[1]许婉秋,曲维光,魏庭新,等.基于类型驱动及模型融合的中文语法纠错研究[J].南京师大学报(自然科学版),2025,48(03):139-148.[doi:10.3969/j.issn.1001-4616.2025.03.016]
　Xu Wanqiu,Qu Weiguang,Wei Tingxin,et al.Research on Chinese Grammar Correction Based on Type-Driven and Model Fusion[J].Journal of Nanjing Normal University(Natural Science Edition),2025,48(03):139-148.[doi:10.3969/j.issn.1001-4616.2025.03.016]
点击复制

基于类型驱动及模型融合的中文语法纠错研究()

《南京师大学报（自然科学版）》[ISSN:1001-4616/CN:32-1239/N]

卷:
48
期数:
2025年03期
页码:
139-148
栏目:
计算机科学与技术
出版日期:
2025-06-20

文章信息/Info

Title:
Research on Chinese Grammar Correction Based on Type-Driven and Model Fusion
文章编号:
1001-4616(2025)03-0139-10
作者:
许婉秋1; 2; 曲维光1; 3; 魏庭新4; 谷静平3; 顾彦慧1; 周俊生1
(1.南京师范大学计算机与电子信息学院/人工智能学院,江苏 南京 210023)
(2.江苏海事职业技术学院信息工程学院,江苏 南京 211170)
(3.南京师范大学中北学院,江苏 丹阳 212334)
(4.南京师范大学国际文化教育学院,江苏 南京 210097)
Author(s):
Xu Wanqiu1; 2; Qu Weiguang1; 3; Wei Tingxin4; Gu Jingping3; Gu Yanhui1; Zhou Junsheng1
(1.School of Computer and Electronic Information/Artificial Intelligence,Nanjing Normal University,Nanjing 210023,China)
(2.School of Information Technology,Jiangsu Maritime Institute,Nanjing 211170,China)
(3.Nanjing Normal University Zhongbei College,Danyang 212334,China)
(4.International College for Chinese Studies,Nanjing Normal University,Nanjing 210097,China)
关键词:
中文语法纠错; 类型依赖关系; 两阶段训练; 大规模语言模型; 模型融合
Keywords:
Chinese grammar error correction; type dependency relationship; two-stage training; large language model; model fusion
分类号:
TP391
DOI:
10.3969/j.issn.1001-4616.2025.03.016
文献标志码:
A
摘要:
中文语法纠错旨在通过模型自动识别并修正中文文本中的语法错误,从而提升文本的准确性和可读性. 然而,现有的中文语法纠错模型在纠错过程中常面临暴露偏差问题,并且对大模型的应用仍显不足,导致纠错效果欠佳. 为此,本文提出了一种基于类型驱动的中文语法纠错模型CTDGC(Chinese Types Driven Grammatical Correction). 该模型通过深入探讨中文四种主要语法错误(冗余、缺失、错词、乱序)之间的依赖关系,设计了两阶段训练策略,有效缓解了训练与预测的不匹配问题,在CGED2020数据集上单模型F0.5达到34.18%,优于以往的方法. 此外,本文还提出了一种基于ChatGLM的中文语法纠错模型CorGLM(Chinese Grammatical Correction Model based on ChatGLM),并对Baichuan大模型设计了特定的Prompt. 通过与CTDGC等模型的融合,F0.5显著提升至40.33%,验证了本文方法的有效性和优越性.
Abstract:
Chinese grammar error correction aims to automatically identify and rectify grammatical errors in Chinese text using models,thereby improving the accuracy and readability of the text. However,existing Chinese grammar correction models often face exposure bias issues during correction,and their application of large models remains inadequate,resulting in suboptimal correction performance. To address this,this paper proposes a Chinese Types Driven Grammatical Correction(CTDGC)model. By thoroughly exploring the dependency relationships among four major types of Chinese grammatical errors(redundancy,omission,incorrect word usage,and word order errors),the model employs a two-stage training strategy that effectively alleviates the mismatch between training and prediction. The CTDGC model achieves an F0.5 score of 34.18% on the CGED2020 dataset,outperforming previous methods. Additionally,this paper introduces a Chinese grammar correction model based on ChatGLM(CorGLM)and designs specific prompts for the Baichuan large model. Through integration with models like CTDGC,the F0.5 score significantly improves to 40.33%,demonstrating the effectiveness and superiority of the proposed approach.

