US specificity of occasion setting: hierarchical or configural learning?
Bottom Line: No effect was observed when two visual "pseudo-occasion setters", C and D (paired with sucrose and oil in a trace relation to the US:C…→suc, D…→oil), were substituted for the occasion setters A and B (C…x, D…x, C…y, D…y; Experiments 2, 3b and 4).These results could not be explained in terms of Pavlovian summation: responding to combinations of Pavlovian CSs paired with same or different outcomes was either the same, or lower when both stimuli had been paired with the same outcome (Experiment 4).Implications of these results for theories of occasion setting and configural learning are discussed.
Affiliation: School of Psychology, University of Nottingham, University Park, Nottingham, UK. email@example.comShow MeSH
Mentions: Standard configural theory cannot explain these findings, because it predicts that test responding depends on the similarity between the configural cue that was conditioned and that present at test – and these configural cues do not contain a representation of the outcome (cf. Honey and Watt, 1999). It follows that an adaptation of configural theory that allowed the outcomes to be represented in the configural cues could explain these results. For example, training A so that it signalled reinforcement of x with sucrose would have resulted in conditioning to the configural cue A/suc/x/suc. If w is subsequently paired with sucrose and z with oil, when A signals w and z at test, the resulting configural cues will be A/suc/w/suc and A/suc/z/oil – and as the former is more similar to the training configure than the latter, this will result in more responding in the former case. An example of such a pseudo-configural theory was proposed by Honey and Watt (1999), who argued that if A signals reinforcement of x with sucrose, A and x become linked to a common hidden unit, p, that also becomes linked to the representation of sucrose. When w is subsequently paired with sucrose, activity in the sucrose representation feeds back to p, and allows w to become associated with it as well (see Fig. 2). A similar associative structure would link B, y and z to the oil representation via a second hidden unit, q. Assuming that two sources of activation to one hidden unit produce more activation than one source of activation to two hidden units (cf. Honey and Watt, 1998, p. 333), then the test results can be explained, as A and w both activate the same hidden unit, p, whereas A and z do not.
Affiliation: School of Psychology, University of Nottingham, University Park, Nottingham, UK. firstname.lastname@example.org