DCC Lecture Jeremy Skipper

Traditional models suggest that a small number of anatomically fixed brain regions are responsible for language functioning, a view deriving from aphasia and neuroimaging studies. The latter employ thresholded measures of central tendency applied to activity patterns resulting from heterogeneous stimuli, potentially obscuring the whole-brain distribution of language processing. ‘Language regions’ might primarily consist of connectivity hubs that coordinate peripheral regions having time varying activity, averaged away after thresholding. This talk provides converging evidence from multiple neuroimaging studies that averaging over heterogeneous words is localised to regions historically associated with language but distributed throughout a whole-brain periphery when not averaging over the linguistic properties of those words. Localised word regions form central connectivity hubs. These are spatiotemporally dynamic, connecting with 44% of the peripheral brain at any second, appearing in aggregate only over time. Encoding models trained using GPT-2 based language representations predict hub-periphery connections much better than activity in individual ‘language’ or peripheral regions. Integrated Information Decomposition (ΦID) shows that, while ‘language regions’ exhibit ‘unique’ information on average, they dynamically form centres of highly ‘synergistic’ or emergent information across the brain. Finally, local field potentials from intracranial recordings in epilepsy patients suggest that linguistic representations are decodable from everywhere in the brain, with ‘language regions’ being more hub-like. Collectively, results suggest that ‘language regions’ and ‘the language network’ are an artefact of indiscriminately averaging across heterogeneous linguistic representations and processes. Instead, they are dynamic connectivity hubs coordinating distributed networks, with ‘language’ emerging from integrated information across the entire brain.