Nelson, Jonathan
Coauthors(s): Prof. Javier Movellan, Institute of Neural Computation and Department of Cognitive Science, UCSD
UCSD
Cognitive Science
UCSD Cognitive Science Dept. 0515 9500 Gilman Dr. La Jolla, CA 92093-0515
cogsci.ucsd.edu/~jnelson

Concept induction in the presence of uncertainty

In many situations, possible explanations for cognitive and perceptual data are vastly underdetermined by available evidence. Humans are adept at finding reasonable explanations in many of these situations, yet understanding this process has been a persistent challenge for cognitive science. Tenenbaum (1998, 1999, 2000) presented a simple probabilistic model to explain how humans could make inferences in situations that appeared underdetermined by available evidence. He found that his model predicted the average behavior of subjects well on several of these tasks. We present a large-scale replication (with 81 subjects, rather than 10) of Tenenbaum's initial study on a number categorization task. We found that his model accounted for 67% of the variance in average subject behavior. We consider the implications of measuring average data across subjects, rather than a model of prototypical individuals, in categorization tasks. Finally, we present an extension of Tenenbaum's model based on the theory of sequential statistical inference, to account for the information gathering behavior of subjects on this task.