In this paper, we have argued that (most versions of) the grammatical frameworks surveyed in
this volume are properly characterized as:
17
• surface-oriented,
• model-theoretic, and
• strongly lexicalist.
In fact it has been an important contribution of work in the LFG, GPSG/HPSG and Categorial
Grammar traditions to show that constraint-based lexicalist grammars are descriptively adequate
or even preferable alternatives to the transformational mainstream of generative grammar.
Grammars with these design properties, we have claimed, can plausibly be embedded
within models of processing that begin to approach psycholinguistic plausibility, allowing, for
example, the incremental computation of partial interpretations. The concerns of psycholinguistic
realism considered here thus add a further dimension of explanatory motivation for the
CBL model of grammar.
Here we have been concerned primarily with issues of sentence processing, but we should
add that similar conclusions about the design of grammatical theory have been obtained by
recent research in language acquisition. See, in particular, Tomasello 2003, 2006.
We have argued that realistic competence grammars – those that can be potentially reconciled
with the facts of performance as we now know them to be – exhibit precisely these
properties. Given that constraint-based lexicalist grammars, as of this writing, also have much
better systematic empirical coverage than any of their transformational competitors, they may
in fact provide the first legitimate grounding for the competence-performance distinction, upon
which all work in generative grammar since Chomsky 1965 has crucially depended.