DivCon: Divergence and convergence in dialogue: The dynamic management of mismatches
Human interaction is deceptively simple to engage in, yet surprisingly challenging to account for theoretically, especially when we consider the full context of a conversation -- including gestures, facial expressions and intonation as well as the words used.
We never share the 'same' language as anyone else, due to differences in our experiences and social communities etc, which raises the question: How can we communicate successfully when individual differences in language use are the norm, not the exception? DivCon aims to address this question by creating a model of language as a skill, not an abstract system of signals, which we will test empirically using a data collection and experiments in text chat, zoom calls and virtual reality. The ultimate aim is that this model will provide the foundations for genuine domain general conversational AI.