Show Summary Details

Page of

PRINTED FROM OXFORD HANDBOOKS ONLINE ( © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice).

date: 19 March 2019

Bantu Phonology

Abstract and Keywords

This chapter surveys the main phonological phenomena found in Bantu languages. One such phenomenon is vowel harmony, where suffixes typically alternate in height depending on the preceding vowel. Combinations of nasal plus consonant are also frequently subject to various modifications; such clusters play an important role in debates over syllable structure, since Bantu languages seem to exhibit the peculiarity that the only onset clusters of consonantals are composed of a nasal followed by an obstruent, reversing the more general linguistic tendency for sonority to rise in the syllable onset. The tone systems of Bantu languages are especially well known for their complex systems of alternations in the form of spreading, dissimilation, and grammatically conditioned melodic alternation. A key feature of interest in the study of Bantu phonology is the considerable variability in the exact details of operation of a highly similar set of rules.

Keywords: phonology, tone, vowel harmony, nasal, syllable, compensatory lengthening, Bantu languages, nasal–consonant clusters

There are around 500 Bantu languages in Africa, spoken in an area ranging from near Nigeria to southern Somalia and south to the tip of the continent. Bantu languages are conventionally classified into zones designated by letters A through S, following criteria and an enumeration of languages in Guthrie (1967–71). These languages are clearly related to neighboring language groups also termed “Bantu,” such as Jarawan, Mamfe, Grassfields, and Ekoid, and constitute a node on a highly branching genetic tree (one constantly being revised) under increasingly broader subgroups such as Bantoid, Benue-Congo, Volta-Congo, and Niger-Congo. See Williamson & Blench (2000) for the position of Bantu within Niger-Congo. Maho (2003) gives a contemporary version of the Guthrie classification of Bantu. Relationships of Bantu languages to one another are quite controversial and unsettled: see Bennett & Sterk (1977), Bastin et al. (1999), and Nurse & Philippson (2003) for proposals.

The reconstructed phoneme inventory of proto-Bantu is simple, being composed of the consonants *p *t *c *k *b *d *g *j *m *ɲ, the vowels *i *ɪ *e *a *o *ʊ *u, and H and L tone (L typically being unmarked). Classical works on Bantu phonological reconstruction are Meinhof (1932) and Guthrie (1967–71). The phonetic values of *c *j are difficult to ascertain, but may have been [s z] or [s j], and *d probably allophonically varied with [l]. The phonetics of the vowel system is open to debate; while it is clear that there are three distinctive vowel heights or degrees, it is not clear whether the phonetic vowels are closer to [i ɪ e a o ʊ u] or to [i e ɛ a ɔ o u]. In synchronic descriptions it is often difficult to assign the second-degree vowels ɪ ʊ firmly to canonical [ɪ ʊ] or to [e o]—the third-degree vowels are in fact most often closest to [ɛ ɔ]. Complicating matters, there is a tradition of reconstructing and describing Bantu with the letters <i̧ i e a o u u̧> (Guthrie 1967–71, Meeussen 1967), where <i̧ u̧> are “super-close” or “super-high,” a property intended to reflect a propensity for the highest vowels to spirantize a preceding stop. Syllables are presumed to have the canonical shape (N)CV(V), with distinctive vowel length and one tone per syllable. NC onset clusters may also be considered to be single phonemes (Schadeberg 2003).

Synchronic inventories tend to preserve these patterns, though the distinction *i, *u versus *ɪ, *ʊ is frequently neutralized to [i, u], and vowel length distinctions are frequently lost. Consonant inventories, on the other hand, tend to expand, especially in connection with the reduction of the 7-vowel system to a 5-vowel system; thus Shona has 35 consonants, and Xhosa has 18 clicks plus 44 nonclick consonants (not counting NC clusters).

Four areas of Bantu phonology are reviewed here. Very many languages have some version of vowel height harmony, discussed in section 1. It is also common to find significant alternations pertaining to NC clusters (section 2). Processes of vowel hiatus resolution, especially involving compensatory lengthening, have played a significant role in the development of theories of syllable structure, as seen in section 3. Bantu languages are well known for their highly variable and complex tonal systems, considered in section 4.

An understanding of Bantu phonology is facilitated by an understanding of Bantu morphology. Meeussen (1967) gives a detailed reconstruction of proto-Bantu grammatical morphemes. Nouns have somewhat arbitrary genders, which are marked with singular versus plural class prefixes, conventionally numbered from 1 to 22 or higher and generally paired so that nouns with a singular in cl. 1 have a plural in cl. 2, and so on. The probable class prefixes and gender pairs of proto-Bantu are given in (1).

  1. (1)Bantu Phonology

Verb morphology is especially rich but also highly variable intense inflection across languages. Example (2) gives a template of the main features of productive verb morphology, with an instantiation in Karanga Shona.1

  1. (2)Bantu Phonology

Prefixes have the shape (C)V-, roots are CVC-, and suffixes are -VC- except for the final vowel, which is V—in other words, the morphology conspires to guarantee that syllables are open.

The root and extensions exhibit the greatest degree of phonological coherence and interaction, and this grouping is often termed the Derivational Stem (DS). The DS then groups with the FV, which expresses tense distinctions, to yield the Inflectional Stem (Downing 1997). Addition of OPs results in a higher domain known as the Macrostem, and finally other prefixes join with the Macrostem to form the fully inflected verb. Proto-Bantu roots may have all been monosyllabic, as are the vast majority of synchronic roots.

The combinatorial riches of Bantu morphology are directly relevant to how the phonology works, because paradigmatic variation in the sounds of morphemes and control over environmental variables can usually be accomplished easily. Thus a possible rule of postnasal voicing can usually be motivated by looking at the 1s vs. 1p present with the prefixes /n-/ vs. /tʊ-/. Sometimes, though, such evidence is not available thanks to reanalysis of grammatical morphemes (e.g. the 1s prefix is /ndi-/ in Shona), in which case a considerable amount of evidence for a phonological process can be lost.

An extensive bibliography of Bantu linguistics is Maho (2009).

1 Vowel harmony

A widespread characteristic of Bantu phonology is vowel height harmony (broadly construed). See Hyman (1999) for more details on historical reconstruction and complications. While any vowel quality can appear in the first root syllable, affixes draw from a more restricted vowel inventory. Typically, affix vowels distinguish only three qualities: [a] and a front/back pair not of the third degree of height, that is, [i u], [ɪ ʊ], or [e o] (but not [ɛ ɔ]), depending on the language. The FV affix is usually drawn from /i~ɪ/ for negation, /ɛ~ɪ/ for subjunctive, and /a/ otherwise. Rules of vowel harmony frequently give rise to surface vowel types in noncanonical positions, such as [ɔ] outside the first root syllable.

A very common historical change is reduction of the 7-vowel system to 5-vowel [i e a o u] (where <e o> may be [ɛ ɔ]). In 5-vowel languages, one almost universally encounters a height-harmony rule whereby suffixal /i/ becomes [e] after [e,o], as exemplified by Shona (Fortune 1955, Beckman 1997) with the applied suffix /-ir-/.2

  1. (3)Bantu Phonology

Seven-vowel languages have the potential for greater variation in vowel harmony. Harmony in Matumbi (Odden 1996) points to two parameters of variation: how many heights a suffix assimilates to and what value a vowel has in the absence of harmony. As we see in (4), the front vowels of the passive and causative suffixes agree completely in height with the preceding nonlow vowel; after a low vowel, suffixes have the highest vowel, [i], which can be assumed to be underlying.

  1. (4)Bantu Phonology

The same harmony arises in the southern Tanzanian languages Matengo (Yoneda 2000) and Ndendeuli (Deo Ngonyani, p.c.), languages with [i e ɛ a ɔ o u]; the latter language is exemplified in (5).

  1. (5)Bantu Phonology

Matumbi, Ndendeuli, and Matengo are relatively close historically and share a number of unusual features.

A variant of this harmony pattern is found in Kinga (Schadeberg 1971 and p.c.), Vwanji (Helen Eaton, p.c.), and Malila (Kutsch Lojenga 2009), also spoken in southern Tanzania, where vowels after a have degree-2 height ([ɪ]).

  1. (6)Bantu Phonology

Outside of southern Tanzania, multiple-height harmony is found in Komo (Thomas 2011), spoken in the Democratic Republic of Congo, and Lugungu (Diprose 2007), spoken in Uganda. Along with Malila, these languages also have bidirectional Advanced Tongue Root (ATR) harmony, which affects prefixes and which allophonically derives [e o] from /ɛ ɔ/; in all of these languages, suffixes with alternating vowels have [ɪ] when harmony does not apply.

