Enhancing Aphasia Assessment Using Fine-Tuned Open-Source Large Language Models and Explainable AI
Brielle Stark
Faculty Mentor
Brielle Stark (College of Arts & Sciences)
Project Description
We want to see how well freely available AI programs that work with text can help speech‑language pathologists judge how clearly a person with aphasia retells a story—and whether we can make those AI programs easier to trust and understand. A student working on this project would start by gathering written versions of patients’ story retells, cleaning up the text, and marking which ideas count as the main points. Next, they would train several free AI programs—think of it as teaching the computer what “good” storytelling looks like—while carefully adjusting the settings to see which choices give the most accurate results. They would then compare these new, trained models to their off‑the‑shelf LLMs.
Technology or Computational Component
The student would be exposed to linguistic/language science as well as AI/LLM, and would work with a professor in speech, language and hearing and with a computer scientist. No prior experience necessary.