Food delivery as a conditional stimulus: Feature-learning and memory in pigeons

Three experiments investigated the learning and memory of discriminations based on presence versus absence of a pre-trial food delivery. In Experiment 1 half the illuminations of a response key were followed by food regardless of the subject's behavior. In one group an extra food delivery preceded only reinforced trials (feature-positive condition), whereas in a second group it preceded only nonreinforced trials (feature-negative condition). Key pecks and approaches revealed more rapid and superior discrimination learning in the first group. Experiment 2 replicated the results of Experiment 1 but yielded no evidence that greater “unexpectedness” of pretrial food conditions facilitates discriminative performance. In Experiment 3, individual pigeons trained on a conditional discrimination exhibited a within-subject feature-positive superiority. Delay between pretrial and trial stimuli interacted with feature-positive versus feature-negative training in both the between-group (Experiment 2) and within-subject (Experiment 3) procedures: performance was decremented at both short and long delays in the feature-positive condition but was decremented only at longer delays in the feature-negative condition. The feature-positive superiority obtained here is incompatible with explanations based on either the general concept of “perceptual organization” or on the conditional nature of feature-negative discriminations.     

The effects of number of sample stimuli and number of choices in a detection task on measures of discriminability

Six pigeons were trained on a conditional discrimination task involving the discrimination of various intensities of yellow light. The research asked whether stimulus—response discriminability measures between any pair of stimuli would remain constant when a third or fourth sample and reinforced response were added. The numbers of different sample stimuli presented and different responses reinforced were two (Part 1), three (Parts 2 and 4), and four (Part 3). Across conditions within parts, the ratios of reinforcers obtainable for correct responses were varied over at least five levels. In Part 5, the numbers of sample stimuli and reinforced responses were varied among two, three, and four, and the reinforcer ratio between consecutive remaining samples was constant at 2:1. It was found that once a particular response had been reinforced, subjects continued to emit that response when the conditional stimulus for that response was no longer presented. Data analysis using a generalization-based detection model indicated that this model was able to describe the data effectively. Four findings were in accord with the theory. First, estimates of stimulus—response discriminability usually decreased as the arranged physical disparity between the sample stimuli decreased. Second, stimulus—response discriminability measures were independent of response—reinforcer discriminability measures, preserving parameter invariance between these measures. Third, stimulus—response discriminability measures for constant pairs of conditional stimuli did not change systematically as conditional stimulus—response alternatives were added. Fourth, log stimulus—response discriminability values between physically adjacent conditional stimuli summed to values that were not significantly different from estimates of the discriminability values for conditional stimuli that were spaced further apart.     

Psychophysics of remembering

We present a new model of remembering in the context of conditional discrimination. For procedures such as delayed matching to sample, the effect of the sample stimuli at the time of remembering is represented by a pair of Thurstonian (normal) distributions of effective stimulus values. The critical assumption of the model is that, based on prior experience, each effective stimulus value is associated with a ratio of reinforcers obtained for previous correct choices of the comparison stimuli. That ratio determines the choice that is made on the basis of the matching law. The standard deviations of the distributions are assumed to increase with increasing retention-interval duration, and the distance between their means is assumed to be a function of other factors that influence overall difficulty of the discrimination. It is a behavioral model in that choice is determined by its reinforcement history. The model predicts that the biasing effects of the reinforcer differential increase with decreasing discriminability and with increasing retention-interval duration. Data from several conditions using a delayed matching-to-sample procedure with pigeons support the predictions.     

Transfer of relational stimulus control in conditional discriminations.

