Natural concepts and generalization classes.

Stemmer (1980) looked at how pigeons sort new pictures. The birds pecked at one key if the picture showed trees, another key if it showed water.

The author argued the birds were not using an inborn idea of tree or water. Instead, they learned a generalization class based on past reinforcement.

The paper says we should drop the word concept and say generalization class to stay true to behavior analysis.

The article is theoretical, so no new data are given. It claims that any animal’s category is just a set of stimuli that control the same response because of shared reinforcement history.

In short, categories are learned, not built-in.

Fields et al. (1991) later showed the idea works with people. After adults formed three-member equivalence classes, they treated brand-new line lengths as part of the class if the lengths looked like trained members. This gives an easy lab model for natural categories.

Carr et al. (2002) seems to disagree. Pigeons that only saw tree pictures did not peck differently when new tree photos popped up. The catch: the birds never saw a second category like water. Once the study added a non-tree category, generalization appeared. So the two papers line up—experience with contrast is needed.

Logue et al. (1986) built on the same theme. They listed five ways to spot real-world reinforcers that keep generalized behavior alive after you leave. Together, the studies turn the 1980 argument into a practical roadmap.

Stop saying concept when you write reports. Say generalization class and link it to the client’s history of reinforcement.

Before you claim a learner knows a category, test novel items that resemble trained ones and check that the response holds.

Also, give experience with at least one contrast category so the generalization class can form. This small wording and testing shift keeps your work inside behavioral boundaries and makes your data cleaner.