Bantu Reconstruction and the Stops Versus Continuants Controversy* KuUkoyelahblli University of Oar es Salaam Introduction In Bantu diachronic studies, continuants -B, -I, -G were proposed by Meinho~ 1910 (1899), and stop -b, -d, -g, by Homburge~(1914)and'Guthriel(1967-71) as Proto- Bantu segements. Meinhof based his pmposal on two principles: frequency of occur- rence and analogical symmetry (1929; 1932:28-31). Guthrie also used the principle of frequency of occurrence, but added another principle: probable direction of sound-shifting (1971. Vol. 1:61-2). These two confli~ recosntructions may be referred to as the 'stops versus continuants' issue in Bantu reconstruction. Although the 'stops versus continuants' issue may hllve been present in the field of Bantu historical phonology sinre 1914, when Homburger reconstructed Proto-Bantu stops instead of usingiMeinhof's:(191O (1899) oontinuants, there hasn't been any prin- cipled discussion of the differences between the two solutions. Grounds for preferring one set of reconstructions over another have never been argued in an expjcit and sys- tematic way. Most researchers in this field have been satisfied with usina either the con- tinuant or stop reconstructions without oommenting at all on the validity of their choice. For instance Tucker 1929), Tucker and Ashton (1942) and Baucom (1974, 1975) use Meinhof's' continuant reconstructions; and Meeussen (1955), Coupez (1954), de Rop ,(1958), Jacobs (1965-66), and others use stop reconstructions without even mentioning the problem. Meinhof himself, in the 1932 English edition of his (1910 (1899) important Nork on Bantu historical phonology, does not even mention the problem. More recently, the stop versus continuant problem has been discussed or comment- ed on by Nurse (1979b), Hinnebusch (1913), Mould (1977) and Hinnebusch et. al. (1981), who think that continuant reconstructions are the proper ones. The arguments given ltre: continuantspccur morelre.~n!ly~n Eastern Bantu languages, and the Meinhof so- lution is more economical than the Guthrie one. Also, Mould (1977), discussina Dahl's Law and othCl" sound shifts in the phonological history of Luyia, constructs a compli- ~ted argument to the effect that these changes can be explained if, and only if, con- Hnuants are posited and a diachronic conspiracy to preserve the redundancy of lfoicg assumed. Howeve~, in the brief and superficial discussions and commentaries which are given this problem by these workers, the methodological principles and their implications are barely touched upon. Thus Hinnebusch's observation that the ...... questim of continu- ants versus stops'basnot,been fully argiied in,tbe literature" (1973:8) stiJ holds. This paper deals with the 'stops VCl"SUS continwmts'!Sproblem in Bantu reconstruc- tion. Arguments that have been used in support of each side will be considered against the background of some assumptions concerning the well-fonnedncss of reconstructions, how sounds change, the strength hierarchies of seaments, etc. A munber of solid con- siderations lead to the condusion that fstop reconstructions result in more plau.ible and economical derivations than continuaJll ones .. The paper discusses three reconstruction principles: the methodoloaical principle 'op.freauel1~f/OcCUlTence, Meinhof's principle of analogiC"'l symmetry; Ed tbeprinci- 1 pIe of SimpJJClty. Considerations from the tneory of strength and lenition hierarchies are also brought to bear. upon the issue, and a critique of Mould's "conspiracy" argu- ment is presented. TIle PrlDdple of Frequeacy .of .OcCUI'l1llCe " Sometimes known as the "majority vote pnnciple" (Zwicky 1973:408) it states that, other things being equal,I!., ... .if the majority of the daughter lfanguagesiagJ'ee in having a certain featute, then that feature is to be attributed to the proto-language" (Zwicky, op. cit.). In imernal reconstruction it is the more frequent altemant that is presumably attributed to the pre-languqe. The problem here, as we will see latc:r, is that othc:r things may not be equal in some cases. Frequency of oa:urrence is uOOoubtedly the most important principle \tied by propo- " nents of continuant proto-segments. The argument based on this principle goes back to Meinhof (1929, 1932) who reconstructs .p, .t, .k, .B, '"l due to ihe apparent fact that they occurred more frequently than other sounds which correspond to them in the languages he investipted. This position has been supported by Mould (1977:389) and Hinnebusch ct. al, (1981:16). ProbielDS wi. tile .... Ullleut As far as Bantu is concerned, the argument based on frequency of occurrence has two probletns. In the fust place, the claim that the continuants B, t, G are more fre- quent than stops b, d, :8 is apparently not wholly supported by the facts. According to Gut~e (1967, Vol. 1:62), the "distributions of the bilabial voiced stop (plain voiced and implosive/glottalic) and the oontinuant are very similar (See also Guthrie's Topo- gram 6:71). Also, the velar voiced stop (plain voiced and implosive/glottalic) is diStributed over a larger area than the corresponding continuant (op. cit., p. 62; see also Topogram 10, p. 75). The only continuant that can be said to be more frequently attested than the cor- responding stop in Bantu ~ t (op. cit., p. 62; see aIs~ Topogram 8, p. 73). Here, d has a clearly restricted distribution. The general situation has been stated thus: " ... the reflect of .nd usually contains d, while that of jd does so in languages ~ far apart as Ganda" E.15, Ngazidya G.44a, Nyanja N.31 and Venda S.21. In addition many languages have di as the reflect of .di" (opp. cit., p. 62). If these distribution statements are correct, then the "majority vote" principle would not automatically rule out b or g as proto- segments. 9nly .d would need some justification on other than this principle. The second problem with the argument b~ on the "majority vote principle" as it has been applied to Bantu by proponents of the continuant solution is that it has been used without regard to other constIaints on the well-formedness of reconstructions. These constraints or principles include simplicity, and phonetic and typological plausibility. The simplicity criterion requires that a description be as concise and utiliz as few con- structs as possible while the criterion of phonetic plausibility requires that the sound changes or rules posited should reflect what is physiologically probable. On the other hand, typological plausibility requires that well-fonned reconstructions reflect the structural types, rule types and sound shifts that are in the daughters, and also agree in important ways with the types that are usually fouOO in languages around the world (cf. Lass 1978:272). «;oueq.