Using Multivariate Pattern Analysis to Identify Conceptual Knowledge Representation in the Brain
Date of Award
Open Access Dissertation
Svetlana V Shinkareva
Representation of semantic knowledge is an important aspect of cognitive function. The processing of concrete (e.g., book) and abstract (e.g., freedom) semantic concepts show systematic differences on various behavioral measures in both healthy and clinical populations. However, previous studies examining the difference in the neural substrates correlating with abstract and concrete concept representations have reached inconsistent conclusions. This dissertation used multiple novel data analyses approaches on functional magnetic resonance imaging (fMRI) data, to investigate representational differences of abstract and concrete concepts and to provide converging evidence that the representations of abstract and concrete semantic knowledge in the brain rely on different mechanisms.
Study 1 used meta-analysis method on a combined sample of 303 participants to quantitatively summarize the published neuroimaging studies on the brain regions with category-specific activations. Results suggested greater engagement of working memory and language system for processing abstract concepts, and greater engagement of the visual perceptual system for processing of concrete concepts, likely via mental imagery. Study 2 showed successful identifications of single trial fMRI data as being associated with the processing of either abstract or concrete concepts based on multivoxel activity patterns in widespread brain areas, suggesting that abstract vs. concrete differences were represented by multiple mechanisms. Study 3 investigated the classification based on condition-specific connectivity patterns. Results showed successful identifications of the
connectivity patterns as abstract or concrete for an individual based on the connectivity patterns of other individuals, both by the connectivity for a priory selected seed regions as well as by the whole-brain voxel-by-voxel connectivity patterns. The results indicated the existence of condition-specific connectivity patterns that were consistent across individuals on a whole-brain scale. Moreover, the results also suggested the representation of abstract and concrete concepts differs from the semantic association perspective in addition to differences on coding forms. Study 4 illustrated the application of MVPA as a cross-modal prediction approach, which is a promising method for further investigation of semantic knowledge representation in the brain, by investigating the role of general semantic system on person-specific knowledge.
Overall, the work described in this dissertation provides converging evidence of the representational difference between abstract and concrete concepts. The differences are suggested to occur at various levels, including the dependence on modality-specific perceptual systems, the organization of associations among different semantic-related systems, and the difficulty and strategy of retrieving contextual information.
Wang, J.(2013). Using Multivariate Pattern Analysis to Identify Conceptual Knowledge Representation in the Brain. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/2697