Learning About Associations: Evidence for a Hierarchical Account of Occasion Setting Charlotte Bonardi and Do ´mhnall Jennings University of Nottingham In 2 experiments rats were trained on a switching discrimination, with 4 occasion setters, A, B, C, and D and 2 target stimuli, x and y. When signaled either by A or by B, x was reinforced with food and y was not, whereas when signaled either by C or by D these reinforcement relations were reversed (i.e., A: 3 x+, A: y 3 –, B: x 3 +, B: y 3 –, C: x 3 –, C: y 3 +, D: x 3 –, D: y 3 +). In a subsequent Stage A was paired with shock, and then the degree to which food–reinforced (Experiment 1a) and nonreinforced (Experiment 1b) presentations of x and y were capable of eliciting fear was assessed. Those conditioned stimulus (CS)/unconditioned stimulus (US) relations that had been operative in the presence of the fear-eliciting occasion setter A (i.e., x 3 +, y 3 –) elicited more fear than the alternative CS/US combinations (i.e., x 3 –, y 3 +). The implications of these findings are discussed with reference to theories of occasion setting and of configural learning. Keywords: learning, associations, occasion setting, hierarchical In a positive occasion-setting discrimination a target stimulus t is reinforced when it is preceded by a feature, F, but not when it is presented alone. Accurate performance on such discriminations cannot always be explained in terms of binary associations be- tween F, t and the outcome; in these cases the feature is termed an occasion setter (cf. Skinner, 1938). One theory that has been proposed to explain this behavior is that an occasion setter acts on the conditioned stimulus (CS)/unconditioned stimulus (US) asso- ciation in a hierarchical fashion, operating as an “and-gate” that facilitates flow of activation between CS and US (e.g., Holland, 1983; cf. Bouton, 1990). Elaborating this idea, Bonardi (e.g., 1989; 1998) proposed that the basis of this behavior might be associa- tive—that the CS/US association is an independent entity that can enter into associations, and that occasion setting is the result of an associative link between the occasion setter and the CS/US asso- ciation. Evidence in support of this idea comes from the demon- stration that occasion setters are subject to blocking—it is more difficult to establish a stimulus as an occasion setter if it is trained in compound with another occasion setter than with some other stimulus (e.g., Bonardi, 1991, 2007). As blocking is viewed as a defining feature of associative learning, this suggests that occasion setting might be associative in nature. However, direct evidence for the proposal that the CS/US association can act as an indepen- dent unit is less forthcoming. The present experiments aimed to provide such evidence. Rats were trained with four occasion setters, A, B, C, and D, and two target stimuli, x and y: A and B signaled the reinforcement of x with food (x+), and the nonreinforcement of y (y–); C and D signaled the opposite (i.e., A: x+, A: y–, B: x+, B: y–, C: x–, C: y+, D: x–, D: y+; cf. Honey & Watt, 1989, 1999). Thus each occasion setter was paired equally often with x and y, and with food and no food; they differed only in the combinations of these events that they signaled. Then A was paired with shock, while C was nonreinforced, and we examined the extent to which the specific pairings x+ and y+ (Experiment 1a) and x– and y– (Experiment 1b) were capable of eliciting fear. Our rationale relies on the assumption that each occasion setter is associated with each of the two CS/outcome relationships that holds in its presence—thus A is associated with x 3 food and y 3 no food, and C with x 3 no food and y 3 food. When A is paired with shock, it should at the same time evoke the representations of the two associations it signals, x 3 food and y 3 no food, which can thus also become associated with the shock. At test the CS/outcome combinations that have been associated with shock should elicit more fear than the alternatives—thus x 3 food should elicit more fear than y3 food, and y 3 no food more fear than x 3 no food. This prediction relies on the associatively activated representa- tion of the CS/outcome association becoming directly associated with the shock. Moreover, it assumes that such learning will occur despite the fact that neither constituent event of the association is actually presented. Evidence for learning of this type, which has been termed mediated conditioning, has been observed when the to-be-associated entity is an individual event (e.g., Holland, 1981). The critical feature of the present experiment is that the associa- tively activated representation that is becoming associated with shock corresponds to an association between two further events. If our prediction is supported, then this will provide evidence that the CS/outcome association is acting as an independent entity, as the hierarchical account predicts. Charlotte Bonardi and Do ´mhnall Jennings, School of Psychology, Uni- versity of Nottingham. Do ´mhnall Jennings is now at the Centre for Behaviour and Evolution, Institute of Neuroscience, Newcastle University. This work was funded by the Biotechnology and Biological Sciences Research Council. We thank Jo Temperton, Stuart Morley, and Richard Wood for technical assistance. Correspondence concerning this article should be addressed to Char- lotte Bonardi, School of Psychology, University of Nottingham, Uni- versity Park, Nottingham NG7 2RD, United Kingdom. E-mail: cmb@psychology.nottingham.ac.uk Journal of Experimental Psychology: © 2009 American Psychological Association Animal Behavior Processes 2009, Vol. 35, No. 3, 440 – 445 0097-7403/09/$12.00 DOI: 10.1037/a0014019 440