Analogical Mapping Method for MT based on Semantic Typology

Satoru Ikehara^*1 Masashi Saraki^*2 Masahiro Miyazaki^*3 Naosi Ikeda^*4
Yoshihiko Nitta^*5 Satoshi Shirai^*6 Katsumasa Shibata^*7

*1 Engineering, Faculty of Engineering, Tottori University
*2 Nagasaki Junshin Catholic University
*3 Faculty of Engineering, Niigata University
*4 Faculty of Engineering, Gifu University
*5 College of Economics, College of Science and Technology, Nihon University
*6 Advanced Technology Co. Ltd
*7 Faculty of Science, Fukuoka University

This paper proposes an Analogical Mapping Method (AMM) for machine translation, in order to break through the limits of conventional technology of Natural Language Processing based on Compositional Semantics. This method is constructed from the following two concepts. One is conceptualized on the basis of the Theory of Semantic Typology (TST). The TST suggests that human cognition to the world is accompanied by an epistemological framework under the influence of a mother tongue, bringing a heuristic approach that semantically non-linear expressions, which are inherent in an individual language, are extracted from the huge corpus of the language and then patterned as a unit of a meaning pattern. Another is derived from the Analogical Mapping Theory (AMT). The authors have applied the AMT to natural language processing and then invented a new method. According to the method, a set of meaning patterns is established from a source language, and then mapped to a set of those from a target language through the common concept to both.

Key Words:

Machine Translation, Semantic Typology, Analogical Mapping, Non-linear Model, Semantic Analysis

[ Technical Report of IEICE, TL2002-34, pp.7-12 (December 2002). ]