NMT Mobile Translation Software Development Based On Paraphrase Model

Each model of translation proposes improvement methods for a certain problem.These methods may cause additional problems,while improving some other problems.At the same time,each method's description of the source language will produce errors and cannot fully express itself.Paraphrase refers to different expressions of the same semantics.For the characteristics of different systems,we use paraphrase method to rehearse the translation to make it more in line with our language habits.Then,we use the system combination method which can combine the advantages of different systems to improve the quality of the translation.Finally,the paper designs the Mongolian-Chinese mobile phone translation software based on translation and paraphrase model.In this paper,experiments were conducted as follows,(1)When rehearsing Statistical Machine Translation(SMT)translations,the corpus was respectively derived from SMT,Recurrent Neural Networks(RNN),and Convolutional Neural Networks(CNN)translations.The models were based on RNN and CNN,respectively.The best result raised the BLEU point from 29.55 to 30.86.When the RNN translations was rehearsed,the corpus was derived from RNN translations.The models used RNN and CNN.The best BLEU value increased from 30.71 to 31.72.(2)Using the MEMT to combine the SMT and RNN candidate translations which raised BLEU point to 31.01 and 31.72,respectively.Experiments proved that paraphrase and the method of system combination can improve the translation performance.Finally,this paper uses the C/S architecture and MVC mode to implement the NMT mobile phone Mongolian-Chinese translation software based on the paraphrase model.The Mongolian sentence was translated first,and then paraphrased by the software translation.Finally the Mongolian was transformed to Chinese.