Classifying pictures and words: Implications for the dual-coding hypothesis