Abstract
The research on independent subproblem learning is reviewed. Subproblems are the individual pairs of stimuli presented to the subject on each trial of a discrimination problem. The present research seeks to determine: (a) do pigeons learn subproblems independently during the original and reversal shift problems? (b) does overtraining have an effect on independent subproblem learning during a reversal shift? and (c) does independent subproblem learning place other important discrimination learning studies such as Krechevsky (1932) and Lawrence and Mason (1955) in a new perspective?Thirty-two domestic pigeons (Columbia livia) were reduced to 75% of their free feeding weights, then given key peck training. Each pigeon was then trained on a series of three discrimination problems, the original problem and two shift problems. There were three independent variables: (a) the relevant dimension of the stimuli (position or color); (b) the degree of training during the original problem (criterion or 200 overtraining trials; and (c) the type of shift following the original problem (reversal or nonreversal). The third problem for all pigeons was a nonreversal shift.During the original learning problem six pigeons showed independent subproblem learning when color was the relevant dimension but only one showed independent subproblem learning when position was the relevant dimension. It was shown that the subproblem functions of pigeons exhibiting independent subproblem learning were similar to the systematic response patterns reported by Krechevsky (1932).During shift learning, with color relevant 38% of the pigeons showed independent subproblem learning; with position relevant 100% of the pigeons showed independent subproblem learning during a nonreversal shift, but zero percent showed independent subproblem learning during a reversal shift. There was no overtraining reversal effect during the reversal shift training. For pigeons given nonreversal shifts, overtrained pigeons did not require more trials to criterion than nonovertrained pigeons.During the probe nonreversal shift learning, with position relevant 100% of the pigeons exhibited independent subproblem learning; with color relevant 38% exhibited independent subproblem learning. There were no significant effects attributable to overtraining on the original problem or type of shift during shift learning. Analysis of the frequency of occurrance of independent subproblem learning during original, shift, and probe learning revealed that shifts from a color relevant to a position relevant problem resulted in independent subproblem learning 100% of the time. Shifts from color to color relevant and from position to color relevant resulted in independent subproblem learning 38% of the time. Shifts from position to position relevant resulted in independent subproblem learning 4% of the time. Comparison of the performance of pigeons learning a probe nonreversal shift problem which repeated or reversed the reinforcement contingencies of the original learning problem revealed that relearning the original problem required significantly fewer errors to criterion than a reversal of the original contingencies (replicating Lawrence and Mason, 1955). There was a trend toward independent subproblem learning occurring more frequently during relearning (75%) than reversal (25%) learning.Independent subproblem learning was discussed in terms of single-stage (continuity) and two-stage (noncontinuity) discrimination learning theory. Single-stage theory seems to provide the most satisfactory account of independent subproblem learning in animals.