Zambezia (1997), XXIV (i).WHEN TWO VOWELS GO WALKING: VOWELCOALESCENCE IN SHONACAROLYN HARFORDDepartment of Linguistics, University of ZimbabweAbstractVowel coalescence is a phonological process in which adjacent vowels causeeach other to change. Processes of vowel coalescence that are widespread inother Bantu languages occur in Shona at the boundaries between words andmonosyllabic affixes. This article argues for two main coalescence triggers inShona: 1) epenthetic i-, used to eliminate monosyllabic words, and 2) anominal initial vowel attested, in other Bantu languages, which is hypothesizedto have once existed in Shona. Coalescence only takes place across asyntactic boundary that occurs within a phonological word.VOWEL COALESCENCE IN BANTUVowel coalescence is a phonological phenomenon in which two adjacentvowels cause each other to change (and sometimes shorten)1. It is acommon phenomenon crosslinguistically and in Bantu languages such asShona. Earlier work on Shona linguistics, such as Fortune (1967; 1985) andDale (1972), point out the existence of vowel coalescence in Shona, butrarely attempt to analyze it (cf. footnote 4). The present article has threegoals: 1) to identify the sources of the vowels that trigger vowel coalescencein Shona, 2) to identify where vowel coalescence does and does not occurin Shona, and 3) to indicate how the different patterns of vowel coalescencefound in Shona dialects are related to the basic pattern presented inconnection with (1) and (2). This article is therefore intended as acontribution to the understanding of Shona phonology and morphology.Vowels that coalesce may come to be adjacent to each other becausethe morphemes they belong to have been joined or because they occur atthe end and the beginning of adjacent words. Examples from Shona appearin (1):1. a. sef!'how?' (like what?)sa- i => se - ilike what1 am grateful to Mr. M. K. Mkanganwi, Mr. Francis Matambirofa, Mr. Munashe Furusa, Dr.Ann Jefferies, Dr. (Catherine Demuth, Dr. Jerome Hachipola and participants in a seminar inthe Department of African Languages and Literature at the University of Zimbabwe, July,1996, for assistance with this article. I am also indebted to Professor M. F. C. Bourdillon forcomments and editorial assistance. All mistakes are my own.6970 VOWEL COALESCENCE IN SHONAb. neP.'by/and/with what?'na-i => ne - iby/and/with whatc. U- ri ku-chemere-i2s be to cry for what'Why are you crying? (You are crying for what?)'The examples in (1) belong to a pattern which is described as occurringfrequently in Bantu languages (Doke, 1943; Meinhof, 1910; Wald, 1973) andwhich is summarised in (2):2. a. a + a => ab. a + i => ec. a + u => oThe examples in (1) show the pattern in (2b). Most of the data dealtwith in this article shows the (2b) type of coalescence, but there are (2a)and (2c) examples as well.To begin with, consider the examples in (3) (the coalescing vowel is inbold):3. a. U-no-bv- e - pi3 pres come from FV where'Where does it come from?'b. ku- pedz - e- zvoinf finish FV those (8)'finishing those' (Zvarevashe, Kurauone, 35)c. V- a- mir- e- po2 RP wait FV there'They waited there.' (Mutswairo, Hamandishe, 28).The examples in (3) involve coalescence at the juncture between thefinal vowel of the verb and the morpheme which immediately follows it.This juncture is indicated by a hyphen in the examples. In addition to thepost-final vowel morphemes in (3), other morphemes which are associatedwith coalescence include locatives -ko and -mo, and possibly -zve. Notealso from the examples in (4) that coalescence in this context is notalways obligatory:4. a. Nd-a- ka-p- a- woIs NP RP give FV also'I gave also.'C. HARFORD 71b. A-chi- bv- a- ko1 cont leave FV there'She leaving there'(Mungoshi, Makunun'unu Maodzamwoyo, 36).I'm not sure exactly where and why coalescence is optional, but itmay be optional only with certain