Interlingua Design for TRANSLATOR

Abstract

The interlingua approach to machine translation (MT) is characterized by the following two stages: 1) translation of the source text into an intermediate representation, an artificial language (interlingua) which is designed to capture the various types of meaning of the source text and 2) translation from the interlingua into the target text. Over the years a number of MT projects tried to develop interlinguabased systems. In these projects the amount of linguistic and encyclopaedic knowledge used to produce intermediate representations was quite limited. However, even at that level difficulties connected with encoding knowledge seemed overwhelming. The TRANSLATOR project at Colgate University benefits from recent developments in knowledge representation techniques. The text of its interlingua text reflects syntactic, lexical, contextual, discourse (including speech situation) and pragmatic meaning of the input This paper discusses the lexicon and grammar of the interlingua used in TRANSLATOR, and touches upon the structure of the bilingual (source language to interlingua) dictionaries. The actual compilation of the interlingua dictionary and additional knowledge bases is an empirical process during which modifications to the original formulations are expected to occur. At all times in the design process the authors were guided by the desire to make decisions that are 'literate' from the point of view of linguistic theory and the experience of knowledge representation in artificial intelligence.