Oscar Woolnough¹ (), Xavier Scherschligt¹, Ellery Wheeler¹, Stanislas Dehaene², Nitin Tandon¹; ¹UTHealth Houston, ²INSERM

Words appear everywhere in our everyday environment. But how does our brain process these words differently while we are actively reading them compared to when they are in the background? Prior work clearly demonstrates enhancement of activation to attended stimuli, but does attention also modulate sensitivity to higher-order word-level statistics within visual cortex? To probe this, we used direct intracranial recordings in six humans as they each performed four different visual-word-based tasks in four separate blocks: (i) Fixation, (ii) One-back, (iii) Semantic decision, and (iv) Lexical decision. Words and pseudowords were presented on screen for 500ms each and participants were asked to push a button when target stimuli for that task appeared on screen. Using linear mixed-effects modelling of broadband gamma activity (70-150Hz), we mapped how these task demands modulated spatiotemporal representations of higher-order lexical statistics, such as lexical frequency and lexicality. We targeted three primary hubs of the reading network in mid-fusiform cortex, inferior parietal sulcus and inferior frontal gyrus. Within individual electrodes we clearly detected task-based modulation of overall activation and sensitivity to both lexicality and lexical frequency. At group level, when performing a lexical decision task, both sensitivities were enhanced across the entire reading network, as compared to fixation. Modulation of lexicality sensitivity manifested as increased activation in response to pseudowords with little modulation of overall known word activation. In contrast, while performing a semantic decision task we observed selective enhancement of lexical frequency sensitivity only in inferior frontal gyrus. This was paired with selective increases in bidirectional Granger causality between mid-fusiform cortex and inferior frontal gyrus, relative to fixation, specifically for known words. This work demonstrates that both local processing of higher-order lexical statistics and inter-regional connectivity throughout the reading network are highly task dependent and are sensitive to the specific demands of a given task.

Acknowledgements: This work was supported by the National Institute of Neurological Disorders and Stroke U01NS128921.