TR-IT-0327 :2000.01

Michael PAUL, Kazuhide YAMAMOTO

Resolution of Referential Expressions within TDMT

Abstract:This report deals with the processing of contextual phenomena within the framework of the spoken-language translation system TDMT(Transfer-Driven Machine Translation). We apply a corpus-based approach to resolve referential expressions in Japanese utterances. In our approach a machine-learning algorithm (decision tree) is utilized to select automatically the attributes from a tagged training set necessary for the resolution task. The task-specific decision tree is applied to the input data and the knowledge about the obtained reference objects is used for a context-adopted translation of the source utterances into English and German.