Fwd: Re: clarification on query

[Luis Casillas]

Oops, forgot to Cc: the list on this.

----- Forwarded message from Luis Casillas <casillas at stanford.edu> -----

Date: Fri, 27 Jun 2003 01:11:30 -0700
From: Luis Casillas <casillas at stanford.edu>
Subject: Re: clarification on query
To: Andrew Carnie <carnie at U.Arizona.EDU>

Let's see if I get all this (roughly) right.  However, I'm not so much
going to try out to answer your question in its own terms, because what
I think one has to do is to see how HPSG works *in its own terms*.

1. The fundamental notion in HPSG is a sign. These are modeled as
   feature structures of type "sign", with PHON and SYNSEM as
   attributes.  Intuititively, a sign is anything that has all of the
   following: a phonological representation, grammatical behavior, and a
   semantic interpretation.

2. Some signs are atomic.  They don't have other signs as proper parts.

3. There are complex or phrasal signs, which have other signs as parts.
   There are many ways of implementing this idea within the framework,
   but I'll just use the one from Pollard and Sag (1994): phrasal signs
   have a feature DTRS ("daughters"), which in turn has features whose
   values are the component signs.  E.g. for a head-complement
   structure, DTRS has subfeatures HEAD-DTR and COMP-DTRS; the first has
   as its value a sign, the other a list of signs.

4. There are constraints on the relationship between the PHON of a
   complex sign and those of the signs in its DTRS feature: linearization.

   A simple-minded approach here would be to have a constraint that
   says, intuitively, "the PHON of a head-complement phrasal sign is
   the PHON of the head daughter followed by those of its complements"
   (for head-first order; reverse this for head-final phrasal types).
   But this assumption can be relaxed, and is indeed relaxed in the work
   of people like Reape and Kathol (see Kathol's _Linear Syntax_ for an
   overview).

5. In general: an HPSG sign is a feature structure that can embed other
   signs as parts, by its DTRS feature.  The use of trees in HPSG is a
   notational convention to represent a complex signs and the simpler
   ones that it embeds.  The "real" theory doesn't properly have
   syntactic trees, as normally understood.  It's a useful analogy, but
   one shouldn't try to push it too hard.

6. Note that some ideas which for linguists (and even for CS people) are
   in the same semantic field as "syntactic tree" just don't apply for
   HPSG.  E.g.: the "string yield" of a tree.  In an HPSG phrasal sign,
   the parts and the embedding structure both have a PHON.  As an image
   to help you see this: it's as if each node in a syntactic tree
   carried its phonological string as a feature, instead of having this
   read off by looking at the terminal nodes that it dominates.  The
   phonological string of a sentence is at the top of the "tree", not
   the bottom.  (And in the work by Reape and Kathol mentioned above,
   the PHON value of a sign does not have to correspond to the string
   yield of the PHON value of its terminal "daughters" in the simple,
   familiar way of context-free grammar; e.g. the distribution of words
   does not have to respect the "boundaries" of "constituents".)

   Similar comments apply to "domination"; HPSG does not use a notion of
   domination to represent grammatical relations in the first place, it
   uses ARG-ST and VAL features.


> I think my previous message wasn't clear. I'm really looking at
> clarification over the notion of domination. What I'm asking is if
> given some node X that dominates Y and Z,  does
> 	-X  *replace* Y and Z (as in  TG); meaning that the representation
> 		contains only Y and Z; X is a historial artifact (or vice
> 		versa, the representation contains only X, and Y and Z are
> 		historical artifacts.

There is no notion of "derivation" in HPSG.


> 	- or does X *contain* Y and Z (as in MP), there is one object in
> 		the representation (X), which contains all the material
> 		formerly in Y and Z. Y and Z no long exist except
> 		derivationally

I like the word "contain", but you use it in a very un-HPSG way here;
the expressions "formerly", "no long" and "derivationally" are out of
place.  Everything is *simultaneous* in HPSG.  There is a sign, and it
contains other signs as parts.  The way this is implemented is by means
of statements in a logical language.  A grammar is a *theory* in the
sense of mathematical logic, i.e. a set of statements, and signs are the
*models* of the theory.

To understand HPSG "from inside", you ought to think less in the
terms of formal language theory, and more in those of model theory in
mathematical logic.  When you consider the set of models for e.g. a
first-order version of Peano's axioms for arithmetic, none of the axioms
"goes first"; the structure has to satisfy them *all at the same time*.
You don't first "derive" the number 0, use it to derive 1, throw away 0,
use 1 to derive 2, etc., to get to 35.  You have a model with the
natural numbers as the domain, and the successor relation.

In a perfect world, a course in formal semantics would be a prerequisite
for one in HPSG.  Scratch that: a "real" course on first-order logic
would be the prerequisite.  ("Real" as in "covers most of the material on
Enderton's textbook").


> 	- or does it *represent* Y and Z (as in GB), X, Y and Z are all
> 		identifiable objects in the representation (and
> 		derivation). They are related through structural relations.

I am, again, kind of ok with the word "represent" until you bring GB in;
I'd say that the sign X "represents" the signs Y and Z *as in HPSG*.
(Yes, this is a perverse statement, but the temptation to try and
understand one framework in terms of the other *must* be rejected.)


> 	-  or something entirely different?

I'm inclined towards this one (though I'm not sure I understand how a
difference can be "entire" in some absolute sense, but I'll leave the
philosophy out of it).  There are three signs, call them X, Y and Z.
X is such that it has a feature DTRS (it is a phrasal sign), and that
among the signs you find in this feature, you find Y and Z.

You really ought to read the first chapter of Pollard and Sag (1994).

--
Luis Casillas
Department of Linguistics
Stanford University


----- End forwarded message -----

--
Luis Casillas
Department of Linguistics
Stanford University