Monolingual Data Augmentation based Chinese to Tibetan Machines Translation

Chinese-Tibetan machine translation has great value to the promotion of general use of national language, cross-ethnic communication in Tibetan area and the implementation and promotion of policies. Recently, neural machine translation has made great progress in the machine translation of Chinese-English, English-French and other languages with the help of large-scale training data and powerful modeling capabilities of deep learning models. Neural machine translation usually relies on large-scale parallel sentence pairs for training. However, the machine translation of Chinese and minority languages such as Chinese-Tibetan, Chinese-Mongolian and other Chinese and minority languages lack large-scale bilingual parallel data, resulting in the lack of performance of Chinese and ethnic language machine translation . Aiming at resolving the problem of lack of training data for Chinese and ethnic language machine translation, this paper utilizes the current situation of large-scale annotation data and mature natural language processing tools in Chinese, and proposes a way to use monolingual data augmentation to expand the scale of Chinese-Tibetan parallel sentence pairs. To improve the quality of Chinese-Tibetan machine translation. This paper has carried out a variety of data augmentation methods. The results on three main neural machine translation models show that the monolingual data augmentation method can effectively improve the quality of machine translation, which is a machine translation between Chinese and national languages. The translation provides a reference