Research on Mongolian-Chinese Neural Machine Translation Based on CSGAN and Co-training
Ren Qing-dao-er-ji, Yujuan Yin, Lixia Wen, He-Ya Sa, Shi Bao · Journal of Physics Conference Series · 2022
Abstract Aiming at the problems of poor translation quality and scarcity of Mongolian-Chinese parallel corpus resources in Mongolian-Chinese neural machine translation, this paper proposes a method based on conditional sequence of generative adversarial networks (CSGAN) and collaborative training. This method uses the CSGAN network based on adversarial training to train the translation model and obtains a better translation. At the same time, in order to make full use of the existing corpus resources, a collaborative training method is adopted on the basis of the above translation model to integrate the translation into the translation results of the high-resource corpus, and further improve the performance of Mongolian-Chinese machine translation. By comparing the BLEU values of different translation models, it is shown that using CSGAN-based and collaborative training methods can significantly improve the quality of Mongolian-Chinese translation models.