Research on Tibetan-Chinese Neural Machine Translation Integrating Syntactic Information

Abstract

In recent years, Neural Networks have gradually replaced other methods in the field of Machine Translation and become the mainstream which have excellent performance in many languages. However, the performance of Neural Machine Translation mainly relies on large-scale parallel corpora, which is not ideal for low-resource languages, especially Tibetan-Chinese Machine Translation. In order to obtain the best translation performance with more external information on the basis of limited corpus, this paper introduces syntactic information, that is, adding part-of-speech(POS) tags as input features in the training process. Experiments verify the effectiveness of this method, which can improve the translation performance to a certain extent.