In order to further improve the translation performance of neural machine translation system,a neural machine transla-tion model enhanced by pre-trained language model is proposed.On the one hand,the mask matrix strategy is introduced to alleviate the catastrophic forgetting problem of BERT pre-trained language model integrated into neural machine language model.On the other hand,the model can make full use of the output information of the optimized BERT and improve the performance of the model through the internal fusion and dynamic weighting of the multi-attention mechanism.The results show that when the Masking matrix coeffi-cient is 0.6,the improved Masking-BERT model with weighted fusion using the gating mechanism has the best test effect on the ex-perimental data set,and in the English-Chinese and Chinese-English translation tasks,Compared with Transformer baseline model,RNNSearch model and RNN-Deliberation model,the BLUE value is increased by 1.88 and 1.41 respectively.7.67,5.77,4.88,4.68,performance improvement is obvious.In the actual English teaching process,the AI artificial intelligence system equipped with the proposed model can not only meet the translation needs of the classroom,but also achieve high manual scores of translation accura-cy and class satisfaction,which is worthy of use and promotion in English teaching.
关键词
人工智能/翻译系统/改进注意力机制/预训练语言模型/英语教学
Key words
artificial intelligence/translation system/improving attention mechanisms/Pre-trained language model/English teaching
维普
万方数据