoroo in the Kon-toru (Fishing River) region or Eastern Delta Fringe (i.e., the portion of the Niger Delta occupied by the Ogoni, Andoni, Opobo, and Nkoroo) before the end of the seventeenth century. One interesting point that seems to emerge from these oral traditions is that Defaka and Nkoroo peoples have been living together as neighbours prior to the establishment of Nkoroo town. Since they were at Iyoba, they have either been one community, or have lived as separate but in geographically contiguous communities. It is also likely that this kind of contact began while both of them were in the Okrika territory. It will therefore be assumed in this paper that the Nkoroo and the Defaka have been in a contact

88

Charles E.W. Jenewari

situation for an extended period of time. In spite of the fact that they are ethnically distinct peoples, the Nkoroo and the Defaka would like to be seen in foreign eyes as one people, going under a common name, Nkoroo. The Nkoroo consider the Kon-toru region an environment infested with enemies; and in this region they are the smallest in terms of numbers. According to the 1963 Nigeria Census, the NkorooDefaka people number only 5,468, distributed as follows: Nkoroo town 4,557, Iwoma Nkoroo 509, Nkoroo villages 420. The Defaka have assimilated Nkoroo culture to such a degree that, today it appears that the only cultural element that clearly bears the stamp of Defaka identity is the language. In almost every other thing - in their economic pursuits, marriage system, religious life, and recreational activities - the Defaka are indistinguishable from their culturally and numerically dominant neighbours. Linguistically, the Defaka are an interesting people. Being in frequent contact with their various numerically superior neighbours, the Defaka cannot but maintain a polyglot tradition. Most Defaka are bi- or even trilingual in the native languages spoken in the Eastern Delta Fringe, namely Nkoroo, Defaka, Igbo, Obolo (Andoni), and Kana (Ogoni). There are in fact •

2

some Defaka people who possess a knowledge of all these languages. Some Defaka people do not even know that there exists another native language, besides Nkoroo, in the community. Those who can speak the native Defaka tongue are mostly middle-aged to elderly people, and these constitute a rapidly decreasing minority. But even among these people the language is rarely spoken; speakers resort to it usually for the purpose of cutting out of communication those who they know are ignorant of it. Children grow up in the two Defaka communities speaking Nkoroo as a first language although there appears to be a stronger Defaka tradition in Iwoma than in Nkoroo town. The Defaka, as they themselves admit, do not care if a man does not speak Defaka but speaks Nkoroo well. In other words, they regard Nkoroo as essentially a language of integration. It therefore appears, looking at the present trend of events, that Defaka is being gradually pushed to extinction. One might even add that a complete shift to Nkoroo is likely to come sooner in Nkoroo town than in Iwoma. The reason for this is fairly obvious. At Iwoma Nkoroo, the Defaka are in the midst of alien people, the Ogoni, and relations between these two groups are far from cordial. Under such conditions, one would expect the Defaka to cling more intimately to their native language. For, "the small language or dialect, Michael West (1958)

Defaka tells us, 'is the natural and important distinguishing feature of the small group. Anything which, while common to the group, tends to differentiate it from other groups tends to intensify the sense of solidarity and of distinctness. Of all instruments for the intensification of group individuality, language is undoubtedly the most powerful1. After Nkoroo, the next most active language among the Nkoroo and Defaka people is Igbo. Igbo is a legacy they inherited from the political influence of Opobo, whose vehicular language is Igbo. Since the days of the Oil Rivers Trade, the Opobo have been the most dominant group in the Eastern Delta Fringe. This influence has manifested itself in the widespread use of Igbo in the whole area. Under the former Eastern Nigeria set-up, the Opobo, Andoni and Nkoroo formed one division, with Opobo town as headquarters. The language of the traditional court at Opobo which served the three communities was Igbo. Most of the teachers who taught in Nkoroo and Andoni schools were of Opobo origin; consequently, Igbo was a language of instruction in these schools. It was also the language of evangelization throughout the whole division. Today, however, Igbo has ceased to be the language of the church and of the lower classes of the elementary school at Nkoroo. But it still functions as the language of trade for the different peoples that inhabit the Kon-toru region, as well as the language of the joint NkorooAndoni court sessions at Unyeada, in Andoni clan. The native languages of the area least spoken by the Defaka are Kana (Ogoni) and Obolo (Andoni). The presence of any of these languages in the linguistic repertoire of a Defaka usually reflects the geographical location of that particular individual. The Defaka at Iwoma, because he is nearer to, and has more direct contact with, the Ogoni than with the Andoni, is more likely to speak Kana than Obolo. Conversely, his counterpart at Nkoroo, by reason of his closer contact with the Andoni, is more likely to speak Obolo than Kana.

Part 2 The various Niger Delta peoples who have been in contact with the Defaka speak languages which fall into six linguistic grcups. These are: (1) Ijo group, (2) Lower Niger group (to which Igbo belongs), (3) Delta Edo group, which comprises Epie, Engenni, and Degema, (4) Ogoni group comprising Kana, Gokana, and Eleme,. (5) Central Delta group, comprising Abua, Ogbia, Kugbo, Bukuma (or Ogbronuagum), Abuloma, etc, (6) Lower Cross group, to which Andoni (Obolo) belongs. These languages belong to the part of Niger-Congo

90

Charles E.W. Jenewari

which Bennett and Sterk (1977) call South Central Niger-Congo, which they subdivide into: Western SCNC, Eastern SCNC, and Ijo.3 The geographical position of Defaka thus suggests that it is a Niger-Congo language, in all probability, a member of SCNC. Admittedly, Bennett and Sterk have not presented an SCNC wordlist. However, an examination of lexical items in Defaka and its six linguistic neighbours (using the Ibadan 400 wordlist) easily reveals that Defaka must be placed within the same broad subgroup of Niger-Congo as its neighbours. For, Defaka shares with these languages, taken as a whole, a fairly large number of apparent cognates. The relevant lexical items are listed in Table I, with the apparent cognates underlined. I have excluded from the comparison items which appear to be exclusively shared by Defaka and Nkoroo (or Eastern Ijo - the subgroup of Ijo which comprises Kalahari, Okrika, Ibani, and Nkoroo). These include items like one (numeral), star} mountain, and certain words for items of material culture. The sources of the data are as follows. For Ijo X have used Williamson's Proto-Ijo reconstructions (personal communication). For Lower Niger, I have used Igbo (comprising the five dialects of Owerri, Ohuhu, Orlu, Umuoji, and Ukwaali) in Armstrong (1967); data are generally cited in the Ohuhu dialect unless indicated otherwise. For Delta Edo, I have used Ben Elugbe's Proto-Edoid reconstructions (personal communication) supplemented with data from Degema (De), Epie (Ep), and Engenni (En) in Thomas and Williamson (1967). For Central Delta, I have cited data from almost all the varieties, more specifically, from Abua (A), Odual (Od), Kugbo (Ku), Eastern Ogbia (or Kolo) (Eo), Western Ogbia (Wo), and Bukuma (unspecified). Abuloma is not cited because for every item on our list it shares a cognate with some other variety of Central Delta, Data on Central Delta excluding Bukuma come from Wolff (1969); data on Bukuma come from the author's own collection. The Ogoni group is represented principally by Kana, which is not referred to by any label in the list; Gokana and Eleme are referred to as (Go) and (El) respectively. Data on Kana come from Williamson Nwinee and other informants; data on Gokana and Eleme come from Brosnahan (1967) and Williamson (1973) respectively. Lower Cross is represented by Andoni (Obolo), and its data come from Reading and Writing Obolo, 1979, and Dr. N.C. Ejituwu (personal communication). Nkoroo forms have also been cited so that one can see the degree of resemblance in the cognate items of Defaka and Nkoroo. Data on Defaka and Nkoroo come from the author's own collection. In the Proto-Ijo and Nkoroo column, the Nkoroo item is always placed last. Note

Defaka that the citing for a single form in a language group does not necessarily imply that the form in question is cognate throughout the group. The following orthographic explanations may also be noted: 6 = b, gb = voiced labio-velar stop (Igbo only), d = d, dj = j, j = y, Ji = ny (Igbo), y = gh (Igbo, Central Delta), ß = bh (Central Delta); a final •n' in a Defaka or Ijo word indicates nasalization of the preceding vowel(s) or approximant(s); bh in Igbo represents an aspirated

b

The tonal notation used for all Defaka, Ijo, Delta Edo, and Central Delta data include high tone ("), downstepped high tone ( j, and low tone (unmarked); in Kana, high and low tones are marked, while mid tone is unmarked; the data in Lower Niger and Lower Cross have been fully tonemarked, since their orthographic tone marking conventions are different from the others. TABLE I Apparent cognates in Defaka and neighbouring languages Defaka 1. two

2. three

tatS

3. four

4. five

PIjo,NK

LNiger

DEdo

CDelta

maami maamt

ábúo

Iva(De)

iwal (Wo) báé

iba

tárú t&árú

ató

isái(De)

isar(Wo) toa

ítá

iní(De)

iñs(Wo)

int

ínóryt néin tuuno

sónór¡ó

Isé(Um) isüwáfi(De) owu(Wo)

Ogoni

LCross

S'SS

gS

sóónó 5. ten

uwSi

Syí) (m) atei Ir! Syí

-gweNi

Sdiobh (Öd)

líb

äköp

6. twenty

sfi

sí sí

Shú

-gheGi

adusúbh (Ku)

tub

äkSp-Ibäétíp

7. father

izulo, da&

daú

finä

odám ósu(De) 'daxx

té

flt§

deí jie

di

dáifi

ólóm

8. husband

olomámu

92

Charles E.W. Jenewari Defaka

PIjo,NK

LNiger

DEdo

CDelta

nörfi

-bo,kfmi kíñí

mádhú

-kwSchi

suunom (Eo)

10. wife

toa

ta toa

nwlnyê

áni(En)

Srlam

11. friend

6k6 bài

ikis" ikiabo

ányl

övre(De)

ôyà-glrâm kód '(Go)

12. guest

ikêrd (tí) '

vgom iken-tbo

ôbhyâ

oklkiä (De)'

isolomB (Wo)

13. animal, ónúmá, meat ami

noma námá

ânwrùânû enáñ (¿e)

ênam

14. fish

tina

iDir¡i ' nji

ázü

êna

bSrl

IrîQ.

15. dog

ebeve

obiri' obiri

ñklta

-bhwa

ayônghS

gbó

ibo

16. goat

óbòr-t

óbórí ôBôri

éghfl

-bhui

ewel/ (Eo)

bòi (Go)

Sbôt

17. fowl, okuna chicken

boko(NK)

ôkiîbô

-khokho

okina

kon

flmân(hen) Sríé(cock)

18. sheep

ánáná

onana ânânâ

áthfiru úsumál "(De )

onugholu

nàànà-f ümän ârôn pêê

19. tortoise

tkâkt

lkakl/tk&- fibékwú, •ghuNi g-t ikâU rtnâbë

aluaulu

kurû

Ikwút

20. elephant

bila

bila' bila

Ri-

Sniìr)

21. crocodile

sako

seki/segi' Sghllyf usSni saku (De)'

merie

îsâkwfit

22. bat

peku

poki '/yogi 'poku

blá

Sfiao

9. person

ényl

•ni

ogbóifi, onwanwS (De)

Sgholu

Ogoni LCross

fíwa

nçncâà ügwú(Go) Ichên ânâm

Defaka DEdo

CDelta

akpala(g)í? Swò, akalám ákirí

otiSo (Ep)

obhom(Eo kárá (Go)

24. housefly

omomó omomó

-khiNa

25. snail

ost osi

éjl, iljùnà

ukpá "(De)

eghoi '(Eo) '

bùé

égúrü, SkpSS

paBaQá ' ápá

ñkü

uvflu(De)

ibhom (Eo)

labS

àk5k5

aldiwo etc.

pàà

àkókò

23. toad, frog

26. wing

Defaka

PIjo,NK

akal&fi

apa

27. feather piSm

LNiger

ípíkó(i)ú)? Sbhübhä ukpro(De) uvun (Ep) p-íko

Ogoni LCross

äkwüök

ànàngln

obü, toguTJö ; okpo

àztì

-k3

òkpìie fldflO (El)'

asi(n) ' asi

öbhSrä

IzSla(De)

m£ i

éjé,ézS

beri ' beri

ñthx

-choGi

to

útóri

31. hair(of j-tmé head)

dtrrré dime

nthùthù

aslnému

zia

Ijéék

32. hand, arm

káa

bära bara

&k&

ubo(De)

ará

bá

úbék

33. head

tóbo

tfbt t'éb'é

Isl

-chiSmhi

Smä

ékábee Ibdt

34. mouth

bU, b'é'é

bibi bebé

6nú

Sgá

ótú

kogoQo (u)

61ú,ónú inyáñ(De)

orné

ÖbSk

bió

inwöm

28. back

okpo

29. blood

30. ear

35. neck

basi

kongo 36. nose

m m nini

Imi

chuaNi

i jonu

94

Charles E.W. Jenewari Defaka

37. tooth

PIjo,NK

LNiger

DEdo

CDelta

Ogoni LCross

Ská

ézé

-kuN

6nái

däa

éjéré

okpo

ékpó

úkwúík

oghúm '(Od )

píi (Go)

àfl, úfóp

38. bone

ifibua

ígbSQiS ifiabá

Skpúkpú úbüó

39. penis

oyom

tógogó

üthü, Smü

tongo

-chua

tebi ' tibi

Shü,ót0, úbe fkpü ' (Ep)

e tu

bi (Go Skpò, édlm

41. finger- ìfrnémé nail

•tmèmèt rfmémé

ifivó Ska úvúnbc (Ep)* "

ewabh (Eo)

pio

àbórj

42. beard

eferu

{•pòri áferü

äfiwä-SnC ukpori ägba(En)

Ileem

fiH

ìfètIràk-ótù

43. breast

ónúó

íD6r¡ú ndó

árá

ivinyáíl (De)

amamám (Ku)

má

ébé

44. tree, stick

t-Cin

té r)í tiin

óslsl

-thaNi

Sréfn

tè

úti

45. grass

Abóbórí

ítuka mbúbónt

áhlhyS

ikpu(En)

íwóogho

ébis

ifibúbét

egeryu owoin

c5r)würú

ìmafu(De)

amugi

soómíS odukuukar)

40. vagina

títtíéé

46. smoke

(Ku)

47. rope

Itkt

MU/dtgt Uki

Óbhü, áríirf, ùdò

•dhuNi, •dhiNi

odidi (Wo)

d.U (Go)

6VCk

48. salt

66

ffln füún

í5nú

ubíicSñ (De )

abom

16

úchí, úkwi

49. oil

pira

pulo(u)'

mánti

-bhi-

n6o

árSi)

50. story

lógo

dúgúj egberi logo

ákfikó

udóú(De)

lok (Go)

ílrók, ñfüfük

asido

Defaka

51. market

95 Defaka

PIjo,NK

LNiger

DEdo

CDelta

Ogoni LCross

kia

fé-yc5

ShyS

-ki

3bua(Eo)

ki

fèê ' 52. town

éwé

(Go)

ama ámá

ôbôdS

ékefn(De)

ema

âmâ

"(Wo)

53. wine

ilo

1WWPU oru

mil, mmányá

udf(Do)

3min(Eo) mlí

mirri

54. thing

yâa

iyé yé

ihyê

inúíS(Do)

abâr(Ku) n 5 )

Ikâlê-Yoruba:

(7:5) O Ô

(5:3) Ijebu-Yoruba:

Nupe:

. Bini:

Kétu-Yoruba:

i t i

e

e a o .. ~ ~ e a o

o

u,_ c , Esan: ~(7:5) u

x l

e

E a o . . e a o

- (7:4)

o

u,., ~(7:5) u

Note that the Nupe situation supports SPR^ stated above, at least on formal grounds. REFERENCES Adeliyi, Stephen. 1872. The Ketu dialect of Yoruba. Long essay for B.A., Yoruba. Department of Linguistics and Nigerian Languages. University of Ibadan, Ibadan. Adeniran, Kunle. 1968. Towards a phonology of the Ijebu dialect of Yoruba. Long essay for Postgraduate Diploma in Linguistics. Department of Linguistics and Nigerian Languages. University of Ibadan, Ibadan. Akpamu, E.G. 1971. Towards a phonology of Esan. Long essay. Postgraduate Diploma in Linguistics. Department of Linguistics and Nigerian Languages. University of Ibadan, Ibadan. Capo, Hounkpatin. 1977. Etude phonologique comparée du Waligbe et du Gengbe. Unpublished M.A. Thesis, Université Nationale du Benin, Cotonou. Capo, Hounkpatin. 1980. "un regroupement des parlers Gbe". Africana

Nasality in GBE Morburgensia 13 (1) : 3-24. Capo, Hounkpatin. 1981a. "Nasality in Gbe: a synchronic interpretation". Studies in African Linguistics 12(1): 1-43. Capo, Hounkpatin. 1981b. A Diachronic Phonology of the "Gbe" dialect cluster. Unpublished doctoral dissertation, University of Ghana, Legon. Dunstan, Elisabeth (ed.). 1969. Twelve Nigerian Languages. Longmans: London. Heine, Bern. 1968. Die Verbreitung und Gliederung der Togorestsprachen. Dietrich Reimer Verlag: Berlin. Hyman, Larry. 1972. "Nasals and nasalization in Kwa." Studies in African Linguistics, 4: 167-205. Isola, Akinwumi. 1969. Brief notes on the phonology of Ikale, a dialect of Yoruba. Long essay for Postgraduate Diploma in Linguistics. Department of Linguistics and Nigerian Languages. University of Ibadan. Ibadan. Ogieriaikhi, E. 1968. Some aspects of Edo (Bini) phonology. Long essay for Postgraduate Diploma in Linguistics. Department of Linguistics and Nigerian Languages. University of Ibadan. Ibadan. Cpubor, Alfred. 1969. "Itsekiri". Chapter 10 in Dunstan (ed.). Smith, N.V. 1969. "Nupe". Chapter 11 in Dunstan (ed.). Stahlke, Herbert. 1971. Topics in Ewe phonology. Ph.D. dissertation. University of California, Los Angeles. (Xeroxed copy). Williamson, Kay. 1973. "More on nasals and nasalization in Kwa." Studies in African Linguistics, 5: 115-138.

142

Hounkpatin C. Capo APPENDIX I . Comparative

Table 4: Dialect

correspondences

data.

and Common-Gbe.

Gloss

Fon

Ajä

Gen

Kpàndo

Wad

Aulan

t o walk

zò

zò

z o»

zò

zò

zò

to

roast

mè

mè

m l/mè

mi

to

love

lò

lo

Io

Io

lo

to uproot

RÒ

RO

RO

RÒ

be good

J1Ò

J1Ó

JlÓ

pot

zé

zé

z *e // z e

sibling

nòvi

nàvi

spoil

gblé

gblé

to drink

nù

nù

t o be

J>i

JH-

II. a) Table

5:

Gbe /bè/

lò

/lò/

RÒ

RÒ

Rò

/Rò/

JTÓ

Ji5

Jl5

Jlò

/ j 6/

zé

zi

za

zé

/zi/

nòvi

novi

n5vi

novi

nòvi

/4òvl/

gblé

gblé

gblä

gbl5

gblé

/gblé/

nù

/ hö gänä

3 k3 nOma ngaa

'it's going to go to Ghana' 'its going to drown carp'

178

William R. Leben

o is the third person pronoun, ko/ho and nUma are verbs, and gana and ng&a are nouns. Note that the non-final forms of ko/ho and nOma differ from the final forms in that the former are always L, even when the latter are H. This is a result of (2), a very general rule of Anyi. Given that, we can take the non-final forms as derived from the final by (2). The basic patterns on the right in (1), which refer to Pro+Verb sequences, need make no direct mention of the non-final forms. (2) NON-FINAL LOWERING3 H

-

L /

Cond.:

X

] ?

