The role of physical identity of the sample and correct comparison stimulus in matching-to-sample paradigms

Concept Learning and Learning Strategies

Concept learning and learning strategies of pigeons were manipulated in a matching-to-sample task. Groups of 4 pigeons responded either 0, 1, 10, or 20 times to a sample stimulus, and then chose between a matching comparison stimulus and a nonmatching comparison stimulus. Tests with unfamiliar arrangements of the three training stimuli showed that learning was not by if-then rules. Tests with novel stimuli showed that as the number of sample responses increased, learning about the configural pattern of each display gave way to more learning about the sample-comparison relationship and more concept learning. Pigeons making the most sample responses showed complete concept learning.

How special is sameness for pigeons and people?

Because of the importance of the sense of sameness for psychological science and because of the tenuous support for this notion in pigeons' matching-to-sample behavior, we experimentally explored the possibly special status of sameness for pigeons. Using photographs from three different natural categories (dogs, fish, and flowers) in a three-alternative matching-to-sample design, we obtained a reliable sameness advantage for pigeons only when the number of correct sample-comparison combinations could have contributed to a sameness advantage; otherwise, no sameness advantage emerged. However, human participants exhibited an immediate and dramatic sameness advantage under essentially the same training and testing conditions as had been given to pigeons. At least under these experimental circumstances, humans exhibit a sameness advantage that far eclipses that of pigeons.

Memory for recent behavior in the pigeon

Variations of the symbolic delayed-matching-to-sample procedure were used to study a pigeon's memory for a small number of pecks. In the first experiment a choice of a left or right sidekey after a delay or retention interval was reinforced if a bird had not pecked at all or had pecked exactly once, before the delay, respectively. In the second experiment a choice of a red or green sidekey, regardless of its position, was reinforced if a bird had not pecked at all or had pecked exactly twice, respectively. In the first experiment a bird could orient toward the correct choice during the delay, whereas it could not in the second experiment. In a third experiment a feature-probing method was used to study a pigeon's memory for a number of pecks in the context of certain other pecks. The results showed that a pigeon can remember a small number of pecks for one-half to one minute or more and that the percent correct is a decreasing function of the log retention interval. When a second number of pecks is different from the first number, memory for the first number lasts only a few seconds. When a second number is the same, memory lasts considerably longer. The more recent number of pecks is remembered better. The results are interpreted in terms of a theory which holds that a reinforcer, in general, may act on a subjects' memory for recent behavior to generate patterns of behavior.

Structure learning in action

'Learning to learn' phenomena have been widely investigated in cognition, perception and more recently also in action. During concept learning tasks, for example, it has been suggested that characteristic features are abstracted from a set of examples with the consequence that learning of similar tasks is facilitated-a process termed 'learning to learn'. From a computational point of view such an extraction of invariants can be regarded as learning of an underlying structure. Here we review the evidence for structure learning as a 'learning to learn' mechanism, especially in sensorimotor control where the motor system has to adapt to variable environments. We review studies demonstrating that common features of variable environments are extracted during sensorimotor learning and exploited for efficient adaptation in novel tasks. We conclude that structure learning plays a fundamental role in skill learning and may underlie the unsurpassed flexibility and adaptability of the motor system.

Acquisition of matching-to-sample performance in rats using visual stimuli on nose keys

Steady and blinking white lights were projected on three nose keys arranged horizontally on one wall. The procedure was a conditional discrimination with a sample stimulus presented on the middle key and comparison stimuli on the side keys. Three rats acquired simultaneous "identity matching." Accuracy reached 80% in about 25 sessions and 90% or higher after about 50 sessions. Acquisition progressed through several stages of repeated errors, alteration between comparison keys from trial to trial, preference of specific keys or stimuli, and a gradual lengthening of strings of consecutive trials with correct responses. An analysis of the acquisition curves for individual trial configurations indicated that the matching-to-sample performance possibly consisted of separate discriminations.

