LEARNING FROM TRIGGERS

In this article we provide a refined analysis of learning in finite pa rameter spaces using the Triggering Learning Algorithm (TLA) of Gibson and Wexler (1994). We show that the behavior of the TLA can be modele d exactly as a Markov chain. This Markov model allows us to (1) descri be formally the conditions for learnability in such spaces, (2) uncove r problematic states in addition to the local maxima described by Gibs on and Wexler, and (3) characterize convergence times for the learning algorithms quantitatively. In addition, we present arguments question ing the psychological plausibility of the TLA as a learning algorithm.