Catalan and Spanish Lexical Simplification and Complexity Prediction Dataset

This is the Spanish and the Catalan part of the MLSP dataset on Lexical Simplification and Lexical Complexity Prediction. They were used in the MLSP (Multilingual Lexical Simplification Pipeline) shared task, held at the BEA (Innovative use of NLP for Building Educational Applications). 

The target selection and data collection process of the datasets for Spanish and Catalan was largely parallel, but there were some differences due to the availability of source texts and annotators. The initial goal was to select 600 target words per language in 200 contexts, with 3 targets per context. An additional 10 contexts (and 30 words) were required for pilot annotations. Due to the sparseness of resources, we had to relax the goal for Catalan to 160 contexts. For each target a minimum of 10 annotations was required which were collected through on-line forms. The annotation process collected two pieces of data for each target word: i) a rating on Lexical Complexity on a 5-point Likert scale (from "very easy" to "very hard") and ii) up to 3 lexical substitutes for the target that fit in the given context. Annotators were asked to simply repeat the target word if they could not find a suitable alternative.

The Catalan dataset consists of 160 context sentences containing 475 target word tokens (454 distinct types). Sentences were selected from the Educational news section of the TeCla corpus4 (Armengol-Estapé et al., 2021) of news texts. The Spanish dataset consists of 625 target words in 210 contexts from texts on educational books on finance.

More details on the creation and compositoin of the datasets can be found in: 

Saggion, Horacio, et al. ‘Lexical Complexity Prediction and Lexical Simplification for Catalan and Spanish: Resource Creation, Quality Assessment, and Ethical Considerations’. Proceedings of the Third Workshop on Text Simplification, Accessibility and Readability (TSAR 2024), edited by Matthew Shardlow et al., Association for Computational Linguistics, 2024, pp. 82–94, https://aclanthology.org/2024.tsar-1.9.

The description of the BEA MLSP Shared task is described in: 

Shardlow, Matthew, et al. ‘The BEA 2024 Shared Task on the Multilingual Lexical Simplification Pipeline’. Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications (BEA), 2024.

Shardlow, Matthew, et al. ‘An Extensible Massively Multilingual Lexical Simplification Pipeline Dataset Using the MultiLS Framework’. Proceedings of the 3rd Workshop on Tools and Resources for People with Reading Difficulties (READI), 2024.