The most widespread Bantu harmony variant in 7-vowel languages is single-height harmony, exemplified by Kikuyu (Peng 2000) in (7) and Nyamwezi (Maganga & Schadeberg 1992) in (8), where suffix vowels have [ɛ] after [ɛ ɔ], and degree-2 vowels otherwise—[e] in Kikuyu and Kamba, and [ɪ] in Nyamwezi, Sukuma, Lega, Rimi, and Langi.3

  1. (7)Bantu Phonology

  1. (8)Bantu Phonology

An interesting complication in harmony is that while the front vowel harmonizes to any vowel, the back vowel usually does not harmonize with e. Virtually all 5-vowel languages with harmony exhibit this backness asymmetry; (9) gives examples in the single-height harmonizing language Bukusu (Mutonyi 2001), which has 5 vowels.

  1. (9)Bantu Phonology

In 7-vowel single-height harmonizing languages (Kikuyu, Nyamwezi, and numerous other languages), the corresponding restriction is that only /ɔ/, and not /ɛ/, triggers vowel harmony.

  1. (10)Bantu Phonology

There is variation in the backness asymmetry among languages with multiple-height harmony. In Ndendeuli (11) and Matengo (12), only back vowels trigger height harmony on the round vowel, which totally assimilates to the height of the preceding vowel. When not harmonized, the back vowel appears as [u], as it does after the nontrigger vowel /a/.

  1. (11)Bantu Phonology

On the other hand, in Matumbi (13) and Kinga (14), as well as Vwanji and Lugungu, only /ɛ/ among the front vowels is excluded from the class of triggers of harmony on back vowels.

  1. (12)Bantu Phonology

  1. (13)Bantu Phonology

The only difference between Matumbi and Kinga (plus Vwanji) in their harmony systems is that the nonharmonizing vowel has degree-1 height in Matumbi but degree-2 height in Kinga.

Though the front/back asymmetry in height harmony is quite widespread, it is not universal, and especially in 7-vowel languages of zones A–C and adjacent languages (Hyman 1999), symmetrical harmony is found. For example, Nkundo lowering harmony applied to /o/ is triggered equally by [ɛ] and [ɔ] (Hulstaert 1965).

  1. (14)Bantu Phonology

Progressive stem-internal height harmony thus exhibits a substantial degree of parametric variation.

There are other kinds of height harmony in Bantu. In the languages of western Lake Victoria, one frequently encounters multiple height-harmony rules. In Kuria (Chacha & Odden 1998) there is progressive stem-internal lowering from degree 2 to degree 3 (oko-rɔ́g-ɛ́r-ǎ ‘to bewitch for’; cf. ugu-súraaŋg-ér-a ‘to praise for’), regressive stem-internal lowering from degree 1 to degree 2 (ogo-séék-ér-á ‘to close for’; cf. ugu-síík-á ‘to close’), regressive raising in all domains from degree 2 to degree 1 (umu-rím-i ‘farmer’; cf. oko-rém-a ‘to farm’), and regressive stem-internal raising from degree 3 to degree 2 (omo-róg-i ‘witch’; cf. oko-rɔg-a ‘to bewitch’).

Closely related Ikoma (Higgins 2012) has prefixal height disharmony and multiple progressive and regressive height harmonies, and is of interest because of rampant asymmetries between front versus back vowels in the harmony system. In (16), prefixes show regressive raising disharmony from degree 2 to degree 1, triggered by the lowest vowels [ɛ ɔ a], which explains the distribution of [o] vs. [u] in the infinitive prefix /ko-/. The applied suffix is /-ɛr-/, contrary to the general pattern that the applied has an underlying degree-2 vowel. In the first column of (16), that suffix vowel raises to degree 2 after a degree-2 vowel—contrast this to the usual Kikuyu-style harmony (see (7)), where degree-2 /e/ lowers to degree 3 only after a degree-3 vowel. The subjunctive FV, on the other hand, is /e/ rather than more typical Bantu /ɛ/ and causes raising of preceding root /ɛ/ to [e], but it is itself lowered to [ɛ] after /ɔ/ (second column). Regressive raising of /ɛ/ is also triggered by the degree-1 causative and passive suffixes /i/ and /u/.4

  1. (15)Bantu Phonology

It is hard to characterize the behavior of the suffixes *-ʊk-, *-ʊl- in terms of phonetically natural processes. After the vowels [u a], [u] appears; after [i e o], [o] appears; after [ɔ], [ɔ] appears.5

  1. (16)Bantu Phonology

Higgins (2012) presents acoustic evidence that /ɛ/ and /e/ completely neutralize to [e] under this regressive raising process and also shows that /ɔ/ is somewhat raised, but without neutralization. In other words, the unphonologized seeds for symmetrical harmony are present in this language.

Regressive harmonies assimilating height in the prefix domain are found predominantly in more northern Bantu languages, especially in zone C, and exist in Kuria, Gusii, Logoori, Lika, Nkundo, Ntomba, Nande, Koyo, Mokwe, Elembe-Nkutu, Tetela, and Malila. In a number of instances, regressive harmony gives rise to an allophonic 9- or 10-vowel system where /i ɪ ɛ a ɔ ʊ u/ expands to [i ɪ e ɛ a ɔ o ʊ u] for instance in Nande (Valinande 1984, Gick et al. 2006).

Finally in the realm of height harmonies, Sotho, Basaa, and Nzebi have what Parkinson (1996) terms “one-step” raising harmony, where vowels are (regressively) raised by one degree. This process is exemplified by raising in Nzebi, which is conditioned by the tense suffix -i associated with the “perfective stem” in Bantu.

  1. (17)Bantu Phonology

A different type of harmony, complete harmony, is found in a number of languages in the northwest zone C area (Leitch 1997). Here the vowel /a/ assimilates completely to preceding /ɛ ɔ/, as exemplified in (19) with the final suffix /-a/ and the passive /-am-/ in Babole.

  1. (18)Bantu Phonology

2 NC clusters

NC clusters, which are usually one of the only two consonantal clusters in Bantu languages, have some interesting phonological properties. Roots generally do not begin with NC clusters (and in proto-Bantu, C is nearly always voiced in tautomorphemic NC clusters—reconstructed roots like *ntu ‘person, thing’ are extremely rare). NC clusters can arise by combining the prefix N- for classes 9 and 10 with a subsequent consonant, and in numerous languages if not in the protolanguage, the verbal prefix for 1s (subject or object) has the shape n or ɲ. Consequently, it is generally easy to combine a nasal with any consonant that can appear at the beginning of the stem, so rules pertaining to NC sequences could easily be learned from surface alternations.

2.1 Nasal + voiced C

One pattern of alternation probably should be attributed to proto-Bantu, namely, the hardening of approximants to voiced stops after a nasal. Commonly, voiced stops and fricatives or approximants are in an allophonic relation, with [b d g] appearing after a nasal in the syllable onset,6 and voiced continuants appearing after vowels and word initially (i.e. syllable initially). Kuria exemplifies this relationship.

  1. (19)Bantu Phonology

This allophonic relation between voiced stops and fricatives is widespread enough that there are competing theories as to what the proto-Bantu consonants fundamentally were—[β l γ] in the reconstruction of Meinhof (1932), but [b d g] according to Guthrie (1967–71).

One frequently finds a synchronic relationship between postnasal voiced stops [b d g] and a related voiced sound such as an approximant, fricative, or implosive. For example, in Matumbi (also Mushunguli, Hehe, and dialects of Pare), voiced stops are regularly implosive unless preceded by a nasal.

  1. (20)Bantu Phonology

Implosive ɗ is not the regular correspondent of proto-Bantu *d in Matumbi; it is actually a rare phoneme found only in loanwords. The etymological correspondent of *d is [l], which, along with the glides [w] and [j], undergoes postnasal hardening.

  1. (21)Bantu Phonology

In other words, the original situation may have involved lenition of voiced stops unless preceded by a nasal, but subsequent addition of new phonemes has led to reanalysis of the process as postnasal hardening.

The voiced NC clusters [mb nd ŋg] are generally not subject to phonological change. An exception is the Makua group of languages, where proto-Bantu NC clusters underwent deletion of the nasal, and voiced stops were devoiced, so proto-Bantu *mbudɪ appears in Makua as epúri ‘goat’ and *ŋgoma appears as ekóma ‘drum’. Since all original sources of NC sequences have been reanalyzed (the cl. 9–10 prefix is /e-/, and the 1s verbal prefix is /ki-/), there is no longer a synchronic basis for the earlier nasal-dropping rule.

In dialects of Kikuyu,7 historical NC clusters (C is always voiced) simplify to C; for example, orthographic kũndora ‘to look at me’ is dialectally [koodɔra]. This results in a somewhat perplexing surface alternation: after the prefix for a 1s object, often assumed to be /n/, root-initial /β r γ/ become [m n ŋ] if a voiced stop (oral or nasal) follows in the next syllable; otherwise, oral consonants (excluding /ð h/) become voiced stops. The alternation is historically explicable: [b d g] derive from mb nd ŋg, and a common rule in Bantu, Meinhof’s Law, dissimilated NCVN to CVN.