morphemes, such as those in (4).Now, consider a second set of examples in (5)2:5. a. no - mukadzi 'with a woman'ne- mukadzi 'with a woman' (Fortune, 1985, iv)(Class 1)b. na - babd 'with father' (Ibid, 9) (Class la)c. na - vakddzi 'with women' (Class 2)d. na - vddzimdi vdngu 'with my wife' (Ibid., 18) (Class 2a)e. na - amdi 'with mother' (Class 2b)f. no - mukaka 'with milk' (Fortune, 1967, 25) (Class 3)ne - mukaka 'with milk'g. ne - miti 'with trees' (Class 4)h. ne - sddza 'with sadza' (Fortune, 1985, 9) (Class 5)i. na - mdpadzd 'with hoes' (Ibid., 11) (Class 6)ne - mdpadzd 'with hoes'j. ne - chigaro 'with a chair' (Fortune, 1967, 25) (Class 7)k. ne - zvtgaro 'with chairs' (Class 8)1. ne - nzira iyi 'by this path' (Fortune, 1985, iii) (Class 9)m. ne - dzipwere 'with various children' (Ibid., 66)(Class 10)n. no - ruoko 'with the hand' (Class 11)ne - ruoko 'with the hand'o. ne - kadukuduku kese 'even the smallest thing'(Ibid, 114)na - kdshoma kdse (Ibid., 115) (Class 12)p. no - twand 'with small children' (Class 13)ne - twand 'with small children'q. no - upfu 'with flour' (Class 14)ne - upfu 'with flour'r. no - kuti 'because' (Ibid., ii) (Class 15)ne - kuti 'because'2 Classes 19 and 20 are in limited distribution dialectally (Fortune, 1985, 82) and are notincluded in this list.72 VOWEL COALESCENCE IN SHONAs. na - paduku pose 'even in the slightest degree'(Ibid, 114) (Class 16)ne Ł paduku pose 'even in the slightest degree't. no - kumusha 'and at home' (Class 17)u. no - mumusha 'and in the home' (Class 18)In these examples, coalescence occurs at the juncture between thepreposition na- and a following nominal. Other morphemes which areassociated with coalescence in this context include the associative particle-a-3, sa- 'like', ha- 'here', the copulative, the locative noun prefixes sometimesand possibly others. Note in particular the alternations found in certaincontexts, such as ne-mukadzi vs. no-mukadzi in Class 1. I will return tothese alternations later.COALESCENCE TRIGGERSConsider now the basis for dividing the data into the two sets representedin (3) and (5). These sets are defined according to coalescence triggers. By'trigger', I mean the vowel whose presence causes coalescence. When weconsider the patterns shown in (2), it might seem difficult to say that onevowel causes the coalescence but not the other. However, this is preciselywhat I am going to say about Shona. When we consider all the placeswhere vowels come together in Shona, there are two cases where aparticular vowel may be considered to be a trigger of coalescence. Thefirst case is the epenthetic /- which appears in examples such as those in (6):6. a. i-dya 'Eat!'EV eatb. i - tsva 'new'i - go 'wasp' (not in Karanga)EV waspd. i - we 'you'EV youe. i - zvo 'these' (8)EV theseThe second case is a morpheme which does not currently exist inShona but which was, I propose, once present in the language: the initial3 I am grateful to Dr. Ann Jefferies for pointing out to me that possessives such as wedu 'our',wenyu 'your', vedu, venyu etc., consist of the associative marker followed by a pronoun (wa- idu, wa - inyu, for example). They conform to the prediction made later in this article thatcoalescence takes place only across a syntactic boundary.C. HARFORD 73vowel which is part of noun class prefixes currently extant in Bantulanguages such as Ndebele and Luganda, as in the examples in (7)4:7. a. u- mu- ntu 'person'IV 1 person(Ndebele: Pelling and Pelling, 1974, 8)b. o - mu - ntu 'person'IV 1 person(Luganda: Ashton, et al, 1954, 20)The basis for the division of data in (3) and (5) is, therefore, that thecoalescences in (3) are triggered by epenthetic i- and the coalescences in(5) are triggered by the extinct initial vowel. Each of these identificationsis justified in turn.Epenthetic i-The obvious question that arises with the claim that the coalescencetrigger in (3) is