Controlled Language and Knowledge-based Machine Translation: Principles and Practice 1

For applications with certain properties (well-deened technical domain, technical vocabulary, simple grammatical style), use of a controlled language can enhance the accuracy of knowledge-based machine translation (Baker, et al., 1994). In the continuing development of the KANT system (Mitamura, et al., 1991; Mitamura and Nyberg, 1995), we have explored diierent sublanguage techniques which limit the complexity of natural language analysis, thus increasing the likelihood of an accurate translation. In this presentation, we will describe some of the general techniques that have been developed for sublan-guage use within the KANT system. 1 Introduction In recent years there has been increasing interest in the use of controlled source languages in MT (cf. Adriaens and Schreurs, 1992 and the references cited there). In this paper, we focus on the use of controlled input language in the KANT translation system (Mitamura, et al., 1991). Controlled English is used to improve the clarity of expression in the source text, and to improve the quality of the translation output. KANT is a knowledge-based, interlingual machine translation system, a descendant of the KBMT-89 system (Goodman and Nirenburg, eds., 1991). KANT uses explicit source language lexicons, grammars and domain semantics to produce an interlingua representation (IR) for each sentence. Each IR is a semantic frame containing features and semantic roles, which may be lled by other IR frames. If a source sentence has more than one possible analysis, KANT produces more than one IR structure. The input to the target language generation module is the set of IRs produced for a source sentence. The decoupling of the analysis and generation phases has