Four adults were trained, using instructions and a matching-to-sample procedure, to select Stimulus B1 in the presence of Stimulus A1, B2 in the presence of A2, and B3 in the presence of A3 (the AB relations). Analogous PQ relations were trained. Afterwards, one stimulus in Set A and another stimulus in Set B appeared together as a sample, and novel Stimuli X1 and X2 were the comparisons. Responses to X1 were reinforced if the two stimuli in the sample had been related in the previous training (e.g., A1 and B1), and responses to X2 were reinforced if the two samples had not been related (e.g., A1 and B2). These were the ABX relations. In a test in which a stimulus of Set P and another of Set Q were the samples and X1 and X2 were the comparisons, 2 subjects selected X1 when the samples were P1 and Q1, P2 and Q2, and P3 and Q3, and selected X2 in the presence of the other six sample combinations (P1Q2, P1Q3, P2Q1, P2Q3, P3Q1, and P3Q2). Another subject showed the same responding after additional training. In the second experiment, 3 adults and an 11-year-old child were trained on AB, PQ, and ABX relations, and they showed the symmetrical relations BA and QP upon testing. Then all 4 of these subjects responded accurately to the PQX test. Results of Experiments 1 and 2 showed novel, consistent comparison selection based on the previously established relation between the two stimuli in the sample. In a third experiment, 3 of the subjects who had shown PQX relations were trained on EFX relations, with pairs of E and F stimuli as samples and X stimuli as comparisons. When the EF relations were tested, all 3 subjects consistently selected F1 in the presence of E1, F2 in the presence of E2, and F3 in the presence of E3 from the first trial. The results of Experiment 3 showed novel stimulus relations after training with a more complex conditional discrimination format.     

Context specificity of conditioned-reinforcement effects on discrimination acquisition

Pigeons were trained on a series of reversals of a simultaneous form discrimination in which the trial outcomes were separated from the choice responses by an 8-s delay interval. Different conditions were defined by the stimuli occurring during the two halves of the delay interval. Discrimination learning was greatly facilitated by having differential stimuli during the delay following correct versus incorrect choices. When the differential stimuli appeared only at the midpoint of the delay, some facilitation occurred relative to when no different stimuli occurred, but there was substantially less facilitation than when the differential stimuli occurred immediately contingent on choice. A reversed-stimulus condition, in which the stimulus at the onset of the delay following a correct choice was the same as that during the last segment of the delay following an incorrect choice, and the stimulus at the onset of the delay following an incorrect choice was the same as that preceding food during the last segment of the delay following a correct choice, also facilitated discrimination learning relative to the nondifferential stimulus conditions.     

Conditional discrimination and equivalence relations: Control by negative stimuli

Three adult subjects were taught the following two-sample, two-comparison conditional discriminations (each sample is shown with its positive and negative comparison, in that order): A1-B1B2, A2-B2B1; B1-C1C2, B2-C2C1; and C1-D1D2, C2-D2D1. A teaching procedure was designed to encourage control by negative comparisons. Subjects were then tested for emergent performances that would indicate whether the baseline conditional discriminations were reflexive, symmetric, and transitive. The tests documented the emergence of two classes of equivalent stimuli: A1, B2, C1, D2 and A2, B1, C2, D1. These were the classes to be expected if the negative comparisons were the controlling comparisons in the baseline conditional discriminations. The negative comparisons, however, were not the comparisons that subjects were recorded as having chosen in the baseline conditional discriminations. Differential test results confirmed predictions arising from a stimulus-control analysis: In reflexivity tests (AA, BB, CC, DD), subjects chose comparisons that differed from the sample; one-node transitivity (AC, BD) and "equivalence" (CA, DB) tests also yielded results that were the opposite of those to be expected from control by positive comparisons; symmetry tests (BA, CB, DC), two-node transitivity (AD) tests, and two-node "equivalence" (DA) tests yielded results that were to be expected from control by either positive or negative comparisons.     

TESTS FOR CONTROL BY EXCLUSION AND NEGATIVE STIMULUS RELATIONS OF ARBITRARY MATCHING TO SAMPLE IN A “SYMMETRY-EMERGENT” CHIMPANZEE

In the present experiments, controlling relations in arbitrary matching-to-sample performance were tested in a 9-year-old female chimpanzee who showed statistically significant emergence of symmetry in previous two-choice conditional discrimination experiments. In Experiment 1, a novel (undefined) sample stimulus was followed by a pair of trained (defined) and undefined comparison stimuli to assess the control by exclusion in arbitrary matching. The chimpanzee selected the undefined shape comparison, excluding the defined one, in color-sample-to-shape-comparison probe trials, although stimulus preferences were relatively stronger than control by exclusion in shape-sample trials. An additional test for control by relations of the sample to the positive comparison (S+ control) showed that her behavior was also under the control of relations of the sample to the positive comparison. In Experiment 2, a defined sample was followed by a pair of negatively defined and undefined comparisons to test control by the relations of the sample to the negative comparison. (S- control). The subject selected undefined comparisons in both color-shape and shape-color test trials. These results clearly indicate that the conditional discrimination behavior of this “symmetry-emergent” chimpanzee was under both S+ and S- control. Furthermore, her performance was also under control by exclusion in color-shape arbitrary matching, unlike other chimpanzees who showed no evidence of symmetry but only S+ control of arbitrary matching.     

