Let AI Read First: Enhancing Reading Abilities for Individuals with Dyslexia through Artificial Intelligence

TL;DR

This work introduces LARF, an AI-driven text annotation approach that preserves original content while highlighting key information to aid readers with dyslexia. Using GPT-4 to generate HTML-based annotations, the authors compare LARF against a control and Bionic Reading in a large randomized study (N=150) and measure recall, retrieval, comprehension, and subjective usability. Results show that LARF improves retrieval and comprehension, with the strongest benefits among participants with severe dyslexia, and it receives more favorable usability ratings than conventional tools. The study discusses practical deployment considerations, privacy, and future directions, suggesting LARF could extend to exams, websites, and other reading contexts to enhance accessibility.

Abstract

Dyslexia, a neurological condition affecting approximately 12% of the global population, presents significant challenges to reading ability and quality of life. Existing assistive technologies are limited by factors such as unsuitability for quiet environments, high costs, and the risk of distorting meaning or failing to provide real-time support. To address these issues, we introduce LARF (Let AI Read First), the first strategy that employs large language models to annotate text and enhance readability while preserving the original content. We evaluated LARF in a large-scale between-subjects experiment, involving 150 participants with dyslexia. The results show that LARF significantly improves reading performance and experience for individuals with dyslexia. Results also prove that LARF is particularly helpful for participants with more severe reading difficulties. Furthermore, this work discusses potential research directions opened up by LARF for the HCI community.

Figures (9)

• Figure 1: Experiment Procedure. Participants are randomly assigned to three conditions, and then they are asked to finish the reading session. They are required to read the same article but with different presentations. Participants in the conventional condition group and LARF condition group are required to finish a subjective evaluation session after they finish the reading session.
• Figure 2: (a) shows the differences in reading time under three different conditions. Though the pattern is not significant, we can observe that users in the LARF group do less "glance over and skip the article." Furthermore, their overall reading time is more concentrated in areas with shorter durations. Subfigure (b) and (c) respectively represent the scores of users in the retrieve and recall phases. It can be observed that compared to other groups, participants reading the LARF-marked texts exhibit better recall ability (marginally significant) and a superior capability to remember the details of the articles (significant).
• Figure 3: The demo of the custom mode of LARF software application on PC.
• Figure 4: An example of the result and parameters of Bionic Reading
• Figure 5: The subjective evaluation result. Participants with dyslexia exhibited a clear preference for LARF, considering text annotated with LARF to be effective, user-friendly, and worthy of broader adoption in various contexts.