Ought and should

The semantics of good places strong constraints on the better-studied and near-synonymous items ought and should. I argue that comparative goodness fact constrain – but do not determine – the interpretation of ought, and propose “Sloman’s Principle”: ought p implies that p is better than all of its alternatives. This is not yet a semantics for ought, but it takes us surprisingly far, as I show by analysing a number of puzzles involving ought and probabilistic information. It is also incompatible with the classically valid inference patterns “agglomeration” and “deontic detachment”. Several examples and an experiment show that this consequence is correct. The account can be strengthened to enforce validities such as the “Smith argument” (Horty 2003) and Weakening (Cariani 2015). Finally, I provide data showing that ought and should are actually the positive forms of gradable verbs, and discuss prospects for deriving their behaviour from the structure of their scales.