This dissertation concerns the expression of epistemic modality across syntactic categories in Kipsigis, an understudied Kalenjin language of Western Kenya. In addition to describing a range of morphosyntactic and semantic properties of Kipsigis, this dissertation takes as its empirical focus two case studies of epistemic modality in the nominal and verbal domains. Through these case studies, this dissertation contributes to the integration of data from understudied languages into semantic theory and suggests that the derivation of epistemic modal content—both across languages and across syntactic domains—must be achieved in a range of different ways.

In the nominal domain, I describe and analyze Kipsigis epistemic indefinites, highlighting the ways in which these forms pose challenges for existing analyses of epistemic indefinites across languages. In particular, I show that Kipsigis epistemic indefinites can signal both first order and higher order ignorance but do not take obligatory wide scope, which challenges the correlation between ignorance type and scope in Dawson (2020). To capture the Kipsigis pattern, I propose that the ignorance effects triggered by use of an epistemic indefinite in Kipsigis arise pragmatically due to competition with another type of Kipsigis nominal, which signals speaker knowledge about the noun’s referent (following work on Russian to and koe indefinites by Geist & Onea 2007).

In the verbal domain, I consider biased belief reports with the Kipsigis verb pɑr ‘think’, which can be used in two very different contexts: 1) to suggest that the reported belief is false or unlikely, and 2) to remind the addressee that the reported belief is true. While these negative bias and reminding functions are independently attested in other languages, the Kipsigis pattern is unique in that it combines these two functions in a single lexical item. To account for this behavior, I propose that, in addition to its basic belief semantics, pɑr contributes an instruction for Common Ground management (Krifka 2008): the reported belief must not be added to the Common Ground.

This dissertation probes the nature of the syntactic operation of Agree through the lens of the morphosyntax of Amahuaca, an endangered Panoan language of the Peruvian Amazon. I take as my empirical focus two interrelated case studies in Amahuaca syntax: 1) the split ergative case system, and 2) the extensive switch-reference system. In the domain of case, I argue that overt ergative case morphology in Amahuaca expones agreement of the transitive subject with multiple functional heads. This leads to a distinction between the features needed for abstract ergative case (agreement only with v), and the features needed to trigger overt ergative case (agreement with both v and T). This distinction between abstract and morphological case factors into the analysis of the switch-reference system of Amahuaca, which I argue is sensitive to abstract case. In addition to case-sensitivity, Amahuaca’s switch-reference system shows the typologically unusual property of tracking the reference of all arguments of the verb, not only subjects. I propose that this system arises through adjunct complementizer agreement that probes both the adjunct and matrix arguments directly through cyclic expansion of the probe. Through these two investigations, I conclude that Amahuaca provides support for a narrowly cyclic model of Agree in which each instance of Merge defines a new cycle of Agree (Rezac 2003, 2004; Béjar and Rezac 2009). Further, the empirical facts can be most straightforwardly accounted for if we assume that some probes are insatiable, agreeing with all possible goals in their search space (Deal 2015b). Finally, despite the fact that some agreement in Amahuaca appears to be long distance, I argue that the data can be captured under the fairly conservative assumption that Agree is always under c-command and is always phase-bound.