  1. (22)Bantu Phonology

It is not clear whether there is sufficient synchronic motivation for claiming that the 1s. prefix is /n/, since its only manifestation is as voicing or nasalization of the following consonant and lengthening of the preceding vowel.8

Deletion of prefixal nasals before underlyingly voiced consonants, though not ones derived from voiceless consonants via postconsonantal voicing, is also found in Mbunga.

  1. (23)Bantu Phonology

The Sotho-Tswana subgroup has an active synchronic process of postnasal devoicing, which poses a problem for theories of phonology holding that phonological processes must be phonetically “natural.” See Hyman (2001b) for discussion of such facts in Tswana; (25) illustrates the process with data from Shekgalagari (Solé et al. 2010).

  1. (24)Bantu Phonology

2.2 Nasal + voiceless C

Voiceless postnasal consonants are frequently subject to allophonic aspiration, as exemplified by the adjectival alternations of Mushunguli (the Zigua dialect of Somalia).

  1. (25)Bantu Phonology

Postnasal aspiration is found in a number of eastern Bantu languages, including Pokomo, Shambaa, Pare, Kagulu, Gogo, Sukuma, Kongo, Kwanyama, Pogolo, Rwanda-Rundi, Tumbuka, and Chewa. The pronunciation of such clusters varies considerably among the languages with this process. The nasal may be partially devoiced (as it is in Mushunguli), and the oral stop may be reduced to the point of deletion so that /mp/ → [m̥], as reported for Kagulu and Gogo.9 In Shona, /mp nt/ become breathy-voiced /m̤ n̤/, and /ŋk/ becomes [ɦ].

  1. (26)Bantu Phonology

The contrary process of postnasal deaspiration is encountered in Nguni languages such as Zulu (Doke 1954, Doke & Vilakazi 1972).10

  1. (27)Bantu Phonology

Another very common rule affecting voiceless stops after nasals is voicing, illustrated in (29) with data from Matumbi. Postnasal voicing is found in Ngindo, Yao, Mwera, Nande, Ndali, Fuliiru, Tembo, Kikuyu, Kamba, Chaga-Taita, nearly all Luyia languages, Herero, and Bangubangu.

  1. (28)Bantu Phonology

2.3 Postnasal fortition and reanalysis

Consonants not preceded by a nasal are frequently subject to diachronic lenition, so earlier *s may become h, *p may become f or h, and so on. Such lenitions are generally unstable and frequently undergo reanalysis as postnasal fortition, especially when new sources of the lenited phoneme enter the language. A typical example is h in Kerewe, which in native words comes from *p, probably via f (which exists in the language only before u and w). As seen in (30), roots can contrast underlying initial /p/ and /h/, and the underlying form is preserved word initially or postvocalically. After a nasal, the distinction between /p/ and /h/ is neutralized to [p].

  1. (29)Bantu Phonology

Given that the choice of [p] versus [h] is not synchronically predictable, the underlying forms of the above roots must contrastively contain both /h/ and /p/. In Kerewe, /p/ as an autonomous phoneme in syllable-initial position is not common and can usually be attributed to the loanword status of words that it appears in.11 So even though the historical source of the h ~ p alternation in Kerewe is a rule where *p → h, the contemporary language has reanalyzed the process as one strengthening /h/ to [p] after a nasal.

An analogous example involving the historical change of *s to [h] in Ntomba (Mamet 1955) reveals that in this language, as a result of new instances of s (which, again, are not numerous in the language), the original lenition process has been reanalyzed as one strengthening /h/ to [s] after a nasal.

  1. (30)Bantu Phonology

A striking example of multiple reanalyses is found in Nyole (Musimami & Diprose 2012). The cl. 10 prefix in this language is /eN/, as is evident from (32a–b). Although Bantu did not contrast [b] and [β], [d] and [l], or [dʒ] and [j], Nyole synchronically contrasts all of these, with /b, d, dʒ/ deriving from secondary sources (e.g. oludʒegere and oludaŋgadi are loanwords from Luganda). Fortition of voiced approximants after a nasal can be seen in (32a). In (32b) we see that synchronic /k/ and /h/12 neutralize to [k] after a nasal, where the nasal deletes before voiceless consonants. The historical source of Nyole [h] is Bantu *k, which lenited to h unless preceded by a nasal; thus the appearance of [k] after a nasal is understandable. However, since k and h now contrast (compare [ohukondola] ‘to commit burglary’, [ohuhondola] ‘to pluck’), what was once “lenition except after a nasal” has become “fortition after a nasal.” In the same vein, in (32c) we see that /ŋ/ and /p/ neutralize to [p] after a nasal. This happens because Bantu *p became /ŋ/ (via intermediate h according to Schadeberg 1989). Since [p] has been reintroduced as a separate phoneme in the language, the synchronic analysis has changed from /p/ → h (except after a nasal) → [ŋ] to /ŋ/ → [p] after a nasal.

  1. (31)Bantu Phonology

2.4 Multiple nasal effects

In most Bantu languages, a single set of postnasal effects is triggered by the prefixes for 1s, cl. 9–10, or possibly the presentative /n/. A number of languages have also introduced a secondary set of NC clusters derived from reduction of prefixes of the shape /mu/ (cl. 1, 3 locative mu, cl. 1 OP and 2 p. SP) before a consonant. Example (33) illustrates such reduction of the cl. 3 noun prefix /mu-/ in Pare. In this case, the reduced nasal constitutes an entire syllable.

  1. (32)Bantu Phonology

Such reduction may be obligatory, as in Swahili, Shambaa, Pare, Pogolo, and Chaga; optional, as in Hehe; or properly part of phonetic implementation since the vowel is not fully deleted, as in Kerewe. Before labial stops, this process can lead to syllabicity minimal pairs such as Swahili [m̩búni] ‘coffee tree’ vs. [mbúni] ‘ostrich’.

Usually, the reduced prefix is invariant syllabic [m̩] that does not interact with the following consonant. In some languages, though, the reduced nasal does interact phonologically. In the Rufiji-Ruvuma languages of southern Tanzania (Odden 2003) there are competing NC effects, depending on whether the nasal is the cl. 9–10 prefix on nouns and adjectives or the 1s SP or OP on verbs, on the one hand, or underlying /mu/ (cl. 3, 4, 18 NCP [noun class prefixes], cl. 1 OP, 2p SP, OP), on the other. In Matumbi, after /n(i)/, glides and liquids harden to voiced stops (34b), nasals degeminate (34c), and voiceless consonants voice (34d), whereas after reduced /mu/, all voiced consonants nasalize (34a–b), geminate nasals remain (34c), and voiceless consonants are unchanged (34d). The forms in (34) have undergone optional vowel deletion in the subject prefix: alternative forms are, for example nibájite ‘I said’, mubájite ‘2p said’.

  1. (33)Bantu Phonology

Similar effects are found in Makonde and Yao.

3 Syllable structure

There are two main issues of recurring interest in Bantu phonology pertaining to syllable structure, namely, the syllabic status of consonant clusters and the resolution of vowel hiatus. See Hyman & Katamba (1999) and Odden & Odden (1999) for descriptions of the syllable in Ganda and Hehe, respectively.

3.1 Consonant clusters

As a general rule, Bantu languages do not have obvious syllable codas (see below for NC clusters), though Sotho languages among others do allow some word-final nasals as a result of vowel deletion applying to /ni/ and /mu/, and more general word- and syllable-final consonants are possible owing to historical and sometimes synchronic vowel deletions in a number of zone A and B languages, such as Basaa and the Beti-Fang languages (e.g. Ntumu mvʌ́t ‘guitar’), as well as a collection of languages spoken in the Democratic Republic of Congo, such as Ruwund, Hungan, and Kanyok. As a result of contact with Afro-Asiatic languages, word-internal clusters that are clearly not onset clusters are found in some eastern Bantu languages (Mushunguli barshi ‘pillow’, Mwiini k-uβl̪a ‘to kill’,13 Swahili γafla ‘suddenly’).

Aside from consonant–glide (CG) and NC clusters, there are generally no onset clusters, though they do arise via deletion of prefixal high vowels in Pogolo (l-tambi ‘branch’, ʃ-kombi ‘cup’) and Mwiini (x-finika ‘to cover’, l-piindo ‘edge’). Ganda has developed geminate consonants, which can appear word initially (ggáalí ‘vehicle’, vvíívî ‘knee’; see Muller 2002 for theoretical analysis).

