General object-based features account for letter perception better than specialized letter features

Abstract

After years of experience, humans become experts at perceiving letters. Is this visual capacity attained by learning specialized letter features, or by reusing general visual features previously learned in service of object categorization? To investigate this question, we first measured the visual representational space for letters in two behavioral tasks, visual search and letter categorization. Then, we created models of specialized letter features and general object-based features by training deep convolutional neural networks on either 26-way letter categorization or 1000-way object categorization, respectively. We found that general object-based features accounted well for the visual similarity of letters measured in both behavioral tasks, while letter-specialized features did not. Further, several approaches to alter object-based features with letter specialization did not improve the match to human behavior. Our findings provide behavioral-computational evidence that the perception of letters depends on general visual features rather than a specialized feature space.