Verj-,

^

e.g. 3 tj gana

-

5 k3 gana

0

Let us temporarily regard the basic patterns on the right in (1) as our analysis, very similar in format to the one McCawley 1970 proposed for Tiv. Then for each tense or aspect there will be a tonal formula, and for the language as a whole we could propose a set of conventions and rules for realizing these formulas on particular verbs. Consider now the Past 1, outlined in (3) and (4). The third singular pronoun o becomes w before a. (3) Past 1

Examples

Basic Patterns

wa hip

wa nUma'

'it has gone'

1

1-syll

2-syll

L+à+H

L+à+LH

it has drowned1

wa h£> gana

wa nUma ngaa

• it has gone to Ghana'

'it has drowned carp'

(4) Past 1 Examples •

amo •you (pi.)' amwa ho?

amwa nUma' 1

' you have gone

' you have drowned1

amwa h£> gana

amwa nUma ngaa

'you have gone to Ghana'

'you have drowned carp'

One complication in this aspect is that the second plural forms behaves differently from all the others, as seen in (4). The second plural pronoun Omo, is invariant in tone throughout Anyi. To account for the peculiar i tonal behavior of h after dmo, we formulate rule (5), which will be

Tone in Anyi

179

familiar to any student of the closely related Akan languages (cf. Schachter and Fromkin 1968). Accounts of the other rules that simplify the fall 4

derived in (5) to H followed by downstep are frequent in the literature. (5)

TONE SPREADING ä

ä

/

H +

e.g. ämo+ä+hö?

amoä h£¡° ámwá hi')

Now take the Habitual: (6)

Habitual Examples 2> nUma'

a. 5 ko? 1

Basio Patterns

it goes'

'it drowns'

5 kö gana

i> nUma ngaa

'it goes to Ghana'

'it drowns carp'

b. ami ko?

1-syll

2-eyll

L+H

L+LH

amo nUma'

'you go'

'you drown'

ämo kä gána

amo nUma ngSa

'you go to Ghana'

'you drown carp'

If we ignored the amo form, we could express the Habitual tone pattern as L+(L)H. But after ¿mo, a monosyllabic verb is always low in the Habitual. To account for the alternation between H and L, we would of course try to posit one alternant as basic and derive the other from it. If we took H as basic and attempted to lower this after ami), the rule would have to be morphologically restricted (cp. Future 1 o feS', which does not change to o ko'). If instead we take L as basic, we can formulate a simple raising rule (H indicates phrase boundary): (7)

FINAL LOW RAISING L

-*•

H / L +

1 „ i ' Verb ™

e.g. 3 +kd

3 kS'

By requiring preceding L in the environment, we prevent (7) from incorrectly applying after amo, which ends in H. Looking back on all the data given so far, we see that this rule will not have to be restricted solely to the Habitual. In fact, it is possible that in Future 2 and Past 1, the surface L+kcrt-H and L+a+H actually result from applying rule (7) to more |k3j basic L +| | + L. One advantage of this is that it helps to explain what

180

William R. Leben

happens in Past 2.

(8)

Past 2

Examples

Basia

3 h3 li

3 nUma li?

'it went'

• it drowned'

3 ho li gana

3 nUma 1

'it went to Ghana 1

Patterns

1-syll

2-syll

L+L+11

L+LH+Z-I

II ngaa

it drowned carp'

The role of rule (7) here is not too hard to see. The disyllabic verb gives some sign that the tone of aspect marker II is linked to that of the preceding disyllabic verb nUma, for when nUmd. ends H, II is H, and when nUma is L, II is L. A way of capturing this is to make II toneless and to copy the tone of nUmd onto II only after Non-Final Lowering (2) has had the chance to apply:

(9)

a.

[3

+

nümä + II]

b. [3

Verb

+

nümä + ÌÌ] . . > Verb

ngäa

(2)

TONE COPY

li

II

If this is correct, then we would of course copy the monosyllabic verb tone L onto II and then II would finally be raised as predicted by (7):

(10) a.

[3

+

h3

+

ll]

b. [3

verb

+

h3 +

li]

yerb

gana

(2)

II

TONE COPY

II

(7)

li

Thus (7) FINAL LOW RAISING helps to explain why the dependence between the final tone of the verb stem and ZJ in (9) is not evident in (10). One other attested affirmative form, the imperative, will not be discussed here because its treatment fits in more naturally with negatives, which will be the subject of a separate paper. The Basic Patterns for the different aspects are drawn together in Table II.

Tone in Anyi

181

Table II. Basic Patterns. 1-syll.

2-syll.

Habitual

L+L

L+LH

Future 1

H+H

H+LH

Future 2

L+feS+L

L+feS+LH

Past 1

L+Ò+L

L+à+LH

Past 2

L+L+ÎJ'

L+LH+W

A generalization emerges here that was not obvious from Table I. This is that the tone of the monosyllabic verb in each aspect agrees with the tone of the pronoun. Where there is a segmental aspect marker, its tone too is predictable. For post-stem II 9 this point was made just above. For ki> and a, note that they agree in tone with the pronoun and monosyllabic verb. In fact, if we consider what must be specified exclusively for eách aspect, it amounts to very little indeed. The tone of disyllabic verbs is invariantly LH (apart from their behavior in non-final position, as described by rule (2)) , and so this can be expressed by a redundancy rule, independently of all the aspects. Furthermore, the segmental form of the pronouns is invariant (apart from vowel harmony and alternations of vowels before vowels, both of which are describable by phonological rule), so these need not be mentioned separately for each aspect. Finally, since the tone of the pronoun in Table II always agrees with the tone of the monosyllabic verb, there is really only one tone that can vary from aspect to aspect. In the Habitual, Future 2, Past 1, and Past 2 it is L, and in Fut 1 it is H. The only other thing that varies from aspect to aspect is the segmental aspect marker. We can in fact collapse these two bits of information into one by considering the tones themselves to be aspectual morphemes. In some cases the tonal morphemes are accompanied by segmental information; e.g. the Future 2 morpheme is kó and the Past 1 morpheme is a. For Past 2, preverbalL and postverbal II might be treated as a discontinuous morpheme. In the remaining cases the aspectual morpheme is purely tonal. For precedents on the specification of purely tonal morphemes there is, of course, a growing literature; see, for example, Welmers 1959 and the recent autosegmental proposals, beginning with Goldsmith 1976. Thus the description of the Anyi verb system becomes quite simple. Underlyingly we have an invariant set of pronouns followed by an invariant set of verbs, with tone pattern LH specified for disyllabic verbs and no tone specified for monosyllabic verbs. All that varies from aspect to aspect is, appropriately enough, the aspectual affix, whose phonological

182

William R. Leben

representation (both segmental and suprasegmental) and whose position with respect to the verb must be stipulated separately for each aspect. Since this suggestion involves positing considerably fewer underlying tones than syllables, we will need some mechanism for assigning tones to phonologically toneless syllables. For this, we may extend the rule of TONE COPY introduced above informally. The needed rule is similar in effect to principles of tone assignment proposed by Goldsmith 1976, Leben 1978, and Clements and Ford 1979, but note that the operation needed for this account of Anyi must be crucially ordered after (2) NON-FINAL LOWERING: (11) TONE COPY Assign the first unassociated tone to the first toneless syllable. Any toneless syllables remaining receive a copy of the immediately preceding tone. To conclude, some sample derivations are given in Table III. TABLE III. Sample derivations Habitual o +

ko

[L] Asp

amo + ko I I HLH + [L] Asp

Future 1

Future 2

D+

o +

nUmä // ngaa I I I [H] LH H Asp

ko, Asp [L] Asp

ko

ho

LENITION numa II L L

NON-FINAL LOWERING (2) TONE COPY (11)

ASP

amo + ko 1 I I HLH + [L ] Asp

o + nUma II \ [Hi L L Asp

ngaa I H

[L] Asp

TONE SPREADING FINAL LOW

? L

+

+ ho o + ko N ^ - ^ Asp i L H

ko / H

= ò kò

= amò kò

-*• ò kò

•*• amò kò9

= ó nUmä ngáa

= ò -y ò

kò hó kò hb">

Tone in Anyi

183

TABLE III (Cont.)

Past 1

Past 2

a"» + [a], A

II

HlH

I

[L]

SP

+ ko

Asp

o + ko + II .^Asp

LENITION

ho

ho

NON-FINAL LOWERING (2) TONE COPY (11)

amo + [a], + ho Asp HLH

TONE SPREADING (5)

amo + [ a]

I

HLH

FINAL LOW

[L]

ASP

amo + [a]

I M HLH

RAISING

A

+ ho

s

L

+ ho

p

H

O + ho + II L

V HI

= 3 hb II -*• amw a ho9

•+ o ho II9

NOTES 1. Work currently in progress in collaboration with Faustin K. Yao and Rebecca La Brum shows that this account can for the most part be extended quite naturally to negatives and to the Anyi serial construction. 2. Also, the symbols [I] and [U] refer to the non-advanced tongue root counterparts of [i] and [u]. The symbol n designates [rj] before g, as in ngaa = [i]gaa ]. Though it is not illustrated in Table I, we will see later that verb-initial k lenites to h in Future 2, Past 1, and Past 2. 3. (1) gives the impression that lowering of tone and elimination of glottal stop go hand in hand. But this is not really so. Other verb forms show that glottal stop is inserted finally, independently of tone lowering. The glottal stop alternation ~ 0) extends to cases where there is no tone alternation (e.g. Neg. Habitual 3 ngi>7 'it doesn't go', 3 ngi> gancX 'it doesn't go to Ghana') and to cases where the non-final verb tone is H, for example before the nomalizing suffix -Iz: i. + hi> + It? 'going'. Thus, I will assume that there is a late rule adding glottal stop postvocalically before pause. 4. A more direct account of this, following Clements and Ford 1979, is: a

I L

—+

/H H IH _ ' L

3

L represents a "floating" L, one which is not pronounced on any syllable

184

William R. Leben

but which is realized internationally, by downstepping a following H. In the same way, ¿5mo will be represented below as amo. 1-1 HLH REFERENCES Clements, G.N., and K.C. Ford. 1979. "Kikuyu Tone Shift and its synchronic consequences". Linguistic Inquiry 10: 179-210. Goldsmith, J.A. 1976. Autosegmental phonology. M.I.T. dissertation, published by Garland Press, N.Y., 1979. Leben, W.R. 1978. "The representation of tone". In V. Fromkin, ed.. Tone: A Linguistic survey. New York: Academic Press. McCawly, J.D. 1970. "A note on tone in Tiv conjugation". Studies in African Linguistics, 1: 123-129. Schachter, P., and V. Fromkin. 1968. A phonology of Akan: Akuapem, Asante, and Fante. Working Papers in Phonetics, 9. U.C.L.A., Los Angeles. Welmers, W.E. 1959. "Tonemics, morphotonemics, and tonal morphemes". General Linguistics 4: 1-9.

Chapter 10

Tone Rules and Derivational History in Edo Phonology Airen Amayo

1. INTRODUCTION This paper describes some Edo* tone rules, drawing special attention to the need for two of them - low tone raising and derived high tone lowering to have access to the derivational history of their input strings. The presence of two other rules, high tone postponement and tonal absorption, is also reported. 2. THE BASIC TONES AND THEIR VARIANTS Edo is a typical two-tone-plus-downstep language, with downdrift very much in evidence. The basic tones, high and low, contrast as in the following examples.^ 1. okpe

"palm-wine tapper"

:

okpe

"flute"

2.

"horn"

:

igho

"money"

igho

A high tone (H) is lowered after a low tone (L) so that if two high tones are separated by one or more low tones the second high is lower than the first E.g. 3.

asele

[ ^

4.

ozo ¡kpolo

"cricket" -J

"Ozo is sweeping"

As evident in the tonetic representation of 4, a succession of low tones interspersed with high tones also drifts downwards. Moreover, whenever a L occurs immediately following a H, the L is realized as a fall from H to L, i.e. H-L. Examples 5 and 6 provide good illustration . 5.

?kpo

-»•

fkp6

(i.e.

6.

o ra kpolo + 6 ra kpolo

"bag" "he's about to sweep"

High tone spreading (HTS) is not obstructed by the presence of word boundary between a H and a following L, as illustrated by example 6. The other variants of the basic tones - the downstepped H and the downstepped L - result from the interaction of the high tone spreading rule and the tonal simplification rule.

186

Airen Amayo

3. TONAL SIMPLIFICATION (TS) When a H-L which is borne by the second vowel in a sequence is followed by another syllable, the H-L gets simplified to a segmental H followed by a floating L, i.e. H".^ Thus while there is no simplification in 7, the H-L in 8 and 9 are simplified. 7.

ek6o

-»-

8.

ebe + eva

ekoS ebeevS

"stomach" ebee~va

ebe ~vS, i.e. ebe!vS)

"book" "two" 9.

ek6o + "kpolo •*• ek66 ~kp616 ->• ekoo" 'kpolo "stomach' "be big" (--

ugbo rue

Here we need to posit a floating low tone between the noun and the possessive pronoun to be able to account for the lowering of the final H of the noun. Thus underlying forms 14 - 16 will be represented as in 14(a) 16(a). 14(a)

owa

+

15(a)

ugbo +

16(a)

ifi

+

vben rue '+

ere

188

Airen

Amayo

We postulate that a floating tone shifts onto the immediately preceding vowel and literally covers up its tone.^ The effect of the tone shifting process is evident in the phonetic realization only when the tone which is replaced by the floating tone is different from it, as when a floating high tone replaces a L or when a floating low tone replaces a H. In all the instances of tone shifting in Edo, the floating tone moves leftwards. All the tone rules so far described interact in the derivation of 11, which is a more complete form of 10. (i) 11(b)

(ii)

(iii)

(iv)

owe + '+ efe •+ oweefe •+• owSefe •» owee "fe

(v) owe!fe

At stage (ii) TSh. applies to give the final vowel of ow| a high tone, HTS takes place at (iii) and TS takes place at stage

(iv).

5. Low Tone Raising (LTR) We have seen that in the associative construction the tone of the final vowel of the head noun becomes high through the shifting of the construction floating high tone on to it. However, if the head noun has an all-low tone pattern, all but the initial L are also raised to H. E.g. 17.

akobe

18.

19.

+

"+ uyi

-»• akobuyl

"iron-trap"

"Uyi's iron trap"

¿gbaleto + '+ Syi

ugbalgtSyi

"head-tie"

"Ayi's head-tie"

oghedd + '+ ugb6

oghedugbo

"plantain"

"Ugbo's plantain"

We therefore require a rule which can raise all but the initial L of a head noun with an all-low tone pattern. This rule will have to be constrained to raise only a L preceding a H derived from an underlying L. It is this constraint on the L-raising rule that prevents the non-initial Ls in words like SkSsSn (maize pudding) edede (old woman), ogedegbe

(personal

name) and ekuy^ (spoon) from being raised. E.g. 20.

ekuye + '+ Sdya

ekuyadya

"spoon"

"Adia's spoon"

The LTR rule can be roughly formalized as a leftward directional rule. 21.

L - > - H /

+

L X - > - H

7

189

Tone in Edo Phonology condition: H Derived from L a (where + = word boundary, and X - zero or any number of Ls) As an iterative rule it keeps reapplying until all non-initial Ls are raised. 6. DERIVED HIGH TONE LOWERING (DHTL) When a polysyllabic head noun with either an all-low or a final HL tone pattern collocates with a L-initial nominal qualifier, the tone resulting from the contraction of the juncture vowels is L (rather than H as in all other cases). E.g. 22.

ababS

+

"+

ozo

•*•

"witchcraft" 23.

ugbaleto

+

"Ozo's witchcraft" "+ ame -+• ugbaletame

"head-tie" 24.

akobe

+

"Ame's head-tie" "+ ofen

"iron trap" 25.

abSbozo

-»•

"rat"

akobof|n "iron-trap for catching rats"

Ighu!hunvbun + '+ ugo -»• ighu! hunvbungo "tax"

"Ugo's tax"

In collocations involving other types of nominals in the associative construction, the tone resulting from the contraction of the juncture vowels is always H. Thus we can contrast 22-25 with 18, and 26 and 27. 18. ugbaleto

+

'+

26. ukpokpd + "+ oz6 "stick" 27. Imi!na

ayl

-»• ugbaletayl ukp6kp6!z6 "Ozo's stick"

+ '+ ame

"dream"

-»• imilnalme "Ame's dream"

It should be noted that words like Imilna and akhalrho (monkey) which end with H!L are underlyingly HLL and not HL; hence the tone resulting from the contraction of such words with a following L-initial noun will be H and not L, as illustrated by 27 above. In deriving the juncture L of 22-25 we assume that the tone shifting rule first applies to give the final vowel of the head noun a high tone. Thus 22 and 23, for example, will have the following intermediate stages in their derivation.

190

Airen Amayo 22(a)

ababe + '+ ozo

-*•

23(a)

ugbaleto + '+ ame -»•

ababe + ozo ugbaleto + ame

In 23(a) LTR further applies to give ugbaleto + ame. The derived final H of the head noun then changes to L if (i) it is immediately preceded by H and (ii) if the initial tone of the qualifier is L. Thus 22(a) and 23(a) will be further derived as in 22(b) and 23(b). 22 (b)

ababe + ozo

-»•

abSbe + oz6

23(b)

ugbaleto + ame

-»•

ugbaleto + ame

The vowel assimilation and contraction rules then apply to derive the phonetic realization; ababozo and ugbaletam?. The rather tight constraints on the lowering of the final tone of the head noun are essential for the derivation of the correct phonetic output. The H which changes to L has to be one which is derived from an underlying L. Otherwise the final H in words such as ukeke (stick) and ugbo (farm) would also be lowered, which is not the case, as illustrated by 28. 28.

ukeke + "+ oz6

•+•

uk£ko!zo

"stick"

"Ozo's stick"

The constraint that the derived H has to be immediately preceded by another H prevents the derived final Hs in cases such as 26 and 27 from changing into L. Moreover, this rule has to be ordered to apply after LTR, otherwise cases like 23 and 24 cannot be accounted for. The derived high tone lowering rule (DHTL) can be roughly formulated as follows (+ = word boundary): 29.

Ha

>-

L

Condition: H

/ a

H

+

L

derived from L.

The rule states that a word-final H derived from L will become L when collocating with a L-initial word. As in LTR, we do not need to specify that the rule applies in the associative construction because it is only in this construction that the conditions for the application of the two rules can be satisfied. 7. EXCEPTIONS TO HTS AND TS High tone spreading (HTS) and tonal simplification (TS) are general tone rules which cut across different grammatical constructions. However, in collocations of the verb + object type there is one instance in which

Tone in Edo Phonology

191

rule inputs which meet the structural description of HTS (and therefore also TS) fail to undergo it, and another instance in which even though HTS and TS apply normally, the expected phonetic output is not realized because another rule intervenes to prevent this. We have to account for these cases by two other rules - high tone postponement and tonal absorption. 7.1. High Tone Postponement (HTP) To set the stage for the introduction of HTP, we need to point out g that (i) the verbal constructions

of Edo are classifiable into two broad

types - the simple completive construction (SCC) and the non-simple completive constructions (NSCCs) and (ii) noun objects are classifiable into three types - H initial, all-L and LH. By a LH noun we mean a noun with an initial L and one or more Hs anywhere else in its structure; thus words like agbalka (crocodile) and edede (old woman) as well as ozo will 9 be regarded as LH nouns. When a verb collocates with a LH object in the NSCCs, the vowel resulting from the contraction across word boundary bears a L rather than a H. E.g. 2

(0)

ka + ukeke

-»• (o)i

kukeke

(he) "count" "stick" "(he) is counting sticks" 30.

