Some Effects of Negative Instances on the Formation of Simple Concepts

Abstract

Smoke (1933) first suggested that the efficiency of concept formation might be affected by presenting instances of what a concept is not rather than instances of what it is. He did not, however, evaluate the amounts of information transmitted by these nedative and instances. Meaningful comparison of efficiency in dealing with instances must be based on the pocential of each type of instance to transmit information about a concept. If a negative instance provides only a slight amount of information relevanc to a concept, it could not be expected to be as effective for problem solution as a positive instance that provides a great deal of information. Tools for evaluating potential usefulness of different instances were proposed by Hovland (1952). If a problem involves an identifiable number of dimensions which may be relevant to the solution, and each dimension can assume a limited number of values, the number of possible concepts of a particular type can be listed. Each positive or negative instance presented to S eliminates some of these possibilities. The potential usefulness of an instance can be measured by the number of possible conceprs eliminated by its presentation. The positive or negative instances necessary to define a concept exactly and without redundancy can be listed. Hovland and Weiss (1953) compared the effectiveness of different combinations of positive and negarive instances in presentation of conjunctive concepts, using exactly the number of instances necessary to define the concept. They found thar problems defined by all positive inscances were more easily solved than those defined by all negative instances. Problems defined by a mixture of positive and negative instances were intermediate in difficulry. In general, positive instances transmit more information per instance in conjunctive concepts than do negative instances. The greater amount of information transmitted per positive instance may be important in ease of solution. The findings of Bruner, Goodnow, and Austin (1956) with disjunctive concepts suggest thar amount of information per instance is indeed important in ease of problem solution. With disjunctive conceprs, the informational value of positive and negative instances reverses, negative instances generally transmitting more information per instance. Here negative inscances result in more efficient problem solution. There would be no discrepancy between amounts of information trans'This investigation was suppozted by a Public Health Service research fellowship (MF12, 979), from the National Institute of Mental Health.