Spontaneous Discovery and Use of Categorical Structures
Abstract
This research deals with unsupervised learning of categories (UL) and how such learning is affected by the sequencing of training instances. Two general models of UL are described, one based on learning explicit associations between correlated features (associative model), and the other based on creating distinct schemas to represent each category without explicit learning of feature correlations (schema-triggering model). An attribute listing paradigm was used as an index of UL in three experiments, each of which manipulated the order in which instances from two different categories were presented and evaluated the effects of this manipulation in terms of the two competing models of UL. Strong evidence was found for the use of a discrete schema-triggering process to learn the categories in these experiments. Moreover, these experiments demonstrate that the attribute listing paradigm can be used to trace learning functions for UL over a series of instances, enabling the future investigation of many independent variables using this task.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 15, 1992
- Accession Number
- ADA248343
Entities
People
- Gordon H. Bower
- John P. Clapper
Organizations
- Stanford University