Relational and absolute stimulus learning by monkeys in a memory task

Three experiments showed stimulus control by either the absolute properties of probe stimuli, relational properties of the probe-list relationship, or both in a serial probe recognition memory task in which a four-item memory list was followed by a single probe (test) item. In Experiment 1, 3 rhesus monkeys received 39 to 75 repetitions of the same 24-trial stimulus sequence. Special tests showed stimulus control by the absolute properties of the probe stimuli. Retention of previous relational control was demonstrated by the good transfer (83%) to novel list and probe stimuli at the beginning of Experiment 2. During Experiment 2, control by absolute properties of the probe stimuli gradually reoccurred. Only a small measure of control by list stimuli could be detected or promoted. In Experiment 3, 4 monkeys were shown to have largely lost their ability to perform on the basis of the list-probe relationship, and were performing primarily on the basis of the absolute properties of the probe stimuli. Over the next 15 weeks, these monkeys were transferred to new stimuli at the beginning of each week. Control by the relational aspects of the task gradually returned. As transfer performance increased, control by the absolute properties of the probe stimuli was eliminated. The results are discussed in terms of stimulus control and performance strategies used by the monkeys.

Use of an ambiguous-sample procedure to establish a cue to forget in pigeons

Pigeons were trained on a variation of the matching-to-sample task in which on double-sample trials two samples, one associated with each of the comparison stimuli, were presented successively. Responding to the comparison associated with the first sample was reinforced on half the double-sample trials, and responding to the comparison associated with the second sample was reinforced on the remaining half. One of two postsample stimuli was presented following the termination of each colored sample. A vertical line was presented after a correct or target sample, and a horizontal line was presented after an incorrect or interfering sample. With extended training, each bird demonstrated above-chance accuracy on double-sample trials, providing prima facie evidence that one or both of the postsample stimuli exerted control over matching behavior. Experiment 2 provided evidence that the horizontal line functioned as a cue to forget the code activated by the preceding sample stimulus. It was concluded that a condition sufficient to establish a postsample stimulus as a cue to forget is that the postsample immediately follow presentation of a sample that, if it were to control test responding, would lead to nonreinforcement.

Contextual control of emergent equivalence relations

Three college students in Experiment 1 and 1 student in Experiment 2 learned visual conditional discriminations under contextual control by tones; the visual comparison stimulus that was correct with a given sample stimulus depended on whether a high tone or a low tone was present. Two of the subjects in Experiment 1 then demonstrated the emergence of two sets of contextually controlled three-member classes of equivalent stimuli, and the subject in Experiment 2 showed the emergence of contextually controlled four-member classes; the class membership of each stimulus varied as a function of the tones. Class membership was demonstrated by the subjects' performance of new conditional discriminations that they had never been taught directly. In Experiment 2, the procedures were intended to ensure that the tones exerted second-order conditional control and did not simply form compounds with each of the visual stimuli, but the subject's verbal description of the tasks suggested that this intention might not have been successful. It could not be ascertained, therefore, whether the tones exerted contextual control as independent second-order conditional stimuli or simply as common elements of auditory-visual stimulus compounds.

The formation of visual stimulus equivalences in children

Four normal children were presented a series of matching-to-sample tasks, using five sets of visual stimuli designated A, B, C, D, and E. Stimulus equivalences were established by matching stimuli from one set to those from another set. Each set consisted of three stimuli, so matching set A to set D meant that each stimulus in set A served as a sample with all three stimuli in set D as comparisons. Subjects were first taught AD and DC matching and were then able to perform AC/CA matching without additional training. After ED was taught directly, CE/EC and AE/EA performances emerged. Following CB training, three new equivalences were demonstrated: AB/BA, EB/BE, and DB/BD. Oral naming of each stimulus showed that subjects had not assigned a common label to stimuli in the same class, indicating that naming is not necessary for the formation of stimulus equivalences. The absence of response mediation suggests that matching to sample can form direct stimulus-stimulus associations. The data also provide support for the notion that generative performances are outcomes of existing stimulus-control relationships.