(0)

koko + agbaika

"feed" "crocodile" 31.

(O)

kpee + ogo

"beat on" "bottle"

(o) kokagbalka "(he) feeds crocodiles" -»• (o) kpogo "(he) is beating on a bottle"

Normally we would have expected HTS and TS to produce a juncture H; this is what happens, for instance, in the (SCC). 32.

(0)

gbe + ewe

"kill" "goat"

(o) gbe!we "(he) killed a goat"

Clearly a different phonological process takes place in NSCCs, apparently to keep the NSCCs separate from the SCC in their phonetic realizations. This therefore is a case of the avoidance of surface homonymy as enunciated by Wilkinson (1975). We shall derive the juncture tone in NSCCs by first lowering the H of the verb stem; the vowel assimilation and contraction rules then apply to derive the phonetic realization. By the lowering rule, 29, for example, will have the following intermediate stage in its derivation.

192

Airen Amayo

29(a)

ka + ukekS

-

ka + ukeke.

The lowering of the H in this case is different from the DHTL encountered in the associative construction (cf section 6). In this case we regard the lowering as resulting from the postponement of a (verb stem) H till the next H is reached.*® We shall formalize the HTP rule as in 33. 33

"

H

—

L

/

]vS(

Ln H

+

NSCC The rule states that the H tone of a verb stem (VS) in a NSCC is lowered when it collocates with a word beginning with any number of Ls followed by H. We do not need to specify that the object could have any number of Hs because it is only the first H that is needed to trigger off HTP. 7.2. Tonal Absorption (TA) Consider the following examples of the collocation of verb stems with objects which have an all-low tone pattern. 34.

35.

sS

+

ame

same

"draw"

"water"

"draw water"

d". With this addition, the representation of kpo | kiri | kpo takes the form (4)

$ 4a

a

A kpo

o

A ki

A ri

a

A kpo

where the pitch-change markers + and + represent pitch relationships between tonal feet. Note that the phonetic content of our basic tonal units, the pitchchange markers, is somewhat different from that of other phonological units, since they represent articulatory gestures (changes in the length, tension, and thickness of the vocal folds) rather than articulatory configurations. It would be possible, I believe, to avoid the use of such units by using relational trees to indicate the tonal relationship between feet, as in the representation shown in (5):

(5) A

H

$/ J kpo

8

J? L

L

A

H

V$ \$

v ^ Nri ki

1

v kpo

(I have omitted the level of the syllable in this representation for the sake of simplicity). The relational tree

R

is used here to indicate

/\ H

L

that the first tonal foot is higher than the second; similarly, the tree R L

indicates that the second foot is lower than the third. In the H

discussion which follows,. I will continue to use pitch-change markers to represent the tonal relationship between adjacent feet; however, the analysis which I will propose is translatable into a relational framework of the sort illustrated in (5). Note that this framework is rather different from the metrical approach proposed by Huang (1980) and Clements (this volume), in which the contour of (5) would be represented as shown in (6) :

Tone Spreading

215

(6)

H H I

L L

H

A

|

kpo ki ri kpo Here only the upper-level H and L are defined relationally (i.e., as "higher" or "lower" with respect to a sister node): the lower-level H's and L's are still defined in the traditional way. This can be seen most clearly for the "H" which directly dominates the final syllable kpo.

This H can-

not be defined as "high in relation to its sister node" because it has no sister. The argument which I will make in this paper is valid only for purely relational systems like those of (4) and (5). Now observe that in a purely relational system, it is not in the end sufficient to represent only pitch relationships among the tonal feet themselves, for we then have no way to distinguish between high-toned and low-toned utterances such as the Igbo words shown in (7), which lack any internal pitch contrast: (7)

a.

$ akwukwo

=

b.

a kwu kwo

$ ngaji 1

'book'

=

n ga ji

1

spoon

Nor can we distinguish between MH and LM contours, as in the Ewe sentences shown in (8): (8)

$ mebuwo ets5 = 'I respected you' 'yest'

$

me^bu^wo + e'^tso 'I respected you yesterday'.

b. ado

wu

afi

'squirrel'

'killed'

'mouse'

=

To distinguish among these contours, we must add to the system some device for indicating the pitch relationship of the overall contour to neutral or rest position. I suggest that we use for this purpose the device of an "anchoring" pcm at the edge of the contour. For example, placing the anchoring pcm at the beginning of the contour, we can distinguish between the two contours of (7), as shown in (9), by placing a + at the beginning of (9a) to indicate that it begins at a higher-than-neutral

216

Mary M. Clark

position, and a + at the beginning of (9b), to indicate that it begins at a lower-than-neutral position: (9)

a.

+

$

/N

---

a kwui kwo i

b. *

$

/N

(= + akwukwo) i i

n ga ji »

(

i

b.

/1\ a kwu kwo

(= akwukwo+)

me bu wo

e

A

$

tso

4.

4

+

/K

n ga ji

a.

(= ngajit)

$

+$

a do wu

(= mewttbo + etsoi)

a fi

(=

There is also a third logically possible system, in which pitch contours are anchored at both ends, as shown in (12): (12)

a.

$

/K

+

a kwu• kwo1 c.

$

b. (=+ akwukwoi) 1 1

t $ 4 -

/K A me bu wo

+

e tso

(= mebuwo t etso + )

$

+

/K

n ga ji1 d. 4-

(= +ngajit) -

$

a

do

(= +ado

I wu

a fi wu

a+fi)

Tone

217

Spreading

In Clark (1978), I argued that the three anchoring-systems illustrated in (9)-(12) all exist (though end-anchoring systems like that of (11) are somewhat rare), and I suggested criteria, including the behavior of "toneless" elements, for deciding which system is used by a given language. I would like to focus here on one particular difference between front-anchoring systems like (9)/(10) and end-anchoring systems like (11). To observe this difference, consider, as an example, a HLH contour, which could be represented in either of the two ways shown in (13), depending on whether the language in question is front-anchoring or end-anchoring: (13)

a. t $ 4- i t $ (=+""+__+"")

b.

$ + $ + $+

(=~~+ _+""+)

Now suppose that the language in which this contour appears has a rule of the form shown in (14) , which deletes a + one syllable to the right of a + : (14)

+

a t

1 2

3 :

Delete 3.

The effect of this rule will differ for the two representations. The application of the rule to (13a), which is front anchored, causes a lowering of the end of the contour, as shown in (15): (15) + $ +

+

+

(= +

+

~,

or HLL)

but the same rule, applied to (13b), which is end-anchored, causes a raising of the beginning of the contour, as shown in (16): (16) $ +

^

t, or EXTRA-H, H, H)

In general, the insertion or deletion of a pcm affects everything to the right of the target site in a front-anchored representation, and everything to the left of the target site in an end-anchoring representation. I will now argue that this property of the system can help us account for an otherwise puzzling distribution of facts regarding rules of tone-spreading. It was observed by Hyman and Schuh (1974) and by Schuh (1978) that there is a tendency for tone to spread to the right, both historically and synchronically. Hyman and Schuh give examples from a number of African language, including Yoruba, Gwari, and Ngizim, and Schuh (1978) adds others, including Bade, Bolanci, Zulu, and Vai. Japanese also furnishes many examples of rightward spreading, as is pointed out by McCawley (1977) and Ostler (1978) and discussed further in Clark (forthcoming). According to Hyman and Schuh, the spreading of tone to the left is non-existent as a diachronic rule, and rare (hence "unnatural") as a synchronic rule.

218

Mary M. Clark I am aware of two sort of exceptions to this principle. First, many

languages exhibit rules which spread a high tone very slightly to the left in the environment of a following low tone. An example from Kikuyu is shown in (17) , where the noun Ktxrioki, which has a rising glide on its final syllable in the isolation form, becomes LH before an initially low-toned word like motya 'good' (Clements and ibrd 1981): (17)

karioki/ (proper name)

kario| ki |mo£"ya

'good Kariuki1

This phenomenon is fairly common; for example, I argued in Clark (1978) that Igbo, Mende, and Mandarin have rules very much like the Kikuyu rule which is illustrated here. We will return to this point below. We will be primarily concerned here with another sort of exception, namely, two rules of Ewe which spread tone freely to both right and left. The first such rule, proposed by Stahlke (1971), spreads a low tone in both directions at the expense of adjacent mids. Low tones do not exist at the underlying level in Ewe, but are created by a rule of Prefix Lowering, which assigns low tone to a noun prefix when the stem begins with an obstruent. This rule is illustrated in (18): (18)

a. ewu H M

afi (underlying form)

ewu afi

MM

Prefix Lowering

HM L M

'He killed a mouse'. b. ame fi azi la MM

M M M

(underlying form)

H

ame fi azi la

prefix Lowering

MM

M L M

H

'A person stole the egg'. Now the low-spreading rule applies, spreading the L tone to both right and left, as shown in (19): (19)

a. ewu afi

\

H M L M

Low Spreading

b. ame fi azi la MM

M L M

H

ewu afi H L L M ame fi azi la

Low Spreading

L L

L L L

H

A general condition on the rule of L-Spreading prevents the L tone from spreading onto a noun stem which begins with a voiceless obstruent, as does fi 'mouse' of (19a). At least one dialect of Ewe, Anlo, as a second bidirectional spreading rule, as shown by Clements (1978). This rule involves a "raised" tone (R) , which is created by several different rules, one of which is stated below in (20), following Clements:

Tone Spreading (20)

M

219 R

/

H

H

The application of this rule, which raises a mid tone to "R" in the environment between two H's, is illustrated in (21): (21)

a. nyonuvi a wo va H M H H

H

H

S nyonuvi a wo va M-Raising

H R H H

H

H

'The girls have come'. b. m3 atoto fl3-g3 L L H H

M

H

\ m3 atoto fl3-g3 M-Raising

L L H H

R

H

'I will buy pineapple'. I will argue below that the rule which creates the R tone should be stated in a different way; however, this is irrelevant to our present concern, the R-Spreading rule, which acts to spread the R tone to both right and left at the expense of surrounding H's, as shown in (22): (22)

a. nyonuvi a wo va H R H H

H

H

^ nyonuvi a wo va R-Spreading

b. m3 atoto fl3-g3 L L H H

R

H

R R R R

R

R

]*> m3 atotofl.3-g3 R-Spreading L L R R

R

R

Finally, a "cadence" rule applies to re-lower the tone of the final syllable to H, producing the surface contours shown in (23): (23)

a. nyonuvi a wo va R R R R R H

b.

m3 atoto fl3 -g3 L L R R R H

In (18) — (22) above, we have seen evidence for two tone-spreading rules of Ewe which apply in both a rightward and a leftward direction. If Hyman and Schuh are correct in their contention that spreading to the left is unnatural, then two questions arise: (1) Under what conditions is leftward spreading possible? (2) Why should we find two instances of leftward spreading in the same language? Is there something about Ewe which makes it

susceptible to leftward spreading? The relational approach to tone can provide an answer to these questi-

ons. Recall that in a language which anchors its tone contours at the end, rules which insert or delete pitch-change markers will have an automatic effect on the contour to the left of the target site. If Ewe is such a language, at least at the underlying level, then we may be able to account for the apparent leftward spreading without recourse to a leftward spreading rule.

220

Mary M. Clark

At the underlying level in Ewe, we find a two-way contrast between neutral (mid) and higher-than-neutral tone. Let us assume that only the high tones are marked. In particular, let us assume that within each lexical item, a string of syllables with higher-than-neutral tone is organized into a tonal foot with a + after it. Thus the strings of (18) have the underlying structure shown in (24): (24)

a. $

b.

4.

I

e

wu

afi

$

I ame

fi

azi

4.

la

The unfooted syllables are then gathered into tonal feet, as shown in (25): (25)

1

a. $ + e

$

wu

b.

afi

^iv i

ame

1

$ +

fi

azi

la

Stahlke's rule of prefix-lowering translates, in this system, as a rule which inserts a pitch-rise marker (+) in the environment between a noun prefix and a noun stem which begins with an obstruent. The application of this rule to the strings of (25) is shown in (26): (26)

a. 4> +

e

$

Prefix Lowering

/K

wu

a

b. ame

fi

fi

$ *

IMf" T

$+

azi

la

t $

I A I

e

$

$

ame

wu

a

0

fi

/f^

+

II

zi

la

$

ame

+

fi

azi

+

i la

I assume here, as a general convention on tone rules, that when a pcm is deleted or moved away, the two tonal feet which were connected by it merge into a single foot. We have now obtained the correct output representations for our two Ewe sentences, with one exception: our high-toned feet are still not high enough in relation to adjacent low-toned feet. To account for the "double" pitch drop or pitch rise between high and low, we must add a "balancing" rule which inserts a 4- at the beginning of a low-toned foot and a t at the beginning of a high-toned foot. I propose to state the balancing rule as shown in (29): (29)

Balancing 0

-»

pcm /

$ pcm

with a general convention that in rules of this form, the inserted pcm must point in the opposite direction from the context pcm. Applying this rule to the output strings of (26a) and (28), we obtain the correct surface contours, as shown in (30): (30)

a

(=+e iwu

ame

fi

azi

I BalanCing

la

ame

fi

(= + ame

fi a+ft)

azi

I

la

fi_ azi

la+)

Note that the double ++ and the double + t are interpreted as a long pitch drop and a long pitch rise, respectively. We have seen that the apparent spread of the low tone in Ewe can be accounted for in this framework without recourse to a leftward spreading

222

Mary M. Clark

rule. I will now show that the same is true for the spreading of the raised tone in Anlo. In the analysis which has been proposed here, the Anlo phrases of (21) above will have the underlying representations shown in (31): (31)

a.

$+$

b.

I I! I I I

nyonuvi

a wo va

$

$ 4. $

$ +

A A II

m3 a t o t o f l a - g a

'The girls have come'.

I will buy pineapple'.

The rules of Prefix-lowering and Balancing then apply, as shown in (32): a.

prefix-Lowering

b. $ + $

not applicable

4 $ 34

1 ms atoto J f 13 -g3 t $4.$+$ +t$ 4-1 $ 4 + $ + M i l 1 nyonuvi a wo va

Balancing

H tt i 1 4 ti+ A A 1 1 ma atoto il3 -g3

and (32a) undergoes a further rule which cancels out 4-t sequences to produce the output form shown in (33): (33)

$

II

nyonuvi

a

4-

wo

...

va

(= + nyo+na—tvi a wo va+)

In other dialects, this would be the end of the derivation. However, the Anlo dialect undergoes a further set of rules, as we have seen, which creates a raised tone in these phrases and spreads it to both right and left. In the framework I am proposing here, the rule which creates the raised tone is most easily stated as a rule which deletes a pitch rise immediately following a pitch drop, as shown in (34):

(34) Streamlining 4-

o t

1 2

3 :

Delete 3.

Condition: 1 is not the first element of a tone group. The application of this rule to the output forms of (32b) and (33) is shown in (35): (35)

a.

f I I nyonuvi

b.

$ a

+1 H

wo tO t

A A l | m3 a t o t o f l 3 - g 3

4

*

Streamlining | va nyonuvi ^ 4-$+ + $ Streamlining

A A m3 a t o t o

a 4

+ wo

va

$4

/ \ f l a -g3

Tone Spreading If we continue to assume that it is the pcm at the end of a tonal unit which is basic to Ewe —

in particular, that it is this pcm which deter-

mines the relationship of the whole contour to neutral, then the final foot in each of these phrases is a higher-than-neutral tone, and the foot before it, which is higher still, is a "raised" tone. Note that, as in the case of the low tone, no rule is needed to account for the "spreading" of this tone to the left. Its spread to the vight is accomplished by means of a +-Shift rule, stated in (36), which exactly parallels our earlier +-Shift rule: (36)

\-Shift 4-

1

2: Move 1 to the right of 2.

The application of this rule to the output strings of (35) is shown in (37) (37)

I^V-.

a. + $ 4-

$

nyonuvi

a

b. + $ t

wo

4va

^

4- $ 4-

+

$

nyonuvi + i> t

AAA

a

4.

wo

4.

va

+ $ 4- 4-

A ^K

m3 atoto fl'3 -g3

m3 atoto /13 -g3

Finally, we come to Clement's Cadence Rule, which translates into this framework as a rule which retracts the first of a sequence of 4- 4-1 s to the preceding syllable boundary. This rule is stated in (38): (38)

Cadence a 4- 4 ] Tone group. 1 2 3 : Move 2 to the left of 1.

and its application to the output forms of (37) is shown in (39): (39)

a.

+ $>

^f^^

nyonuvi

a

+4wo

t $ va

nyonuvi (= 1

b. 4 - $ +

+$4-+

A /As

m3 atoto fl3 -g3

a

+fiyonuvi

4 - $ + + $

I

+ 0 4wo a

va wo 4-va 4-

4-4)4-

A A\ I

m3 atoto Jl3 -g3

(= 4- m3_a 'toto /l3 -g34) In conclusion, what I have tried to show here is that by adopting a purely relational/dynamic approach to tone, we can account for the apparent

Mam M. Clark

224

leftward spread of the low and raised tones in Ewe without resorting to leftward-spreading rules

(which in this framework would take the form of

rules which shift the position of a pitch-change marker to the left). The reason Ewe appeal's

to have leftward spreading rules is that it is a

language which anchors its pitch contours on the right. In such a language, rules which insert a pcm (e.g. the Ewe rule of Prefix Lowering) or delete a pcm (e.g. the Ewe Streamlining Rule) have an automatic effect on the contour to the left of the focus, an effect which may give the appearance of leftward spread. Note that it would be too strong to say that we have succeeded in eliminating all

rules which shift the position of a pitch change to the

left; the Ewe Cadence Rule (38) and the rule of Kikuyu which was illustrated in (17) above both involve the leftward movement of a pcm. The Kikuyu example is repeated in (40) , along with a statement of the rule which accounts for the alternation: (40)

+ $ t 4-

$

As moeya A

karioki Rule:

a 1 2

+

+ $

+ $ +

4,

, kario A kiI moeya A

+ 3 :

Move 2 to the left of 1.

These examples, and others I have looked at, suggest that rules of leftward shift can be restricted very sharply. In particular, it appears that such rules always apply in a "provoking" environment (often in the environ2 ment before a +,) left —

and that they shift a pcm only very minimally to the

to the preceding syllable/mora boundary. Rules of this type, which

I will call "retraction" rules, are quite distinct from the true tonespreading rules, which take the form shown in (41): (41) pcm $ 1

2

: Move 1 to the right of 2.

