A General Learning Theory and its Application to Schema Abstraction.
Abstract
ACT is a computer simulation program that uses a propositional network to represent knowledge of general facts and a set of productions (condition - action rules) to represent knowledge of procedures. There are currently four different mechanisms by which ACT can make additions and modifications to its set of productions as required for procedural learning: designation, strengthening, generalization, and discrimination. Designation refers to the ability of productions to call for the creation of new productions. Strengthening a production may have important consequences for performance, since a production's strength determines the amount of system resources that will be allocated to its processing. Finally, generalization and discrimination refer to complementary processes that produce better performance by either extending or restricting the range of situations in which a production will apply. These learning mechanisms are used to simulate experiments on schema abstraction by Franks and Bransford (1971), Hayes-Roth and Hayes-Roth (1977), and Medin and Schaffer (1978). The mechanisms are used to predict recognition trials to criterion, as well as final test recognition and classification. ACT successfully accounts for the effects of distance of instances from a central tendency, frequency of individual instances, and inter-item similarity. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 15, 1978
- Accession Number
- ADA064820
Entities
People
- Charles M. Beasley Jr
- John R. Anderson
- Paul J. Kline
Organizations
- Carnegie Mellon University