Multi-Perspective Document Revision

Proceedings of the 29th International Conference on Computational Linguistics, pages 6128–6138 October 12–17, 2022. 6128 Multi-Perspective Document Revision Mana Ihori and Hiroshi Sato and Tomohiro Tanaka and Ryo Masumura NTT Computer and Data Science Laboratories, NTT Corporation 1-1 Hikarinooka, Yokosuka-Shi, Kanagawa 239-0847, Japan mana.ihori.kx@hco.ntt.co.jp Abstract This paper presents a novel multi-perspective document revision task. In conventional stud- ies on document revision, tasks such as gram- matical error correction, sentence reorder- ing, and discourse relation classification have been performed individually; however, these tasks simultaneously should be revised to im- prove the readability and clarity of a whole document. Thus, our study defines multi- perspective document revision as a task that simultaneously revises multiple perspectives. To model the task, we design a novel Japanese multi-perspective document revision dataset that simultaneously handles seven perspec- tives to improve the readability and clarity of a document. Although a large amount of data that simultaneously handles multiple per- spectives is needed to model multi-perspective document revision elaborately, it is difficult to prepare such a large amount of this data. Therefore, our study offers a multi-perspective document revision modeling method that can use a limited amount of matched data (i.e., data for the multi-perspective document revi- sion task) and external partially-matched data (e.g., data for the grammatical error correction task). Experiments using our created dataset demonstrate the effectiveness of using multiple partially-matched datasets to model the multi- perspective document revision task. 1 Introduction With the advance of natural language processing technology using deep learning, applications for writing support systems have been developed (Tsai et al., 2020; Ito et al., 2020). Such systems often implement a grammatical error correction task that corrects errors such as typos and mistakes in in- flected verb forms (Rothe et al., 2021). It is easy for the reader to understand an error-free docu- ment, and the lack of errors can allow for smooth text communication. In addition, it is crucial to revise a document automatically because it is dif- ficult to read one’s writing objectively, and it is time-consuming for a third party to revise the doc- ument. The document revision task has been studied in the natural language processing field by being broken down into partial tasks. Grammatical error correction tasks have been studied actively (Sawai et al., 2013; Mizumoto and Matsumoto, 2016; Junczys-Dowmunt and Grundkiewicz, 2016), and modeling with the sequence-to-sequence (seq2seq) model has achieved high performance with the ad- vance of deep learning (Yuan and Briscoe, 2016; Junczys-Dowmunt et al., 2018; Rothe et al., 2021). In addition, tasks considering the relationship be- tween sentences in a document are sentence order- ing (Yin et al., 2019; Wang and Wan, 2019) and discourse relation classification (Liu et al., 2016; Dai and Huang, 2018). These tasks have achieved high performance using deep learning, as with the grammatical error correction task. However, studies that simultaneously revise mul- tiple perspectives have not been well considered. To advance writing support, it is important not only to correct grammatical errors in a single sentence but also to …

Content truncated to 500 words