THEORY: phonemes and Optimality Theory tutorial

On Mon, 6 Nov 2000, jesse stephen bangs wrote:
>
> dirk elzinga sikayal:
>
> > Hey.
> >
> > Interesting that in the discussion on pedagogical methodology this
> > provocative statement of John's slipped by without comment ...
> >
> > > In particular, the phoneme may be dead [...]
> >
> > Dirk (who, BTW, agrees with this sentiment)
>
> You shouldn't have baited me <evil grin>.  I let it go the first time
> because I was looking forward to my weekend and didn't feel like
> responding, but . . .
>
> I doubt that the phoneme is dead, it's just been disguised a little.  The
> old, structuralist requirements of biuniqueness and minimal pairs have, of
> course, been abandoned, but the idea of a "phoneme" as a unit of language
> information still exists.  Even in the most radical generative approaches
> to language, there is still a finite set of underlying symbols to be acted
> on: phonemes, but more abstract ones.
>
> I don't know if there's a term for it, but it does seem that there needs
> to be a way to distinguish the classical phoneme from the generative
> phoneme.  One is the phonetic information relevant to the speaker's
> comprehension, and the other is the underlying unit from which the surface
> phonetics are decided.
And on Thu, 9 Nov 2000, Jeff Jones wrote:

> I got a couple days behind and waiting until I was caught up before posting.
> Aren't phonemes defined as the sounds of a particular language which a
> naive monolingual can distinguish? If so, then phonemes definitely have a
> psychological reality.
I thought I'd reply to both of these comments at once.

A way of thinking about the phoneme is to consider it to be the
minimal unit of sound which serves contrastive function. Thus any
feature or property of a sound which does not function contrastively
may not be part of the phoneme. This actually allows quite a bit of
latitude. In Shoshoni for example, the 'phoneme' /p/ is realized
variously as [p], [b], [B], and [F] (the last two voiced and voiceless
bilabial fricatives, respectively). So what is necessary for the
Shoshoni speaker? Not the fact that /p/ is voiceless, since there is
no /b/ which contrasts. Not the fact that it is a stop, since there
are no /B/ or /F/ which contrast. Only the fact that it is bilabial
and oral (rather than nasal; there is a contrasting /m/) seems to be
relevant.

In the mid-80's, this would have been seen as an argument for under-
specification; a theory of phonology which includes underspecification
only specifies those features which are sufficient to distinguish one
phoneme from another or whose presence can be predicted. However,
nowadays many phonologists subscribe to a view of phonology that
maintains that significant phonotactic generalizations are to be made
on the basis of surface patterns alone. The grammar on this view
consists of a "filter" which excludes candidate forms which do not
optimize the realization of phonotactic constraints. This means that
*any* underlying structure which gives rise to the correct surface
pattern is legitimate. Thus in Shoshoni, any of [p], [b], [B], or [F]
(or variously underspecified versions of them) are viable candidates
for the underlying phoneme, since they all can function contrastively.
This was known in the American Structuralist tradition as the
"non-uniqueness" problem (as first described by Yuen-Ren Chao in
1934).

Optimality Theory is a current phonological theory which takes the
notion of surface constraints seriously (there are others), and which
provides a way of rethinking the phoneme. Here's how it works. Assume
that there is a phonotactic constraint operative in Shoshoni which
requires that obstruents be voiceless. This is a reasonable
requirement and seems to be true of a great many languages. Let's call
it OBS/VOI. Now take a word such as [pia] 'mother'. Now let's take
/pia/ to be the underlying form and "generate" some candidate surface
forms [pia], [bia]. (This "generation" of candidates is actually quite
important. The theory requires that there be a set of forms for the
grammar to select from; these forms are generated by making various
changes to the underlying form.) It should be clear that the
constraint OBS/VOI will only allow the candidate [pia], since the
other candidate [bia] has a voiced obstruent. This can be represented
graphically in a table (best viewed in monowidth display):

           /pia/ | OBS/VOI
           ------+--------
        -> [pia] |
           [bia] |   *!

The arrow indicates the attested surface form, "*" indicates
a violation of a constraint, and "!" a violation which eliminates a
candidate from consideration. Now, given the underlying form /bia/,
and the same set of candidates, the same result is obtained:

           /bia/ | OBS/VOI
           ------+--------
        -> [pia] |
           [bia] |   *!

What these two examples show is that it doesn't matter if the
underlying form contains a /b/ or a /p/; the grammar picks the same
candidate in each case.

Shoshoni does have surface [b], contrary to the requirements of
OBS/VOI, so how do we account for its presence? Besides the
requirement that obstruents be voiceless, there also is a phonotactic
constraint requiring obstruents (stops in particular) to be voiced
following nasals. Again, there is good evidence for such a move. Let's
call this constraint VOI:N_. Since there are voiced stops following
nasals in Shoshoni, this constraint obviously takes precedence over
OBS/VOI. To see this, let's consider the word [nambe] 'foot'. Assume
an underlying form /nampe/, and two candidates [nampe] and [nambe].
The table looks like this:

           /nampe/ | VOI:N_ | OBS/VOI
           --------+--------+--------
           [nampe] |   *!   |
        -> [nambe] |        |   *

Since VOI:N_ outranks OBS/VOI, any violation of it will disqualify a
candidate in favor of one which satisfies it, even at the cost of
violating lower-ranked OBS/VOI. Thus, the grammar selects [nambe] as
the best surface form.

Again, if we assume a different underlying form, viz. /nambe/, the
same candidate is selected as the surface form:

           /nambe/ | VOI:N_ | OBS/VOI
           --------+--------+--------
           [nampe] |   *!   |
        -> [nambe] |        |   *

Since the choice between underlying /p/ and /b/ doesn't seem to
matter, either one's status as a phoneme in the structuralist or
generative sense seems to be questionable. Thus Optimality Theory
elevates the "non-uniqueness problem" to the status of a grammatical
principle by making requirements only on surface forms and takes the
wind out of the phonemic sails.

Dirk

--
Dirk Elzinga
dirk.elzinga@m.cc.utah.edu