2. Distance-based network analysis

A distance-based network analysis is a quantitative analysis used as “an introductory visual means of data exploration” (Pelkey, Reference Pelkey2011:279). The Neighbor-Netalgorithm, as developed by Bryant & Moulton (Reference Bryant and Moulton2004) is applied to a distance matrix, or a standard lexicostatistical matrix with the figures converted into their opposite values. For example, the Isikisya and Ndala lects below are 0.63735 similar and 0.36265 dissimilar (see the lexicostatistical matrix for the majority of the corpus languages in Appendix B). Splits Tree 4 (4.11.3) is the software program used in this paper to implement the algorithm and display the results (Huson & Bryant, Reference Huson and Bryant2010).

Bantu languages, and East African Bantu languages in particular, are characterized by the dual historical processes of divergence and convergence. Standard tree models are only able to represent divergence. “Unlike trees, which only permit branching and divergence among taxa, networks can also have reticulations among branches, making it possible to show more than one evolutionary pathway on a single graph” (Holden & Gray, Reference Holden and Gray2006:24). The diagram, or splits-graph, does appear more tree-like if the distance-based relationships are unambiguous. However, if the relationships are ambiguous, the splits are weighted with the relative weight (or confidence in the split) indicated by the length of the branch. Refer to Holden & Gray (Reference Holden and Gray2006) for more on the Neighbor-Netalgorithm as applied to other Bantu languages. Figure 1 below illustrates the competing relationships of the Nyamwezi dialects (the Konongo lects of Ilunde, Inyonga, and Utende are not included).

Figure 1 Distance-based network of 10 Nyamwezi varieties.

Based on the distance-based network in Figure 1, the evidence points towards at least three main dialects within Nyamwezi: a Tabora dialect consisting of Igalula, Kitunda, and Makingi, a Sikonge dialect consisting of Igigwa, Ipole, and Mkolye, and an Urambo-Usoke dialect. From the distance-based network, Isikisya appears to also belong to the Tabora dialect along with Makingi, but is shown in §3 to be a part of Sukuma along with the Ndala lect on the basis of shared phonological innovations.

3. Implicational hierarchy and dynamic wave analysis

In this section, I use certain concepts within Baileyean dialectology (see Bailey, Reference Bailey1996), namely an implicational hierarchy and an accompanying dynamic wave model, to examine the validity of the dialectal relationships proposed in §2 and refine them.

The difficulty with examining isogloss patterns is that “isoglosses typically fail to bundle in topological or social space and scattered transitional dialects are the norm” (Mühlhäusler, Reference Mühlhäusler1996:11). An implicational hierarchy rearranges those isoglosses so that each lect is distinguished by only one feature. A plus sign represents the presence of a particular feature, while the minus sign represents its absence. The resulting distribution is implicational in the sense that if a lect at the far right of the diagram has a particular feature, those to its left necessarily have that feature as well.

The goal is to see how innovations have spread across dialects in space and time. At a minimum, the most innovative dialects can be separated from the more conservative ones. However, as mentioned in §1, Nyamwezi has a reputation for being extremely conservative. Also, innovations such as *p-lenition, Dahl's Law, and spirantization have applied inconsistently across Sukuma and Nyamwezi (Batibo, Reference Batibo2000; Nurse, Reference Nurse1999:10). The implicational hierarchy can aid in seeing how these innovations have spread across dialects. This implicational pattern can also be represented in a dynamic wave model, another way of illustrating this spread of innovations. These methods yield both quantitative and qualitative measures, and the conclusions are predictive, testable, and falsifiable.

Table 4 below presents the implicational hierarchy for the Nyamwezi lects along with the Utende variety of Konongo. The wordlist data supporting each of the features in Table 4 are included in Appendix A. Isikisya and Ndala (data from Maganga & Schadeberg, Reference Maganga and Schadeberg1992) are so similar in terms of shared innovations that they have been combined. For the lexical isogloss ‘hunter’, a minus symbol (−) was given to those lects with the reflexes of *-bɪnd,Footnote ⁹ while a (+) was given to those lects with the reflexes of either *-guɪm or *-pɪig. The majority of the spirantization features are included in the separated bottom portion of the table as an indication that the spread/diffusion of these features occurred differently (see below). Please note the different ordering of the lects in the bottom portion as a result.

