Abstract
An enduring debate regarding the functional architecture of the cortex is whether different kinds of information are represented by distributed and overlapping neural circuits or are restricted to networks specialized for particular domains of information. Neural recordings, lesions, and stimulation show that the perception of particular visual categories is causally related to the activity in category-selective patches of ventral temporal cortex. On the other hand, visual deficits caused by lesions are rarely, if ever, “pure,” and information about categories can be found outside of patches selective for those categories. Given that category-level discrimination is generally spared in various agnosias, a critical tension between domain-specific vs. distributed models is whether individual-level discrimination can be found outside of putative category-selective areas. To address this tension, intracranial recordings from 17 epilepsy patients were used to assess the spatiotemporal representation for individual faces and words in human temporal cortex. Recordings from a category localizer task were used to measure category selectivity in all electrodes located in ventral temporal cortex. Multivariate classification was used to analyze the spatiotemporal dynamics of individual face or word discrimination inside and outside these category-selective cortical patches. The results of this analysis show that single faces and words can be individuated both within and outside of category-selective patches, but the respective representations emerge approximately 200 ms earlier inside than outside the selective patches. Further analyses reveal that the information represented outside of category-selective patches is non-redundant with the information within these patches, and thus the non-selective regions contribute to the overall neural representation though in a later stage of processing. These results provide a potential resolution between domain-specific and distributed models of visual perception by suggesting that the cortical representation is dynamic, with processing first primarily restricted to domain-specific networks followed by a distributed processing stage.