-ces of pllc.aetlc ell.. illrelaUoa to freqllellCy of oceurre .. Why should the pinciple of fIeqUCDcyof oa:urrence not be given precedence without exploring the other possibilities? The reason is simple and oommonplace to every lin- quist who has worked with phonological evolution: it has to do with the consequences I 2 of long-term phonetic change. Given the necessary consequences of long-term phonetic change, the original or input segment might come to represent the minority vote. This can be exemplified, first, from languages with written traditions: Spanish and Greek. Ferguson (1978) reports two interesting cases of the historical spirantization of *d in Spanish and Greek. In both languages, *d changed to the labio-dental continuat D ex- cept mainly post nasally. In Spanish, the continuant is "...a little more than twice as frequent as the phone (illin text occurrence..." (p. 410), while in Greek, the "frequen- cy of / D / is close to three times that of / d / " (p. 414). Another example may be taken from Rimi, a Bantu language. All Bantuists are agreed that the proto-voiceless stops are: *p, *t, *k . Rimi has the alternations: p/$, t/j, and kx/x (Note that the facts are more complex in the case of the reflexes of *k\ cf. Olson 1964; Kahigi 1988: 132ff). Diachronically, *p&, %*, *k>kx>x. Synchronically, the distribution of stops is very restricted: in the case of p and t, they only occur post" nasally; k occurs mainly before j , i, and y. (kx occurs postnasally). The continuants are overwhelmingly more frequent. Here again, what is taken to be uncontroversially "origi- nal" has been overtaken by the historically derived segment in terms of frequency of occurrence. It is clear that the Spanish, Greek and Rimi examples speak for themselves. Fre- quency of occurrence cannot always stand as proof of the direction of a change, i.e the most frequent is not necessarily the 'original' segment. In some cases the less fre quent alternant or segment in a correspondence set may represent the original segment, if all other things are equal. The Principle of 'Analogical' Symmetry Meinhof refers to this principle' as 'analogy'. The principle may be defined as fol- lows. If the majority of the segments (or features) in a series (e.g. stops) are reconstruct- ed on solid evidence, the remaining segment (or feature) may be posited on analogy with the others even if there is no sufficient evidence to support such a reconstruction, It should be noted here that the synchronic counterpart of this principle, known as "pat- tern congruity" in structuralist literature, has been criticized (cf. Hyman 1975:94). This principle was used by Meinhof to reconstruct *G. He reconstructed this seg- ment 'hypothetically', i.e. it was not present in the data he was using; he did so on 'anal- ogy' with the continuants *B, *1, which he had reconstructed on the basis of the "majority vote" principle. One methodological objection here is that Meinhof does not say why he uses 'analogical symmetry' rather than 'frequency of occurrence' in the reconstruc- tion of the voiced velar. When one adopts a new method the least one should do is say why. This was in fact necessary in the case of Meinhof since if he had used the 'frequen- cy of occurrence' principle he would have ended up with the voiced velar stop instead of the continuant. This is because the former is more frequent man the latter in the languages listed on his map (facing p. 248). (Meinhof s map includes many of Guthrie's zone D, E, F, G, N, and P languages which have the voiced velar stop — cf. Guthrie, Vol. 1967:75, Topogram 10). In objecting to Meinhof s use of analogical symmetry we are not claiming that sym- metry should never be used in reconstruction. After all, symmetry is a common charac- teristic of linguistic systems (cf. King 1969:62, 191-9), although it is also not uncommon to find asymmetries in languages. However, methodological decisions in reconstruction, as elsewhere in the social sciences, should be based on data, and should always be justi- fied in a principled way. Undoubtedly, Meinhof's approach leaves a lot to be desired. The Principle of Simplicity The importance of simplicity or economy in description was stated by Halle: Given two alternative descriptions of a particular body of data, the description con- taining fewer ..... symbds will be regarded as simpler md will, thererore, be preferred over the oth~s" (l962:SS) .. In dia\:hronic description, it is a measure of alternative solutions In terms of the number and comp1eXay or'the'pro{o-segments, ciachronic rules and mecluuUms of chllnF posit- ed. It should however be noted that simplicity is meaningful only when applied with due regard to considerations such as phonetic and typological plausibility. That is, the segments, rules, and the mechatUims posited should be b~ on a reconstruction - projection and mapping - methodology guided by the simplicity criterion and plausi- bility considerations. In this view, the simpler solution should also be the more plausi- ble, phonetically and typologically. The issue of simplicity in Bamu diachronic description has been broached by Hin- ' nebusch (1973:8-9), who claims that the continuants reconstructed by'Meinhof have an advantage over Guthrie's stop reconstruction as far as the Kenyan Coastal languages are concenred since the type so( changes that have occurred historically in the Kenyan CoastallanguaBe5 are mOR: easily predicted in terms of Meinhof's recmstructions than (;uthne's, in that an CIltra step would be required, lBing Guthrie's forms, in gettiqj from the deep fl'rtnS to the surface forms (p. 8). Hinnebusch gives $ an example the following mapping to suppOrt his claim'; (1) Meinhof's reconstruction: .B)w~ Guthrie's reConstruction: .b>B>w'fI Flere Meinhof's reconstruction appears simpler since it has fewer steps. However, it is easy to see that these mappings, in isolation from related mappings and other parts of the 'historical grammars' of thelanguqes in question can not prove the claim of rela- tive simplicity. In order to demonstrate that a solution is simpler than anotl1ei, die com- peting solutions have to be assessed in terms of the 'whole grammar' vis-a-vis the constrains of phonetic and typological plausibility. F.f With the exception of cases of weakening and loss (which appear to be 'natural'), it can be.said that the Meinbof solution relies heavily on the mechanism of strengthen- ing, and'may be refaTed to as the strengthening solution. As formulated by Meinbof (1932:31), tlm solution proceeds on the wumPtion of the 'half-plosivity' ofthe posited continuants and their 'inherent tendency to become plosive' . There is context-sensitive and c?Jrtext-free strengthening: the former takes place before the close vowels j and y, after J, and postnasally; the latter 0CC\lI$ in cases of e.g. implosion as in the Kongo case noted above. Examples of streogthenina before/after the high close vowels are : Before :: .oCsotbo ~ (dental d); .Ji)dzi (N)'lUDeZi, Sbona); before Q: .Ju)Sotho, du (retroflex 4 d); * :Iy, *ly)Venda bvu (Meinhof, 26-7). After j: *-G~S.)Swahili -ib- 'steal' (with loss :>f*G) (Meinhof, 121). Postnasally, *NB>Nb, N1>Nd, *NG>Ng. In addition, Meinhof jlOSitSanother mechanism, analogy. This mechanism is supposed to accol.Ult for certain Jccurrenc.esofvoiced stops, e.g. *-(1)ngo>-gongo 'back of body' (Sumbwa, Sukuma Nywamwezi). The "G)8 change here was sUPPosedio haveQccurred in analogy withthe 'lOstnasal stop. In sum, Meinhof's continuant' reconstructions give'rise to the follo~inl mapping rules: (3) (i) *B,b, G; bv, B, w, v,.1f (ii) *l)d, J: dz, d!, bv, J : ., •• V, Z, 11 (iii)*G> g"G,' {, bv, (&.1., k, .... v .. z, .If . As noted above, in this solution, the stop and affricate reflexes result from strengthc:n- ing, while the rest of the reflexes are either retentions or results of different types of sound changes. In contrast to Meinhof, Guthrie's solution heavily relies on the mechanicsm of weakening, and may be referred to as the weakening solution. To accol.Ult for the oc: currence of continuants and affricates or loss Guthrie (1967:55-80; 1971!30-64) posits weakening rules of the type: *b)w, *1>';" .d)jIl, .cbl,.d)dz ..• g1J, .