A case for restricted-domain relational learning

Monkeys and pigeons learned a same/different task with pairs that were selected from a training set of eight picture stimuli. They showed no novel-stimulus transfer and hence no abstract-concept learning. They were also tested with novel pairs of the eight training pictures (i.e., combinations that had not been used in training) and with inverted pictures of the training pairs. If the subjects had learned the task item-specifically (e.g., if—then or configural learning), they should have failed these tests, but they performed well with novel combinations of training pictures and inverted pictures, suggesting that they learned the task relationally (i.e., on the basis of the relationship between the two pictures that were presented in each trial). This somewhat paradoxical conclusion of relational learning in the absence of abstract-concept learning is contrary to most theories of abstract-concept learning. The implications of this conclusion are discussed in the context of restricted-domain relational learning.     

Same/different abstract-concept learning by pigeons

Eight pigeons were trained and tested in a simultaneous same/different task. After pecking an upper picture, they pecked a lower picture to indicate same or a white rectangle to indicate different. Increases in the training set size from 8 to 1,024 items produced improved transfer from 51.3% to 84.6%. This is the first evidence that pigeons can perform a two-item same/different task as accurately with novel items as training items and both above 80% correct. Fixed-set control groups ruled out training time or transfer testing as producing the high level of abstract-concept learning. Comparisons with similar experiments with rhesus and capuchin monkeys showed that the ability to learn the same/different abstract concept was similar but that pigeons require more training exemplars.     

Matching-to-sample abstract-concept learning by pigeons

Abstract concepts--rules that transcend training stimuli--have been argued to be unique to some species. Pigeons, a focus of much concept-learning research, were tested for learning a matching-to-sample abstract concept. Five pigeons were trained with three cartoon stimuli. Pigeons pecked a sample 10 times and then chose which of two simultaneously presented comparison stimuli matched the sample. After acquisition, abstract-concept learning was tested by presenting novel cartoons on 12 out of 96 trials for 4 consecutive sessions. A cycle of doubling the training set followed by retraining and novel-testing was repeated eight times, increasing the set size from 3 to 768 items. Transfer performance improved from chance (i.e., no abstract-concept learning) to a level equivalent to baseline performance (>80%) and was similar to an equivalent function for same/different abstract-concept learning. Analyses assessed the possibility that item-specific choice strategies accounted for acquisition and transfer performance. These analyses converged to rule out item-specific strategies at all but the smallest set-sizes (3-24 items). Ruling out these possibilities adds to the evidence that pigeons learned the relational abstract concept of matching-to-sample.     

Learning and transfer of relational matching-to-sample by pigeons

We trained pigeons on a relational matching-to-sample task to see whether a nonprimate species can discriminate higher-order “relations between relations.” We required the birds to relationally match arrays of 16 items that were chosen from five nonoverlapping sets of 20 colored computer icons. On each trial, randomly selected icons from one set were placed into a 4×4 grid to form a sample; onsame trials, all 16 icons were identical to each other, whereas ondifferent trials, all 16 icons were different from each other. After 10-20 pecks, 16-itemsame anddifferent testing arrays were presented that were created from an entirely different icon set. Because no icons were common to the sample and testing arrays, discriminating higher-order relations was required for success on the tests. As have primates in similar tasks, pigeons successfully learned and transferred this relational discrimination, suggesting that both birds and mammals possess the cognitive antecedents of analogical reasoning.     

Abstract-concept learning in Black-billed magpies (<Emphasis Type="Italic">Pica hudsonia</Emphasis>)

Abstract relational concepts depend upon relationships between stimuli (e.g., same vs. different) and transcend features of the training stimuli. Recent evidence shows that learning abstract concepts is shared across a variety species including birds. Our recent work with a highly-skilled food-storing bird, Clark’s nutcracker, revealed superior same/different abstract-concept learning compared to rhesus monkeys, capuchin monkeys, and pigeons. Here we test a more social, but less reliant on food-storing, corvid species, the Black-billed magpie (Pica hudsonia). We used the same procedures and training exemplars (eight pairs of the same rule, and 56 pairs of the different rule) as were used to test the other species. Magpies (n = 10) showed a level of abstract-concept learning that was equivalent to nutcrackers and greater than the primates and pigeons tested with these same exemplars. These findings suggest that superior initial abstract-concept learning abilities may be shared across corvids generally, rather than confined to those strongly reliant on spatial memory.     

