Word Meanings in Semantic Memory: Separating, Defining and Characteristic Features

Three experiments were conducted to obtain some empirical generalizations about the structure of semantic memory. In the first experiment subjects were asked to give features for five category names as well as four instances per category. Instances varied on degree of relatedness to the category. In the second experiment subjects were asked to separate the features that they had given into (a) those that were essential for the meaning of a word (defining), and (b) those that were not absolutely necessary for the word (characteristic). In the third experiment a new set of subjects were given the same five category names and 2 instances out of the previous four, each word followed by a common pool of features that were generated by its category and instances in experiment l, and were asked to separate these features into essential, non-essential and not a property for each stimulus word. The predictions tested were that (l) subjects would produce a greater number of defining than characteristic features, (2) the first feature generated would be more frequently a defining than a characteristic feature, (3) defining features of the category would be fewer in number than defining features of the instances, and (4) the degree of relationship between the defining and characteristic features of the instance and the defining and characteristic features of its category would vary with the degree of relatedness of the instance to the category. Prediction l was not supported. Rather, it was found that characteristic features were more prevalent than defining features for both categories and instances. Prediction 2 was confirmed: there were more defining than characteristic features given as first response. Prediction 3 did not receive any support. Prediction 4 was confirmed for defining features, where it was found that decreasing levels of relatedness lowered the overlap between the instance and its category; however, the relationship between the characteristic features of the instance and its category remained at about the same high level at all degrees of relatedness. These findings, along with the fact that the categorization of features for instance words was not altered when Ss were given the category names1 features first, argue against the view that category names function as superordinates for instance in semantic -memory and in favor of the view that natural language categories possess an analog structure and that category names act as best exemplars of each category