Lecture by Ellen Lau: One semantic hub, two working memory systems

Alan Baddeley famously proposed the existence of multiple, independent neurocognitive working memory systems, most importantly distinguishing between the 'phonological loop' and the 'visuo-spatial sketchpad'. Here I argue that routine language comprehension depends on not one but *both* of these systems, in a way that resolves some ongoing puzzles in the neuroimaging literature. Specifically, I suggest that language comprehension makes use of two independent, capacity-limited working memory systems: a posterior temporal - frontal circuit that buffers phonological-articulatory information in sequence form, and an anterior temporal - inferior parietal circuit that buffers semantic information in 'scene' form. While the latter circuit initially developed for guiding visual information processing of scenes, on this account it came to also support non-spatial content as amodal conceptual representations in anterior temporal cortex became increasingly elaborate. This 'semantic sketchpad' hypothesis provides a new perspective on an ongoing debate about why both anterior temporal cortex and the angular gyrus are frequently associated with semantic processing: I argue that these observations follow from their complementary but wholly distinct roles in long-term memory encoding and short-term buffering. In this account, the angular gyrus is not a 'hub' for long-term semantic knowledge, but is rather engaged whenever the task or the semantic information to be encoded is complex enough that buffering is required. This contrasts with existing 'dual hub' views that distribute long-term semantic knowledge across both regions according to whether the knowledge is taxonomic or thematic (Schwartz et al. 2011; Binder & Desai 2011), as well as 'single hub' views that include angular gyrus as part of a generic 'semantic control' circuit (Lambon Ralph et al. 2016).