An all-or-none theory for learning on both the paired-associate and concept levels

Abstract A theory is proposed for paired-associate learning tasks in which several items are interrelated to facilitate learning. The theory is a synthesis of the all-or-none paired-associate and concept models. It assumes that when an item is presented either nothing is learned, or that item alone is learned, or the group of related items is learned. Thus the theory postulates both instance learning and rule learning. From the general theory, a number of learning models are derived for various item presentation schedules and levels of data analysis. In particular, the serial learning, study-test, anticipation, and Bernoulli procedures are studied in conjunction with the individual item level and concept level error-success processes.