from Chris Kennedy

[Ivan A. Sag]

FYI...

-Ivan

Date: Mon, 30 Apr 2001 02:46:36 -0500
To: "Ivan A. Sag" <sag at csli.Stanford.EDU>
From: Chris Kennedy <kennedy at northwestern.edu>
Subject: Re: good news from generative grammarians (fwd)

>PS:  I just noticed that my submission to the HPSG list got bounced,
>because I'm not a member.  I've attached it below --- would you mind
>posting it for me if my original msg doesn't go through?  Thanks.
-------

I hear that I am indirectly responsible for a spirited discussion about the
relation between HPSG and P&P on the HPSG list, though I've only actually seen
Ivan's, Ash's and Carl's messages, in addition to the original one from Liz.
Even from this small selection, though, I see that a few misconceptions about
my views of HPSG may have arisen, so I'd like to clarify things a bit.

First and foremost, let me make it absolutely clear that I have never, will
never, and would never make a distinction between "generative grammar"
on the one hand and HPSG on the other (as though only people who work in
the current Chomskyan framework are doing generative grammar).  Frankly,
I feel kind of stupid having to say this at all, though I understand why the
issue is a sensitive one:  the fact that Liz would assume that this distinction
even makes sense is an indication of how robust the political divisions in
syntax still are, and how they manage to seep down even to the undergraduate
level, despite the best efforts of some of us to neutralize them.  I don't
know what the content of the other postings to the list is, but judging from
the messages from Carl and Ash, I can see that Liz's choice of words
has triggered an (understandably) strong reaction, since they seem to
describe some "die-hard *GB/MP* guy" making pronouncements on HPSG.
This is not completely accurate, however, either in the description
of the guy or in the topic of discussion.

I guess the main substantive point I want to add to the discussion here that
"HPSG" in Liz's message should be replaced with "AVM(s)".  That is, the topic
under discussion in this class was attribute value matrices as *formalisms*
for representing lexical and syntactic information, not HPSG as a grammatical
*theory*.  The main point I was trying to make in class was that the sort of
GB/MP-type framework that most of the students in the class were familiar
with, ***if done right***, needs to make use of a formal system of feature
organization and interaction (this is the point that Ash is getting at in his
message, too), and that AVMs could provide one formalism for implementing such
a system.  Indeed, this is the only way I can make sense of claims about
"feature checking" and whatnot.

At a higher level, then, it is very much the case that people working in such
frameworks presuppose some kind of formal system of feature representation/
interaction --- possibly, in fact *typically*, without consciously realizing
it, and almost always without having any idea what the structure of the system
should be!!!  (We as linguists should know very well that presupposition need
not be a conscious activity.)  I am not saying that this is a good thing ---
in fact my goal in bringing this point up in a computational linguistics class
was to emphasize the fact that we should not settle for such tacit
presupposition: we need to ensure that the theoretical proposals we make can
be grounded in a well-articulated and clearly defined formal system.  This is
true both for HPSG, where people are --- thankfully and to their credit!! ---
concerned with the "messy details", and of GB/MP, where, to their and the
framework's deficit, people often are not.  (Thus I quite agree that
"abstract" in GB/MP typically equals "vague".)

Hopefully it's clear now that the "compatibility" of frameworks that I was
talking about was really a question of whether (at least some) theoretical
claims in GB/MP about lexical properties, syntactic structures, etc. can be
formally captured using AVMs (or similar representational frameworks).

Let me conclude by saying that I do very much think that the larger question
Liz asks about the relation between the grammatical frameworks HPSG and GB/MP
is a pretty damn interesting one, and definitely worth exploring. It's
something that I've often asked myself, and it's a question that I've
(obviously) tried to encourage my students to ask.  (Though I also think that
it's important to have consistency in framework when training students to do
syntactic analysis --- maybe this is why I'm a "die-hard generative grammar
guy".)  I highly doubt that the relation is going to turn out to be one of
equivalence or notational variance, though, so I think Ivan is going to have
to wait a long time for his proofs.

