From Exemplars to Categories: Generalization of Action-Effect Learning in a Free-choice Task

A fundamental aspect of intentional action is the ability to generalize learned response-effect (R-E) associations to novel situations. However, the mechanisms underlying such category-based generalization, particularly in the context of voluntary action selection, remain contested. Across four free-choice experiments, we systematically investigated this issue. Participants learned associations between keypresses and pictorial effects, where the learning experience was manipulated from a single, repeated exemplar to numerous unique exemplars presented only once. The categories ranged from basic-level (e.g., dogs vs. cars) to abstract superordinate levels (e.g., animals vs. machines). Results consistently demonstrated a significant generalization effect: participants preferentially selected actions congruent with the category of novel test stimuli, even when exemplar-specific learning was minimized. This provides evidence for a direct response-category binding mechanism. However, this group-level effect was consistently driven by a subset of participants exhibiting near-deterministic choice patterns, and the statistical evidence for this effect was substantially weakened after excluding them. These findings suggest that R-E generalization in voluntary action is not a uniform, automatic process but relies on the acquisition and strategic application of explicit, propositional knowledge regarding action-effect contingencies. This highlights the critical role of conscious reasoning in goal-directed behavior, refining current theories of action control.