Journal of Experimental Psychology: Learning, Memory, and Cognition 1985, Vol. II, No. 1, 85-93 Copyright 1985 by the American Psychological Association, Inc. O278-7393/85/SOO.75 Effects of Semantic Relatedness on Same-Different Decisions in a Good-Bad Categorization Task James A. Hampton and Peter J. Taylor The City University, London, England Schvaneveldt, Durso, and Mukherji (1982) investigated the effect of relatedness on six kinds of same-different categorization tasks. They discovered two distinct patterns of results. For tasks involving surface features of words, relatedness facilitated both same and different judgments equally, whereas for tasks requiring a semantic analysis, relatedness facilitated same judgments but had no effect on different judgments. The only task that did not conform to this division was judgment of good versus bad, which showed the same pattern as surface-feature tasks. The present two experiments showed that this anomaly was due to the use of antonym word pairs for this task. When nonantonyms are used, there is no facilitation of different judgments by relatedness. The nature of antonymy as a semantic relation is discussed. In a recent article, Schvaneveldt, Durso, and Mukherji (1982) discovered two distinct classes of same-different categorization tasks. They were investigating the effects of semantic relatedness on same-different category judg- ments, using six different tasks that varied in the depth of processing required. The partic- ular advance that they made over previous research on this question (Glass, Holyoak, & O'Dell, 1974; Schaeffer & Wallace, 1970) was in devising materials for which relatedness could be manipulated independently of whether a same or a different response had to be made. Thus, a lack of semantic relat- edness between a pair of words could not be used as the basis for a different decision (as had been possible in earlier studies). They discovered that for judgments involving vowel-consonant (as initial letter of the word), word-nonword (where the nonword was a misspelled word), and good-bad decisions, a particular pattern of results could be obtained. For these three tasks, semantic relatedness facilitated both same and different responses equally. For judgments involving plant-ani- mal, natural-manmade, or noun-verb, how- ever, a different pattern was found. For these tasks, relatedness facilitated same decisions, The authors acknowledge the help of Frances Guy, who collected the data for Experiment 2. Requests for reprints should be sent to James A. Hampton, Psychology Division, The City University, Northampton Square, London EC IV OHB England. but had a slight inhibitive effect on different decisions. The six tasks thus formed two distinct classes, distinguished by the effect of relatedness on the different judgments. In discussing their results, Schvaneveldt et al. (1982) ruled out an explanation of the different patterns in terms of a distinction between semantic and nonsemantic tasks be- cause of the good-bad task, which produced a result similar to the other surface-feature tasks. After considering various accounts of the results, they finally proposed a spreading- activation intersection model, in which acti- vation spreads out from each concept and activates both related concepts and related semantic features. Features activated by both words in a pair will fall in the intersection of activation and must then be scanned for information relevant to the particular decision required. Initially, therefore, lexical retrieval of the two words will be faster if they are related because they will activate each other. In the case where the decision is based on purely lexical information—which in the two tasks used by Schvaneveldt et al. (1982) was orthographic information (initial letter, or correct spelling)—there is no need to scan the intersecting semantic connections between the words. Thus, relatedness has a small and equal facilitatory effect on both same and different responses. However, when the cate- gorization involves a semantic analysis, the existence of irrelevant semantic information 85