What about the PRED features?

[Joan Bresnan]

Concerning the discussion of the PRED initiated by Avery Andrews and
recently added to by Ron Kaplan, I would add the following remarks.

The original *linguistic* idea of the PRED feature was to define the
linguistic concept of "nucleus of predication", a grammatical
construct echoing ideas of traditional grammar (e.g. Jespersen's nexus
theory) and certainly playing a fundamental role in syntactic matters
(not just semantics).  The nucleus of predication, or nucleus for
short, can be defined in LFG as a region of f-structure in relation to a
single predicator (PRED).  It defines possible binding domains (as
TENSE also does), it plays a role in case phenomena (case can "spread"
to nonarguments in a nucleus in some languages); it explains certain
properties of "small clauses" without assuming artifactual c-structure
nodes.  In short, it has grammatical motivation, not just semantic
motivation.

Exactly the same doubts about having a PRED feature in f-structure can
be raised about having a TENSE feature in f-structure.  Again the
answer is the same.  Many *grammatical* phenomena depend on knowing the
locus of tense in relation to other grammatical functions.  The fact
that there is a semantics of tense as well does not diminish its
grammatical role and hence its motivation for being in f-structure.

Joan