基于SCOPE的英文语法纠错技术研究  

Research on English grammar error correction technology based on SCOPE

作　　者：蒋旭 努尔麦麦提·尤鲁瓦斯[1] JIANG Xu;NURMEMET Yolwas(School of Computer Science and technology,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学计算机与科学技术学院,乌鲁木齐830046

出　　处：《中央民族大学学报(自然科学版)》2025年第1期55-63,共9页Journal of Minzu University of China(Natural Sciences Edition)

摘　　要：文本校对是一项基本的自然语言处理任务,用于检测和纠正文本中存在的语法错误。语法错误中的部分拼写错误是由发音相似引起的。本文研究了发音预测模型,提出了基于SCOPE的英文文本校对模型。模型在共享编码器的基础上设置了两个并行的解码器,一个用于语法错误纠正任务,另一个用于预测细粒度的英文发音任务,并引入了自适应加权方案以平衡这两个任务。模型在CoNLL-2014数据集中的精度、召回和F0.5分数分别为69.1、42.3和61.3,在F0.5分数上相对基线模型提高了1.5%。实验结果表明,通过预测英文单词的细粒度发音(元音、辅音和轻重音),有效提升了模型拼写纠错的准确率,为中、西亚语言的拼写纠错研究提供了借鉴。Text proofreading is a basic natural language processing task for detecting and correcting grammatical errors present in text.A part of spelling errors among grammatical errors is caused by pronunciation similarity.In this paper,the pronunciation prediction model is investigated to address the above problem and a SCOPE⁃based English text proofreading model is proposed.The model pro⁃posed in this paper sets two parallel decoders on top of the shared encoder,one for the task of gram⁃matical error correction and one for the task of predicting fine⁃grained English pronunciation,and in⁃troduces an adaptive weighting scheme to balance these two tasks.The precision,recall,and F0.5 scores of the model proposed in this paper in the CoNLL-2014 spelling error test data are 69.1,42.3,and 61.3,respectively,which are relatively higher than the baseline model in terms of F0.5 score by 1.5 percentage points.The analysis of the experimental results shows that the model pro⁃posed in this paper improves the accuracy of the model's spelling error correction by predicting the fine⁃grained pronunciation of English words,i.e.,as vowels,consonants,and light and heavy sounds.The method provides ideas for subsequent research on spelling error correction in Central and West Asian languages.

关 键 词：语法错误纠正 英文发音预测 拼写错误纠正 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]