É«É«À²

Title of the thesis: Neural and computational mechanisms of flexible word reading

Thesis defender: Jiaxin You 
Opponent: Professor Ole Jensen, University of Oxford, the United Kingdom
Custos: Aalto Professor Riitta Salmelin, Aalto University School of Science

Reading is surprisingly flexible: we can often understand a word even when some of its letters are wrong. This thesis investigated how the brain achieves this flexibility. The aim was to identify the neural mechanisms that support flexible word reading and to test whether they can be explained by predictive coding, the idea that the brain continuously generates predictions about incoming sensory information and minimizes the mismatch between those predictions and the actual input. 

The research combined magnetoencephalography (MEG), analyses of information flow between brain regions, and computational modelling. The results showed that misspelled words do not appear to activate a separate error-correction system. Instead, the brain adapts activity within the same reading network used for normal reading. A key finding was that reading is not purely a one-way process from visual input to language understanding. For correctly spelled words, information mainly flowed forward from visual areas towards language-related regions. For misspelled words, later feedback signals also travelled from higher-level language regions back towards areas involved in visual word recognition. This feedback was stronger when the written input was more uncertain. A computational model based on predictive coding showed a similar advantage. Adding feedback interactions improved its recognition of misspelled words and made its internal representations more similar to human brain responses. 

Together, the findings suggest that flexible reading relies on late, feedback-mediated modulation within the classical language network, governed by predictive coding principles. The results provide new evidence for theories of predictive brain function and help bridge research in language, computational neuroscience, and artificial intelligence (AI). They may also inspire the development of more human-like AI models of cognition.

Key words: magnetoencephalography, word reading, functional connectivity, computational model

Thesis available for public display 7 days prior to the defence at .