Posted on Tuesday, September 28, 2004 11:09 PM
John Harris on
'Vowel reduction as information loss'. The abstract:
Vowel reduction degrades phonetic information in the speech signal and should be understood as having an analogous impact on phonological representations. Reduction follows two apparently contradictory routes in vowel space, yielding either centralised values (the ‘centripetal’ pattern) or the corner values a, i, u (the ‘centrifugal’ pattern). What unifies these vowels is the relative simplicity of their acoustic spectra compared to those of mid peripheral vowels. Spectral complexity can be taken as one measure of the amount of phonetic information present in a speech signal at a given time. On this basis, centripetal and centrifugal reduction can both be construed as resulting in a loss of phonetic information. In selectively targeting non-prominent positions, reduction has the effect of enhancing syntagmatic contrasts among vowels. Representing vowels in terms of three basic components manifested as a, i and u allows informational asymmetries of this sort are to be directly recorded in phonological grammars.
It is true that it seems more elegant to analyse reduction to schwa and reduction to peripheral vowels in the same way. It is also true that the AIU framework seems most suitable to this -- Ben Hermans has argued this independently of Harris in a number of talks he gave on reduction in Slavic. It is also interesting to see that Harris and Hermans also seem to have come to the conclusion that Bulgarian is an important example of this. There is one thing which Hermans did discuss, but which Harris leaves aside, and this is the issue of why it is exactly the [A] element which is lost. Hermans tries to relate this to the fact that [A] is the most 'sonorous' element, as far as I remember. I am not sure whether he has written this up somewhere. He should have.
Another thing is that Harris unfortunately ignores some of the problems which Crosswhite raised for an AIU approach. Here is how Hermans explained one of these problems:
[An] impressive example is Pavlikianski Bulgarian. This system has the vowel inventory {i,u,e,o,ɛ,ɔ, a}. Of these vowels only i reduces to schwa!
There are a few other examples which need discussion. It is a pity that Harris does not provide this. (From what I can tell from an abstract I have, Hermans' own solution involves interaction of a constraint which wants all vowels to reduce to schwa with OT type constraints which require vowels which contain A and U to remain faithful.)