How do word frequency and semantic diversity affect selection of representations in word processing?