Observational Category Learning as a Path to More Robust Generative Knowledge

Models and theories of category learning may exaggerate the extent to which people adopt discriminative strategies because of a reliance on the traditional supervised classification task. In the present experiment, this task is contrasted with supervised observational learning as a way of exploring differences between discriminative and generative learning. Categories were defined by a simple unidimensional rule with a second dimension that was either less diagnostic (than the simple rule on the first dimension) or non-diagnostic. When the second dimension was less diagnostic, observational learners were more sensitive to its distributional properties compared to classification learners (though classification accuracy at test did not differ). Observational learners were also consistently more sensitive to distributional information about the highly diagnostic dimension. When the second dimension was non-diagnostic, neither learning group showed sensitivity to the distributional properties of this dimension.