Research On Neural Machine Translation Of Academic Thesis Based On Multi-Branch Tree Neural Network

Based on the rapid development of computer technology,especially the growing employment of Graphics Processing Unit(GPU)in recent years,neural networks are no longer unreachable as when they were first proposed.Neural machine translation(NMT),i.e.the application of neural networks to the traditional machine translation,exploits the nature of neural network to implement the autonomous machine learning,including the acquaintance of translation rules and completion of translation tasks as a substitute for human.Under the compressive impact of the deep learning technology,related researchers commence a problem-solving progress from a new perspective.Based on that,the research on machine translation for academic thesis based on the multi-branch tree deep learning network.We hope that future researchers can be inspired by our preliminary exploration of related fields.Based on the research of traditional neural machine translation networks and the application analysis of tree-structure deep learning network,this research aims to build a multi-branch tree-based neural machine translation(MbTbNMT)that combines traditional neural networks with multi-branch tree deep learning networks,which can capture the syntactic information of words by re-modeling the multi-branch tree deep learning networks.First of all,this thesis introduced the traditional neural machine translation and tree-structure deep learning network,mainly including the principle and detailed construction process of Tree-LSTM,and the applied algorithms were roughly introduced.Then,the implementation method of MbTbNMT model is introduced in detail.By modeling the hidden state nodes of the LSTM encoder,the upper node contains more child nodes information.Combined with the Attention mechanism,the model can extract the grammatical information of academic papers.Next,a new corpus construction method based on web crawler was introduced.This research was built and implemented on the Tensorflow deep learning platform,where the new corpus was employed to train the model.After detecting the translation performance of the model to specific corpora about academic thesis,the experimental results showed that the translation performance has been improved.In order to speed up the convergence process during training,a stochastic gradient descent(SGD)was applied to optimize the model.Finally,a comparative analysis between MbTbNMT model and two classic machine translation models based on the translation results were given when the same corpus were applied.Compared with those classic models,MbTbNMT model cannot only extract the syntactic information of the corpus related to academic thesis,but also has lower system perplexity,and better translation performance.