From form to concept: a dynamic spatiotemporal brain network cascade supports language comprehension

Abstract

Abstract Encoding new information through language underlies much of human learning, yet the neural dynamics that transform linguistic input into lasting conceptual representations remain unclear. Methodological constraints in neuroimaging have siloed the study of language comprehension into isolated subprocesses (word-, sentence-, and discourse-level), obscuring how information travels across the cortex in real time. Here, we overcome these constraints using a fusion of functional Magnetic Resonance Imaging and electroencephalogram (EEG) in healthy adults and introduce the Form-to-Concept framework. Focusing on canonical language EEG components, we identified a sequence of brain network activations: an occipitotemporal perceptual word processing network (250 ms), a temporoparietal semantic retrieval network (400 ms), a posterior default mode inferential network (500 ms), a frontotemporal syntactic-semantic integration network (600 ms), and a prefrontal-midline default mode network concept coherence network (700 ms). Each stage was differentially tuned based on context, and transitions were mediated by overlapping hub regions, particularly the left temporoparietal junction. Individuals with stronger language comprehension ability exhibited distinct engagement along the spatiotemporal cascade, showing enhanced weighting of mid-to-late inferential processes relative to top-down semantic control. This work bridges the longstanding spatiotemporal gap, revealing that canonical event-related potentials reflect a temporally structured cascade of distributed brain networks that dynamically transform linguistic input into conceptual knowledge.