Complementary neural representations for faces and words: A computational exploration

Abstract

A key issue that continues to generate controversy concerns the nature of the psychological, computational, and neural mechanisms that support the visual recognition of objects such as faces and words. While some researchers claim that visual recognition is accomplished by category-specific modules dedicated to processing distinct object classes, other researchers have argued for a more distributed system with only partially specialized cortical regions. Considerable evidence from both functional neuroimaging and neuropsychology would seem to favour the modular view, and yet close examination of those data reveals rather graded patterns of specialization that support a more distributed account. This paper explores a theoretical middle ground in which the functional specialization of brain regions arises from general principles and constraints on neural representation and learning that operate throughout cortex but that nonetheless have distinct implications for different classes of stimuli. The account is supported by a computational simulation, in the form of an artificial neural network, that illustrates how cooperative and competitive interactions in the formation of neural representations for faces and words account for both their shared and distinctive properties. We set out a series of empirical predictions, which are also examined, and consider the further implications of this account.

Keywords:

• Prosopagnosia
• Alexia
• Neural substrate

Acknowledgments

This work was funded by National Science Foundation (NSF) Grant BCS0923763 to Behrmann and Plaut and by NSF Grant SBE-0542013 to the Temporal Dynamics of Learning Center, an NSF Science of Learning Center. We thank Jennifer Brace for her work in creating the stimuli used in the reported simulation.

Notes

¹ We use the terms “module” and “modular” not in the strict senses in which Fodor (1983) defined them, but to denote a general class of theoretical commitments in which domain-specific cognitive processes, such as face recognition, are each carried out by a neuroanatomically identifiable cortical area, such as the FFA. To the extent that multiple cortical areas are involved in a given cognitive process, it would mitigate against a modular account of that process but might still be consistent with modular accounts of localized subprocesses.

² Although not reported here in detail, these qualitative results are stable over changes to nonessential aspects of the network architecture and training methods, including variations (within reasonable limits) in random initial weights, learning parameters, and numbers of hidden units.

³ The exact nature of the change over developmental time remains somewhat controversial with some studies showing changes in the volume of activation for one category over another and others showing a change in functional/effective connectivity over the course of development. The studies are also not entirely consistent with each other (see Cantlon et al., 2011, showing adult-like activation to faces as well as sensitivity to alphanumeric symbols in four-year-olds although volume/cluster size was not evaluated in this study). These empirical discrepancies remain to be resolved.