Multiple Exemplar Training: Some Strengths and Limitations.

Abstract

In successful multiple exemplar training, a set of exemplars that sample the range of stimulus and response topographies is trained, and the full range of performances results. Examples abound in experimental psychology and include abstraction and concept learning, responding to relations, identity matching, rule following, behavioral variability, responding to wh-questions, describing past events, learning sets, and continuous repertoires. Thus, behavior analysts often allude to a history of multiple exemplar training to account for different generalized performances. It is easy to see how a strategy of multiple exemplar training can work with many types of performances, even performances that involve relations between objects or events. However, there appear to be at least two exceptions, where direct multiple exemplar training does not work well: (1) when there are no physical dimensions at all along which generalized performances can emerge, and (2) when the relation between a stimulus and an effective response is complex. Interpretation of the latter type of cases in terms of mediated generalization is outlined and discussed. An experimental and conceptual research program should produce an account of the general limits of multiple exemplar training, and guidelines for the most effective training for generalized skills.