Higher-order conditioning: A critical review and computational model.

Higher-order conditioning results from a simple training procedure: Pairing two relatively neutral conditioned stimuli, A and X, allows properties separately conditioned to X (e.g., through pairing it with an unconditioned stimulus, US) to be evident during A. The phenomenon extends the range of ways in which Pavlovian conditioned responding can be expressed and increases its translational relevance. Given this relevance and the wealth of available behavioral analysis, it is a surprisingly underdeveloped territory for formal theoretical analysis. Here, we first provide a critical review of two (informal) classes of account for higher-order conditioning that reflect either: (a) processes that are analogous to Pavlovian conditioning, but involving associatively activated representations (e.g., A→US); or (b) the formation of an associative chain (e.g., A→X, and X→US). Our review first identifies fundamental theoretical and empirical challenges to both classes of account. We then develop a new computational model of higher-order conditioning that meets the challenges identified by showing: how reciprocal associations between A, X, and the US are formed and affect performance; and how the similarity of stimuli, their traces, and associatively retrieved representations modulate this process. The model generates a wealth of novel predictions, providing a platform for further empirical and theoretical analysis. (PsycInfo Database Record (c) 2023 APA, all rights reserved).