Is OT a theory of syntax?

[Joan Bresnan]

Another response to Dan Everett's reply to Nigel Vincent's remarks on OT.

The important thing about OT is that it is a system for evaluating
candidate outputs (of any sort).  That's why it is sometimes referred
to as a theory of comparative grammaticality (e.g. by Grimshaw,
Prince, and Smolensky).  Comparison of forms, rather than derivation
of forms, is the key.  A grammar is, at bottom, just a harmonic
ordering of the infinite set of candidate structures (Smolensky,
"Generalizing optimization in OT: a competence theory of grammar
`use'", paper presented at the Workshop on Optimality Theory and
Cognition, Center for the Study of Language and Information, Stanford
University, December 6, 1996).

The "input" is simply a baseline or index to the infinite set of
candidate outputs, selecting the candidates most relevant for the
evaluation along certain dimensions.  There is no derivational
relation implied between input and output.  For this reason, Legendre,
Smolensky, and Wilson, in their MIT (Is the Best Good Enough?) paper,
don't even use the term "input", but speak instead of the "index".
Another alternative to "input" would be "(relevant) content".

While OT is a theory of the comparison of structures, there must also
be a theory of structures (a theory of GEN) and a substantive theory
of the constraints (CON) by which structures are evaluated.
OT does not dictate either the content of the constraints proposed, or
choice of GEN.  That is left to theories like P&P, LFG, functional
syntax, etc.  

Quite a bit of interesting work in functional/typological theories of
language supports the existence of markedness hierarchies in
morphosyntax parallel to what has been made good use of in OT
phonology (see, eg. Aissen's recent work making use of the person,
relational, and animacy hierarchies).  I personally think this is one
of the more promising directions in which to search for a theory of CON.

So I think Nigel is right that OT is compatible with a variety of
syntactic theories, not a competitor theory.  


Joan

> Nigel Vincent questions whether OT is monostratal or in fact whether terms
> like monostratal or monolevel ought to apply to it, since it is a way of
> interpreting other theories. 
> 
> But it seems to me that it is misleading to think that OT is so
> chameleon-like that it takes on the architecture of whatever theory it is
> interpreting. No matter what theory it is interpreting, OT will require
> that the theory have an input, an output, and some statement on the GEN
> function's range (i.e. the inviolable constraints of that theory). 
> 
> This is the minimum. That part is invariant. So I do not think that I can
> agree if what Nigel is claiming is that one cannot characterize OT as
> having X-levels. It has two, plus GEN, and this doesn't vary. Other strata
> (e.g. if one were rendering an RG analysis in OT terms) would have to be
> stated in terms of constraint-defining properties of the output, not as
> separate strata. This is the cost of doing business in OT. 
>  
> Or have I missed something?
> 
> -- Dan