epenthetic /- is how we can assume that the two are thesame, given that the environments in which the two appear are apparentlyso different. It is usually assumed that epenthetic i- is used to prevent amonosyllabic morpheme from becoming a monosyllabic word, as seen inthe examples in (6) above.5 However, all of the examples in (3) arepolysyllabic, so the presence of epenthetic i- would appear to beunmotivated.The assumption that the function of epenthetic i- is to eliminatemonosyllables is certainly correct, but it is not the whole story. Epenthetic/- turns up in all kinds of places where its presence is not required toeliminate a monosyllable, illustrated in the examples in (8):8. a. / - tyd - i - she (name)EV fear EV Godb. / - da- i - she (name)EV loveEV Godc. i- mbo-tarisa6EV Aux look'Just have a look now.' (Dale, 1972, 92)4 Fortune (1967, 25) discusses a 'latent' vowel as being responsible for vowel coalescence,but stops short of identifying it with the Bantu initial vowel.5 Certain ideophones are monosyllabic, one of a number of ways in which this word class isphonologically exceptional (Fortune, 1962, 29). Many Bantu languages have a similarconstraint against monosyllabic words. For a discussion of this constraint in Swahili, cf.Park (1995).6 Compare: Nyatso - nditarisa 'Have a good look at me' (Ibid.'). Here the auxiliary verb isdisyllabic and epenthetic i- is not inserted. Also, Dale notes -do- as an exception to thegeneralisation that monosyllabic auxiliary verbs ('infix verbs' in Dale's terminology) inimperatives take epenthetic ('prothetic' in Dale's terminology) vowels (p. 92).74 VOWEL COALESCENCE IN SHONAd. vana-i- dyd -ndigere2a EV eat I have stayed'rich people with income' (Fortune, 1985, 106)e. va-tsvd-i- tsva2 new EV new'very new ones' (Fortune, 1985, 110)Note now the elements that are adjacent to the coalescing final vowelin the examples in (3). They are all monosyllables and, therefore, cannotoccur as separate words. We have already seen that the insertion ofepenthetic i- is a strategy for avoiding monosyllabic words. However,there is also a second strategy: cliticisation to a preceding or followingword. (Cliticisation is a process where a syntactically independent elementbecomes phonologically dependent on a neighbouring word.) I proposethat epenthetic i- and cliticisation are used separately or together,depending on the morpheme involved, but also, probably, on factors suchas dialect and idiolect. We see cliticisation to the preceding word in thecase of particles like ko, and cliticisation to the following word in the caseof the preposition na. (It is not clear to me why one strategy is used incertain cases but not others.) In the case of particles like ko, cliticisation isthe required strategy, but the option of epenthetic i- is also available7.This option gives rise to the variation we see in (3) and (4). It is possible toapply cliticisation alone, as in the examples in (4), or cliticisation andepenthesis, as in the examples in (3).The interaction of these two strategies may be modelled appropriatelyby any phonological theory which employs rule ordering, such as thetheory of Lexical Phonology (Kiparsky, 1982; Mohanan, 1986). Themorphemes pi, i, ko and wo are phonologically independent prior to theapplication of the strategies which enforce the prosodic constraint againstmonosyllabic words. The variation we observe is produced by the twostrategies applying in opposite orders, as illustrated in the examples in(9):9. a. Epenthetic /- insertion followed by cliticisation:u-no-bv-a pi (example (3a)Epenthetic i- insertion =>u-no-bv-a ipiCliticisation =>u-no-bv-a4pi (coalescing vowels are in bold)The strategies for eliminating monosyllables are not sensitive to syntactic category. Theyare sensitive only to position (pre-noun phrase or post-final vowel). Note, however, that -wo'also' appears never to