From the top portion of Table 4, Isikisya and Ndala are shown to be the most innovative lects in the corpus, while Konongo is the most conservative. One could interpret the implicational pattern in Table 4 as evidence for a dialect continuum from Isikisya and Ndala in the north all the way to Konongo in the south. It is important to realize, however, that the implicational hierarchy represents the spread of linguistic innovation over the space and time dimensions, and does not make any inherent judgments regarding language/dialect, dialect chains, transitional dialects, subgrouping, etc. As Bailey, quoted in Mühlhäusler, says: “We must distinguish the relative constancy of the linguistic pattern (i.e. in the grammar) from the many ways in which the variants can be distributed at different times and different places” (1996:11).

The ordering of lects within the implicational hierarchy in Table 4 confirms the conclusion from the distance-based network in §2 that there are at least three main dialects within Nyamwezi: A Tabora dialect consisting of Igalula, Kitunda, and Makingi, a Sikonge dialect consisting of Igigwa, Ipole, and Mkolye, and an Urambo-Usoke dialect. The same groupings can be derived from the lower portion of Table 4 dealing with the distribution of Bantu spirantization features (see discussion below). The inclusion of Kitunda with the Tabora dialect, although the lect is spoken much further south (see Map 1), can be explained by the fact that speakers from this area claim they originally came from Tabora (time depth unknown). The linguistic evidence supports this claim, and also explains their use of mu-βend-i ‘hunter’, a reflex of *bɪnd, and not a reflex of either *guɪm or *pɪig.

Figure 2 below uses the data from the top portion of the implicational hierarchy to show the wave-like spread of innovation over each of the lects.

Figure 2 Dynamic wave model of Nyamwezi dialects.

The Isikisya and Ndala lects should be considered Sukuma linguistically, while the Sikonge lects are distinctly Nyamwezi. Makingi and Igalula should be considered transitional dialects, or mediators between the more historically innovative Sukuma lects and the more historically conservative Nyamwezi lects (see §5). The most representative example is the fact that each of the Tabora dialects has remnants of Dahl's Law, while the Urambo-Usoke and Sikonge dialects have no traces.

As previously mentioned, the agentive and causative spirantization features follow a different pattern, and are included in the separated bottom portion of Table 4 as a result. The key innovations which do not fit into the top portion include *d causative spirantization and the lexicalized, spirantized mlozi ‘witch’. The implication for the Bantu spirantization innovations is that their origin was in the Nyamwezi lects (Konongo, Urambo-Usoke and Sikonge) and not Sukuma. What might this process have looked like?

Bantu spirantization in both Sukuma and Nyamwezi is considered phonologically restricted. Whereas in other languages with Bantu spirantization all or most stops are replaced by fricatives in the proper phonological/morphological environment, in Sukuma and Nyamwezi only certain stops are replaced by fricatives, e.g. in Sukuma, *p, *t, *k, *d morpheme-internally and *d before the agentive (Batibo, Reference Batibo2000; Bostoen, Reference Bostoen2008:322). This corresponds with the restrictions to causative and agentive Bantu spirantization in the bottom portion of Table 4 as well. Furthermore, there is evidence from Bantu spirantization typology which proposes that “[Bantu spirantization] originally affected all plosives across the board, but only became morphologized in the case of the most commonly spirantized plosives” (Bostoen, Reference Bostoen2008:341). There also seems to be a progression of Bantu spirantization in which coronals (*t/*d) are affected first, then velars (*k/*g) and labials (*p/*b). Bostoen (Reference Bostoen2008) goes on to say that if this is true, “then it seems plausible that at a certain stage, the majority of spirantized agent nouns were derived from verb roots with a final coronal consonant” (341).

Bostoen says that “as regards the diachronic evolution of Agent Noun Spirantization (ANS), ‘phonologically restricted ANS’ seems to represent an initial stage in the morphological conditioning of the sound change” (2008:340). Both Sukuma and Nyamwezi are 7V languages, and the retention of a 7V system also hinders the morphologization of Bantu spirantization (Bostoen, Reference Bostoen2008:343). For Sukuma and Nyamwezi this means that agentive nouns with *d (*d > l) such as mtoozi ‘groom’ (-toola ‘to marry’) and msuzi ‘blacksmith’ (-sula ‘to forge’) are fossilized evidence of the possible beginnings of morphologization and subsequent lexicalization of these forms before the language had a chance to level them away. For Konongo and most of Urambo-Usoke and Sikonge, this process was able to take place for at least one agentive noun with *g, mlogi>mlozi. It is possible that such ANS could have been diffused from nearby language groups to the south or west (Bende-Tongwe, Rungwa or Ha).

Table 5 summarizes the relevant changes for each dialect area from the implicational hierarchy data in Table 4.

