Heterogeneity and uniformity in the evidential domain

The dissertation is devoted to the formal mechanisms that govern the use of evidentials, expressions of natural language that denote the source of information for the proposition conveyed by a sentence. Specifically, I am concerned with putative cases of semantic variation in evidentiality and with its previously unnoticed semantic uniformity.An ongoing debate in this area concerns the relation between evidentiality and epistemic modality. According to one line of research, all evidentials are garden variety epistemic modals. According to another, evidentials across languages fall into two semantic classes: (i) modal evidentials; and (ii) illocutionary evidentials, which deal with the structure of speech acts. The dissertation provides a long-overdue discussion of analytical options proposed for evidentials, and shows that the debate is lacking formally-explicit tools that would differentiate between the two classes. Current theories, even though motivated by superficially different data, make in fact very similar predictions. I reduce the cases of apparent semantic variation to factors independent from evidentiality, such as the syntax of clausal complementation, and show that these cases do not resolve the modal-illocutionary debate. I further propose novel empirical diagnostics that would identify modal-hood and speech-act-hood.I then turn to the many traits that evidentials within and across languages have in common. I argue that evidentials belong to the class of subjective expressions, along with first-person pain and attitude reports, and attribute to them a unified semantics of first-person mental states. The subjectivity of evidentials is contributed by two components: (i) the first-person component that is part of the conventional meaning of evidentials, analyzed as indexicality; and (ii) the mental state component that is rooted in the properties of cognitive processes described by evidentials (and other subjective expressions), such as perception and introspection.I show that the subjectivity of evidentials restricts their behavior across a range of environments in a uniform way. In dialogues, subjectivity accounts for the resistance to direct denials, a property known as non-challengeability and previously seen as supporting the not-at-issue analysis of evidentiality. In attitude reports, subjectivity disallows ascribing evidence to a third party and bans evidentials from amnesiac scenarios, used in the literature on attitudes as a litmus test for ‘de se’. In information-seeking questions, subjectivity creates an effect of obligatory shift to the addressee because it is incompatible with speaker-oriented interpretations wherein the speaker does not have access to their own epistemic state. I further show that evidentials may be speaker-oriented in non-canonical questions. That evidentials shift has been previously hardwired to their syntax and/or semantics, which fails to explain the lack of shift in non-canonical questions.If language is in some ways a window on the mind, evidentiality is a natural meeting point for several areas, including at least linguistics, philosophy of language, philosophy of mind, and epistemology. But so far, expressions of evidentiality have been studied in-depth almost exclusively within formal semantics. Current linguistic theories of evidentiality are disconnected from theories of knowledge and models of reasoning. By deriving the linguistic behavior of evidentials from non-linguistic properties of experiences they describe, this dissertation makes a necessary first step towards filling this gap