take epenthetic i-.C. HARFORD 75Vowel coalescence (and vowel shortening)u-no-bv-e-pib. Cliticisation followed by epenthetic i- insertion:a<hi-bv-a ko (example (4b)Cliticisation =>a<hi-bv-a-ko(Epenthetic /- insertion does not apply, since ko isno longer a monosyllable. Likewise, vowel coalescencedoes not apply since there is no [i] to coalesce with the [a].)Note that although cliticisation is a process which eliminatesmonosyllables, it is not constrained to apply only to monosyllables. Hence,it may cliticise a formerly monosyllabic morpheme to which epenthetic i-insertion has already applied, as in (9a). Epenthetic /- insertion, on theother hand, only applies to monosyllables, as shown by its failure to applyin (9b). Therefore, cliticisation may bleed epenthetic i- insertion, but notvice versa. This analysis shows how epenthetic i- can produce the vowelcoalescence seen in the examples in (3).Initial vowelTurning now to the initial vowel, there are two pieces of evidence for theidea that Shona once had a now-extinct initial vowel that is responsible forthe vowel coalescence seen in the examples in (5).8 First, the vowels thatwould have to be present at some point to produce the coalescences in (5)according to the patterns in (2) are copies of the vowels which appear inthe noun prefixes themselves, a pattern attested in other Bantu languageswith extant initial vowels. Second, there are certain syntactic contextswhere these coalescences don't take place which are the same contexts inwhich initial vowels are absent in languages which have them. Each ofthese arguments is discussed in turn.Initial vowels as copies of noun prefix vowelsThe goal in the following discussion is to develop a plausible scenario inwhich Shona had initial vowels at a previous stage, based on patterns inextant Bantu languages with initial vowels. This scenario, in order to bereasonable, does not have to conform in all its details to the particulars ofany extant language and there are probably extant languages which do notconform to the pattern posited for Shona, given the large amount ofvariation that exists (Hyman and Katamba, 1993; de Blois, 1970).The argument in this article is not an attempt to enter the debate about whether Proto-Bantu had initial vowels. I assume that initial vowels belong to a recent phase in the historyof Shona.76 VOWEL COALESCENCE IN SHONANevertheless, as long as there is a possible scenario which could haveproduced the patterns seen in present-day Shona, the argument for initialvowels at an earlier stage of the language is strengthened.Shona noun class prefixes contain the following vowels:10. a. [a]: Classes 2, 6, 13, 16b. [i]: Classes 4, 5, 7, 8, 9, 10c. [u]: Classes 1, 3, 11, 12, 14, 15, 17, 18The class 5 noun prefix cannot be analysed as containing a vowel(Lafon, 1994), but other Class 5 agreement markers in Shona contain [i].Also, Class 5 noun prefixes in other Bantu languages such as (again)Ndebele and Luganda contain [i]. So we will put Class 5 in Shona intocategory (10b). Pursuing the same reasoning, we will also put Class 9 into(10b), since its agreement markers contain [i].If the now-extinct initial vowel in Shona was, for each noun class, acopy of the vowel in the extant noun prefix, the prefixation of thepreposition na- would produce the following coalescences according tothe pattern in (2) (coalescing vowels are in bold):11. a. na + a-noun prefix => na-noun prefix (10a)IVb. na + i-noun prefix- => ne-noun prefix (10b)IVc. na + u-noun prefix- => no-noun prefix (10c)IVExceptions to this pattern in the data in (5), such as those in Classes 1,11, 12, 13, 14 and 15, will be dealt with below.The idea that Shona had a now-extinct initial vowel which was a copyof the noun prefix vowel becomes more plausible when we consider thatthe same pattern occurs in Bantu languages which currently have initialvowels, such as