Uni- and Supramodal Categorical Representations

The neural routing of information into meaningful categorical representations is an essential prerequisite for multiple cognitive processes. It implies two sequential levels, grouping elements according to given categories and, within categories, identification and discrimination of individual exemples. While localized neural correlates of the former operation have already been identified in many functional neuroimaging studies, there are still few data available that demonstrate coding at the subordinate level of identity representation. Using functional magnetic resonance imaging in healthy human subjects, we addressed these issues for two behaviorally important categories: faces and numerical magnitudes. In the case of faces, we exploited perceptual invariance across spatial scales. From full-range face images we constructed non-overlapping high- or low-pass filtered face images. We then tested where in the brain repetition of individual face identity but with images using different spatial scales would still result in response adaptation. Adaptation is a signature that the neuronal sub-populations within a given brain area that are involved by an individual stimulus process the two versions as identical. In our case, this would mean that they invariantly respond to the identity of a face despite the profound differences between the images conveying this information. We found such functional behavior in fusiform but not occipital face-sensitive areas (Eger et al. NeuroImage 2004; 22: 232–242). The latter only adapted when identical image versions of an individual face were repeated. To study the representation of numerical magnitudes we exploited the fact that across the visual and auditory modality Arabic numerals and number words are always automatically recoded to representations of the corresponding numerical magnitude even when this is not required for the task on hand. Behaviorally, this transpires into the so-called numerical distance effect. If, for example, the task requires subjects to report number 6 as a target and all other numbers shown as non-targets, the reporting of non-targets will be slowed down the closer numerically this non-target is to the target. In other words, it will take longer to report 5 as a non-target than 2. In the control categories we used, letters and colors, we found corresponding behavioral distance effects in relation to alphabetic and chromatic distance, respectively. Using an event-related design with randomly interleaved item presentations across all three categories and auditory and visual modalities, we could establish that a cortical region deep in the posterior intraparietal sulcus responds selectively to numbers in a supramodal way (Eger et al. Neuron 2003; 37: 719–725). This finding converges with a body of data from human and non-human primates that have implied this region in the explicit processing of numerical magnitude