~. etc. Some of the rules posited by Guthrie can be summarized as tbe fOllowing lenition chainshifts: (4) (i) (a).b) B)w)Jt (b) • b)bv)l (ii) I(a) 1 *a)l)Y)lf (~ii)(a) *J1}(1)IJ . (b) .g)g):)dZ>z 'fhe output of each shift can be found in some Bantu languages( s), and can be regaraed as a stage in the phonological development of Bantu. In this view, different languages in Bantu can be seen as reflecting v;u;yiniJ time depths. Examples of some of the data that ledJ9 the position of *b, *d, .g are: • 5(a) C.S. 5a ~-babud-.'singe' V.t. Bobangi -babol-, Lega -babul-, Nyankore -BaBul-, Ganda .-baQul-, Sumbwa -BaBuh Luchazi "-Haul-. LUba-Kasai -bobu/-, Luba- Katanga -babul-, Bemba -BaBul-, Ma,N,anjaj-waul-;Herero ,-Baur-, Zulu - 6aGul- (Guthrie Vol. 3, p. 18, except Sumbwa example, which is from KahiJi 1988). (b) C.S: 591 .-dj"Toot' Ombo..bo-l~'R'mdi umu-dzi, Gandaomu-zi, Kikuyu mo-ri, Kamba moo', S'ukuma (n) .d!i,2 Hungu mu-"!i, Mbunda mu-Di, Kwanyama omu-di. (c) C.S. 771 *-Rambo.'affair' Teteladi-kambo, Rundi i-dzambo, Nyoroek,{-~mb(), Tongwe e-Gambo, Sumbwa i-gambo, Swahili (Unguja) ki'gambo, Luchezi ts-ambo. Luba-Kasai di-ambu, Kahonde c-ambo. I1a k-ambo, Matengo [j:gambo , (Guthri~ Vol. 3, p. 205, except Sumbwa eXl!II1plewhich is from Kahigi 1988). Examples in 5(a) illustrate some of the reflexes of .b, 5(b). those of *d, and 5(c) those of .g. In 5(a) C.S. Sa has the following correspondences; tlBlw~ root-initially, andb/B/"f/~ root-medially. In Guthrie's view, .b. is the mast probable p-ota-segment not only because it occurs in many Bantu languages, but alro because shifts like -b)Jt\ v.H1 are more probable tMn B)b. Frequen~ of occurrence and the probable direction of sound shifting3 also mctivate the positing of -g in S(c) (C.S. 771),1 Here the correspon- dence are: k/dz/g/(IG/~ g occurs more freguentlythan G (in tM and other compara- tive series) and.also the direction of the shifts- .ttGIII (~ening) and .g)dY (paIahlli~tion -a weakening process) is more prObable than the Meinhof alternative. Now, turning to -d in 5(b) the correspondences here are: V dlJ./r/1.I~/d:i17.1d/D (C;S. 591). Arguments to reconstruct -Ci instead of a liquid are: (a) d (and relevant environment) can acCount for affricates d1, dz, spiranW~!, z and' the liquids I, T, in the most economical and phoneticaBy plausible way4. (b) The widespread occurence of d postnasalJy; which, as will be shown uuection\3.4. is jlenenilly a position ofretention in Bant\1~ .The retention of d corresponds with . the reteiition of b and g in this position; these, in t1a'n, paralJel the widespread post- nasal retention of ~p, ~. and -kS_ s \c) The parallelism between(d>I,'r .. and *t,)!,r. .. in Bantu. All Bantuists who have dealt with phonological reconstruction have at least reconstruaed *t)I, r. .. This sonoran- tization of *t, though not as widespread and far-reaching as that of *d, occurs in 10 of the 15 Bantu zones set up by Guthrie (fr. Vol. 2; 30-64).6 It is no doubt an important .shift in Bantu. What is interesting about the parallelism of *t)I, rand *d) I, r is that the former implies the latter. That is, if a language has shifted *t to I or r, it will also h~vel shifted *d to I or rete. )'his implicational relationship suggests that the shift involving *d antedated that of *t7 In addition to lenition shifts, Guthrie posits devoicing (which is a 'strengthening' -cf. section 3.0) to take care of cases such as those affecting his *g in some languages (eg. Tetela in 5(h) above), i.e. *g)\{. (For more examples see Guthrie 1970, Vol. 3) . . Guthrie also posits rules of implosion, i.e. *b:>6, *d>&',*d, which take care of de- velopments in Swahili and other Bantu languages.S Now given these considerations, which solution or analysis is simpler? For our pur- poses, the foJlowingshouId be noted. First, Meinhof's notion of the 'half-plosivity' of continuants and their 'inherent tendency to become plosive' (1932 p. 31) is highly questionable. It is evidently posited so as to trigger some strengthenings. Thus, *G changes into? g in Kinga due to its" .... ten- dency to become plosivc" (p. 31). However, all that is known about sounds and their Jynamic tendencies does not support the idea of a continuant or non-continuant having an "inherent tendency" to become some other sound. Sounds hav," been noted to change dueto struct ural, phySiologiCai."psychological,language..aquisitional,andcontact factors. A sound does not, in and of-itself, have any tendency to become some other sound; the activating factor of a sound change has to come from somewhere else. In addition to this problematic aspect of Meinhof's solution, there is the issue of the satus of 'strengthening as a mechanism of sound change. Strengthening, according to Meinhof, occurs befare \. and y after 1, postnasaBy, and in other environments. It should, however, be noted 1hat strengthening before and after high cbse vowels is difficult to defend, in view of known history and other Bantu-internal facts. Known history tells ~¥iIi!li;theenvironment of high vowels (especially j) is.~enerally, a ~eakening environ- ment. Taking i as an example, this is, universaIfy, a palataliziIlienvironment (Foley 1977:90-106). Palatalization is commonly followed by assibiIation. These two processes are the usual historical sources of dy, d~, dz, ~, z; ty, tr, ts, ~, s, etc. Bantu-internal evidence also supports this position. According to Meinhof (I 932:26fo, Guthrie (1971, Vol. 2:30-64), and others, there occurred, in Bantu prehistory, widespread weakenings before'~ and II in the voiceless stop series. These weakenings gave rise to a lot of palatal sounds 8nd spirants. In some cases, however, this environment has not caused any palatali- zation or assibiIation, e.g. in Makua, *JJ9Phi, *py>Phu (Meinhof, p. 27). Thes~ few facts about the diachronic evolutioft of the vOiceless stop series point to the fact that weaken- ing (or no c~ange), ~ot strengthening, is the solution that makes more phonetic and typo- logical sense'in the envirooment in question. This, in turn, points to the validity of Guthrie's stQp reconstructions. In Gutluie's view, weakening mayor may not occur in i the environment before andlj or after~. This, however, does not mean that strength- ening does not occur in this environment. What it ;means is that, if it occurs at all, it should somehow be the exception, not the generaI rule .• A related consideration concerns Meinhofs 'analogy' part of his solution. In this case, the postted cominuants bewme stops on 'analogy' with postnasal stops. Thus, in Swahili .Q)g "by analogy with it change after a nasal, viZ.n + Gmg" (p. 31). What Mcinhw means by 'analogy' here is vague, but his move to resort to this "'mechanism' .Pl this casc is auite undcrandable: after all, he has to account for such cases of non- 6 postnasal 'strengthening'. The move is necessitated by the very analysis be adopts. But anllIogy (whether phonetic or morphological) operates on the basis of the same- ness/ similarity of the relevant elements and environmems. The cases supposed by Mdnhof to have been affected by analogy lack the neCessary oondition on which this mf!dum; ..... is based: the sameness/similarity of environments. Guthrie's approach is of course sim- pler and more plausible: to him this is simply a case of retention. Retentions of this sort are widespread allover the Bantu area (c,f Guthrie's vol. 3). Although this dmcussion does not exhaust the problem of simplicity in Bantu projec- tiOn and mapping, it has shown that Meinhof's solution is very problemati.: since it gives rise to mappingS which are not in line with phonetic and typological plausibiity. Gutiu:ie's solution, however, makes sense phonetically and typologicaly, and is, on the evidenct, simpler. WeakeRing .. d Strengthming of Segments It was noted in the foregoing that Guthrie us~ the principles of the probable direc- tion of sound shifts and frequency of occurrence to posit PO stops, both voiceless and voiced, from which the current reflexes are denved. His SOIUtiOllaIlS been referred to as the 'weakening' solution, represented by tbe chainshifts below: phi) (a) "6>~v»w>R .. (b) b)bv", (ii) r(a) .. d>d!