Ndebele. Initial vowels in Ndebele noun prefixes can beseen to be copies of the vowels of the noun prefixes which have vowels, asillustrated below:12. a. u-mu- (Classes 1,3)b. a-ba- (Class 2)c. i-mi- (Class 4)d. /-//-(Class 5)e. a-ma- (Class 6)f. i-si- (Class 7)g. i-zi- (Class 10)C. HARFORD 77h. u-lu- (Class 11)i. u-bu- (Class 14)j. u-ku- (Class 15)The same is true in Luganda, as reported in Hyman and Katamba(1993). Also de Blois (1970), as cited in Hyman and Katamba (1993), hasreconstructed the Bantu initial vowels with copied vowels.Now, the major objection to this hypothesis is that these are notalways the patterns attested in the language. There is a certain amount ofdialectal and idiolectal variation in the coalesced vowel away from thepattern predicted above. In particular, recall ne-mukadzi and nomukadziin Class 1. Only no-mukadzi is predicted by the analysis just given.This sort of variation is due to the diachronic nature of the trigger.Originally, coalescence was phonologically motivated by the presence ofthe initial vowel. Later, the initial vowel disappeared, leaving the grinwithout the cat. Once the original motivation disappeared, the formscould be shifted into environments for which they were not originallymotivated. A certain amount of shifting has taken place and probablycontinues to take place. One shift that has probably taken place is one infavour of the use of [e] to the exclusion of [a] and [o], at least in thepreposition na-. What this means is that we can expect to see [e] wherethe pattern would predict [a] or [o], but not vice versa: we would notexpect to see [a] or [o] where the pattern would predict [e]. Here lies thesignificance of the observation that variations like no-mukadzi are onlyused by a minority of speakers. Given the shifting that has taken place, thecurrent status of such forms within the mosaic of dialects and idiolectsdoesn't matter; they still provide evidence about the original state of thesystem.Syntactic contexts for initial vowelsSo far, we have examined how coalescence works in examples in which apreposition is cliticised to a following noun. In particular, we have examinedthe preposition na-. Note now that na- also functions as a kind of quasi-verb, in which case it takes subject noun class markers, as in the followingexample:13. Ndi-ne mart. 'I have money.'Is have moneyNote that the vowel of the preposition coalesces to [e] when followedby the Class 9 noun, as expected. This coalescence is obligatory:14. *Ndi-na mart.Is have money78 VOWEL COALESCENCE IN SHONANote now that when the form is negative, coalescence is forbidden:915. a. Ha-ndind marl 'I don't have money.'neg Is have moneyb. *Ha- ndi-ne mart.neg Is have moneyWhat is the explanation for this pattern? Consider the followingcrosslinguistic evidence. In languages like Luganda, initial vowels neverappear on the objects of negative verbs. The following examples are fromLuganda:16. a. Te - baawa bdana bitabo.(Luganda: Hyman and Katamba, 1993, 224)Neg they gave children books'They didn't give the children books.'b. *a - bdana e - bitaboIV IVc. *a - bdana bitaboIVd. * bdana e - bitaboIVNote that neither of the objects of the negated double-object verb inthis example may take an initial vowel.Suppose now that Shona at an earlier stage had an initial vowel whichwas subject to the same constraint. There would then have been no initialvowel to trigger coalescence in this syntactic environment. We would thenexpect to find that coalescence does not take place in this environment inthe present stage of the language.Another example comes from object relative clauses (in which thehead noun is the object of the relative clause), exemplified as follows:17. a. nhivi idzi mbiri dza - wdrevasides these two which you have mentioned(Fortune, 1967, 140).'these two sides which you have mentioned'b. ne - chinhu cha - ndinoddwith thing which