Relevance is a fundamental feature of human cognition: when we talk, we don’t produce random utterances; rather, we carefully pick our utterances so as to convey relevant information. But what makes a piece of information relevant? Formal pragmatics has an answer to this question: roughly, a proposition is relevant iff it eliminates at least some of the possibilities under consideration at a given time (Groenendijk and Stokhof 1984; Roberts  2012).
In this talk, I do three things. First, I draw attention to a limitation of this approach: it is not able to account for the fact that sometimes a proposition is perceived to be relevant despite eliminating no live possibility. Second, I discuss recent work on relevance (including my own work) that tries to overcome this problem by using probabilities and information-theoretic notions (e.g. Kullback-Leibler Divergence).
I end the talk with a cautionary note: current accounts of relevance might not be accounts of relevance at all.