Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!udel!gatech!howland.reston.ans.net!EU.net!sun4nl!news.nic.surfnet.nl!tudelft.nl!news.twi.tudelft.nl!joho
From: joho@dutiag.twi.tudelft.nl (Job Honig)
Subject: Re: Syntax/Semantics Interface (Object-Oriented Representations)
Message-ID: <CvG1LE.4rM@dutiws.twi.tudelft.nl>
Summary: tree manipulation language
Sender: J.Honig
Organization: Delft University of Technology
References: <341qdh$5d2@news.cs.tu-berlin.de>
Date: Thu, 1 Sep 1994 09:12:49 GMT
Lines: 56

In article <341qdh$5d2@news.cs.tu-berlin.de>,
Joachim Quantz <jjq@cs.tu-berlin.de> wrote:

>I assume that the output of a syntactic parser will be something
>like a phrase structure tree. It seems that in most existing systems
>the semantic module takes this phrase structure tree as input and
>PARSES it in order to build up a semantic representation. 
>
>1. Are there any systems (or is there any theoretical work), in which
>   the phrase structure tree is treated as an abstract data type for
>   which access predicates are provided?
>   I would be especially interested in syntactic representations which
>   treat the nodes in the phrase structure trees (i.e. the words and
>   phrases) in an object-centered way. `Functional predicates' would then
>   return, for example, 
>	- the subject of a sentence
>	- the daughters of a phrase
>	- the topic of a sentence
>	- the direct object of a verb
>	- the preposition of a prepositional phrase
>	- the case of a noun phrase
>	- etc.
>   Obviously, which predicates are applicable at a specific object (node)
>   depends on the type of the object (e.g. sentence, noun phrase, verb).
>
>   (I know that it is in principle possible to build such an interface
>   on top of, for example, an HPSG-like feature structure. I'd be interested
>   to know whether this has ACTUALLY been done already.)

Several years ago, we developed a system which uses a special purpose
functional programming language in which a parse tree is a built-in
data type. The language has a large set of operators to access nodes
relative to a given node. Functions can be written to retrieve any
node or tree, based on the tree's topology and/or the decorations of
the nodes.

The system was designed to collect contextual information to be used for
lexical transfer (in an MT system), but we even (mis-)used it to generate
database queries from (simple) Dutch sentences. 

>2. If such interfaces exist, are they designed for a specific syntactic
>   theory, or can they be used for several different syntactic theories?
>
>   Alternatively, is there any work concerned with identifying linguistic 
>   relations (as exemplified by the ones above) which are used in 
>   several/many/most/all syntactic theories?

It can be used with any theory. However, the system is highly integrated 
with our language analysis software.

>Thanks for any information on this,
>	Joachim Quantz

Job Honig
Applied Linguistics Group
Delft University of Technology
