A Mono-lingual Corpus-Based Machine Translation of the Interlingua Method

This paper describes a prototype of an example-based machine translation system. In this system, key language resources are EDR corpus and concept classification dictionary. The corpus consists of a pair of sentences, their morphological representations, their syntactic representations, and their semantic representations. The semantic representations are described by an interlingua. Therefore the corpus can be viewed either as a mono-lingual corpus or as a parallel corpus between a natural language and an interlingua. The system analyses source sentences and generates target sentences by example databases. Similarity calculations play essential roles in analysis and generation phases. These calculations uses the concept classification dictionary. The translation system is realized by directly combining a source language analysis and a target language generation without a transfer phase. The system has been implemented and the state of the current prototype showed evaluation data which suggested the corpus-based MT approach would be good prospects.