270 is in many respects like Kuria.
96 97
Note that both Simbiti and Ngoreme share Kuria’s unusual root vowel lowering which occurs with the applicative and inversive extensions.
The one-step raising and lowering in Ikizu, Kuria and Simbiti prefixes and stems is quite a different pattern than either of the more typical cases of [-ATR] or [+ATR]
dominance. They exhibit what could be considered a combination of both height and [ATR] harmony, and I refer to them in this thesis as height harmony languages. The
incremental raising in these languages e.g. [e o], and e o to [i u] could be a
remnant of [+ATR] spreading, which would have raised both degree 2 and degree 3 vowels, since both sets would have been [-ATR] if the languages historically had a
7VH system. It is not uncommon for sound changes to cause a change from a more expected [ATR] alternation to a less-usual height alternation van der Hulst and van de
Weijer 1995.
7.3.2 Three possible histories of Mara’s seven-vowel languages
Based on the data in the previous section as well as a variety of other data from these languages and other work on Bantu historical linguistics, we can entertain at least three
possibilities concerning the history of Ikoma and the other JE40 Mara languages. Two of the hypotheses assume an immediate genetic relationship between all of the 7V Mara
languages, whereas one does not. This is an important issue, since there is neither sufficient synchronic nor diachronic work on all of the languages to know how valid this
96
One difference between Kuria and Simbiti is that Simbiti only has the five vowels i a u underlyingly in the verb system, even though it has the additional vowels e o in the noun system. The upper mid vowels
[e o] occur allophonically in verbs, but not underlyingly. Interestingly, there is evidence that Ngoreme has also lost e o as phonemes in verbs but similarly retains [e o] allophonically.
97
The prefix alternations in these languages appear to be quite similar to a group of Kru and “Togo- remnant languages” referred to by van der Hulst and van de Weijer 1995 520. They show alternations
which occur in the language Santrokofi, in which i u occur in affixes with i u stems, e o occur with e o stems, and occur with a stems. These three-way alternations could be analyzed as combination of
both [ATR] and [low] spreading. See van der Hulst et al. 1986.
271 grouping is from a genetic not merely a geographic standpoint. Recall that in the
introduction §1.3, I note that Nurse 1999 calls into question the lack of evidence available for evaluating the accuracy of the current classification of Great Lakes Bantu
languages in particular. How sure are we that these languages constitute an immediate genetic family?
More specifically, could the 7V languages of Mara be split into two or more groups which have different historical origins, so that genetic relationships between them are
farther up the tree than previously assumed? There is enough variation in the region for this to be possible.
98
If there is not genetic unity in Mara, we have little room to speculate concerning Ikoma’s history, since it is not even a straightforward exercise to determine
which of the other languages Ikoma is most likely related to. Because the languages have been in close contact for so long, without going through the comparative method it is
difficult to tell which features of the languages are inherited from the same source and which are borrowed. And since there are both 7VH and 7VM systems in the region,
which exhibit [+ATR] dominance and total height i.e. raising and lowering harmony, respectively, the genetic origin which we posit for Ikoma would certainly affect our
understanding of how it has developed its inventory and harmony patterns. This is a difficult issue which we simply cannot be sure of without further comparative work.
Nonetheless, without making any claims concerning the genetic unity or lack thereof of
98
At the same time, however, there are many curiosities which confuse the picture. One example is from a lexicostatistics study reported in a survey report Hill et al. 2007 which reports that Ikizu and Zanaki have
87 lexical similarity, whereas Ikizu has only 61 lexical similarity with a Kuria dialect in the region. However, recall from the previous section that Ikizu and Kuria share many more similarities in terms of
their vowel systems. Other features such as Dahl’s Law, the tenseaspect systems, noun classes, etc. often suggest different historical divisions. Therefore, there is no clear answer on the surface about the history of
the region’s linguistic situation. Without using the comparative method and other means of digging deeper into these patterns which is beyond the scope of this thesis we cannot reliably determine the subgroupings
or even the level of genetic unity of this group of languages.
272 the Mara languages, I propose two other historical possibilities which assume some level
of genetic unity. In particular, it assumes that Ikoma and Zanaki are daughters of the same proto language.
In the remainder of this section, I begin with the hypothesis which I find most plausible, based on the most well-attested types of sound change patterns. In this
hypothesis, Ikoma would have historically come from a proto language with the 7VH inventory i a u and [+ATR] harmony, much like present-day Zanaki as well as
Kinande, Rangi, etc. This hypothesis largely follows Stewart’s 2000 argument that the Proto-Bantu PB vowel system was i a u, not i e a o u as is argued for by
Hyman 1999.
99
Stewart argues that even though i e a o u the inventory of the non- Savannah Bantu languages, i.e. Guthrie’s zones A, B and C is more common cross-
linguistically, the rest of Bantu i.e. Savannah Bantu generally has the i a u system. He also argues that the most natural and plausible inventory change is , e,
o, not e, o , . With this background, it makes more sense to posit and as PB’s second degree vowels. If this is the case, it is likely that Zanaki, as well as a number of
other present-day Savannah Bantu languages with the same inventory, have simply retained the 7V PB inventory.