>z (b) d)l)(y», (iii) (a) "~)Q)y ,(b) *g)gy)dl)z As is evident in these chamshifts, the shifting is in the dirett'.on of the wedges. On the other hand, Meinhof's continuant reconstructions result in rules which have a reverse :lirection to some of Guthrie's rules, e.g. *O>b, *l)d, "G>g etc. This type of solution has been referred to as the 'strengthening' solution. Strength I1ierarchies Strengthening and weakening have been defined within the theory of strength hier- archies (cf. Zwicky 1972; Lass and Anderson 1975; 148-187; Hooper 1976: 195-242; Foley 1977, and Lass 1984:177-183). Briefly, this is a theory according to which segments, environments of the'word (initial, medial, final), and positional classes(labial, alveo- lar, palatal, velar) are characterized in terms of relative strength in relation to synchron- iC and diachronic processes in language. For our purposes here, we shall CO&X:em ourselves with segmental and environmental hierarchies;'we shall not deal with positional hier- archies. Some relations holding among segments are s~ematized as follows: (7) (i) zero>vowel)glide;mquid~nasal;>fricative.!asfPfi~at~~op ~ ncat~- (i1) voiced obstruent}voiceless obstruent (iii) S}SS or ~~(where S = Segment; SS = geminate, and Si Sj = cluster) In these schemata, the direction of the wedges shows increltiing strength. In (i), stops are the strongest segments, and vowels the weakest. Here it should be noted that the notion 'strength' is associated with the phonetic parameters of openness and sonority. The most open and S(>Dorantsegments, (i.e. vowels) are the weakest, while the least open and sonorant (I.e. voiceless stops) are the strongest (cf. Lass and Anderson 1975:151; Hopper 1976: 198). In (ii) voiceless obstruents (e.g. t, ts, s) are stronger than voiced ones (e.g. d, dz,z). In (Hi), a gerninae (SS) or a cluster (81 Sj) is stronger than a single seg- ment. Thus, in terms of strength, a voiceless geminate stop is strooger than its single- segment counterpart, which in turn is stronger than its voiced counterpart, and so on. Now, if a segment's rank changes in the direction of tbe wedges, the process is called 7 streogthening; if its ranJt dUlD&esin the reverse direction, it is a weakening. These are key notions in the theory of strength hierarchies. An important task of the theory is to specify the contexts in which these proces$:S. take place. Segments commonly weaken or strengthen in specifIC envlfonments. Such environ- ments have been referred to as weak andlstrong environments, respectively, and will be considered from two angles: the environments of the word, i.e. initial, intervocalic, and fInal, and other specifIc phdological envronments, e.g. the pre-C or post-C environment. laitial, iDtervocalk: _ ("mal enliroaments Word-initial environments are assumed to be strong. This assumption is supported by the universal fact that all contrasts of consonants in a language may occur initially, while contrasts in the fmal position are fewer and tend to be neutralized (cf. Hooper 1976:2(0). FUrthermore, apparent strengthenings have been observed to occur in this environment, e.g. Sp(anis~ huevo 'egg' has two pronunciations: [we~ebQt; huerto 'garden' iU'ronounced as~ertcj o~rt~); L(atin) vita [wit~ 'life' Sp. vida ~iDij; Sp. Vivo lI'iBQJ'a1ive'. The intervocalic position is the preferred one for weakening. It is a w~ environ- ment; here, wnsonants take on some of the qualities of the surrounding vowels - for example voicing and continuancy. In their development from Latin, Spanish and French segments have undergone weakening in specifIc intervocalic environments: Latin Span,ish French agua agua ttaGual eau to 1 'water' amicaQamika) amiga damlliltr ami 'friend' legere . leer lire. tli:r':J 'read' credere creer crOIre (jkrwa:r') 'believe' In these examples, the 'input' intervocalic stops are: .g, .k, and.d. The diachronic (weakening) rules are: .g'Ki (Spanish), .IogkJ (Spanish), .~ (Spanish and French), and .d>D)0 (Spanish and French). The word-final environment is regarded as the weakest since loss, the logical con- clusion of weakening, has been observed to occur here (in closed syllables) before oc- curring in the other environments. The reason for the weakness of the word-final environment is that final segments are more often pronounced weakly. rhe weakening of consonants in this environment typically proceeds by devoicing, then g1ottalization, which gives way to total loss eventually. Examples: weakening and loss of final .p, .t, .k in French, Maori, and Chinese; also, in many dialects of English, final voiceless oc- clusives, especially t and k, are being replaed by a glottal stop, which is. iust one step from total loss (cf. Aitchison 1981: 32-33). Therefore; hierarchically, word-initial positions are strong, intervocaic ones weak, and final ones the weakest. The concept of 'protection' We now turn to the consideration of strong and weak environments as they relate to the concept of 'protection'. One key .sumption in the theory of strength hierarchies is that when strengthening occurs, strong segments are affected fust "and most exten'- sively and preferentially in-strong environments," ~hile weakening occurs to wea~ seg- ments first "and most extensively and preferentially in weak environments" (Foley 1977:107). We have already noted that theintervocaIic and the final positions in a word are weak. The initial environment is strong. Additional environments which have been observed to behave as strong ones we pre-C, post-C, and after a stressed vowel (cr. Anttila 1972:66; Meinhof 1932:29, 59; Foley 1977:91). One characteristic of strong environments is that tbey tend to resist weakenings which commonly occur in weak environments. 8 In other words, they tend to 'protect' the relevant segments from phoneiic attrition for as lonitas possible. For l;xample, consider the evolution of Proto-Indo-European voice- less stops which spirantized in all environments except when they occurred in a post-C environment, e.g. Latin captivus 'captive'; Old High German spiwan 'spit'; GothicflSks 'fish' (Anttila 1972:66) The only probable reason why there was no change in this en- vironment is that the segments were protected from weakening by the first consonant in:thecluster.Anadditioqal example of a protective environment is the Bantu postnasal environment. This can be said to have protected segments. from weakening or further weakening. Examples have already been given of diachronic shifts which have occurred in Rimi, a sister of Sumbwa. In Rimi, * p and *t have weakened intervocalically, includ- ing the initial position, but not postnasally. Examples are: iefo/mpefo 'cold, wind' ( PB *-pepo), -iik-/mpiko 'arrive/arrival' ( PB *-pik-). Additional examples are fron. Haya (E. 22; from research notes) and Kongo (cL Meinhof 1932:158-9); in this case, Proto-Bantu*p lenited to h in Haya and B in Kongo, except in the postnasal envirc.n- ment. Haya examples:-h- 'give', (mpa) 'give me' ( PB *-pa-); -hulil- 'hear' , (mpulile) 'that I may hear' ( PB *pudid-); Kongo examples: -Ban- 'give', (mpeni) 