Categories, Generalization, and Vagueness

Since the publication of Lewis (1969), philosophers have used signaling games to better understand the development of language and information transfer. Increasingly, these games have proved to be a powerful conceptual tool in a number of fields. Part of the reason these games have proved so powerful, and so broadly applicable, is the highly simplified nature of the model. It pairs down signaling scenarios to the barest relevant aspects, thus allowing researchers to understand phenomena that are usually too complex to pull apart. Because the model is so highly simplified, one research avenue that has been successful in the past involves the introduction of aspects to the game intended to better model real world signaling.;One simplification of the Lewis signaling game regards its treatment of payoff. In real world signaling situations sometimes actions will be appropriate in multiple states of the world, or appropriate to varying degrees in different states. Here I explore a modified signaling game---the sim-max game introduced by Jager (2007)---that is a model of such situations. Actors in the sim-max game receive perfect payoffs for choosing the correct act for a state, but they also receive payoffs for choosing an act that is nearly appropriate. This is done by building a similarity structure over the state space of the sim-max game and assuming that acts will be somewhat appropriate for states that are similar to the ideal state for that act. In this way, payoffs can vary gradually over various state-act combinations.;In this dissertation, I use this modified game, and a related decision problem, to address several topics of interest to philosophers and biologists. In chapter 2, I discuss the evolution of linguistic vagueness. In chapter 3 I address how learning generalization can evolve. And in chapter 4 I discuss the ways evolutionary models can inform the connection between perceptual categories and real world structure.