Towards Inclusive Reading: A Neural Text Generation Framework for Dyslexia Accessibility
Elham Madjidi, Christopher Crick · 2024
This paper presents a novel approach to generating dyslexia-friendly text using neural text generation techniques.We propose a framework that leverages transformer-based language models, specifically GPT and T5, and incorporates syllable and morphological analysis to enhance the readability and comprehension of text for dyslexic readers.Our approach involves fine-tuning the language models on a curated dataset of dyslexia-friendly text, validated through human assessments and feedback from individuals with dyslexia.We conduct a two-phase experiment with 14 undergraduate students with dyslexia to evaluate the effectiveness of our generated text.The results demonstrate improvements in reading time for participants presented with the refined dyslexia-friendly passages, while also highlighting the importance of individual preferences and text engagement.Furthermore, we provide insights into the specific challenges faced by dyslexic readers and propose targeted approaches to address these issues.This research contributes to the advancement of text accessibility by automating the process of converting standard text into dyslexiafriendly formats.The insights gained from this study inform the design of dyslexia-friendly materials and emphasize the importance of a holistic approach to text accessibility.Our framework has the potential to increase the availability of dyslexia-friendly content and support individuals with dyslexia in accessing written information.