AI人工智能翻译中结合模糊算法与改进注意力机制的分析  

Analysis of artificial intelligence translation combining fuzzy algorithms with improved attention mechanisms

作　　者：张菊玲[1] ZHANG Juling(Xi’an Siyuan University,Xi’an 710038,China)

机构地区：[1]西安思源学院,西安710038

出　　处：《自动化与仪器仪表》2024年第8期223-227,共5页Automation & Instrumentation

基　　金：2022年度陕西社科联青年专项研究项目《高质量视域下学前教育专业文化建设研究》(2023QN0380);2022年度陕西省教育科学“十四五”规划课题《地方文化融入幼儿园游戏活动的实践研究——以陕西地方文化为例》(SGH22Y1868)。

摘　　要：为了进一步提高神经机器翻译系统的翻译性能,提出一种融入预训练语言模型增强的神经机器翻译模型。一方面针对BERT预训练语言模型在融入神经机器语言模型中存在的灾难性遗忘问题,引入掩码矩阵策略进行缓解;另一方面通过对多注意力机制的内部融合和动态加权,使模型可以充分利用经过优化的BERT的输出信息,从而提高模型性能。结果表明,当掩映矩阵系数度为0.6时,使用门控机制进行加权融合的改进Masking-BERT模型,在实验数据集上的测试效果最佳,在英→中与中→英翻译任务中,相较于Transformer基线模型、RNNSearch模型与RNN-Deliberation模型,BLUE值分别提高了1.88、1.41;7.67、5.77和4.88、4.68,性能提升明显。在实际的英语教学过程中,搭载所提模型的AI人工智能系统,不仅可以满足课堂的翻译需求,且翻译准确度和课堂满意度的人工评分都较高,值得在英语教学中使用和推广。In order to further improve the translation performance of neural machine translation system,a neural machine translation 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 coefficient is 0.6,the improved Masking-BERT model with weighted fusion using the gating mechanism has the best test effect on the experimental 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 accuracy 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 

分 类 号：F399[经济管理—产业经济] TP29[自动化与计算机技术—检测技术与自动化装置]