How more/less abstract is HPSG?

[Raúl Aranovich]

Hi all. I've been following this discussion with interest. One remark by
Emily Bender caught my attention. She writes:
"From my point of view, GB/MP is more abstract than HPSG in at least
two ways: (1) it is less precise, in that many of the principles
etc. that make a GB/MP grammar "go" are only codified in prose, and
often not explicitly spelled out at all and (2) while HPSG is shaped
by an aspiration to performance-plausible competence grammar, GB/MP
isn't."

My comment is about Bender's remark (1). It si true that, compared to all
the claims made in Minimalist analyses, HPSG has an explicit explicitly
spelled out way of dealing with features and unification. Moreover,
eveything in the theory has an explicit ontological status, which is where
I find the robustness of the theory: every object in the theory is assigned
to a type. Nothing of that sort can be said about GB/Minimalism/PP.
Presumably, linguistic principles, stated as constraints on types, should
also find their place in the ontology of HPSG. The principle that requires
the value of ARGST to be identical to the concatenation of SUBJ and COMPS,
for instance, is one such explicit statement of a theoretical constraint on
all signs (it is part of a theory because things could have been otherwise,
right? It is not a constraint that sprouts from the formalism adopted, it
is imposed by us as a hypothesis about what a possible sign is). However,
there are many other principles that have not been yet formalized as types,
and are still in "prose". As far as I know, that is still the case with the
Binding Theory in HPSG. It is very clear what the notion of o-command
employed in Pollard & Sag 1992 is, and how to define it over an AVM, but
Principle A (for instance) is still stated in prose (one can formalize it
as a set of AVMs with all possible occurrences of antecedents an anaphors
in o-command relations, but that is not a principle any more!). My feeling
is that HPSG has been steadily enriched with 'functional' or 'relational'
notions and principles which have a hard time finding their place among the
traditional set of types and constraints. They may be suitable for
computational implementations, but that does not mean that they have a
place in the formal ontology of the theory. That is just prose written in
binary code.

I have muloled over these issues for a long time, so I would appreciate
comments, corrections, or suggestions on these ideas. Even though a
complete formalization of the theory as constraints on types is the Holly
Grail which all HPSGers should pursue, I think we should keep ourselves
hones, and realize that we are not there yet.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/hpsg-l/attachments/20010428/84dbf7d1/attachment.htm>