This dissertation sketches a theory of expected conversational roles. An expected conversational role (ECR) is the trajectory you expect a person will take when contributing to a conversation about a certain topic. ECRs are used to explain instances of surprise that arise in response to what people say. I distinguish Content Directed Surprise (surprise about the content of an utterance) from Speaker-Oriented Conversational Surprise (surprise that a particular person has said what they have said). We see the latter in, say, a sexist math professor’s surprise that a young woman has given a correct answer. Here the object of his surprise is not her answer (content he knows), but that she had it. Instances of speaker-oriented conversational surprise occur when an ECR is violated (since ECRs amount to beliefs about what people might, could, and should add to conversation). I draw attention to two familiar ways we form them. ECRs are formed through practicing what we will say to certain people given certain subject matters (e.g., in medical ethics some have written guidelines for practitioners’ future participation in difficult conversations like those delivering negative prognoses or aimed at garnering consent for a risky operation). ECRs are also passively formed via tacit beliefs we come to have about people based on their social identity (here I appeal to research in social epistemology about the morally and politically relevant ways that prejudice arises in us and can shape the credibility assessments we make of others). Last, I give an account of the normative profile of instances of speaker-oriented conversational surprise. This account distinguishes the assessment from the phenomenological charge of an instance of speaker-oriented conversational surprise. The assessment reflects how an utterance has violated expectations while the phenomenological charge reflects the personal importance this violation carries for the subject of speaker-oriented conversational surprise.