A Study On Chinese-Mongolian Hybrid Machine Translation And The Related Technologies

Each of the Machine Translation (MT) methods has its own advantages and limitations. The purpose of hybrid MT method is to make full use of various MT's advantages, avoid their shortages, optimize translation result, and improve the whole performance of the MT system.In view of the resources available to us, after referencing the research achievements of related researchers, this paper studies about the Chinese-Mongolian hybrid machine translation system. We built a phrase-based Chinese-Mongolian SMT system using the existing open source tools. At the same time, we have established language resources for automatic evaluation platform of Chinese-Mongolian machine translation system. In order to improve the performance of the phrase-based Chinese-Mongolian SMT system, this thesis made the following research and experiments:(1)By joining the Chinese-Mongolian bilingual dictionary and doing morphological analysis for additional components of which Mongolian nouns cases, nouns plural forms and genitive cases to solve unknown word problems in translation.(2) We proposed Chinese sentence reordering method based on the Mongolian word order and handled the large number of word order errors that appeared in the phrase-based SMT. First of all, syntactic analysis needs to be done to the Chinese sentences. Then, according to the reordering rules Chinese sentences have to be converted to a form which similar to Mongolian sentence word order. Finally, the reordered Chinese sentence is sent to statistical decoder for monotonous decoding.(3) In the study of phrase-based Chinese-Mongolian SMT, we noticed that there are some errors in the Chinese-Mongolian quantifier translation. We compared Chinese-Mongolian quantifier translation methods and concluded one-to-one, many-to-one, one-to-zero and one-to-many relationships of translation between Chinese quantifier and Mongolian quantifier.Machine Translation Evaluation plays an important role to the development of Machine Translation technology. We provided the training corpus, development set and test set for Chinese-Mongolian daily evaluation task to CWMT2009 machine translation evaluation. In order to prepare corpus we have developed a rule-based Mongolian sentences automatic segmentation program and a converting program from Mongolian Latin to UTF-8 code. This thesis introduced the method and the process for developing these programs. Finally, we present the Chinese-Mongolian hybrid machine translation system's experiments and results analysis.