Newsgroups: sci.lang
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!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: <DDB0AE.5It@cogsci.ed.ac.uk>
Organization: Centre for Cognitive Science, Edinburgh, UK
References: <4084i9$dml@newsbf02.news.aol.com> <DD5CLH.2nJ@cogsci.ed.ac.uk> <40makg$fea@medici.trl.OZ.AU>
Date: Mon, 14 Aug 1995 14:00:44 GMT
Lines: 97

jbm@newsserver.trl.oz.au (Jacques Guy) writes:

>I cannot resist to try and give some answers, right off the top of my
>head, to these questions from Steve Finch. (Off the the top my head is 
>important: I do not want to think to long about them, I want to capture
>my seat-of-the-pants grasp of those problems)

Thanks Jaques, I have a few comments.

>>1.  Why does our use of words satisfy Zipf's law and how does the HSPM
>>make language satisfy Zipf's law?  Why do the forms of syntactic
>>construction we use also satisfy a Zipf-Mandelbrot type law?

> Dimly, it must have to do with general statistical properties information. 
> Those are the true universals.

This may be the case, but it needs further investigation.  Surely any
processing system would exploit such a strong regularity which appears
at all levels in natural language (and other information sources).
How is this exploited?

>>3.  Why are many of the traditional linguistic categories in language
>>apparent statistically, and of what importance is this observation for
>>acquisition and processing?

>  First part of the question: because like occur with like (similia
>  apud similia). An article, for instance, will hardly ever precede
>  a preposition or an article in English. Clustering from a matrix
>  of co-occurrence frequencies, then, ought to produce clusters
>  corresponding to grammatical categories. Steven Finch has tried it,
>  anyway, and found that it worked.

>  Second part of the question: Another question: processing which way?
>  Translating utterances into their referents, or encoding meanings
>  into utterances? The two must be very distinct and separate processes.
>  The skills needed to read a map are not at all those needed to draw a
>  map.

But map making and map reading are far from independent skills; they
have in common knowledge of the way relevant information is encoded.

>>5.  What processes might lead us to acquire knowledge of how to use
>>words from hearing/reading them once only?  

>  
>  So Laelius returned to Scipio with a ***** from Syphax that no
>  danger would attend his visit, though nothing further was 
>  garanteed.  
>		 (Livy: The War With Hannibal, book XXVIII)

>  You've guessed the meaning of *****, of course. Same process.

>>What is needed from
>>language in order to be able to do this?

>Nothing to do with language, I should think. Only with the predictability
>of situations.

But language has to be such that we can guess what **** is from the
description of the rest of the sentence, no?  Even if the way we
intuit doing this is to apply "world knowledge" to the task, it still
leads to a regularity (which we can use for prediction) in observed
language.

Again, as with the map-making/reading analogy, the skill to do this is
not unrelated to the skill needed to identify **** if we don't hear it
clearly or if it appears as a typo in text.  The problem of
identifying that we haven't heard it clearly, or realising it is a
typo is large (and intimately connected with our ability to process
language).  Again, in the general case this might require extensive
world knowledge, but in the typical case it's about putting sources of
evidence together, and relies on the relative unambiguity
(predictability) of language in order to work.

>>I also believe the answers to these
>>questions will be universal across languages.

>Strangely, I find myself agreeing there. "Strangely" because I
>generally disbelieve anything and everything about "language
>universals". But this is not narrowly constricted to "language".
>It must hold of any information, whatever its medium.

So this leaves a obvious question: what is *special* about language as
an information source?  Of course we're going to make full use of
whatever cognitive tools we have available to deal with language, but
what makes humans uniquely able to use natural language to
communicate?

IMHO, this might possibly have something to do with the answer to Q8.
But equally it might just be about an innate facility of humans to
apply their IP bag of tools in a way which other animals don't, and
the study of the bag of tools is then central.  Certainly the second
view has been neglected so far.

Cheers,

Steve.
