A Study of Tibetan-Chinese Machine Translation Based on Bidirectional Training Strategy
Chao Wang, Songsi Yan, Jie Zhu, Zehui Xu, Yashan Liu, Zezhou Xu · 2022
The lack of Tibetan-Chinese parallel corpus is the main reason that affects the effectiveness of Tibetan-Chinese neural machine translation. In this paper, bidirectional training strategy is invoked for the study of Tibetan-Chinese Transformer machine translation model, and the quality of translation is improved by combining with back translation method. The experimental results show that the BLEU values of the translations obtained by this method are improved by 1.27%, 1.25%, 1.09%, 0.90%, 1.21%, 1.16%, respectively, compared with the baseline experiments.