Embodied agents must perform reference resolution if they are to achieve sufficient language understanding with humans. But sit- uated interaction introduces social norms, which are often overlooked yet critically need to be reasoned together with language to resolve ref- erences. To address this issue, we offer a novel normative-based rea- soning approach to reference resolution and provide a proof-of-concept implementation in a cognitive robotic architecture with natural lan- guage human-robot interaction capabilities. We discuss reference res- olution problems that require different levels of normative reasoning, demonstrate how a large language model, GPT-3, struggles to consis- tently identify target referents when normative reasoning is needed, pro- vide a user study to show how humans perform norm-guided reference resolution, and demonstrate the successful operation of our proposed ar- chitecture on a fully autonomous assistive robot interacting with human instructors in natural language.
@inproceedings{abramsetal24icsr,
title={Robots That Perform Norm-Based Reference Resolution},
author={Mitchell Abrams and Christopher Thierauf and Matthias Scheutz},
year={2024},
booktitle={Proceedings of ICSR},
url={https://hrilab.tufts.edu/publications/abramsetal24icsr.pdf}
}