The randomization procedure in the study of categorization of multidimensional stimuli by pigeons.

Abstract

Pigeons categorized rectangles varying in both height and width in an adaptation of a method used by Ashby and colleagues for the cognitive and neuropsychological analysis of human decision bounds for ill-defined categories. Optimal decision bounds were defined in a stimulus space in which the point (x,y) corresponded to a rectangle with width x and height y. Four tasks defined the following 4 optimal bounds: x = y, x = c, x = y + d, and (x-a)2 + (y-b)2 = r2, where a, b, c, d, and r were constants given by a task. Estimated decision bounds for individual pigeons conformed approximately to the optimal decision bound in each of the 4 tasks. The new method suggests a way to (a) integrate the disparate literature on ill-defined visual concepts and on optimal performances in nonhuman animals; (b) compare how humans and nonhuman animals categorize ambiguous, multidimensional configural stimuli; (c) model how nonhuman animals categorize naturalistic stimuli; and (d) infer that pigeons' categorizations of naturalistic stimuli may be remarkably close to optimal.

Evidence for a procedural-learning-based system in perceptual category learning

“…In fact, the unidimensional and conjunctive rule-based tasks are of lower category discriminability, and thus by all relevant objective measures they are more difficult. In addition, pigeons have no more difficulty learning information-integration categories than rule-based categories (Herbranson, Fremouw, & Shimp, 1999). A second problem with the difficulty argument is that it predicts that any experimental manipulation disrupting category learning should have greater effects with information-integration than with rule-based tasks.…”

Procedural learning in perceptual categorization

“…For example, there is evidence that pigeons can learn both types of category structures (Herbranson, Fremouw, & Shimp, 1999), but no one would claim that they learn rule-based categories via an explicit reasoning process. The question of how people learn rule-based and information-integration categories is strictly empirical.…”

Modeling unidimensional categorization in monkeys

“…Herbranson et al found that the boundaries estimated from the pigeons' performance approximated the boundaries necessary to perform the task optimally, a result previously found with humans (Ashby & Gott, 1988;Ashby & Maddox, 1992;. Surprisingly, this was the case even when the optimal boundary was nonlinear (see, e.g., Figure 10 of Herbranson et al, 1999). Although the experiment was not intended to test any particular model, the patterns of boundary use suggest that complex representations may underlie the categorization behavior of animals, representations that perhaps approach the complexity of those underlying humans' performance on such perceptual tasks (Herbranson et al, 1999).…”

“…Furthermore, Shimp et al found a nonlinear relationship between the likelihood that a sample was generated by a particular coin and the probability of responding in favor of that coin, indicating that the pigeons did not simply match probabilities. In a similar vein, Herbranson et al (1999) tested pigeons on a multidimensional categorization task in which the members of each category, rectangles, were normally distributed along two stimulus dimensions, width and height. This experiment was a replication of tasks previously employed to examine decision bound accounts of human perceptual categorization (Ashby & Gott, 1988;Ashby & Maddox, 1992).…”

“…Although animal categorization has not enjoyed the extensive theoretical development witnessed in human categorization (Shimp, Long, & Fremouw, 1996), some recent research has taken advantage of the advanced state of theory in the human literature by focusing on examining animals' categorization under conditions previously examined in humans (Fagot, Kruschke, Dépy, & Vauclair, 1998;Herbranson, Fremouw, & Shimp, 1999;Medin & Dewey, 1984;Sigala, Gabbiani, & Logothetis, 2002;Smith, Minda, & Washburn, 2004). In several cases, this work has been complemented by application of well-developed quantitative theories of human categorization to the behavior of animals (Fagot et al, 1998;Herbranson et al, 1999;.…”