optimality

[Brian MacWhinney]

Optimality Theory is certainly interesting in its own right.  Although
it is a repackaging of things that have been around for quite awhile in
the work of the natural phonologists and morphologists, it is still a
nice recent statement, I would think.  For grammatical functionalism, it
seems to be a glass that is both half empty and half full.  The good
side is that functionalists can increase the scientific rigor of their
work by using formalisms, particularly when they are not "empty
formalisms".  Moreover, there are pieces of OT that make it look like it
is saying something interesting about competitive processing and
constraint satisfaction mechanisms.

However, the actual implementation of constraint satisfaction in OT does
not seem to be as robust in terms of empirical predictions as one might
wish.  And this appears to be the half (maybe even more than half?) view
of the glass.  In particular, as Paul Smolensky noted in response to
questions at Cognitive Science in Boulder three years back, OT has not
yet addressed issues of processing, acquisition, functional distribution
or variation.  In the first versions of the OT book, there was lots of
interesting stuff on Harmony Theory, but later versions seemed to have
this mostly excised and the project seemed focus on just an ordered set
of universal constraints.  I am not competent enough in phonology to
really evaluate these formulations as contributions to phonological
theory, but I understand from those who do understand these things that
some advances have been scored in this area.

On the language learning and learnability front, there are some
interesting accounts for phonological development that work through
successive imposition of the highest ordered or most universal
constraints and a fairly tightly constrained finite model that reorders
constraints one by one.  The algorithm will work to identify languages,
but I doubt that it will predict the right sequence of developmental
data.  If more attention were paid to the dynamics of competing
constraints in connectionist systems, perhaps OT would come closer to
what connectionists typically believe.  What the truth is in the area of
computational modeling of phonological development, is still up for
grabs, given the paucity of model-ready empirical data.

In terms of language processing, I don't see anything in OT that matches
anything I am familiar with.

In terms of variation theory, I would imagine that a Labovian
variationist framework could eventually get hooked up to OT, but right
now, the places to insert those hooks are pretty unclear.

In terms of the extension of OT to functional grammar, there have been
some promising initial attempts by folks at Colorado and perhaps
elsewhere.  At the very least, it seems that OT provides a nice
descriptive framework.  You get to know a lot about implicational
universals.  If a language has X, then it also has Y.  But the reason
for this ordering of constraints seems to lie outside of OT.

We need to move in the general direction of something like OT, but I
think we should probably take the current version of OT as simply a
metaphor for some properties that a universalist model might eventually
have.

Please note that all of the above are not presented as scientific
findings, only as initial reactions of a potentially interested consumer
who has not yet made a purchase.
--Brian MacWhinney