Flexible Configural Learning of Non-Linear Discriminations and Detection of Stimulus Compounds

Abstract

Previous work showed that prior experience with discriminations requiring configural solutions (e.g. biconditional discrimination) confers an advantage for the learning of new configural discriminations (e.g. negative patterning) in comparison to prior experience with elemental discriminations. This effect is well established but its mechanism is not well understood. In the studies described below we assessed whether the saliences of configural and element cues were affected by prior training. We observed positive transfer to a new configural discrimination after configural pre-training but we were unable to find evidence for changes in cue salience using a signal-detection task. Our results confirm previous work by demonstrating experience dependent flexibility in cue processing but they also suggest that this flexibility occurs at a point in the stimulus processing pipeline later than 1-2s after the presentation of stimulus inputs. Successful learning implies acquisition of specific responses to specific stimulus situations but may also involve acquisition of general strategies which can be deployed in new tasks. For example, better discrimination performance after an intra-dimensional, rather than after an extra-dimensional shift condition (e.g. Roberts, Robbins, & Everitt, 1988) suggests that something is learned about the stimulus dimensions and this generalises between tasks. Intra-dimensional shift superiority could involve enhanced attention to the relevant dimension, ignoring the irrelevant dimension, or a combination of both. In a similar vein Alvarado and Rudy (1992) found evidence for changes in configural information processing with transverse patterning problems in rats. In transverse patterning a set of simple simultaneous discriminations is learned.For example, whenever cues A and B are presented reinforcement is obtained for selection of A (A+/B-trials). Whenever cues B and C are presented reinforcement is contingent on selection of B (B+/C-trials). Finally, whenever cues C and A are presented, reinforcement is given for selection of C (C+/A-trials). Although these discriminations are individually simple, successful performance in transverse patterning cannot be explained by simple elemental models of associative learning such as the Rescorla-Wagner Model (RWM, Rescorla & Wagner, 1972). In such models the associative strengths of stimuli A, B, and C would be the same due to the fact that they each occur on reinforced and non-reinforced trials equally often. In contrast, to explain performance in transverse patterning, it seems necessary for subjects to process the stimulus configurations (e.g. Pearce, 1987Pearce, , 1994Rescorla, 1973;) that are present on each trial in order to select the appropriate response.However, Alvarado and Rudy's contribution was not the demonstration that rats could solve transverse patterning problems per se. It was the demonstration that components of transverse patterning problems could be solved more readily after 'configural' than after 'elemental' pre-trai...

Fluid Abilities and Rule Learning: Patterning and Biconditional Discriminations

“…Similar to previous studies [12,13,26,30], our results suggest that configural processing is flexible and increases after exposure to configural discriminations. This flexibility is thought to be advantageous, as it allows individuals to adapt their processing style depending on which strategy is more useful for learning contingencies between stimuli in the environment.…”

“…If so, it would mean that fluid abilities not only influence generalization strategies in ambiguous situations (e.g., on a transfer test in which participants are asked to predict the outcome to a combination of cues they have never experienced) but also in unambiguous situations in which participants receive feedback. Indeed, it is known that learning to solve one discrimination influences performance on a subsequent discrimination [12,13,14]. Our aim was to investigate the role of fluid abilities on the extent to which learning one discrimination influences the acquisition of new discriminations.…”

“…This might explain why individuals are able to solve these discriminations. It has been suggested that configural and elemental processing lie on a continuum and that individuals can flexibly change their processing style depending on their experience [12,15,16]. That is, it is possible that experience with discriminations that can be solved elementally increases elemental processing, whereas discriminations that can only be solved configurally increase configural processing.…”

“…That is, it is possible that experience with discriminations that can be solved elementally increases elemental processing, whereas discriminations that can only be solved configurally increase configural processing. This can explain why solving a discrimination that has only a configural solution (i.e., cannot be solved on the basis of the individual cues alone) facilitates the acquisition of subsequent patterning discriminations [12,13,14]. To investigate whether fluid abilities influence performance on a second discrimination by inducing a shift towards configural processing or by increasing the likelihood that a learned rule will be applied to the new discrimination, we tested whether fluid abilities could predict an improvement in performance in situations in which a rule could be learned and applied to the second discrimination (Experiments 1 and 2).…”

Category relevance attenuates overshadowing in human predictive learning.

“…Additionally, it has often been observed that previous experience solving a discrimination in which a configural solution is necessary, facilitates solving a second discrimination with a configural requirement (Glautier et al, 2016; Mehta & Russell, 2009). Therefore, previous experience solving a discrimination modulates subsequent behavior (i.e., transfers) concerning new discriminations (Urcelay & Miller, 2010).…”