Machine translation using corpus-based acquisition of transfer rules

In this paper we present a Japanese-English transfer- based machine translation system. Our main research contribution is that the transfer rules are not handcrafted but are learnt automatically from a parallel corpus. We learn specific transfer rules from sentence pairs, which are then generalized in a consolidation phase to avoid overtraining. The system has been implemented in Amzi! Prolog, which offers scalability for large rule bases, full Unicode support for Japanese characters, and several APIs for the seamless integration of the translation functionality into common office environments, e.g. Microsoft Word. The dynamic nature of our system allows for an easy customization of the rule base according to the user´s personal preferences.