the right track vs. the wrong track

[Stephen M. Wechsler]

At 1:59 PM -0400 5/3/01, Robert Levine wrote:
>Like many people doing PSG on a long-term basis, I'm in general agreement
>with Steve's points and the kind of examples he's bringing up. But saying
>that `only subjects can raise' is a little dicey. If argument-compostion
>analyses as pioneered by Erhard Hinrichs and Tsuneko Nakazawa, and now
>employed widely in analyses of Romance languages, Korean etc., are valid,
>then don't we have non-subjects raising, in effect? Specifically, if you
>assume that a verb can take a lexical head as one of its complement and
>some sublist of the valence elements selected by that head, doesn't this
>allow non-subject raising along any number of lines?
>
>Just curious---
>
>Bob

Good point-- I wasn't thinking about those (odd since I'll be
presenting such an analysis myself at Trondheim...).

It's a little tricky to reformulate the generalization properly, but
it's something like this:    You can't raise a complement while
leaving the subject to be realized 'in situ' in the complement
clause, as in:

	*Honey tends [bears to like].

	*I expect honey [bears to like]

This follows from a principle that the SUBJ item cannot be cancelled
(via the Valence Principle) before all the COMPS items are (OK after
COMPS; or all at the same level as in inversion and VSO).  I don't
know that anyone has formulated such a principle explicitly, but
maybe we should, since it seems to be tacitly assumed.  It did follow
from the old SUBCAT system (HPSG2?), when 'subject' was defined as
the last item cancelled.  Such a principle differs from the P&P
proposals I mentioned in that it is needed even for simplex clauses.
It is virtually definitional for SUBJ and COMPS.  Or so it seems to
me.

I believe the argument-attraction cases are consistent with this, but
I haven't checked.

--Steve