摘要
针对英语翻译错误检测中的语法错误检测问题,构建了一个以图神经网络为基础的语法错误检测模型。从序列标注任务的性质,以及对图神经网络建模所具有的优势这两方面进行考虑,然后构建的语法错误检测框架以双向LSTM模型为基础,并在其中融合了使用门控图神经网络对错误语句的依存句法树建模的方法和预测上下文的多任务训练方法,使模型可以更好地理解语义和上下文关系。实验结果表明:GGNN模型在测试中,验证集上的错误检测准确率提高了19.9%,召回也提高了25.9%,测试集上的F值提升了30.1%。证明了基于图神经网络的语法错误检测模型在检测语法错误上具有可行性和有效性。
Aiming at the problem of grammar error detection in English translation error detection, a grammar error detection model based on graph neural network is constructed. Considering the nature of sequence tagging task and the advantages of graph neural network modeling, the syntax error detection framework is based on the bidirectional LSTM model, which integrates the method of modeling the dependent syntax tree of error statements using gated graph neural network and the multi task training method of predicting context, so that the model can better understand the semantics and context. The experimental results show that in the test of ggnn model, the error detection accuracy on the verification set is increased by 19.9%, the recall is also increased by 25.9%, and the F value on the test set is increased by 30.1%. It is proved that the grammar error detection model based on graph neural network is feasible and effective in detecting grammar errors.
作者
李洁莹
马佳瑛
LI Jieying;MA Jiaying(Xi’an Fanyi University,Xi’an 710105,China)
出处
《自动化与仪器仪表》
2023年第2期242-246,共5页
Automation & Instrumentation
基金
2015年陕西省社会科学艺术项目课题《陕西地方戏曲‘走出去’中的方言与术语英译研究》(2015SY059)
2020年陕西省高等教育学会项目《疫情背景下外语应急语言服务人才培养研究》(XGH20313)
《西安翻译学院“一带一路”语言与文化研究基地》(20KYJD02)。
关键词
语法错误检测
图神经网络
双向LSTM
依存句法树
syntax error detection
Fig.neural network
Bidirectional LSTM
Dependency syntax tree
