Abstract

Conditional sentences pose a problem for semantic accounts in terms of truth conditions. On the one hand, evidence about the validity of inference patterns suggests that there is a close relationship between conditionals and conditional probability. On the other hand, while probability is typically interpreted as the probability that a proposition is true, that connection has proven elusive in the case of conditionals. Some probabilistic accounts accommodate these obstacles by concluding either that conditionals do not have truth conditions, or that the usual connection between truth and probability does not hold for them. Both of these approaches remain problematic; moreover, definitions of the probability of conditionals directly in terms of conditional probability do not extend easily to complex and embedded conditionals, a limitation of coverage imposed solely by the probabilistic calculus. ;In this dissertation I develop a way to reinstate the connection between conditional probability and probability of truth by adopting a new definition of the latter. Building on previous work on the formal aspects of the problem, I suggest that conditionals denote random variables which, unlike the denotations of other sentences, may take values between 1 and 0 ; the probability of a conditional is the expectation of its values. I explore the rationale for such values and the predictions made by the theory in detail for three common types of conditionals: epistemic, predictive, and counterfactual. Under the proposed theory, conditionals are interpreted in models of the evolution of objective chance and subjective probabilities, both of which are the sources of the values of different kinds of conditionals. The time of evaluation thus plays a crucial part in determining those values. ;The theory predicts values for complex and embedded conditionals as well, but the expectation of these values in a given situation does not always accord with intuitions about what the probability of the sentence should be. I show that this problem can be successfully addressed by making the interpretation sensitive to causal dependencies. This addition opens up new insights into the relationship between counterfactual conditionals and their epistemic and predictive counterparts: They are equivalent but not equiprobable