Feature Overlap on Categorization in a Binomial Labeling Paradigm

Event Center A&B

This research investigates how common and distinctive features are weighed in free categorization, an essential human learning ability. Participants viewed 16 novel objects arranged in 4 x 4 arrays and were asked to divide them into categories by providing a binomial (family, species) label for each item.The objects were naturally divisible into three broad families. Members of each family were structurally alignable while sharing no specific features, with the exception of a single pair that shared from one to four (out of four) features in common. People were sensitive to overall alignability, being much more likely to assign the same family label to alignable than to non-alignable objects. People also show sensitivity to matching features within each family; the probability of assigning two instances their own distinctive species label increased exponentially with the number of features they shared. In a second experiment, participants rated the similarity of selected pairs of objects from the same 4 x 4 arrays on a 20- point scale. Alignable objects were rated as being much more similar than non-alignable objects, indicating a sensitivity to alignability-based categories. In contrast to the labeling data, rated similarity increased linearly, rather than exponentially, as the number of features shared by two alignable objects increased from one to four. Possible explanations for these somewhat divergent similarity effects, as well as future lines of research, will be discussed.