Modeling the Visual Word Form Area Using a Deep Convolutional Neural Network
- Author(s): Wiraatmadja, Sandy
- et al.
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 in functional magnetic resonance imaging (fMRI). Since then, there has been an increased number of neuroimaging studies to understand the VWFA further for its properties. Because it is still relatively recent, there are disagreements in some properties of the VWFA. One such disagreement is regarding whether or not the VWFA is highly more selective for whole real words than pseudowords. A recent study provided evidences that the neurons in the VWFA are tuned to be more selective to real words. This contradicts past studies which hypothesize that the VWFA is tuned to sublexical structure of visual words, and therefore has no preference for real words over pseudowords. The goal of this project is to develop a realistic model of the VWFA by training a deep convolutional neural network to map printed words to their labels. We then analyzed this network to see if we could observe the same selectivity the recent study found for whole real words. On the test set, the network that we trained from scratch is able to achieve an accuracy of 98.5%. Furthermore, we notice the same trends in our network, as in the results of the study, that show how the VWFA is highly selective for whole real words