Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!dsinc!spool.mu.edu!usenet.eel.ufl.edu!bofh.dot!warwick!bham!sunserver1.aston.ac.uk!usenet
From: rupraibs <rupraibs@aston.ac.uk>
Subject: Re: Transition Network for English Grammar. Possible or not?
X-Nntp-Posting-Host: austria.aston.ac.uk
Content-Type: text/plain; charset=us-ascii
Message-ID: <DrEMsq.EL6@aston.ac.uk>
To: chrisu@wrox.demon.co.uk
Sender: usenet@aston.ac.uk
Content-Transfer-Encoding: 7bit
Organization: Aston University
References: <DqsH03.IL3@aston.ac.uk> <831782552snx@sv.span.com> <DrCxzn.G8D@aston.ac.uk> <4n9i17$no3@wrox.demon.co.uk>
Mime-Version: 1.0
Date: Tue, 14 May 1996 16:50:50 GMT
X-Mailer: Mozilla 1.1N (X11; I; SunOS 5.4 sun4c)
X-Url: news:4n9i17$no3@wrox.demon.co.uk
Lines: 35

chrisu@wrox.demon.co.uk (Chris Ullman) wrote:
>In article <DrCxzn.G8D@aston.ac.uk>, rupraibs <rupraibs@aston.ac.uk> says:
>>
>>I assume you are talking about 'real-world' knowledge?
>>Forgive me if I'm wrong but I thought that if you just wanted to parse 
>>a language using a grammar, you don't have to worry about semantics. If you
>>just wanted to check that the arrangement of the words were in a valid order, 
>>then the meaning of the sentence is irrelevent, isn't it? 
>
>Unfortunately this isn't the case, as far as I understand. The English Language
>has many words which have dual meanings and their type cannot just be determined 
>from grammar order. The word 'green'is a classic example. Take two sentences 'He's a 
>little green', it could be taken to mean he's a little inexperienced, or it could 
>mean he's a small ecologist. There is no way from determining,just by positioning in 
>a sentence, whether the word is an adjective or a noun. Extra semantic information 
>would be needed to parse this sentence. Ok you could employ some sort of lookahead 
>search but this brings in all sorts of other problems. 
>
>Chris

I can see that your above example would pose a problem for parsing a sentence
for correct placement of capital letters (assuming the person being talked
about was a member of The Green Party).
Instead of some sort of lookahead, you could just employ some form of 
backtracking. You would have a Recursive Transition Network (say) with several
ways that a sentence could be parsed, if one route through the network failed,
then the system would backtrack and try another until it succeeded or failed.
Note, I am not concerned with implementation difficulties if that is what you
mean by "all sorts of other problems", but whether or not the issue is 
theoretically possible with no hitches.

I'd welcome any further points you have to make on the issue.

Roop.