Here there is no provoking environment and no limit on the length of the segmental string which the target pcm may move over. I believe rules of this form apply only in a rightward direction. In closing, I would like to point out another advantage of the analysis of Ewe which has been proposed here, besides the fact that it allows us to eliminate leftward spreading rules, namely that this analysis gives a better account of the introduction of the raised tone in AnlO. This is because the rule which introduces this tone, repated below in (42):

Tone Spreading

(42) Streamlining 4- a 1 2

+ 3 :

Delete 3.

is a very common and natural tone rule. In fact, it is the same rule which produces downstepped high tones in many African languages by converting a HLH sequence to H'HH, as shown in (43): (43) t"

"+

H L H

H ' H H

This is in contrast to Clements' analysis, where the rule which produces the raised tone (by converting M to R in the environment H

H) is a very

uncommon rule. Note that these two advantages of the analysis —

a better account of

the production of the raised tone and the elimination of leftward spreading rules —

are not independent results, but both derive from our

original assumption that Ewe is a language which anchors its tone contours at the end. If we had assumed, on the contrary, that Ewe was a fronttt 11 u rr it 1 anchoring language, then the phrase nyonuvi a WO Va (39a) would have the surface-level representation shown in (44): (44)

+ +$

+$

/rtV^ I

nyonuvi

a

wo

va

To obtain this form from the underlying (45)

+ 4 +*

+i+

nyonuvi

a

wo

va

we would need both an "unnatural" raising rule like that shown in (46): (46)

+1 2 o

t : 3

Delete 1.

and a leftward tone-shifting rule, to shift the + back to the beginning of the phrase. NOTES 1. The analysis which will be presented here is somewhat oversimplified in that I have presented the rules as if they applied all in a block, to the surface form. In fact, I believe the rules apply cyclically, subject to Kean's Strict Cyclicity Principle, beginning at the lexical level

226

Mary M. Clark

and proceeding to the level of the phrase and finally of the utterance. The assumption of cyclical application would require some small changes in the analysis, and one big change in the overall framework — namely, it will be necessary to introduce a device for distinguishing between pcm's which are associated with the end of a foot and those which are associated with the beginning of a foot. 2. Retraction rules are also found to apply at the end of an utterance. BIBLIOGRAPHY Clark, Mary M. (1978) A Dynamic Treatment of Tone, with Special Attention to the Tonal System of Igbo. Unpublished University of Massachusetts Ph.D. dissertation. Distributed by the Indiana University Linguistics Club. Clark, Mary M. (forthcoming) "On the Representation of Tone in Japanese". Clements, George N. (1978) "Tone and Syntax in Ewe", in Donna Jo Napoli, ed., Elements of Tone, Stress, and Intonation. Georgetown University Press, Washington, D.C. 21-99. Clements, George N. (1982) "The Hierarchical Representation of Tone Features". In this volume. Clements, George N. and Kevin C. Ford (1981) "On the Phonological Status of Downstep in Kikuyu", in D.L. Goyvaerts, ed., Phonology in the 1980's. Story-Scientia, Ghent, 309-357. Huang, Cheng-Teh James (1980) "The Metrical Structure of Terraced-level Tones", in John T. Jensen, ed., NELS 10 (Cahiers Linguistiques d'Ottawa, 9). University of Ottawa, 257-70. Hyman, Larry and Russell G. Schuh (1974) "Universals of Tone Rules: Evidence from West Africa", Linguistic Inquiry 5(1), 81-115. McCawley, James D. (1977) "Accent in Japanese", in Larry M. Hyman, ed.. Studies in Stress and Accent (Southern California Occassional Papers in Linguistics, 4). University of Southern California, Los Angeles, 261-302. Ostler, Nicholas (1978) "Autosegmental Theory and Japanese Tone Spread", in Mark J. Stein, ed., NELS 8, Department of Linguistics, University of Massachusetts at Amherst, 141-53. Schuh, Russell G. (1978) "Tone Rules", in Victoria A. Fromkin, ed.. Tone: a Linguistic Survey, Academic Press, N.Y., 221-256. Stahlke, Herbert (1971) Topics in Ewe Phonology, unpublished U.C.L.A. Ph.D. dissertation. Distributed by University Microfilms, Ann Arbor, Mich.

Chapter

13

Accent in Tonga: An Autosegmental Account John Goldsmith

1. INTRODUCTION In this paper, I would like to present a sketch of the major elements of the accentual and tonal system of Zambian Tonga. In this, my work is much indebted to that of Hazel Carter and A.E. Meeussen, and has been stimulated by Mark of Michael Cohen and James McCawley as well. There are a number of particularly interesting conclusions that may be drawn from the entire system, including the following: (1) Tonga is very much an accentual language, having undergone a significant reanalysis distinguishing it from non-accentual tone languages among its neighbours on the African continent. (2) Since reanalysis as an accentual language, Tonga has undergone a further change, in which the basic tone melody remained High-Low, but in which the accented tone shifted from High to Low. Thus in the modern form, it is the Low tone which is associated with the accented syllable, much as in the intonation associated with yes-no questions in English. (3) In the complex verbal morphology, a series of accent deletion rules strictly precedes the stage at which the tone melodies are inserted. This observation is consistent with the hypothesis that McCawley has suggested much along these lines. (4) If we adopt the autosegmental framework suggested in Goldsmith (1976), essentially the entirety of Tonga's tonat pattern is predictable, given the correct accent patterns, thus strongly supporting this view of the structure of the tonal system. 2. NOMINAL SYSTEM Turning to some of the data from Tonga, we will first look at some of the properties of the nominal system, after which we shall consider the verbal system. The standard argument for the accentual treatment of a language is that for words of n-syllables, one finds n classes of tonal patterns available in the language, corresponding to accent on each of the available syllables. In the case of Tonga, this argument is neither available nor necessary. It is not available because unaccented nouns are found, as well as multi-accented words (i.e., those with more than one accent).

228

John Goldsmith

There are thus considerably more tonal classes, and not too many seriously polysyllabic nouns with which to make the sort of analysis as we see for Japanese attractive. The argument for an accentual treatment of the nominal system is rather different. We shall see that all tone patterns can and must be factored into one or more (in fact, no more than two) sequences of a High tone plus a Low tone. Furthermore, while the High tone may "hang off" on the left, so to speak, we can always specify lexically which vowel is associated with the Low tone - - which is accented, that is. The largest group of nouns falls into Carter's Type A, as in (1), or Type B (2). 1. Type A:

ibu-si

'smoke

fma-kani

'news'

£ku-pa

'to give'

Ikl-sum3

'to sew'

initial accent: 2. Type B:

Imu-simbI

/si/,/kani/ 1

im-pongo

girl'

second accent: /simbi/, /pongo/

'goat'

In the first, all stem-vowels are Low-toned, but prefix tones are High; these are initial-accented words. In group (2), accent falls on the second vowel; again, all preceding vowels are High in tone. Both classes are produced automatically by the principles of autosegmental phonology as we *

*

see in (3). The L of H L is linked by rule (4) to the accented vowel of the word, and the other vowels are automatically associated with their correct tones by the Vtell-formedness Condition. 3.

ibu-si ^ 1 I \ i N ! H L

5.

imu-simbi x

\ \\

\ (

I

i I* H L

tombela

'lizard'

/tombela/

Imlba5k3nl

'flag*

/baakani/

In (5) we see examples of multi-accented words; as tombela shows, Tonga has a rule of tonal simplification (6). 6. Simplification V

tombela

A

H L

H L

tombela H L

Accent in Tonga

229

In (7) we see some illustrative examples of accentless stems, which are all Low in tone. Following a suggestion of Bill Poser's, we shall tentatively analyse these as undergoing a "recessive accent" rule reminiscent of Slavic and Japanese, as in (8), which, along with rule (6), gives the correct result (see (9)). 7. Accentless:

8. V — > v*/ #

imu-ntu

'person'

/ntu/

1-da

'stomach1

/da/

Iba-sankwa

'boys, men'

:

9. imu-ntu -•—»

where

X #

imu-ntu

imu-ntu

(8)

,

/sankwa/ *

X r V * imu-ntu

(6)

H

Thus in all the cases so far — accented class —

except, for obvious reasons, the un-

the tone pattern may be broken up into a sequence of High

and Low. I would like to make it clear now that the High that we see on the class-prefixes is in fact the realization of a High contributed by the accent on the stem rather than being the inherent tone of the prefix. This is suggested by the Low tone that appears on the class-prefixes in (7), but much stronger evidence is available, in fact. Vfe can see this when we turn to the other kind of prefix that is available, one that sharply contrasts tonally (and accentually) with the prefixes that we have seen so far. These prefixes, the so-called stabilizing elements, are accented, and thus will be associated with another H L melody, as in (10). 10. a.

ngu-taata A

* ^

0 0

/#

, _verb

istem I Iprefix

But the point to focus on is that while the exceptional behavior here is behavior of the High tone, its exceptionality derives from the position not of the High tone itself, but of the accent, or, if you will, the Low tone — the fact that the accent and Low tone is on the suffix. Thus the behavior of the High tone is governed by the location of the righthand accent, that accent of which the High is part of the accentual melody. The last example in (18) illustrates the source of downstep in Tonga. As has been noted by all earlier workers, what would otherwise be a Low flanked by Highs on both sides is realized as downstep. This is due to the

Accent in Tonga

233

effect of rule (19), which feeds rule (11), motivated above while looking at the nominal system. 19.

V

V

V

V

V

V

or,

I I I -> I N

H L H

H L H

V

V

V

H

L

H

IN

4. HISTORICAL ORIGINS In closing, I would like to briefly sketch how Tonga must have gotten to where it is today. Using comparative data provided by Carter and Guthrie, we can note the following basic changes: 20.

Original

Present

L L

(L+)

L L

(accentless)

V

V

H L

(H+)

L L

initial*

V

V

L H

(H+)

H L

second*

V

V

H H

(H+)

L L

initial*

V

V

The third and fourth lines look like tonal inversion: LH to HL; HH to LL; but the first line has remained constant (LL) and the second seems odd. Looking at the tones gives no clue as to what must have happened. But if we look at where aooent falls today, on the basis of the analysis presented here, we see that accent still falls on the vowel that would naturally have sounded most prominent in the original tonal system. Thus clearly what has happened is this: 21. Original tonal system Reanalysis as accent, with accent perceived on High tone; * LHL accentual melody. HL and HH fall together. Accent system *

.

H L becomes

*

H L.

Modern system Only one point remains to be made. We note that original HH and HL (as in Shona) were both interpreted as initial accent —

that is, is a HH

sequence, the first is perceived as more prominent. A striking synchronic trace can be found of this effect. The verbal morphology presents an extremely complex set of cases in which adjacent accents are simplified when a verb is composed of several morphemes, each of which is underlyingly accented. While the principles are complex and described in a longer version of this paper, one generalization holds for all the accent deletion rules: they all delete the accent on the right when two accents are adjacent. Thus, viewing this as a reflex of the change from a tonal to an accentual

234

John Goldsmith

system, and recognizing that the currently accented morphemes are those that were High in the original system, we see that these accent reduction rules reflect the generalization that consecutive Highs were perceived not as consecutive accents, in the original reanalysis, but rather as a sequence of Accented plus Unaccented. REFERENCES Carter, H. 1962. Notes on the Tonal System of Northern Rhodesian Plateau Tonga, Her Majesty's Stationery Office, London. Cohen, M. 1974. "Topics in Tonga Tone"

Unpublished paper, MIT.

Goldsmith, J. 1976. Autosegmental Phonology [Published, Garland Press, N.Y., 1979 I|. McCawley, J. 1978. "What is a Tone Language?" in V. Fromkin, ed. Tone: A Linguist-La Survey, Academic Press, 1978, 113-131. Meeussen, A.E. 1963. "Morphotonology of the Tonga Verb". Journal of African Languages Vol. 2, Part 1.

Chapter

14

The Asante Twi Tone Shift John M . Stewart

1. INTRODUCTION Schachter and Fromkin (hereafter S and F), in their phonology of the three best-known dialects of Akan (Akuapem Twi (am), Asante Twi (As), Standard Fante (Fa)) (1968), set up a large number of separate non-general synchronic tone rules which have the general effect of moving the tone pattern of the word one syllable to the right in Asante but not in Fante. In this paper I attempt to explain this rightward movement in-terms of a diachronic tone shift similar to that posited by Clements and Ford (1979) as having affected Kikuyu, and I indicate how this rightward tone shift appears to have interacted with two other tonal innovations in the shaping of the highly complex synchronic situation which we find today. 2. THE ASANTE RIGHTWARD TONE S H U T The Asante rightward tone shift is tentatively formulated as in (1): (1)

aL >

-aL




aL




L, +L: low tone H, -L: high tone L, H : floating L, H o o floating

: carried by zero syllable

zero syllable: syllable containing no segments (Applies simultaneously to all aL -aL sequences within the word). This says that where two successive syllables differ in tone register (i.e. where they display the tone register sequence H L or L H), the change in tone register is deferred from the beginning to the end of the second syllable; and that if the second syllable is word-final, a word-final zero syllable (i.e. a syllable containing no segments) is added to carry the original tone of the original word-final syllable. The examples in (2) illustrate the predicted correspondence Fa H L = As H H (L) : o (2)

=

As H H (L)

a. kofi na akoma^

Fa H L

kofi akoma

Kofi's heart

kofi abSgye

Kofi's chin

kofi na ab£>dwe

236

John M. Stewart b. kofi nà àdàkàà

kòfi adakàa

Kofi's box

kòfl nä àbifòé

kòfi äböföö

Kofi's messengers

mà kòfi mfér£~

c. mà kòfiriiférEnò

nò mà kòfi mmisà"

mà kòfiriibisànò

nò

Kofi should call him

Kofi should ask him

The L of the Asante examples in (2c) has no surface realization of its o own but is manifested by the blocking of the assimilation of the L of the object pronoun to the H preceding the L; cf. kofi + fer£ + no ->• kofi fer£ no 'Kofi calls him', which illustrates S and F's rule of 'pronoun tone raising' (P17, pp. 236-9, 249) which affects Asante but not Akuapem or Standard Fante. The examples in (3) illustrate the predicted correspondence Fa L H = As L L (H) : o (3)

Fa

As L L (H) o

L H

a. ò ròkShwé

ò rèkàhwé

he is going to look at it

b. ò nhw£ nò

ò nhwE" no

he does not look at him

The H of the Asante example in (3b) has no surface realization of its own o but is manifested by the assimilation to it of the L of the object pronoun by the 'pronoun tone raising' rule to which I have just referred; cf. hwE + no +

hwE no

1

look at him 1 .

The examples in (4) illustrate the predicted correspondences Fa L = As H L (cf. (2) above) and Fa L H = As L L (cf. (3) above) simultaneously: o (4)

Fa

L H L

As L L H L o

a. ò ribisà mè

ò ribisà" mè

he is asking me

b. £> ròbfihwè mè

0 rèb£hwé~ mè

he is going to beat me

The correspondences illustrated by the examples in (5) are essentially the same as those already seen, but the first tone is carried by a zero syllable in Fante and the corresponding zero syllable has been automatically deleted in Asante as it has ceased to differ in tone from the following syllable: (5) a.

Fa

H L o

As H (L) o

kòfi né "sère

kòfi sér£

Kofi's thigh

kòfi nó "kòkò

kòfi kókò

Kofi's chest

Asante Twi Tone Shift

237

köfl nó 'bàfôé

kôfi hôfôô

Kofi 1 s messenger

köfi né 'sèkàn

kôfï sékáñ

Kofi's knife

kôfi 'hw£ mè

köfi hw£~ mè

Kofi looks at me

yàw "hw£ mè

yàw hw£~ mè

Yaw looks at me

b. Fa L H o

As L

5 "kShwê

ö kàhwê

he goes and looks at it

köfi ~ ( ' )köhw£

köfi kä(")hw£

Kofi goes and looks at it

[-" "I c. Fa L H L o £> ~bôhwè mê i kôfi ~(')bôhwè mè

As L H L o £> b £hwé " mè

he comes and beats me

kôfi b£(')hwé~

Kofi comes and beats me

mè E"--] ö ~bisà mè i köfi ~(')bisà mè

ô bisa" mè i kofí bi(*)sá" mè

5 ~rôhw£• mè kôfi '(')rôhwÈ mè

he asks me Kofi asks me

J

à rèhw£" mè i kôfi rè(")hwê~

he is looking at me Kofi is looking at me

mè

The Fante word-initial zero syllables have no surface realization of their own; in the last example in (5a) the zero syllable with H tone is manifested by the assimilation to it of the final L-tone syllable of

yaw

1963: 68-70; S and F 1968: 121-2); in the four examples with

kofi

(Stewart as

subject in (5b) and (5c) the zero syllable with L tone is manifested by the downstep (S and F's 'downdrift') which it conditions (Stewart 1963: 57-8; S and F 1968: 121-2). Asante word-final H L is realized simply as H in pause, e.g. (6)

¡5 + b£hw£ + mè

ö b£hw£ mé

he comes and looks at me

i) + bÈhwé~+ mè

î> bÈhwé mê

he comes and beats me

à + b£hw£

S b£hw£

he comes and looks at it

S + b£hwé~

î> b£hwé

he comes and beats it

but

I argue below (section 4) that the tone of the zero syllable is assimilated to that of the preceding non-zero syllable; the zero syllable is thereupon automatically deleted as its tone is then identical to that of

238

John M. Stewart

an adjacent syllable. Asante word-final L H also is realized as H in pause, e.g. o (7)

i> + nhwÈ" + nò

5 nhwÈ no

he does not look at him

ì> nhw£

he does not look at it

but Ò + nhwè'

Here we can say that the tone of the non-zero syllable is assimilated to that of the following zero syllable; the zero syllable is thereupon automatically deleted as before. It will be seen that this latter assimilation has the effect of reversing the change from L H to L L H by the diachronic tone shift. This suggests a solution to the problem that in certain cases in which our tentative formulation of the diachronic tone shift predicts the correspondence Fa L H = As L L H we find L H in both Fante and Asante, e.g. (8)

Fa L H

=

As L H

a. o + kyer£ + me

o + kyer£ + me

-»• O kyer £ me

he shows me

i> kyer£ mS (not: *£> + kyèrè' + mè *ò kyèrè mé)

b. ò + kyèr£ •*• Ò kyèré

ò + kyèr£ •+

he shows

ì> kyèrÉ

(not: *ò + kyèr£' ò kyèr£) In such cases it would appear that the reversal has been extended analogically from prepausal and other positions in which it is phonologically regular (cf. (8b)) to the positions such as that before a prepausal object pronoun in which it is not phonologically regular (cf. (8a)), with the result that all trace of the shift has been obliterated. This would go a long way towards explaining why there appears to be no evidence at all of the tone shift in nouns which are not in the possessed form (which is illustrated in (2a-b) and (5a)), as the tone pattern L^ H (a high tone preceded by one or more low tones) predominates in the uninflected form of the noun. S and F account as follows for the examples quoted in (5c) above of the correspondence F a L H L (9)

=

AsLHL:

a.

5 b£hwe me Am-As: Tone raising in monosyllabic VRs

£> b£hwé mè

Asante Twi Tone Shift

239

(VR: verb root) (P19, pp. 178-81, 249) Am-As: Ingressive low tone

As

ù b£hwé mè

(P47, pp. 231, 256) Cf. Fa ò ~bóhwè mè

=

As

5 bÈhwé ~ mè

in (5c) above. b.

ò bisa mè Am-: Disyllabic VR tone reversal (P20,

ò bisá mè

pp. 195-7, 249) Tone incorporation

ò bisâ mè

(p85, pp. 111-5, 264) Tone simplification

As

ò bisa mè

(p87, pp. 111-5, 265) Cf. Fa ò "bisà mè

=

As

ò bìsà~ mè

in (5c above). c.

