Newsgroups: sci.lang
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!news.mathworks.com!newsfeed.internetmci.com!news.sprintlink.net!dispatch.news.demon.net!demon!uknet!newsfeed.ed.ac.uk!edcogsci!steve
From: steve@cogsci.ed.ac.uk (Steve Finch)
Subject: Re: Chomksy, Significance, and Current Trends
Message-ID: <DD7F77.CKw@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>
Date: Sat, 12 Aug 1995 15:32:41 GMT
Lines: 79

David Pesetsky <pesetsk@mit.edu> writes:
>steve@cogsci.ed.ac.uk (Steve Finch) wrote:
>
>>Indeed.  Here are some small questions of crucial importance to
>>applied linguists either ignored by "real" linguists or for which the
>>analysis of "real" linguists is largely unapplicable.
>>
>
>>
>>3.  Why are many of the traditional linguistic categories in language
>>apparent statistically, and of what importance is this observation for
>>acquisition and processing?
>
>Two years or so ago, you brought this question up in this very forum.  I kept 
>trying to present an answer that was not at all inimical to your concerns, and
 you 
>kept ignoring that fact -- perhaps because it was inconsistent with your conti
nued 
>insistence on a sharp intellectual and political break between your kind of 
>"useful" linguistics and what you now call "'real' linguistics" with scare-quo
tes.  
>The answer is and was: let Universal Grammar be as rich as one thinks, there 
>remains the problem of how the child bootstraps into the system.  The existenc
e of 
>statistical evidence for categorization, as shown in your own dissertation, 
>provides a hint as to one way this bootstrapping might take place.  We need 
>empirical work on children's speech to tell us more.
>

Well yes, of course, and I have never said this isn't the case
(athough I may have been silent on the issue since I have seen no
convincing empirical evidence for the details of UG, but I would say
that if the presuppositions of the UG story turn out to be correct,
then so is your analysis).  But there is a lot more that is
statistically evident distributionally than is the traditional domain
of Universal Grammar.  And the efficacy of such statistical methods
relies on many facts of the nature of language which aren't considered
by traditional approaches, such as the observation about Zipf's law,
the lack of ambiguity in language, the ability of one part of a
construction to be statistically informative of another part of the
construction, the robustness of language and many more interesting
features of all natural languages which are effectively the most
direct precinct of a statistical analysis of language.

BTW, by your "let UG be as rich as one thinks" comment, am I to
presume you are *defining* UG to be "that which facilitates language
acquisition and processing in humans"?  If so, then it must account
for all language-universal observations, some of which I raised in the
post you are replying to, rather than simply the ability to
discriminate sentences.

>>5.  What processes might lead us to acquire knowledge of how to use
>>words from hearing/reading them once only?  What is needed from
>>language in order to be able to do this?
>
>See the work of Lila Gleitman and colleages.  Also Steven Pinkers's MIT Press 
book 
>"Learnability and Cognition".  Also various papers and books by Beth Levin wit
h and 
>without Malka Rappaport (Hovav). All squarely within the tradition that you th
ink 
>has nothing to offer you, and all extremely careful and interesting.

Computer implementable theories of one time learning from corpora
which work as well as human one-time-learning does are nonexistant.  I
want one of those and believe such is attainable.  I can't implement
such a system from these references.  That does not mean they are
uninteresting, just unapplicable.

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?

Or to put it another way: what is the science of the language we find
in large language corpora, if not linguistics?

Steve.
