Research On Dependency-to-String Model For Japanese To Chinese Statistical Machine Translation

As one of the most popular topic in natural language processing, statistical machine translation (SMT) has very important academic and commercial values. So far, SMT has experienced nearly two decades of research. One of the most encouraging research problems in SMT is to integrate linguistics knowledge into SMT model. Motivated by case grammar and Japanese dependency tree, this paper proposes a new Japanese dependency-to-string model for Japanese to Chinese machine translation. The main contribution are identified as follows:(1) Definition of Japanese case grammar. The new Japanese case grammar is modified with case grammar and Japanese dependency tree. Case grammar reordering grammar is derived from this grammar, which could be integrated into SMT model easily.(2) Definition of dependency-to-string model. New definition of rules for SMT (Japanese case grammar reordering rules and lexical phrase translation rules) and the acquisition algorithm of these rules are proposed.(3) Decoder algorithm. A bottom to up charting-parsing is introduced for Japanese to Chinese dependency-to-string model with an addition phrase translation.This paper proposes a method of integrating linguistics knowledge into SMT. Experimental results prove that our method performs well on long structural reordering and lexical translation and gains better BLEU score than other SMT model in Japanese to Chinese machine translation, and it’s also proved in Japanese to English machine translation.