£> ròhw£ mè Am-Fa: Monosyllabic VR tone reversal

Fa

ò róhwÈ mè

(p50, pp. 217, 257) Cf. Fa £> ~róhw£ mè

=

As

ò rèhw£~ mè

in (5c) above. They account as follows for the example Fa S nhw£ nò i> nhw£' no

•+ i> nhwE nó

=

As 5 nhwE" nò

quoted in (3b) above of the correspondence Fa

L H = As L L H : o i) ñhw£ nò

(10)

As: Pronoun tone raising (already seen) As: Tone lowering before high tone

S ñhwÉ nó As

Ò ñhwé nó

pronoun (pl8, pp. 238-9, 249) As H H L such as o those quoted in (2c) above by making their rule of 'optative tone' (P49,

They account for examples of the correspondence Fa H L

pp. 222, 257) dialect-sensitive. As they confine their attention almost exclusively to the verb they disregard examples of the correspondence Fa H L = As H H such as those quoted in (2a-b) and (5a) above. 3. THE AKAN LEFTWARD TONE SHIFT In the case of the first two of the three contexts in which I have noted the correspondence Fa L H L = As L H L (see (5c) and (9) above)

I am in

agreement with S and F in taking the Fante to be the original, though I treat it as a diachronic rather than as a synchronic original; in the case

240

John M. Stewart

of the third context, however, in which the second non-zero syllable is a H-tone VR, whereas I still take the Fante to be the diachronic (protoAkan) original, S and F take the Asante as the synchronic original and derive the Fante from it by a leftward tone-shift rule. In this section 1 make explicit an assumption which is implicit in the treatment in section 2 and which accounts for the apparent disagreement with S and F, namely the assumption that S and F's synchronic leftward tone-shift rule was already in operation as a synchronic rule in proto-Akan. Consider the examples in (11): Fante

(11)

L-tone VR

H-tone VR

Prefix L-tone

S ròhwè mè

b. i> ~róhw£ mè

he is beating me H-tone

he is looking at me d. ä "bóhwÉ mè

5 ~ bóhwè mè he comes and beats me

he comes and looks at me

Asante L-tone VR

H-tone VR

Prefix L-tone

e. 5 rehwe me

H-tone

g- i> b£hwe~ me

f. 5 rehw£' me h. 5 bEhwf me

It will be seen that even if we disregard Asante altogether we still need a synchronic leftward tone-shift rule to account for Fa t> "rohwE me. My assumption is that the Asante rightward tone shift with which I am chiefly concerned in this paper operated on the output of this rule and thus had the effect of reversing the Akan leftward tone shift which gave rise to the rule. The leftward tone shift has clearly affected all the non-Asante dialects which I have examined, including not only the Fante and Akuapem Twi dialects spoken to the south of Asante but also the Brong dialect spoken to the north; since there is no evidence of common innovations affecting the non-Asante dialects alone, and since we must in any case posit a rightward shift in Asante to account for examples such as those in (llg-h), the simplest explanation is that a leftward shift affected the whole language and was subsequently reversed in Asante by a rightward shift of more general scope.

Asante

Twi

Tone

Shift

241

Since there are no Akan dialects which have not undergone the leftward shift the task of formulating that shift cannot be expected to be a simple one, and it must therefore be left to some future occasion. 4. ABURA FANTE PREPAUSAL H-LAG It was seen in connection with the examples in (2) and (3) above that Asante differs from Akuapem and Standard Fante in having a rule (S and F's 'pronoun tone raising') whereby a low-tone object pronoun, when in pause, is assimilated in tone to an immediately preceding high-tone syllable, whether that syllable is zero or not; the examples in (12) illustrate: (12)

o + réhwé + me 5 + behwé + me

+

5 réhwé me

he is beating me

o b£hw£ mé

he will look at me

5 + réhwg" + mé -+• o réhwé mé

he is looking at me

5 + ñhwé'

he doesn't look at me

+ mé

o ñhwé mé

It was noted that in such examples the surface tone of the object pronoun serves as a surface indicator of the presence or absence of an underlying verb-final zero syllable. This situation suggests the possibility of a historical relationship between the introduction of the tonal assimilation rule on the one hand and the rightward tone shift on the other: the first can be seen as either precipitated by, or creating favorable conditions for, the second, and both have affected Asante without affecting either of the other two dialects mentioned. In this section I offer one argument for preferring the drag-chain to the push-chain interpretation: there is a dialect, namely Abura Fante as described by myself (Stewart 1963), which displays the tonal assimilation of prepausal object pronouns (pp. 97-101) but which has not undergone the rightward tone shift; the examples in (13) (which show only the surface tones) illustrate: (13)

Standard

Abura

Asante

Fante

Fante

o róhwé nó

O róhwé nó

O rehwe no

he is beating him

5 kóhwé nó

O kóhwé nó

O kohwe no

he goes and beats him

5 kóhwé nó

O kohwé nó

o kohwe no

he goes and looks at him

O ñhwé nó

o ñhwé nó

O nhwi: nif kAtAm ] "you must cut it with a knife" { l i t . "you must take knife cut it"). Grammatical similarities between the Manding languages and the Atlantic Creoles (especially Krio) were dealt with in the earlier paper; the present examination is primarily lexical.

The Manding

in Sierra

Leone

Ethnic groups speaking dialects of the Manding language in Sierra Leone include the Bambara,

Dyula,

Kuranko,

Mandinka,

Maninka

and Mcmyaka.

By far the most widely represented of these is the Maninka, usually referred to as Mandinka or Mandingo in reference to Sierra Leone (but Madtrjka

in

Krio*). Mandinka, Bambara and Dyula speakers are not permanently resident in the country, coming from the Gambia, Mali, and Upper Volta/Ivory Coast. The Mandinka entered Sierra Leone during the latter half of the 15th century primarily as traders, but with the Susu served the additional purpose of spreading Islam throughout the area. They settled in several areas, including Forekaria (the "Mandingo Country" of early European writers), Karina, Fintonia and Kabala, where they established themselves as traders, especially in cattle, and disseminators of Islam. Their influence has been felt further south also; Little (1951:28) has noted Maninka influence upon 2 the Mende physical type as well as in Mende social organization . Like the Yoruba, the Maninka are noted for their preference for closely-knit community life. During the first half of the 19th century, many left their northern settlements and migrated to Freetown, where they established themselves around the western limits of the city. Here they acted as brokers for Maninka traders resident upcountry. By 1850, most had moved from the fringes of town to Magazine Cut at Destruction Bay, because of its proximity to the main Freetown market. Here they were more easily able to carry out their trade, especially as butchers and cattle-dealers, and to establish Koranic schools. The right to operate Koranic schools came at the end of a difficult period for the Maninka. During the Yoruba uprisings of the 1830s, the disciplinarian governor Findlay misguidedly blamed the Maninka and the Fulani for causing the unrest, claiming that they had led the

250

Ian F. Hancock

Yoruba astray with Koranic doctrine^. In 1833, he banned all Maninka from the (mainly Yoruba) village of Waterloo, and shortly thereafter forbade any Muslim of any ethnic background into the villages, at the same time banning all but European dress. Despite his efforts, the Yoruba allied themselves with the Maninka and Fulani alfas in Freetown, and Islam became firmly established in the colony. The Maninka have always been characterized by their independence; while not themselves actively engaged in the Temne attack upon Freetown in 1801, they allowed King Tom's men to take refuge among them in northern Sierra Leone. During the slave trade, the Maninka were much in demand for their reputed mental and physical dexterity, especially in Spanish America. Slave revolts, such as that at the Scarcies River, were frequently Maninkainspired. They were also noted for their humanitarian attitude toward their own slaves (Little 1951:38), though they were also known to bring children taken in slavery to Freetown, ostensibly to be raised in "white man fashion" but in fact to be used as unpaid servants (Fyfe 1962:270). Some of these were later lured out of the colony and traded for cattle. The Fulani were similarly involved in this practice. During this period, many of the Freetown and village (especially Gloucester) Maninka permanently adopted the Creole life-style, taking European names, and assimilating to the extent that their descendants may no longer be identified as Manding. Those who are still identifiable are probably all indigenous to the area, and not the descendants of the Recaptives; they are nowadays employed as weavers and dyers, goldsmiths, silversmiths, cattle traders and mosque officials in Freetown. In the Provinces the Maninka are usually traders, though since the decline of the diamond industry, many have migrated north to Guinea (Dalby 1962:63).

The Manding

languages

Manding constitutes a member of the northern branch of the Mande unit, others including Vai and Susu. As well as the dialects mentioned above, a Maninka-based lingua franca has developed throughout the area as the normal medium of intercourse among most of the Manding-speaking communities. This is known as Kceqgbe, that is, "clear language" (Dalby, toe. eit., and Westermann & Bryan 1970:71), and it is probably from this that most lexical adoptions into Krio have come. There are something over 241,415 Manding spea4 kers (all dialects) in Sierra Leone at the present time .

Manding and Sierra Leone Krio Maninka Phonology Maninka phonology differs little from that of the other Manding dialects; such differences as do occur are mainly tonal or, less commonly, consonantal (e.g. Maninka has the phoneme /j/ where all other dialects have /y/. The Consonants Maninka consonants may be tabulated as follows: Bilabial

Plosive

p

Nasal

Labio-

Dental-

Dental

Alveolar

b m

k

n f

1

Roll

r w

Glottal

g

gb

j

s

Lateral

Labio-

ng eh

Fricative

Velar

Velar

t d

Affricate

Semivowel

Palatal

h

y

The Vowels Oral

Nasal

No sounds occur in Manding (based largely on Mandinka forms where Krio is concerned) which do not also occur in Krio^; that forms in the two languages are therefore not more frequently alike is due to the number of items which have entered Krio via Temne, whose phonology differs in several respects from that of Manding. Like Krio, Manding operates a two level tonal system, viz. high and low, although in the latter language these may both occur on the same syllable, i.e. as an upward glide, without significant lengthening: nisi "cow", sa "snake". Indications are that Krio tonal behaviour is most like that of Manding, e.g. in its two levels and terracing, and in its final-syllable high tone which operates on English-derived items and may have been incorporated

252

Ian F. Hancock

prior to the multilingual situation in 19th century Freetown^. Even items derived from Yoruba, the African language which has had the most extensive effect upon Krio, have lost their historical three-level distinction, and there are several examples of tone-reversal on disyllabic African-derived 7 items . Phonological Modification Very little modification appears to have taken place for items adopted directly by Krio from Manding; such modifications as do appear are generally indicative of such items having entered Krio via Temne: MDG garaQg(b)a

> TEM k=araQgba

> KRIO karaQgba

"body louse"

MDG kodo MDG kSsarjge

> TEM u=k0th0

> KRIO koto

"Fulani trader"

> TEM k=asar)ke

> KRIO kasalJkS

"shroud"

Where the Krio form remains unmodified, this suggests an immediate acquisition from Manding: MDG buya

> KRIO bQya

(of.

TEM

=boya)

"Lagniappe as an inducement to further custom or as a favour to a valued client" MDG santigi

KRIO santigi

(of.

TEM

=santh3kl) "chief's advisor"

Routes of Semantic Adoption No intermediary language other than Temne seems to have acquired an item before it was adopted into Krio. While Susu shares many cognates with Manding, and is spoken throughout a large area of northern Sierra Leone, the Manding are more numerous, and more frequently in contact with the Creoles, and many Susu themselves employ the Manding-based lingua franca Karjgbe in their commercial dealings, rather than their own language. Thus Manding items in Krio have presumably entered directly from Manding, or via Temne. Where cognate forms occur in other languages spoken in Sierra Leone, such as Hausa or Fula, e.g. nono, bilya, k3ya, &c., the same loanwords also appear to be present in Temne. Semantic Areas of Adoption The Manding (mainly Maninka) in Sierra Leone came primarily as dispensers of Islam, and as traders and entertainers, and it is in these areas of the vocabulary that their language has most influenced Krio. It is significant that while highly regarded in their religious and mercantile capacities by indigenous Sierra Leoneans, the Manding are poorly thought of as entertainers, a profession usually considered to be synonymous with begging. The Creole population associates the Manding primarily with their strong

Manding and Sierra Leone Krio

253

adherence to Islam, and this is reflected by the fact that several items in Krio Muslim vocabulary are of indigenous Manding origin rather than of ultimate Arabic derivation. The only other language to have provided g non-Arabic Islamic vocabulary is Yoruba . Names of persons, and their behaviour, are well represented, and so are anatomical items and terms for articles of dress. A number of opprobrious terms have also been adopted {e.g. kamaboro, mokofe, fontoba, &c.), as well as a pejorative interpretation for a small number of items: soriba, a given name in Manding, but "houseboy" in Krio, or karagke "leather worker" in Manding, but "bad cobbler" in Krio. A. Religion and Magic bofema bofima karam3k3 karamS

A powerful composite charm. Cf. MDG bori "medicine" + fima "black" A teacher of Koranic doctrine. Cf. MDG karS-m3(g)5/-m5(X)3, "teacher"

kamS

- W

/

[ Q

a

Condition: Q does not contain a heavy syllable. b. [ 1 CVCWCV (V )C J lr , [ plural] This new version of the rule generates the correct results for all broken plurals considered thus far. In particular, it provides for a difference in the locus of W-insertion in the two triliteral noun types, the [CWCV(V)C] and the [CVCVVC] singulars. The unassociated C position created by restructuring in the former (17) follows the first syllable while it follows the second syllable in the latter (23c). This accounts for the tvro different slots where W (or the

7

derived from w) can appear.

•¿.4 Nouns [CVC(V)C\ Like the nouns of the preceding section, these forms also show a discrepancy in plural formation between masculine and feminine gender. Something over 90% of the masculine nouns with the singular template [CVCC] take one of the three broken plural patterns in (25); a smaller but significant percentage of feminine nouns of the same pattern take these plurals, whereas others, under lexical government, have a different broken plural type disg cussed below in section 4.3. CaCC naf s

nufuus

'soul'

fcahr

bihaar

• sea*

farx

?afraax

'young of a bird

qasi»at

qisaa9

'dish'

hukm

'ahkaam

'judgment'

rumh

rimaah

'spear'

burd

buruud

'robe'

CuCC

Arabic Broken Plurals

305

ruq9at c.

riqaaö

'scrap of paper'

himl

'ahraaal

'load'

qidh

qidaah

'arrow'

dirs

duruus

'molar'

hiqbat

huquub

'period of time'

CiCC

The varying broken plural patterns of masculine forms have been listed in order of frequency, from greatest to least. Thus, nouns CaCC are more likely to have plurals CuCuuC than are singulars CuCC. Apart from these relatively uninteresting aspects of statistical preponderance, there is no effect of the singular vocalism on the form of the plural, and all three plural types occur with each singular type. For the less common feminine broken plurals, I have given just one example of the most frequent plural of each type. Again, some considerations of allomorphy considerably simplify the morphological analysis. Two of the three broken plural types in (25) share the same prosodic template [CVCWC]; the third has initial 7 a with a following [CCWCJcanonical pattern. Levy (1971) has argued that the underlying form of this last type is CaCaaC, but subject to a morphologically-restricted rule of methathesis and

9

epenthesis which she formulates. Such a rule

has considerable independent support from other aspects of Arabic morphology, and I will assume its application here. We have, then, three basic broken plural patterns in this class: CiCaaC, CuCuuC, and CaCaaC. All are obviously formed on the prosodic template [CVCWC] and consequently differ only in the vowel melody associated with it. The three melodies are given in (26); it is clear that selection among them is entirely lexical:^ (26)

Vowel Melodies a.

[i

b.

[u],

c.

[a]

a]

These melodies present no problems of association; the universal conventions in (2) will w r k without additional stipulations. There is an apparent similarity between the [CVCWC] template of this plural class and the templates generated by the Plural Template rule (24). As is clear in (27), insertion of VV after the only heavy syllable of

306

John J. McCarthy

[CVCC] singulars yields the correct plural form: (27)

O

A + [CVCC1

\ll

nfs

°

°

A A [cvcwc ]

\\/ nfs

This noun type, then, is subject to the same insertion rule as the previous classes. It can be fully incorporated into the analysis by slightly broadening the scope of the Plural Template Filter as in (24b'): (24)

b'. [ C V C W C (V(V)C)] [ p l u r a l ]

No other distiction between the different singular types need be made. Disyllabic singulars [CVCVC], however, do offer a fairly interesting difficulty. Although this singular class is significantly less common than the [CVCC] type, nevertheless it quite regularly forms plurals along the lines in (28):

(28)

CVCVC qadam

7

aqdaam

1

foot step 1

9inab

,

a9naab

1

grape 1

jabal

jibaal

•hill'

kabid

kubuud

1

liver'

riqaab

1

neck'

raqabat

Clearly these forms are identical to the plurals of [CVCC] singulars. The most direct analysis would be to transform the template [CVCVC] to [CVCC] by a new rule, ordered before VV-insertion: (29)

Vowel Ellipsis V

—*• 0 /

[ CVC

C ]

/

Plural

After Ellipsis applies to the prosodic template, the derivation of broken plurals from erstwhile [CVCVC] singulars will proceed exactly as in (27) above. The rule of Vowel Ellipsis has independent motivation from a somewhat unexpected quarter —

a singular/plural type we have not yet discussed.

Arabic has a large number of adjectives referring to colors and bodily defects, related to a corresponding verbal derivational class. The masculine

Arabic Broken Plurals

307

singulars of these words are of the form 9aCCaC

(with irregular feminine

CaCCaa) and the common gender plural is CuCC. This pattern is exemplified in (30): (30)

?ahmar

humr

'red'

'ahdab

hudb

'hump-backed'

The masculine singulars are derived from underlying [CVCVC] templates by a rule of metathesis and '-epenthesis we have already discussed (Levy 1971), so this pattern, together with the Ja] melody, comprises the characteristic morphology of adjectives of color or defect. In the plural this template is subject to Vowel Ellipsis (29) and it arguably receives the [u] melody in (26b). A comparison of the ordinary [CVCVC] noun pattern and the formally identical color and defect category is revealing. Although both types are subject to a morphological rule of ellipsis in the plural, only the former also undergoes W-insertion, bringing it into conformity with the template filter. Technically, the color and defect class is exceptional with respect to insertion and the filter, and permits only one of the three vowel melodies for disyllabic plurals.

