Peak shift and rules in human generalization.

Abstract

Two experiments tested whether a peak-shifted generalization gradient could be explained by the averaging of distinct gradients displayed in subgroups reporting different generalization rules. Across experiments using a causal judgment task (Experiment 1) and a fear conditioning paradigm (Experiment 2), we found a close concordance between self-reported rules and generalization gradients using a continuous stimulus dimension (hue). Both experiments also showed an overall peak-shifted gradient after differential conditioning, but not after single cue conditioning. Importantly, the peak shift could be decomposed into linear and peaked gradients when participants were divided into rule subgroups. Our results highlight the need to consider individual differences in the rules that participants derive in human generalization studies and suggest that in some situations, peak shift may be a consequence of averaging across diverse rule subgroups. (PsycINFO Database Record (c) 2018 APA, all rights reserved).

To perceive or not to perceive? Individual differences in perception predict rule induction and response generalization

“…That is, in the majority of datasets the similarity rule was not only reported by those with a more accurate perception, but it was more likely then the linear rule in those participants. This study is novel in reporting a relationship between perception and induction using cluster analysis, and suggests one avenue by which individual differences in inductive rules reported in recent work with these stimuli (Lee et al, 2018;Lovibond et al, 2020) could arise. It may be that participants who are better able (or happen) to attend to the features of the CS+ during training go on to generalize using these features and therefore report a similarity rule.…”

“…When aggregated across subjects, responses tend to form a bellshaped gradient across the stimulus dimension that peaks at the location of the CS+. If instead of simple conditioning, differential conditioning is conducted with an additional similar cue (e.g., a slightly higher or lower tone) that predicts US absence (CS-), the gradient often becomes asymmetrical and sigmoidal in shape (Ghirlanda & Enquist, 2003;Mednick & Freedman, 1960), and the peak of the gradient can sometimes shift (Lee, Hayes, & Lovibond, 2018;Purtle, 1973;Zaman, Struyf, Ceulemans, Vervliet, & Beckers, 2020).…”

“…The identification and differentiation regarding these and other mechanisms based solely on patterns of generalized responding, however, remains difficult as different mechanisms can yield the same behavioral phenotype (Struyf, Zaman, Vervliet, & Van Diest, 2015). The second complexity arises from the fact that response generalization in humans displays a high amount of variability (Lee, Mills, Hayes, & Livesey, 2020;Zaman, Chalkia, et al, 2020), with large individual differences in inductive rules (Lee et al, 2018;Lovibond et al, 2020;Wong & Lovibond, 2017) and perceptual ability (Struyf, Zaman, Hermans, & Vervliet, 2017;Struyf et al, 2015;Zaman, Ceulemans, Hermans, & Beckers, 2019;Zaman, Struyf, Ceulemans, Beckers, & Vervliet, 2019;Zaman, Chalkia, et al, 2020;Zaman, stimulus features. This leads to the prediction that more accurate stimulus perception should lead to a higher likelihood of deriving a similarity over a linear generalization rule.…”

Rule-based generalization and peak shift in the presence of simple relational rules

“…It is tempting to conclude that the peak shift shown in the Inconsistent group on trials where the relational rule was not applicable on the previous trial is due to the same basic associative processes used to explain peak shift in animals. Error-correction learning models account for the peak shift phenomenon with impressive quantitative precision ([ 7 , 10 , 14 ], but see [ 15 ], for an alternative rule-based explanation]. When considered in light of the human literature on peak shift, our results seem wholly consistent with the idea that associative processes always operate in human categorization or discrimination studies, but are usually masked by higher-order rule-learning which dominates performance at test ([ 20 ]).…”

“…Peak shift describes the situation in which the peak of the generalization gradient shifts from the location of the S+ in the direction away from the S-. This effect is well accounted for by associative models that employ elemental representation and conceive of the S+ and S- as a series of overlapping elements on a continuum ([ 7 , 10 , 14 ], but see [ 15 ], for an alternative rule-based explanation). Peak shift can be predicted if it is assumed that the S+ and S- activate elements on the continuum in a graded (Gaussian) manner during discrimination training.…”