Working memory represents phonological and semantic information differently

In verbal working memory (WM), lists of phonologically similar items, such as letters or words, are recalled less accurately than lists composed of phonologically dissimilar items. This effects arises because humans tend to confuse similar information being held in memory. This phonological similarity effect is an important empirical benchmark: It demonstrates that WM encodes information in a phonological format. In contrast, the way semantic is represented has received relatively little attention. I will show that WM encodes semantic information in a fundamentally different way than phonological information. This is informed by experiments manipulating semantic similarity, where participants learn lists composed of semantically similar (e.g., leopard, cheetah, lion, tiger) and dissimilar (e.g., desk, sky, monkey, saw) items. Studies using this manipulation have consistently shown, across a wide range of paradigms, that semantic similarity does not induce confusion errors. This striking dissociation between phonological and semantic similarity challenges current models of WM and suggests the need for alternative approaches to representing phonological and semantic information.