Crosslinguistically, expressions of ability exhibit a curious duality of interpretation, in some contexts describing the abilities and potential of an agent, and in others simply describing what the agent did on a particular occasion. In languages that mark grammatical aspect, the alternation between ability and action extends to abilitative uses of the possibility modal, and is governed by aspectual marking (Bhatt, 1999). For instance, imperfectively marked uses of the French modal 'pouvoir' ('can') are compatible with pure, potentially unrealized ability interpretations; by contrast, perfectively marked 'pouvoir' gives rise to actuality entailments, requiring the realization of its complement, and seemingly very little else. An influential line of work seeks to derive actuality entailments in the composition of modality and aspect, treating ability as a type of circumstantial possibility operator, and the perfective aspect as imposing temporal boundaries on eventualities in its scope (Hacquard, 2006, a.o.). This dissertation lays the groundwork for an account that links both the ability and actuality interpretations to a novel component in the semantics of ability: causal dependence. The main idea is that ability modals describe a complex causal structure, in which the (circumstantial) possibility that an agent S will realize an event A(S) obtains in view of the causal dependence of A(S) on an available choice or action for S. This proposal is motivated by philosophical work on ability, which suggests that abilitative possibilities have stronger truth conditions than pure circumstantial possibilities (Kenny, 1976; Brown, 1988). I develop the argument for a causal account of ability by comparing actuality inferences to the interpretation of two other types of complement-taking predicates: implicative verbs (e.g., 'manage'; Karttunen, 1971) and 'enough' and 'too' predicates (e.g., 'be fast enough'; Meier, 2003). I show that, in both cases, complement inferences follow from the combination of two things: (i) the presupposition that some prerequisite action for an agent is causally necessary and causally sufficient for the complement, and (ii) a determination of whether or not the prerequisite action occurred. Implicative verbs resolve the prerequisite as asserted content, deriving their characteristic complement entailments as causal consequences. 'Enough' and 'too' constructions, by contrast, simply indicate that the prerequisite action is available to the agent. Drawing on theories of aspectual coercion (Moens and Steedman, 1988, a.o.), I argue that perfective aspect interacts with this 'availability' assertion by systematically forcing an interpretation on which the agent instantiates the prerequisite. As a result, imperfectively marked 'enough' and 'too' constructions imply that their complements are possible, but perfectively marked constructions entail their complements as causal consequences of the prerequisite, in the same way as implicative verbs. 'Enough' and 'too' constructions thus represent a special type of ability attribution, which is specific about the nature of the causal prerequisite for the ability-complement. Pursuing this analysis, the actuality inferences of ability modals result not just from the composition of modality and aspect, but more specifically from the composition of aspect with the specific type of complex causal possibility conveyed by ability predicates. I formalize causal dependence relations over the structure of a causal model which represents causal connections between events as directed links in a graphical network (Pearl, 2000; Schulz, 2011; Kaufmann, 2013). In such a model, the felicity conditions imposed by causal necessity/sufficiency presuppositions depend crucially on the discourse background. Grammatical aspect then selects for a particular interpretation of the abilitative causal structure by selecting for a particular type of background. I argue that this view of ability is a natural extension of the standard modal theory, and suggest that formal models of causation are one way of representing reasoning about the 'normal' developments of situations ('stereotypicality'; Kratzer 1981).