Study On Chinese-mongolian Machine Translation Integrated With Morphological Analysis

Statistical technique is a mainstream technology of machine translationresearch currently. Precondition of statistical machine translation research isto have an adequate bilingual parallel corpus. In recent years,Chinese-Mongolian machine translation has made a certain achievements, butcomparing with English and the other languages, there is a big gap inMongolian machine translation. It mainly reflected in several ways: at first, inthe aspect of parallel corpus’s scale. There are great differences betweenChinese-Mongolian, European languages corpus’s scale andChinese-Mongolian bilingual corpus. Secondly, in Basic research like lexicalanalysis. Comparing with languages like English, Mongolian lexical analysisstarts lately, so it is not mature. These have already severely limited thedevelopment of the Chinese-Mongolian Machine Translation.Mongolian is a kind of language which is agglutinative, while Chinese isan isolating language and almost no morphological changes. There is a certaindegree of difficulty in machine translation for morphologically rich languages,but using morphological information reasonable will solve some problemseffectively, such as translation word form errors and the sparse data caused bythe small-scale corpus in machine translation research.From the perspective of Mongolian morphological information. Firstly,the research builds affix-based unsupervised morphological segmentationsystem and then uses the rules for post-processing to improve the accuracy ofsystem segmentation. The biggest advantage of the system is that it candirectly use the raw corpus for training; secondly, on the basis of theMongolian morphological analysis, we did a series of experiments whichapply target language’s morphological information to the Chinese-Mongoliantranslation system by using factor model. The experimental results show that the factor translation model with Mongolian morphological information is farbetter than the baseline experiments and it achieve the expected results.