Researched On Mongolian-Chinese Statistical Machine Translation Based On String To Tree Translation Model

Recently, as the booming growth of the economy of Inner Mongolia, the economic and cultural exchanges between the Mongolian people and the Han people becomes intimate gradually. This leads to the further demands of the translation between the traditional Mongolian and the Chinese, the research on the technology of machine translation between these two languages have the tremendous significance in mitigating this pressure. This paper combines a number of the merits of the translation between the traditional Mongolian and Chinese together, and has done an intensive research on the translating methods of Mongolian to Chinese machine translation based on the string to tree translation model.In order to alleviate the data sparse problem in Mongolian and Chinese words alignment, this paper has made some intensive researches on the Mongolian and Chinese words alignment of log linear model. This paper has integrated several features that is in the traditional Mongolian and Chinese translation into the log linear model framework, lowering the ratio of the alignment error rate. Simultaneously, this paper has done an intensive research on the translation method of Mongolia to Chinese machine translation based on the string to tree translation model. This method can draw out Mongolian string and Chinese syntactic tree alignment templates from the Mongolian and Chinese words alignment information and the Chinese syntactic information automatically, translates the traditional Mongolia to Chinese by using the templates and the decoding algorithm based on CYK, and makes use of the bilingual phrases to do some reprocessing on the candidate items as well. This paper has rewritten some of the artificial-made rules to the template base, and achieved the integration between the statistical method and the artificial method. It is proved by experiments that the model that based on the string to tree translating does the task of the translation between the traditional Mongolian and Chinese more appropriately, achieving a raising of the BLEU by 1.5% on the test set. the model that based on the string to tree translating is a kind of open model, and achieve the integration of statistical method and the artificial method more easily under the string to tree translation Framework.