forwarded from Ron Kaplan Re: Filler-gap mismatches

[Carl Pollard]

>From kaplan at parc.xerox.com Fri May  4 13:42:16 2001
From: kaplan at parc.xerox.com
Subject: Re: Filler-gap mismatches
To: pollard at ling.ohio-state.edu
Date: Fri, 4 May 2001 10:42:17 PDT

(Hi Carl, I tried to send this to the list, but it won't accept messages
from me.  Could you forward it.  Thanks--Ron)

Hi,

  I've been eavesdropping as a friendly observer to the recent set of
interesting discussions, and I actually have comments from a slightly
different perspective on a number of issues.   I will send a longer
message when I have a little more time, hopefully before the trail gets
too cold, but I thought I would just quickly respond to the "he didn't
think of" question.

  From the LFG perspective, these kinds of functional relocations that
seem to violate categorial requirements are not particularly mysterious.
 Categories are c-structure properties that are mostly but not uniformly
correlated with f-structure relations, and subcategorization is defined
on the f-structure and not the c-structure.  So the system naturally
allows for some slippage--things that are realized in different
categories in different positions can be assigned the same functions.
(Indeed, we are very happy when a little slippage is observed, since it
tends to add further support for the modularity assumptions of the LFG
architecture.)  In the original LFG paper (Kaplan & Bresnan, 1982),
long-distance dependencies were characterized as functional as opposed
to categorial identities based on c-structure configurations of traces,
and these "think of" examples were cited as compatible with this
formalization.

  Kaplan and Zaenen (1989) introduced  functional uncertainty as a
better way of characterizing long-distance dependencies, purely in
functional terms and without making use of traces at all (what has now
become a common approach outside of the transformational tradition).
That paper also discusses the "think of " examples, actually in slightly
more detail and with a little more review of the literature on this
topic.  We claimed that the observed patterns follow from two facts:
COMP cannot be at the bottom of an English uncertainty and English
phrase-structure does not allow for S's inside PP's.

  I'm not sure whether or how this proposal could be translated into
current HPSG technology--it may just be a natural way of characterizing
these things only with respect to the rest of the theoretical apparatus
(although of course I don't believe that).

Ron