Abstraction of covariation in category learning: a critical note on K. Richardson's studies

Several studies directed by K. Richardson (Richardson, 1986, 1987; Richardson & Carthy, 1989, 1990) claimed that people are able to abstract imperfect covariations among feature variables defining artificial categories, and to use that information in various subsequent tasks. We challenge this claim on a twofold basis. First, a large part of the performance variance Richardson attributed to covariation knowledge may be accounted for by a very simple exemplar model, which assumes no abstractive processes. Second, the residual influence of relational information on performance may be attributed to the knowledge subjects have acquired in real-world situations before their training with the study exemplars.