Can you think of an experiment where you could get better accuracy on non-word trials than on word trials?