The aim of this section was to see how the early period innovations spread across dialects in space and time. We saw from the implicational hierarchy (and dynamic wave model) how innovations spread across dialects in the space dimension. Within the current data corpus, the Isikisya and Ndala area seems to be the focal area for many of the early period innovations (except Bantu spirantization). Those innovations spread to the Tabora dialect area, and then later to Urambo-Usoke and to some extent Sikonge. The implicational hierarchy also makes a claim regarding the relative ordering of innovations across the time dimension. From the top portion of Table 4, this indicates *c/*j fricativization occurred first, then *p-lenition and subsequently Dahl's Law.

In §4, I set out to refine the relative chronology for the early period innovations in Sukuma and Nyamwezi. The purpose is to use the insights from this research on Nyamwezi dialectology to begin to untangle Nyamwezi's historical relationship with Sukuma and the rest of WT.

4. Towards a refined relative chronology

We saw in §1.2 and §3 that most of the early period innovations within Sukuma and Nyamwezi have inconsistent/incomplete application, including *p-lenition, Dahl's Law and Bantu spirantization. Batibo (Reference Batibo2000) offers solid evidence that the inconsistent application of these phonological innovations is due to the timing of their activity/inactivity, bleeding effects, and the incorporation of lexical items from other sources. In regard to Bantu spirantization in particular, we saw from Bostoen's (2008) Agent Noun Spirantization typology that the processes of morphologization, lexicalization, and analogical levelling were also at work.

In §3, we saw that the implicational hierarchy makes an implicit claim regarding the relative ordering of innovations across the time dimension. In the case of Sukuma/Nyamwezi, *c/*j fricativization occurred first, then *p-lenition, and subsequently Dahl's Law. Batibo (Reference Batibo2000) also proposes a limited relative chronology, which is summarized in Figure 3. There is essentially no substantial difference between my proposal and Batibo's (2000) proposal in regard to the relative ordering of *c/*j fricativization, *p-lenition and Dahl's Law.

Figure 3 Summary of Batibo's (2000) relative chronology.

To summarize Batibo's (2000) proposal, the Sukuma and Nyamwezi were still an intact group after their split from West Tanzania and stayed that way until around 1100-1300ad. The WT split occurred around 500ad: the Nilamba/Nyaturu went to the northeast, the Kimbu to the southeast, and the Sukuma/Nyamwezi to the northwest (Batibo, Reference Batibo2000:24). The fricativization of *c/*j took place before the WT split, and Dahl's Law, *p-lenition, and spirantization all were activated before the split of Sukuma and Nyamwezi. The fact that Nilamba/Nyaturu and Kimbu do not have Dahl's Law supports this conclusion (Batibo, Reference Batibo2000:24). Dahl's Law must have taken place after the WT split but before spirantization, while *p-lenition could have occurred anytime after the WT split but before the Sukuma-Nyamwezi split (Batibo, Reference Batibo2000:24–5).

However, §3 of this paper has presented phonological evidence contrary to Batibo (Reference Batibo2000) and to some extent Masele (Reference Masele2001). Based on the data from the implicational hierarchy in Table 4, many of the Nyamwezi dialects do not share the same innovations as Sukuma. For instance, the only sign the Urambo-Usoke and Sikonge dialects show of Dahl's Law is *kupɪ > ŋguhɪ. The process was perhaps “undone in its early stages” as Nurse proposes for some of the Great Lakes (GL) languages (1999:28). It is just as likely that the later diffusion of Dahl's Law never reached the more southern Nyamwezi dialects and that the lexical item ŋguhɪ/ ŋgupɪ ‘short’ was borrowed and maintained at a later point. Along with the fact that Nyaturu, Nilamba, and Kimbu show limited (or no?) signs of Dahl's Law (Nurse, Reference Nurse1999:10), this is fairly good indication that the innovation did not spread any further southward.

Batibo claims that Dahl's Law is/was active across morpheme boundaries in Nyamwezi, giving the example k℧-tool-a > g℧-tool-a (2000:26). This type of Dahl's Law across morpheme boundaries was not found in any of the corpus lects. Combined with the fact that these dialects have inconsistencies with the northern Nyamwezi dialects/Sukuma in regard to *p-lenition and Bantu spirantization as well, it means at very least the Urambo-Usoke and Sikonge dialects did not travel the same path as northern Nyamwezi/Sukuma for as long.

