Translating Easy-to-Read Standards into Prompts: An Empirical Study of LLM-Based Text Simplification Strategies

Abstract

Text simplification remains a significant challenge, especially when texts must comply with Easy-to-Read (E2R) standards. This paper explores how normative E2R guidelines can be translated into prompt configurations for Large Language Models (LLMs) to support text simplification. We compare six prompting strategies, ranging from a baseline prompt without explicitly including the UNE guidelines to one-shot and few-shot strategies. Using Gemma 3 4B, we evaluate the generated simplifications on a mixed corpus of human-written and synthetic texts, through automatic linguistic complexity metrics, and human evaluation of guideline compliance and content preservation. Results indicate that prompting strategies that explicitly incorporate the guidelines, particularly those including examples, improve compliance with UNE-based E2R requirements compared to baseline configurations. Overall, this study provides initial evidence of the potential of LLMs as a support tool for generating E2R adapted content, and also highlights the importance of future evaluation with end users.