From nobody@cse.Buffalo.EDU Thu Oct 22 13:14 EDT 1998
From: nobody@cse.Buffalo.EDU
Date: Thu, 22 Oct 1998 13:14:46 -0400 (EDT)
To: techreps@cse.Buffalo.EDU
Subject: techrep: POST request
Content-Type: text
Content-Length: 2747

Comments: This was Min-Hung's MS thesis
ContactPerson: rapaport@cse.buffalo.edu
Remote host: adara.cs.buffalo.edu
Remote ident: rapaport
### Begin Citation ### Do not delete this line ###
%R 97-16
%U /projects/rapaport/MT/tech-report.ps
%A Liao, Min-Hung
%T Chinese to English Machine Translation Using SNePS as an Interlingua
%D December 1, 1997
%I Department of Computer Science and Engineering, SUNY Buffalo
%K machine translation, interlingua, semantic networks
%X An interlingua machine translation is a two-stage operation: from  
source language to an interlingua, and from the interlingua to the  
target language.  The idea of translating natural-language texts using  
an interlingua, an intermediate common language, is based on the belief  
that while languages differ greatly in surface structures, they share a  
common deep structure.  I propose a way of automatically translating  
texts using SNePS as an interlingua. The representation of the meaning  
of the source-language input is intended to be language-independent,  
and this same representation is used to synthesize the target-language  
output. As an interlingua, SNePS fulfills the requirements of being  
formal, language-independent, and a powerful medium for representing  
meaning; thus, it can handle ambiguities.  SNePS can be used to  
translate texts automatically as follows.  The user inputs a sentence  
of the source language to a generalized augmented-transition-network  
(GATN) parser-generator. The parser fragment of the GATN  
parser-generator updates an existing knowledge base containing semantic  
networks to represent the system's understanding of the input sentence.  
The node newly built to represent the proposition is then passed to the  
generator fragment of the GATN parser-generator, which generates a  
sentence of the target language expressing the proposition in the  
context of the knowledge base.  The parsing of Chinese relies more on  
semantic information than syntactic relations because, first, the word  
order is determined primarily by semantic factors rather than syntactic  
ones, second, there is a lack of morphological inflections and  
syntactic clues. A series of noun phrases and verb phrases can be  
juxtaposed without syntactic glues such as function words or variation  
of verb forms to make the linking. These linguistic properties cause  
lexical and structural ambiguities. Besides being an adequate  
interlingua representation, SNePS is also a computational environment  
particularly suitable for processing Chinese because it provides the  
facilities for building, retrieving, and deducing semantic information  
that guides the parsing and resolves ambiguities.