Align and Shine: Building High-Quality Sentence-Aligned Corpora for Multilingual Text Simplification

Researchers have released a new study titled "Align and Shine," addressing the scarcity of high-quality datasets for text simplification in languages other than English. Text simplification is vital for enhancing accessibility for language learners and individuals with limited literacy. The paper details an experimental methodology for collecting and processing crowd-sourced simplification data from comparable corpora. The primary contribution is the construction of a robust corpus suitable for both training and evaluating text simplification systems across five languages: Catalan, English, French, Italian, and Spanish. A key technical achievement reported is the development of mechanisms for accurate sentence-level alignment derived from document-level data. This innovation allows for more precise model training and testing. The resulting dataset of aligned sentence pairs has been made publicly available to support further research and development in natural language processing. By providing these resources, the authors aim to bridge the gap in multilingual text simplification tools, promoting greater inclusivity and comprehensibility of written information for diverse global audiences.