Machine translation method research based on G-Transformer

The invention discloses a machine translation method research based on G-Transform, and relates to the field of neural machine translations, the machine translation method research based on G-Transformer is designed by means of a recurrent neural network GRU and a complete self-attention network Transform, the recurrent neural network GRU is responsible for recoding word vectors used for machine translation, vectors considering context information of sentences are output by utilizing the characteristics of the recurrent neural network, and the complete self-attention network Transformer is responsible for recoding word vectors generated by the recurrent neural network, so that subsequent translation is realized. 本发明公开了一种基于G-Transformer的机器翻译方法研究，涉及神经机器翻译领域，该基于G-Transformer的机器翻译方法研究借助循环神经网络GRU和完全的自注意力网络Tranformer进行设计，循环神经网络GRU负责对用于机器翻译的词向量进行重编码，利用循环神经网络的特性输出考虑了句子上下文信息的向量，完全的自注意力网络Tranformer负责对前者生成的词向量进行再编码，从而实现后续翻译。