Memory-basedsame-different conceptualization by pigeons

We first trained pigeons to peck one button (same) after the successive presentation of 16 identical pictures and to peck a second button (different) after the successive presentation of 16 nonidentical pictures. Later, we tested the birds with other lists ofsame anddifferent items composed of completely novel pictures. Accuracy to the testing lists reliably exceeded chance levels, thus demonstratingsame-different conceptualization by pigeons under conditions that, for the first time, (1) eliminated purely perceptual mechanisms of discrimination learning and transfer and (2) required memory-based processing of the experimental stimuli.     

Matching and oddity relational learning by pigeons (<Emphasis Type="Italic">Columba livia</Emphasis>): transfer from color to shape

Relational learning, as opposed to perceptual learning, is based on the abstract properties of the stimuli. Although at present there is no doubt that pigeons are capable of relational behavior, this study aims to further disclose the conditions under which it occurs. Pigeons were trained in an outdoor cage on a matching-to-sample or an oddity-from-sample task, with colored cardboard stimuli presented horizontally. The apparatus involved three sliding lids on which the stimuli were drawn and which, when displaced, revealed the reinforcement. The lids were either adjacent to each other or somewhat separated. Training sessions involved two colors, and test sessions six different colors (same dimension test), or six different shapes (different dimension test). One group of birds trained under the ‘adjacent’ condition failed when tested with new stimuli, but succeeded in both dimension tests after training under the ‘separate’ condition. Two other groups of birds succeeded in all tests after training under the latter condition. These results show that depending on procedural details, pigeons are or are not able to transfer from one visual dimension to another, thus extending previous related findings.     

Generalization hypothesis of abstract-concept learning: learning strategies and related issues in Macaca mulatta, Cebus apella, and Columba livia

The generalization hypothesis of abstract-concept learning was tested with a meta-analysis of rhesus monkeys (Macaca mulatta), capuchin monkeys (Cebus apella), and pigeons (Columba livia) learning a same/different (S/D) task with expanding training sets. The generalization hypothesis states that as the number of training items increases, generalization from the training pairs will increase and could explain the subjects' accurate novel-stimulus transfer. By contrast, concept learning is learning the relationship between each pair of items; with more training items subjects learn more exemplars of the rule and transfer better. Having to learn the stimulus pairs (the generalization hypothesis) would require more training as the set size increases, whereas learning the concept might require less training because subjects would be learning an abstract rule. The results strongly support concept or rule learning despite severely relaxing the generalization-hypothesis parameters. Thus, generalization was not a factor in the transfer from these experiments, adding to the evidence that these subjects were learning the S/D abstract concept.     

Mechanisms of same/different abstract-concept learning by rhesus monkeys (Macaca mulatta)

Experiments with 9 rhesus monkeys (Macaca mulatta) showed, for the first time, that abstract-concept learning varied with the training stimulus set size. In a same/different task, monkeys required to touch a top picture before choosing a bottom picture (same) or white rectangle (different) learned rapidly. Monkeys not required to touch the top picture or presented with the top picture for a fixed time learned slowly or not at all. No abstract-concept learning occurred after 8-item training but progressively improved with larger set sizes and was complete following 128-item training. A control monkey with a constant 8-item set ruled out repeated training and testing. Contrary to the unique-species account, it is argued that different species have quantitative, not qualitative, differences in abstract-concept learning.     

Concept learning by monkeys with video picture images and a touch screen.

Two rhesus monkeys were trained in a same/different task to discriminate digitized computer-stored picture stimuli. The pictures were digitized from 35-mm slides and presented in pairs on a computer monitor. The monkeys were required to touch the pictures and then make a choice response to indicate whether the pictures were identical or nonidentical. The response areas and stimuli were located to the sides of the picture stimuli. Responses were defined and monitored by an infrared matrix touch screen. After learning the same/different task, both monkeys showed performance accuracy with novel picture stimuli similar to that with training picture stimuli. This accurate novel-picture transfer indicates that a same/different concept had been learned, a concept similar to the one they had previously demonstrated in a different apparatus with rear-projected slide stimuli and a response lever.     

Development of relational learning: Effects of instruction and delay of transfer

Preschool children were trained on a color relevant oddity problem by one of three methods: increased salience of the oddity relationship, instruction on the solution rule, or a combination of increased salience and rule instruction. The youngest Ss solved the training problem when the solution rule was provided but not under the salience procedure. The two older groups solved in all conditions and errors decreased with age. The generality of the solution was determined by performance on a form relevant transfer task. The youngest Ss made fewer errors on the transfer problem following rule instruction training than following training in the combined procedure. The intermediate age group made fewer errors when the transfer task was administered one week after training than one minute after training. The transfer problem was easily solved by the oldest Ss and performance was independent of training procedure and amount of delay. These results were interpreted as reflecting developmental differences in relational concept learning.     

