Generalization

Description of the Strategy

Generalization as an effect has been noted since researchers began discussing learning. Original experiments in classical conditioning found that animals conditioned to a particular tone would initially respond to similar tones. Operant conditioning has shown that an individual who is reinforced for a particular response is more likely to make that response in similar situations. Both these examples illustrate the general concept of generalization. Most often, during learning trials an organism learns not only to respond to the target stimuli but also to not respond to other, somewhat similar stimuli. This later phenomenon is discrimination, and it is unusual to see a description of generalization without a concomitant discussion of discrimination.

At the most basic level, all instances of learning may be considered generalization. In this most broad conceptualization, whenever an organism predicts that because a past behavior was reinforced the same behavior will be reinforced in the future, the organism shows generalization. Thus, according to this broad view, any instance of learning is generalization.

Most researchers, however, are not referring to the broad conceptualization when they discuss generalization. In general, when discussing generalization, researchers and practitioners are focusing on “one-many” generalization, also known as “stimulus generalization.” Stimulus generalization refers to an organism responding to similar stimuli or situations in the same manner, albeit weaker, as it did to the original stimuli or situation.

Basic research with generalization has noted that generalization follows a predicted generalization gradient. For stimuli that are nearly identical to the original stimuli, the organism will respond in nearly the same way as it did to the original stimuli. Thus, in this instance the generalization gradient is flat, indicating that the response will be nearly as strong to similar stimuli as to the original stimuli, but the response gets somewhat weaker as the new stimulus deviates from the initial stimuli. In contrast, the greater the new stimuli differs from the initial stimuli, a more steep generalization gradient is noted. This means that the individual's response is likely to be weaker than the initial response, and the response becomes less strong the more the new stimuli differs. Learning theorists argue that what happens during generalization training is that the individual responds to a population of stimuli. Each population has a mean value for the response(s), and there is typically variability around the mean. Thus, the stimulus gradient follows a predictable pattern and can be seen for all types of stimuli, including visual and auditory stimuli, among others. In addition, this phenomena is observed during training using more basic stimuli (color hue, auditory pitch) and more complex phenomena (word learning). Overall, generalization is a robust phenomena that is observed among all organisms.

Studies conducted over many decades have shown the existence of generalization and the parameters affecting its acquisition for basic concepts. However, generalization has an important part in the applied literature as well. In this context, generalization is best [Page 862]described as the continued display of the behavior once the treatment program has been reduced or eliminated. The conceptualization of generalization within a treatment context indicates that there are two main aspects: (1) continuation of treatment effects after the elimination or reduction of the intervention and (2) presence of treatment effects in situations other than the one that was the focus of the intervention. Within the treatment literature, some authors have suggested that generalization and discrimination are so important to behavioral analysis that these principles should be primary and the investigation of ways in which behavior change is effected should be secondary.

...