UC Berkeley

Formal developments of normative theories typically claim that the guidance they give is universal: for any agent, and any way the world could be, there is a way she is permitted (according to that theory) to act. Yet when we consider an agent facing an open, indeterminate future, cases are possible in which what is permitpted depends on what she actually does. These situations follow the letter of the law while seeming to violate its spirit. A famous example, discussed by Gibbard and Harper (1978), comes from Somerset Maugham: while in Damascus, you learn that Death is coming to collect your soul. Your one option is to flee to Aleppo. But you are confident that Death never misses her quarry: if you flee to Aleppo, Death will be there. But if you stay in Damascus, Death will be there too.

If Death is going to Damascus, you should go to Aleppo, and if Death is going to Aleppo, you should go to Damascus. So for any way the world could be, there is a way you should act. Yet there is a clear sense in which there is nothing you can do: since Death's destination depends on yours, no act is such that you ought to have done it, *given that you do it*. The norms of rationality in cases of this structure---and cases with the opposite structure, where available acts deontically validate themselves---are the subject of much recent work in ethics, decision theory, and the metaphysics of persons. I show how a model theory for the natural language modals Ought and May can incorporate these notions of deontic validation and self-defeat. Because May tracks the concept of permissibility brought out by act-dependent cases, its inferential properties reflect the language-*independent* intuitions we have about choiceworthiness, in cases like Death in Damascus.

This theorizing makes contact with natural language in the form of my solutions to two infamous puzzles about deontic modal language, free choice permission (Kamp 1973) and Ross's puzzle (Ross 1941). Free choice permission is the apparent validity of the classically *invalid* inference from May([phi] or [psi]) to both May[phi] and May[psi], and Ross's paradox is the apparent *in*-validity of the classically valid inference from Ought[phi] to Ought([phi] or [psi]). The first step to a unified solution to these puzzles is precisely to leverage the notion of permissibility corresponding to deontic self-validation. The second component is a generalization of classical logic. On my account, the interpretation of a disjunction depends on which of its atomic disjuncts are true at the actual state---where the 'actual state' can be keyed to the *future*-actual state an agent chooses when she acts. While classical consequence is preserved for sentences without modals, this analysis sets up a match between the act one brings about and the contents of statements describing that act's deontic status: for example, in futures where the agent chooses to [phi], it is [phi] that she has permission to do. This allows the stronger-than-classical conclusions of free choice permission to follow, and blocks the inference in Ross's puzzle. It also predicts the positive entailment properties of disjunction under Ought, while preserving the inviolability of the role classical disjunction plays in our reasoning. This intuitively appealing combination has, in previous work, proved difficult to achieve.