Evaluación de modelos de inteligencia artificial en simplificación automática de textos en español para personas con discapacidades visuales.

The present proposal aims to evaluate the efficacy of artificial intelligence models in automatic text simplification. We will construct datasets comprising simple and complex sentences for training and evaluating the models. We will apply fine-tuning and transfer learning techniques to pre-trained models, such as Llama 2, Llama 3, and Vicuna, to optimize their performance in text simplification. Additionally, we will develop new evaluation metrics that incorporate lexical and syntactic aspects specific to Spanish and relevant to the target population. Expert judges will validate the results through empirical analyses, assessing the readability and comprehensibility of the generated texts.

This research aims to bridge the existing gap in the accessibility of Spanish texts, thereby benefiting vulnerable populations and promoting greater inclusion in access to knowledge. The project addresses the needs identified in previous studies on inclusion and disability. Furthermore, this work will advance the state of the art in automatic text simplification and natural language processing

The project's results will facilitate the creation of tools and resources that educators, researchers, and professionals in the field of accessibility can use. This effort will enhance the educational and professional inclusion of visually impaired individuals. The project's outputs will include accessible datasets for the scientific community, scientific articles, optimized algorithms, and new validation instruments

Evaluar modelos de inteligencia artificial en simplificación automática de textos en
español.

1. Empleo de las tecnologías para el análisis del discurso, el tratamiento, la
enseñanza y el aprendizaje de lenguas.
2. Prácticas comunicativas en la lengua materna y otros idiomas