The Emergence of Theoretical Beliefs as Constraints on Concepts
Frank C. Keil Cornell University
Cognitive psychology has recently embraced a view of concepts that has had a long tradition in the philosophy of science, namely that coherent sets of core beliefs, or "theories," are essential to a full specification of concept structure. Concepts cannot be represented merely in terms of probabilistic distributions of features or as passive reflections of feature frequencies and correlations in the world. Some of the most compelling demonstrations involve illusory correlations where prior theories cause people to create or enhance correlations that are central to their theories and ignore or discount equally strong correlations that are more peripheral to that theory. This phenomenon has been known for some time in the social and clinical psychology literature, such as in the illusory correlations in diagnoses made by clinical psychologists (e.g., Chapman & Chapman, 1969); but its greater relevance to most concepts is now being widely recognized ( Murphy & Medin, 1985).
There are many other problems with mere probabilistic models, such as demonstrations that equally typical (i.e., equally probabilistically associated) features may be dramatically different in how they affect judgments about the goodness of exemplars. Thus, Medin and Shoben ( 1988) have shown that, although curvedness is judged to be equally typical of bananas and boomerangs, straight boomerangs are considered to be much more anomalous members of the boomerang family than straight bananas in their family, because curvedness is seen as theoretically more central, that is, causally more critical to the "essence" of boomerangs. This finding is also further evidence against real- world correlations exclusively driving concept structure because, empirically, there are, in fact, some straight boomerangs and no straight bananas.