Chris Kennedy

PS: Those of you who are interested in the actual content of this
computational linguistics class can check out the website:

http://ling.nwu.edu/~kennedy/Classes/Syllabi/346-W01.html


>Hi Chris,
>
>I was wondering, along with Ash, whether the professor in question here was
>you. If so, I congratulate you on training a student with such curiosity and
>dedication.
>
>In various settings, I've encountered claims about the notational variance,
>inclusion, etc. of HPSG and/in GB, P&P and MP.  I would be very interested if
>you had some results (proofs, etc.) about the equivalence of MP and HPSG. I
>also wonder if you have seen the excellent (pointed) discussion of the
>relation between constraint-based frameworks like HPSG on one hand and the
>`Minimalist Program' on the other by Johnson and Lappin:
>
>David Johnson and Shalom Lappin. 1999. Local Constraints vs. Economy.
>Stanford: CSLI Publications.
>
>This is a must-read for anyone entering into a discussion of the relation
>between these frameworks, in my opinion.
>
>I hope we can have a productive discussion of these issues.
>
>All best,
>Ivan
>
>
>
>------- Forwarded Message
>From: Ash Asudeh <asudeh at csli.Stanford.EDU>
>To: Liz Coppock <lizziecoppock at yahoo.com>
>cc: HPSG <hpsg-l at lists.Stanford.EDU>
>Subject: Re: good news from generative grammarians
>
>
>Hi Liz
>
>I can see the usefulness of your professor's comments (Chris Kennedy, by
>an chance?) in trying to relate unfamiliar formalisms to a more familiar
>one ("generative grammar" -- I agree with Tom Wasow that this is annoying,
>because although at this point the term has largely become a positive
>buzzword, by using it in opposing GB/MP to HPSG, LFG, RG, you name it, it
>pumps up the former, at the expense of the latter. I thought we were doing
>science, not politics).
>
>However, just because there are analogies between various frameworks, that
>doesn't mean that one can be reduced to the other. Yes, to some extent
>these are just differences in formalisms, but when theoretical constructs
>are stated in formal terms (and "formal" could be ordinary English, but at
>a sufficient level of precision), these differences matter. For example,
>in many versions of HPSG, there are no traces and no movement. Movement is
>the key explanatory operation in GB/MP. Although there are versions that
>do without it (e.g., Koster, Brody), they still rely on traces mediating
>the filler and the gap.
>
>Also, too often the abstract nature of GB/MP has consisted purely of lack
>of precision. But these are different concepts. A theory can be abstract
>and still be precise in its predictions (general relativity comes to mind;
>just about any theory in the hard sciences is abstract, but precise). I
>think it is mischaracterizing HPSG to say that it is not abstract. After
>all, nobody believes you have directed acyclic graphs in your head. But
>HPSG is precise (as Emily's email pointed out).
>
>It's fine to ignore the "messy details", but only if your theory is
>*precise* enough to someday include more of the mess. I feel somewhat
>confident in the abilities of HPSG, LFG, and to an extent GB (some
>precision wouldn't hurt it, but there have been interesting attempts by
>Stabler, Rogers, and others) to do this, but less confident in MP. That's
>why I do the former and not the latter.
>
>Here's the kind of lack of precision or espousal of abstractness, if you
>would rather put it that way, that I think is problematic with Minimalism,
>for instance: it is a theory where half the work is done by
>feature-checking and movement, yet there is not any theory of features,
>feature projection, feature checking, or feature movement that is precise
>in any sense of the word.
>
>Best,
>Ash



