Celso Camilo-Junior
2024
Evaluating the Simplification of Brazilian Legal Rulings in LLMs Using Readability Scores as a Target
Antonio Flavio Paula
|
Celso Camilo-Junior
Proceedings of the Third Workshop on Text Simplification, Accessibility and Readability (TSAR 2024)
Legal documents are often characterized by complex language, including jargon and technical terms, making them challenging for Natural Language Processing (NLP) applications. We apply the readability-controlled text modification task with an emphasis on legal texts simplification. Additionally, our work explores an evaluation based on the comparison of word complexity in the documents using Zipf scale, demonstrating the models’ ability to simplify text according to the target readability scores, while also identifying a limit to this capability. Our results with Llama-3 and Sabiá-2 show that while the complexity score decreases with higher readability targets, there is a trade-off with reduced semantic similarity.