I want (Ibid., 143)'the thing which I want'9 I am informed by M. K. Mkanganwi that (15b) is grammatical in his dialect. This observationis in line with my earlier hypothesis that [e] forms are replacing [a] and [o] forms.C. HARFORD 79According to Fortune (Ibid., 140), the object relative marker is theassociative particle (which he refers to as the possessive concord)mentioned in connection with (5) above as one of the morphemes whichcoalesces when prefixed to a noun. Examples are as follows:18.a. chigarochairb. mharadzidestroyerche -ofyo-ofhuniwood (Ibid., 133)munhua person (Ibid., 133)However, judging from Fortune's examples, among others, it appearsthat the associative particle never shows coalescence when it is used asan object relative marker. Why not?We can again draw on a comparison with Luganda to explain thispattern. According to Hyman and Katamba (1993, 212), subject relativeclauses in Luganda may take initial vowels:19. a. o - IwaagwaIV one that fell'the one that fell' (Ibid)b. Iwaagwaone that fell'the one that fell' (Ibid.)Object relative clauses, on the other hand, do not (Ibid., 213). Notethat in example (20b), the object relative marker may not take an initialvowel:20. a. e- bikopd bye bddlabdIV cups that they saw'the cups that they saw'b. * e - bikopd e-bye bddlabdIV cups IV that they sawWe can follow the same line of reasoning used above for the negativeforms of na-. If Shona once had an initial vowel which patterned the sameway in relatives as the initial vowel currently does in Luganda, then wewould not expect to find coalescence in this environment.Why epenthetic i- and the initial vowel?If we accept the preceding arguments, we see that epenthetic /- and theinitial vowel are almost entirely responsible for vowel coalescence inShona. However, there are also cases where [a] does not coalesce with anadjacent [i] or [u]. These are listed in (21)-(29) (relevant vowels are inbold):80 VOWEL COALESCENCE IN SHONA21. Tense/aspect/modality morphemes:a. Nd-ai- gara (habitual -ai-~)Is hab stay'I stayed'b. Nd-a- ibd (near past -a- + verb stem)Is NP steal'I stole'c. Nd-a- kd - ibd (remote past -ka- + verb stem)Is NP RP steal'I stole'd. Ndi<hd Ł ibd (future -cha- + verb stem)Is fut steal'I will steal'e. Nd-a- i- tengd (near past -a- + object marker)Is NP OM buy'I bought it'22. Negative morphemes:a. Mu- sd -ib - e2p neg steal FV'Don't steal'23. Subject marker + verb stem:a. A-ib-eSM steal FV'(that) s/he steal'24. Plural-J10:a. Pindd-i!enter FV PL'Enter!'25. Deideophonic morphemes:a. -tsvdir - asweep FV26. Noun class prefixes + stems:a. ma-uchi6 honey'inferior honey, lots of honey' (Fortune, 1985, 55)10 Note that the form pindaiwo indicates that -i does not occupy the slot following the finalvowel.C. HARFORD 81b. ma-usiku6 night'long tedious nights' (Ibid., 55)c. ma-uta6 bow'bows' (Ibid., 75)27. Phonotactically:a. a - iwa 'no'Note also that final vowel -a does not coalesce when immediatelypreceding a noun:28. Final vowel -a + NPU:a. T-a-fir- a nyika.lp NP die for FV country'We died for the country.'The same is true of the ownership phrases described in Fortune(1985, 157-158):29. Ownership phrasesa. sa<higaroOP chair'chairman' (Fortune, 1985, 157)b. nya-ku-bd zvavdmweOP to steal others' things'one who steals others' things (Ibid.)c. mu-zvina-guhwa1 OP talebearing'talebearer' (Ibid., 158)How can we account for the differences here? When we look at all theplaces in Shona where vowels come together, it turns out that coalescenceonly takes place across a syntactic boundary that occurs within aphonological word. The examples in (3) and (5) are all examples of thistype: in (3), coalescence takes place across the boundary separating thefinal vowel of the verb from the morpheme that occupies the immediatelyfollowing slot. In (5), coalescence takes place across the boundaryseparating prepositions from the following nouns. In