Excursus: Some Properties of Weak Roots Weak roots radicals —

—

those that contain a high glide W or y as one of their

contribute substantially to the allomorphy of Arabic. A full

study of this problem far exceeds our goals here, as it involves many other aspects of nominal morphology as well as the verb (see Brame 1970, Levy 1971). But certain facets of this complex whole do touch directly on the representation of broken plurals as well as contribute to our understanding of the prosodic theory. This analysis is, then, by no means exhaustive. Roots known as Ill-weak — radical —

meaning they have a high glide as the third

present relatively simple alternations in the plurals of the

[CWCVC ] singular base. The forms in (31) are representative: (31)

baadiyat

bawaadii

'desert'

jaariyat

jawaarii

'girl'

As these plural forms are cited, they include one of the case desinences u 'nominative' or i 'genitive' suffixed to the stem. Therefore, bawaadii might be analyzed as bauaadiy + u or i, with the final vowel showing the

308

John J. McCarthy

effects of progessive assimilation. (The accusative, with suffix a, is baadiya.) The final root consonant y is lost by virtue of a rule that deletes a C on the template intervocalicallV after a short vowel, providing that C is associated with a high glide:

(32)

C

-»

0

/

CV

V 1

r-consi [_+highJ A final formulation of this rule would clearly be much more complicated, since it must exclude sequences like iya as well as others, but this version will suffice. Since rule (32) applies on the template level only, it leaves the associated root material, the high glide itself, stranded without an association. The unassociated material receives no phonetic realization, the equivalent of being deleted:

(33)

a i /K I [cvcwcvl+v \ / fady

The unassociated C position then undergoes U-insertion (18). Although formulating this glide deletion rule prosodically offers no improvement over a segmental rule for the forms in (31), it does make a very interesting difference with the plurals of [ C V C W C ] nouns formed on weak roots. Final radical high glides render the broken plurals of this noun type extremely opaque:

(34)

a. Ill-y

Roots

hadiiyat

hadaayaa

'present'

nuqaayat

naqaayaa

'pick, choice'

b. III-W

Roots

'itaawat

°ataawaa

'tax'

hiraawat

haraawaa

'stout stick'

Although there is much that is puzzling in these plural forms, note in particular that, in the third C position of the template, instead of

9

derived

from inserted W, we find the root glide W or y. This observation is funda-

Arabic Broken Plurals mental to our analysis of these forms. Consider the representation of one of these broken plurals immediately before the application of U-insertion:

(35)

a

i

A \ [cvcwcvc]+v whdy/

The most important assumption we will make is that this form is then subject to rule (32), ordered before W-insertion. By association convention (2b), the now-stranded final high glide of the root will reassociate with the adjacent free C position in the template, yielding the output in (36). In other words, the root material persists despite the deletion of the template position with which it was associated:

(36)

a

i

[ CVCWCV]+V

whdy /

This reassociation, which preserves the final high glide of the root, does not occur with the plurals of [CWCVC] nouns, as in (33), because of the prohibition against many-to-one associations. In that case, the stranded y or W cannot map onto the already occupied C-position on its left, and therefore remains unassociated. The reason for the final long aa in hodao.yaa, viiich is invariant in all three cases, is unclear. Conceivably some new rule deletes the melodic element i, with the result that the a spreads to the stem-final vowel and triggers assimilation of the desinential vowel. But it is difficult to find independent support for such a process, and so the problem must remain 8 open. 3. DIMINUTIVES There are very striking similarities between diminutives and the broken plural patterns discussed above. These similarities are apparent from the data in (37), which are representative of the diminutive pattern of essentially all nouns regardless of their broken plural pattern:

309

310

John J. McCarthy Singular a. jundab sultaan b. &ankabuut

Plural

Diminutive

janaadib

junaydib

salaatiin

sulaytiin

Sanaakib

9unaykib/ 9unaykiib

safaarij

sufayrij/sufayriij

xawaatim

xuwaytim

jaamuus

javaamiis

juwaymiis

d. diinaar

danaaniir

dunayniir

e. jaziirat

jazaa^ir

juzayyir

samaa'il

sumayyil

nufuus

nufaysat

hukm

'ahkaam

hukaym

qidh

qidaah

qudayh

qadam

'aqdaam

qudaym

safarjal c. xaatam

simaal f. naf s

9

The diminutive is an entirely productive category in Classical Arabic; it is therefore of some interest to incorporate this system into our analysis. In fact, the addition of diminutives to the grammar is almost trivially simple. It is apparent from all the forms in (37) except (37f) that the vocalic melody of the diminutive is [u a i] . It is also apparent that the diminutive prosodic template is identical to that of the broken plural except for the presence of an internal VC sequence (phonetically ay) rather than a W

sequence. Let us suppose that the diminutive template is

actually the same as the plural one. A left-to-right napping of the [u a i] melody onto this template, purely in accordance with the conventions in (2), will yield the output in (38): (38 )

a. f CVCWCVC 1 \ I V u a i b. [ C V C W C W C 1

I

IV

u a i

c. [ CVC W C 1

I

IV

u a i

It is a fact of Arabic syllable structure that ai and au sequences are always realized as vowel plus glide combinations —

that is, the high portion

is nonsyllabic. Thus we may consider the change of V to C in the prosodic template to be a further consequence of the minimal restructuring needed

Arabic Broken Plurals

311

to bring the representation into conformity with the canons of Arabic syllabification. The output will appear as in (39): (39)

a. [CVCVCCVC]

I IV

u a i

b. [CVCVCCWC1

I IV

u a i

c. [CVCVCC]

I IV

u a i

To achieve this result, it is also necessary to designate the diminutive melody as an exception to the Vowel Association rule (6). This ensures that only the universal conventions apply to yield a left-to-right pattern of association. It follows, then, that for the diminutive precisely the same template apparatus as the broken plural is sufficient. The grammar must only stipulate the diminutive melody [u a i]. For completeness we will go through each diminutive type in detail. The quadriliteral diminutives in (37a) need no comment, but we can note that the quinqueliteral ones (37b) show the same pattern of loss of consonantal material described in section 2.1, as well as the indeterminacy of final vowel length predicted by their arbitrary selection of template. The diminutives of [CWCV(V)C] nouns have the same pattern of inserted W as the broken plurals, whereas the exceptional broken plurals of this class (37d) retain their special pattern of root-to-template association in the diminutive, showing that Minor Association (21a) applies to diminutives as well. The diminutives of nouns [CVCWC] have a somewhat suprising geminate y between the second and third syllables. Since these forms are subject to W-insertion, we would expect a sequence yw like *ju.zayG)ir. Not surprisingly, a rule /yw/->yy is not difficult to motivate on independent grounds. For example, the diminutive of 9urwat 'handle', in which the W is radical rather than inserted, is 9urayyat from /9uraywat/. So these forms present no special complications. Finally, the diminutives in (37f) have several interesting characteristics. The diminutive of hukm shows the underlying form of the prosodic template, before the application of metathesis and '-insertion in the broken plurals. This metathesis is inapplicable in the diminutive because the first vowel

312

John J. McCarthy

is u rather than a.

The lack of any surface i vowel in the diminutive fol-

lows simply from the shortage of V positions in the disyllabic prosodie template. The left-to-right association, then, accounts for this difference between the vowel patterns of diminutives depending on their length. The diminutive of qadam Includes prior application of the Vowel Ellipsis rule (29), which must therefore include the diminutive category in its scope. The end result is as claimed: diminutivès require no addition to the descriptive apparatus other than a new vocalic melody. This is fairly striking confirmation for the analysis. 4. CITHER PLURAL TYPES A number of other broken plural types occur, some as options with noun classes already discussed, some confined to particular singular patterns, and some with wide distribution but very low frequency. All are systematic exceptions to the W-insertion rule and template filter in (24) . We will very briefly consider the most interesting of these here, and in the concluding subsection outline a few of the others.

4.1.

[CWCVC]

Participles

Masculine active participles of the first verbal derivational class take the pattern [CWCVC] with [a i] vocalism. When entirely productively derived they form sound or suffixing plurals, as described in section 1. But when at least partly lexicalized, the majority of these words select, under lexical government, one of the two broken plural patterns in (40):^ (40)

a. saajid saamir

sujjad

'prostrating oneself'

summar

'conversing at night'

b. haakim

hukkaam

'judge'

jaahil

juhhaal

'ignorant'

No other nouns except occasional feminines of the same type form their plurals in this way. Clearly we could simply stipulate, as any analysis must, that these plurals are formed this way, and leave it at that. But there are significant generalizations to be captured if we attempt a more thorough treatment under the prosodic theory. It is apparent that the plurals in (40a) have almost exactly the same canonical pattern as the corresponding

Arabic Broken Plurals singulars, but that they substitute gemination of the medial consonant for length of the first vowel. This generalization can be expressed by an autosegmental rule (41) that adds an association of the middle root consonant with the second V position of the prosodic template: (41)

Participle Plural Rule f CWCVC 1 [

]

The addition of this line of association yields a representation like that in (42), which will be automatically restructured by changing the V associated with consonantal material to a C: (42) [ CWCVC 1

\ // smr

*

[ CVCCVC 1

w/

smr

About half of these forms are also subject to an additional rule lengthening the vowel of the second syllable in the plural (40b). Some lexical items undergo this rule or not, in free variation. The other aspect of the problem presented by this plural type is the vowel melody [u aj. Although this is unrelated to any other vowel melody developed so far, we shall see shortly that it is characteristic of certain kinds of plurals with human referents, like all the participles [CWCVC]. 4.2 Nouns [CVCWC]: Other Plural Types Although, as we saw in section 2.3, most feminine nouns [CVCWC] have trisyllabic plural stems, as do a few masculines, nevertheless most masculine nouns of this type form their plurals rather differently. There are two main lexical types. In the first, a basic distinction is made between human and nonhuman referents, with about three-fourths of the former taking the broken plurals in (43a) and a large majority of the latter taking those in (43b) (many of the remainder exceptionally have the same plural pattern as the feminine nouns in (22)) . (43)

a. Human waziir

wuzaraa?

'vizier'

?amiir

?umaraa?

'commander'

314

John J. McCarthy baxiil

buxalaa7

'stingy'

hakiim

hukamaa?

'wise'

janaab

'ajnibat

'wing'

h imaar

7

9amuud

7

b. Nonhuman

qadiib

ahmi rat

'ass'

a9midat

'pillar'

'aqdibat

'branch'

The plural forms in (43) all have a feminine suffix: aa9 in the human nouns and at in the nonhuman ones. Let us first consider the vowel melodies associated with these broken plural types. The melody [u a] is characteristic of human broken plurals, both of these nouns and the participles discussed in the preceding section. It is an obvious virtue of this analysis that we can isolate this melody, apart from other aspects of the plural forms, as a human plural morpheme on an autosegmental tier. The other, nonhuman melody [a ij can be identified with the vocalism of the trisyllabic plural stems given in (5). So neither melody actually represents a morpheme not independently needed. The treatment of the prosodic template of these forms also does not present any truly new difficulties. We can, in fact, consider the stem template of the broken plural to be identical to that of the singular [CVCWC], subject only to suffixation of one of the two feminine endings. Two rules that have already been motivated then apply. The first, discussed in section 2.1, shortens stem-final vowels before feminine suffixes in the plural. It is applicable in some quadriliteral plurals as well as the forms in (43):

(44)

V

->• 0 /

V

C~] + Feminine! Plural J J

A second rule, which is needed independently for the broken plurals of section 2.4, takes underlying /CaCVC/ to /?aCCVC/ by metathesis and insertion of a glottal stop. This'rule will apply in (43b) to yield the correct plural forms. The other major plural type for [CVCWC] nouns is illustrated in (45): (45)

kitaab

kutub

'book'

sariir

surur

'throne'

safiinat

sufun

'ship'

Clearly we can identify the [u] plural melody of this pattern with that of

Arabic Broken Plurals [CVC (V) C ] nouns in (26b) as well as that of the color and defect adjectives But the prosodic template of the plural is rather problematic, since no independently motivated rule will account for the shortening of the vowel in the second syllable. We might formulate such a rule nevertheless, or we might suppose that the feminine suffix has been truncated in these forms, but neither alternative is particularly compelling. It may be that this plural type, like those below in 4.4, represents a case in which a plural template simply replaces the singular one. 4.3 Feminine [CVCC] Nouns of this type have two basic modes of plural formation, largely under lexical government. One, a complication of the feminine sound pliiral, is applicable regardless of the stem vocalism. The other, a broken plural form is confined chiefly to those nouns with a high stem vowel in the singular. The special property of [CVCC] feminine nouns is that their plurals retain the vocalic melody of the singular, unlike any other pattern discussed thus far. The sound plural suffix aat, when applied to this class, induces insertion of a vowel into the stem-final cluster: a. ?ard jafnat b. kisrat sidrat c. gurfat fiulmat

?aradaat

'earth*

jafanaat

•dish'

kisiraat/kisaraat

' fragment'

sidiraat/sidaraat

'lotus tree'

gurufat/gurafat

'upper chamber

6ulumaat/6ulamaat

'darkness'

As is apparent from (46), this inserted vowel can be either a copy of the stem vowel or a. It is unclear whether this variation was free or dialectal in Classical Arabic (in the contemporary language a has prevailed). The insertion itself is also variable, at least with the forms in (46b) and (46c) Clearly a rule like (47) is needed to account for these facts: (47)

0

->-

V

/ [ CVC

C] /

Feminine Plural

Rule (47) is systematically suppressed with II-weak roots (jawzat, jawzaat 'nut') and biliteral roots, which have gemination of the second radical isaddat, saddaat 'charge').** These factors should ultimately be incorpo-

316

John J. McCarthy

rated into the environment of the rule. There are at this point two options. In one case the inserted vowel is bound directly to a, yielding forms like kisaraat. In the other case, where the inserted element is unspecified on the melodic tier, the stem vocalism spreads by the operation of the association conventions to this new V position. This is illustrated in (48): (48)

i

A [CVCVC ]

(+aat)

\ I/ ksr It is obviously unnecessary to stipulate that the inserted vowel copies the stem vocalism, since that property can be derived from the prosodic model. The other type of plural formation, which predominates with [CVCC] feminines with characteristic vowel u and i, involves insertion of a into the stem-final cluster, but without the sound plural suffix: a.

b.

rukbat

rukab

'knee'

?ummat

'umam

'nation'

kuurat

kuwar

'district'

qit&at

qita9

'piece1

limmat

limam

'lock of hair'

siirat

siyar

'mode of walking'

These forms are subject to insertion of a V bound to a in the plural, and nothing else. This rule is apparently distinct from (47), since it is not variable, it lacks the possibility of inserting an unspecified V, and it can (unlike (47)) apply to II-weak and biliteral roots. 4.4 Other Plural Types There is quite a large number of broken plural patterns that have not been treated in this study, perhaps as many as fifty. The great majority of these are of extremely low frequency, occurring with no more than twenty different words, and then usually as one of several options. There are, however, some patterns of somewhat higher frequency that merit at least brief notice. A few are given in (50):

Arabic Broken Pluarals a.

b.

c.

317

fals

'aflus

'copper coin

wajh

'awjuh

• face'

9anaaq

?a5nuq

'female kid'

yamiin

'aymun

'right hand'

waral

wirlaan

'lizard'

jura6

jirfiaan

'field-rat'

gulaam

gilmaan

•boy'

9abd

9ubdaan

'slave'

balad

buldaan

'town'

xaliil

xullaan

•friend*

These data are not unanalyzable. First, it is reasonable to suppose that the plural pattern in (50a) has the underlying template [ CVCVC] , subject to metathesis, with a melody [a u]. Second, the grammar should also note that the patterns in (50b,c) differ only in the stem melody or the plural, and in fact [uj and [i] are used almost interchangeably here. What is absent, however, is a significant connection between the form of the singular and the form of the plural. In words like these, singular and plural apparently share only the same root, with neither the vowel melody nor the prosodic template carried over in any form to the plural. This feature of retaining only the root in a morphological relationship is characteristic of Arabic derivational processes (McCarthy 1979a, 1981). It is met with in broken plural formation only rarely, as in these minor patterns. 5. CONCLUSION This analysis has provided an account of all major properties of Classical Arabic broken plural formation in terms of the prosodic theory of morphology. The major features of the analysis include a unified formal statement of the several broken plural patterns described in section 2 and a recognition in the grammar of the close connection between broken plural and diminutive formation. Other aspects of importance are the use of melodic levels to characterize regularities in weak root allomorphy and distribution of broken plural vocalism. Finally, various minor plural patterns involve reasonable extensions of the formal apparatus needed for more widespread plural types.

318

John J. McCarthy

NOTES I am indebted to Nick Clements, Morris Halle, and Alan Prince for their assistance at various stages in the development of this work. A preliminary version of some of this material appears in McCarthy (1979a). The transcription system used here has its usual values except for the following. 5 and h are the voiced and voiceless pharyngeal glides, respectively, i is the voiced alveopalatal affricate and 6 is the voiced interdental spirant. A subscripted dot in t , d, s, and 6 indicates pharyngealization, often known as emphasis. Long vowels are analyzed as bimoraic. 1. Strictly speaking, the units a and ktb in representations like (1) are not segments, but archisegments undefined for the features [segmental] and [syllabic]. For convenience, however, these archisegmental feature bundles will be notated as if they were fully defined segments. 2. Sound plurals are formed by suffixing nominative uuna or genitive/accusative Una in the masculine and aat plus case desinence in the feminine. (These suffixes are analyzed prosodically in McCarthy (1979a).) 3. The position I take here may appear to be inconsistent — how, one may ask, can a rule be lexical or diacritically governed yet also productive and applying to a vast majority of the possible cases? This alleged inconsistency follows only from entirely unwarranted and unsupported assumptions about what characterizes a class of rules called minor. Clearly productivity is orthogonal to lexical government, if only because productivity is scalar (Aronoff 1976) and lexical government is binary — yes or no. And statistical preponderance, although a useful tool, cannot be an absolute measure, as it is too sensitive to accidental frequency effects (as in the strong verbs in English). 4. Sporadic forms are inconsistent with this latter property of the plural template, like muftir, mafaatiir 'one who breaks the fast', '¿Ssaar, ? a5aasir 'dust-storm'. 5. Root-initial w is subject to a special rule of dissimilation to 9 in the broken plural pattern exemplified in (15): waasilat, ^auaasil 'joining'. " ' 6. Possibly as an accident of such a small category, the application of Minor Association (21a) is restricted to [CWCWC] singular nouns with [i a] vocalism. A variant of w-insertion attaches y to the free position in a few nouns of this type: diibaax, dayaabiix 'brocade' (also dabaabiix). 7. Roots whose first or second radical is y do not occur with the [i a] plural melody, presumably to avoid yi and iy sequences. 8. For reasons that I do not fully understand, doubly-weak roots — those II—U and III-w or y — have identical broken plural forms for [CWCVC] and [CVCWC ] singulars: zaauiyat, zaaaayaa 'corner'; hawiiyat, hawaayaa 'intestine'. A straightforward analysis along the lines in (36) accounts for the latter but not for the former without some new stipulation. 9. nafs is a feminine noun which exceptionally lacks the at suffix of

Arabic Broken Plurals

319

the feminine. This exceptionality is lost under derivation, as in the. diminutive. 10. A few [CWCVC ] active participles form plurals with the pattern CaCaC+at, a minor type. Ill-weak roots inexplicably fail to form the broken plural in (40), instead showing up as, for example, qudaat from qaadii 'judge. This is apparently identical to the other human plural pattern below in (33a), but with the feminine suffix at rather than aa 11. The biliteral analysis of these so-called geminate roots is developed and supported in McCarthy (1981b). REFERENCES Brame, M. (1970) Arabia Phonology: Implications for Phonological Theory and Historiaal Semitic, Doctoral dissertation, MIT, Cambridge, Massachusetts . Chomsky, N. (1951) The Morphophonemics of Modern Hebrew, Master's thesis, University of Pennsylvania, Philadelphia. (Published by Garland Press, N.Y.). Clements, G.N. and K. Ford (1979) "Kikuyu Tone Shift and its Synchronic Consequences," Linguistic Inquiry 10, 179-210. Goldsmith, J. (1976) Autosegmental Phonology, Doctoral dissertation, MIT, Cambridge, Massachusetts. (Published by Garland Press, N.Y.). Halle, M. and J.R. Vergnaud (1978) "Metrical Structures in Phonology," unpublished paper, MIT, Cambridge, Massachusetts. Harris, Z.(1941) "Linguistic Structure of Hebrew," Journal of the American Oriental Society 62, 143-167. Hayes, B. (1980) A Metrical Theory of Stress Rules, Doctoral dissertation, MIT, Cambridge, Massachusetts. Kahn, D. (1976) Syllable-based Generalizations in English Phonology, Doctoral dissertation, MIT, Cambridge, Massachusetts. Kiparsky, P. (1979) "Metrical Structure Assignment is Cyclic," Linguistic Inquiry 10, 421-441. Levy, M. (1971) The Plural of the Noun in Modern Standard Arabia, Doctoral dissertation, University of Michigan, Ann Arbor, Michigan. McCarthy, J. (1979a) Formal Problems in Semitic Phonology and Morphology, Doctoral dissertation, MIT, Cambridge, Massachusetts. (1979b) "On Stress and Syllabification," Linguistic Inquiry 10, 443-465. (1981a) "The Representation of Consonant Length in Hebrew," Inquiry 12, 322-327.

Linguistic

320

John J. McCarthy (1981b) "A Prosodic Theory of Nonconcatenative Morphology," Linguistic Inquiry 12, 373-418.

McCawley, J. (1970) "A Note on Tone in Tiv Conjugation," Studies in African Linquistias 1, 123-129. Murtonen, A. (1964) Broken Plurals: Origin and Development of the System, E.J. Brill, Leiden. Palmer, F. (1962) The Morphology of the Tigre Noun, Cambridge University Press, Cambridge. Selkirk, E. (forthcoming) Phonology and Syntax: The Relation between Sound and Structure, MIT Press, Cambridge, Massachusetts.

