ARG-ST as a head feature

[Carl Pollard]

Hi Martin,

>
But you still agree with Bayer
that there is a fundamental difference between valence features and
other features.
>>

Of course.

>
This will have empirical consequences that make it
quite different from 1994 HPSG.
>>

Yes, like that 1994 HPSG does not allow (at least not as far as I can
see) any kind of reasonable analysis of coordination and syncretism.

>
But how does it differ from Bayer and
Johnson empirically?
>>

I've only read Bayer, and know Bayer and Johnson only from what
others have said about them, but I will return to this question
below.

>
How does one decide what values go into set P?
>>

I was taking P to be a bounded-complete meet semilattice isomorphic
to the subsumption order of (not necessarily totally) well-typed
abstract feature structures of sort _head_ along the lines of HPSG
1994. But the feature representation is unimportant.

>
> This differs from Bayer/Johnson because they CONFLATE
> conjunction/disjunction with neutralization/coordination,

I'm guessing that they would insist that this is intentional and
desirable -- not only is this not a problem in CG, it is the essence
of their analysis.
>>

I know it is the essence of their analysis, but I'm not sure it
is not a problem in CG.

>
> with the undesirable consequences noted by Heylen. (Roughly, you
> can't describe a man with a dissociated personality, where one of
> the personalities is a vegetarian optometrist and the other is a
> coke-snorting gunrunner: the gunrunner gets turned into a
> vegetarian. In case the metaphor is too opaque, being a vegetarian
> optometrist is conjunction; being schizophrenic is
> neutralization.)

I'm assuming you're referring to the discussion of e.g. "her" (which
is ambiguous, not neutral) etc. at the end of ch. 9 of Heylen's
dissertation (also in a separate paper).
>>

No, this is not what I am referring to. I am aware of the distinction
between neutrality and ambiguity, which for Bayer is the difference
between using meet vs. having two lexical entries.

>
Very briefly, in CG it is
possible (and in this case mandated by empirical data) to assign two
different types to "her" (personal pron. vs. possessive pron.).  Not
so in the HPSG lexicon: basically, there is only one lexical entry for
each word, though it may be disjunctive.  This can lead to
overgeneration in conjunction constructions, as Heylen notes.
>>

HPSG94 never had any way to represent neutrality and coordination and
the duality between them, an unpleasant and long-known fact which is
one of the chief reasons I stopped working in this framework several
years ago. This is not at issue. (Not the only reason -- there is also
the long-standing lack of any kind of interpretation of CONTENT
values.)  Calcagno and Levine have a way to do neutrality by
elaborating the hierachy of possible feature values, but this does not
take care of coordination and its dual status relative to
neutralization. (Actually, I think their construction is really the
first half of my construction of the Levy lattice, but is missing
the second half.)

But my second-hand knowledge of Johnson and Bayer is that they use
\wedge for two different things, neither of which is ambiguity: (1)
logical conjunction of properties, as in "HAS is finite and
auxiliary", and (2) neutralization "FRAUEN is accusative and
dative". These are two different kinds of AND's: with respect to the
first one, accusative and dative are inconsistent; with respect to the
second they are not. To construct an example that would be problematic
for J&B, one would look for a form FOO that was neutral (NOT
AMBIGUOUS) between the two feature specifications [F a, G b] and [F c,
G d] (where, in terms of logical conjunction, a and c are inconsistent
and also b and d are inconsistent), and then look for a form BAR that
selected [F a, G d]. That would overgenerate the ungrammatical FOO BAR
(in a head-final language).

Go back to the schizophrenic vegan-optometrist \neut coke-snorting
gunrunner. S/he would be able to get into a club whose membership
was limited to vegan gunrunners, even though s\he had no
personality in which these two properties were co-instantiated.

>
You're proposing a solution to this problem (note that the problem is
specific to a particular formulation of the HPSG framework).  That's
why I was saying that it seems that now you've caught up with CG
again.
>>

Actually, I have not really been thinking of HPSG as I have been
thinking about these problems. Did I mention I don't work in that
framework as of several years standing? It is an empirical linguistic
problem. The whole discussion began with the question of whether
ARG-ST is a  head feature, but the real question is whether the
kind of info that ARG-ST encodes should be manipulated in the
same way as familiar morphosyntactic features, and if so how
to do so in a way that does not run afoul of coordination. The
solution I gave could be coded up pretty easily in either a
TLG-ish or an HPSG-ish way.

>
(BTW, why not go for a simple solution like allowing multiple
distinct lexical entries?  Would we see an empirical difference?)
>>


See above.

> The treatment of ARG-ST as a head feature using the freeze construct
> is not very much like anything I know of in categorial grammar.

>
CG does have mechanisms for locking and unlocking things, though they
have been used for different purposes.
>>

You can always add stuff to any framework to solve any problem!

>
I was trying to epxress concern about how you would market your
proposal to the formal grammar community.  What your treatment is
useful for could be done in CG already.
>>

See above.

>
My concern is not so much of
a theoretical kind (it has been demonstrated that there are
Turing-complete versions of CG, so you can state any HPSG theory in
them, and vice versa), but a practical: how easy is it to do things?
what price do you pay for it in terms of technical machinery?  It
seems that CG is still strong in both respects.
>>

Your position seems to be: if you need something, develop a form
of CG that has it, and that establishes that CG is good. Whereas
if you follow the same strategy in HPSG, then you say "so big deal,
CG has that". I don't really care about the sociological division of
nontransformational grammar. I am interested in finding the right
category to do linguistics in, and have little doubt that whatever
it is, there will be functors embedding it into the category of
suitably-enriched CGs on the one hand and suitably enriched HPSGs
on the other hand.

On a somewhat different note, with respect to your question "how easy
is it to do things", ease is relative to what you know and what is
intuitively easy for a given person to work with. If you learn about
resource-sensitive logics different things are easy than if you learn
about dcpos and lattices. To return the algebraic metaphor, the
set of easy theorems about boolean algebras are not the same for
a ring theorist as for a lattice theorist.

Also, different people have different linguistic intuitions. For
example, most generative grammarians (I use this term in a way that
includes HPSG but not CG) have an intuition that linguistic
expressions should have structural representations. Most CGists seem
to have the intuition that the logic of categories includes
hypothetical reasoning.  [Full disclosure: I don't feel strongly about
either one.]  And what kinds of intuitions people have make different
kinds of analyses easy or hard, for example, to very different kinds
of "easy" analyses of right NCC (BOOKS TO KIM AND RECORDS TO SANDY),
e.g. Steedman/Dowty vs. Stump/Sag et al. The first way makes the
semantics easier, but the second generalizes more easily to gapping.
(No doubt people on  both sides will tell me I am wrong, because
there are things you can do to the frameworks to make
them easier.)

>
But in the end it is the empirical side that counts:
>>

No argument there.

>
can you do more
than CG can do?
>>

That is an unnecessarily tendentious way to put it, in light
of what I just said. There is no fixed official version of CG.
Forget about HPSG vs. CG. The gist of my proposal is just that
the algebra of categories carry two cross-cutting distributive
lattice structures, one for logical conjunction/disjunction and
another for coordination/neutralization. This is different, as far
as I know, from what has been proposed so far on either side
of the sociological divide (but only as far as I know).

Carl