The structure of pigeon multiple-class same-different learning

Three experiments examined the structure of the decision framework used by pigeons in learning a multiple-class same-different task. Using a same-different choice task requiring the discrimination of odd-item different displays (one or more of the display's component elements differed) from same displays (all display components identical), pigeons were concurrently trained with sets of four discriminable display types. In each experiment, the consistent group was tested such that the same and different displays of four display types were consistently mapped onto their choice alternatives. The inconsistent group received a conflicting mapping of the same and different displays and the choice alternatives that differed across the four display types but were consistent within a display type. Experiment 1 tested experienced pigeons, and Experiment 2 tested naive pigeons. In both experiments, the consistent group learned their discrimination faster and to a higher level of choice accuracy than did the inconsistent group, which performed poorly in general. Only in the consistent group was the discrimination transferred to novel stimuli, indicative of concept formation in that group. A third experiment documented that the different display classes were discriminable from one another. These results suggest that pigeons attempt to generate a single discriminative rule when learning this type of task, and that this general rule can cover a large variety of stimulus elements and organizations, consistent with previous evidence suggesting that pigeons may be capable of learning relatively unbounded relational same-different concepts.     

What is learned when concept learning fails?—A theory of restricted-domain relational learning

Two matching-to-sample (MTS) and four same/different (S/D) experiments employed tests to distinguish between item-specific learning and relational learning. One MTS experiment showed item-specific learning when concept learning failed (i.e., no novel-stimulus transfer). Another MTS experiment showed item-specific learning when pigeons’ novel-stimulus transfer decreased because they chose familiar training comparisons instead of matching novel comparisons. In 8-item and 3-item S/D tasks, pigeons and monkeys were accurate with unfamiliar training-stimulus pairings, stimulus inversions, and distorted stimuli, suggesting relational learning within a domain restricted to the training stimuli (i.e., no novel-stimulus transfer). In 32-item S/D tasks, pigeons with previous 8-item training showed less transfer than those without prior training, suggesting a carryover of restricted-domain relational learning. Pigeons shifted from 1024-item to 8-item S/D tasks showed reinstatement of restricted-domain relational learning. These findings are important in specifying which types of learning occur in these tasks, showing that subjects failing novel-stimulus transfer are not required to switch from item-specific to relational learning as a training set is expanded, and demonstrating that concept learning failure is not proof of item-specific learning.     

Pigeons can discriminate locations presented in pictures

The present experiments were designed to teach pigeons to discriminate two locations represented by color photographs. Two sets of photographs were taken at two distinctive locations on a university campus. These sets represented several standpoints at each location. For the true-discrimination group, pictures from the two locations were differentially associated with reward; for the pseudodiscrimination group, half of the views from each location were arbitrarily but consistently associated with reward. The former group acquired the discrimination much more rapidly. These birds also showed good transfer to new views from the standpoints used in training and to a new standpoint at each location not used in training. In a second experiment, another group of pigeons could terminate any training trial by pecking an “advance” key. Three of 4 subjects used this option to reduce the duration of trials in which pictures from the negative location were presented. These data suggest that pigeons can integrate views shown in pictures into a “concept” of a location. The method used here may be the experimental analogue of a common, natural process by which animals learn to identify locations.     

Perceptual identification thresholds for 150 fragmented pictures from the Snodgrass and Vanderwart picture set

This paper reports perceptual identification thresholds for 150 pictures from the 1980 Snodgrass and Vanderwart picture set. These pictures were fragmented and presented on the Apple Macintosh microcomputer in a picture-fragment completion task in which identification thresholds were obtained at three phases of learning: Train (initial presentation), New (initial presentation after training on a different set), and Old (repeated presentation of the Train set). Pictures were divided into five sets of two subsets of 15 pictures each, which served alternately as the Train and New sets. A total of 100 subjects participated in the task, with 10 subjects assigned to each subset. Individual thresholds for each picture at each phase of learning are presented, along with the fragmented pictures identified by 35% of the subjects across the Train and New learning phases. This set of fragmented pictures is provided for use in experiments in which a single level of fragmented image is presented for identification after a priming phase. Correlations between the Snodgrass and Vanderwart norms and identification thresholds at the three phases of learning are also reported.     

