Re: Webpage

On Tue, 6 Aug 2002, Thomas R. Wier wrote:

> Quoting Marcus Smith <smithma@...>:
> > The vowel systems it generates are rather simplistic (no nasalization,
> > glottalization, etc). It can't generate enough vowels to be natural, and
> > my dispersion algorithm isn't very sophisticated, but sometimes it gets
> > interesting results.
>
> Yeah, the distribution of the vowels that I got on several tries
> came out with a skew for a five vowel system, but four of those
> are high vowels (/i/, /I/, /u/, /U/)!  So, maybe it just needs some
> tweaking.
The system produces 5 vowel systems roughly 25% of the time, lower than
the real word tendancy, which is about 31%. I don't recall now why I gave
it that slant. As for the high-vowel slant: I know why it does that, but
I'm not sure how to fix it exactly.

The consonant generator works in terms of manner and place of
articulation. The vowel generator doesn't know anything about height,
backness, roundedness, etc; it works entirely on F1 and F2 frequencies.
After the vowel system is determined on that basis, it maps each vowel to
a phonetic symbol. This is one place where things (sometimes) go wrong.
The boundary between /e/ and /I/ is 420 Hz, so at 419 Hz the vowel is /e/
but at 421 Hz the vowel is /I/. Arbitrary.

This issue doesn't arise much in front-central-back distinctions, except
with the low vowels. There's a strong tendancy to see back-/A/ rather than
central-/a/, partially (but not mainly) due to the arbitrary borderline
between central and back.

I'm not sure how to fix this problem without some complicated scheme of
assigning vowel labels depending not only on the frequency of the vowel,
but also on what other vowel symbols have been assigned, which starts
getting more complicated than the algorithm for figuring out the values in
the first place. *Sigh*

I'm still working on this part (as well as trying to fit in nasalization
and stuff).

Marcus

Reply