Research On Term Automatic Translation Technique For Japanese-Chinese Machine Translation System

As the crystallization of human wisdom, terms concentrate the core knowledge of the specific field. With the development of the information age, the study of the terminology has increasingly become the focus of attention. Term translation, as the research focus of terminology, is widely used in the machine translation, cross-language information retrieval, compilation of the bilingual dictionaries, etc.The thesis mainly analyses characteristics of Japanese terms and Chinese terms, deeply researches on international patent classification information of patent document which includes many terms, common character information in Japanese-Chinese term words, and relative conditional entropy information beneficial to term words collocation, etc. And we employs Moses decoder to implement automatic translation system from Japanese to Chinese uses the features which are above information as the features, N-gram language model feature, and phrase probability features, ect. The work of this thesis mainly includes:First, add the international patent classification information as the domain information feature of the term to the translation model. Because the term domain information has important effect on term translation, this thesis makes statistic both international patent classification number and word frequency of the patent documents including terms and term components, and affiliates the term domain information as one of features to translation cost computing.Second, regard similar character information in Japanese-Chinese term words as a feature of translate model. The thesis picks up common character information in the Japanese-Chinese term as a feature to join in state-shifting cost computing, and help select words from some translating candidates.Third, regard relative conditional entropy information beneficial to term words collocation as a feature of translation model. This feature which associates with language model, distinguishs the best translating candidates from others.The experimental results indicate that the effective integration of the features improves the translation quality of the Japanese-Chinese term translation. Besides, the thesis deeply researches on stack-based decoding process and translating result evaluation, including translation options generation, cost estimation, search with pruning, translation results generation, evaluating standard, etc. Finally, this thesis summarizes effects of each feature on term translation in the term automatic translation, and looks forward to the future work of term translation, including adding semantic information to term translation, etc.