Recent machine translation technology has achieved the level where several systems are in daily use. Conventional approaches to system enhancement are mostly aimed at handling written text, such as technical documents, which mainly contain assertive sentences. However, dialogue utterances involve various kinds of intention expressions and are often fragmentary depending on the situation. To improve conventional systems so that they can translate these utterances, translation examples of both situation dependent expressions and idiomatic expressions for daily use must be flexibly applied. Conventional technology is not completely satisfactory because conventional analytic methods can neither give a unique structure from whicha target sentence can be easily generated in a systemaatic manner nor handle illfomed speech recognition results including misrecognized parts. The author addresses the problem of both spoken language translation technologies and speech translation.
展开▼