Machine translation is one of the important research directions in natural language processing. In recent years, neural machine translation methods have surpassed traditional statistical machine translation methods in translation performance of most of language and have become the mainstream methods of machine translation. In this paper, we proposed syllable segmentation in Tibetan translation tasks for the first time and achieved better results than Tibetan word segmentation. Four kinds of neural machine translation methods, which are influential in recent years, are compared and analyzed in Tibetan-Chinese corpus. Experimental results showed that the translation model based on the complete self-attention mechanism performed best in the translation task of Tibetan-Chinese corpus, and performance of the most of the neural machine translation methods surpassed performance of the traditional statistical machine translation methods.
展开▼