Browse Prior Art Database

Transfer Mechanism in Machine Translation Using Dependency Structures As a Knowledge Base

IP.com Disclosure Number: IPCOM000101813D
Original Publication Date: 1990-Sep-01
Included in the Prior Art Database: 2005-Mar-16
Document File: 4 page(s) / 128K

Publishing Venue

IBM

Related People

Nagao, K: AUTHOR [+2]

Abstract

Disclosed is a transfer mechanism using dependency structures as a knowledge base in machine translation based on the transfer approach (Fig. 1).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Transfer Mechanism in Machine Translation Using Dependency Structures As a Knowledge Base

       Disclosed is a transfer mechanism using dependency
structures as a knowledge base in machine translation based on the
transfer approach (Fig. 1).

      The transfer phase of this system uses a dictionary to select
the most preferable patterns. To get the best transfer pattern, it is
not enough to consider syntactic information. Semantic information,
such as the semantic relations among words, context, and common
knowledge, is necessary for a natural translation.

      There are two problems in realizing a practical semantic
processing: constructing a huge knowledge base efficiently, and
providing a mechanism that can access such a knowledge base
efficiently.

      We propose a mechanism to resolve these two problems.
Configuration

      Our system assumes that the transfer phase receives dependency
structures (DSs) in the source language and outputs DSs in the target
language. A DS is a tree that expresses the governor and governee
relations among words in the input sentence.

      Fig. 2 shows the configuration of our system. The system
consists of two major process: the pattern selection and
transformation.

      In the next section, we give the details of the knowledge base
and the transfer dictionary, and after that, we describe the two
major processes of the transfer phase. Knowledge

      All knowledge used in our system is expressed in the form of
DSs. A DS is a tree that expresses governor and governee
relationships among words (Fig. 3).

      There are two kinds of knowledge; the commmon-sense knowledge,
which is the hierarchical relationships among words (Fig. 3), and
contextual knowledge.

      Common-sense knowledge is used in reference and is constructed
in advance. Contextual knowledge is used in transfer pattern
selection and is constructed during the processing of documents.
Transfer Dictionary

      The transfer dictionary consists of transfer patterns, each of
which is a pair of a transfer condition and a transformational
description. The transfer conditions (TCs) are DSs described in the
source language, whose nodes contain a word, a case, and other
features in the source language. The transformational descriptions
(TDs) are also DSs, but in the target la...