'I have given' ( PB *-pa-), -Bol- 'be cold,' and (mpolo), 'cooling' ( PB *-pod-). It is important to note that weak and strong environments only constitute "preferred" environments for the respective processes, (Lass and Anderson 1975:159ff; Foley 19~7:107ff). Segments need not weaken or change at all in a weak environment, nor need they strengthen in a strong environment. In other words, weak and sU'oI}genviron- ments do'not constitute thenecessaryana,sutficient conditio~ for the respective!process~s. The theory of strength hifrarchies and Bantu reconstruction Having clarified some "aspects of strength hierarchies, we will now related the the- r ory to Bantu reconstructions .. One way to find out which environments are relatively strong or weak in language or group of languages is to consider reconstructed history with a view to pinpointing the weakenings, retentions orstrengthening. In Bantu, this can be done by considering the diachronic shifts of some proto-segments.on which all Ba.ntuists are agreed upon, i.e., *p, *t, *k, which,' according to the theory of strength hierarchies, are strong seg- . ments. In Proto-Bantu/they were the strongest ,I regardless of whether one adopts Mein- hof's or Guthrie's reconstructions. Given their relative strength, massive weakenings affecting them in many languages would be highly diagnostic of the weakness of the environments involved. Thus a weak environment would be one which, on the basis of internal and comparative evidence, could be shown to have indu~e'd/extensive weaken- ings 'of *p, *t, and *k in many Bantu languages. A strong environment would be one which could be shown to have either induced strengthenings or protected segments from phonetic attrition. After iqentifying the weak and strong environments for the voice- less stops, we'shall apply this consideration to the issue of Meinhof's and Guthrie's ree n- structions. ~ First, however, a few facts about Bantu should be clarified. In general, Bantu Ian guages tend to favor open syllabI~s, and consequently Proto-Bantu has been reconstructed thus. This means that, for the ~ata we are dealing with, there are only two consonant environments in the word: the initial and the intervocalic. The former is l'epresr.nted by Guthrie as C I. and the latter as C2.AIthough tTie d,istinction between these two environ- ments m~ be regarded as unimportant in very many languages due to theopertness of their syllables, we shall, for our purposes, maintain it. Another relevant fact to note : here is that, apart from NC clusters, there were no other structural CC sequences in Proto-BllJ;ltu. The reflexes of *p, *t, *k in relations to weak and stro.ng e/lviromnents The recoffitnicted histories of these segments indicate that they have weakened in various environments. For example, two of the five languages used in Meinhof (1932) (Pedi and Kongo) show weakening of voiceless stops in C, and C2 posi- tions, while all of them show weakening before high close vowels. Some examples are given below: Pedi (pp. 58-81): weakening in CI and C2, and also before high close vowels . • p>i: -fa (r: -raro (4'B .-tatu) 'three', -phiri «PB.("-Plty 'hyena'; k>x: -xam- a>B .-kam-) 'milk', -xura (~B .-kyta); .k>s: inosi ( PB .-okj> 'smoke', mo-siJa (f:-fik- (PB ._p~c-) 'hide'; *t}s: -sima (PB *-tjrna) 'pool'; .~s: mW-is(<.pB*-oky 'smoke'. Before *1j:.p, *t, .k)f: -fukul- (PB .pykul-) 'dig out', -fuku (PB .t1jku) 'night', -futa ($,f, s, in various environments and *Np?Nb (Shaghala and Dawaida) (cf. Slavikova 1975:36, 53). Here the postnasal *p gets voiced (i.e. weakens one step), but the environment itself is still strongei in relation to other environments. Other changes in the postnasal environment include the so-called Meinhof rule by which *Nb, *Nd, *Ng>NN, i.e. the stop ac- quires the nasalization of the adjacent nasal, if the following syllable consists of nasal + voiced stop. This has applied in Sukuma, Ganda, and other Bantu lan- guage (Meinhof 1932:183-4). Other weakenings affecting stops in the postnasal environment are: *Np, *Nt, *Nk>Nh in Nyamwezi and Sukuma; affrication in Pedi, e.g. *Nk>Nkxh; aspiration, deocclusivization, and voicing in some Southern Bantu languages, e.g. Tswa *rprJmph}mh. From the above evidence ar^d considerations, the following inferences are in order: ,. The preferred process intervocalically andbef ore high close vowels is weaken- ing. The initial environment behaves like the intervocalic in very many lan- guages (probably due to the open-syllable character of Bantu languages), although in some languages, e.g. Ngazidya* it appears to be strongar than the jntervocalic. 2. The postnasal environment is relatively strong. Here, segments get retained which weaken in other environments (cf. Haya and Kongo examples above). However, as n o ted above, postnasal segments are not immune qo sound change. Some implications for reconstruction of the voiced series Recall that the reconstructions in dispute are voiced stops and continuants. Note further that these segments are weaker than the uncontroversial voiceless stops considerd above from the point of view of weakening and strengthening. Now, if. "....weaker elements weaken first and most extensively and preferentially in weak environments" as is assumed in the theory of strength hierarchies (cf. Foley 1977:107), then the follow- ing further inferences are in order: 1 If stronger segments (in this case the Proto-Bantu voiceless stops) weaken in an environment, weaker segments should also weaken in this environment. That is, in. principle, such an environment cannot be a weakening one for strong elements, and a strengthening one for weaker elements. This is in accord with what is known about the weakening of segments in general. 11 2. :Smce the postnasal environment is a strong one for stronger segments, it snould also be a strong one for weaker segments. These typological statements may be used in the attempt to make a decision as to which solution is the proper one, Meinhof's or Guthrie's. Apart from stating weak and strong environments, they also state the direction of change in these environments. A proper solution wquld be one which would be in line with these typological statements (if all other things are equal, of course). In what follows, Meinhof's and Guthrie's solu- tionswill be considered in the light of these inferences of weakenings and retentions in Bantu prenistory. First, in Meinhof's solution, voiced continuants are posited and then mapped on to the various reflexes in modern Bantu languages, including voiced stops and affri- cates. Strengthenings occur postnasally, before *\, *1J, after *j, stem-initially and inter- vocalically. However, some initial strengthenings e.g. *Gongo> -gongo back of body' (in Sumbwa, Nyamwezi, Swahili, etc.) are supposed to be explained by what Meinhof .calls 'analogy'. The following rules summarize these changes: L [;onD > C~gn~J 2. [+ cn~>r=-c~t .' L+ ImploslOuJ 3. (+cnD>l:cnQ/-VNCV Now, postnasal ;trengthening is..phonetically, a plausible process, since the condition~ ing environment is strong. But strengthening in the intervocalic environment and before*l and *.9' and after *i (and also stem-initially in many languages) seems to be unm~tivat- ed, typologically and phonetically, given inference (I) above and preceding conSIdera- tions. Continuants, being weak consonants. are not expected to strengthen in weakening environments. If anything, they are expected to weaken further in such enviroaments (cL Foley 1977: 107). In effect, Meinhof's solution has