The two kinds of (apparent) consonant clusters that have a central status in Bantu, namely, homorganic NC sequences, often termed “prenasalized consonants,” and CG sequences, can overlap to yield triconsonantal NCG sequences. Typical examples are Matumbi mbaáŋgo ‘warthog’, bwoóga ‘mushroom’, and mbwá ‘dog’. The analysis of such clusters has proven to be nontrivial. Rather than being treated as bisegmental clusters, supposed CG sequences are occasionally analyzed as secondary articulations on single consonants. Hargus & da Conceição (1999) propose that Ronga has distinctively labialized consonants (nwala ‘fingernail’), rejecting a cluster analysis on the grounds that there are no other onset clusters. Likewise, van Otterloo (2011) treats Fuliiru’s potential clusters of the type pj, kw as involving secondarily articulated palatalized and labialized consonants [pj, kw]. One can find competing analyses of such clusters: Tucker (1929) proposes that the Sotho-Tswana languages have labialized consonants, and Kunene (1961) holds that these languages have CG clusters. Herbert (1986) discusses criteria applied to the unit/cluster debate and pre-autosegmental theories of representation, especially in Bantu.

The main argument for a bisegmental treatment of such sequences is that they frequently have an obviously bisegmental synchronic source, namely, glide formation applied to high vowels. For instance, the cl. 5 prefix of Matumbi /li-/, seen in li-nóolo ‘sharpening stone’, surfaces as [lj] before a vowel, as in lj-óoli ‘tear’. This CG sequence is a consequence of a general glide-formation rule whereby /u,i/ becomes [w,j] before a vowel, uncontroversially observed in the alternations of the cl. 8 prefix /i-/ (cf. [i-wó] ‘deaths’, [j-ɪɪ́mbe] ‘knives’). Such patterns would seem to challenge the secondary-articulation theory of CG sequences, because under that theory such sequences would necessitate a further complication whereby, for example, derived ljóoli becomes [ljóoli].

Another argument for a cluster analysis is that multiple secondary articulations within a segment are unordered, whereas there can be evidence for ordering of multiple glides. As seen in (35a), in Kerewe the vowels /i u o/ always become glides before another vowel, even between words. Under the no-cluster theory of CG sequences, this change involves converting palatality and labiality into a secondary articulation on the preceding consonant. A word-final vowel may itself be preceded by a CG cluster, as in (35b), where we would also expect glide formation in the consonant-cluster theory of CG formation, and this results in contrasting orders of segments in the intermediate form: endosjw íísátu vs. ekizwj éétʃo. A glide never follows w in Kerewe, and j deletes in the surface form [ekizw éétʃo].

  1. (34)Bantu Phonology

The problem for the secondary-articulation theory is that the differential treatment of intermediate Cjw vs. Cwj requires an ordering between the two vocalic elements, which is generally held to be impossible within a single segment. Since it is very hard to generate consecutive glides within a word in Bantu, and most languages do not have phrase-level glide formation, this argument cannot be made very often.

Turning to NC sequences, there are two separate issues, namely, whether NC is a single segment or a cluster, and if it is a cluster, how the components syllabify. Herbert (1986) and Downing (2005) discuss issues and evidence relating to these problems. A basic reason to treat [nt nd] and similar sequences as bisegmental is that they are frequently bimorphemic and arise by concatenation of an autonomous nasal with another consonant. As we have seen in (34), Matumbi [mb] in [mbájite] ‘I said’ derives from /ni+bájite/, which is optionally realized as [nibájite], and [mp] in [mpáaŋgite] ‘2p did’ derives from /mu+páaŋgite/, likewise optionally realized as [mupáaŋgite]. A basic reason not to treat [nt nd] and the like as bisegmental is that doing so would endow Bantu languages with typologically uncommon syllable structure, with onset clusters that violate the Sonority Sequencing Principle (Sievers 1881, Clements 1990). Although authors may employ the term “prenasalized stop” to refer to NC (e.g. Heath 2003:336 writes, “Makaa has twenty-two simple consonants, plus eight prenasalized stops …,” and Alnet 2009:47 lists a series of “pre-nasalized” consonants in Shimaore), it is generally not clear what representational claim they are making. Sometimes the theoretical interpretation is clear: Clements (1986) refers to a process of “prenasalization” in Ganda that prosodically realigns the nasal segment of a preconsonantal nasal cluster so that it shares a C-slot with the following onset consonant.

  1. (35)Bantu Phonology

A recurrent and puzzling fact about NC sequences in Bantu is that vowels are frequently lengthened before them. The classic example of this is Ganda, where vowels are always long before NC clusters, as exemplified in (36) by muuntu. Such a length regularity holds in numerous languages, such as Holoholo, Luyia, Kerewe, Nyambo, Haya, Kamba, Kikuyu, Kuria, Sukuma, Yao, Hehe, Lungu, Bemba, and Ruwund. The theoretical puzzle is just what mechanism accounts for this regularity. The standard analysis since Clements (1978, 1986) has been to assume that a preconsonantal nasal has a special prosodic status such that it is dominated by V rather than C or, translated into moraic terms, that preconsonantal nasals are moraic. This status would normally result in the nasal being syllabified into the coda of the preceding syllable, but the fact that words do not end with any consonant14 is taken to argue against positing nasals in the coda. The nasal is then syllabified into the onset of the following syllable, which leads to compensatory lengthening of the preceding vowel by reassociation of the stranded timing unit.

As Herbert (1986) and Downing (2005) point out, this analysis depends crucially on assuming that the nasal in a /VNC/ sequence must be in the onset, as has been assumed in nonlinear analyses (Clements 1986, Bickmore 1989, Downing 1990). Pre-NC lengthening is then treated as compensatory lengthening resulting from the fact that moving the nasal into the onset deprives it of its V-slot or mora. Further assumptions have to be made to explain the special timing slot on the nasal, especially if nasals are not syllabified in the coda. In the Clements (1986) analysis of Ganda, it is simply an unexpressed generalization that nasals are dominated by V if they precede C in underlying forms. Bickmore (1989) explains pre-NC lengthening in Nyambo by making a preconsonantal nasal be a full syllable (hence moraic) and then restructures the nasal to be in the onset of the following C. For Jita, Downing (1990) postulates a rule assigning a mora to any consonant sharing place features with the following consonant.

The phonological evidence for onset status of N in NC is not absolutely clear. The only phonological argument that N is in the onset is that no words end in a consonant, even a nasal, which leads to the conclusion that syllables cannot end with consonants. The theoretical validity of that argument is challenged by languages with stricter conditions on word edges than word internally. For example, Chadic Kotoko disallows all word-final obstruents and inserts a schwa when such codas would arise ([hàm-à] ‘yawn (inf.)’, [hám] ‘yawn (past)’ but [làb-à] ‘tell (inf.)’, [làbə̀] ‘tell (past)’), while allowing word-internal obstruent codas ([àbγwàj] ‘stomach’, [bə̀tkúm] ‘punctured’, [àksùm] ‘rat’). Following the Coda Filter approach of Ito (1989), there should be a language that strictly prohibits autonomous place specifications in the coda without the mechanism of final-exceptionality. This perfectly describes Bantu syllable structure: place specifications in the coda are allowed only when they are shared with the onset.

Drawing on an idea from phonetic studies of duration in Bantu VNC sequences (e.g. Hubbard 1995), Downing (2005) posits that long vowels before NC share a mora with the nasal and that the moraicity of the nasal derives from a version of “Weight By Position.” Other Optimality Theory–specific machinery then guarantees that the nasal’s mora associates to the preceding vowel. The resulting surface form would be something like (37), a Nyambo example from Hubbard (1995).15

  1. (36)Bantu Phonology

3.2 Vowel hiatus

It is very common in Bantu languages for morpheme concatenation to give rise to underlying sequences of vowels that are partially or completely eliminated at the surface. At one end of the logical continuum of hiatus resolution, Kerewe has absolutely no vowel sequences,16 even at the phrasal level. Underlyingly, such sequences arise by combination of vowel-final prefixes with vowel-initial roots or prefixes or by combination of any word (necessarily vowel final) with a vowel-initial word. Within words, if V1 is any of /i e o u/, it becomes the corresponding glide, and the following vowel is lengthened. High vowels undergoing glide formation are illustrated with noun class prefixes in (38).17

  1. (37)Bantu Phonology

Glide formation also applies to the output of optional deletion of root-initial /j/, as seen in (39). Since subject prefixes include mid vowels, this gives nonhigh vowels undergoing glide formation.18

  1. (38)Bantu Phonology

The combination of /a/ plus a vowel results in a lengthened nonhigh vowel.

  1. (39)Bantu Phonology

Sentence-level vowel sequences also are eliminated, though by somewhat different rules. The vowels /i u o/ undergo glide formation with compensatory lengthening.

  1. (40)Bantu Phonology

However, the vowels /e a/ simply delete (again, with compensatory lengthening).

