Transfer of color matching in pigeons

Abstract

Two pigeons were tested for transfer of matching to novel colors after a three-phase training procedure: First, the birds were trained to match colors; second, they were trained to choose odd colors; and third, they were retrained to match colors. The third phase included additional trials in which pecking at novel colors was reinforced. A subsequent transfer test showed that performance on trials with novel colors was slightly above chance but below levels observed on trials with only familiar colors. In addition, there was no difference between performance on trials in which both the sample and correct comparisons were novel and trials in which only the incorrect comparison was novel. These results provide some evidence for the acquisition of a general matching strategy, but they suggest that when pigeons are presented novel stimuli in a test for transfer of matching, their behavior displays biases that can lead to misinterpretations of what is learned in matching tasks.In a matching-to-sample task, a subject is exposed to a sample stimulus and then to two comparison stimuli. A response to the comparison stimulus that is similar to the sample is reinforced, and a response to the comparison that is different from the sample is not reinforced. Several experiments designed to identify what pigeons learn in the matching problem have been based on a transfer-of-training paradigm (Cumming & Berryman, 1965;Farthing & Opuda, 1974). In this paradigm, subjects are trained to match with one set of stimuli. After a high level of matching performance is achieved, these stimuli are said to be familiar. Subsequently, a transfer test is administered with a novel set of stimuli. The results show that on transfer trials in which the sample and correct comparison are familiar and the incorrect comparison is novel, performance is as accurate as on trials in which all three stimuli are familiar. However, when the sample and correct comparison are novel, performance is no better than chance, and it is often significantly below chance. Both Cumming and Berryman (1965) and Farthing and Opuda (1974) have used results such as these to argue that pigeons do not attend to the incorrect comparison in matching-to-sample training. Rather, pigeons learn a conditional stimulus response chain in which a specific sample stimulus directs responses to a specific comparison stimulus.However, there is an alternative description of transfer performance: On all trials in which the correct comparison is familiar, performance is good; on all trials in which the correct comparison is novel, performance is poor. Thus, pigeons may simply have a bias against pecking novel stimuli. This bias might mask the performance of whatever is learned in matching-to-sample training. The purpose of the present experiment was to evaluate this hypothesis. Here, we tried to neutralize this potential bias by reinforcing pecks at novel stimuli before those stimuli were used in the transfer test.Previous experiments have investigated this hypothesis and have obtained contrad...