LeSS: A Computationally-Light Lexical Simplifier for Spanish

Proceedings of Recent Advances in Natural Language Processing, pages 1132–1142 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_120 1132 LeSS: A Computationally-Light Lexical Simplifier for Spanish Sanja ˇStajner1, Daniel Ib´a˜nez2, Horacio Saggion3 1Karlsruhe, Germany, stajner.sanja@gmail.com 2Sevilla, Spain, danielibanezgarcia@gmail.com 3LaSTUS Lab / TALN Group, Universitat Pompeu Fabra, Spain, horacio.saggion@upf.edu Abstract Due to having knowledge of only basic vocabu- lary, many people cannot understand up-to-date written information and thus make informed decisions and fully participate in the society. We propose LeSS, a modular lexical simplifi- cation architecture that outperforms state-of- the-art lexical simplification systems for Span- ish. In addition to its state-of-the-art perfor- mance, LeSS is computationally light, using much less disk space, CPU and GPU, and hav- ing faster loading and execution time than the transformer-based lexical simplification mod- els which are predominant in the field. 1 Introduction Even in the highly-developed countries, many peo- ple (16.7% on average) only have a knowledge of a basic vocabulary thus encountering difficulties in understanding written information on a daily basis (OECD, 2013). This limits their active partic- ipation in the society and can negatively influence their life choices. According to the Adult Literacy Report from 2013 (OECD, 2013), this problem is particularly prominent in Spain, where 28.3% of people are in this situation (ˇStajner, 2021). Lexical simplification (LS) is the process of sub- stituting complex words or phrases with their sim- pler variants. It is an important factor in making texts more accessible for people with aphasia (Car- roll et al., 1998; Devlin and Unthank, 2006; Devlin and Tait, 1998), dyslexia (Rello et al., 2013b,a), autism spectrum disorders (Or˘asan et al., 2018), cognitive impairments (Feng et al., 2009; Saggion et al., 2015), low literacy levels (Alu´ısio et al., 2008; Watanabe et al., 2010), deaf and hard-of- hearing people (Inui et al., 2003; Alonzo et al., 2020), children (De Belder et al., 2010), and non- native speakers (Hirsh and Nation, 1992; Heilman et al., 2007). Due to its evident potential for great social im- pact, lexical simplification has been attracting a growing attention from the natural language pro- cessing (NLP) community (Paetzold and Specia, 2017; ˇStajner, 2021). SemEval-2021 Task 1 on Lexical Complexity Prediction attracted 198 teams, out of which 91 submitted their systems for one of the two sub-tasks, single-word or multi-word lexi- cal complexity prediction (Shardlow et al., 2021). The recent TSAR-2022 shared task on Multilin- gual Lexical Simplification received 33 system sub- missions for English, 17 for Spanish, and 16 for (Brazilian) Portuguese (Saggion et al., 2022). While it is generally considered that the more frequent words are easier to understand for every- one, which words should be considered as com- plex or simple can vary from one target population to another (Yimam et al., 2017), and is subjec- tive even within one target group (Yimam et al., 2018). Manual lexical simplification requires an extensive knowledge of the language and the par- ticular simplification needs of the target user, thus being expensive and time-consuming. Automatic text simplification systems, in contrast, could of- fer a …

Content truncated to 500 words