contrast, none of the1' Compare the name Tafire - nyika, where the verb + noun form a single word and coalescencetakes nlare.82 VOWEL COALESCENCE IN SHONArelevant vowel junctures in (21)-(29) involves the same kind of syntacticboundary within a word. It is this factor which isolates epenthetic /- andthe initial vowel as coalescence triggers.There is an apparent exception to the generalisation that vowelcoalescence takes place only across a syntactic boundary. In the plurals ofthe nouns ziso 'eye' and zino 'tooth', vowel coalescence takes place at theboundary between a noun prefix and a noun stem, which is not a syntacticboundary12 (the coalescing vowels are in bold):30. a. ziso 'eye'meso 'eyes' (ma - iso; there is also the regular maziso)zino 'tooth'meno 'teeth' (ma - ino, there is also the regular mazino)Note that this boundary is the same as for the examples in (26a-c),where vowel coalescence is correctly predicted not to occur. Do theseplurals negate the generalisation that coalescence takes place only atsyntactic boundaries?If we say that these plurals mean that vowel coalescence operatesacross syntactic and non-syntactic boundaries alike, then we are left withno generalisation concerning the domain of vowel coalescence. We haveno way of distinguishing the environments in which vowel coalescencedoes occur from those in which it doesn't. However, it is counterintuitiveto say that no such regularity exists in the data presented in this article.On the other hand, if we classify meso and meno as genuine exceptions,we can account for their presence in the current Shona lexicon. I proposethat these nouns are holdovers from an earlier stage in the history ofShona at which vowel coalescence could take place across the boundarybetween noun prefix and noun stem. It is an attested phenomenon inlanguage change for relics of an earlier system to persist into a later stagein which the rules that created them no longer operate. In this respect,meso and meno are like modern English irregular plurals like mice andoxen, whose regularities are apparent only when we consider the historyof the language and whose presence in the modern lexicon does notobscure the existence of the regular -s plural.In support of this proposal, note that vowel coalescence appears incognates of the same pair of plurals in other Bantu languages. In ZambianTonga, both plurals are irregular, as in Shona (example 31a). In Swahili,'teeth' is irregular (example 31b), as is 'eyes' in Isizulu (example 31c) (thecoalesced vowel is in bold):12 Myers (1987) proposes that the boundary between noun prefix and noun stem in Shona is,in fact, a syntactic boundary. I accept the arguments made in Bresnan and Mchombo (1995)against this analysis and similar analyses of other Bantu languages in Baker (1988a; b);Carstens (1991) and Kinyalolo (1991).C.HARFORD 8331. a. meso 'eyes' (ma Ł isd)menyo 'teeth' (ma Ł inyo)(Tonga: Collins, 1962, 14)b. meno 'teeth' (ma Ł ind)(Swahili: Ashton, 1944, 64)c. amehlo 'eyes' (ama Ł ihld)(Isizulu: Taljaard and Bosch, 1988, 18)It is reasonable to suppose that the examples in (31), as well as theShona plurals, derive from a common historical source, rather thanresembling each other accidentally. If so, then the rule permitttingcoalescence in this environment was productive in this historical source,prior to the development of the separate daughter languages. This inferencelends credence to the idea that the Shona plurals are not the product ofthe synchronic grammar but are inherited from an earlier stage in thedevelopment of Bantu languages.CONCLUSIONIn conclusion, this article has argued for two main sources for Shonavowel coalescence. The first is epenthetic /-, a strategy to eliminatemonosyllabic words which also triggers coalescence. The second is ahypothesised initial vowel which is no longer extant in Shona but whosepresence is still