A New Dataset and Empirical Study for Sentence Simplification in Chinese

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 8306–8321 July 9-14, 2023 ©2023 Association for Computational Linguistics A New Dataset and Empirical Study for Sentence Simplification in Chinese Shiping Yang*† , Renliang Sun*, Xiaojun Wan Wangxuan Institute of Computer Technology, Peking University Center for Data Science, Peking University The MOE Key Laboratory of Computational Linguistics, Peking University yangshipingnlp@gmail.com sunrenliang@stu.pku.edu.cn wanxiaojun@pku.edu.cn Abstract Sentence Simplification is a valuable technique that can benefit language learners and children a lot. However, current research focuses more on English sentence simplification. The devel- opment of Chinese sentence simplification is relatively slow due to the lack of data. To alle- viate this limitation, this paper introduces CSS, a new dataset for assessing sentence simpli- fication in Chinese. We collect manual sim- plifications from human annotators and per- form data analysis to show the difference be- tween English and Chinese sentence simpli- fications. Furthermore, we test several unsu- pervised and zero/few-shot learning methods on CSS and analyze the automatic evaluation and human evaluation results. In the end, we explore whether Large Language Models can serve as high-quality Chinese sentence simpli- fication systems by evaluating them on CSS. 1 Introduction Sentence Simplification (SS) is the task of mod- ifying a sentence to make it easier to understand and improve its accessibility for a wider audience, while retaining most of its original meaning (Alva- Manchego et al., 2020b). Automatic SS systems provide reading assistance to children (De Belder and Moens, 2010; Kajiwara et al., 2013), non- native readers (Paetzold and Specia, 2016b), and people with reading disabilities (Carroll et al., 1998; Rello et al., 2013; Evans et al., 2014). Currently, there are multiple datasets for English sentence simplification to choose from, such as WikiSmall (Zhu et al., 2010), WikiLarge (Zhang and Lapata, 2017) and Newsela (Xu et al., 2015). However, few simplification-specific datasets are available for other languages. Research on other popular languages like Spanish (Štajner et al., 2015; Saggion et al., 2015), French (Gala et al., 2020), * Equal contribution † Work done during internship Russian (Sakhovskiy et al., 2021; Dmitrieva et al., 2021) and Italian (Brunato et al., 2015; Tonelli et al., 2016) has gained momentum in recent years. Unfortunately, research on Chinese sentence sim- plification is still scarce: to the best of our knowl- edge, there is currently no publicly available simpli- fication corpora for training, even lacking a dataset to evaluate the ability of simplification models. These limitations greatly prevent the development of Chinese sentence simplification. Creating a parallel dataset for Chinese sentence simplification poses considerable challenges. Most datasets for SS use the automatic sentence align- ment method to construct (Zhang and Lapata, 2017; Dmitrieva et al., 2021), which relies on the exis- tence of large-scale simplification corpus like Sim- ple English Wikipedia1. However, there are no suitable data sources in Chinese. Another possible option is to translate the English dataset to Chinese through neural machine translation (Sakhovskiy et al., 2021; Li et al., 2022). Nevertheless, Chi- …

Content truncated to 500 words