Exploring AI Tools for Accessible Scientific Documents

Authors: Christopher Power & Eleazar Vasquez III

https://doi.org/10.51388/20.500.12265/326

Abstract: This field note addresses the critical challenge of accessing mathematical content in PDF documents for students using screenreader technology. Standard PDF exports often discard structural tags and fail to convey the complex meaning of mathematical notation, hindering full student participation in learning activities. To overcome these barriers, this exploratory project investigates improved document-production workflows that preserve mathematical structure from source code to final output. Utilizing Large Language Models (LLMs) like Anthropic’s Claude Code, the project prototyped methods for embedding structural information within LaTeX source code to generate inherently accessible PDFs and HTML documents featuring MathML. Furthermore, the note outlines a comprehensive future evaluation framework involving subject experts, accessibility specialists, and skilled screenreader users to rigorously validate these prototypes against real-world educational tasks. Ultimately, by both human expertise and AI, this initiative aims to develop an authoring toolchain for educators. This approach ensures mathematical documents are accessible upon creation, thereby enabling independent and more equitable participation for all students in mathematics education.

Suggested Citation: Powers, C. & Vasquez, E. (2026). Exploring AI Tools for Accessible Scientific Documents. ATS Hub Field Note Series. https://doi.org/10.51388/20.500.12265/326

This material was developed under The Institute of Education Sciences Award R305N250006 from the U.S. Department of Education. Any opinions, findings, conclusions, or recommendations are those of the authors and do not necessarily reflect the views of IES.