Abstract
It has been assumed that when pigeons learn how to match to sample, they learn simple stimulus-response chains but not the concept of sameness. However, transfer to novel stimuli has been influenced by pigeons’ tendency to be neophobic. We trained pigeons on matching (n = 7) and mismatching (n = 8) with colors as samples and, with each sample, one color as the nonmatching comparison. We then replaced either the matching or the nonmatching stimulus with a familiar stimulus never presented with that sample. Results suggest that for both matching and mismatching, pigeons locate the stimulus that matches the sample: If the task involves matching, they chose it; if it involves mismatching, they avoid it. Thus, the concept of sameness is the basis for correct choice with both tasks. This finding suggests that sameness is a basic concept that does not have to be learned and may have evolved in many species, including humans.
In 1950, B. F. Skinner boldly proposed that when pigeons learn how to match to sample (e.g., when the center light is red, choose the red comparison light; when the center light is green, choose the green comparison light), what they learn is two simple stimulus-response chains: to peck red after observing red and green after observing green. Furthermore, Skinner proposed that mismatching (pecking green after observing red and red after observing green) should be no more difficult to learn than matching because they are arbitrary associations. The implication is that when pigeons learn either task, they learn two relatively simple but arbitrary stimulus-response chains in which sameness plays no role. This hypothesis was supported by early research that found that, after matching training, if one introduced a novel color as both the sample stimulus and the correct comparison stimulus, matching accuracy fell to chance, whereas if one replaced the incorrect stimulus with the same novel stimulus, there was no reduction in matching accuracy (see Cumming & Berryman, 1961). However, this research failed to recognize the fact that pigeons tend to avoid novel stimuli, and during testing trials, the researchers always gave the pigeon a choice between a familiar and a novel comparison stimulus. Thus, when the sample and the correct stimulus were both novel, the pigeons tended to choose the familiar incorrect stimulus, whereas when the incorrect stimulus was novel, the pigeons chose correctly because the correct stimulus was familiar.
Zentall and Hogan (1974, 1976) attempted to correct the Cumming and Berryman (1961) bias by training pigeons on matching and mismatching (i.e., reinforcement for choosing the comparison that does not match the sample) and transferring the pigeons to all novel stimuli. This procedure produced significant evidence of concept transfer (both same and different), but it likely underestimated the degree of concept transfer, because transferring to all novel stimuli is likely to be disruptive.
To determine the role of sameness in matching and mismatching without using novel stimuli in testing, Zentall, Edwards, Moore, and Hogan (1981) trained pigeons on matching and mismatching using four colors as samples, but in each case using only two of the remaining colors as the nonmatching comparison in training. To determine the basis of learning, they replaced either the correct or incorrect stimulus with the remaining familiar color. For the matching task, as expected, replacing the correct matching stimulus resulted in a large decrease in accuracy, whereas replacing the incorrect nonmatching stimulus had little effect on accuracy. For the mismatching task, however, replacing the correct nonmatching stimulus had little effect on accuracy, whereas replacing the incorrect matching stimulus resulted in a large decrease in accuracy. Thus, for both groups, the results suggested that when the matching stimulus was replaced, accuracy decreased, whereas when the nonmatching stimulus was replaced, accuracy remained high. For mismatching this was surprising, because replacing the incorrect matching stimulus meant that the correct stimulus from training was still present, yet accuracy suffered. The results suggested that when pigeons acquire mismatching, they look for the matching comparison stimulus and avoid it.
A potential problem with this study, however, was that for mismatching trials, the subjects would have had to learn two sample–comparison relations (stimulus-response chains) for each sample, whereas for matching trials, they would have had to learn only one. This may have biased the mismatching pigeons against learning the relations between the sample and the correct comparison stimulus. Thus, the purpose of the present study was to make the potential learning for both tasks symmetrical by making the potential number of learned relations between the sample and the correct comparison stimulus the same for both the matching and mismatching groups.
Method
Subjects
The subjects were 16 unsexed White Carneau pigeons, 5 to 8 years old, obtained from the Palmetto Pigeon Plant in Sumter, South Carolina. One of the pigeons in the matching group died during the course of training, and those data were not included in any analysis. Four to eight pigeons per group are typically used in research of this kind. All of the pigeons had prior experience with simultaneous color discriminations. The pigeons had free access to water and grit in a climate-controlled colony room that was maintained on a 12:12-hr light/dark cycle. During the experiment, all pigeons were maintained at 85% of their free-feeding body weight and were cared for in accordance with the University of Kentucky’s animal-care guidelines.
