Global constraints [Was: "Generative" serves them right]

[kay at cogsci.berkeley.edu]

On Sat, 28 Apr 2001, Emily Bender wrote:

> Re: The issue of global constraints and Raul's question about
> where the constraints that make up the Binding Theory "live"
> (i.e., how they fit into the ontology)...
>
> I've always liked the idea of a grammar consisting of different
> pieces (lexical items and constructions) that fit together however
> they can.  All possible combinations license grammatical structures,
> although only some of those structures (e.g., those that unify with
> the initial symbol) count as sentences, and some others count as
> non-sentential stand-alone types.  What I find appealing in this
> is what it says about incremental processing:  When you've only
> dealt with the first three words of a sentence, you can already
> pick some possible structures as possible.
>
> In contrast, any constraint that involves checking the whole thing
> when you're done (e.g., Completeness and Coherence in LFG, the idea
> that all features must have a value in Construction Grammar) means
> that you can't know immediately if a partial structure will turn out
> to be grammatical in the end.  The Binding Theory constraints seem
> to be to have a similar flavor.


Emily,

I think Completeness and Coherence in LFG don't depend on "checking the
whole thing" in the way your words suggest because F-structures can be
very local and C & C aply to individual F-strs. If memory serves, this
point was emphasized by Joan and Ron (and perhaps others) at early days of
LFG, as part of an argument that LFG could provide a story about many
"non-sentential stand-alone types." (I haven't checked 20 yr old
sources to make sure i've got that exactly right, but I think something
like that is reasonably accurate.)  Something not entirely dissimilar can
be said about teh const. Gram. requirement that all features get values.
If you have a word or phrase in which all features have gotten values,
then that is a brass bottomed construct that any CG parser (if there were
any) could assign a full syntactic and semantic analysis to.

p
__________________________________________________________
 Paul Kay                      Department of Linguistics
 kay at cogsci.berkeley.edu       University of California
 www.icsi.berkeley.edu/~kay    Berkeley, CA 94720, USA