“TRANSITIVE INFERENCE” IN MULTIPLE CONDITIONAL DISCRIMINATIONS

We used multiple conditional discriminations to study the inferential abilities of pigeons. Using a five-term stimulus series, pigeons were trained to respond differentially to four overlapping pairs of concurrently presented stimuli: A+B?, B+C?, C+D?, and D+E?, where plus and minus indicate the stimulus associated with reinforcement and extinction, respectively. Transitive inference in such situations has been defined as a preference for Stimulus B over Stimulus D in a transfer test. We measured this and other untrained preferences (A vs. C, A vs. D, B vs. E, etc.) during nonreinforced test trials. In three experiments using a novel, rapid training procedure (termed autorun), we attempted to identify the necessary and sufficient conditions for transitive inference. We used two versions of autorun: response-based, in which the subject was repeatedly presented with the least well-discriminated stimulus pair; and time-based, in which the subject was repeatedly presented with the least-experienced stimulus pair. In Experiment 1, using response-based autorun, we showed that subjects learned the four stimulus pairs faster than, but at a level comparable to, a previous study on transitive inference in pigeons (Fersen, Wynne, Delius, & Staddon, 1991), but our animals failed to show transitive inference. Experiments 2 and 3 compared time- and response-based autorun. Discrimination performance was maintained, but transitive inference was observed only on the second exposure to the response-based procedure. These results show that inferential behavior in pigeons is not a reliable concomitant of good performance on a series of overlapping discriminations. The necessary and sufficient conditions for transitive inference in pigeons remain to be fully defined.     

Conditional discrimination in mentally retarded subjects: programming acquisition and learning set.

In Experiment 1, 3 subjects with retardation were exposed to two visual-visual arbitrary matching-to-sample problems each day. One conditional discrimination was presented under trial-and-error conditions, and the other was presented under a component training procedure. The latter began by establishing the comparison discrimination and its rapid reversal. The successive discrimination between the sample stimuli was established through differential naming. Then, sample naming was maintained in conditional discrimination sessions in which the same sample was presented in blocks of consecutive trials. Block size was decreased across sessions until sample presentation was randomized as in trial-and-error training (but with naming maintained). Two subjects initially learned only with component training. The performance of the 3rd subject was inconsistent across conditional discriminations. One of the successful subjects ultimately learned rapidly and consistently with trial-and-error procedures. Experiment 2 sought to demonstrate learning set in the other 2 subjects. Elements of the component training procedure were withdrawn over successive conditional discriminations. Ultimately, 1 subject nearly always learned under trial-and-error conditions, and the other learned under trial-and-error conditions combined with differential sample naming.     

Penalization approaches in the conditional maximum likelihood and Rasch modelling context

Recent detection methods for Differential Item Functioning (DIF) include approaches like Rasch Trees, DIFlasso, GPCMlasso and Item Focussed Trees, all of which - in contrast to well established methods - can handle metric covariates inducing DIF. A new estimation method shall address their downsides by mainly aiming at combining three central virtues: the use of conditional likelihood for estimation, the incorporation of linear influence of metric covariates on item difficulty and the possibility to detect different DIF types: certain items showing DIF, certain covariates inducing DIF, or certain covariates inducing DIF in certain items. Each of the approaches mentioned lacks in two of these aspects. We introduce a method for DIF detection, which firstly utilizes the conditional likelihood for estimation combined with group Lasso-penalization for item or variable selection and L1-penalization for interaction selection, secondly incorporates linear effects instead of approximation through step functions, and thirdly provides the possibility to investigate any of the three DIF types. The method is described theoretically, challenges in implementation are discussed. A dataset is analysed for all DIF types and shows comparable results between methods. Simulation studies per DIF type reveal competitive performance of cmlDIFlasso, particularly when selecting interactions in case of large sample sizes and numbers of parameters. Coupled with low computation times, cmlDIFlasso seems a worthwhile option for applied DIF detection.