Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!newsfeed.internetmci.com!tank.news.pipex.net!pipex!uknet!newsfeed.ed.ac.uk!edcogsci!usenet
From: chrisbr@cogsci.ed.ac.uk (Chris Brew)
Subject: Re: GPSG as introductory grammar formalism
In-Reply-To: andrich@infko.uni-koblenz.de's message of 4 Aug 1995 14:42:54 GMT
Message-ID: <CHRISBR.95Aug7105713@blair.cogsci.ed.ac.uk>
Sender: usenet@cogsci.ed.ac.uk (C News Software)
Nntp-Posting-Host: blair
Organization: Centre for Cognitive Science, University of Edinburgh
References: <3vtble$i9n@newshost.uni-koblenz.de>
Date: Mon, 7 Aug 1995 09:57:13 GMT
Lines: 57

In article <3vtble$i9n@newshost.uni-koblenz.de> andrich@infko.uni-koblenz.de (Oliver Andrich) writes:

   Path: edcogsci!newsfeed.ed.ac.uk!leeds.ac.uk!bradford.ac.uk!strath-cs!tank.news.pipex.net!pipex!howland.reston.ans.net!xlink.net!newshost.uni-koblenz.de!andrich
   From: andrich@infko.uni-koblenz.de (Oliver Andrich)
   Newsgroups: comp.ai.nat-lang
   Date: 4 Aug 1995 14:42:54 GMT
   Organization: University Koblenz / Germany
   Lines: 25
   NNTP-Posting-Host: schiller.uni-koblenz.de
   X-Newsreader: TIN [version 1.2 PL2]

   Hello everybody!

   Can anybody please give me any good reason for using GPSG as an
   introductory grammar formalism for computational linguists? I mean,
   the grammar formalism is somewhat complicated and some people even
   think, that GPSG is the worst choice, cause the parsing seems to be np
   hard. And so the GPSG think seems to irrelevant for computational
   linguists, cause they want to implement something like that on a
   computer.

Even if parsing is NP complete for some version of GPSG, as suggested
by Barton, Berwick and Ristad, it certainly does not follow that GPSG
is irrelevant for computational linguists. For one thing, the worst
case behaviour may not arise in the grammars you write for a
particular application, and for another GPSG is a (reasonably)
precisely stated theory whose computational consequences follow (fairly)
directly from the publicly available description in the book. This
gives it a useful edge over some other grammatical theories, especially
as an introduction to linguistics for people who are going to build 
artefacts rather than advance linguistic theory. Plausible candidates
to take its place, with much the same desirable properties, include TAG,
HPSG and some variant of categorial grammar, but none of these is without
its problems, all of them involve a degree of complication in one way or
another, and I wouldn't want to claim that any of them, on its own, would
be an ideal introduction for computational linguists. Personally, I would
encourage people to read Peter Sells' Introduction to Contemporary Syntactic
Theories first, then maybe Michael Moortgat's thesis book, then something
about HPSG. But not every computationalist may wish to read all that.

Finally, GPSG has formed the basis of several decent computer systems, 
including the Alvey Natural Language Toolkit, and the MT system built
in Berlin by Christa Hauenschild's group. So it can't be completely
terrible as a basis for computational artefacts. 

Chris
--
------------------------------------------------------------------
Email: Chris.Brew@edinburgh.ac.uk
Address:  Language Technology Group, 
          HCRC, 2 Buccleuch Place,  Edinburgh EH8 9LW 
          Scotland
Telephone: +44 131 650 4631 
Fax:       +44 131 650 4587
------------------------------------------------------------------


