For most people, reading is effortless. Yet, the ease hides a cluster of complex and connected cognitive processes, many of which remain unknown. The EU-funded VIFER (The visual front-end of reading) project examined the visual perception of written words. Researchers considered how statistical learning shapes such perception. The team produced a comprehensive learning-based neuro-computational explanation for word perception. Researchers also integrated the neuro-computational approach with electrophysiology using an electroencephalogram (EEG). This method provided novel types of neural signatures, which allowed characterisation of timing and spatial aspects of the cortical processing involved in reading. Most importantly, the team developed several advanced neuro-computational models able to describe the fine details of visual word perception. Initial results concerned the processes involved in perception of letters. Simulations showed a plausible hierarchical structure of visual features that emerged during unsupervised perceptual learning. The study supported the recycling hypothesis, whereby cortical structures evolved to support general vision and perception of letters. The team additionally investigated the cognitive mechanisms that combine letters into words, and the emergence of lexical statistical regularities. An extension of the same model used for letter perception easily acquired lexical grammars representing discrete words. The result demonstrated that the generative approach successfully applies to sequential letter processing, and that the emergent statistical model effectively simulates psycholinguistic patterns. Work on whole-word perception involved development of a further model able to simulate many neural-level details in visual word perception. One such detail was spatial normalisation. The model also explained how normalisation produced various psycholinguistic phenomena, including letter transposition. Simulations revealed details of the hierarchical structures allowing word perception. Single letters are the essential top-level perceptual element. Advanced analysis of EEG signatures revealed the complex dynamics of visual perception processes associated with reading. Such processes could theoretically decode perceived letters/words from cortical activity. Nevertheless, testing of the hypothesis showed that EEG signals could only allow discrimination of low-level visual features. Instead, the team explored steady-state visually evoked potentials, and their use as tools for investigation of cognitive mechanisms. VIFER results helped reveal the deep cognitive functions involved in reading, meaning potential new treatments of reading disorders.
Reading, word perception, VIFER, visual perception, neuro-computational, EEG