Abstract
Intentional action is guided by learned relations between actions and their effects, but it remains unclear which features of an effect support generalization of such learning. We tested whether phonological information can retrieve learned action-effect relations and bias voluntary action selection. Japanese homophones provide a stringent test because words can share pronunciation while differing in meaning and visual or orthographic form. In four preregistered experiments, Japanese participants learned arbitrary pairings between left/right keypresses and pictures or Kanji characters. They then completed a Go/No-go free-choice test measuring whether each test stimulus biased them to choose the key previously paired with a stimulus of the same pronunciation. Experiment 1a showed that repeating the learned picture biased choices toward the associated key. Experiment 1b found no reliable transfer to homophone pictures. Experiments 2a and 2b used written Kanji characters, which should make phonological codes more directly available. Experiment 2a showed a reliable bias when the learned Kanji characters were repeated in the test. Experiment 2b found no reliable transfer to homophone Kanji. These findings suggest that, in free-choice action-effect learning, phonological overlap alone is insufficient to retrieve or apply newly acquired action-effect relations when meaning and visual or orthographic form differ. Generalization therefore appears to require richer overlap with the original learning episode.