Translation Unit for Simultaneous Japanese-English Spoken Dialogue Translation

It is strongly required to develop a simultaneous translation system that supports the smooth cross-lingual communications. We are putting forward a framework for simultaneous translation of incrementally executing detection, translation and generation processing for a translation unit which is smaller than a sentence. In this paper, we propose a translation unit which can be translated independently and immediately. Moreover, we define how to segment a source sentence into the translation units by using a parallel corpus. To confirm that such the translation unit is effective for simultaneous translation, we evaluated the translation unit from the viewpoints of the length and the automatic detectability.