121
4.1 Vowel co-occurrence patterns
As noted in §3.1, all seven phonemes, both long and short, occur in V1 position in both nouns and verbs. We also saw that, as expected in Bantu languages, for each vowel there
are examples of canonical CVCV noun stems in which both vowels are identical. There are, however, many other possible combinations of vowels in multisyllabic stems, as well
as restrictions on which vowels can co-occur. The first syllable of the stem has a privileged position in Bantu, and it is common
for all vowels to be allowed in that position Hyman 2003; Beckman 1997. On the other hand, the second syllable is less prominent, and restrictions on the second vowel often
depend on the first vowel of the stem. Hyman notes concerning Bantu languages “Underlying vowel distribution within specific morphological slots and morphological or
prosodic domains is highly restricted in both 7V and 5V languages” 2003 45. Though these tendencies are true in Ikoma, we also see an interesting pattern in which some stem-
initial vowels have been affected by raising as a result of a following high vowel. Therefore, we might argue that the stem-initial position in Ikoma is not quite as
prominent as that in other languages in which the stem vowels are invariant. This section explores the underlying and surface vowel distributions of noun
stems, not only for the purpose of documentation but also as evidence for diagnosing [ATR] markedness asymmetries. Co-occurrence patterns in a seven-vowel system can be
helpful clues in understanding the nature of vowel harmony in the language, as discussed further below. Some of the key questions are the following Can e o co-occur with ?
Can a occur symmetrically with i.e. both before and after all other vowels? Can the high vowels i u occur symmetrically with all other vowels? The table in 95 below
122 answers these questions, showing the co-occurrence patterns in bisyllabic CVCV noun
stems. I have excluded long vowels in these counts, because there are some distributional differences, as discussed below.
47
The total number of bisyllabic nouns represented below is 221.
48
95 Co-occurrence patterns in CVCV noun stems
V2 V1
i u
e o
a i
7 -
3 16
13
u 7
14 6
3 11
e 7
3 10
2 6
o 15
2 6
10 1
8 5
5 2
15 1
a 7
3 7
6 1?
2 18
As the table shows, mid vowels must agree in [ATR] value. Disharmonic combinations such as [ …e] or [o… ] are not allowed. These blocks are shaded in dark
gray, indicating that these gaps are expected and that the patterns are unattested. It is common for languages exhibiting harmony to prohibit the occurrence of contrasting
values of the harmonic feature within the same word. For example, in the Bantu C languages, which have the same 7V inventory as Ikoma, Leitch 1997 notes that the
overwhelming majority of languages avoid mid-vowel sequences with different [ATR] values. The languages vary on the strictness of harmony with other vowel combinations,
but the tendency is to have much stronger restrictions on these mid vowel sequences. If [e o] and [ ] were allowed in the same stem, that could be evidence that the
degree 2 vowels are actually [ ], not [e o], since in that case the degree 2 and 3 vowels
47
In particular, short [ ] do not occur before [i u] because of raising. However, long [ ] and [ ] are not subject to the same raising, allowing them to occur on the surface before the high vowels.
48
Note that this analysis is based on many more than 221 nouns. The figure here is limited here by the exclusion of long vowels, and of monosyllabic and trisyllabic stems.
123 would both be [-ATR]. If they were both [-ATR], there would be no reason for these
vowels not to co-occur. We would, however, expect restrictions on the combination of degree 1 and 2 vowels, since those would be the contrasting [ATR] pairs. Note, however,
that Ikoma shows clear evidence of symmetric co-occurrence of the degree 1 vowels [i u] with the degree 2 vowels [e o]. Therefore, the distribution patterns further support the
hypothesis that Ikoma’s degree 2 vowels are the [+ATR] vowels [e o].
49
Concerning the low vowel combinations, there is generally symmetric co- occurrence of [a] with all other vowels, though [a... ] and especially [a… ] are
questionable. As the example numbers above show, it is much more common in Ikoma to have the combination [a…eo] than [a… ]. Nonetheless, two examples have been found
which seem to have [a… ] in the stem, and there is one possible example of [a… ] as well.
