Three experiments investigated categorical discrimination and generalization in pigeons. Multiple fixed interval-extinction training was conducted with a pool of 48 different negative discriminative stimuli (12 slides each of people, flowers, cars, and chairs). The most errors were committed to negative stimuli (S-s) from the same category as the 12 positive stimulus (S+) slides. Such categorical generalization was stronger when the 12 S+s entailed 1 copy of 12 different slides (Experiment 2) than when the S+s entailed 12 copies of 1 slide (Experiment 1). In addition, reliable but incomplete loss of inhibitory control was observed to novel stimuli chosen from the same category as the S-slides (Experiment 3). These results are consistent with perceptual theories of categorical coherence, according to which preexisting similarities among stimuli chiefly determine the acquisition and application of categories.We thank Don Dorfman,

Four experiments used a four-choice discrimination learning paradigm to explore the pigeon's recognition of line drawings of four objects (an airplane, a chair, a desk lamp, and a flashlight) that were rotated in depth. The pigeons reliably generalized discriminative responding to pictorial stimuli over all untrained depth rotations, despite the birds' having been trained at only a single depth orientation. These generalization gradients closely resembled those found in prior research that used other stimulus dimensions. Increasing the number of different vantage points in the training set from one to three broadened the range of generalized testing performance, with wider spacing of the training orientations more effectively broadening generalized responding. Template and geon theories of visual recognition are applied to these empirical results.

Training associated pairs of perceptually dissimilar stimulus classes with a common delay or probability of food reinforcement in pigeons. Then, different choice responses were trained to 1 component class in each pair. In a choice test, the untrained class in each pair occasioned the same response as did the choice-trained class. In a 3rd experiment, 2 classes had reinforcement delays of 1 s and 15 s, respectively, and 2 other classes had reinforcement probabilities of 0.1 and 0.9. Then, 1 choice response was reinforced to a class previously associated with a better condition of reinforcement (e.g., 1-s delay or 1.0 probability), and a different response was reinforced to a class previously associated with a worse condition of reinforcement (0.1 probability or 0-s delay). Testing with all classes suggested that categorization was based on the relative reinforcement or hedonic value and not on the parametric details of reinforcement.

Superordinate categorization via association with a common response was studied in pigeons. Original training paired disparate classes (e.g.,people + chairs and cars + flowers) with a common response (Responses 1and 2,respectively). Reassignment training taught new responses (Responses 3 and 4,respectively) to one component class from each pair (e.g., people and cars). Superordinate categorization was documented in testing when the pigeons made the same responses to the stimuli that were withheld in reassignment training (e.g., chairs and flowers) as they did to the reassigned stimuli themselves (e.g., people and cars) and when the birds transferred these discriminative responses to novel stimuli from all four component classes. Reassignment training with novel stimuli produced effects that were similar to those of reassignment training with familiar stimuli. Superordinate categorization via association with a common response is thus a robust effect that generalizes to novel stimuli from each of the component classes.

No abstract