Cree, G. & Benko, J.
Department of Psychology, University of Toronto Scarborough, Toronto, ON, Canada, M1C 1A4
It has been clearly demonstrated that the human semantic system picks-up on the statistical regularities with which features (e.g., <flies>) pattern across concrete noun concepts (e.g., hawk, eagle, toaster), and that these regularities influence processing in both on-line (e.g., priming) and off-line (e.g., typicality rating) behavioral tasks. It has been a challenge to extend this line of inquiry to abstract noun concepts because we lack a clear understanding of what the features of these concepts might be. An understanding will allow us to design better behavioural experiments, better computational models, and develop a better understanding of patient deficits. We will first present data from an ongoing serious of studies that explore the validity of different methods for deriving the features of abstract noun concepts, including feature norming, feature rating, and the mining of representations derived from semantic models based on analysis of word co-occurrence in large text databases. We then compare the concept similarity structures of representations derived from these methods, and compare these to results from a new method that derives saliency scores for semantic dimensions of abstract noun concept representations that can be tightly linked to known neural processing pathways/systems in the brain.