some unnatural consequences as far as the theory of weakening is concerned. Guthrie, on the other hand, posits voiced stops *b, *d, and *g. As noted earlier, Guthrie takes into account two considerations: (I) frequency of occurrence, and (2) the probable direction of sound shifts. It is the second consideration, that of the principle of the direction of change, that makes Guthrie's solution agree with the conclusions based on the theory of strength hierarchies. The typological perspective of strength hierarchies, then, favours Guthrie's voiced stop reconstructions rather than Meinhof's continuant reconstructions. Mould's 'conspiracy' argUba~b~ Before concluding we need to address Mould's (1977) argument for reconstructing continuants for the voiced series, especially since Hinnebusch, Nurse and Mould (1981: 16) refer to it as an additional argument why Meinbof's reconstructions are the proper ones. Mould'saim is to adduce evidence to the effect that, firstly, the featurelvoic~ was redundant in PB; and secondly, there has been a 'diachronic conspiracy' which has func- tioned to preserve the redUndancy of this feature in some Bantu languages. A conspira- cy is a situation whereby two or more seemingly independent rules/changes appear to work towards a particular structural effect or target (Kisseberth 1970). An example is two Sumbwarules, devocalization and vowel assimilation (cL Kahigi 1988, chapter 3), 12 which may roughly be written as: (13) V~ (-syl)/mV (14) Vi + Vj ~ Vj t+ h!J chi) 8- loni These rules may be said to 'conspire' against VV-sequences in phonetic representa- tions in the language. This is, of course, a 'synchronic conspiracy'. In a diachronic con- spiracy, an additional variable, that of time, is involved, and here the rules/cl1anges are pressumed to work toward maintaining a specific situation, or as in Mould's case, the supposed redundancy of lYoic~. Mould's evidence for the lY~icc;}redundancy and the conspiracy "consists of motivating and explaining subsequent phonological changes in various Bantu languages" (p. 389). Mould proceeds on the presupposition that three cate~;ories of phonological phenomena need to be explained as far as Bantu prehistory is concerned: assimilations, dissimilations, and chainshifts (similar to Grimm's_Law). To him, assimilations pose no problem: "ease of articulation is motivation enOl,lgh" «p. 389). However, dissimila- .tions (such as Dahl's Law) and chain shifts (such as ones that presumably occurred in Luyia) are more difficult to explain. In order to explain these changes, Mould (assum- ing Meinhof's reconstructions) proposes a redundancy of 1Y0ic~ in PB and to consider the changes in question as constituting a "conspiracy to preserve the predictability of (Yoicij"(p. 389). A~cording to Mould, the conspiracy was "most thoroughly carried' out" in Luyia (p. 390). Mould's evidence for the redundancy of (Yoicij in PB concerns Dahl's Law. This isa law (or rule) whose original formulation by Edmund Dahl was that the first of two voiceless aspirates in two neighbouring syllables dissimilated by losing the aspiration and getting voiced (cf. Meinhof 1932:181). Mould thinks ease of perception is "weak and insufficient" as an explanation of such a dissimilation; to him there is a much im- portant factor: ... what is more important is what made Dalh's law possible, and that is that there were no voiced stops already present, and therefo~e there was a lot of phonologi- cal space encouraging free variation, which, together with the motivation for per- ceptual ease led to phonologization (p. 390), Now, since DaIh' s law is supposed (in Mould's view) to presuppose the absence of voiced stops in the system (that is, its basic motivation) it is easy to see why it would supposed- ly support the redundancy of fyoicij, thus endorsing Meinhof's reconstructions .. But Mould's 'explanation' of Dahl's law poses a problem which stems from, inter alia, the fact that his explanation depends on the supposition that the variation that gave rise to dissimilation depended on the ab;ence of one of the members (the voiced mem- ber) from thephonernic inventory. Linguists and other students ofIanguage have known for a long time that variation is an immanent quality of language on all levels. As far as the sound level is concerned, it is known that articulatory, perceptual, and other fac- tors (cf. Essen 1964, Jeffers 1974, Ohala 1974a, 1974b; Labov 1981, 1982) are responsi- ble for the variation that occurs. In view of this, it is too strong a claim to say, as Mould does, that free vaFiation in respect to a feature, in this case, tyoicej, presup-poses the abs~nce of contrast involving the feature. To be sure, the ab;ence of distinctiveness of a . feature, i.e. its redundancy, makes it available for variation, but so does the presence of distinctiveness of a feature. Unless there is independent evidence poinv ing to the absenc~ of distinctiveness of a feature being the motivating factor of varia- tion, it is safer to avoid such ~ supposition, especially in prehistoric reconstruction. Moreover, Mould's supposition that Dahl's Law was possible due to the absence of voiced stops is contradicted by what is known about dissimilations in general. Con~ siderations of known cases of dissimilation from Neogrammarian times to tho present have consistently shown that dissimilations are "changes by phonemes," Le. the dis- 13 simiJated segment beannes an instance of a ditIt'.cent phoneme: (Bloomfidd 193~.390), and do not produce sounds that Ire not alre~.dy in the phonemic inventory (Hoenig- swald 1918:17j-81). ~ Qthcr words, ~,""imilations comtitute phonetic mergers (cf. Bennett 1967:137). Dqthis point, one student of language categorically states: " ... /0 dissimila- tion ne ale p/Js phonemes nouvet!llX ... "Grammont 1933:270; cf. also Vendryes 1925:62). All kn0Wl4issimilations seem to have obeyed this principle. As Hcenigswald (1978: 177) notes, su& ageneralization "must ... be taken as typological in nature and hence as subject to empiricalconimnation. .. " The opposite claim, that dissimilations give rise to new phonemes, must also be supported by evidence, which, in the case of Bantu prehistory, may not be forthcoming. The rest of Mould's evidence consists of shifts which he thinks support the con- spiracy to preserve the redundan<.y of [yoic9. These include: (I) the shift of *c to the alveolar ar~ (afterwhich it spiramized) in response to the supposed phonemicization of *j «*0); (2) the devoicing of *j in Gusii and most of Luyia dialects; (3) the spir8:I1tiza- tion of voiceless stops and the devoicing of voiced ones (Luyia Law), supposedly moti- vated by the need to level out a contrast that had begw to occur in the velar area (where, presumably, the reconstructed *G had already shifted to j, y,~, and g, there~ causing a contrast between k «..*k)and g «.