融合跨语言记忆网络与语义信息的神经机器翻译系统架构设计研究
Research on the Architecture Design of a Neural Machine Translation System Integrating Cross Language Memory Networks and Semantic Information
白雯1
作者信息
摘要
当前的神经机器翻译系统在关键外文传统文化语义信息翻译时存在翻译错漏问题,因此,研究在具备跨语言记忆的转换器网络模型的基础上融入关键语义信息,并以此设计系统架构,同时对其进行了验证.实验结果表明,研究模型的分数始终高于对模型,整体上在中英数据集上增长了 1.60～2.25分左右,在英德数据集上增长了 0.38～2.36分左右.构建的系统在中英翻译数据集上的双语评估得分为37.6分,优于对比系统且同等翻译量下最低耗时0.36 s,准确率达到85.4％.同时其在传统文化诗歌翻译中总分数值最高达到0.90,高于对比方法.综合来看,研究设计的神经机器翻译系统在外文传统文化相关翻译中具备较高的性能,且在传统文化实际的实时翻译中具备有效性.
Abstract
The current neural machine translation system suffers from translation errors and omissions when translating key tradi-tional cultural semantic information in foreign languages.Therefore,this study integrates key semantic information into a converter network model with cross language memory,designs a system architecture based on this,and verifies it.The experimental results showed that the scores of the research model were consistently higher than those of the control model,with an overall increase of about 1.60-2.25 points on the Chinese and English dataset and about 0.38-2.36 points on the British and German traditional culture data-set.The bilingual evaluation score of the constructed system on the Chinese English translation dataset is 37.6 points,which is better than the comparison system,and the lowest time consumption is 0.36 seconds under the same translation volume,with an accuracy rate of 85.4％.At the same time,its total score in traditional cultural poetry translation reached the highest of 0.90,higher than the comparative method.Overall,the neural machine translation system designed in the study has high performance in translation related to foreign traditional culture,and is effective in real-time translation of traditional culture.
关键词
Transformer/语义信息/NMTS/BLEU/准确率Key words
transformer/semantic information/NMTS/BLEU/accuracy引用本文复制引用
出版年
2024
NETL
NSTL
万方数据