Abstract
In this paper we present a formal learning algorithm based both on the Occam?s razor and on Chomsky?s classification of languages. Since Chomsky proposes that the generation of language (and, indirectly, any mental process) can be expressed through a kind of formal language, we assume that cognitive processes can be formulated by means of the formalisms that can express those languages. We apply this idea to the simplest languages according to Chomsky?s classification, the regular languages, which can be expressed by finite state machines. Besides, we apply the Occam?s razor principle, which says that when data do not allow to distinguish between two theories, the simplest one should be chosen. This principle, basic in science, is implicitly applied in the human brain. We apply these concepts to construct an algorithm that provides the simplest finite state machine (that is, the simplest cognitive theory) that fits into some given world observation. Thus, the resulting machine is the most preferable theory for the observer, according to the Occam?s razor criterion.