Creating Structured Task-sets from Categorical Stimuli
Christina Bejjani1, Tobias Egner1; 1Duke University
Adaptive behavior is facilitated by our ability to discover and leverage rules for classifying stimuli and linking them to appropriate actions. Previous studies have shown that when humans learn stimulus-response associations for a small set of multi-dimensional stimuli (e.g., colored shapes), they will spontaneously form and generalize abstract rule structures, even in the absence of inherent structure and performance benefit. Here, we tested some determinants and boundary conditions of such spontaneous task-set building. Specifically, we tested whether this effect could also be observed at the level of stimulus categories (trial-unique face stimuli), the degree to which it is biased by how stimulus categories map onto responses (random vs. grouped by dimension), and whether the order in which these biases are introduced affects task-set structure. Participants performed a feedback-based learning task that allowed for hierarchical clustering of stimulus-action rules according to face stimulus dimensions (age, gender). In “flat learning” blocks, the stimulus-response mapping was arbitrary; in “hierarchical learning” blocks, the choice was “motor-biased” such that the buttons assigned to each image category were clustered according to a particular higher-level dimension (e.g., gender). We found dimensional switch-costs in both blocks, but costs were increased in the hierarchical learning condition, regardless of the order in which the conditions were introduced. These results document that humans generate hierarchical task-sets for grouping trial-unique stimuli into categories, even in the absence of inherent structure and performance advantages, and this tendency is robustly enhanced when stimulus-response mappings encourage dimensional grouping.
Topic Area: EXECUTIVE PROCESSES: Goal maintenance & switching