Family Resemblance in Unsupervised Categorization: A Dissociation Between Production and Evaluation

AbstractA plurality of the categories we hold exhibit family resemblance (FR; i.e., many characteristic but few defining features), suggesting FR may occupy a central role in human category formation. However, research in unsupervised learning has shown that when people are asked to sort an array of novel items into categories, they ubiquitously use a unidimensional (UNI) rule – despite the availability of a FR solution. This work suggests that, perhaps, FR similarity is not a core tendency in category formation. Here, we question whether the UNI bias is a result of the sorting paradigm. Specifically, we speculate the paradigm conflates two components vital for category formation: production and evaluation. Across three experiments we show that when evaluation is separated from generation – by using a novel forced-choice task that pits different category organizational schemes against one another – people exhibit a FR over UNI preference. The implications of these results are discussed.


Return to previous page