Learning and generalization as a function of complexity, parity, and abstraction within two primitive Boolean families

by Hammerly, Mark D.

Abstract (Summary)
The question of what makes a given Boolean concept harder or easier to learn than another has been extensively investigated. Recently, Feldman (2000) argued persuasively that much of the difficulty inherent in learning a Boolean rule stems from two orthogonal factors: complexity, which is a measure of the logical incompressibility of the Boolean function, and parity, which refers to the number exemplars placed in the positive category. The present experiment sought to determine the individual contribution of each to the learning process by holding one constant and varying the other. 195 subjects were randomly assigned to one of fifteen different groups formed from the factorial combination of rule order (complexity increasing, complexity decreasing, parity increasing, parity decreasing, control with both forces static) and stimulus set abstraction (shapes, gauges, medical diagnoses). Subjects learned three bidimensional rules from two primitive Boolean families theorized to be ‘building blocks’ for more complex concepts (Neisser and Weene, 1962). They were then asked to transfer this knowledge to a rule within the same stimulus set that had higher complexity and a third relevant dimension. Dependent measures were learning speed and transfer of training. Non-parametric statistics were used to carry out omnibus analyses and planned comparisons due to a pronounced skew in the data. Complexity overall had a much greater affect on learning speed than either parity or a control. It is speculated that differences between parity and a control might only be revealed in the presence of a covarying index of complexity. There was no order, increasing or decreasing, across either complexity or parity, which promoted transfer. Medical diagnoses were significantly easier to learn that either shapes or gauges, while training on gauges led to significantly worse transfer of training. The former result is theorized to arise from the need to translate visual stimuli into coherent linguistic statements suitable for hypothesis-testing, while the later is ascribed to the unique combination of visual complexity and this ‘translation factor.’
Bibliographical Information:


School:Miami University

School Location:USA - Ohio

Source Type:Master's Thesis

Keywords:learning generalization transfer of training boolean logic concept attainment stimulus abstraction


Date of Publication:01/01/2003

© 2009 All Rights Reserved.