Research On Machine Translation Technology Based On Deep Learning

Neural Machine Translation(NMT)continues to bring exciting news to the Machine Translation(MT)field since its inception.However,NMT uses only a single Neural Network to convert natural languages.Although this method can simplify the translation process,it has two shortcomings:(1)Compared with the Statistical Machine Translation,NMT is more sensitive to sentence length;(2)The end-to-end implementation fails to make explicit use of linguistic knowledge for better translation performance.To address the above problems,we put forward corresponding methods as follows:In view of the problem one,this paper proposes a NMT method based on divide and conquer strategy.Based on the idea of the strategy mentioned above,the method identifies and extracts the Maximal-length Noun Phrase(MNP)in a sentence,and retains the special mark or head word to form the sentence frame with the rest words.Through translating the MNP and the sentence frame by NMT system,and recombining the translations,the method alleviates the problem that NMT has poor translation performance on long sentences.Experimental results show that the BLEU score obtained by our method is improved by 0.89 compared with the baseline method.In view of the problem two,this paper proposes a NMT method based on multiple sequence encoding.This method can encode the lexical and syntactic linguistic knowledge in the form of sequence and construct an intermediate vector together with the source sentence to participate in the training and decoding process.In this paper,two kinds of encoding networks are implemented: multiple gated Recurrent Neural Network(Multi-GRNN)and multiple bidirectional RNN(Multi-BiRNN).The methods are respectively applied to traditional NMT and attentional NMT.Experimental results show that compared with the baseline method,the BLEU scores obtained by our method are increased by 1.35 and 1.14 respectively.Finally,we integrate the two methods proposed in this paper to form a Chinese-English NMT system.The BLEU score of this system is increased by 1.82 compared with the baseline system.Experimental results show that the two methods can effectively enhance the performance of NMT system.