DescriptionMany languages grammatically mark evidentiality, i.e., the source of information. In assertions, evidentials indicate the source of information of the speaker while in questions they indicate the expected source of information of the addressee. This dissertation examines the semantics and pragmatics of evidentiality and illocutionary mood, set within formal theories of meaning and discourse. The empirical focus is the evidential system of Cheyenne (Algonquian: Montana), which is analyzed based on several years of fieldwork by the author. In Cheyenne, evidentials are part of the illocutionary mood paradigm. Based on this grammatical system and crosslinguistic data in the literature, I propose a new theory of evidentials. I argue that evidentials contribute not-at-issue content, which cannot be directly challenged or denied. This content is added directly to the common ground, without negotiation. In contrast, at-issue content, the main point of a sentence, is proposed to the common ground, up for negotiation. This analysis of evidentials implies a more articulated theory of assertion and other speech acts. In particular, I argue that all speech acts are structured into three components: presentation of the at-issue proposition, a non-negotiable update that directly restricts the common ground, and a negotiable update that imposes structure on the common ground. I implement this proposal in an update semantics with individual, modal, and propositional discourse referents. The distinction between at- issue and not-at-issue information comes out as an instance of grammatical centering in the modal domain. The presentation of the at-issue proposition is modeled as the introduction of a propositional discourse referent. This predicts that only the at-issue proposition can be referred to in subsequent discourse, and the non-challengeability of the evidential falls out as a special case of propositional anaphora. The proposed analysis can be extended to evidentials and related phenomena in other languages. While there are real crosslinguistic differences in the behavior of evidentials, there are also many commonalities. The proposed analysis captures the properties that all evidential systems share, but is fine-grained enough to account for variation. On this analysis, evidentials crosslinguistically form a natural semantic class.