Practices in Dialogue

We explore the idea that conversational episodes not only ground facts, but also establish practices (know how). We apply this idea to the well-studied phenomenon of lexical entrainment. We provide a formal model of situations giving rise to lexical entrainment, and use it to make precise implications of extant attempts to explain this phenomenon (the lexical pact model, and the interactive alignment model) and to make precise where our attempt differs: it is an automatic and non-strategic, but goal-driven account, which has a place for partner-specificity and group forming. We define a learning mechanism for practices, and test it in simulation. We close with a discussion of further implications of the model and possible extensions.