Dissecting the cortical stages of invariant word recognition

Abstract

Fluent reading requires the brain to precisely encode the positions of letters within words, distinguishing for instance FORM and FROM across variations in size, position, and font. While early visual areas are known to encode retinotopic positions, how these representations transform into invariant neural codes remains unclear. Building on a computational model of reading, we used 7T fMRI and MEG to reveal a cortical hierarchy in which early visual areas (V1-V4) predominantly encode retinotopic information, whereas higher-level regions, including the Visual Word Form Area (VWFA), transition to an ordinal letter-position code. MEG analyses confirm that retinotopic encoding emerges early (60-200 ms), followed by a shift toward ordinal representations in later time windows (220-450 ms). Despite this transition, word position remained a dominant factor across all time points, suggesting a concurrent coding of both retinotopic and abstract positional information. These findings uncover the spatiotemporal dynamics by which the human brain transforms visual input into structured linguistic representations, shedding light on the cortical stages of reading and their developmental and clinical implications.