DescriptionThis dissertation centers on two questions: (1) Can we explain epistemic facts in terms of non-epistemic facts? (2) What is the most explanatorily basic notion within the epistemic realm? Many philosophers are attracted to the idea that the epistemic is reducible to the natural: facts about epistemic justification, knowledge, and the like can be explained in terms of non-epistemic facts. How could such a reduction be achieved? Chapter One explores the two leading proposals in the literature: process reliabilism and mentalist evidentialism. I argue that both of these approaches flounder when it comes to explaining epistemic defeat (cases where an individual gets some evidence in favor of a belief, which is then trumped by countervailing evidence). The standard process reliabilist treatment of defeat faces counterexamples, and leading evidentialist treatments of defeat either fall victim to the same fate or fail to be reductive. Chapter Two explores a different reductive strategy. I suggest that we should pursue the path of semantic ascent: we should focus on epistemic linguistic expressions and seek to define them without recourse to epistemic notions. Specifically, I develop an ‘attitudinal’ semantics for a variety of epistemic expressions, according to which epistemic expressions are analyzed in terms of the conative attitudes that give rise to them. The resulting semantics is reductive; in addition, it offers to explain some of the striking commonalities between epistemic and ethical discourse. Chapter Three considers reduction within the epistemic domain. Recent epistemology has seen the rise of ‘Knowledge First’ epistemology, according to which knowledge is the most basic explanatorily basic epistemic notion. I advance an alternative picture, according to which epistemic certainty is explanatorily fundamental. After developing a context-sensitive semantics for certainty ascriptions, I go on to put certainty to explanatory work. I argue that a wide range of epistemic phenomena—including epistemic modals, evidential probability, and knowledge itself—can be profitably analyzed in terms of certainty.