Saturday, November 14, 2009

More Adventures in Phonology

Last month in our phonology class we analyzed the phenomenon of consonant dissimilation in Kikuyu (Gikuyu). This is the exercise given on p. 107 of Morris Halle's Problem Book in Phonology (1983).

The first part of the exercise is to determine the base form of the infinitive prefix, stating a rule that accounts for its pattern of alternation. The two alternating forms of the prefix are [ko] and [ɣo]. [ko] appears before a verbal root beginning with the [+cont] consonants {r, m, n, h} and the consonant [ɣ]. [ɣo] appears before a verbal root beginning with the [-cont] consonants {t, c, k} and the consonant [ð]. From the observance of other data (particularly that in part two), it is obvious that /ko/ is the base form while [ɣo] is the alternate. Therefore, we are dealing with something like the following preliminary rule for dissimilation:

/k/ → [ɣ] / __ V(V){t, c, k}V

/k/ also becomes [ɣ] before [ð]. What distinguishes [ð] from {t, c, k} above is that [ð] is [+cont, +voi]. From the data (for example the loanwords from English in part three), we see that Kikuyu has very few voiced consonants, and the consonant /g/ always undergoes spirantization, except when prenasalized. [ð] is a bit more difficult to analyze. We can note that fricatives in Kikuyu are always voiced. And the English loanwords in nos. 46 and 47 show that English [s] is pronounced as [ð] in Kikuyu. Thus an /s/, /d/, or even a /t/ could be the underlying representation for [ð].

From the dissimilation rule above, it appears that the wildcard [ð] should be analyzed as an underlying /t/, as this gives the rule a clearly-definable phonetic context. And as a voiced /g/ likely lies behind [ɣ], we can modify our dissimilation rule above and add a spirantization rule as follows:

(1) Voicing Dissimilation Rule

/k/ → [g] / __ V(V){t, c, k}V, where [t] = [t] or [ð]

(2) Spirantization Rule

C[-cont, +voi] → C[+cont]

Thus these ordered rules state that (1) in two consecutive [-cont, -voi] consonants, the first consonant is dissimilated with regard to voice, and (2) voiced consonants are spirantized in all environments except after nasals (post-nasal hardening).

Applied to the data, it looks like this:

/ko-geera/ /ko-taaka/
(1) ko-geera (1) go-taaka
(2) ko-ɣeera ( ) go-daaka
(2) ɣo-ðaaka

For the second step of the /ko-taaka/ form, the voicing of the /t/ appears to be due to some factor not specified. As no. 12 [ko-hetoka] shows, such voicing doesn't always occur. The forms that do have voicing, i.e., no. 10 [ɣo-ðaaka] and no. 20 [ɣo-ðɛka], show voicing of /t/ when it appears at the beginning of a verbal root. Such voicing could be the result of morphological or other phonetic factors such as accent or tone. I wish the author of this book would fill us in on what is going on here.

When I first wrestled with this problem, I had assumed the following dissimilation rule:

/k/ → [ɣ] / __ V(V){t, c, k, d}V

The rule states that dissimilation occurs only in [-cont] consonant sequences, omitting the [-voi] condition seen in the rule above. I also assumed that [ð] came from an underlying /d/. Note that the essence of this rule is not about voicing dissimilation, but about obstruent dissimilation.

The problem with this view is why a similar dissimilation process does not occur before [ɣ], which, parallel to /d/ → [ð], should be analyzed as coming from /g/. I had assumed that the underlying representation, in contrast to /d/, could be analyzed as /ɣ/. And since a dissimilation of /k/ to [ɣ] before a following /ɣ/ would by its very nature defeat the purpose of dissimilation (it would actually be a process of assimilation!), I considered /k/ to be exempt from the dissimilation rule in such cases:

/ko-ɣeera/
(1) Not Applicable

The problem with this analysis is that it fails to give a unified account for the ordering of our Dissimilation Rule (1) and our Spirantization Rule (2). For if we assume /ko-ɣeera/ for the underlying representation as above, and not /ko-geera/, we contradict ourselves by setting up at the same time the underlying form /ko-daaka/. We could say that spirantization of voiced consonants occurred at different stages for /d/ and /g/, but that would be ad hoc. The merit of our first analysis is that consistently unspirantized forms can be set up in the underlying representation and spirantization is assumed to occur at one step in the process. See the forms below:

/ko-daaka/ /ko-geera/
(1) ɣo-daaka (1) *ɣo-geera
(2) ɣo-ðaaka (2) *ɣo-ɣeera

In this analysis, dissimilation in stage (1) already generates spirantization for /g/. It is strange then to postulate another spirantization rule in stage (2). The merit of this approach, though, is that we don't have to assume anything for the changes in [ɣoðaaka] and [ɣoðɛka], but this is outweighed by the other problems with such an analysis.

A search on the internet reveals that this process of dissimilation in Kikukyu has been defined as Dahl's Law, which states that the first of two voiceless consonants will be dissimilated to its voiced version in certain contexts.