This project is directed at understanding how the brain represents and processes conceptual knowledge about words, objects, actions, and people. Semantic knowledge mediates among perceptual, linguistic, and action based representations, abstracting away from the surface structure of each modality to capture the underlying functional relationships among entities in the world. A thorough understanding of the functional and neural organization of semantic memory would facilitate the formulation of effective remediation and compensatory strategies for individuals with semantic impairments, such as those suffering from Alzheimer's disease. The proposed research focuses on observed dissociations among specific semantic categories (e.g., animals vs. artifacts, objects vs. people) in the behavior of brain-damaged patients and in the patterns of brain activation of normal subjects. These dissociations suggest that a number of brain regions are partially specialized for different types of semantic information, and that different categories of entities place distinctive demands on various parts of this system. Our hypothesis is that semantics is a learned, internal representation with graded functional specialization that arises from the effects of two constraints on learning: 1) available connectivity among brain areas, both within and between hemispheres, and 2) the statistical structure among the surface representations of various categories of entities. The empirical work will combine analyses of a) the structural distribution of lesions in patients with semantic impairments, b) the distribution of their behavioral impairments across categories, c) distributions of brain activations in normal subjects performing the same behavioral tasks, and d) abnormal activation patterns in structurally normal brain regions in the patients. Findings from these studies will inform the development of a computational model employing realistically structured perceptual, linguistic, and action-based surface representations and incorporating explicit neuroanatomic structure. Hemispheric differences will be built into the model but the relative degree of participation of brain areas within each hemisphere will emerge from the operation of general learning principles in concert with architectural constraints. The model will be evaluated both in terms of its ability to exhibit the appropriate selective semantic impairments following damage, and whether it exhibits the type of functional specialization across brain areas identified in the empirical studies.
Showing the most recent 10 out of 98 publications
