Variations in the Accuracy of Old–New Judgements Following Compound Learned-Predictiveness Training

Connectionist theories of psychology propose that the accuracy of stimulus representations is a function of the extent to which its elements become interconnected, or associated, as a consequence of experience. To explore this idea, across three experiments, participants were required to solve a compound learned-predictiveness task, in which pairs of stimuli in each compound were established as either predictive or non-predictive of a subsequent outcome. Outcome probability was further manipulated as either probabilistic or deterministic. It was hypothesized that the propensity for one stimulus to connect with another within the compound would be greater when (a) it possessed higher relative predictive validity for an outcome and (b) was followed by an uncertain outcome relative to a deterministic outcome. Consequently, participants accuracy in recognising the conjunction of the previously exposed stimulus pairs should vary as a function of these variables. Across three experiments, which used old/new judgements following compound learned-predictiveness training, a consistent effect of predictive validity was observed - participants' hit rate for detecting a change in the conjunction of predictive stimuli was superior to changes in the conjunction of non-predictive stimuli. However, there was no effect of outcome uncertainty. The implications of these results for theories of learning and representation are discussed.