>-----
>Date: Fri, 27 Apr 2001 22:51:27 -0400 (EDT)
>From: Carl Pollard <pollard at ling.ohio-state.edu>
>Message-Id: <200104280251.WAA23869 at julius.ling.ohio-state.edu>
>To: hpsg-l at lists.Stanford.EDU, lizziecoppock at yahoo.com
>Subject: Re: good news from generative grammarians
>
>Dear Elizabeth,
>
>As a `die-hard generative guy' myself, I can't resist pointing out
>that an HPSG **IS** a generative grammar, i.e. a formal specification
>of a set of structural representations (Pollard 1999, `Strong
>generative capacity in HPSG'). The same can be said of an LFG, and if
>you regard a proof tree as a kind of structural representation, the
>same can be said of a categorial grammar. On the other hand, with the
>exception of Ed Stabler's formalization of GB theory in first-order
>logic, I'm not aware of any manifestation of Chomskyan linguistics
>after 1965 that could be characterized as `generative' in this sense.
>
>To ask what the relationship between HPSG and generative grammar
>is actually quite insulting to a practicioner of HPSG. It is like
>asking a Buddhist what the relationship is between his/her faith
>and a REAL religion (namely the questioner's).
>
>As for `we presuppos[ing] all this stuff', that sounds very
>impressive, but it is actually devoid of content absent any precise
>specfication of what is being presupposed. It is pretentious and
>intellectually dishonest. It also creates the false impression that it
>doesn't matter what the details are, that all formalized theories are
>equivalent, that the differences between them don't matter (because
>`we' are too important to be bothered learning the math so we can tell
>them apart). In fact, different formalized theories in science make
>distinct empirical claims. By contrast, vague, imprecise, or
>not-yet-worked-out `theories' don't make any empirical claims at
>all. As Stabler (1992) put it: "It really is quite important to get
>the details right. Otherwise the theory is just descriptively
>inadequate." Geoff Pullum, writing in 1989, put it this way:
>
>      It really is true that for most varieties of grammatical theory
>      being practiced today, there is no way to determine from the
>      published literature what counts as a sentence or a structural
>      description (hence a language), or a rule or a grammar or a
>      universal principle (notice that these are purely syntactic
>      questions about the form of grammars); and also no way to
>      determine what a rule or universal principle actually says about
>      sentences, structural descriptions, languages, or grammars (these
>      being semantic questions about the content of grammars).
>
>I think things are actually WORSE now than when Pullum was writing.
>
>  >
>He indicated a conception of the two frameworks
>wherein one was just more abstract than the other, not
>concerned with all the messy details, but essentially
>compatible.
>>>
>
>You are using `abstract' here as a synonymn for `vague', `imprecise',
>`not really worked out yet' or `sloppy'. An HPSG as currently
>formalized is a theory (in the logical sense) stated in a certain
>formal language (a kind of feature logic), and generates a set of
>abstract feature structures. This is extremely abstract, in the same
>sense that (say) abstract algebra is abstract. However, it is NOT
>`abstract' in the sense in which you employed the word.  In your
>usage, if instead of asserting that E = mc^2, Einstein had asserted
>that `energy is in an appropriate licensing relationship with the mass
>and the speed of light' that would have been more `abstract', and
>therfore preferable because it didn't stoop to handling `messy
>details.'
>
>>
>In what sense(s) are the two frameworks compatible?
>Could there exist a mapping from any generative
>grammar theory to a corresponding HPSG theory?
>>>
>
>You can't map between two theories unless you HAVE two theories.  An
>HPSG actually IS a theory. What other `generative grammar' theory do
>you want to try to map into it?
>
>This is probably not the kind of reply you expected, but unfortunately
>you pushed one of my buttons. Several of them, actually.
>
>Regards,
>
>Carl



>On Fri, 27 Apr 2001, Liz Coppock wrote:
>
>>  One of my professors (a die-hard generative grammar
>>  guy) said, wrt HPSG (in a computational linguistics
>  > class), "we presuppose all this stuff."
>>
>>  He indicated a conception of the two frameworks
>>  wherein one was just more abstract than the other, not
>>  concerned with all the messy details, but essentially
>>  compatible.
>>
>>  I wonder what the people on this list have to say
>>  about the relationship between HPSG and generative
>>  grammar --
>>
>>  Is his view accurate?
>>  In what sense(s) are the two frameworks compatible?
>>  Could there exist a mapping from any generative
>>  grammar theory to a corresponding HPSG theory?
>>  Is one more abstract than the other?
>>
>>
>>  Elizabeth Coppock
>  > Northwestern University

- --

- --------------------------------------------------
Department of Linguistics                tel:  847.491.8054
Northwestern University                fax:  847.497.3770
2016 Sheridan Road
Evanston, IL  60208

------- End of Forwarded Message