Research And Implementation Of Neural Machine Translation Model Based On Fusion Of Dependency Syntactic Information

With the development of the artificial neural network,the neural machine translation model has shown great potential and good development prospects and has become the mainstream of machine translation research in academia and industry.The neural machine translation model has surpassed the traditional statistical machine translation model in some translation tasks and achieved the effect of SOTA(State-of-the-art).The neural machine translation model has high translation fluency,but there are some problems such as low translation accuracy,easy over-translating/omitting,low interpretability of model,and difficulty in long sentence translation.Under this circumstance,more and more works start to use linguistic knowledge to improve the performance of the neural machine translation model.The results of many pieces of research show that the integration of linguistic knowledge enriches the translation information that can be learned from neural machine translation models and improves the performance of the neural machine translation model.However,how to efficiently integrate the additional,regular and variable linguistic knowledge into the model has become an important topic in the research of neural machine translation.Therefore,this paper mainly studies the inefficiencies and ways of introducing linguistic knowledge into neural machine translation models.The specific research work is as follows.1)In views of the low accuracy of the neural machine translation model,the difficulty of long sentence translation and the existing linguistic knowledge fusion mode,this paper proposes the method of “Dependency-based Local Attention Approach to Neural Machine Translation” model.This model combines the dependency information in the linguistic knowledge with the local attention mechanism,and uses the linguistic knowledge to improve the attention mechanism,so as to integrate the dependency information into the neural machine translation model in a more accurate and effective way to improve the translation effect.2)In views of the inefficiency fusion of linguistic knowledge and the extra cost in the process of citation,based on the encoder-decoder neural machine translation model,this paper proposes the method of the neural machine translation model equipped with a multi-layer attention mechanism.This model changes the usual way uses linguistic knowledge as additional information and attempts to fully utilize the linguistic knowledgelearned by the neural machine translation model itself.The neural machine translation model has been improved in a more efficient and faster way.