基于神经机器翻译的英语语法错误纠正方法分析  

Analysis of English Grammar Error Correction Methods Based on Neural Machine Translation

作　　者：王蕊[1] WANG Rui(Xi'an Siyuan University,Xi'an 710038 China)

机构地区：[1]西安思源学院,陕西西安710038

出　　处：《自动化技术与应用》2021年第8期57-60,74,共5页Techniques of Automation and Applications

摘　　要：本文基于神经机器翻译提出了英语语法错误纠正方法,并以实验进行了验证分析,结果表明,利用sampling解码策略的back-translation数据增强方法,提高了模型纠错性能;通过反向模型生成伪错误句子时,sampling解码策略效果更好;对抗训练有利于反向模型英语语法错误生成,保障了伪错误-纠正平行句对真实性,且推进了正向模型训练开展。总之,本文设计方法的效率与性能都实现了明显提高与优化。In this paper, an English grammar error correction method is proposed based on neural machine translation and verified and analyzed by experiments. The results show that the back-translation data enhancement method using the sampling decoding strategy can improve the model error correction performance. When false error sentences are generated by the reverse model, the sampling decoding strategy is more effective. The antagonistic training is beneficial to the generation of grammatical errors in the reverse model, to ensure the authenticity of pseudo-error-corrected parallel sentence pairs, and to promote the development of forward model training. In a word, the efficiency and performance of the design method in this paper are improved and optimized obviously.

关 键 词：英语语法 神经机器翻译 错误纠正 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]