The Empirical Case for Dual Coding