Modeling the Visual Word Form Area Using a Deep Convolutional Neural Network

The visual word form area (VWFA) is a region of the cortex located in the left fusiform gyrus, that appears to be a waystation in the reading pathway. The discovery of the VWFA occurred in the late twentieth century with the advancement of functional magnetic resonance imaging (fMRI). Since then, there has been an increasing number of neuroimaging studies to understand the VWFA, and there are disagreements as to its properties. One such disagreement is regarding whether or not the VWFA is more selective for real words over pseudowords1. A recent study using fMRI adaptation (Glezer, et al., 2009) provided evidence that neurons in the VWFA are selectively tuned to real words. This contradicts the hypothesis that the VWFA is tuned to the sublexical structure of visual words, and therefore has no preference for real words over pseudowords. In this paper, we develop a realistic model of the VWFA by training a deep convolutional neural network to map printed words to their labels. The network is able to achieve an accuracy of 98.5% on the test set. We then analyze this network to see if it can account for the data Glezer et al. found for words and pseudowords, and find that it does.