  1. (41)Bantu Phonology

Bantu languages may have less thorough systems of hiatus resolution. Matumbi, for example, has a number of morpheme-internal vowel sequences (baaɪ́ ‘end’, fáidika ‘to profit’, kalaái ‘basin’, kibáo ‘stool’, kikói ‘clothing’, kiléu ‘chin’, lubáu ‘rib’). Glide formation obligatorily desyllabifies prevocalic high vowels in word-initial prefixes such as noun class and subject prefixes (43a) but is optional with a prevocalic high vowel in a noninitial syllable (43b).

  1. (42)Bantu Phonology

Glide formation is optional or blocked in a number of contexts largely having to do with H tone on the first vowel, for example, kí-ukumú ‘the Ukumus’ and niibweni mú-aandʒú ‘I saw it in the firewood’ (cf. mwaandʒú ‘in the firewood’). Contraction of /a-i, a-u/ to [ee, oo] is optional and limited to prefix vowels.

  1. (43)Bantu Phonology

There is obligatory merger of /a-a/ into a long vowel, a change that is discernible tonally, but this merger only takes place in word-initial syllables.

  1. (44)Bantu Phonology

There is virtually no phrasal hiatus reduction in the language Matumbi. Systems of hiatus reduction are analyzed in Clements (1986) for Ganda, Nash (1992) for Ruwund, Roberts-Kohno (2000) for Kamba, Bickmore (2007) for Lungu, and Mtenje (2007) for Ndali, Tonga, and Nyika.

4 Tone

Bantu languages are especially known for their often complex tone systems. The vast majority, perhaps 97%, have distinctive tone. Nearly all tonal Bantu languages are fundamentally two-level languages with one tone per syllable, frequently augmented with downstep. A few western languages (e.g. Tuki and Yabassi) that have undergone final-vowel deletion have true three-level systems. Some languages have expanded the surface inventory via contextually restricted modifications of the basic H/L contrast; for example, Kamba has developed now-contrastive “extreme” tones at the ends of words, where H becomes super-high and L becomes super-low. Some western Bantu languages (Ntumu, Kombe, and Bafia) have a contrast between prepausal falling L and nonfalling L (arising from final L vs. rise), and Swati has developed an “extra-low” from pitch-depression effects induced by preceding voiced consonants (which has become phonologized and now exists where there was no consonantal cause).

It has been recognized at least since Stevick (1969) that there is a behavioral asymmetry between surface H and L in Bantu, which frequently leads to an analysis of the system in terms of a privative H/Ø opposition rather than H/L. Hyman (2001a) provides a critical discussion of arguments for privativity. For example, the location of H is frequently quite restricted, but the location of L is free; tense inflection is typically characterized by addition of H and almost never by the addition of L. The sequence /HH/ is often dissimilated to [HL] or [LH], but /LL/ never dissimilates in like fashion. Myers (1998) thus proposes that Chewa never phonologically specifies L. Bickmore (2000) pursues an Optimality Theory analysis of Namwanga that dispenses entirely with surface L tones, even in the face of contrastive downstep,19 and Hyman & Ngunga (1994) propose for Yao that L tone is not specified except at the very end of the derivation to represent downstep and for the formulation of a late rule that changes long rising tone into level H. On the other hand, even though Stevick’s privative analysis is based on Ganda tone, that same language provides evidence for specified L being the result of H-dissimilation, since surface-L syllables block a general leftward H-spreading rule (Hyman 1982). For example, underlying /a-bá-tá-lí-láb-il-il-a bapákasi/ undergoes Meeussen’s Rule (section 4.2) becoming abátàlìlàbilila bapákasi, and then the last H spreads leftward, but only to underlyingly toneless syllables, resulting in [abátàlìlàbílílá bápákasi] ‘they who will not look after porters’. Following common practice, H is indicated here with acute accent and L—whatever its phonological status—is unmarked.

Surface-distinctive rise, fall, and level H on long syllables are not rare, but usually the position of H in a long syllable is rule governed, albeit with some opacity. In Tachoni, long penults with H have falling tone (oxutéexa ‘to cook’), and before the penult a long H syllable is level H (oxutééxela ‘to cook for’). Rising tones always become level H, so underlying /oxu-íra/ undergoes glide formation, which should yield oxwiíra, but this instead surfaces as [oxwííra] ‘to kill’. This process gives rise to a limited derivational contrast between level H versus fall on the penult. Matumbi has no long level-H tones (except in the Swahili loanword sáána ‘very’) but has both rise and fall on long vowels. The choice, however, is ultimately rule governed. Rise results from retracting final H to a long vowel (/lu-seendʒéma/ → [luseéndʒemá]20 ‘mosquito’), when a mora-counting rule targets the second half of a long vowel (H is assigned to the 3rd Macrostem mora in the subjunctive, so /u-kalaange/ → [ukalaánge] ‘you should fry’), or when a H vowel is compensatorily lengthened after a L-toned prefix vowel via glide formation (/ki-átí/ → [kjaátí] ‘family farm’). Fall is the regular realization of lexical or phrasal assignment of H to a syllable with a long vowel (/ki-bwéeja/ → [kibwéejá] ‘jackal’, /ki-goombo/ → [kigóombo] ‘load’) and also arises in the system of mora-counting tone assignment (/u-ki-kalaaŋge/ → [ukikaláaŋge] ‘you should fry it7’).

As a general rule, Bantu noun stems can exhibit the combinatorial limit on H and L freely distributed over the two syllables of canonical stems, and class prefixes have no underlying tone. Certain patterns may be rare because of historical tone changes (e.g. proto-Bantu stems with the tone pattern HL have the pattern HH in Shona, so *ŋkáta → [hátá] ‘head pad’, but HL exists in loanwords, such as [tʃikóro] ‘school’). In a few languages there are more root-governed surface contrasts in nouns than can be attributed to the number of TBUs (tone-bearing unit) in the root; for example, Shimakonde (Liphola 2001) has eight surface possibilities in disyllabic stems. On the other hand, in Kuria, noun tone is almost completely predictable except for a contrast between [CV́V́CV́] and [CV́V́CV] roots. Compared to nouns, verbs have very limited underlying contrasts but great potential for surface differences. Roots have only a two-way distinction between H and L tones regardless of length, with no underlying distinctions in where the tone is realized; some languages lack even that contrast. Prefixes may contrast H vs. L tone, but the choice often depends on the verb’s tense; for example 3rd person subject prefixes are H in Shona (átʃatórá ‘he will take’) except in subject-headed relative clause verbs (atʃátórá ‘he who will take’). Stems may have additional floating tones—“melodic Hs”—which partially mark verb tense (section 4.3).

Especially in verbs, the highly agglutinative morphology of Bantu makes possible great variation in the position of an underlying tone relative to other tones and ends of words. As a result the environment in which a tone appears is subject to easily controllable variation, often giving rise to complex tonal morphophonemics. Marlo (2013) presents an overview of the many factors at play in Bantu tonology, expressed as a questionnaire methodology for investigating Bantu verbal tone. Here we consider three factors contributing to tone complexity, namely, tone mobility, H-dissimilation, and systems of melodic tones.

4.1 Tone mobility

Very many Bantu languages have processes expanding the domain of H to the right or to the left. The most common expansion is to the right, except that final tones frequently expand to the left. Kerewe has a tone-doubling rule that spreads H one syllable to the right as long as another syllable follows.

  1. (45)Bantu Phonology

Rightward tone doubling is found as a general rule in a number of languages, such as Chewa, Makua, Yao, Kuria, Gusii, Holoholo, and Sotho-Tswana, and in specific environments in Karanga Shona, Logoori and other Luyia languages, Makonde, and Kwanyama.

Unbounded rightward spread occurs in Shambaa (not affecting word-final vowels).

  1. (46)Bantu Phonology

Unbounded spreadings, sometimes with domain restrictions (e.g. stems only or prefixes only), are found in Nguni, Shona, Lungu, Namwanga, and Digo and as part of the system of melodic tones in Kamba and Pare.

An unusual rule of tone tripling is found in Shona, where H roots spread the H to the following two syllables within the word, but no further.

  1. (47)Bantu Phonology

This process is also found in Venda, Kanye Tswana, Tsonga, and Copperbelt Bemba.

Related to tone spreading, a number of languages have rules of tone shifting. The most common form of shift is rightward shift by one syllable, illustrated in Jita (Downing 1990), where H moves to the following non-final syllable (if there is one).

  1. (48)Bantu Phonology

Such rightward shift also occurs in Holoholo, Nyamwesi, Taita, Chaga, Rimi, Kikuyu, and Kwanyama.

Long-distance tone shift is found in Zigua (Kenstowicz & Kisseberth 1990, Kisseberth 1992). In (50), H shifts from the root-initial syllable of /lómbez/ and the H subject prefix /á/ to the penult; the source of the H is underlined.

  1. (49)Bantu Phonology

Similar tone shifting is found in Digo (shifting to the final vowel) and Nguni (shifting to the antepenult).

Leftward movement of H tone is significantly less common. Leftward shift by one syllable occurs in Totela (also Nande and Tonga); thus the toneless OP /mu/ has H before the H root /bíik/, and /la/ has H before the H-toned OP /bá/ (Crane 2011).

