基于深度学习的英语短语译文智能校对系统  

Intelligent proofreading system of English phrase translation based on deep learning

作　　者：杨冰[1] YANG Bing(Xi’an Fanyi University,Xi’an 710105,China)

机构地区：[1]西安翻译学院,西安710105

出　　处：《自动化与仪器仪表》2022年第8期185-188,193,共5页Automation & Instrumentation

基　　金：陕西省教育科学“十三五”规划2020年度课题《新时代下高校英语教学中的听说能力培养方式研究》(SGH20Y1517)。

摘　　要：针对英语短语译文智能校对查错F1-score值低问题,提出基于深度学习的英语短语译文智能校对系统。硬件方面,针对晶振电路和接口电路进行设计。软件方面,根据自动翻译流程提取语义融合的特征参量,创建语义本体概念树。从实体词信息、逻辑关系信息、短语长度信息三个维度抽取句子特征,设计英汉短语对齐算法。依托于深度学习原理和Seq2Seq模型,加入了Bi-LSTM单元和注意力机制,通过网络训练得出自动化短语译文校对规则,生成智能校对模型。最后,采用Teacher Forcing强制训练的方式,构造解码端的训练与预测函数,完成智能校对系统的整体设计。系统测试结果表明:所提系统的查错F1-score值为0.93,相比文献[3]、文献[4]系统提升了26%与24%。Aiming at the problem of low F1 score of intelligent proofreading of English phrase translation,an intelligent proofreading system of English phrase translation based on deep learning is proposed.In terms of hardware,the crystal oscillator circuit and interface circuit are designed.In terms of software,the feature parameters of semantic fusion are extracted according to the automatic translation process,and the semantic ontology concept tree is created.Sentence features are extracted from three dimensions:entity word information,logical relationship information and phrase length information,and an English Chinese phrase alignment algorithm is designed.Relying on the principle of deep learning and seq2 seq model,Bi LSTM unit and attention mechanism are added.The automatic phrase translation proofreading rules are obtained through network training,and the intelligent proofreading model is generated.Finally,the training and prediction function of the decoding end is constructed by using the teacher forcing forced training method,and the overall design of the intelligent proofreading system is completed.The system test results show that the error checking F1 score value of the proposed system is 0.93,which is 26% and 24% higher than that of literature [3]and literature [4].

关 键 词：深度学习 译文校对 短语翻译 神经网络 系统设计 语义本体模型 

分 类 号：TP39[自动化与计算机技术—计算机应用技术] TN99[自动化与计算机技术—计算机科学与技术]