Research On Chinese-Vietnamese Machine Translation Based On Neural Network

In the context of the "Belt and Road Initiative",the 19 th China-ASEAN Expo and the China-ASEAN Business and Investment Summit were successfully concluded in Guangxi,China,and the cooperation between China and ASEAN countries has become increasingly close.Vietnam is one of the ten ASEAN countries.It has close exchanges and cooperation with China in politics,economy,culture and other aspects.It inevitably faces the problem of Chinese-Vietnamese mutual translation in the processing of many businesses.Currently,the performance of Chinese-Vietnamese neural machine translation is not well.Against this background,studying Chinese-Vietnamese machine translation has great significance and practical value.The current research work on Chinese Vietnamese machine translation has some problems,such as incomplete information capture of Chinese sentences,defects in model structure design,and insufficient corpus,which lead to poor translation performance.Therefore,this paper conducts research work from three aspects: improvement of word embedding methods,model structure design,and parameter optimization:(1)A word embedding method incorporating pinyin features is proposed.The pinyin of Chinese characters carries part of the syntactic and semantic information.Previous Chinese-Vietnamese machine translation research often ignores the processing of pinyin information of Chinese characters.Therefore,this paper designs a fusion word embedding layer that can capture pinyin information to enable the model to acquire the ability to process pinyin information..The experimental results prove that this method helps to improve the accuracy of the translation model.(2)A Chinese-Vietnamese machine translation fusion model that introduces convolutional structure,recurrent structure and self-attention mechanism is proposed,named TC-TL model.At present,the neural machine translation model is mainly implemented based on the encoder-decoder framework,and there are three sub-network structures: recurrent neural network,convolutional neural network,and Transformer.By analyzing the advantages and disadvantages of the three subnetwork structures,this paper proposes a new model that integrates a convolutional neural network structure and a self attention mechanism at the encoder end,and a cyclic neural network structure and a self attention mechanism at the decoder end,for Chinese tasks other than machine translation.The experimental results show that the translation performance of the TC-TL model has been greatly improved compared to the baseline model.(3)A Chinese Vietnamese machine translation method based on multi task transfer learning optimization is proposed.Through the idea of knowledge sharing and knowledge transfer,the fusion method of multi task learning and transfer learning and the transfer learning method of using only English as the medium language are used to pre train the parameters of the Chinese Vietnamese machine translation model.The language knowledge is learned using high resource corpus data,solving the problem of insufficient performance of Chinese Vietnamese machine translation in low resource environments,and further improving the translation performance of the model.