Innovative user experiences brought by statistical machine translation

Machine translation research has changed dramatically in the last 10 years. There have been two major innovations. The first is statistical machine translation. Unlike conventional rule-based translation where the translation rules are made manually by experts, here the translation rules are learned automatically from a bilingual corpus. The second is the automatic evaluation of translation accuracy, a low-cost and short-turnaround alternative to human subjective evaluation. In this talk, I will present our work on statistical machine translation and its automatic evaluation, in particular, between English and Japanese, which is one of the most difficult-to-translate language pairs in the world.