  1. (50)Bantu Phonology

Unbounded leftward spread is found in a large set of Lacustrine languages, such as the Luyia language Logoori.

  1. (51)Bantu Phonology

Interestingly, there seem to be no attested cases in Bantu of unbounded leftward shift or of leftward doubling.

4.2 H-dissimilation

Concatenations of Hs are frequently modified by deletion (lowering) of one of the Hs or by fusing them into a single tone. As with tone movement, deletion comes in a number of varieties according to which tone is affected and the maximum distance allowed between the tones.

The most common form of tone dissimilation is deletion of H immediately after H, a process commonly termed Meeussen’s Rule (MR; Goldsmith 1982). Although this process is found in quite a number of languages, it is typically difficult to motivate directly, and usually the evidence for MR has to be filtered through an understanding of tone movements that obscure the application of MR. An example is Kerewe, which as we saw in (46) has rightward tone doubling. OPs have underlying H, as seen when they come before toneless verb roots in (53a). When an OP stands before a H root (53b), the root H is deleted, though later application of tone doubling resupplies that syllable with H. Likewise, after --, a root H or an OP H is deleted. That there is no H on the syllable after the root-initial syllable or after the OP is, in light of the doubling rule, evidence that H has been deleted. The same argument holds of an OP after another OP, seen in (53c). Even clearer evidence for deletion of H comes from the surface pattern arising when three or more underlyingly H-toned syllables are concatenated. As (53d) shows, all of the underlyingly H-toned syllables in a sequence of Hs are surface L except the first H and the immediately following H, which is due to tone doubling.

  1. (52)Bantu Phonology

Versions of this rule are found in many languages, including Tonga, Ganda, the Luyia languages, Jita, Digo, Chewa, Hehe, and some Makonde dialects. Significantly morphologized or domain-restricted versions are found in Shona, Venda, Matumbi, and Nkore.

A less frequent H-dissimilation process, sometimes termed Reverse Meeussen’s Rule (RMR), turns H into L before H. This process arises in Xhosa and other Nguni languages as well as in Rimi. In Xhosa (Claughton 1992, Cassimjee 1998), OPs have underlying H, seen before a monosyllabic L root as in [nija-wáá-lw-a]21 ‘2p are fighting them’. When prefixed to a long-enough L verb, the H of the OP shifts to the antepenult, so /nija-wá-ʃukumisa/ → [nijawaʃukúmiisa] ‘2p are shaking them’ (cf. [nijaʃukumiisa] ‘2p are shaking’), because of a general shift of pre-antepenultimate H to the antepenult. If H appears underlyingly in the antepenult, it shifts to the penult; thus /nija-wá-bala/ → [nijawabáála] ‘2p are counting them’ (cf. [nijabaala] ‘2p are counting’). Likewise, H of the root /bón/ shifts to the penult when underlyingly in the antepenult, as in [nijaboníísa] ‘2p are demonstrating’ (cf. [nijabóóna] ‘2p see’). RMR, the lowering of the first H in a sequence of underlying Hs, is seen in H roots with an OP, such as [nijawabóóna] ‘2p see them’ and [nijawaboníísa] ‘2p are demonstrating them’, from /nija-wá-bóna/ and /nijawá-bónisa/.

Often, both of these H-deletion processes coexist in a language (though not in Nguni). The Nkore data (Poletto 1998) in (54) show that OPs are H toned, and when an OP comes before another H (that of a root or another OP), only the last of the Hs surfaces, indicating that RMR operates on OPs.

  1. (53)Bantu Phonology

MR, deletion of the second in a sequence of Hs, applies to root-initial Hs, affecting H–H sequences that do not undergo RMR. The reflexive prefix -é- does not undergo RMR, nor do the tense prefixes -rí-, -rá-, and the combination of such prefixes plus a H root results in lowering of the root H. If an OP comes after -rí-, -rá-, the OP’s H is unaffected, though as the last form in (55) shows, OP Hs may be lowered by application of RMR.

  1. (54)Bantu Phonology

Lowering of H in the context of another H generally affects only adjacent Hs, but a few long-distance lowerings are attested. In Matumbi, a word-final H in a verb is deleted phrase medially as long as another H precedes it somewhere within the stem.

  1. (55)Bantu Phonology

Long-distance RMR is actually relatively well attested in connection with inflectional melodic Hs (section 4.3). In many languages, all Hs within the Macrostem or word are lowered if a melodic H is added toward the end of the verb. This is illustrated in (57) with data from Kerewe, where in certain tenses H is assigned to the penult, and in others it is assigned to the final vowel and then spreads to the left, prepausally.

  1. (56)Bantu Phonology

Such deletion is widespread in the Lacustrine group.

4.3 Melodic tones

In apparently all tonal Bantu languages, the tonal system is augmented by tone patterns associated with certain grammatical categories, especially verb tenses, which are usually realized as the positioning of additional tones in some position in the stem. These are referred to as melodic H patterns. See Odden & Bickmore (2014) and the associated papers in Africana Linguistica XX for further exemplification of melodic tone patterns in Bantu. For example, in Kerewe, the remote past and hodiernal perfective in (58a) have whatever H is lexically present on the root (plus tone doubling). In the simple past and hesternal past (58b), H is added to the final vowel, and that H causes deletion of preceding Hs.22 The habitual negative and near future (58c) have an added H on the penult, which likewise causes deletion of preceding Hs.

  1. (57)Bantu Phonology

The positioning of the melodic H may be significantly affected by the underlying H vs. L status of the verb root. For example, in Tachoni, the melodic H of the remote future appears from the second stem syllable to the penult with L verb roots, but from the third syllable to the penult with H roots (59a). In the negative imperative, there is H on the syllable containing the second stem mora in toneless roots but no surface H in underlyingly H-toned roots (59b). In the positive imperative, L roots do not realize the melodic tone, but H roots have H from the third syllable to the penult (59c). In all cases, underlying root Hs (underlined) are deleted when a melodic H is added.

  1. (58)Bantu Phonology

Systems of melodic tone inflection can become quite complex. In Marachi (Marlo 2007) there are nearly a dozen tense-determined patterns, involving parameters such as whether H appears at the right versus left edge of the stem, shifting of H into the stem from an OP, deletion of root H, and complete deletion of all Hs.

In some languages, the lexical contrast between H and L roots was historically neutralized in verbs, and all surface tones in verbs come from assignment of melodic Hs according to tense. Matumbi exemplifies this pattern, with Hs assigned according to inflectional construction to the initial, final, third, second, second and final, or initial and third vowels of the stem.

  1. (59)Bantu Phonology

Such systems are found in a range of languages of southern Tanzania, Kuria, southwestern Luyia languages such as Khayo, Ruwund, and Kanyok.

5 Summary

A desideratum of general linguistic theorizing is some method of validating claims about the nature of human language. In the realm of grammatical theory, it is often difficult to fill this lacuna as it pertains to phonology, since it is impossible to construct specific grammars to see if they are learnable or to see whether one kind of fact entails another. Instead, we must observe what types of languages exist. The phonology of Bantu languages is of particular interest for testing grammatical theories, because on the one hand one can find a high degree of sameness in many aspects of their grammars, but on the other hand one also encounters a high degree of difference among languages in specific details. Thus Bantu languages as a whole constitute a naturally occurring controlled experiment that varies the building blocks of phonological systems. Variations in Bantu tone systems are well known and have attracted much theoretical attention. Similarly, the existence of asymmetries in vowel harmony, where the conditions for harmony involve a complex of factors regarding the target and the trigger and where languages differ in the specific requirements for harmony, poses a problem for the theory of phonological rules: theories of rules are founded on the assumption of symmetry in the class of elements participating in a rule, so how do we account for asymmetry? Even the simple matter that the synchronic phoneme /h/ can harden postnasally to [p], [s], [k], or [ŋ] constitutes a challenge to phonetically based theories of representation. These and many similar facts will continue to feed the highly productive interaction between language description and linguistic theorizing for the foreseeable future.


My thanks go to Lee Bickmore, Helen Eaton, Larry Hyman, Constance Kutsch Lojenga, Mike Marlo, Deo Ngonyani, Mary Paster, Thilo Schadeberg, and an anonymous reviewer for their helpful comments and data. Unsourced data are from my notes.


