Inaugural Lecture: Spatial Attention and Working Memory Mechanisms in Natural Language Processing

Attention and working memory are well characterized in simplified tasks, but natural reading and speech unfold rapidly and impose stricter temporal demands. I will present human MEG studies probing these mechanisms during naturalistic reading and language comprehension. When listening to language, working memory supports dependencies across embedded structures. We have identified multivariate and representationally specific signatures, along with temporal dynamics, associated with the maintenance of linguistic content and its reactivation when required later in a sentence. In reading, spatial attention is allocated to upcoming parafoveal targets. We conducted a series of studies demonstrating that readers extract orthographic, phonological, and semantic information from the next saccade target within ~100 ms of fixating the current word. Saccade timing is phase-locked to ongoing alpha (~10 Hz) oscillations involving the frontal-eye field, suggesting that alpha rhythms gate visuo-motor coordination during natural reading. I will make explicit how these findings constrain mechanistic models supporting working memory and spatial attention. In future work, we will use OPM-MEG to uncover when these mechanisms emerge in children acquiring language and reading abilities.