THEORY: vowel harmony (longish)

On Thu, 24 Jun 1999, I wrote:

> I wouldn't put any money on it if I were you. It has already been noted
> in recent discussions that the most basic contrast in vowels is one of
> height (thus some have argued that Kabardian really has only two vowels:
> <schwa> and [a]; I personally find the arguments compelling). Height
> harmony would neutralize this very basic kind of contrast by requiring
> all vowels within a harmonic domain to be of the same height. As far as
> I know, this situation does not exist in natural languages. Features
> which do trigger harmony include: fronting/backing, rounding, or tongue
> root advancement/retraction. Height is sometimes involved in harmony;
> there are systems where vowels harmonize for some feature (usually
> rounding) iff they are of like height--Yawelmani/Yowlumni is like
> this--but height contrasts are never neutralized in such a system.
To which Matt replied:

> Aren't there languages with height dissimilation, such that all the
> vowels in a word must be either mid or non-mid?  E.g., /e/ and /o/ can
> co-occur, as can /a/, /i/, and /u/, but /e/ cannot co-occur with
> /i/, for example.  I seem to remember that there are such languages.
> Could that be considered a sort of vowel harmony?
Which reminded me that a paper (and dissertation) recently appeared
which contains a discussion of a height-harmony system in Shona. In that
language, some prefixes have two forms: Ce- and Ci-. The form with the
mid vowel occurs on a stem whose initial vowel is also mid; the other
form occurs elsewhere. (I think stems also harmonize in this way--either
all vowels are mid, or none of them are). So I stand corrected; there is
height harmony. But I stand by my original statement; height contrasts
are never neutralized in such a harmony system. (Of course, this is true
in any harmony system; contrasts will never be neutralized.)

I then went on to describe the harmony system of Nez Perce, where the
harmonizing sets don't seem to share much in the way of phonetic
similarity, making the harmonizing feature rather unobvious. These sets
are:

        R = {i, ae, u}
        D = {i, a, o}

Recall that, within a Nez Perce word, if any morpheme contains a vowel
of set D, then all vowels in the word will be from this set; otherwise,
all vowels in a word will be of set R.

Several people guessed that tense/lax might be the harmonizing feature,
and this is indeed my analysis of it, although I use the somewhat
different feature of tongue root retraction rather than "lax", since
"lax" implies the absence of a tensing gesture, while what is happening
in Nez Perce is an active retraction of the tongue root. The twist is
that the vowel [i] belongs to both sets. This vowel is pronounced with a
great deal of tongue root advancement (rather than retraction), so it
doesn't seem to belong in the D set.

One of the properties of natural language phonology that linguists are
coming to understand more and more is the role that physiology and
acoustics plays in the more abstract sound pattern of a language. (This
has always been appreciated, but not well understood, and never
formalized.) Nez Perce provides a nice example of this; it's just easier
to make a [i] with tongue root advancement, so even though the
harmonizing feature is tongue root retraction, [i] will be an exception
and will be pronounced with tongue root advancement everywhere.

Whew! That's a lot of phonology. Gotta luv it!

Dirk

--
Dirk Elzinga
dirk.elzinga@m.cc.utah.edu                   "All grammars leak."
http://www.u.arizona.edu/~elzinga/                 -Edward Sapir