TemplateGEC: Improving Grammatical Error Correction with Detection Template

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6878–6892 July 9-14, 2023 ©2023 Association for Computational Linguistics TemplateGEC: Improving Grammatical Error Correction with Detection Template Yinghao Li1∗ Xuebo Liu2 Shuo Wang3 Peiyuan Gong1 Derek F. Wong4 Yang Gao1 Heyan Huang1 Min Zhang2 1School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China {yhli,pygong,gyang,hhy63}@bit.edu.cn 2Institute of Computing and Intelligence, Harbin Institute of Technology, Shenzhen, China {liuxuebo,zhangmin2021}@hit.edu.cn 3Tsinghua University, Beijing, China 4University of Macau, Macau, China wangshuo.thu@gmail.com, derekfw@um.edu.mo Abstract Grammatical error correction (GEC) can be divided into sequence-to-edit (Seq2Edit) and sequence-to-sequence (Seq2Seq) frameworks, both of which have their pros and cons. To utilize the strengths and make up for the short- comings of these frameworks, this paper pro- poses a novel method, TemplateGEC, which capitalizes on the capabilities of both Seq2Edit and Seq2Seq frameworks in error detection and correction respectively. TemplateGEC utilizes the detection labels from a Seq2Edit model, to construct the template as the input. A Seq2Seq model is employed to enforce consistency be- tween the predictions of different templates by utilizing consistency learning. Experimen- tal results on the Chinese NLPCC18, English BEA19 and CoNLL14 benchmarks show the effectiveness and robustness of TemplateGEC. Further analysis reveals the potential of our method in performing human-in-the-loop GEC. Source code and scripts are available at https: //github.com/li-aolong/TemplateGEC. 1 Introduction Grammatical error correction (GEC) is a fundamen- tal task in natural language processing that focuses on identifying and correcting grammatical errors in written text (Ng et al., 2013, 2014). The uti- lization of GEC is wide-ranging, including but not limited to, improving the quality of machine trans- lation (Popovi´c, 2018), increasing the readability of text (Liao et al., 2020), and assisting non-native speakers in language proficiency (Knill et al., 2019). There has been a significant amount of research in the field of GEC (Yuan and Briscoe, 2016; Bryant et al., 2017a; Ren et al., 2018; Zhou et al., 2018; Awasthi et al., 2019; Lai et al., 2022; Gong et al., 2022; Zhang et al., 2022b), which can be broadly ∗ Work done when Yinghao Li was a remote intern at Harbin Institute of Technology, Shenzhen. Detection Correction TemplateGEC He prefer study in home. He prefers to study at home. Figure 1: TemplateGEC takes the best of both worlds by utilizing the detection ability of the Seq2Edit framework and the correction ability of the Seq2Seq framework. classified into two categories: Sequence-to-Edit (Seq2Edit) and Sequence-to-Sequence (Seq2Seq). Seq2Edit GEC typically involves converting a source sentence into a sequence of editing opera- tions (Stahlberg and Kumar, 2020; Omelianchuk et al., 2020). Despite certain limitations, such as the manual selection of edits and the use of a dic- tionary (Awasthi et al., 2019; Malmi et al., 2019), Seq2Edit GEC have specific advantages for gram- matical error detection due to its high understand- ing ability (Omelianchuk et al., 2020). Seq2Seq GEC, on the other hand, which approaches GEC as a monolingual translation problem (Ge et al., 2018; Sun et al., …

Content truncated to 500 words