Satoshi Shirai, Akio Yokoo, Hiromi Nakaiwa, Satoru Ikehara & Masahiro Miyazaki, Technical Report of IEICE, SIG-NLC, July 24-25, 1997

A Semantic Valency Dictionary for Japanose-to-English Machine Translation

Satoshi SHIRAI^*1, Akio YOKOO^*1, Hiromi NAKAIWA^*1,
{shirai,ayokoo,nakaiwa}@cslab.kecl.ntt.co.jp

Satoru IKEHARA^*2, and Masahiro MIYAZAKI^*3
ikehara@ike.tottori-u.ac.jp miyazaki@info.eng.niigata-u.ac.jp

^*1NTT Communication Science Laboratories, ^*2Faculty of Engineering, Tottori Universify, and
^*3Faculty of Engineering, Niigata University

Abstract

In order to improve the qualify of machine translation, the valency pattern method of analyzing the semantic relationship between verbs and nouns is known to be effective. We compiled about 12,000 general patterns (6,000 verbs) and 8,000 idiomatic patterns for a Japanese-to-English machine translation system based on Japanese-to-English dictionaries for human use and Japanese and English example sentences. We evaluated the patterns and found that at least 20,000 general patterns (10,000 verbs) and 5,000 idiomatic patterns were needed. In this paper, we introduce the semantic valency dictionary and discuss how to complete it.

Key words

machine translation, Japanese-to-English translation, semantic valency dictionary, va1ency pattern pair, semantic analysis

[ Technical Report of IEICE, NLC97-14, pp.45-52 (July, 1997). ]