Encoding in visual pattern perception

Abstract

A number of investigators have suggested that humans make use of the schematic (or redundant) aspects of a family of patterns in tasks involving memory by encoding those aspects which partially describe every member of the family. The patterns used in the present research were generated by a computer program, VARGUS 7, which allowed quantification of the patterns in information terms and independent manipulation of several relevant variables. The schema in patterns produced by VARGUS 7 is defined as the most probable sequence (MPS) of column heights as determined by the transitional probabilities of a seven element Markov process. Redundancy (schema) is determined by the magnitude of the probability associated with the MPS. The two experiments comprising the present research focused on schema learning and schema transfer. A transfer design was used in Experiment 1 to investigate schema learning (redundancy encoding) in a reproduction task. An experimental group reproduced 67% redundant patterns, containing the same MPS, in the train and teat phases while a control group reproduced 0% redundant patterns in the train phase and 67% redundant patterns in the test phase. Hypothesis l proposed that schema learning would be evidenced by the following results: (a) the experimental group would perform significantly better in the teat phase than in the train phase (b) in the test phase, the experimental group would be superior to the control group. The results confirmed these predictions and further indicated that schema learning apparently occurs without awareness and does not depend upon external reinforcement (knowledge of results). Because the patterns used in Experiment 1 constituted a random sample from a defined population, the results were expected to apply to samples of patterns drawn from other such populations. Hypothesis 2 proposed that the equation describing reproduction performance as a function of trials in Experiment 1 would accurately predict reproduction performance with the sample of patterns used in Experiment 2. The results confirmed hypothesis 2, showing that (with adjustment for level of difficulty) performance was predicted in samples of patterns drawn from different populations with only 4% error variance. Experiment 2 was designed to determine whether practice in learning successive schemata leads to a general improvement in schema learning. A transfer group reproduced 67% redundant patterns containing each of three successively presented schemata in the train phase and 67% redundant patterns containing a fourth schema in the test phase. A control group reproduced 0% redundant patterns in the train phase and 67% redundant patterns containing the fourth schema in the test phase. Hypothesis 3 proposed that schematic transfer would be evidenced by the transfer group's performing significantly better than the control group in the teat phase. The results confirmed this hypothesis. The results also indicated that investigators using traditional transfer of training designs might have reached opposite conclusions if additional tasks had been introduced or if performance on different training trials had been evaluated.