Prior beliefs influence symmetrical or asymmetrical generalizations in human causal learning.

The generalization decrement between element A and compound AX has shown both symmetrical (Thorwart & Lachnit, 2009) and asymmetrical (Glautier, 2004) patterns in human contingency learning. In a series of experiments we examined the hypothesis that prior beliefs about the relationship between a distinctive element X and an outcome are important for determining the different generalization patterns. Participants learned which given enterobacteria caused a negative or a positive effect on gastrointestinal conditions. Subsequently, they were asked to evaluate learned cues and novel cues in which distinctive elements were added to or removed from the enterobacteria. The results generally demonstrated that relatedness between the elements and outcomes, such as negative features combined with a negative outcome or positive features combined with a positive outcome, resulted in asymmetrical generalization patterns. By contrast, unrelated combinations, such as positive features and a negative outcome, produced symmetrical patterns of generalization. Configural and elemental models of stimulus generalization are discussed.