*G); (4) postnasal neutralization of 1Y0ic~, and (5) devoiciJlg of strident fricatives, i.e. *v, *Qf, s. Since Dabl's Law does not appear to support Mould's theory of redundancy of [yoice] in PI, the hYPOthesis of a diachronic conspiracy going all the way back to PB seems to be'baseless. Besides, the above shifts will have to be accounted for in terms of either the processes affecting Ealtern Bantu, e.g. *C», spirantization of voiceless stops, ~c., or those having to do with the diachronic phonology of Luyia (and neighbouring languages such as Kikuyu), e.g. postnasal neutralization of 1Y0ice] (cf. Guthrie 1971, Vol. 2:3O-M). It should be noted that Bennett (1967) provides an account of Dahl's Law which takes into account the above shifts (for the Kikuyu group, Luyia and Gusii); the account is consistent with Guthrie's reconstructions. This account, very plausible and consistent with the simplicity criterion, cannot be passed over in silence in favor of an unmotivated and unprovable conspira<.y. Mould's conspiracy hypothesis could be considered from a different angle. As an "explanation", it is teleological. A teleological explanation differs from a causal one in that instead of the causal structure: "y because of x" it has the strucfure: "y in order that x" (Vincent 1978:4(9). For imtance, in Mould's c.onspiracy, some changes suppos- edly occur in order to preserve the redundancy of lYoi~. There are two types of teleol- ogy:functional,and;purposeful. A functional teleology refers to the function of an element in a system (Anderson 1973:789). As far as the phonologic system is concerned, the ele- ment may be a distinctive feature, a phoneme, or rule. As an example, consider the Sumb- wa rule of devocalization referred to earlier, roughly stated as: V -.tsyll/-.;\{ t+h3 I [Iii] A fUnctional teleolOgical explanation ot the addition of this rule to Pre-sumbwa pho. nology would take- into account .at least two thingS; first, the new relationship that wiU be introduced between the phonologic rcpresentatiom lu-V, i-V, 0¥1 and theirphonetict, representations [wV', yV, wVj; second, the rdationship of this rule to other rules in" the system - for imtance, the vowel assimilation rule, which, together with the rule in question, function to eliminate VV-sequenceS'in the phonetic representations in the language. A purposeful teleology refers to the intentionality and goal-directedness-evident in the attainment of a target, or iI\ our case, in the implementation of change (Vincent 14 1918:409-IU, Andersen 1973:780-1, 789-.Ji, not (n)dli as recorded here by Guthrie (personal communicatim). 3This will be discussed in detail in a section below in terms of the weakening and strengthening of segments. 40iven that .l»zis the usual path of diachronic derivation, it is easy to see ~at th,e affricates (dl, dz) point back to their source, i.e. d. and the spirants also point back to the!r ulti- mate so ... "". u.hcouPllhe lIitennediary, the,affncates. LIkewise since *dJD and .d)l/r are natura! shifts; the contillUants can plausibly be assumed to pOint back to .d. 5Weakening of .d in non-postnasaJ environments has already been naed to have ocCtrred in Spanish and Greek,IRil section)il,ove. This type of weakeni~ (Le. sonoramization of .d or its voiceless counterpart *0 is not uncommon; it has been reported to have OCQIrred in other languages, e.g. !agalog (Schac~ter and Otanes 1972:25) and other Austronesian languages(cf. DahI1976:55-6~); ~ some dialects ill Enaland, t and d have become r intlervocalically (Wright 1995:230,232); in DraVId- Ian languages. *t and d. have changed to either I or one of 'r' -type sounds (Caldwell 1961:153. 154 ff); also. In Mande *d has changed to I imervocalically while being retained postnasally (per- sonal communiction from David Dwyer). 6The zones in question are: A. B. C, E ~Msaba-Luhyagroup Chaga grol11'1 F (F .32 Rirni, a siste, of Sumbwa). G (G.44a Ngazidya. g. 44b Nzwani), H. (H. 13 Kunyi, Zaire), IqK.21 Lozi), P (P.3O Makua group). S (S.20 Venda group, S. 30 Solho- Tswana gro",. S. SO Tswa-Ronga group, S.6O.) 7Accordi~g to so~e ass1;1mption in the theory of strength hierarchies d is weaker than t in the al- yeoler regJO~, and thus It is expected to " .... weaken first and most extensively and preferentially In weak envlfonments" (Foley 1977:107). 8There are two theories that have been proposed in connection with the rise of Baltu implosion: (I)the theory of external origin pro{lOSCd by Doke 1931, and (2) the theory of 'inta-nal evolution' proposed bv Greenberg (1970). AcclX'ding to Doke 1931:48 Bantu Impl()slon" ... owes its origm 16 to Indian influence." This theory is unacceptable as there' is no known intense contact betwe~n Bantu and Indian languages. The theory of "Internal evolution' proposed by Greenberg I970s: 134 ff proceeds by assuming that one of t'lle ways implosion n.ilY originate in a language iSJhrough the acquisition of the feature (+ implisoIj) by plain voiced stops. On this view imploded variants must have occurred ,in free variation with plain variants before the feature r+ implosiOlj] got phonologized. This seems to be a plausible Cl{planation of the rise of. implosion in Bantu languages. Refe~ncel Aitchison, 1. 1981. Language Change. Fontana Books. Andersen, H. 1973. 'Abductive and Deductive Change,', Language 49,4: 765-793. Anderson, 1.M. and C. lones, oos., 1974. Historical l;inguistics. 2. vas. Amsterdam: North-Holland Publishing Company. Anderson, S. 1974. The Organization of Phonology. New York: Academic Press. Anttila, R. 1972. An Introduction to Historical and Comparative Linlf!Jistics. New YeWch-Ostafrika', in Zeitschrift fur Kolonidsprochen, VoL V: 227-253, 210-298; VOl. VI: 1-27, 102-23 .. Doke, C.M. 1963. Textbook of Zulu Grammar. Cape Town: Longrnans. Dwyer, David 1969. Coirs61U11'1t Mutation in Memk. M.A. Thesis. East Lmsing: Michigan State Uniw:rsity. , Ferguson, C. 1978. 'Phonoillgical Processes. In Gremberg, ed., 1978: 4m-442. Fisiak, J., ed., 1978. Recent developments of historicJl/ phonology., Jbe Hape: Mouton. f91ey, J. 1977. Foundations of theoretical phonology .. Cabridge: CUP; Grammont, M. 1933. Traite ck phonetique. Paris: De!al.rllVe. Greenberg, T."1948. ,.the tonal system of p-oto-Bantu,' Word 4, 3: J4J6-:Il8. _____ 1966. 'Synchronic and diachronic univedals in phondogy.' Langua/Jfe 42, 2:508-17. _____ . 1970: 'Some gen~ons, CP~ giottalic consmants, ~y implosives', Intel7llltiond Journal of :American Linguistics 36: 123-145. _____ . ed., 1978. UniversalsqJ~umguage. Vol. 2: Phonology. Stanford: Univ. of Califor- nia Press. 17 Guthrie, M. 1967-71. Comparative Bantu.4 vols. Farnborough, Hants: The Gregg Press. Halle, M. 1962. 'Phonology in Generative Grammer'. Word 18: 07.-'12. Heffner, R.M.S. 1969 (1949). General Phonetics. Madison: Univ. of 'Wisconsin Press. Hermann, 1904. 'Lusiba, die Sprache der LaenderKisiba, Bugaou, Kjamtwara, Kjanja, und Ihangiro'. In: Mitteilungen: des Seminars fuer Orientalische Sprachen, Vol. VII: 150-~; Berlin. Hinnebusch,: T .• 973. Prefixes, sound chantieS, and sub-grouping in the coastal Kenyan Bantu Lcnguages. Ph. D. Dissertation, V~LA: Univ. Microfilms Ann Arbor. ,D. Nurse, and M. Mould, 1981. Studies in the Classijirotion of Eastern Bantu Languages. Humburg. Hoenigswald, H. 1960. Language Change and Linguistic Reconstruction. Univ. of Chicago Press. ____ . 1964. 'Graduality, sporadicity, ani the minor sound change processes', Phonetica II, 202:215. _______ 1978 'Secondary split, typology and universals'. In: Fisiak, ed., 1978: 173-181. Homburger, L. 1914. Etude sur la phonetique historique du Bantou. Paris. Hooper, J. 1976a. An Introduction to Natural Generative Phonology. New York: Academic Press. Hudgins, C.V. and L.M. Di Carlo 1939. 'An Experimental Study of Assimilation between abut- ting consonants'. In:. The Journal of General Psychology 20:449-469 Hyman, L. 1975. PhOflology: Theory and Analysis. N.Y:: Holt Jacobs, I J. 1965 and 1966. 'De reflexen van de Oerbantoe-Consonantfonemen in de talen van de zone C'. In Orienta/ia Gandensla 2: 259-98 (1965); 3: 341-4(11966). Jakobson, R. 1931. 'Principles of Historical Phonology" In Keiler, e.d, 1972:121-138 .. ______ 1958. 