“…However, recently, evidence from psychophysics and time-resolved intracranial recordings suggests that frequent bigrams may not contribute to recognition, at least during the first ~300 ms of word recognition, where only frequent letters are separated from rare letters or non-letters (Agrawal et al, 2020;McCloskey et al, 2013;Woolnough et al, 2020). What develops with literacy is the compositionality of visual word representation, which increases the dissimilarity and independence between individual letters at nearby locations (Agrawal et al, 2019(Agrawal et al, , 2022. Compared to contextual bigram coding, ordinal letter coding provides a better fit to the psychophysical distance between letter strings (Agrawal et al, 2020), the early intracranial responses to written words (Woolnough et al, 2020), and the responses of word selective units in artificial deep networks (Hannagan et al, 2021).…”