50
Some Bantu C50 languages e.g. Mombesa, Likile, Lokele, Olombo have sequences of [a] followed by both [+ATR] and [-ATR] mid vowels as well. Leitch
considers the [a… ] pattern to be quite anomalous in comparison to the other Bantu C languages and hypothesizes that it is the result of a historical retracted suffix, not any sort
of phonological conditioning factors 2006:193, 195. Depending on the formal account one adopts for Ikoma, we might want to treat the few [a… ] forms as exceptional.
51
49
Another possibility which we should consider is that there is no stem harmony at all. If there were no constraints on vowel co-occurrence in stems, then we could not use these patterns as evidence for the
inventory and harmony type. However, the fact that the co-occurrence table shows clear patterns is highly suggestive of stem harmony. Also, there is clear evidence of both prefix and suffix harmony, as described
in the following chapters, and many have noted that stem harmony is often a prerequisite to affix harmony Leitch 1997; Hyman 1999.
50
The allowance of the [-ATR] vowels [ ] following [a] is different than the common patterns in Bantu C languages, in which [a… ] is often not allowed. However, in Bantu C there is also a clear hostility
between the features [low] and [-ATR], which we do not see in Ikoma. Instead, in Ikoma [ a ] generally all pattern together as [-ATR] vowels.
51
As previously mentioned, the mid vowels can be surprisingly difficult to distinguish, and much more so when listening to word-final vowels which are often of shorter duration and sometimes even voiceless.
Measuring the formants of word-final vowels also proved to be tricky and sometimes unreliable. For these
124 Looking back at the table in 95, the other boxes shaded in lighter gray indicate
unattested combinations of [ ] with high vowels. There is neutralization of the [ATR] contrast in mid vowels both preceding and following high vowels, so that only [e o] can
occur with [i u]. Though it is hard to be sure, it seems likely that the cause of neutralization preceding [i u] is perhaps different than the cause of neutralization
following [i u]. Neutralization of to [e o] in the stem-initial syllable is the focus of the following section §4.2, in which I give evidence of at least historical or lexicalized
raising of , which explains the lack of examples of [ …iu]. However, concerning neutralization of the second stem vowel, it is difficult to know for sure whether or not the
lack of [ ] is a result of raising as well in this case, it would be rightward-spreading [+ATR] assimilation, or if it is because the [-ATR] mid vowels are “marked” and thus
have a more restricted distribution.
52
Where we see raising elsewhere in the language, it is always leftward, so rightward [+ATR] spreading would not be the clearest explanation
for the lack of [iu… ] examples. Another way to look at it is that the neutralization of the mid-vowel contrast to [e
o] following [i u] could be a case of either “emergence of the unmarked” passive neutralization or “submergence of the unmarked” asymmetric assimilation following
Rice 2007. If we go with the hypothesis that [-ATR] is marked in the stem, I suggest that the avoidance of [iu… ] sequences is an instance of passive neutralization.
53
Formally, it could be explained by a requirement that [-ATR] is left-aligned in the stem. In the case
reasons, I acknowledge that more data could solidify the existence of some of the less attested combinations.
52
If the [-ATR] set were unmarked, we would expect to have the wider distribution and to occur in instances of passive neutralization.
53
As I discuss later, additional evidence of [-ATR] markedness is found especially in the [-ATR] vowels triggering prefix alternations and in distributional restrictions on [-ATR] vowels.
125 of a high vowel in the stem-initial syllable, [-ATR] cannot be left-aligned, and if [-ATR]
cannot associate to left, then it also cannot associate to the vowel to the right that is, to the second syllable.
A final note concerning high vowels is that they are not altered by any surrounding vowels; they do not become [-ATR] when adjacent to [-ATR] vowels. Also,
the gap for the sequence [i…u] is unexplained, since it should in principle be a possibility in any type of [ATR] or height harmony system.
4.2 Mid vowel raising