The simplest kind of stimulus class is one defined simply by a common trained response. Other stimulus classes do not require training of each of the members but they derive their membership from the physical similarity among class members through the process of stimulus generalization. Stimulus classes of greater interest are nonsimilarity based classes for which membership implies properties not explicitly experienced, or so-called emergent properties. Nonsimilarity-based or arbitrary classes can develop when two or more arbitrary stimuli become directly associated through their pairing with a common event. These classes appear to develop during matching-to-sample training when two or more samples are associated with the same comparison (i.e., many-to-one matching).
Convergent evidence will be presented for the development of emergent relations between these samples (or for their common coding) in the form of: (1) Transfer: When a subset of these samples is associated with new comparisons there is evidence that the remaining samples are also associated with the new comparisons. (2) Interference/facilitation: When a subset of these samples is associated with new comparisons and a delay is inserted between the samples and new comparisons, interpolation of one of the remaining samples into the delay will interfere with or will facilitate matching accuracy depending on whether the interpolated sample was paired with the same comparison as the current sample in original many-to-one training. (3) Partial versus total reversal: When originally trained sample-comparison associations are reversed, reversals are faster when all of the associations are reversed than when only half of the associations are reversed (and the rest remain relevant). (4) Samples associated with the same comparison are harder to discriminate from each other (in a simple successive discrimination) than are samples associated with different comparisons. (5) The retention functions (matching accuracy as a function of delay inserted between the sample offset and comparison onset) for samples associated with the same comparison are much more similar in slope than those same retention functions when samples are associated with different comparisons.
The relation between this kind of stimulus class formation and formal (Sidman) stimulus equivalence will be discussed. Specifically, direct evidence for the three defining characteristics of stimulus equivalence (reflexivity, symmetry, and transitivity) in animals will be presented. Finally, the relation between stimulus class formation and human language will be addressed.
78A stimulus class is defined by any group of stimuli that have some property in common. The simplest kind of stimulus class is one defined by a common response. All of the items in a refrigerator can be thought of as a stimulus class because they share a common location and can each be obtained by making a similar response (opening the door of the refrigerator). Stimulus classes of this type typically derive their membership from specific experience (i.e., having placed something in the refrigerator). Representing the members of a class by a common tag or label can serve as a useful and efficient means of storing, retrieving, and communicating information (e.g., about where the item can be found). But if the only relation among members of such a stimulus class is their physical location, such a class would have few implications for the organization of information beyond what was already known about specific experienced responses.
Stimulus classes of greater interest are those for which membership implies properties not explicitly experienced. In the case of the refrigerator example, if I knew that an novel food object had been placed in the refrigerator, I might conclude that it would likely spoil relatively quickly at room temperature. This would be an inference, or emergent property of the novel substance, based on learning about other objects that are typically placed in a refrigerator. The kind of stimulus class of interest here is that for which untrained properties can be shown to emerge. In other words, they are classes in which members are not just related to the class but are also related to each other, such that, unless explicitly trained otherwise, whatever happens to one member of the class will have an effect on the other members.
Consider for a moment the stimulus class DOG as it might be used by humans. Not only might this class contain experiences with a wide variety of dog breeds, but it might also contain pictures of dogs, the sound of the spoken word “dog,” and the written word “DOG.” The value of such as stimulus class is in our ability as humans to use information about one member of the class to make inferences about other members of the class. If you saw a dog bark, you might assume (whether correctly or not) that other dogs bark as well. Similarly, if you were told that there was a widespread outbreak of rabies among dogs, that might change your behavior in the presence of dogs that you encountered (other members of that class).
Perhaps the most important benefit to an organism of emergent relations among members of a stimulus class is when the value of one member of the class changes (e.g., is associated with fear or pain) and the organism can react appropriately, not only to that member of the class, but to other members of the class as well. Thus, emergent relations among members of a stimulus class allow a specific learning experience involving a change in value to have wide spread effects on the value of unexperienced events in the class. The capacity for emergent relations allows organisms to extend learning far beyond the events experienced.
In the remained of this chapter I will describe the various emergent relations that can develop among stimuli in a class and will present evidence of these relations. I will also address some of the controversies in the literature concerning emergent relations among stimuli in a class by humans and nonhumans.