参考文献/References:

[1]RAO G,ZHANG B,XUN E,et al. IJCNLP-2017 task 1:Chinese grammatical error diagnosis[C]//Proceedings of the IJCNLP 2017,Shared Tasks. Taipei,China:Asian Federation of Natural Language Processing,2017:1-8.
[2]LAI S,ZHOU Q,ZENG J,et al. Type-driven multi-turn corrections for grammatical error correction[C]//Findings of the Association for Computational Linguistics. Dublin,Ireland:Association for Computational Linguistics,2022:3225-3236.
[3]FU K,HUANG J,DUAN Y. Youdao's winning solution to the NLPCC-2018 task 2 challenge:a neural machine translation approach to Chinese grammatical error correction[C]//Proceedings of NLPCC-2018. Hihhot,China:Springer,Cham,2018:341-350.
[4]REN H,YANG L,XUN E. A sequence to sequence learning for Chinese grammatical error correction[C]//Natural Language Processing and Chinese Computing. Hihhot,China:Springer,Cham,2018:401-410.
[5]WANG H,KUROSAWA M,KATSUMATA S,et al. Chinese grammatical correction using BERT-based pre-trained model[C]//Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing. Suzhou,China:Association for Computational Linguistics,2020:163-168.
[6]LI P,SHI S. Tail-to-Tail non-autoregressive sequence prediction for Chinese grammatical error correction[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing. Online:Association for Computational Linguistics,2021:4973-4984.
[7]WU X,WU Y. From spelling to grammar:a new framework for Chinese grammatical error correction[C]//Conference on Empirical Methods in Natural Language Processing. Abu Dhabi,United Arab Emirates:Association for Computational Linguistics,2022:889-902.
[8]CHEN P,WU S,CHEN L,et al. Chinese grammatical error diagnosis by conditional random fields[C]//Proceedings of the 2nd Workshop on Natural Language Processing Techniques for Educational Applications. Beijing,China:Association for Computational Linguistics,2015:7-14.
[9]OMELIANCHUK K,ATRASEVYCH V,CHERNODUB A,et al. GECToR-grammatical error correction:tag,not rewrite[C]//Proceedings of the Fifteenth Workshop on Innovative Use of NLP for Building Educational Applications. Seattle,WA,USA:Association for Computational Linguistics,2020:163-170.
[10]LIANG D,ZHENG C,GUO L,et al. BERT enhanced neural machine translation and sequence tagging model for Chinese grammatical error diagnosis[C]//Proceedings of the 6th Workshop on Natural Language Processing Techniques for Educational Applications. Suzhou,China:Association for Computational Linguistics,2020:57-66.
[11]LI J,GUO J,ZHU Y,et al. Sequence-to-Action:grammatical error correction with action guided sequence generation[C]//AAAI Conference on Artificial Intelligence. Online:AAAI press,2022:10974-10982.
[12]ZHANG Y,LI Z,BAO Z,et al. MuCGEC:a multi-reference multi-source evaluation dataset for Chinese grammatical error correction[C]//Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies. Seattle,United States:Association for Computational Linguistics,2022:3118-3130.
[13]TANG C,WU X,WU Y. Are pre-trained language models useful for model ensemble in Chinese grammatical error correction[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. New York,USA:AAAI Press,2023:893-901.
[14]KANEKO M,MITA M,KIYONO S,et al. Encoder-Decoder models can benefit from pre-trained masked language models in grammatical error correction[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Toronto,Canada:Association for Computational Linguistics,2020:4248-4254.
[15]ZHAO Z,WANG H. MaskGEC:improving neural grammatical error correction via dynamic masking[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Toronto,Canada:Association for Computational Linguistics. Online:Association for Computational Linguistics,2020,34(1):1226-1233.

备注/Memo

备注/Memo:
收稿日期:2024-06-26.
基金项目:国家社会科学基金重大项目(21&ZD288)、江苏省教育厅哲社项目(2024SJYB1630).
通讯作者:曲维光,博士,教授,研究方向:自然语言处理,计算语言学、语言工程、人工智能. E-mail:wgqu_nj@163.com
更新日期/Last Update: 2025-06-20