Elaborative Simplification: Content Addition and Explanation Generation in Text Simplification

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 5123–5137 August 1–6, 2021. ©2021 Association for Computational Linguistics 5123 Elaborative Simplification: Content Addition and Explanation Generation in Text Simplification Neha Srikanth Department of Computer Science The University of Texas at Austin nehasrik@utexas.edu Junyi Jessy Li Department of Linguistics The University of Texas at Austin jessy@austin.utexas.edu Abstract Much of modern-day text simplification re- search focuses on sentence-level simplifica- tion, transforming original, more complex sen- tences into simplified versions. However, adding content can often be useful when dif- ficult concepts and reasoning need to be ex- plained. In this work, we present the first data- driven study of content addition in text simpli- fication, which we call elaborative simplifica- tion. We introduce a new annotated dataset of 1.3K instances of elaborative simplification in the Newsela corpus, and analyze how entities, ideas, and concepts are elaborated through the lens of contextual specificity. We estab- lish baselines for elaboration generation using large-scale pre-trained language models, and demonstrate that considering contextual speci- ficity during generation can improve perfor- mance. Our results illustrate the complexities of elaborative simplification, suggesting many interesting directions for future work. 1 Introduction Text simplification aims to help audiences read and understand a piece of text through lexical, syn- tactic, and discourse modifications, while remain- ing faithful to its central idea and meaning (Sid- dharthan, 2014). It remains an important task, im- proving text accessibility for children (De Belder and Moens, 2010; Kajiwara et al., 2013), language learners (Yano et al., 1994; Petersen and Osten- dorf, 2007; Pellow and Eskenazi, 2014; Paetzold, 2016), and those with language impairments (Car- roll et al., 1998; Rello et al., 2013). Text simplifi- cation can also be a useful pre-processing step for other NLP tasks such as machine translation (Chen et al., 2012; ˇStajner and Popovic, 2016) and sum- marization (Vanderwende et al., 2007; Silveira and Branco, 2012). With the introduction of large, parallel cor- pora (Zhu et al., 2010; Woodsend and Lapata, Original Text Results, she said, “could help the team better un- derstand ancient Egyptian health” and, correspond- ingly, modern-day health. For instance, some mum- mies still have arteries in their mummified remains, Miller-Thomas said. And, sometimes, scientists can tell if those arteries had hardened. Simplified Text The scans could help the team understand about an- cient Egyptians’ health. For example, some mum- mies still have arteries. An artery is a tube that moves blood through the body. The artery could show if the person had been healthy or not. Figure 1: Elaborative simplification with two elabora- tions of varying contextual specificity. 2011; Coster and Kauchak, 2011; Xu et al., 2015), text simplification research has rapidly advanced in recent years, especially in sentence simplifi- cation (Alva-Manchego et al., 2020). However, document simplification involves rich linguistic phenomena that cannot be easily characterized by sentence-level transformations of text, e.g., the omission and addition of content (Petersen and Ostendorf, 2007; Siddharthan, 2014). This paper presents the first data-driven, dedi- cated study of elaborative simplification, which involves …

Content truncated to 500 words