Multi-Word Lexical Simplification

Proceedings of the 28th International Conference on Computational Linguistics, pages 1435–1446 Barcelona, Spain (Online), December 8-13, 2020 1435 Multi-Word Lexical Simplification Piotr Przybyła Institute of Computer Science, Polish Academy of Sciences Warsaw, Poland piotr.przybyla@ipipan.waw.pl Matthew Shardlow Department of Computing and Mathematics Manchester Metropolitan University Manchester, UK M.Shardlow@mmu.ac.uk Abstract In this work we propose the task of multi-word lexical simplification, in which a sentence in nat- ural language is made easier to understand by replacing its fragment with a simpler alternative, both of which can consist of many words. In order to explore this new direction, we contribute a corpus (MWLS1), including 1462 sentences in English from various sources with 7059 simplifi- cations provided by human annotators. We also propose an automatic solution (Plainifier) based on a purpose-trained neural language model and evaluate its performance, comparing to human and resource-based baselines. 1 Introduction Text simplification is the task of automatically modifying natural language text to improve its ease of understanding, whilst preserving the overall meaning. The diverse applications range from second lan- guage learners to lay readers of scientific texts to stroke victims. This challenge is often defined in the framework of lexical text simplification (LS), where individual words are replaced with their simpler equivalents (Paetzold and Specia, 2017b). However, single-word substitutions do not cover the full complexity of techniques humans use to approach text simplification, including replacements, deletions, addition and sentence splits. They could be modelled implicitly, by sentence-to-sentence monolingual translation approaches (Zhu et al., 2010; Zhang and Lapata, 2017), or explicitly, by collecting data (Alva-Manchego et al., 2020) and developing methods (Dong et al., 2019) focusing on particular types of transformations. With a similar motivation, we propose in this work a new task going beyond single word substitution, namely Multi-Word Lexical Simplification (MWLS). The aim of MWLS is to replace a given fragment (short sequence of words) with its simpler version, so that the enclosing sentence retains its meaning, but becomes easier to understand. See table 1 for several examples from our corpus. Note how these fragment-to-fragment replacements can involve words substitutions, expansions, deletions and other restructuring. This paper explores the task of MWLS in two directions. Firstly, we contribute a dataset of 1462 sentences with 7059 simplifications obtained through crowdsourcing. Secondly, we design a method for generating such simplifications automatically. Our solution, called the Plainifier, is inspired by a recently proposed method for LS utilising language models (Qiang et al., 2020), which we extend so that multi-word simplifications can be obtained. In order to encourage more research on the problem, we make the dataset1, the language model2 and the Plainifer code3 openly available. 2 Background Typical lexical simplification systems have followed a four stage pipeline of complex word identification (Shardlow, 2013), substitution generation, word sense disambiguation and synonym ranking (Paetzold This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http:// creativecommons.org/licenses/by/4.0/. 1https://github.com/piotrmp/mwls1 2https://github.com/piotrmp/tersebert 3https://github.com/piotrmp/plainifier 1436 and Specia, 2017a). However, in the past these systems have mostly focused on identifying and replacing single words …

Content truncated to 500 words