SWiPE: A Dataset for Document-Level Simplification of Wikipedia Pages

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10674–10695 July 9-14, 2023 ©2023 Association for Computational Linguistics SWIPE: A Dataset for Document-Level Simplification of Wikipedia Pages Philippe Laban Jesse Vig Wojciech Kryscinski Shafiq Joty Caiming Xiong Salesforce AIChien-Sheng Jason Wu {plaban, jvig, wojciech.kryscinski, sjoty, cxiong, wu.jason}@salesforce.com Abstract Text simplification research has mostly focused on sentence-level simplification, even though many desirable edits—such as adding relevant background information or reordering content— may require document-level context. Prior work has also predominantly framed simplifica- tion as a single-step, input-to-output task, only implicitly modeling the fine-grained, span-level edits that elucidate the simplification process. To address both gaps, we introduce the SWIPE dataset, which reconstructs the document-level editing process from English Wikipedia (EW) articles to paired Simple Wikipedia (SEW) ar- ticles. In contrast to prior work, SWIPE lever- ages the entire revision history when pairing pages in order to better identify simplification edits. We work with Wikipedia editors to anno- tate 5,000 EW-SEW document pairs, labeling more than 40,000 edits with proposed 19 cate- gories. To scale our efforts, we propose several models to automatically label edits, achieving an F-1 score of up to 70.6, indicating that this is a tractable but challenging NLU task. Finally, we categorize the edits produced by several sim- plification models and find that SWIPE-trained models generate more complex edits while re- ducing unwanted edits. 1 Introduction Text simplification (TS) aims to make complex documents accessible to larger audiences by lower- ing the barrier of reading for children, non-native speakers, and novice readers in technical domains. TS has primarily been approached in a sentence- level sequence-to-sequence (seq2seq) manner, fol- lowing the methodology of mature NLG tasks such as machine translation. Prior work framed at the sentence level has focused on simplification edits that occur within sentence units, such as lexical replacements (Glavaš and Štajner, 2015) and sen- tence splitting (Narayan and Gardent, 2015; Sulem et al., 2018). Yet, many simplification operations, such as background elaborations (Srikanth and Li, Tchaikovsky, Mussorgsky, and Rimsky-Korsakov received their premieres. Since Yuri Temirkanov's retirement in 1988, the The conductor Valery Gergiev has been the served as The Mariinsky Theatre is a very famous historic theatre of opera and ballet in Saint Petersburg, Russia. It was opened Opened in 1860 and soon, it became the most important preeminent music theatre for opera and ballet in Russia. Many great operas and ballets of the late 19th century were first performed there Russia, including where many of the greatest stage works by masterpieces of Lexical Syntactic Edit Lexical Elaboration Semantic Deletion Discourse Reordering Lexical Specific-to-general Lexical Lexical Insertion Only in Simple Eng. Wiki Page Deletion Only in English Wiki Page Unchanged In both Wiki Pages Annotated Edit Group Mariinsky Theater - Wikipedia Page Lexical Legend its general director there since 1988. Figure 1: Sample from SWIPE, a Wikipedia-based dataset for document-level simplification. Many edits in SWIPE require document-level context. 2020) or content reordering (Zhong et al., 2020) require document-level context. …

Content truncated to 500 words