![[2509.24597] Inducing Dyslexia in Vision Language Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2509.24597] Inducing Dyslexia in Vision Language Models
Summary
The paper explores how vision-language models can simulate dyslexia by disrupting word processing mechanisms, providing insights into reading impairments and their neural correlates.
Why It Matters
This research is significant as it leverages advanced AI models to investigate dyslexia, a condition that affects many individuals. By simulating dyslexia in vision-language models, the study offers a novel approach to understanding the underlying mechanisms of reading difficulties, which could inform future interventions and educational strategies.
Key Takeaways
- The study uses vision-language models to simulate dyslexia.
- Disruption of visual-word-form-selective units leads to reading impairments.
- The model replicates key characteristics of dyslexia, including phonological deficits.
- Findings provide a computational framework for investigating brain disorders.
- This approach may enhance understanding of reading difficulties and inform future research.
Computer Science > Computation and Language arXiv:2509.24597 (cs) [Submitted on 29 Sep 2025 (v1), last revised 26 Feb 2026 (this version, v3)] Title:Inducing Dyslexia in Vision Language Models Authors:Melika Honarmand, Ayati Sharma, Badr AlKhamissi, Johannes Mehrer, Martin Schrimpf View a PDF of the paper titled Inducing Dyslexia in Vision Language Models, by Melika Honarmand and 4 other authors View PDF HTML (experimental) Abstract:Dyslexia, a neurodevelopmental disorder characterized by persistent reading difficulties, is often linked to reduced activity of the visual word form area (VWFA) in the ventral occipito-temporal cortex. Traditional approaches to studying dyslexia, such as behavioral and neuroimaging methods, have provided valuable insights but remain limited in their ability to test causal hypotheses about the underlying mechanisms of reading impairments. In this study, we use large-scale vision-language models (VLMs) to simulate dyslexia by functionally identifying and perturbing artificial analogues of word processing. Using stimuli from cognitive neuroscience, we identify visual-word-form-selective units within VLMs and demonstrate that they predict human VWFA neural responses. Ablating model VWF units leads to selective impairments in reading tasks while general visual and language comprehension abilities remain intact. In particular, the resulting model matches dyslexic humans' phonological deficits without a significant change in orthographic processing, ...
