Poster D53, Monday, March 26, 8:00-10:00 am, Exhibit Hall C
How abstract concepts are neurally represented
Robert Vargas1, Marcel Just1; 1Carnegie Mellon University
Abstract concepts are typically defined indirectly as the opposite of concreteness, failing to specify their neural or cognitive basis. Contemporary neuroimaging techniques were applied to more directly characterize abstractness. Neural representations of 28 abstract concepts (defined as activation levels across 120 voxels with a stable semantic tuning curve across concepts) were assessed using fMRI data in 9 participants. A classifier (Gaussian Naive Bayes) trained on neural signatures in a subset of the data for each subject decoded the concepts in an independent subset (mean rank accuracy was 0.82, chance threshold = 0.53, p < 0.01). Representational commonality across participants was indicated by successful identification of concepts when trained on all but a left-out test participant (mean rank accuracy was 0.74, chance threshold = 0.54, p < 0.01). Factor analysis revealed 3 semantic dimensions underlying the activation patterns, providing a brain-based ontology for this set of abstract concepts. The 3 dimensions are: Verbal Characterization (degree a concept is defined in terms of other concepts (e.g. faith [more verbal] vs. gravity [less verbal]); Externality/Internality of the concept to the perceiver (e.g. deity [external] vs. sadness [internal]); and the Social Content of the concept (e.g. gossip [social] vs. heat [asocial]). A cross-validated generative model, using behavioral ratings for the 28 concepts’ along the 3 dimensions, provided converging evidence for the interpretation (mean rank accuracy = 0.71). In conclusion, representation of abstract concepts requires the activation of more complex neurocognitive functions (i.e. language, self representation, and social interaction) rather than separation from perceptual information.
Topic Area: LANGUAGE: Semantic