Those adjectives

[Mark_Johnson at Brown.edu]

I may get flamed for suggesting this, but I think that most modern
linguistic theories, LFG included, have very little to say about
argument structure except in verbal (and deverbal) projections.
The exception to this is the work in model-theoretic semantics,
which while not oriented towards explaining syntactic phenomena
such as agreement, does provide a systematic account of the 
argument-structure relationships in attributive and predicative
adjectival constructions. 

Our model-theoretic semanticist friends, since Montague at least and
probably back to Russell, have been analysing both attributive and
predicative adjectives (as well as nouns and verbs) as having a single
open argument slot.  They use a positional argument notation (the
order in which the lambda-abstractions appear basically determines
argument structure) whereas LFG uses a named argument notation (i.e.,
argument slots are unordered, and identified by names such as SUBJ,
OBJ, etc).  But modulo this difference, one might try to mimic the
model-theoretic semantic analysis in LFG as follows.

If we name the single open argument slot in the model-theoretic
semantic analysis the SUBJ argument, then the model-theoretic analysis
can be viewed as claiming that _all_ nouns and adjectives have a SUBJ
argument.  Their account of attributive constructions such as Dick
Hudson's ``the good book'' is that the open (SUBJ?) arguments associated
with ``good'' and ``book'' are identified (unified), and in effect
saturated by the determiner ``the''.  

The full generalized quantifier analysis of NPs now standard in the
model-theoretic semantic literature is somewhat more complex, and it
is not clear exactly how it should be formulated using the unification
machinery of LFG.  However, it is straight-forward to formulate the
generalized quantifier analysis of NPs in the simplified version of
LFG I have been developing called R-LFG.

My paper in the LFG'97 proceedings, and the paper I have submitted to
the forthcoming book on glue language approaches in LFG, give more
details on R-LFG (you can fetch them from my Web page on
www.cog.brown.edu), but the basic approach is that by simplifying the
formal machinery of LFG (rather than complicating it, as has been the
trend over the past decade or so), one actually obtains a better
system for linguistic description.

Comments?

Mark

Mark Johnson
Cognitive and Linguistic Sciences, Box 1978
Brown University, Providence, RI 02912 
(401) 863 1670, fax (401) 863 2255