Newsgroups: sci.lang
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!newsfeed.internetmci.com!news.sprintlink.net!howland.reston.ans.net!tank.news.pipex.net!pipex!uknet!newsfeed.ed.ac.uk!edcogsci!steve
From: steve@cogsci.ed.ac.uk (Steve Finch)
Subject: Re: Chomksy, Significance, and Current Trends
Message-ID: <DDAtxL.1z4@cogsci.ed.ac.uk>
Organization: Centre for Cognitive Science, Edinburgh, UK
References: <4084i9$dml@newsbf02.news.aol.com> <DD5CLH.2nJ@cogsci.ed.ac.uk> <40flg8$6cf@senator-bedfellow.MIT.EDU> <DD7F77.CKw@cogsci.ed.ac.uk> <40ioik$579@senator-bedfellow.MIT.EDU>
Date: Mon, 14 Aug 1995 11:43:29 GMT
Lines: 113

David Pesetsky <pesetsk@mit.edu> writes:

>steve@cogsci.ed.ac.uk (Steve Finch) wrote:

>>If there is an intellectual divide (I try to keep out of politics but
>>I accept that, in this game, intellectual and political differences
>>seem to be correlated) it is on the question: is everything that is
>>scientifically interesting about the nature and structure of language
>>derivable from an analysis of acceptability decisions?

>What nonsense! That's never been anyone's question.

Then the presupposition must be false and there is no intellectual
divide :-)

>Who on earth has ever taken the stand that "everything that is
>scientifically interesting about the nature and structure of language 
>[is] derivable from an analysis of acceptability decisions?"  True, 
>acceptability decisions were for many years the best available source 
>of evidence on certain issues -- e.g. the nature of long-distance 
>dependencies or the rules governing anaphora.  But who has ever made a 
>virtue of this necessity? Is it the same people who view with great 
>interest the slowly increasing availability of brain-imaging data on 
>sentence processing?  Or is it the same people who construct 
>painstaking act-out experiments with children or count constructions 
>in the CHILDES corpus of child utterances in order to understand how 
>*children* differ from adults on matters like anaphora?  Or the 
>students in our Research Training Grant program who work out parsing 
>theories that can explain the anaphora judgments that flow from 
>certain modern syntactic theories?  Or the members of Bob Berwick's 
>group here at MIT who work on linguistically informed, computationally 
>efficient parsers as a goal in its own right and as an aid to 
>syntactic theory?

>Clearly not.

>Then who are these people who put things the way you say they do?

My "nothing of value comment" (which I had assumed you had in mind
when talking of an `intellectual divide') was specifically about the
utility of logical form, to which none of these quoted areas apply.
You seem to be taking a wider interpretation of what I said (maybe to
make me appear more radical than I am).  Psycho-linguistics, and
neuropsychophysical linguistics is certainly a valuable and
interesting part of the scientific study of language.  However
experiments in these fields always have to be interpreted within a
theory, and that theory should at least be capable of giving
explanations for the universal observations about languages I made in
my initial post.  In its initial form, UG was certainly a theory
derived from the science of acceptability judgements.

>No, I think the intellectual side of the dispute is something else 
>entirely.  It's nothing less tham the following question: Should  
>language be approached first and foremost as a cognitive ability of 
>human beings, to be studied the way one studies other properties of 
>human beings -- or should we should put our research energy into 
>studying linguistic artifacts (corpora) independent of any 
>investigation of the mechanisms that produce and interpret these 
>artifacts?

But corpora are the single largest source of evidence we have about
the input and output of the HSPM (and indeed there is a move in
psycho-linguistics to study corpora and statistics of corpora in order
to test processing hypotheses, particularly w.r.t. interpretation
preference).  Psychological and psychophysical data are just that;
data.  However, there are a lot of interesting facts about corpora
which current theory doesn't seem to address and need scientific
explanation and investigation, and consequently a lot of psychological
and psychophysical data is uninterpretable within current theory.

Corpora are interesting for many reasons, but in order to study them
we need a theory which can explain many of the interesting phenomena
we observe in them.  For example, it would be a somewhat stupid HSPM
which consistently made attachment decisions which turned out to be
wrong (garden pathed) in the 90% case in corpora, since a better HSPM
would be one which didn't.  This leads to the hypothesis "We garden
path because on the basis of the information we have encoded at the
particular point where we need to make the decision, our best guess is
that this attaches there".  One interesting question for attachment
decisions would be "what information could the HSPM use to decide
attachment?" And this question can be investigated by studying corpora
alone (although maybe not completely).  The interesting question for
the UG view of the world would then surely be "what *can't* you get by
looking at corpora?", and this question has to be more thoroughly
investigated than has been done until now.  But you still need to
investigate corpora to achieve this.

In general, corpora are artifacts which represent an abstraction over
the output and input of the HSPM, and as such study of corpora can in
principle throw light on the HSPM by allowing us to construct theories
of the same by observing what works for the processing of corpora.
Moreover, constucting corpus processors are likely to have direct
application to language engineering tasks and lead to alternative
theoretical frameworks.  These theories can then be tested
psycho-linguistically and psycho-physically in much the same way as
current theories are.  In essence we can say "here's a computational
artifact that works for corpora: what are its similarities and
differences to the way humans process corpora?"  As we refine our
artifact, we then have a hypothesis refinement programme of research
which has the potential to be useful in the scientific study of the
HSPM (and perhaps many other H*PMs).  And what's wrong with that?

>I also think there's a very real political side: practical research is 
>always more easily sold to the outside world (including funding 
>agencies and industry) than basic, less obviously practical research.  
>But to make the sale, one has to stress and even exaggerate the 
>likelihood of quick results with one approach and the unlikelihood of 
>quick results with the other.

I see a difference in emphasis, not a divide.  Structure is important,
but then so is much else.  Is structure axiomatic or derived?

Steve.
