Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!news.kei.com!travelers.mail.cornell.edu!cornell!chrisb
From: chrisb@cs.cornell.edu (Chris Buckley)
Subject: Re: Books on Intro. natural language proces
Message-ID: <1994Sep29.063344.29297@cs.cornell.edu>
Organization: Cornell Univ. CS Dept, Ithaca NY 14853
References: <3608t2$eo9@news.cais.com> <1994Sep25.174751.28787@cs.cornell.edu> <366k2s$500@redwood.cs.scarolina.edu> <36ci4i$if6@narnia.ccs.neu.edu>
Date: Thu, 29 Sep 1994 06:33:44 GMT
Lines: 42

hafner@ccs.neu.edu (Carole Hafner) writes:

>Now, to enter the fray: IR systems based on statistical matching are
>certainly "text processing" systems but are not "natural language"
>systems (as the term is used by those in the NL field).  NLP (including both
>understanding and generation) is concerned with trying to figure out
>how natural language "works".  It is an article of faith (well justified
>by other scientific endeavors) that success in figuring out how natural
>language works will eventually enable us to create interesting and
>useful systems.

I'll agree with at least the last sentence (and the parenthesized part
of the first sentence!).  But it seems strange to try and figure out
how natural language works while ignoring an area that not only is
very successful dealing with natural language, but even offers ways
of evaluating how well natural language is being understood!

>There has been slow but steady progress in understanding how natural
>language "works" during the past 30 years, and systems have been created
>in limited domains, performing tasks that statistical matching could
>not even approach. 

This doesn't agree with my view of the past couple of years of MUC and
TIPSTER.  I see wholescale abandonment of the deep parsing and
associated linguistic theories of the past 30 years in favor of
        1. statistical part-of-speech tagging
        2. pattern matching
        3. template filling 
While it may be true that there's been "slow but steady progress", it
doesn't seem to be reflected in the TIPSTER systems.  They certainly
owe a debt to what had been done before, but not to any general
"understanding of how natural language works."

Those groups seem to agree that full parsing, at least, is not needed
to understand text (at least understand at the level of the current
state of the art).

                                ChrisB
-- 
Chris Buckley                   Dept of Computer Science
chrisb@cs.cornell.edu           Upson Hall, Cornell University
(609)  275-4691                 Ithaca, NY   14852
