Research on Mongolian-Chinese Machine Translation Based on Generative Method

Neural machine translation methods based on auto-regressive Recurrent Neural Networks (Auto-regressive RNN) are constrained by sequential decoding methods during translation prediction, which in turn results in severe translation delays. In addition, the error transmission during training will lead to error accumulation, which will affect the quality of the model. Therefore, in this paper, using the generative method, the latent vector space based on the source language and the target language is constructed. The advantage is that the translation decoding is only subject to the independent distribution in the potential space, and is not affected by the historical target translation, so the translation delay of the target language is alleviated, and the error accumulation problem of the traditional method is also avoided. Moreover, the CWMT2019 Mongolian-Chinese translation task is used to verify the constructed model, and the results show that it has excellent results, in which way its effectiveness is fully demonstrated.