Filler-gap mismatches

[Carl Pollard]

>
There's an attempt to clarify this in Andrews and Manning (1999),
where f-structure (and other info) is decomposed into a whole series
of functional projections of different kinds that may (or may not) be
shared between nodes.  The information usually stuffed altogether in
f-structure is broken down into a projection containing the semantic
(syntactico-semantic?) PREDicate and its arguments, a projection
containing grammatical functions (valence), and a "morphosyntactic"
projection containing features such as TENSE and CASE.  (There's also
a category projection and a bar-level projection; don't know about
PFORM.)  The details of what goes in what projection are arguable, but
the basic intuition is that standard LFG f-structure could be seen as
a shorthand for several types of thing.
>>

This seems precisely parallel to the way it has always been done in
HPSG, where the feature structure representing syntactic-semantic
information (roughly, everything except phonology and consituent
structure, what in HPSG is called a synsem object) distinguishes
between CONTENT features (analogous to PRED), VALENCE features, and
HEAD (morphosyntactic) features. This is a perfect example of just the
kind of concergence I referred to in my replies to Yehuda.

Of course there are differences, e.g. in HPSG the part of speech
is a head feature.

Carl