output filters in HPSG?

[Stephen M. Wechsler]

For a paper on case I'm working on, I want to posit a condition of the
following type:

"A node labeled _case_ must be the terminus of some path satisfying the
following description:

|case-inflected-word                   |  "
|SYNSEM|LOC|CAT|HEAD|CONCORD|CASE  case|

Any word showing case inflection (noun, determiner, adjective, etc.) will
be of type case-inflected-word.  The purpose of this constraint is to
capture the following generalization, found in many languages: the case
feature on an NP must be inflectionally realized in at least one place, but
there is some variability as to where.  E.g. in Serbo-Croatian special
uninflected nouns can't appear in dative or instrumental case positions
unless modified by an inflected determiner, adjective, etc.  Related facts
are found in a variety of other languages (German, Choctaw, Gooniyandi,
...).  Since all the elements involved in case-concord share a single
_case_ value, the constraint above does the trick nicely.

However, if I'm not mistaken the condition above introduces some
non-monotonicity into the syntax.  It is, in effect, an "output filter".  I
have not been able to formulate it as an implicational AVM that unifies
with all objects. For example, this won't work--

|                                      |  =>
|SYNSEM|LOC|CAT|HEAD|CONCORD|CASE  case|

|case-inflected-word                   |
|SYNSEM|LOC|CAT|HEAD|CONCORD|CASE  case|

--because this would mean that EVERY case-concord element must be
inflected, which is wrong.  Rather, the condition we want is that AT LEAST
ONE such element must be inflected, for a given CASE feature.

This non-monotonicity (if that's what it is) worried me somewhat until I
noticed that the Trace Principle (Pollard & Sag 1994, p. 400) has a similar
status:

"The SYNSEM value of any trace must be a (non-initial) member of the SUBCAT
list of a substantive word."

So whatever the consequences, at least I'm in good company.

Any reactions to this?  Can someone help me out here?  I realize there are
many versions of HPSG.  Maybe this conforms to some variants but not
others?  Reading suggestions welcome, too.

Steve
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/enriched
Size: 2152 bytes
Desc: not available
URL: <http://listserv.linguistlist.org/pipermail/hpsg-l/attachments/20000602/817b0872/attachment.bin>