Apparatus
The experiment took place in a BRS/LVE (Laurel, MD) standard sound-attenuating operant test chamber measuring 34 cm high, 30 cm wide, and 35 cm across the response panel. Three circular response keys (2.54 cm in diameter) were horizontally aligned on the response panel (spaced 6.0 cm apart from edge to edge) and were located 25 cm from the floor. A 12-stimulus in-line projector (Industrial Electronics Engineering, Van Nuys, CA) with 28-V, 0.1-A lamps (GE 1820) was mounted behind each of the three response keys so as to project red, yellow, blue, and green hues on each of the three keys. Reinforcement consisted of 1.5-s access to mixed grain (Purina Pro Grains, a mixture of corn, wheat, peas, kaffir, and vetch) provided from a feeder. A 28-V, 0.04-A lamp illuminated the feeder when reinforcement was delivered. Experimental events were controlled by a microcomputer and interface located in an adjacent room.
Procedure
Pretraining
The pigeons were first trained to peck all four colors on each of the three response keys for reinforcement. Pecks to the stimuli on the center key were gradually increased from one to five.
Training
The pigeons were randomly assigned to the matching and mismatching groups such that there were eight pigeons in each group. At the start of each trial, the center key (sample) was lit. Five pecks to the sample illuminated the side (comparison) keys (the sample remained on). A single peck to either comparison key terminated the trial and initiated a 5-s intertrial interval. Correct comparison-key responses were reinforced. For pigeons in the matching condition, a peck to the stimulus that matched the hue of the sample was defined as correct. For pigeons in the mismatching condition, a peck to the stimulus that differed in hue from that of the sample was defined as correct. For each sample color, one of the other colors was chosen to be the nonmatching color. Thus, when the sample was red, the comparison stimuli were red and yellow. When the sample was yellow, the comparison stimuli were yellow and green. When the sample was green, the comparison stimuli were green and blue. When the sample was blue, the comparison stimuli were blue and red. The side of the matching stimulus was varied randomly over trials. Each session consisted of 72 trials. Each sample color occurred randomly, 18 times per session. Training consisted of 50 sessions.
Testing
All pigeons were presented with new-incorrect trials (32 per session) involving configurations in which the correct stimulus was replaced by one of the stimuli they had not seen before with that sample (e.g., for the matching task, a red sample was presented with red and blue stimuli or with red and green stimuli; for the mismatching task, a red sample was presented with blue and yellow stimuli or with green and yellow stimuli; see Fig. 1). They were also presented with new-correct trials (32 per session) involving configurations in which the correct stimulus was replaced by one of the stimuli they had not seen before with that sample (e.g., for the matching task, a red sample was presented with yellow and green stimuli or with yellow and blue stimuli; for the mismatching task, a red sample was presented with red and blue stimuli or with red and green comparison stimuli). During test sessions, responses to the correct stimulus from training or the hue that replaced the correct stimulus from training were reinforced. In all other respects, test trials were the same as training trials. The test session consisted of 64 trials.
Design of the experiment. In training (top row), each sample color appeared with one other color (counterbalanced for position); in this example, the stimulus on the left is correct in the matching condition, and the stimulus on the right is correct in the mismatching condition. On test trials (bottom four rows), one of the colors not used in training replaced either the matching or the mismatching color from training. In each condition, the correct stimulus appeared equally often on the left and right.
Results
As expected, in the matching condition, replacing the correct stimulus with a new but familiar stimulus resulted in a drop in accuracy to chance level, but replacing the incorrect stimulus with a new but familiar stimulus resulted in only a small drop in accuracy. In the mismatching condition, however, the reverse was true. Replacing the correct stimulus with a new but familiar stimulus resulted in only a small drop in accuracy, whereas replacing the incorrect (matching) stimulus with a new but familiar stimulus resulted in a drop in accuracy to chance level. These results, together with baseline accuracy from the last training session, are presented in Figure 2.
Proportion of correct responses on test trials as a function of trial type, separately for the matching and mismatching conditions. On new-correct trials, the correct comparison stimulus from training was replaced by one of the two colors not seen with that sample during training. On new-incorrect trials, the incorrect comparison stimulus from training was replaced by one of the two colors not seen with that sample during training. Baseline accuracy is presented for comparison. Error bars indicate standard errors of the mean.
A mixed-factor analysis of variance with condition as a between-subjects factor and trial type (new correct vs. new incorrect) as a within-subjects factor indicated that there was no effect of condition, F < 1, but there was a significant effect of trial type, F(2, 26) = 48.0, p < .0001, as well as a significant Condition × Trial Type interaction, F(2, 26) = 31.0, p < .0001. Of greatest interest is the interaction, because it indicates that although the basis of task learning by pigeons in the matching condition was the relation between the sample stimulus and the correct comparison stimulus, the basis of task learning by pigeons in the mismatching condition was the relation between the sample stimulus and the incorrect (matching) stimulus. The effect of trial type can be attributed to the drop in accuracy from baseline on test trials. That conclusion was confirmed by a mixed-factor analysis of variance, excluding baseline accuracy, which indicated that the effect of trial type was not statistically significant, F(1, 26) = 1.0, p = .33, but there was still a significant Condition × Trial Type interaction, F(1, 26) = 61.0, p < .0001.
