ANALYSIS, GENERATION AND SEMANTIC REPRESENTATION IN CONTRAST - A CONTEXT-BASED MACHINE TRANSLATION SYSTEM

This paper describes both a Japanese language analysis system and an E nglish language generation system. The knowledge utilized by these sys tems has been incorporated into a context-based machine translation sy stem (CONTRAST) developed by the Electrotechnical Laboratory (ETL). Th e Japanese language analysis component of the CONTRAST system consists of individual modules to analyze syntax, semantics and context. The s yntactic analysis module is a completely parallel parsing system based on Earley's algorithm. The semantic analysis module conducts analysis based on case relations. The contextual analysis module creates an in termediate representation that correspond to original text. It does th is by placing events included in the semantic analysis results for eac h individual sentence within a framework that is read out from the gen eral knowledge described as a process model in the concept dictionary. The English text generation system of CONTRAST begins with this inter mediate representation. Then, referring to the concept dictionary (com mon with the analysis module), it creates English text according to ge neration rules created from the real life text data dependent knowledg e relating to English newspaper articles. The format of the concept di ctionary, which consists of descriptions of general knowledge, and the format of the intermediate representation, which is a description of the contextual information contained in each body of text, are the sam e. The outline of the conceptual system employed and studies on the pr ocess of creating this system are also discussed.