Phonological Overlap Alone Does Not Mediate Generalization in Action-Effect Learning: Evidence from Homophones

The theory of event coding posits that action selection is guided by anticipating action effects. A key question is how newly acquired action–effect relations generalize beyond the specific learning context. Whereas transfer based on perceptual and semantic overlap has been reported, it remains unclear whether phonological overlap alone can mediate generalization. Using Japanese homophones that share segmental phonology but differ in meaning and orthography, we tested phonological generalization in a free-choice action–effect learning paradigm. Participants first learned arbitrary mappings between left and right keypresses and two action-effect stimuli. In a subsequent Go/No-go test, an effect-related stimulus was presented, and participants freely selected a response; the critical outcome was whether choices were biased toward the response previously associated with the corresponding phonological category. Across experiments using pictures and Kanji words, homophone test stimuli did not reliably bias voluntary choices. In contrast, a control condition with identical learning and test stimuli showed a clear learning-related choice bias, indicating that the paradigm can detect action–effect learning under full stimulus overlap. Together, these results suggest that segmental phonological overlap, when isolated from semantic and orthographic similarity, is insufficient to drive transfer in voluntary action selection. The findings are consistent with accounts emphasizing holistic event-file retrieval and/or strategy-based conceptual rules in free-choice tasks.