Newsgroups: sci.lang
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!nntp.sei.cmu.edu!cis.ohio-state.edu!math.ohio-state.edu!cs.utexas.edu!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: <DD7CpI.BoA@cogsci.ed.ac.uk>
Organization: Centre for Cognitive Science, Edinburgh, UK
References: <4084i9$dml@newsbf02.news.aol.com> <DD5CLH.2nJ@cogsci.ed.ac.uk> <40gcuc$rl9@ruccs.rutgers.edu>
Date: Sat, 12 Aug 1995 14:38:51 GMT
Lines: 75

jirifkin@ruccs.rutgers.edu (Jay Rifkin) writes:

>In sci.lang Steve Finch writes:

>>analyse if we are to be able to perform useful tasks in
>>computationally applied research such as translation, message
>>extraction, information abstraction and retrieval from natural
>>language, speech recognition, and so on and so forth.

>He then goes on to list a number of questions, most of which relate
>to how we understand language as it comes in (process it).  He suggests
>that "real" linguists aren't interested or have nothing to say about these
>issues, and implies that these are the issues of interest to folks in
>the "applied" realm.

>As Pesetsky notes, the bit about linguists simply isn't true.

>But I'd like to add that these are scarcely the only important issues
>in the applications Finch mentioned, and that for some of them, they're
>not such important issues at all.  For example, someone interested in
>text translation doesn't really need to be all that concerned with noisy
>input (unless the text is a transcript of a written dialogue), one-shot word
>learning, on-line ambiguity (the computer doesn't have to be limited to
>reading linearly), or learning/bootstrapping issues.

Well, this is simply not true.  What are you going to translate if it
is not written by anyone?  Ambiguity, it strikes me, is one of THE
major issues in machine translation with words having many plausible
translations according to context, and one way to approach MT is by
training a system on parallel texts (e.g. IBM's stuff which performs
at `state-of-the-art' level for free text) for which one shot learning
(or learning from sparse data sets) will be a large issue.

>In fact, they may well be more interested in the sorts of things that Finch's
>"real" linguists do, such as grammatical structure and argument structure.

And I'm sorry if I appeared to belittle this enterprise (I'm NOT
arguing for an abandonment of it), but it's simply not sufficient for
performing the tasks.

My favourite example is aligning parallel texts; the task of saying
for direct translations "this sentence is a translation of that
sentence (or those two sentences, etc)".  One would think the
vocabulary used in the sentences might have the most important
influence here.  Not a bit of it.  By far the most effective way found
so far is to align on sentence lengths (number of words).  Some might
say "but that's just a hack".  It works far too well to be "just a
hack".  There is something about the nature of languages which
translates short sentences to short sentences and long ones to long
ones and traditional linguistics simply does not address that part of
the nature of language at the level of detail required to build an
aligner.

>Finch also comments on the "lack" of useful findings in the Generative
>framework.  But that seems rather short-sighted. Much of the math and the
>cutting-edge theoretical physics being done today doesn't seem to have
>much relevance today, but I wouldn't expect to see engineers complaining
>because there's still plenty of work to be done applying the stuff
>mathematicians and physics were thinking about 60-100 or more years ago.
>I'm not saying that one shouldn't look at linguistics with a critical
>eye, I'm just saying that one shouldn't expect immediate gratification.
>Sometimes there's a long bridge to be built between theory and software.

I'm not saying the generative framework is a useless enterprise.  Far
from it; it clearly describes some important and interesting features
of the structure of natural language.  Nor is research into formalisms
which can elegantly describe such structure "useless".  I use
feature-value logics extensively, for example.  To say, however, that
it constitutes the central paradigm in the scientific study of the
nature and structure of language will prove to be false well within my
lifetime.


Steve.

