Not Learning a Complex but Learnable Category
David J. Danks · eScholarship (California Digital Library) · 2006
Recent theoretical research has argued that multiple psychological theories of categorization are mathematically identical to inference in probabilistic graphical models (a framework developed in statistics and computer science).These results imply that the major extant psychological theories can all be represented mathematically as special cases of inference in (subclasses of) chain graphs, a particular type of probabilistic graphical models.These formal results suggest that people should be capable of learning significantly more complicated category structures than can be expressed in the standard psychological theories.In this paper, we present an experiment in which people apparently failed to learn the complex category, though a significant group of participants seemed to have learned something about the contrast category.Although inferences to cognitive failure are notoriously problematic, these results suggest that the hypergeneral theory useful for the mathematical equivalencies does not accurately describe human categorization.