Chinese-based Dependency Grammar - The Naxi Language Statistical Machine Translation Research

Machine translation is the hot and difficult area in the field of natural language processing, and it is important to communicate between different languages. In recent years, our country gives more attention to information application of the Naxi minority language, so it is very important to realize intercommunication between Naxi language and Chinese. Using the machine translation to accomplish this goal is an important method. Because the Chinese language and the Naxi language are different in the syntactic structure, using a simple statistical machine translation method based on word or phrase can’t get a good effect on translation. Therefore, due to the syntactic features between Chinese and Naxi language, this paper uses the dependency relation of the two languages in the syntactic structure to research the Chinese-Naxi statistical machine translation. The main results got by this paper are as following:(1) The method of Chinese-Naxi statistical machine translation based on improved dependency tree-to-string translation template. Translation template is an important basis and method to improve the effect of statistical machine translation system. This paper takes the Naxi language and Chinese as the research object, and constructs the dependency tree-to-string translation template which is required in Chinese-Naxi statistical machine translation system. In the construction of translation template, we propose the template extraction method which adds merge operation and decoding algorithm, and according to the decoding algorithm, inosculate the translation templates into the decoding stage in Chinese-Naxi statistical machine translation to complete the translation. The experiments show that in the support of certain scale of translation template based on improved dependency-tree to string, the effect of Chinese-Naxi language based on statistical machine translation increases greatly.(2) Construction of the dependency language model based on the Naxi language and constraint on the decoding results. According to the syntactic features of the Naxi language, inosculate the Naxi language syntactic structure information into the model and put forward the method of training. When choosing the candidate translation decoding results, we calculate the score of the normally decoded NBEST candidate translation results, and adjust their sequence in order to improving the accuracy of translation results. The final results show that the proposed language model based on dependency relations has a great help for the selection of the best translation results.(3) Using the software of word alignment and syntactic analysis, combining the improved dependency tree-to-string translation template, decoding algorithm and dependency language model, building Chinese-Naxi statistical machine translation system based on dependency syntactic.