Acquiring Generalizations to Organize Human Databases
Abstract
This report describes a program of empirical and theoretical research on how category-level generalizations facilitate human performance in learning tasks, especially when the categories are acquired in an unsupervised environment. An information-processing model is described in which people are assumed to spontaneously search for patterns and regularities among the training instances, and use them as a basis for forming general concepts. These concepts, in turn, enable learners to economize their encoding of further instances by focusing selectively on their most informative features. The resulting memory organization appear to optimize later access to the information from long-term memory. Several memory experiments are described which provide strong support for these claims. Two similarity experiments are also reported; these demonstrate that concept learning affects the evaluation and judgement of training instances as predicted by our theory, specifically, that comparison are strongly affected by the informative (surprising or unusual) features of the objects being compared. We also introduce a new procedure for observing the spontaneous acquisition of concepts in an unsupervised task. This task provides a trial-by-trial index of the strength of subjects' default generalizations about the concept. Keywords: Attention; Mental model; concepts/categories; Unsupervised learning; Memory encoding; Memory retrieval.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 26, 1990
- Accession Number
- ADA219835
Entities
People
- Gordon H. Bower
- John P. Clapper
Organizations
- Stanford University