SIMSUM: Document-level Text Simplification via Simultaneous Summarization

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 9927–9944 July 9-14, 2023 ©2023 Association for Computational Linguistics SIMSUM: Document-level Text Simplification via Simultaneous Summarization Sofia Blinova∗ EPFL sofia.blinova@epfl.ch Xinyu Zhou∗ EPFL xinyu.zhou@epfl.ch Martin Jaggi EPFL martin.jaggi@epfl.ch Carsten Eickhoff University of Tübingen carsten.eickhoff@uni-tuebingen.de Seyed Ali Bahrainian EPFL, University of Tübingen seyed-ali.bahreinian@uni-tuebingen.de Abstract Document-level text simplification is a spe- cific type of simplification which involves sim- plifying documents consisting of several sen- tences by rewriting them into fewer or more sentences. In this paper, we propose a new two-stage framework SIMSUM for automated document-level text simplification. Our model is designed with explicit summarization and simplification models and guides the genera- tion using the main keywords of a source text. In order to evaluate our new model, we use two existing benchmark datasets for simplifica- tion, namely D-Wikipedia and Wiki-Doc. We compare our model’s performance with state of the art and show that SIMSUM achieves top re- sults on the D-Wikipedia dataset SARI (+1.20), D-SARI (+1.64), and FKGL (-0.35) scores, im- proving over the best baseline models. In order to evaluate the quality of the generated text, we analyze the outputs from different models qual- itatively and demonstrate the merit of our new model. Our code and datasets are available 1. 1 Introduction Text simplification is an important technique that aims to simplify a document to make it more under- standable and accessible for people at different ed- ucation and reading levels while still retaining the content of the original text (Woodsend and Lapata, 2011). It concentrates on lexical simplification (i.e., using simpler vocabulary and including definitions that provide explanations in simple terms) as well as syntactic simplification (i.e., using less compli- cated sentence structures and grammar) (Saggion, 2017). Simplification is considered as a sequence- to-sequence text generation problem, closely re- sembling other NLP generation tasks such as text summarization (Dong et al., 2018; Cao et al., 2020; Miller, 2019) and paraphrasing (Zhao et al., 2018). ∗ Equal Contribution. 1https://github.com/epfml/easy-summary/tree/ main The applications of text simplification are broad. It can be an important tool for assisting children (Kajiwara et al., 2013) and non-native speakers (Glavaš and Štajner, 2015; Paetzold, 2016) to under- stand advanced texts with ease. Additionally, it is helpful for enabling people suffering from aphasia (Carroll et al., 1999), autism (Barbu et al., 2015), or dyslexia (Rello et al., 2013). Moreover, text simplification can be applied as a pre-processing step in other downstream NLP tasks such as Pars- ing (Chandrasekar et al., 1996), Information Ex- traction (Miwa et al., 2010), Text Summarization (Siddharthan et al., 2004) and Machine Translation (Štajner and Popovi´c, 2016). Two types of simplification can be defined based on the source text: sentence simplification and doc- ument simplification (Sun et al., 2021). Sentence simplification can be applied to texts with several sentences, one at a time, meaning that the num- ber of sentences in the input and output would be the same. Conversely, document simplification can reduce the number …

Content truncated to 500 words