Cognitive neuroscience research on conceptual knowledge often is discussed with respect to “embodiment” or “grounding.” We tried to disentangle at least three distinct claims made using these terms. One of these, the view that concepts are entirely reducible to sensory-motor representations, is untenable and diminishing in the literature. A second is the view that concepts and sensory-motor representations “interact,” and a third view addresses the question of how concepts are neurally organized—the neural partitions among concepts of different kinds, and where these partitions are localized in cortex. We argue that towards the second and third issues, much fruitful research can be pursued, but that no position on them is specifically related to “grounding.” Furthermore, to move forward on them, it is important to precisely distinguish different kinds of representations—conceptual vs. sensory-motor—from each other theoretically and empirically. Neuroimaging evidence often lacks such specificity. We take an approach that distinguishes conceptual from sensory-motor representations by virtue of two properties: broad generality and tolerance to the absence of sensory-motor associations. We review three of our recent experiments that employ these criteria in order to localize neural representations of several specific kinds of nonsensory attributes: functions, intentions, and belief traits. Building on past work, we find that neuroimaging evidence can be used fruitfully to distinguish interesting hypotheses about neural organization. On the other hand, most such evidence does not speak to any clear notion of “grounding” or “embodiment,” because these terms do not make clear, specific, empirical predictions. We argue that cognitive neuroscience will proceed most fruitfully by relinquishing these terms.

This research studies the neural systems underlying two integration processes that take place during natural discourse comprehension: consistency evaluation and passive comprehension. Evaluation was operationalized with a consistency judgment task and passive comprehension with a passive listening task. Using fMRI, the experiment examined the integration of incoming sentences with more recent, local context and with more distal, global context in these two tasks. The stimuli were stories in which we manipulated the consistency of the endings with the local context and the relevance of the global context for the integration of the endings. A whole-brain analysis revealed several differences between the two tasks. Two networks previously associated with semantic processing and attention orienting showed more activation during the judgment than the passive listening task. A network previously associated with episodic memory retrieval and construction of mental scenes showed greater activity when global context was relevant, but only during the judgment task. This suggests that evaluation, more than passive listening, triggers the reinstantiation of global context and the construction of a rich mental model for the story. Finally, a network previously linked to fluent updating of a knowledge base showed greater activity for locally consistent endings than inconsistent ones, but only during passive listening, suggesting a mode of comprehension that relies on a local scope approach to language processing. Taken together, these results show that consistency evaluation and passive comprehension weigh differently on distal and local information and are implemented, in part, by different brain networks.

The perception of apparent body movement sometimes follows biologically plausible paths rather than paths along the shortest distance as in the case for inanimate objects. For numerous authors, this demonstrates that the somatosensory and motor representations of the observer's own body support and constrain the perception of others’ body movements. In this paper, we report evidence that calls for a re-examination of this account. We presented an apparent upper limb movement perception task to typically developed participants and five individuals born without upper limbs who were, therefore, totally deprived of somatosensory or motor representations of those limbs. Like the typically developed participants, they showed the typical bias toward long and biomechanically plausible path. This finding suggests that the computations underlying the biomechanical bias in apparent body movement perception is intrinsic to the visual system.

Segmentation of the world into meaningful units has typically been described as object individuation, emphasizing the spatially disconnected quality that comes as a result of objecthood. This segmentation can occur rapidly, even in parallel for multiple objects. It remains unclear whether objecthood is a necessary requirement for parallel individuation, or whether target features in distinct locations, such as object parts, may also be individuated in parallel. In a series of six experiments, we used a rapid enumeration task to test whether subitizing, a phenomenon believed to result from parallel individuation, occurs over object parts. We found that subitizing and individuation occur over connected object parts as well as physically separate objects of varied shapes and sizes. We also observed subitizing when target items are indents, features intrinsic to the shape of the object, and when cues for occlusion were removed. The results of these studies suggest that parallel individuation is not bound to objecthood, and can occur over object parts existing in separate locations.

The nature of domain-specific organization in higher-order visual cortex (ventral occipital temporal cortex, VOTC) has been investigated both in the case of visual experience deprivation and of modality of stimulation in sighted individuals. Object domain interacts in an intriguing and revelatory way with visual experience and modality of stimulation: selectivity for artifacts and scene domains is largely immune to visual deprivation and is multi-modal, whereas selectivity for animate items in lateral posterior fusiform gyrus is present only with visual stimulation. This domain-by-modality interaction is not readily accommodated by existing theories of VOTC representation. We conjecture that these effects reflect a distinction between the visual features that characterize different object domains and their interaction with different types of downstream computational systems.

Recognizing the identity of a face is computationally challenging, because it requires distinguishing between similar images depicting different people, while recognizing even very different images depicting a same person. Previous human fMRI studies investigated representations of face identity in the presence of changes in viewpoint and in expression. Despite the importance of holistic processing for face recognition, an investigation of representations of face identity across different face parts is missing. To fill this gap, we investigated representations of face identity and their invariance across different face halves. Information about face identity with invariance across changes in the face half was individuated in the right anterior temporal lobe, indicating this region as the most plausible candidate brain area for the representation of face identity. In a complementary analysis, information distinguishing between different face halves was found to decline along the posterior to anterior axis in the ventral stream.