Testing complex animal cognition: Concept learning,proactive interference,and list memory

This article describes an approach for assessing and comparing complex cognition in rhesus monkeys and pigeons by training them in a sequence of synergistic tasks, each yielding a whole function for enhanced comparisons. These species were trained in similar same/different tasks with expanding training sets (8, 16, 32, 64, 128 … 1024 pictures) followed by novel‐stimulus transfer eventually resulting in full abstract‐concept learning. Concept‐learning functions revealed better rhesus transfer throughout and full concept learning at the 128 set, versus pigeons at the 256 set. They were then tested in delayed same/different tasks for proactive interference by inserting occasional tests within trial‐unique sessions where the test stimulus matched a previous sample stimulus (1, 2, 4, 8, 16 trials prior). Proactive‐interference functions revealed time‐based interference for pigeons (1, 10 s delays), but event‐based interference for rhesus (no effect of 1, 10, 20 s delays). They were then tested in list‐memory tasks by expanding the sample to four samples in trial‐unique sessions (minimizing proactive interference). The four‐item, list‐memory functions revealed strong recency memory at short delays, gradually changing to strong primacy memory at long delays over 30 s for rhesus, and 10 s for pigeons. Other species comparisons and future directions are discussed.     

Abstract-concept learning and list-memory processing by capuchin and rhesus monkeys

Three capuchin monkeys (Cebus apella) touched the lower of 2 pictures (same) or a white rectangle (different), increased same/different abstract-concept learning (52% to 87%) with set-size increases (8 to 128 pictures), and were better than 3 rhesus monkeys (Macaca mulatta). Three other rhesus that touched the top picture before choices learned similar to capuchins but were better at list-memory learning. Both species' serial position functions were similar in shape and changes with retention delays. Other species showed qualitatively similar shape changes but quantitatively different time-course changes. In abstract-concept learning, qualitative similarity was shown by complete concept learning, whereas a quantitative difference would have been a set-size slope difference. Qualitative similarity is discussed in relation to general-process versus modular cognitive accounts.     

Transfer of relational rules in matching and oddity learning by pigeons and corvids

Three experiments compared the performance of pigeons and corvids when they were given the opportunity to transfer the relational rule underlying matching or oddity discriminations to new sets of stimuli. In the first, pigeons and jackdaws were initially trained either on a matching or on a non-relational conditional discrimination and then transferred to a new matching discrimination. In the second, pigeons and jays were trained on a series of three matching (or oddity) discriminations with three different pairs of colours and finally tested, either with the same or the reversed rule, on matching or oddity to line orientations. In the third, pigeons and rooks were trained to perform one response when two coloured panels were the same and a different response when the two colours were different and then transferred, either with the same or the reversed rule, to a new set of colour stimuli. All three experiments produced the same result: no evidence of transfer of the relational rule by pigeons, but substantial and significant transfer by corvids.     

Failure to find evidence of stimulus generalization within pictorial categories in pigeons

Pigeons' key pecks were reinforced in the presence of pictures from one of two categories, cats or cars. A single picture associated with reinforcement was used in Experiment 1, and 20 pictures from the same category were associated with reinforcement in Experiment 2. Pigeons then were presented with novel test pictures from the training category and from the other, previously unseen, category. During Session 1 of testing, pigeons pecked no more often at pictures from the reinforced category than at pictures from the previously unseen category. When pigeons were trained with pictures associated with reinforcement or its absence from different categories in Experiment 3, differential responding to novel pictures from different categories appeared during Session 1. These findings argue against a process of automatic stimulus generalization within natural categories and in favor of the position that category distinctions are not made until members of at least two categories are compared with one another.     

Evidence for a conceptual account of same-different discrimination learning in the pigeon

We trained pigeons to peck two different buttons in response to 16-iconsame arrays versus 16-icondifferent arrays. In thesame arrays, the icons were all the same as one another, whereas in thedifferent arrays, the icons were all different from one another. In Experiment 1, we upset the spatial regularities of the displays by disarranging the icons—randomly displacing each icon to reduce the degree of perceptual order. The pigeons’ discriminative performance was unaffected by disarranging. In Experiment 2, spatial regularities were disturbed by varying the rotation of the icons within a display. Again, no disruption in discriminative performance was observed. These and other findings suggest that pigeons treat the 16 icons as either the same or different despite changes in the spatial organization or orientation of the icons, thus implicating a conceptual rather than a perceptual process in same—different discrimination.