Research on Statistical Model of Japanese-Chinese Name Translation for Japanese Kana

Abstract

In view of the lack of bilingual resources in Japanese-Chinese translation, and the naming entity of Chinese character naming entities, especially Japanese pseudonyms, it is difficult to use, so the Japanese-Chinese pure pseudonym translation is a statistical translation model. The model has been applied so far and there have been many problems. The first is limited by the size and quality of the parallel corpus, and the second is the lower accuracy of the translation. Based on this, the paper proposes a statistical model of Japanese pseudonymous Japanese name translation based on inductive learning method. This method can extract Japanese and Chinese named entities from Japanese and Chinese corpora, and convert them into Roman alphabet and Pinyin sequence. Example screening after similarity calculation. Then, the instance-based induction learning method is used to automatically obtain the Japanese-Chinese transliteration rule base of the named entity, and iteratively reconstructs the transliteration rule base through feedback learning. Through experimental research, it is found that the Japanese-Han name translation statistical model method for Japanese kana is simple and efficient, and the translation accuracy is overcome while overcoming the dependence on bilingual data.