Proceedings of the Annual Meeting of the Cognitive Science Society

Humans align to one another in conversation—adopting shared conventions that ease communication. We test whether LLMs form the same kinds of conventions in a multimodal communication game. Both humans and LLMs displayed evidence of convention-formation (increasing the accuracy and consistency of their turns) when communicating in same-type dyads (humans with humans, AI with AI), though AI-AI pairs show limited reduction in length. Heterogeneous human-AI pairs failed to converge on effective referring expressions, suggesting differences in communicative tendencies. In Experiment 2, we prompting LLMs to produce superficially humanlike behavior. While the length of prompted models' messages matched that of human pairs, accuracy and lexical overlap in human-AI pairs continued to lag behind that of both human-human and AI-AI pairs. These results suggest that conversational alignment requires more than just the ability to mimic previous interactions, but also shared interpretative biases toward the meanings that are conveyed.