NeuroBridge simulates an AI conversation partner that communicates in a direct and literal manner, a style common among some autistic individuals. The system delivers scenario-specific feedback that helps neurotypical users interpret this communication style and choose responses that are clearer and more appropriate
NeuroBridge centers the simulated conversation around four common communication scenarios that may pose a challenge in mixed-neurotype settings:
- Indirect speech acts (implicit requests vs literal questions)
- Figurative expressions (idioms/metaphors interpreted literally vs interpreted figuratively)
- Emojis with variable interpretations (emojis with fluid meaning dependent on context)
- Being misperceived as blunt (direct statements read as rude vs honest)
Users enter basic info (e.g., name/pronouns) and a topic they are interested in. Using this information, NeuroBridge generates a personalized scenario and context for the conversation.
For each user-written message, NeuroBridge generates three semantically similar options that differ in tone/clarity/phrasing based on the given scenario. Users select which option to send.
After the character responds, NeuroBridge provides scenario-specific feedback:
- Constructive feedback explains the intent mismatch, identifies the most appropriate option, and why it’s best—then provides an empathic follow-up message the user can send.
- Positive feedback reinforces good choices and explains why other options could cause misunderstanding.
NeuroBridge was co-designed with an advisory board of autistic individuals and evaluated in an in-lab study with 12 neurotypical participants.
For frontend:
cd frontend
npm installCreate a .env file at the root of the frontend folder, and set the environment variable REACT_APP_API_HOST.
For backend, follow the instructions in api/README.md .
cd frontend
npm startNeuroBridge: Using Generative AI to Bridge Cross-neurotype Communication Differences with Neurotypical Perspective-taking
Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025)
Best Student Paper 🏆
PDF: https://rukhshan23.github.io/assets-2025.pdf
@inproceedings{haroon2025neurobridge,
author = {Haroon, Rukhshan and Wigdor, Kyle and Yang, Katie and Toumanios, Nicole and Crehan, Eileen T. and Dogar, Fahad},
title = {NeuroBridge: Using Generative AI to Bridge Cross-neurotype Communication Differences with Neurotypical Perspective-taking},
booktitle = {Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025)},
address = {Denver, CO, USA},
year = {2025},
url = {https://rukhshan23.github.io/assets-2025.pdf}
}