attested in the form of coalescence.ABBREVIATIONS USED IN THIS ARTICLEAux:cont:EV:fut:FV:inf:IV:neg:NP:lp:2s:2p:auxiliary verbcontinuousepenthetic vowelfuturefinal vowelinfinitiveinitial vowelnegativenear past1st person plural2nd person singular2nd person pluralOM:OP:PL:prep:QP:RP:SM:TAM:Is:object markerownership phrasepluralprepositionquestion particleremote pastsubject markertense/aspect/modality1st person singularOtherwise, numbers in the glosses represent the Bleek-Meinhofnumbers of noun prefixes and agreement morphemes.Acute accent indicates high tone; no accent indicates low tone. TheLuganda examples from Hyman and Katamba (1993) contain grave accents84 VOWEL COALESCENCE IN SHONAindicating low tones. Examples from Fortune (1985) are marked withFortune's tone marks. The remaining tones were kindly provided by FrancisMatambirofa. The only unmarked examples are the Luganda example in(7b) and the Tonga and Isizulu examples in (31). Swahili is not a tonelanguage.REFERENCESASHTON, E. 0. (1944) Swahili Grammar (London, Longman Group Ltd.).ASHTON', E. 0., E. M. K. MULIRA; E. G. M. NDAWULA AND A. N. TUCKER (1954), ALuganda Grammar (London, Longmans, Green and Co.).BAKER, M. (1988a) Incorporation: A Theory of Grammatical Function Changing(Chicago, University of Chicago Press).BAKER, M. (1988b) 'Theta theory and the syntax of applicatives in Chichewa',Natural Language and Linguistic Theory, VI, 353-89.de Blois, K. F. (1970) 'The augment in the Bantu languages', AfricanaLinguistica, IV, 85-165.BRESNAN, J. AND S. MCHOMBO, (1995) 'The lexical integrity principle: Evidencefrom Bantu', Natural Language and Linguistic Theory, XIII, 181-254.CARSTENS, V. M. (1991) 'The Morphology and Syntax of Determiner Phrasesin Kiswahili' (Los Angeles, Ph.D dissertation, University of California).COLLINS, B. S. J. (1962) Tonga Grammar (Lusaka, National EducationalCompany of Zambia Ltd.).DALE, D. S. J. (1972) Shona Companion (Gweru, Mambo Press).DOKE, C. M. (1943) Outline Grammar of Bantu (Grahamstown,Communication #12, Department of African Languages, RhodesUniversity).FORTUNE, G. (1962) Ideophones in Shona (London, Oxford University Press).FORTUNE, G. (1967) Elements of Shona (Harare, 2nd edition, Longman).FORTUNE, G. (1985) Shona Grammatical Constructions, Vol. I (Harare, MercuryPress).HYMAN, L. AND F. KATAMBA (1993) 'The augment in Luganda: Syntax orpragmatics?' in S. Mchombo (ed.) Theoretical Aspects of BantuGrammar (Stanford CA, Center for the Study of Language andInformation, Stanford University), 209-56.KINYALOLO, K. (1991) 'Syntactic Dependencies and the SPEC-head AgreementHypothesis in Kilega' (Los Angeles, Ph.D dissertation, University ofCalifornia).KIPARSKY, P. (1982) 'Lexical morphology and phonology', in I. S. Yang (ed.)Linguistics in the Morning Calm (Seoul, Hanshin).LAFON, M. (1994) 'Shona class 5 revisited: A case against *ri- as class 5nominal prefix', Zambezia, XXI (i), 51-80.MEINHOF, C. (1910) GrundrilS einer Lautlehre der Bantusprachen (Berlin,Verlag von Dietrich Reimer).MOHANAN, K. P. (1986) The Theory of Lexical Phonology (Dordrecht, DReidel Publishing Co.).C. HARFORD 85MUNGOSHI, C. L. (1970) Makunun'unu Maodzamwoyo (Harare, The CollegePress).MUTSWAIRO, S. (1991) Hamandishe (Gweru, Mambo Press).MYERS, S. (1987) 'Tone and the Structure of Words in Shona' (Amherst,Ph.D dissertation, The University of Massachusetts).PARK, J-I, (1995) 'Minimality effects in Swahili', in A. Akinlabi (ed.) TheoreticalApproaches to African Linguistics (Trenton, New Jersey, Africa WorldPress).PELLING, J. AND P. PELLING. (1974) Lessons in Ndebele (Harare, LongmanZimbabwe).TAUAARD, P. C. AND S. E. BOSCH, (1988) Handbook oflsizulu (Pretoria, J. L.van Schaik (Pty) Ltd.).WALD, B. (1973) 'Syntactic change in the Lake Languages of NortheastBantu' Studies in African Linguistics, IV, (iii), 237-68.ZVAREVASHE, I. M. (1976) Kurauone (Harare, The College Press).