Alnet, Aimee Johansen. 2009. The clause structure of the Shimaore dialect of Comorian (Bantu). Urbana, IL: University of Illinois dissertation.Find this resource:

Bastin, Yvonne, André Coupez & Michael Mann. 1999. Continuity and divergence in the Bantu languages (Annales Sciences Humaines 162). Tervuren, Belgium: Musée Royale de l’Afrique Centrale.Find this resource:

Batibo, Herman. 1985. Le Kesukuma (langue Bantu de Tanzanie): Phonologie, morphologie. Paris: Editions Recherche Sur Les Civilisations.Find this resource:

Beckman, Jill. 1997. Positional faithfulness, positional neutralisation, and Shona vowel harmony. Phonology 14. 1–46.Find this resource:

Bennett, Patrick & Jan Sterk 1977. South Central Niger-Congo: A reclassification. Studies in African Linguistics 8. 241–273.Find this resource:

Bickmore, Lee. 1989. Kinyambo prosody. Los Angeles, CA: University of California dissertation.Find this resource:

Bickmore, Lee, 2000. Downstep and Fusion in Namwanga. Phonology 17. 297–333.Find this resource:

Bickmore, Lee. 2007. Cilungu phonology. Stanford: Center for the Study of Language and Communication.Find this resource:

Cassimjee, Farida. 1998. Isixhosa tonology: An Optimal Domains Theory analysis. Munich: Lincom Europa.Find this resource:

Chacha, Chacha & David Odden. 1998. The phonology of vocalic height in Kikuria. Studies in African Linguistics 27. 129–158.Find this resource:

Claughton, John. 1992. The tonology of Xhosa. Grahamstown, South Africa: Rhodes University dissertation.Find this resource:

Clements, G. N. 1978. Syllable and mora in Luganda (draft version). Cambridge, MA: Harvard University, MS.Find this resource:

Clements, G. N. 1986. Compensatory lengthening and consonant geminaton in luganda. In Leo Wetzels & Engin Sezer (eds.), Studies in compensatory lengthening, 37–77. Dordrecht: Foris.Find this resource:

Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary Beckman (eds.). Papers in laboratory phonology I: between the grammar and the physics of speech, 283–333. Cambridge, UK: Cambridge University Press.Find this resource:

Crane, Thera. 2011. Beyond time: Temporal and extra-temporal functions of tense and aspect marking in Totela, a Bantu language of Zambia. Berkeley, CA: University of California dissertation.Find this resource:

Diprose, Martin. 2007. Lugungu phonology statement.

Doke, Clement. 1954. Textbook of Zulu grammar. Johannesburg: Longmans.Find this resource:

Doke, Clement & Benedict Vilakazi. 1972. Zulu–English dictionary. Johannesburg: Witwatersrand University Press.Find this resource:

Downing, Laura. 1990. Problems in Jita tonology. Urbana, IL: University of Illinois dissertation.Find this resource:

Downing, Laura. 1997. Correspondence effects in Siswati reduplication. Studies in the Linguistic Sciences 25. 81–95.Find this resource:

Downing, Laura. 2005. On the ambiguous segmental status of nasals in homorganic NC sequences. In Marc van Oostendorp & Jeroen van de Weijer (eds.), The internal organization of phonological segments, 183–216. Berlin: De Gruyter.Find this resource:

Fortune, George. 1955. An analytical grammar of Shona. Capetown & New York: Longmans, Green.Find this resource:

Gick, Brian, Douglas Pulleyblank, Fiona Campbell & Ngessimo Mutaka. 2006. Low vowels and transparency in Kinande vowel harmony. Phonology 23. 1–20.Find this resource:

Goldsmith, John. 1982. Accent systems. In Harry van der Hulst & Norval Smith (eds.), The structure of phonological representations, vol. 1, 47–63. Dordrecht: Foris.Find this resource:

Grosserhode, Ralf. 2000. Stimmlose Nasale im Gogo. In Rainer Vossen, Angelika Mietzner & Antje Meissner (eds.), “Mehr als nur Worte”: Afrikanistische Beiträge zum 65. Geburtstag von Franz Rottland, 239–246. Köln: Rüdiger Köppe.Find this resource:

Guthrie, Malcolm. 1967–71. Comparative Bantu: An introduction to the comparative linguistics and prehistory of the Bantu languages. 4 vols. Farnborough: Gregg Press.Find this resource:

Hargus, Sharon & Manuel da Conceição. 1999. Consonant labialization in Ronga. University of Washington Working Papers in Linguistics 16. 31–50.Find this resource:

Heath, Teresa. 2003. Makaa (A83). In Derek Nurse & Gérard Philippson (eds.), The Bantu languages, 335–348. London & New York: Routledge.Find this resource:

Herbert, Robert. 1986. Language universals, markedness theory, and natural phonetic processes. Berlin: De Gruyter.Find this resource:

Higgins, Holly Ann. 2012. Ikoma vowel harmony: Phonetics and phonology (SIL e-Books 43). this resource:

Hubbard, Kathleen. 1995. “Prenasalised consonants” and syllable timing: Evidence from Runyambo and Luganda. Phonology 12. 235–256.Find this resource:

Hulstaert, Gustave. 1965. Grammaire du Lomongo. Deuxieme partie: Morphologie. Tervuren, Belgium: Musée Royale de l’Afrique Centrale.Find this resource:

Hyman, Larry & Francis Katamba. 1999. The syllable in Luganda phonology and morphology. In Harry van der Hulst & Nancy Ritter (eds.), The syllable: Views and facts, 349–416. Berlin: Mouton de Gruyter.Find this resource:

Hyman, Larry M. 1982. Globality and the accentual analysis of Luganda tone. Journal of Linguistic Research 2. 1–40.Find this resource:

Hyman, Larry & Armindo Ngunga. 1994. On the non-universality of tonal association conventions: Evidence from Ciyao. Phonology 11. 25–68.Find this resource:

Hyman, Larry. 1999. The historical interpretation of vowel harmony in Bantu. In Jean-Marie Hombert & Larry Hyman (eds.), Bantu historical linguistics, 235–295. Stanford, CA: CSLI.Find this resource:

Hyman, Larry M. 2001a. Privative tone in Bantu. In Shigeki Kaji (ed.), Crosslinguistic studies of tonal phenomena: Tonogenesis, Japanese accentology, and other topics, 237–257. Tokyo: Institute for the Study of Languages and Cultures, Tokyo University of Foreign Studies.Find this resource:

Hyman, Larry. 2001b. The limits of phonetic determinism in phonology: *NC revisited. In Elizabeth Hume & Keith Johnson (eds.), The role of speech perception in phonology, 141–165. New York: Academic Press.Find this resource:

Ito, Junko. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7. 217–259.Find this resource:

Kenstowicz, Michael & Charles Kisseberth. 1990. Chizigula tonology: The word and beyond. In Sharon Inkelas & Draga Zec (eds.), The phonologysyntax connection, 163–194. Chicago, IL: University of Chicago Press.Find this resource:

Kisseberth, Charles. 1992. Metrical structure in Zigula tonology. In Derek Gowlett (ed.), African linguistic contributions, 227–259. Pretoria: Via Afrika.Find this resource:

Kisseberth, Charles & Mohammad Abasheikh. 2004. The Chimwiini lexicon exemplified. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies.Find this resource:

Kunene, Daniel. 1961. The sound system of Southern Sotho. Cape Town: University of Cape Town dissertation.Find this resource:

Kutsch Lojenga, Constance. 2009. Two types of vowel harmony in Malila (M.24). In Margarida Petter & Ronald Beline Mendes (eds.), Proceedings of the Special World Congress of African LinguisticsSão Paolo 2008, 109–128. São Paolo: Humanitas.Find this resource:

Leitch, Myles. 1997. Vowel harmonies of the Congo basin: An Optimality Theory analysis of variation in the Bantu zone C. Vancouver, BC: University of British Columbia dissertation.Find this resource:

Liphola, Marcelino. 2001. Aspects of phonology and morphology of Shimakonde. Columbus, OH: Ohio State University dissertation.Find this resource:

Maganga, Clement & Thilo Schadeberg. 1992. Kinyamwezi: Grammar, texts, vocabulary. Köln: Köppe.Find this resource:

Maho, Jouni. 2003. A classification of the Bantu languages: An update of Guthrie’s referential system. In Derek Nurse & Gérard Philippson (eds.), The Bantu languages, 639–651. London & New York: Routledge.Find this resource:

Maho, Jouni. 2009. Bantu online bibliography.