More specifically, in the matching condition, planned comparisons indicated that accuracy on new-correct trials (M = .50) was significantly lower than accuracy on new-incorrect trials (M = .77), t(6) = 4.37, p = .04, Cohen’s d = 1.65 (exact probability = .008), whereas in the mismatching condition, accuracy on new-correct trials (M = .78) was significantly higher than accuracy on new-incorrect trials (M = .48), t(7) = 2.97, p = .02, Cohen’s d = 1.05 (exact probability = .004). These t tests were two-tailed planned comparisons.
Although accuracy on new-incorrect trials in the matching condition and new-correct trials in the mismatching condition was well above chance, it was not as high as on baseline trials. For the matching condition, the difference between baseline accuracy (M = .92) and accuracy on new-incorrect trials (M = .77) was significant, t(6) = 2.64, p = .04, Cohen’s d = 1.00, (exact probability = .23), as was the difference between baseline accuracy (M = .96) and accuracy on new-correct trials (M = .78) for the mismatching condition, t(7) = 2.77, p = .03, Cohen’s d = 0.98 (exact probability = .035). All t tests were two-tailed.
Discussion
According to traditional views of matching and mismatching (Cumming & Berryman, 1961; Skinner, 1950), only the correct comparison stimulus should enter into learning. The results of the present experiment demonstrate, however, that pigeons acquire both matching and mismatching by recognizing the similarity between the sample color and the color of the matching comparison stimulus. Further, if the task is matching, they choose the matching comparison stimulus, but if the task is mismatching, they avoid the matching comparison. But as the stimulus displays were exactly the same for both tasks, one can conclude that the concept of sameness was the basis for learning both the matching and mismatching tasks.
It should be noted that the pigeons also learned something about the nonmatching comparison stimulus, because when that stimulus was replaced, accuracy declined below that of the baseline level (see Fig. 2). Thus, even though the nonmatching comparison stimulus was not critical for correct comparison-stimulus choice, some learning about it occurred, and task accuracy declined when it was replaced.
A similar sameness predisposition was recently reported in human infants by Hochmann, Mody, and Carey (2016), using eye-tracker technology. In Experiment 1, infants trained on matching anticipated the location of a matching stimulus (having just seen a matching stimulus), whereas those trained on mismatching anticipated the location of a mismatching stimulus (having just seen a mismatching stimulus). However, anticipation of the matching stimulus occurred sooner than anticipation of the matching stimulus. In Experiment 2, infants were trained with animation of the matching or mismatching stimulus. With novel stimuli, those trained with animation of the matching stimulus looked more often at the location where they had previously seen a stimulus that matched the now-present sample (but not when they had previously seen a stimulus that did not match the now-present sample). Surprisingly, however, those trained with mismatching looked more often at the other location when they had previously seen a stimulus that matched the now-present sample (but did not look more often at the location where they had previously seen a stimulus that mismatched the now-present sample). Thus, for infants in both groups, sameness appeared to serve as a better cue than difference as to where to look to anticipate animation.
The present study provides the first clear results of the importance of the concept of sameness for pigeons while avoiding the possibility of stimulus generalization between the training and testing stimuli, as may have been present in studies that used complex photographs (Katz & Wright, 2006) or other complex images, such as icons (see Wasserman, Young, & Cook, 2004), as stimuli. However, the generality of these results may be limited by the species tested (pigeons) and by the stimuli used (colors). To the extent that the results of Hochmann et al. (2016) are an indication of the generality of this effect to young humans using very different procedures, it appears that sameness may be a basic concept that does not have to be learned and is likely to have evolved in many species, including humans.
Footnotes
Acknowledgements
This research was conducted under a protocol approved by the Institutional Animal Care and Use Committee.
Action Editor
Ralph Adolphs served as action editor for this article.
Author Contributions
T. R. Zentall developed the study concept. All authors contributed to the study design. Testing and data collection were performed by D. M. Andrews and J. P. Case. T. R. Zentall analyzed the data and drafted the manuscript. D. M. Andrews and J. P. Case provided revisions. All authors approved the final version of the manuscript for submission.
Declaration of Conflicting Interests
The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.
Open Practices
The data and materials have not been made publicly available, but detailed descriptions of them can be obtained from the first author. Data and analysis plans for this study were not preregistered.