Decoding levels of representation in reading: A representational similarity approach.

Multiple levels of representation are involved in reading single words: visual representations of letter shape, orthographic representations of letter identity and order, phonological representations of the word's pronunciation, and semantic representations of its meaning. Previous lesion and neuroimaging studies have identified a network of regions recruited during word reading, including ventral occipital-temporal regions and the angular gyrus (AG). However, there is still debate about what information is being represented and processed in these regions. This study has two aims. The first is to help adjudicate between competing hypotheses concerning the role of ventral occipital cortex in reading. The second is to adjudicate between competing hypotheses concerning the role of the AG in reading. Participants read words in the scanner while performing a proper name detection task and we use a multivariate pattern analysis technique for analyzing fMRI data - representational similarity analysis (RSA) - to decode the type of information being represented in these regions based on computationally explicit theories. Distributed patterns of activation in the left ventral occipitotemporal cortex (vOT) and the AG show evidence of some type of orthographic processing, while the right hemisphere homologues of the vOT supports visual, but not orthographic, information processing of letter strings. In addition, there is evidence of left-lateralized semantic processing in the lvOT and evidence of top-down feedback in the lvOT. Taken together, these results suggest an interactive activation theory of visual word processing in which both the lvOT and lAG are neural loci of an orthographic level of representations.