LINGUIST List 3.751

Wed 07 Oct 1992

Disc: Unification vs Constraint-Based Grammars

Editor for this issue: <>

Directory

1. Avery Andrews, unification- vs. constraint- based grammar

There doesn't seem to have been much discussion on Linguist recently,
so here's an attempted starter ...

People who work with what used to be called `unification-based' theories
(LFG, GPSG, HPSG, CUG, ...) have recently taken to describing them
instead as `constraint based'. It seems to me that this is a
mistake, since *all* current more-or-less formal syntactic theories
that I know about are constraint-based in the intended sense,
including especially, GB, but also RG, systemic functional grammar,
Michael Kac's CORG, etc. (among other things, it seems to me that
one of the original appeals of Lakoff's Global Rules approach to TG
was that he presented it as a constraint-based view. At least,
that's something I remember liking about it at the time).

What distinguishes the `unification-based' theories from the others
is precisely the extensive use of unification, which can be thought
of as a mechanism of omni-directional feature-copying & consistency-checking,
which really does behave differently from the sorts of formal devices
that generative grammarians are used to working with. The idea that
the effects of movement, binding, adsorption (of case, etc),
feature-percolation, & various kinds of co-indexing are all the results
of unification might be wrong-headed (maybe even an instance of the
Postalian Best Theory fallacy), but it is an actual idea that one can
regard the various u.b.t.'s as follow-ups to, and one that is arguably
worth working on, since unification is a rather simple minded idea
(anybody can write a unifier, but implementing the effects of the above
list of GB mechanisms would be a pretty challenging task, I think).

If you de-emphasize or abandon the central role of unification in
these theories, it's not clear to me that they will then amount to
much more than balking at one or another aspects of currently
popular GB analysis, which may be OK in itself, but it maybe not
enough to base a whole theory on.

 Avery.Andrewsanu.edu.au