A computational framework for AI-driven reading comprehension rehabilitation for neurodiverse learners

Artificial intelligence is opening exciting new possibilities in education and speech and language therapy. It can offer meaningful support to students, especially those with special educational needs and disabilities who struggle with reading comprehension. Although some students demonstrate progress in word recognition, their comprehension often remains slow and effortful, impacting primary and secondary education achievement, but also personal and daily life. This Ph.D. thesis aims to enhance reading comprehension learning and rehabilitation by integrating psycholinguistic theory with computational tasks. The aim is to formalize comprehension processes, translate them into Natural Language Processing (NLP) tasks, and apply these tasks in practical educational and clinical settings. This work is structured around three main steps. First, the choice of computational methods is grounded in established psycholinguistic theories of language comprehension. Second, the chosen computational methods and evaluation framework are introduced and implemented, including sentence simplification, coherence classification, keyword extraction, text-to-pictogram translation, and concept map construction. This results in a toolbox of usable components that not only target distinct aspects of comprehension processes but also demonstrate singularly measurable improvements, particularly in concept map generation and text simplification. Finally, these technologies are combined into an online interface, developed with stakeholder feedback and evaluated through user studies in educational and clinical settings. The users' studies demonstrate the clinical validity of the Artificial Intelligence (AI)-driven framework for the rehabilitation of Special Educational Needs and Disabilities learners. In summary, this thesis enhances the theoretical, computational, and practical understanding of how AI can support reading comprehension for neurodiverse learners, while also contributing to novel methods for several NLP tasks, an integrated framework for AI-driven reading comprehension interventions, and its empirical validation.