Mamet, M. 1955. La langue Ntomba telle qu’elle est Parlée au lac Tumba et dans la région avoisinante (Annales Sciences Humaines 11). Tervuren, Belgium: Musée Royale de l’Afrique Centrale.Find this resource:

Marlo, Michael. 2007. The verbal tonology of Lunyala and Lumarachi: Two dialects of Luluyia (Bantu, J.30, Kenya). Ann Arbor, MI: University of Michigan dissertation.Find this resource:

Marlo, Michael. 2013. Verb tone in Bantu languages: Micro-typological patterns and research methods. Africana Linguistica XIX: 137–234.Find this resource:

Meeussen, Achiel E. 1967. Bantu grammatical reconstructions. Africana Linguistica III. 79–121.Find this resource:

Meinhof, Carl. 1932. Introduction to the phonology of the Bantu languages. Translation and revision by Nicolaas J. van Warmelo & Alice Werner. Berlin: Reimer.Find this resource:

Morris, H. F. 1963. A note on Lunyole. Uganda Journal 27. 127–134.Find this resource:

Mtenje, Al. 2007. On recent trends in phonology: Vowel sequences in Bantu. SOAS Working Papers in Linguistics 15. 33–48.Find this resource:

Muller, Jen. 2002. The phonology and phonetics of word-initial geminates. Columbus, OH: Ohio State University dissertation.Find this resource:

Musimami, Sylvester & Martin Diprose. 2012. Ehyagi hyʼebibono byʼolunyole. Lunyole dictionary. Entebbe: Lunyole Language Association & SIL International.Find this resource:

Mutonyi, Nasiombe. 2001. Aspects of Bukusu morphology and phonology. Columbus, OH: Ohio State University dissertation.Find this resource:

Myers, Scott. 1998. Surface underspecification of tone in Chichewa. Phonology 15. 367–392.Find this resource:

Nash, Jay. 1992. Aspects of Ruwund grammar. Urbana, IL: University of Illinois dissertation.Find this resource:

Nurse, Derek & Gèrard Philippson. 2003. Towards a historical classification of the Bantu languages. In Derek Nurse & Gèrard Philippson (eds.), The Bantu Languages, 164–181. London & New York: Routledge.Find this resource:

Odden, David. 1982. Tonal phenomena in Kishambaa. Studies in African Linguistics 13(2). 177–208.Find this resource:

Odden, David. 1996. The phonology and morphology of Kimatuumbi. Oxford: Clarendon.Find this resource:

Odden, David & Mary Odden. 1999. Kihehe syllable structure. In Harry van der Hulst & Nancy Ritter (eds.), The syllable: Views and facts, 417–445. Berlin: Mouton de Gruyter.Find this resource:

Odden, David. 2003. Rufiji-Ruvuma (N10, P10-20). In Derek Nurse & Gèrard Philippson (eds.), The Bantu Languages, 529–545. London & New York: Routledge.Find this resource:

Odden, David & Lee Bickmore. 2014. Melodic tone in Bantu: Overview. Africana Linguistica XX: 3–13.Find this resource:

Overton, Harold. 1972. A generative-transformational grammar of the Kikuyu language based on the Nyeri dialect. Baton Rouge, LA: Louisiana State University dissertation.Find this resource:

Parkinson, Frederick. 1996. The phonological representation of vowel height, its phonetic correlates and role in constraint-based phonology. Columbus, OH: Ohio State University dissertation.Find this resource:

Peng, Long. 2000. Kikuyu vowel harmony. South African Journal of African Languages 20. 370–384.Find this resource:

Petzell, Malin. 2008. The Kagulu language of Tanzania: Grammar, texts and vocabulary. Köln: Rüdiger Köppe.Find this resource:

Poletto, Robert. 1998. Topics in Runyankore phonology. Columbus, OH: Ohio State University dissertation.Find this resource:

Roberts-Kohno, R. Ruth. 2000. Kikamba phonology and morphology. Columbus, OH: Ohio State University dissertation.Find this resource:

Schadeberg, Thilo. 1971. Zur Lautstruktur des Kinga (Tanzania). Marburg, Germany: Philipps-Universität dissertation.Find this resource:

Schadeberg, Thilo. 1989. The velar nasal in Nyole (E. 35). Annales Aequatoria 10. 169–179.Find this resource:

Schadeberg, Thilo. 2003. Historical linguistics. In Derek Nurse & Gèrard Philippson (eds.), The Bantu languages, 143–163. London & New York: Routledge.Find this resource:

Sievers, Eduard. 1881. Grundzüge der Phonetik. Leipzig: Breitkopf & Hartel.Find this resource:

Solé, Maria-Josep, Larry Hyman & Kemmonye Monaka. 2010. More on post-nasal devoicing: The case of Shekgalagari. Journal of Phonetics 38. 604–615.Find this resource:

Stevick, Earl. 1969. Tone in Bantu. International Journal of American Linguistics 35. 330–341.Find this resource:

Thomas, John Paul. 2011. A morphophonology of Komo: Non-tonal phenomena. SIL Electronic Working Papers 2011-006.

Tucker, Archibald N. 1929. The comparative phonetics of the Suto-Chuana group of Bantu languages. London: Longmans, Green.Find this resource:

Valinande, Ngessimo. 1984. The structure of Kinande. Washington, DC: Georgetown University dissertation.Find this resource:

van Otterloo, Karen. 2011. The Kifuliiru language: Volume 1. Phonology, tone, and morphological derivation. Dallas: SIL International.Find this resource:

Williamson, Kay & Roger Blench. 2000. Niger-Congo. In Bernd Heine and Derek Nurse (eds.), African languagesAn introduction, 11–42. Cambridge, UK: Cambridge University Press.Find this resource:

Yoneda, N. 2000. A descriptive study of Matengo, a Bantu language of Tanzania: With focus on verbal structure. Tokyo: Tokyo University of Foreign Studies dissertation.Find this resource:

Yukawa, Yasutoshi. 1989. A tonological study of Sukuma verbs. Bantu linguistics, 2: Studies in Tanzanian languages, 339–404. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies.Find this resource:


(1) SP = Subject Prefix, TNS = tense-aspect-mood-polarity-clause inflection, OP = Object Prefix, EXT = extension, FV = final inflection.

(2) Authors frequently give examples using abstractions such as roots or stems, to which various prefixes and tones would be added to form specific words. Usually, a verb stem ending in /-a/ with the right tone pattern happens to form a singular imperative.

(3) Caution is called for in interpreting the orthographic distinction <e o> vs. <ɪ ʊ> as the phonetic value of degree-2 vowels. The degree-2 vowels <e o> employed in works on Kamba by Odden and Roberts-Kohno are somewhere between canonical [ɪ ʊ] and [e o], being noticeably higher than the corresponding vowels of Kikuyu but lower than those of Matumbi. In this case the choice of symbols is arbitrary. Batibo (1985) likewise represents the degree-2 Sukuma vowels phonemically as <e o> but describes them phonetically as [ɪ ʊ], which agrees with the description of Yukawa (1989).

(4) Unusually, the expected final vowel /a/ is lacking after these vowel-final extensions.

(5) There is no clear example of /ɛ/ followed by either of these suffixes.

(6) See section 3.1 for discussion of syllable structure.

(7) Overton (1972) identifies this as a typical feature of the Nyeri dialect.

(8) Vowel-initial roots receive an epenthetic palatal [dʒ] or [ɲ], depending on whether a voiced stop appears in the following syllable, so such roots do not provide adjudicating evidence for a prefix nasal.

(9) See Petzell (2008) for phonetic evidence from Kagulu. The judgment regarding Gogo is based on my own experience and is supported by Grosserhode (2000), who observed dialectal variation between [mph], [m̥ph], and [m̥].

(10) The unaspirated voiceless stops are weakly and variably ejective (compared with ejection found in e.g. Salishan, Ethiopian Semitic, or Chechen). There is no clear evidence indicating whether ejection is a phonological characteristic of unaspirated stops.

(11) The verbs -paang- and -haang- both derive from proto-Bantu *pang- ‘do, make’, and Kerewe -paang- is borrowed from Swahili.

(12) Morris (1963) transcribes <x> where other sources transcribe <h>, which may reflect a dialect difference.

(13) This particular root is from Bantu *budag (Kisseberth & Abasheikh 2004).

(14) This argument would not hold in languages that allow final consonants. These languages, however, do not seem to have pre-NC lengthening.

(15) Downing does not give a fully syllabified output example and does not give candidates with NC onsets in her formal analysis, so this representation is a conjecture based on various statements in the article. She also claims that word-initial preconsonantal nasals are (surface) moraic, which is a possible analysis in some languages like Jita, where syllabicity/moraicity of initial preconsonantal nasals is noncontrastive, but is not tenable for languages like Swahili, where it is contrastive ([m̩buni] ‘coffee tree’, [mbuni] ‘ostrich’).

(16) Long vowels as found in [kutééka] ‘to cook’ resemble vowel sequences only orthographically: they are single long segments.

(17) Glides always delete before homorganic high vowels.

(18) It is also possible that these prefixes are underlyingly /i,u/ and word-initial vowels are lowered—the only initial high vowel, the cl. 5 marker [i-], derives from /eli-/.

(19) Because Optimality Theory has no means of characterizing underlying properties, input-specified L cannot be explicitly ruled out. Downstep is analyzed as the phonetic implementation of consecutive H autosegments; see Odden (1982) for arguments in Shambaa showing that downstep does not necessarily require specified L.

(20) H is assigned to the final vowel when the penult is underlyingly H toned.

(21) Penults are automatically lengthened.

(22) This final H spreads leftward when prepausal.