The activation of Bantu spirantization is assumed by Batibo to have followed Dahl's Law (and perhaps *p-lenition?) because of possible bleeding effects. At least one of Batibo's examples of the interaction of Dahl's Law and Bantu spirantization was found to be inconclusive. He argues that *-puta > -buta ‘to cut’ prevented the form -futa from occurring (2000:24). However, the Urambo-Usoke and Sikonge dialects retain -puta (see Appendix A). Data for analyzing morpheme-internal spirantization were not collected for the corpus data (see §6 for ideas for further research). If the bleeding effects are inconsistent, it is possible that the activation of Bantu spirantization took place prior to or simultaneous with Dahl's Law and *p-lenition (see discussion in Nurse, Reference Nurse1999:21–2).

Essentially, Batibo holds to a divergence model for the linguistic history of Sukuma and Nyamwezi. Batibo assumes a coherent Proto-WT which included Nilamba/Nyaturu, Kimbu, and Sukuma/Nyamwezi. He also assumes a Proto-Sukuma/Nyamwezi and assigns approximate dates to various splits. We could represent this fairly easily with a Stammbaum diagram. At the risk of oversimplification, Batibo's (2000) divergence proposal is a valid hypothesis. Of course, there is much more complexity, as we have seen, but if we further posited an older dialect chain and the transitional dialects around Tabora, this would make sense of much of the data. However, it does not explain the palatalization of the *d causative in Isikisya and Ndala (e.g. ku-guja ‘to sell’ with /j/ as a voiced palatal stop; see Note 1 to Appendix A) rather than the spirantization which occurs in the *d causative in the Nyamwezi lects (e.g. ku-gu z ya ‘to sell’). These are completely different outputs and are difficult to explain in a divergence scenario like the one described above.

There is another equally valid hypothesis which also explains the data and owes more to convergence (see Masele, Reference Masele2001). Under this scenario, we do not have to assume the unity of a WT (although this approach is not incompatible with a unified WT). Instead, we assume that either Sukuma or Nyamwezi was already in situ, while the other came from somewhere else, and came into close contact with the in situ variety. This scenario seems to make the most sense if Sukuma were the in situ variety, and Nyamwezi came from elsewhere and came into close contact with Sukuma. The supporting evidence for such a scenario would be the spread of *p-lenition and Dahl's Law which for the most part only reached Sukuma and the Tabora transitional dialects, as well as the differences in Bantu spirantization (*d causative, *g agentive) between Sukuma/Tabora and the rest of Nyamwezi. This scenario would also be compatible with a situation in which Sukuma experienced the diffusion of Dahl's Law and *p-lenition from nearby Great Lakes/Interlacustrine languages to the north and/or west.

5. Synthesis

Thus far, I have demonstrated through the combination of a distance-based network analysis, implicational hierarchy and dynamic wave analysis that Nyamwezi should be considered to have at least three main dialects: Tabora, Sikonge, and Urambo-Usoke. I have also shown that the Isikisya and Ndala lects should be considered Sukuma on the basis of lexical and phonological evidence. By implication, this evidence also points toward a rough linguistic border between Nyamwezi and Sukuma just north of Tabora. Lexically, the Isikisya lect could very well be grouped with Makingi according to the distance-based network analysis. However, the Isikisya lect is also not lexically incompatible with Ndala (see Figure 1). Moreover, the phonological overlap between the Isikisya and Ndala lects is coupled with several key phonological differences with the majority of their southern neighbors: Dahl's Law, voiceless nasals, and lack of *d causative spirantization. The presence of such a linguistic border just to the north of Tabora is in direct contradiction to Maganga and Schadeberg's evaluation that “there seems to be no strict linguistic border which would separate it [Kinyamwezi] from its northern neighbor Kisukuma” (1992:11).

Consequently, the linguistic border just north of Tabora does not match up with the perceived ethnic border around Nzega. The result is a “no-man's land” between Tabora and Nzega where people identify themselves as Nyamwezi but actually speak Sukuma (see Map 3 below).

In Map 3 the borders for each of the dialects, as well as the linguistic and ethnic borders for Nyamwezi and Sukuma, are intended to be approximate.

Instead of positing a dialect continuum the length of the entire Nyamwezi and Sukuma language areas, though, I argue that the Makingi and Igalula lects around Tabora should be considered transitional dialects. Within such transitional dialects, features of a more innovative language variety and a more conservative language variety are negotiated. Language change, however, moves in only one direction, from the more innovative variety to the more conservative. In the case of Nyamwezi/Sukuma, language change filtered from the Sukuma lects influenced the Tabora dialects, e.g. *p-lenition and Dahl's Law. This transitional-dialect analysis explains both why “speakers…refer to any more northern variety…as Kisukuma”, and the fact “they also recognize that ‘Nyamwezi’ and ‘Sukuma’ are two distinct ethnic identities, each with its own language” (Maganga & Schadeberg, Reference Maganga and Schadeberg1992:11).

