Attention-guided Graph Convolutional Network For Chinese-English Machine Translation

With the maturation of information technology in the new era,people realize new text translation through computers and networks.Compared with traditional manual translation,this method not only reduces the consumption of human resources but also greatly reduces the proportion of time taken up.In addition,it also improves the accuracy of translation and truly liberates useless human consumption.However,the web-based translation method only translates the time series statements,ignoring the integrity and failing to accurately translate polysemy and paragraph.Therefore,how to use the new model for machine translation optimization is the current focus,but is also difficult.In foreign language translation,long and short sentences occupy the main length of articles,and these sentence patterns have various semantic information and complex logical structure,but also have a close connection with the context.Chinese-English translation of a single sentence not only loses semantic information but also destroys the complex logical structure.In other words,data sparsity,mistranslation,interpretability,and discourse translation are not in line with the needs of contemporary machine translation,which makes the overall relevance of machine translation a challenging and relatively independent project.At present,the neural network has occupied the mainstream position in machine translation models,but the development history and challenges of machine translation require further optimization of the model.In view of this,a novel machine translation optimization model based on an attention-guided graph convolution network is proposed,which can preserve word element features and paragraph hierarchy information through the combination of a multi-attentional mechanism and graph convolution neural network structure.The model consists of multiple identical blocks,and each block contains three types of layers: attentional guidance,dense connection,and linear combination.Three layers are used to solve the problems of omission,interpretability,and text translation in neural network-based translation.To further optimize the problem of data sparsity in the field of machine translation,this paper uses a new data augmentation method based on back-translation and multivariate noise mixing.This method combines the characteristics of back translation and noise-based data augmentation and selects the most suitable multivariate noise mixing and optimization model fusion based on the WMT21 data set selected in the specific experiment.Finally,a large number of experiments are carried out on the data set,and the BLEU index is used to test whether it is superior to other traditional algorithms.Through the analysis of multiple results such as comparative experiment,interpretability experiment,and ablation experiment.In an example analysis,it is intuitively shown that the optimization model achieves the ideal effect.