Generics as Modals

A common view of generics is that they express universal quantification over normal individuals. I argue against this approach, and provide evidence that the quantification is neither universal nor over normal individuals. Regarding universality, generics fail standard tests for universal quantification. As an alternative, I propose that generics express measures of high probability. I formalize this notion using a modal extension of first order probability logic, and demonstrate how this system can account for embedded generics and inference with generics. As for normality, I contrast it with another notion, that of uniformity: generics are evaluated with respect to worlds whose future resembles their past. Using standard substitution intensionality tests, I demonstrate that generics are evaluated with respect to uniform, rather than normal worlds. The end result is that a generic such as Birds fly does not mean "All normal birds fly"; rather, it means that the probability for a randomly chosen bird to fly is high, and this tendency is expected to continue.