Research topic?

Li Hui-Feng <hflee@madonna.postech.ac.kr>
From: Li Hui-Feng <hflee@madonna.postech.ac.kr>
Message-id: <199606171335.XAA29913@madonna.postech.ac.kr>
Subject: Research topic?
To: mann@cse.unsw.edu.au, ged@cs.rmit.edu.au, srkb@cs.umbc.edu, cg@cs.umn.edu,
        kaw@swi.psy.uva.nl, nl-kr@snyside.sunnyside.com,
        pazienza@info.utovrm.it, sowa@west.poly.edu, cgkee@neumann.une.edu.au
Date: Mon, 17 Jun 1996 23:35:18 +0900 (GMT+9:00)
X-Mailer: ELM [version 2.4 PL21-h4]
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-2022-kr
Content-Transfer-Encoding: 7bit
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk

Dear Sirs, and CGers,

I am now considering the topic for PhD dissertation proposal related to 
CG representation for Korean language sentence that will be used as 
interlingua for machine translationi of Korean-Chinese MT that may extended to 
multilingual MT. I have noticed there are many research works on natural language 
generation from CGs, but less research on CG generation from NL sentence.

Here are some questions that is needed to be confirm.

1) First of all, I would like to ask CGers tell me if the following title and content 
   is valuable for PhD dissertation topic. 

   "Semantic analysis for conceptual graph generation from syntactic dependency 
   graphs"

   This topic concerns CG generation as interlingua representation for MT which will 
   read the syntactic parser's result as a clue to do semantic analysis. We need to 
   map word sense to a conceptual type and transform syntactic relation to a conceptual 
   relation. The syntactic paser of ours uses dependency grammar for Korean.

2) I have noticed there are few papers on CG generation from natural language
   sentences for meaning representation, although CG is a good model for semantic 
   representation. Please write to me the main problems remained for CG generation from 
   NL sentence.

3) I have read several papers by Sowa, Pazienza, Bornerand....  

   Sowa(1986) wrote in his paper for short the method of semantic analysis, that is 
   mainly using CG join operation with syntactic tree in bottom-up way. 

   -- I am not sure if CG operation and canonical graph can disambiguate word sense and 
   structural ambigutiy, and accomplish the semantic analysis, such as analyzing word 
   usage: metonymy, metaphor,..., and so on.

   Canonical graphs can be used to represent selectional restriction for verb, like 
   case frame.  Then how to represent the constraint between nouns that may be used in a 
   phrase? Do we need to define only conceptual type to a noun?

   -- I think it will be complicated to do semantic analysis using only CG operation(join, 
   simplificatin, copy, restriction), since we need to define in detail the canonical graph
   which is difficult to design and which will increase the complexity of JOIN operation.

   -- We observed that there will be several conceptua relation for a syntactic relation 
   such as subj to { agent, experience, patient,...} that will need to be dealt with verb 
   semantic knowledge of each language. We may need to attach some constraint to CG canonical 
   graphs for syntactic to semantic correspondence.

   -- So, we may need to design canonical graphs and other lexical knowledge base that describe 
   in more detail the word usages, and implement hybrid semantic analysis module which may 
   include word sense disambiguation(using collocation pattern, world knowledge,...), semantic
   verification module, and CG operation module(mainly using unification?..), and so on...

   What do you think of these opinion? We think we could represent world knowldge with 
   schemata, but it may raise difficult in diffinition and usage for real MT system... 

4) Most of all, I would like to know the state of art of the current research 
   of CG generation, its main problem remained to be solved, the possible way that 
   do semantic analysis for CG generation.

I hope dear CGers give me guidence and hint for research on this topic. 

The early mail will be highly appreciated.

With best regards,

        Hui-Feng Li(6/17)

PIRL, Knowledge & Language Engineering Lab. 
Pohang University of Science and Technology, 
San 31, Hyoja-Dong, Pohang, 790-784, Republic of Korea

Tel: +82-562-279-5638, -279-5656
FAX: +82-562-279-5699
email:hflee@madonna.postech.ac.kr