INCREMENTAL TRANSFER IN ENGLISH-JAPANESE MACHINE TRANSLATION

Since spontaneously spoken language expressions appear continuously, t he transfer stage of a spoken language machine translation system have to work incrementally. In such the system. the high degree of increme ntality Is also strongly required rather than that of quality. This pa per proposes an incremental machine translation system. which translat es English spoken words into Japanese in accordance with the order of appearances of them. The system is composed of three modules: incremen tal parsing, transfer and generation, which work synchronously. The tr ansfer module utilizes some features and phenomena characterizing Japa nese spoken language: flexible word-order, ellipses, repetitions and s o forth. This is influenced by the observational facts that such chara cteristics frequently appear in Japanese uttered by English-Japanese i nterpreters. Their frequent utilization is the kev to success of the e xceedingly incremental translation between English and Japanese, which have different word-order. We have implemented a prototype system Syn c/Trans, which parses English dialogues incrementally and generates Ja panese immediately. To evaluate Sync/Trans we have made an experiment with the conversations consisting of 27 dialogues and 218 sentences. 1 90 of the sentences are correct, providing a success rate of 87.2%. Th is result shows our incremental method to be a promising technique for spoken language translation with acceptable accuracy and high real-ti me nature.