The Research on Morpheme-Based Mongolian-Chinese Neural Machine Translation

In view of the rich morphology of Mongolian language and the limited vocabulary of neural machine translation, this paper firstly segmenting Mongolian words from different granularity, which are the segmentation of separates morphological suffixes and the segmentation of Ligatures morphological suffixes. For Chinese, we use word segmentation and word division. Then, we studied the morpheme-based Mongolian-Chinese end-to-end neural machine translation under the framework of bidirectional encoder and attention-based decoder. The experimental results show that the segmentation of Mongolian word effectively solves the data sparsity of Mongolian, and the morpheme-based Mongolian-Chinese neural machine translation model can improve the quality of machine translation. The best NIST and BLEU values of the morpheme-based Mongolian-Chinese Neural Machine Translation results were respectively reached 9.4216 and 0.6320.