Erik M. Bollt, United States Naval Academy, Mathematics Department, Annapolis, MD
Michael A. Jones, Montclair State University, Upper Montclair, NJ

Abstract: In experimental psychology, artificial grammars, generated by directed graphs, are used to test the ability of subjects to implicitly learn the structure of complex rules. We introduce the necessary notation and mathematics to view an artificial grammar as the sequence space of a dynamical system. The complexity of the artificial grammar is equated with the topological entropy of the dynamical system and is computed by finding the largest eigenvalue of an associated transition matrix. We develop the necessary mathematics and include relevant examples (one from the implicit learning literature) to show that topological entropy is easy to compute, well defined, and intuitive and, thereby, provides a quantitative measure of complexity that can be used to compare data across different implicit learning experiments.

Keywords: artificial grammars, implicit learning, symbolic dynamics, complexity, topological entropy