Stewart’s analysis could be good reason to argue that Ikoma most likely had the same 7VH system as present-day Zanaki and then underwent the , e, o sound
change. In support of this hypothesis, note that Hyman 1999 refers to Mould 1981 187, who suggests that “Gusii and evidentally all of East Nyanza e.g. Kuria…
show other adjustments to the system i, i, e, u, u, o, a i, e, , u, o , a.” Therefore, we
99
Note, however, that Hyman reconsiders this position, and in Hyman 2003 he concedes that i a u, with an [ATR] contrast in the high vowels, was likely the proto system, exemplified by present-day
Kinande, and well-attested in eastern Bantu languages.
273 see that others have posited this inventory change as well. It is difficult to know exactly
how that change affected vowel harmony, but we can speculate. If the [+ATR] dominance which we now see in Zanaki was also present in an earlier Ikoma vowel
system, perhaps some of the [+ATR] dominance patterns remained even after the inventory change. The process of change might have looked something like that
described in 215 below. 215
Possible stages of language change in Mara JE40 languages •
Mara JE40 Proto language has inherited inventory i a u same as PB, possibly with [+ATR] harmony
• Sound change from e o in all languages but Zanaki
• Non-low prefix vowels previously had underlying height 2 vowels which were
[-ATR] , but with the sound change, underlying height 2 prefix vowels became [+ATR] e o
• Vestiges of [+ATR] harmony remain in Ikoma, with leftward [+ATR]
spreading from i u •
Ikoma develops [-ATR] spreading and prefix dissimilation before [-ATR] vowels
• Kuria, Simbiti, and Ikizu keep underlying e o in prefixes but have a three-
way alternation, with i u and surfacing in prefixes as well. Kuria and Simbiti also exhibit raising of root vowels before i u and lowering of high
root vowels before mid vowels in verbal extensions
100
i.e. roots with i u become [ ] before .
This can be represented graphically in the tree in 216 below.
100
As mentioned in footnote 96 above, Simbiti has only the lower mid vowels in verbs, whereas Kuria has all seven vowels in both verbs and nouns. Both Kuria and Simbiti have lowering of high vowels before
the applicative and inversive extensions, but the patterns are a little different because of the different vowel systems in verbs. In Kuria, high root vowels lower to [e o] before these mid-vowels extensions. However,
in Simbiti, the high vowels lower as far as [ ] before [ ] in extensions.
274 216
Mara JE40 language change hypothesis
Proto Mara JE40
i a u
Zanaki e o
i a u [+ATR] harmony
pfxs: m - and omu- Ikoma, Nata, Ngoreme Kuria, Simbiti, Ikizu
i e a o u i e a o u
prefixes omo- and omu- prefixes omo-, m -, umu-
leftward [+ATR] from i u leftward [+ATR] from i u
Kuria Simbiti
Ikizu
i u [e o] before [e o] extensions
i u [ ] before [ ] extensions
The second hypothesis is the opposite picture of that presented above: Proto-Mara JE40 had the i e a o u system, and Zanaki is the innovating language, with the sound
change e o . This second hypothesis has a number of difficulties. As Stewart suggests, a sound change in this direction is not nearly as well attested. Also, it would
not explain why the majority of the 7V Mara languages have curious instances of [+ATR] spreading or raising in the case of the height harmony languages even though they have
vowel inventories which are not only suggestive of [-ATR] dominance but also exhibit [-ATR] or lowering harmony elsewhere in the language. On the contrary, Zanaki’s
vowel system is in many ways very typical of 7V Bantu languages and has no known instances of the sorts of less usual harmony patterns as we find in the other 7V Mara
languages. Therefore, all evidence points to Zanaki as the conservative language of the region, whereas the others are innovators of the 7VM system and of different types of
[-ATR] or height harmony.
275 Much more historical work, as well as more detailed work on more of the Mara
vowel systems, must be done in order to confirm which hypothesis is correct. The important point, however, is to note that Stewart’s conclusion that the PB system was
most likely i a u sheds light on the possible history of Ikoma’s vowel system, which in turn helps us to understand its unusual synchronic state. If, in the language’s
history, the second degree vowels were instead of e o, with corresponding [+ATR] dominance, it could explain the presence of superficial [+ATR] spreading.
276
Chapter 8: Concluding remarks
Ikoma is a fascinating language, typical of other Bantu languages in many respects, but quite unusual in other ways. Some of the fascinating features include the height-changing
prefixes, the [ATR] markedness paradox and the frontback [ATR] asymmetry in verb stem harmony. The [ATR] markedness issue could have interesting implications for a
theory of vowel harmony systems, since it at first glance it appears to be a potential counterexample to Casali’s inventory-dependent [ATR] dominance hypothesis.
Nonetheless, it seems possible that there could be a formal account which posits [-ATR] as the systematically dominant value, thus conforming to the hypothesis. Ikoma provides
an intriguing new set of data for theoretical linguists, and it has much to teach us about dominance reversal. Further comparative work could also bring significant insights to our
understanding of how languages and vowel systems in particular change over time.
8.1 Summary of previous chapters
In this final section, I briefly summarize the contents of each chapter and then suggest areas for future research. In Chapter 1, I introduced the Ikoma language and situated it
within Tanzania’s linguistically-diverse Mara Region. I also introduced typological issues and vowel formant analysis methodology, both of which are important throughout this
work. Chapter 2 provides additional introductory material, but here I focused on giving background to other aspects of the language outside of the vowel system. I described
nominal and verbal morphology, consonant phonology and tone. Chapter 3 began the analysis of the vowel system, arguing that the language has a
7V inventory with e o as the degree 2 vowels, as well as both phonemic and conditioned vowel length. This analysis was supported by plenteous examples showing contrast in