ABSTRACT
This chapter aims to address why some properties are more generalizable than others. It argues that centrality is concept-relative. “Roundness”, for instance, is central for Basketballs but not for Cantaloupes. The chapter shows that the more central a feature in a category’s representation, the higher its projectibility among concepts that share common structure. Models based on feature- similarity appeal only to the relations among categories to predict inductive strength; centrality has no place in their equations. Also, models based on structural alignment fail to account for such effects because for such models to work predicates, as well as their relation to other predicates, must all be clearly specified. Since the dependencies of the predicates in the study were left vague, it is unclear how such models could apply.