Theoretically, the transitional-dialect analysis does not preclude an internal dialect continuum within either Nyamwezi or Sukuma. It is still likely that Sukuma exists internally as a dialect continuum, stretching from just north of Tabora to the north of Mwanza. Nevertheless, a full dialect survey of Sukuma would need to be done to confirm that hypothesis. For Nyamwezi, this is much less likely. However, if one considers Konongo a dialect of Nyamwezi, it may be possible to consider Konongo and Sikonge the beginning of a dialect chain northward.

In addition, I also made a claim regarding the relative ordering of innovations: First *c/*j fricativization, and then *p-lenition followed by Dahl's Law. There is essentially no substantial difference between my proposal and Batibo's (Reference Batibo2000) proposal in regard to the relative ordering of these three innovations. No firm claim was made, however, in regard to the timing of Bantu spirantization. Contrary to Batibo (Reference Batibo2000), though, my data/analysis is not incompatible with Bantu spirantization operating concurrently with *p-lenition and/or Dahl's Law. Furthermore, a convergence scenario in which Sukuma was already in situ, with Nyamwezi arriving at a later point and coming into close contact with Sukuma was found to correspond more closely with the historical linguistic evidence.

6. Conclusion and further research

This paper set out to remedy the lack of information regarding the dialectology of Nyamwezi, a well-known East African Bantu language. At least three main dialects were identified within Nyamwezi: (1) a wider Tabora dialect including the Igalula, Makingi, and Kitunda lects, (2) a Sikonge dialect consisting of the Mkolye, Ipole, and Igigwa lects, and (3) a dialect consisting of Urambo and Usoke. These three main dialects were identified using both quantitative and qualitative methods: A distance-based network analysis, an implicational hierarchy, and dynamic wave analysis. Konongo could also be considered a dialect of Nyamwezi, but the situation remains unclear. In addition, the Isikisya and Ndala lects were identified as Sukuma, and a rough linguistic border was identified just north of Tabora. The Tabora lects were found to be transitional dialects, mediating innovation from Sukuma to Nyamwezi. Nyamwezi and Sukuma are not dialects of one another, and do not exist in a dialect continuum with one another as has been traditionally assumed. This transitional-dialect analysis better incorporates the new Nyamwezi survey data.

This was not dialectology for the sake of dialectology, however. The additional aim of this paper was to apply a Baileyean dialectology to the early period innovations in Sukuma and Nyamwezi to conceptualize the spread of linguistic innovation. The situation with Nyamwezi and Sukuma was seen to be extremely complex with phonological innovation characterized by activity/inactivity, bleeding effects, lexical borrowing, morphologization, lexicalization, and analogical levelling. The hope is that having used the Nyamwezi-Sukuma situation as a sort of paradigmatic test case in the region, related research can be done in other nearby Bantu languages. If so, we can begin to refine our conception of Proto-Bantu and the early movements of the Bantu peoples from the bottom-up. However, there is much more to do even within Nyamwezi and Sukuma.

As mentioned at the outset in §1, this study was based on data from a rapid dialect survey of the Nyamwezi area, focusing only on segmental phonological features and lexicostatistics from relatively short wordlists. Areas for further research should firstly include a comparative study on the lexical and grammatical tone patterns in Nyamwezi and Sukuma. Much work has already been done on tone in Sukuma (e.g. Batibo, Reference Batibo1985; Maganga & Schadeberg, Reference Maganga and Schadeberg1992). Now that the main dialect areas have been identified one could do research more strategically in fewer locations. Another area for further research would be a comparative study on tense/aspect morphology in these same varieties.

Yet another idea for further research is a full dialect survey of Sukuma. Not only would this provide information regarding the “‘focal’ and ‘relic’ areas” (Hock, Reference Hock1991) of Sukuma, but the results would be a crucial piece of the puzzle for reconstructing the historical scenario of WT and the relative chronology of Bantu spirantization in relation to *p-lenition and Dahl's Law. Further research into the dialects of Nyamwezi could include the examination of morpheme-internal spirantization, agent noun spirantization, and core vocabulary from the Swadesh lists. More of the data from Batibo (Reference Batibo2000) could also be checked for any inconsistencies in each of the major dialect areas of Nyamwezi now that they have been identified.

Another area for research would be a proper phonological analysis of the vowel system in Konongo. Intelligibility testing could also be done between Konongo and other Nyamwezi varieties. Further research is also needed to compare Konongo with Kimbu lects to the south as this would aid in uncovering the exact nature of Konongo's historical relationship within a hypothetical WT.