Chapter 19

Anaphora, Cataphora, and Topic Focusing: Functions of the Object Prefix in Swahili Keith Allan

As in most Bantu languages, a Swahili verb may take a cliticized object prefix. The usual constituent order in Swahili (as relevant to the discussion which follows) is given in (1): (1)

SNP

SP-T-OP-V-sfxs

ONP

(XNP)

SNP = subject NP, SP = subject prefix, T = tense/aspect, OP = object prefix, V = Verb radical, sfxs = extensional* and modal suffixes, ONP = object NP, XNP = object chômeur. E.g. (2)

Juma

a-li-mw-ib-ish-a

m-toto

pesa

SNP

SP-T-OP-V-sfxs

ONP

XNP

Juma [el] SPl-Past-OPl-steal-Caus-Aff

Cl-child

money [c9]

'Juma made the child steal money' The primary function of the OP is pronominal. This is clear when the verb to which it is clitized does not predicate an ONP coreferential with it. In such circumstances it will either be anaphoric to a coreferential full NP somewhere in the preceding discourse, or cataphoric to one in some following clause, or alternatively it will have a referent identifiable in the nonverbal situation. Compare the sequences in (3) and (4). (3)

Jana

mw-alimu

a-li-nunua

ki-tabu.

Yesterday Cl-teacher SPl-Past-buy C7-book.

A-me-ki-soma. SPl-has-0P7-read.

'Yesterday the teacher bought a book. He has read it (4)

already '.

Jana mwalimu alinunua kitabu. Ame(ki)soma kitabu.

The repetition of "kitabu" in (4) violates the Gricean maxim of quantity (Grice 1975) if the two instances of the NP are intended to be coreferential; just as does the English version in (5): (5)

Yesterday the teacher bought a book. He has read the book.

And if (4) were intended to carry the message that the teacher bought one book and read another, then it violates the Gricean maxim of manner (or

324

Keith Allan

worse), since such a message should be rendered something like (6). (6)

Jana mwalimu alinunua kitabu, lakini a-li-(ki-)soma Ki-ngine. Yesterday teacher bought book, but he-Past-(0P7-) read C7-another. 'Yesterday the teacher bought one book, and read another'.

Ex. (3) exemplifies the contextually introduced referent for the OP as anaphoric pronoun. There are also the situationally dependent ones such as those emphasized in (7) . (7)

a. Nitafcuona kesho.

'I'll see you tomorrow'

b. Ali alinirudishia vitabu. 'Ali returned the books for me'. Speaker and addressee are necessary participants in the speech situation, and as a rule need no further identification. When the OP appears without a coreferential ONP predicated by the same clause predicate, it functions pronominally. But presumably when the OP co-occurs with the ONP it must have some other function. One possibility is that the OP, like the SP, is simply a mark of agreement between the grammatical relation and the verb. But whereas every SNP has a concordial SP to mark SNP-Verb agreement, not every ONP has a corresponding OP. And so we conclude that the OP is not the mechanical indicator of Object-Verb agreement. It is often said that the Swahili OP specifies the ONP as definite (cf. Driever 1976, 23; Hinnebusch 1979, 218; Wilson 1970, 186) without any account being given of what is meant by the notion of definiteness. I will assume that a NP is definite when the speaker takes its designatum to be given, i.e. to need no further identification than is presented within the NP in question. The reasons for this may be either that the designatum has been introduced already into the discourse; or that the designatum is present in the situation of utterance; or that its identity is fully predictable within the world being spoken of. Although definiticity is obligatorily indicated in English NP, it is only a contingent category in the NP of Swahili and other Bantu languages; that is to say, contingent on there being for instance a demonstrative or a numerative determining the NP. Otherwise, definiticity is a grammatical irrelevance. It is hardly surprising, then, to find textbooks warning the learner of a Bantu language not to expect an OP with every ONP that would be definite in English. For example, none of the underlined ONP on (8) have a cataphoric OP although the English equivalents are necessarily definite.

The Object Prefix in Swahili (8)

325

a. Huwezi kununua motokaa hi~L bila fedha nyingi. 'You can't buy this car without a lot of money' b. Asma alishonesha nguo zake Nairobi. 1

Asma had her dresses sewn in Nairobi'

c. Tulipokwenda Dar tulitembelea ohuo kikuu. I the

'When in Dar we visited i 2f university' [there is only the one] I % I d. Tunaweza kuona bahari kutoka kwenye hoteli yetu. , the | 'We can see j j sea from our hotel' e. Tupeleke ghavama yote, au nusu? Afadhali tupeleke yote. the 'Shall we send *a

whole amount, or half

of it ?

We'd better send [it] all' In none of the sentences of (8) is the OP necessary in Swahili, although the ONP are definite, and have to be marked definite in English. It cannot therefore be the function of the Swahili OP to indicate definiteness in the ONP. But note that this is not to deny the possibility that the OP itself is definite; and that is a point to which I will return later. There are many instances of the Swahili OP being used with ONP whose English translation is indefinite. For some of these, it would seem that the ONP is referential, and it is sometimes said that the OP marks referentiality. For example it is apparently possible to differentiate the sentences in (9) from those in (10) on the basis of the referentiality of their ONP: (9)

a. Nilidhani sitampata mtu wa kuniongoza. 'I thought I wouldn't get someone to direct me' b. Ilikuwa tabia ya mwalimu kumruhusu mtoto kufunga mlango. 'It was the habit of the teacher to allow a (certain) child to close the door'

(10)

a. Nilidhani sitapata mtu wa kuniongoza. 'I thought I wouldn't get anyone to direct me' b. Ilikuwa tabia ya mwalimu kuruhusu mtoto kufunga mlango. 'It was the habit of the teacher to allow some child (or other) to close the door'

326

Keith Allan

The presentation of this data is, however, misleading: the translation given (10a) could equally well be given (9a), and vice versa. And what we have here can be given a somewhat different interpretation as I shall show later. Givon (1976, 159) quite justifiably described the ONP in (11) as nonreferential although it is customarily accompanied by an OP. (11)

Simuoni mtu ye yote.

(12)

?Sioni mtu ye yote.

'I don't see anyone'

So the evidence that the OP marks a concomitant ONP referential is at best equivocal; and I don't believe this is the function of the Swahili OP, any more than it is its function to mark ONP definiteness. Ashton (1947, 44) says "Direction of emphasis dictates the position of the noun object, and the use or omission of the object prefix"; and again "Note the use of the Object Prefix, especially when attention is directed to the person or thing in question" (op. cit., 303). Or Polom£ (1967, 160) "in the non-relative forms the infix slot is only filled when the attention is obviously focussed on the object". These observations seem to me to point to the most satisfactory analysis of the OP: that the OP co-occurs with the ONP when the latter - or rather its designatum - is the topic of discourse, and the OP focuses on this topic status. Topics are defined as being in some sense the center of attention, and they are supposedly definite and referential (cf. Keenan 1976, 297; Li & Thompson 1976, 464) : these are characteristics traditionally associated with the Bantu OP, although we have already noted that concomitant ONP are not necessarily either definite or referential. However, both definiteness and referentiality appear to be characteristics of the OP itself, and this is a point I will return to later. According to Givon (1976, 157) the topicalized Object is obligatorily accompanied by the OP; but in fact there are exceptions like (13): (13)

a. Hata risasi nimenunua. Even bullets I have bought. b. Je, maharagwe mabichi, hutaki? So, fresh beans you don't want? c. Shati hili umepasua! This shirt, you've torn [it]!

Nonetheless most left shifted ONP are accompanied by OP. E.g.

The Object Prefix in Swahili (14)

a. Chombo amefcipanza mwamba. The vessel, he's run it on a rock [he's ruined the scheme^]. b. Suruali hizi, umeziharibu. These trousers, you've ruined them. c. Hizo nyingine tutazifuata baadaye. Those others, we111 come for them later. 4

I am told

that all the exx. in (14) sound better with the OP than without

it. Focus on the ONP as a result of right dislocation is also instantiated, and in such examples the OP seems to be obligatory if meaning is to be held constant. Cf. (15)

a. Msichana alijaza ndoo maji. girl caused-full bucket water 'The girl filled the bucket with water' b. Msichana aliijaza maji ndoo. girl caused-it-full water bucket 'The girl filled it with water, the bucket' c. *Msichana alijaza maji ndoo. girl caused-full water bucket

(16)

a. Msichana alikama ng'ombe maziwa. girl squeezed [-from] cow milk 'The girl milked a/the cow' b. Msichana alimkama maziwa ng'ombe. girl squeezed [-from] -it milk cow 'The girl milked it, the cow' c. 'Msichana alikama maziwa ng'ombe. girl squeezed [^-fromj milk cow

(17)

a. Mama alitona binti zake wanja. Mum put [-on| girls her eyeshadow 'Mum put eyeshadow on her girls' b. Mama aliwatona wanja binti zake. mum put [-on] -them eyeshadow girls her 'Mum put eyeshadow on them, her girls' c. *Mama alitona wanja binti zake. mum put [on] eyeshadow girls her

327

328

Keith Allan

Right dislocation of the ONP is a means of focusing on it as an afterthought topic, and perhaps what Chafe (1976, 53) calls an 'antitopic'. My use of the term topic for a noninitial phrase may be misleading; but dislocation of a NP to either right or left has much the same discourse function, i.e. to center the attention on the phrase - or more correctly, on its designatum. Since dislocation of the Swahili ONP is normally, if not necessarily, accompanied by the use of an OP, we may conclude that the OP is normally required where the ONP is topic. The topicalization of ONP coreferents in cleft sentences like (18) generally involves the use of OPs, although they are not obligatory. (18)

a. Hi-vyo Dem-C8

ndi-vyo vl-atu Cop-C8

ni-vi-taka-vyo.

C8-shoe I-OP8-want-Rel8

'Those are the shoes that I want' b. Ha-pa Dem-C16

ndi-po

mahali

tu-pa-penda-po.

Cop-C16

place [cl6]

we-OP16-like-Rell6

'Here is the place we like' The OP in (18) is directly controlled by the Relative which holds the Object relation within the relative clause; in turn the Relative is controlled by the coreferential NP "viatu" in (a) and "mahali" in (b), and these NP - or rather their designata - are topicalized. So, once again, the OP is found to be concomitant with an ONP whose designatum is the discourse topic. In a sentence like (19) there is no structural evidence of ONP topicalization such as dislocation or clefting; nevertheless, the OP cross references a topic. (19)

Juma

amejiumiza

kwa

panga

a-li-lo-li-nunua

jana.

Juma hurt-himself with panga [c5] he-Past-Rel5-OP5-buy yesterday 'Juma hurt himself with the panga he bought yesterday' The center of attention here - and thus the topic - is what Juma hurt himself with: namely, the panga he bought yesterday. The NP "panga" is coreferential with the Relative infix "-lo-" which holds the Object relation in the relative clause; and which in turn is cross referenced by the OP "-li-". Thus the OP occurs concomitantly with an Object that is topic. But the only structural mark of the topic status of this Object is the OP itself. This shows up if we make the person hurt become the topic of discourse, cf.

329

The Object Prefix in Swahili (20)

Juma alijiumiza mwenyewe kwa panga alilonunua jana. 'It was himself that Juma hurt with the panga he bought yesterday'

In this circumstance there is no OP cross referencing with panga. The OP in (19) has been described as a 'resumptive pronoun1 by e.g. Morolong & Hyman (1977, 205). In IE languages resumptive pronouns turn up with dislocations: in other words with coreferential topicalized NP, e.g. I don't like him, that waiter; Cet homme, ¿e (ne) I'aime pas. Hetzron (1971), following Getatchew (1971), has exemplified a resumptive pronoun in Amharic which focuses on the ONP as topic: *ganzab ssatatallah™ (21)

betu-n Almaz ba-matragyaw katarracc-9W,

n9$uh yahonall * ysssabbarall * I will give her money

the-house-Acc. Almaz with-the-broom if-she-cleaned-it,

it will be clean *it [broom] will be broken

The resumptive pronoun "-aw" focuses on "betun" as topic, and so only certain discourse sequences are possible; in (21) one possible sequence is exemplified along with two that Hetzron claims are impossible. So resumptive pronouns occur together with a coreferential topic. This would appear a proper description of the Swahili OP in (15-19). The evidence, therefore, points clearly to the OP in Swahili being an indicator of topic focus on the ONP; or, shall we say, being concomitant with an ONP whose designatum holds the center of attention. Otherwise, in the absence of an ONP, the OP functions as a regular pronoun. What additional evidence can be adduced for the hypothesis that the Swahili OP is a pronoun signalling the ONP as focused topic? Consider four short texts that seem to confirm the accuracy of this analysis. The first I composed myself; the second is quoted from a letter dated 1911 and published in Beech (1918); the third and fourth are taken from recordings of spontaneous conversations made in Dar es Salaam, and published in Stevick, Mlela & Njenga (1963). (22)

Tuliwtafuta mshale kwa masaa kadhaa ndipo Njeri akapaaza kwa sauti "Mshale, nauona!" 'We had searched for the arrow for several hours when Njeri shouted "The arrow, I see it!"'

In (22) the topic of discourse is clearly the arrow, and this is focused

330

Keith Allan

on through cross referencing by the OP "-U-". (23)

Wabaadu nakuarifu nyumba yangu ni rahani katika 'Court' nami nataka kuiuza hiyo nyumba ao kuiweka rahani nipate fedha kulipa deni niwiwayo katika 'Court' - ... (Quoted in Beech 1918, 18) ' X beg to inform you that my house is held as security by the Court and I want either to sell the house, or use it as security to borrow money so that I can pay the debt I owe the Court'.

The topic in (23) is the writer's house and it is therefore cross referenced by the OP "-i-" in the second clause. Notice the OP does not appear in the first clause which introduces the house as a topic. The use of the OP in the first clause of (22) suggests that the arrow had been introduced as a topic earlier in the discourse. (24)

Y: Ningeshukuru sana kama ungefika nyumbani kwangu. [1] Hivyo tafadhali fika.[2] 'I'd be most grateful if you'd drop by the house. So please do come'. Z: Hilo ni jambo zuri sana Bw. Yusufu.[3]Lakini, kama unavyojua sisi watu wa mjini hatukuwaii kuhama.[4] Hivi wakati huu uko mtaa gani?[5] 'That's very kind Y. But as you know we city dwellers change our residence frequently. So what street are you living in right now?' Y: Mimi sasa ninakaa Msimbazi Street[6]Nyumba yangu nimejenga mwenyewe. [7] Nyumba hiyo ni nzuri na ina orofa moja.[8] 'Right now I'm living on M. Street. I built my own house. That house is very nice and it has an upstairs'. Z: Lo! Bw. Yusufu umekuwa tajiri sana.[9] Mara umekwisha jenga nyumba![10] 'Hey, Y; you've been doing alright for yourself. You've already built a house!' Y: Ehe, kwa sababu ya kulima![ll] 'Yep, thanks to farming' Z: Aaa, kumbe sisi ambao tunahamishwa kutoka hapa kwenda hapa tunapata hasara tu.[12] Nitafurahi sana kuiona nyumba yako. [13] 'Ah, those of us who are transferred hither and thither just lose

The Object Prefix in Swahili

331

out. I'll be very glad to see your house'. Y: Vyema! Na nitakueleza mipango mingi. 'Fine! I'll fill you in on all the details'. (Stevick, Mlela & Njenga 1963, 519-20) Throughout this passage the topic is very clearly Y's house, but not until [13]do we have an OP when the house is designatum of the ONP. In [ 1] the house is first introduced into the discussion and so no OP cross referencing is to be expected. In any case "nyumbani kwangu" is locative and not open to OP cross referencing. Furthermore, the topic of this sentence is rather a visit to the house than the house itself. The next direct reference to the house is in [7] . Here "nyumba yangu" is the ONP and moreover it is topicalized; but there is no OP (we do not have nimeijenga). There are a couple of possible explanations: one is that the topic is being reintroduced at this point, and not of itself focused on; the other is that Nyumba yangu, nimeijenga lmenyewe sounds boastful rather like the English /Is for my house, I built it myself. In [10] "nyumba" is (indefinite) generic and does not designate Y's house in particular, so the ONP is not the discourse topic. But in [13] we do have focus on the topic in the ONP "nyumba yako"; and it is here that we find the OP used. So the text of (24) bears out the claim that the OP is used concomitantly with an ONP when there is focus on the ONP designatum as topic. Consider a final text. (25)

B: Ningependa gauni mbili kwa bibi yangu. [1] Je, fundi wako anaweza kushona vizuri itakiwavyo?[2] 'I'd like 2 dresses for my wife. Say, your tailor, he sews as well as one would want?' S: Oh, kabisa! [3] Pia kuna mishono mingi. [4] Kuna mshono wa kibwebwe na mingine mingi ya kila aina. [5] Mshono gani mkeo anapenda?[6] 1

Of course! And there are a lot of styles [to choose from].

There's the kibwebwe style, and many others of every kind. What style does your wife like?' B: Naona kama anapenda mchinjo. [7] Waelewa mchinjo ni nini, bwana? [8] 'I think she likes the mchinjo style. Do you know what the mchinjo is like, bwana?' S: Naelewa sana, maana mahitaji ya watu wengi nayapata hapa.[9] 'I know very well, because I cater for the needs of many people

332

Keith Allan here1 . B: Aha! Basi nataka mchinjo.[10] Mpime bwana.[ll] 'OK. Then I want the mchinjo. Will you measure her, bwana?' S: Nampima kifua thelathini na sita.[12] Pia nampima kiuno ishirini na sita.[13] Pia nampima mabega kumi na nane.[l4] Urefu wataka, tuseme, kama arobaini na moja?[l5] 'Bust 36; waist 26; shoulders 18. The length you want? Shall we say about 41?' B: Tuseme mpaka hapa.[16] Inatosha.[17] Je, lini ataaifuata nguo hizi?[18 ] 'Let's say to here. This is alright. And when shall she come for these clothes?' (Stevick, Mlela & Njenga 1963, 524-5)

No detailed exposition of (25) is necessary. The discourse topic is the clothes for B's wife, and in [18] this topic is brought into focus through the OP "-zi-" accompanying the ONP "nguo hizi". These texts verify that where the topic is an ONP and focused on, the OP is used. I conclude that where the OP is concomitant with a coreferential ONP it is a cataphoric resumptive pronoun causing focus to fall on the ONP designatum where the latter is discourse topic. This description of OP function accounts for the presence of the OP with dislocated ONP and cleft ONP. It categorizes the OP as a pronoun whether or not there is a concomitant ONP. And it will account for the fact that OP are definite and referential^ even though the coreferential ONP may not be. However, this is not the whole story. The conclusions just drawn about the function of the Swahili OP apply only to those of a class numbered higher than 2. Class 1/2 OP (i.e. the -m(w)-/-wa-classes) obligatorily accompany appropriate ONP unless the focus of information falls on the predicate, such that the combination of Verb and ONP has a semantic unity reminiscent of phrasal verbs. In evidence of the customary use of class 1/2 OP compare (26) with (27): (26)

a. Simwoni mtu ye yote.

'I don't see anyone'

b. ?Sioni mtu ye yote. (27)

a. Sikioni kitu cho chote. 'I don't see anything' b. Sioni kitu cho chote.

Now compare the following pairs of sentences in which the (a) example

The Object Prefix in Swahili

333

contains a class 1/2 OP and ONP, while the (b) example contains only a class 1/2 ONP: (28)

(29)

(30)

a. Mke wangu alimpata mtoto jana.

'My wife had the baby yesterday'

b. Mke wangu alipata mtoto.

'My wife gave birth'

a. Nilimuona mganga.

1

b. Niliona mganga.