'Typological studies and their contribution to historical and comparatIve linguistics'. In Keiler 1972: 299-305. Jeffers, R.J. 1974. 'On the notion 'explanation" in historical linguistics' . In Anderson and Jones, eds., 1974: 231-255. ______ and Ilse Lehiste 1979. Principles and Methods for Histocia/ Linguistics. TM MIT Press. Johnson, M.R. 1974. 'Abstract analysis and Bantu reconstruction: A Luganda example'. Studies in African linguistics 3:325-337 . Johnston, H.H. 1919. A Comparative Study of the Bantu and Semi-Baltu Languages. 2 vols. Oxford. Jordan, A.C. 1966. A Practical Course in Xhosa. Cape Town: Longmans. Kahigi. K. 1977. Sumbwa Phonology. M.A. Thesis. University of Dar es Salaam. ... ______ 1984. 'Stops or Continuants in Proto-Bantu?' Paper JYesented to the MichIgan Lm- guistic Society annual meeting, 19 Oct., 1984, East Lansing, Michigan _______ 1988. Aspects of Sumb_ Diachronic Phonology. Ph.D. Dissertation, Michi- gan State University, Mi., East umsing .. Kahler-Meyer, E. 1971. 'Niger-Congo, Eastern Bantu' in Sebeok, T.A. ed., Current Trends In Linguistics, Vol. 7: Linguistics in Sub-Shaharan Africa. The Itague: Mouton, 307-356. Keiler, Allall, cd., 1972. A Reader in Historical and Comparative Linguistics. New York: Holt. Kent, R.G. 1936. 'Assimilation and Dissimilation~ Language. 12: 245-258. King, R. 1969. Historical Linguistics and Generative Grammar. New Jersey: Prentice-Hall. Kiparsky, P. 1970. 'Historical Linguistics'. In J. Lyons, ed, New HorizO(IS in Linguistics, Pen- guin, 1970:302-315. Kislleberth, C. 1970 'On the functional unity of phonological rules' . Linquistic Inquiry I: 291-306. ----_ and M. Kenstowicz 1979. Generative Phonology. Academic Press. Koenen, M. «Rev., W.F.) n.d.New Kisukuma G'rammar of the Diocesaof M_nza. (mimeo) Kotey, P.l'.A. and _H. Der-Houssikian, eds., 1977. LaflguagCo>.indLinguistic Problems in Africa. Labov, W. 1972. Socidinguistic Patterns, University of Pennyslvma Press. _______ 1981. 'Resolving the Neogrammarian controversy'. Language 57,2: 267-308. ------ 1982. 'Building on eqJiricaJ foundations'. In Lehmann and Mallriel, cds., 1982: 17-92 . Lass, R. 1978, 'Mapping constraints in phonobgical Teconstroction: on climang down trees without falli~ out of them'. In: Fisiak, 1978:245-286. ----- 1980. On Explaining Language Change. London: CUP. ----- 1981. 'Explaining Sound C~e: The Future of an Illusion' . In: Phonologica 1980, ed. W.U, DTessler, O.E. Pfeiffer, and J.R. Rennison, Innsburdc. 1981, pp. 257-73 .. ------ 1984. Phonology. Cambridl!lC Univ. Press. _____ lI1d J. AnderSQll . 1975. Old English Phonology. CUP. Lehmann, W.P. and Y. Malkiel, eds., 1968. DirectiOflSfor Historical Linguistics. Austin: Univ. of Texas Press. ------ 1982. Perspectivess on Historical Linguistics. John Benja- mins, B.V. Mann, M. 1973. 'SouncJ..correspondences and sound-mifts'. A/rkan Languages Studes, XlV:26-35 . .18 Meeussen, E.A. 1955. 'Les phonemes du ganda et du bamu commull', Africa XXV, 2: 170-1110. _____ 1967. 'Bantu Grammatical Reconstruttions', Africana Linguistica m, pp. 79-121. ______ 11973. 'Comparative Bantu: Test Cases for Method'. African Language Sudies, XIV:6-18. Meinhof, C. 1899. Grundriss einer Lautlehre du Bantusprochen. Berlin. ________ 1929. 'The Basis of Bantu Philology'. Africa II, pp. 39-56. ______ and N. van Wamerlo 1932. An Introduction to Bantu Phonology. London (Ab- breviated Meinhof 1932). Mould, M. 1977. 'The proto-Bantu voiced cOll'lonants and the feature (voice)'. In: Kotey and Der- Houssikian, eds., 1977:389-394. Nurse, D. I 979a. Description of Sample Bantu Languages from Tanzania. African Lan- guages/ Langues Africaines Vol. 5, No.1. International Africa Institute, London. (Earlier Version titled A Phonological and Morphological Sketch of 15 of the Principal Lan- guages of Tanzania published by Institute of Kiswahili Research. Dar es Salaam. n.d.) _______ 1979b. The Classification of the Chaga Dialects. Hamburg. _______ and G. Phillipson, 1980. 'The Bantu Languages of East African: A Lexicostatisti- cal Survey'. In Polome and Hill 1980: 26-67. Ohala, J. 1974a. 'Phonetic explanation in phonology'. In Bruck, et aI., .1974: 251-274. _______ 1974b. 'Experimental historical phonology.' In Anderson and Jones, eds., 353-389 .• Olson, H.W. 1964. The Phonology and Morphology of Rimi Ph.D Thesil, Hartford Seminary Foundauou .. Paul, H. 1891 (1888). Principles of the History of Language. London: Longmans. Ponelis, Fritz ~974. 'On the dynamics of velarization and labialization: some Bantu evidence' Studies in African Linguistics 5, 1:27-58. Posner, R.R. 1961: Consonantal Dissimilation in the ~omance Languages. Oxford: Blackwell. Ralph, Bo 1977. 'Rule Extension in Hiltorica1 Phonology'. In: Studio Linguistica XXXI, II: 164-191. Rehse, H. 1912-13. 'Die Sprache der Baziba in Deutsch-Ostafrika'. In: Zeitschriftfuer Kolonial- sprachen. Vol. 111: 1-33, 81-123, 201-229. Berlin. Richardson, I. 1966. 'A Vocabulary of Sukuma,' ed. by M. Mann, African Language Studies VII: 1-79. Robinson, O. and F. van Coetsem 1973. 'Review of King 1969' Lingua 31: 331-369. Romaine, S. 1983. 'Historical linguistics and language change: progress or decay? (Review) Lan- guage in Society 12:223-237. Rop, A. de 1958. Elements de phonetique historique du Lomongo. Leopoldville. Ruhlen, Merritt 1976. A Guide to the Languages of the World Standford University Language Universals Proiect. Saussure, F. de 1959 (l9i6). Q)urse in General Linguistics. New York: Philosophical Press. Schachter, P. and Fe T. Otanes 1972. Tagalog Reference Grammar. Los Angdes: Univ. of California Press. Schregel, W. 1913. 'Abriss einer Grammatik cler Kinyaturu-sprache', in Mitteilungen des Seminars fur orientalische Sprachen, Vol. XVI. 60-62. Berlin. Siavikova, M. 1975. Relalonships among Six N.E. Bantu Language Ph. D. Thesis, Umv. of London. ______ and M.A. Bryan 1973. 'Comparative Bantu: The Case of Two Swahili Dialects'. African Language Studies 14: 53-81. S0AS, London. Stigand, C.H. 1915. A Grammar of Dialect Changes in the Kiswahili Language. Cambridge: At the Univ. Press. Sturtevant, E.H. 1971. Linguistic Change. Chicag'). Univ. of Chicago Press. _____ 1947. An Introduction to Linguistic Science. New Haven: Yale Univ. Press. Taylor, C. 1959. A Simplified Runyankore-Rukiga-English and English-Runyakore-Rukiga Dic- tiontry. Kampala: East African literature Bureau. Tucker, A.N. 1929. The Comparative Phonflics of the Suto-Chuana group of languages. Long- mans, Green and Q). _____ and E.O. Ashton 1942. 'Swahili Phonetics'. Reprint from African Studi2S, Vol. I, 2:77-103; 3:161-182. _____ and M. Bryan 1970. 'Tonal cllL'lsification of Nouns in Ngazija' African Longuagu Stuaes, 11:351-385. Velten, C. 1901. Grammotik des Kinyamuesi. Gottingen: Vandenhoeck und Ruprecht. Vendryes, J. 1925. Language. New York: Knopf. Vennernann, T. 1972b. 'Phonetic analogy nd conceptual analogy'. In: Vennemann and Wilbur 1972: 181-204. _____ and. T. Wilbur 1972. Schuchardt, the neo-grammarions and the tronsformational theory of phonological chtmge. Frankfurt a.M. 19 Vin~ent, N. 1978.\ 'Is sound ch.ange teleologi~?'. I~: ~i~iak 1978: 409-430. Whiteley, W.H. llfd A.E. Gutkind, 1954. A LmgulStlC BlblOgraphy of East Africa. Kampala: East African Swahili Committee and the E.A. Institute of Social Welfare. Wright. J. 1905. The English Dialect Grammar. Oxford. Zwicky, A. 1972. 'Note on a JitonologicaI Hierarchy in Englisr'. In: Stockwell and MacCaulary 1972: 275-301. ______ 1973. 'Lin~tics as Chemistry: The Substance Theory of Semantic Primes'. In: S. Anderson and p. Kiparsky,"ed., A FestschriftforMorris Halle. New York: Holt, 1973: 467-485. 20