ARG-ST on phrases (long)

[Ivan A. Sag]

Hi Stefan,

> I agree that one has to have locality restrictions. Selecting for synsem
> objects made nonlocal selection more difficult, but it didn't make it
> impossible. If one allows for relational constraints, one can dig around
> in structures and find the daughter that corresponds to a selected
> element. Doing this recursively one can implement selections as the one
> above.

In my previous message, I wasn't concerned about the question of what can be
stated in a grammar formalism. There's nothing in the framework (whatever
foundations one assumes) that requires:

1. the introduction of the type synsem,
2. the typing of the SUBCAT (or ARG-ST) values as list(synsem), and
3. stating the Subcategorization Principle in terms of the SYNSEM value
   of the non-head daughters.

But 1-3 together constitute an empirical hypothesis about the way natural
languages work.

One might object that this hypothesis doesn't have enough teeth. Suppose
someone found a selectional dependency in some language like the one I
described in my previous message as never occurring.  Of course one could
posit some HEAD feature --call it FOO-- and rig things so that [FOO +] heads
always select for, let's say, an NP[ablative] complement.  In this way a
higher verb V could select for a [FOO +] sentential complement and that would
guarantee that the sentential complement was headed by a verb that selected an
NP[ablative] complement. The nonlocal selection would be rendered local via
the introduction of an ad hoc feature.

Maybe I'm wrong, but I've never thought that this kind of objection cut
very deep, because there is an aesthetic judgment involved in any linguistic
theory (in any scientific theory).  If, in language after language, one had to
keep introducing new features to transmit information nonlocally and if there
were no independent motivation found for those features, then one would start
to question the whole selectional locality hypothesis. The hypothesis in 1-3
entails at least that any non-local selectional dependency must be mediated by
some feature for which independent motivation should be sought.  That's why
the issue of tough-adjectives and the SLASH analysis is interesting: the
feature that the selectional locality hypothesis forces you to assume turns
out to provide an account of lots of interesting data. That is, it is
independently motivated.

Is there some different take on these issues that I'm not considering?

> To me the locality principle more looks like a guideline for linguists.

As a linguist, I'm interested in something more than this. If selectional
locality had to be a mere guideline, I would feel like the framework I was
using failed to provide me with the proper tools for stating my theory of
language.

Carl, can I ask what alternative account of selectional locality you have in
mind?

All best,
Ivan

-------------------------------------------------------------------------
Ivan A. Sag
Professor of Linguistics
Director: Symbolic Systems Program (2000-2001)

Email: sag at csli.stanford.edu
WWW: http://www-csli.stanford.edu/~sag/sag.html

Dept. of Linguistics              CSLI - Ventura Hall
Fax:   650-723-5666               Fax: 650-725-2166
Office: MJH 040B                  Office: Cordura 228
Phone: 650-725-2323               Phone: 650-723-2876
-------------------------------------------------------------------------
Stanford University - Stanford, CA  94305  USA
-------------------------------------------------------------------------