'I sought medical advice'

1 saw the doctor'

a. Nilidhani sitampata mtu wa kuniongoza. 'I thought I wouldn't find someone to direct me' b. Nilidhani sitapata mtu wa kuniongoza. 'I thought no-one would be found to direct me1

The (a) sentences of (26-30) involve significant reference to the individual holding the Object relation - which is doubtless the effect of the resumptive pronoun, i.e. the OP. The (b) sentences refer to an event that comprehends an ONP having no significant separate status. Thus the event of giving birth is what is significant in alipata mtoto, and not the child; in ali-mpata mtoto, however, there is as much focus on the child as on the event of its birth.^ The semantic unity of the Verb-Object combination in the (b) sentences of (26-30) is reminiscent of phrasal verbs, and of that which holds within such expressions as kuwa na njaa 'be hungry' or kuona kiu 'be thirsty', and which cannot have an OP coreferencing "njaa" or "kiu". The peculiar characteristics of class 1/2 OP in Swahili put them at a midpoint in a personal pronoun hierarchy (cf. Hawkinson & Hyman 1974) . The hierarchy is laid out in (31). (31)

Obligatory OP

1st & 2nd

OP customary unless ONP

OP contingent on ONP

part of phrasal verb type

referent being focused

expression

as discourse topic

class 1/2 (animate)

classes*2 (inanimate)

m(w)/wa

u/i,li/ya,etc.

persons ni/tu/ku/wa

We see from (31) that conditions on the use of OP get more stringent as we move down the personal hierarchy; which is exactly what any universalist would have predicted. Let me conclude by saying that although I have restricted discussion in

334

Keith Allan

this paper to Kiswahili, the characteristics of the Swahili OP seem to 7 be very much the same as those of OPs in other Bantu languages too. NOTES 1. The extensions are: Applicative (marking dative or benefactive ONP); Reciprocal, Reversive; Causative; Stative; Passive. 2. The asterisk here indicates inappropriacy under existing pragmatic conditions rather than an ungrammatical construction. 3. Cf. Ashton (1947, 45). 4. By Gido Mapunda, a native of Mbinga in south western Tanzania. His mother tongue is Kimatengo. I would like to thank him for his help with all the Kiswahili examples. 5. I am uncertain whether pronouns are necessarily referential. In (i) the English pronoun "it" proes the nonreferential "a car", and whether it is itself referential remains equivocal. (i) John says he want to buy a car and it must have a t.v. installed to entertain back seat passengers. 6. Ashton (1947, 44-5) remarked of the sequence Q: Umeleta ohakula? A: Nimeleta, Bwana. 'Have you brought the food? 'I have brought [it], Bw.' "The important element here is the action not the object". She contrasted this with Q: Chakula umekileta? A: Nimekileta, Bwana 'The food, have you brought it?' 'I have brought it, Bw. ' Thus use of the anaphoric pronoun is guided by the same principle. 7. It was only after presenting this paper at the conference that I became aware of Benji Wald's 'The development of the Swahili object marker: a study of the interaction of syntax and discourse' (1979). His and my conclusions about the functions of the OP agree in all essentials. REFERENCES Ashton, Edith 0. 1947. Swahili Grammar. London Longmans. (2nd Edn.) Beech, Mervyn W.H. 1918. Aids to the study of Ki-Swahili. London: Kegan Paul, Trench s Trubner. Chafe, Wallace L. 1976. Givenness, contrastiveness, definiteness, subjects, topics, and point of view. Subject and Topic, pp. 25-55. ed. by Charles N. Li. New York: Academic Press. Driever, Dorothea. 1976. Aspects of a Case Grammar of Mombasa Swahili Hamburg: Helmut Buske Verlag. Getatchew, Haile. 1971. The suffix pronouns in Amharic. Papers in African Linguistics, pp. 101-12, ed. by Chin-Wu Kim & Herbert Stahlke. Edmonton-Champaign: Linguistic Research.

The Object Prefix in Swahili Givön, Talmy. 1976. Topic, pronoun and grammatical agreement. Subject and Topic, pp. 149-188, ed. by Charles N. Li. New York: Academic Press. Grice, H. Paul. 1975. Logic and conversation. Syntax and Semantics 3: Speech Acts, pp. 41-58, ed. by Peter Cole & Jerry L. Morgan. New York: Academic Press. Hawkinson, Annie K. & Larry M. Hyman. 1974. Hierarchies of natural topic in Shona. Studies in African linguistics 5, pp. 147-170. Hetzron, Robert. 1971. Presentative function and presentative movement. Studies in African linguistics, Supplement 2, pp. 79-105. Hinnebusch, Thomas J. 1979. Swahili. Languages and Their Status, pp. 209-293, ed. by Timothy Shopen. Cambridge, MA: Winthrop. Keenan, Edward L. 1976. Remarkable subjects in Malagasy. Subject and Topic, pp. 247-301, ed. by Charles N. Li. New York: Academic Press. Li, Charles N. & Sandra A. Thompson. 1976. Subject.and Topic: a new typology of language. Subject and Topic, pp. 457-489, ed. by Charles N. Li. New York: Academic Press. Morolong, Malillo & Larry M. Hyman. 1977. Animacy, objects and clitics in Sesotho. Studies in African Linguistics 8, pp. 199-218. PolomS, Edgar c. 1967. Swahili Language Handbook. Washington D.C.: Center for Applied Linguistics. Stevick, E.W., J.G. Mlela & F.N. Njenga. 1963. Swahili Basic Course Washington D.C.: Department of State FSI. Wald, Benji. 1979. The development of the Swahili object marker: a study of the interaction of syntax and discourse. Syntax and Semantics 12: Discourse and Syntax, pp.. 505-524, ed. by Talmy Grivon. New York: Academic Press. Wilson, P.M. 1970. Simplified Swahili. Nairobi-Dar es Salaam-Kampala: East African Literature Bureau.

335

Chapter

20

The Complex Structure Conspiracy and the Grammar of Mandingo Mallafé Dramé

1. INTRODUCTION The present study is an attempt to provide a coherent explanation to the behavior of complement clauses in Mandingo, a Mande language spoken mainly in Sénégal, Gambia and Guinée-Bissau.^ In particular, we want to further investigate a hypothesis, formulated in an earlier version of this 2

paper , namely the Complex Structure Conspiracy (CSC), which was developed as follows: Based on distributional facts, it is claimed that Mandingo complement clauses are not instances of complex NPs. By convention they shall be referred to as Complex Structures (CS). The Complex Structure Conspiracy then simply says that no Complex Structure can hold a grammatical relation in Mandingo. In this paper, we will attempt to extend the CSC analysis to relative clause formation. The predictions made by the Complex Structure Conspiracy and the consequences that fall out of it will be discussed. However, before we can start the discussion on CSC, a number of preliminary remarks must be made. Subsequently, the following subdivision is proposed: Section (1.0) will present a brief introduction on simple sentence structure, focusing more specifically on word order, tense/aspect marking and adverbial constructions. Section (2.0) will be devoted to complementation proper. It will be shown that Mandingo has basically two types of complementizers, clause-initial (CIC) and non-clause-initial (NIC), and that the choice of these complementizers seems to depend on various semantic and syntactic factors. An attempt will be made to determine the deep structure underlying these complement clause and to account for the expletive pronoun a 'it', which surfaces in the direct object position of some main verbs. We will also discuss alternative approaches on how to derive Mandingo complementizers. The Complex Structure Conspiracy will be formulated and tested in Section (3.0). The final section (4.0) will sum up and conclude the paper. 1.0. Simple Sentence Structure One salient feature of simple sentences that interacts crucially with

Mallafé

338

Dramé

338

Mandingo complementation is word order. In light of this, it is necessary to say a few words about the word order of Mandingo simple sentences before undertaking the analysis of complement clauses. 1.1. Word ovder: The basic word order in a Mandingo sentence is SOV, and no variation is permitted, as can be seen in (1): 1.

a. Moussa ye dlndirjo

je^

Moussa TA the child

see

(Moussa saw the child) b. *Moussa ye 0 Subj.

je

TA

dindigo

V

DO

c. 'Moussa dindiQo ye 0 je Subj . D O

TA

V

d. *Dindigo (,) Moussa ye 0 je DO

Subj. TA

V

The Mandingo direct object is not only preverbal, but it must occur between the tense-aspect marker (TA) and the verb, as in (la). Any violation of this word order results in an ungrammatical sentence, as shown in (la through d). This word order remains unchanged under questioning and negation, and regardless of the tense of the verb (cf Dramé, 1979). When the sentence contains an inherently double-object verb, the direct object remains preverbal, and the indirect object occurs after the verb, followed by a postposition. Sentences (2b-e) are ungrammatical because they violate this basic constituent ordering: 2.

a. Keó

kitaaboo

dii

the man TA the book

ye

give

karandirìlàa

la

the teacher

to

(The man gave the book to the teacher) b. *Keó

ye

Subj. TA c. *Kéo

ye

Subj. TA d. *Ke6

ye

Subj. TA e. *Keó

ye

Subj. TA

karandirilaa IO karandirilaa

DO

PtP

V

10

V

DO

DO

karandirilaa

dii

kitaàboo

kitaàboo

PtP

DO kitaaboo

dii

(la)

10 kitaaboo

(la)

V (la)

dii

PtP

V

karandirilaa 10

dii

0 -0

The Complex Structure Conspiracy Not only is the postposition la inomissible, but the direct and the indirect objects cannot switch places or be moved elsewhere, as illustrated in (2a-e). Given such a restricted word order, one expects to find the so-called direct object complement clause in direct object position, that is between the TA and the verb. But, as we shall see, that is not where it occurs. 1.2. Tense-aspeet marking (TA): Another constituent of the sentence that needs to be briefly discussed is the tense-aspect marker (TA), which, unlike in many African languages (e.g. Bantu languages) is an independent morpheme in Mandingo. With the exception of the future tense and the present habitual, which have a double tense marking (the second TA functioning as a verbal suffix), and the past tense of intransitive verbs (which is postverbal and is realized as ta), TA is preverbal and pre-DO, as was shown in (1-2) . (3) and (4) are illustrations of the distribution of the future and past tense-markers: 3.

a. Babacar Babacar

be

bun-daa

yele-la

TA^ the door

open TA^

(Babacar will open the door) b. *Babacar

bun-daa

Subj.

DO la

bun-daa

Subj.

TA 2

DO

yele-la V

V be

Subj.

TA

tiinaT) tâ

TA 2

yele-be

Babacar

DO a. Kambaanôô

0

TAj

c. *Babacar

d. *Bun-daa (,)

4.

be

TA^ 0

yele-la V

kuljkoo

TA to

the boy spend the day TA the farm at (The boy spent.the day at the farm). b. * Kambaanôô Subj. c. *Kambaan

(na-mir), tuma-mig, da-miQ)

S I Z|

The variable Z stands for other sentence-final materials such as adverbial expressions, which are usually ordered after NICs. If this solution is adopted, then the conclusion would be that Mandingo has two types of complement clauses, and this based on the surface distribution of their complementizers. However, as we have shown, the two complement types have striking similarities that one would like to capture not only in principle but in the formulation of the rules that generate them. Such an approach would lead to generating both complementizer types in the same position, say in initial position by S and moving NICs in sentence-final position by a later rule. Of course the NIC movement rule will have to be motivated, but one could argue that such an analysis serves to preserve the basic word order of Mandingo in that NICs are semantically understood as adverbs, and adverbs in this language occur sentence-finally. Thus NICs would be marked with the feature t+ Advl and CICs with the feature [- Adv]. In fact one structure that should be included in this proposed generalization is the structure formed by the relative clause together with its head NP. As is shown in (19) , this structure cannot function as subject or direct object: 19. a. Moussa ye

kitaaboo-lu

Moussa TA the book

pi

min-nu

sa^, Samba ye i

wh

buy

pi

S.

cika

TA them pick

(Lit: The books that Moussa bought, Samba picked them).

The Complex Structure Conspiracy b. *Samba ye [Moussa ye kitaaboo-lu min-nu sal]] cika Subj.

DO

c. Moussa ye kitaaboo-lu Moussa TA the book pi

V

min-nu

sag, i

Wh

buy they be the table on

pi

be taabuloo kar)

(Lit: The books that Moussa bought, they are on the table) d. [*Moussa ye kitaaboo-lu min-nu sag ] be taabuloo Subj.

V

kar)

PP

The ungrammaticality of (19b & d) shows that the traditionally labelled NP formed by the relative clause and the head NP cannot function as subject or as direct object. It has to be put in a focus position while a pronoun agreing in number with the head holds the subject or direct object position. The claim that is been made in this paper is that whatever principle is involved in preventing complement clauses from surfacing in subject or direct object position is also involved in preventing the NP formed by the relative clause and its head NP from surfacing in subject and object positions. Consequently, our grammar of Mandingo complementation must be able to capture this generalization on embedded clauses. This will constitute the main focus of the next section. 3. SEMANTIC CHARACTERIZATION Leaving open the problem of determining the right framework for the derivation of Mandingo complementizers, let us now try to formulate a tentative definition of what we shall call a Complex Structure: 20.

A Complex Structure (CS) is any structure S NP, or any structure dominated by S.

The notion of Complex Structure must be understood as defined on deep structures since the S NP structure represents the left-branched deep structure of relative clauses. If the definition of S were extended to include relative clauses, say by adopting a WH-movement framework, then the notion of Complex Structure would be simpler since it would simply mean any structure dominated by S. Complex Structures must also be understood as different from Complex NPs, since the Complex Structure Conspiracy, that will be formug lated shortly, does not cover complex NPs.

Nominalized sentences are also

not defined as Complex Structures. As a consequence one claim that we are making is that there is no direct transformational device relating embedded sentences to their supposedly nominalized counterparts. Given this definition of Complex Structure, the Complex Structure Conspiracy can be

352

Mallafê Drainé

stated as follows: 21.

Complex Structure Conspiracy (CSC): No complex structure can bear a grammatical relation in Mandingo.

(21) states that, whenever a complex structure occurs in Mandingo, it will be in a chômeur position, e.g. no complex structure can appear anywhere on the Accessibility Hierarchy, as defined by Keenan and Comrie (1977). We have already shown that no complex structure can function either as subject or direct object. In the next subsection, we will attempt to show that complex structures cannot bear any of the remaining grammatical relations defined in the Accessibility Hierarchy. 3.1. That no complex structure can be an indirect object is attested by the ungrammaticality of (22a, c & e): 22. a.'Moussa ye kunfââ yitandi [musô eo

tli+o dookuu+o • saatée+o kordââ+o -

tio dookuo -> saatSo —* kordaa

'the 'the 'the 'the

feather' work, the job' village' house'

356

Mallafê Drame

8. Complex NPs formed by a noun and accompanying epithets do bear all grammatical relations that simples nouns bears. DSnnoo ye kidoo soso : Dânnâ bulu-dil-maa ye kidoo soso:

'The hunter charged the gun' 'The skilled hunter charged the gun' FJ jaga-yaa ta wo musu nimmaa ti : I am taller than that beautiful woman' 9. There are three nominal possession constructions, ordinarily referred to as alienable possession and inalienable possession (I) and (II). Their structure is as follows: AP : Possessor la possessee : Moussa la dondikôo : 'Moussa's shirt' IP1 : Possessor possessee : Moussa sïQo = : 'Moussa's leg' IP2 : Possessor possessee-maa: Moussa musu-mââ : 'Moussa's wife' Their occurrence is dictated by a network of semantic relations that must hold between the possessor and the possessee. 10. A simple case of comparative sentence would be the following one: Tootôô suti-yââ ta a faa-maa ti Tootôô short TA he father be (Tootôô is shorter than his father) REFERENCES Bokamba, Eyamba G. and Dramé, Mallafé. 1978. Where do relative clauses come from in Mandlngo? In CLS 14. 1978 PP 28-43. Bresnan, Joan W. 1972. Theory of Complementation in English Syntax. MIT Ph.D. dissertation. 1972. Creissels, Denis 1979. Notes de recherches sur un parler Handing de la Gambie. Université des langues et lettres de Grenoble, Section de linguistique générale et didactique générale des langues. Dramé, Mallafé. 1979. Aspects of Mandingo Complementation. In Studies in the Linguistic Sciences. Volume 9, Number 2, Fall 1979. Green, Georgia M. and Morgan, Jerry L. 1972. A guide to the study of Syntax. Karttunen, L.. 1974. Presupposition and Linguistic context. In Theoretical Linguistics 1. PP 181-194. Keenan, E.L. and Comrie, B. 1977. Noun Phrase Accessibility and universal grammar. Linguistic Inquiry, 8,1: PP 63-99. Kiparsky, P. and Kiparsky, C. 1971. Fact. In D.D. Steinberg and L.A. Jacobovits, eds.. Semantics. Cambridge University Press. 1971. Levi, Judith N. 1978. The Syntax and Semantics of Complex Nominals. Academic Press. 1978. Long, R.W. 1971. A comparative study of Northern Mande languages. Indiana University Ph. D. dissertation. Published by University Microfilms Ltd. Ann Arbor, Michigan.

The Comples Structure

357

Conspiracy

Rosenbaum, Peter s. 1967. The grammar of English Predicate Complement constructions.

Cambridge Mass., MIT Press.

Ross, John R. 1967. Constraints on Variables in Syntax, unpublished MIT dissertation. Rowlands, E.G. 1959. A Grammar of Gambian Mandinka.

S.O.A.S. London

Touré, Amadou. 1975. La grammaire des phrases bambara à complément prédicatif. In Annales de l'université d'Abidjan, Serie H., t. X. 1977.

List of Contributors

Keith Allan, Department of Linguistics, Monash University, Clayton, Victoria. 3168 Australia Airen Amayo, English Department, University of Ibadan, Ibadan, Nigeria M. Lionel Bender, Department of Anthropology, Southern Illinois University, Carbondale, 111. 62901. U.S.A. Patrick R . Bennett, Department of African Languages and Linguistics, University of Wisconsin, Madison, Wise. 53706, U.S.A. Eyamba G. Bokamba, Department of Linguistics, University of Illinois, Urbana, Illinois 61801, U.S.A. Ellen Broselow, Department of Linguistics, University of Iowa, Iowa City, Iowa 5224-2, U.S.A. Hounkpati C. Capo, Department of Linguistics and African Languages, University of Benin, Benin City, Nigeria Mary M. Clark, English Department, University of New Hampshire, Durham, New Hampshire 03824, U.S.A. George N. Clements, Department of Linguistics, Harvard University, Cambridge, Mass. 02138, U.S.A. Mallafe Drame, Department of Linguistics, University of Illinois, Urbana, Illinois 61801, U.S.A. David Dwyer, African Studies Center, Michigan State University, East Lansing, Michigan 48823, U.S.A. Ben Elugbe, Department of Linguistics and Nigerian Languages, University of Ibadan, Ibadan, Nigeria James F. Fordyce, Department of Linguistics, U.C.L.A., Los Angeles, California 90024, U.S.A. John Goldsmith, Department of Linguistics, Indiana University, Bloomington, Indiana 47401, U.S.A. Joseph H. Greenberg, Department of Anthropology, Stanford University, Stanford, California 94305. U.S.A. Ian F. Hancock, Department of English, University of Texas, Austin, Texas 78712, U. Charles E. W. Jenewari, School of Humanities, University of Port Harcourt, Port Harcourt, Nigeria

360 William R . Leben, Department of Linguistics, Stanford University, Stanford, California 9^305. U.S.A. John J. McCarthy, Department of Linguistics, University of Texas, Austin, Texas ?8?12 U S J. W. Snyman, Department of African Languages, University of South Africa, Pretoria, South Africa John M. Stewart, Afrikaanse Taalkunde, University of Leiden, The Netherlands