Does activation really spread?

Abstract

In this article we consider three versions of the spreading activation model for retrieval of information from memory. Predictions are derived for activation spreading down a linear network structure as a function of time. To test these predictions, two experiments were performed in which facilitation as a function of time was measured for target words in a linearly structured paragraph. A target word was primed either by a word near to the target in the paragraph structure or by a word far from the target. Results showed that facilitation begins at about the same time for the far and near conditions. This is inconsistent with the predictions derived from the models. Implications of these results are discussed and alternative conceptions of the activation process are described. In this article we discuss the theoretical and empirical status of the spreading activation process. It is important to investigate this process, because it forms the central mechanism of two current models of memory—a model proposed by Anderson (1976) in the areas of memory, language, and thought, and a model proposed by Collins and Loftus (1975) in the area of semantic memory—and because it is becoming accepted as a mechanism for processes involved in text comprehension (Kieras, 1981; Miller, 1981). Spreading activation has also formed the basis of models in artificial intelligence (Fahlman, 1981; Hinton, 1981; Quillian, 1967), problem solving (Levin, 1976), language understanding (McDonald & Hayes-Roth, 1978), word recognition (McClelland & Rumelhart, 1981), and word production (Dell, 1980; Dell & Reich, 1977). Although we are concerned in this article with spreading activation as a process of retrieving information from memory, and therefore are most directly concerned with the models of Anderson and Collins and Loftus, the discussion presented here should also have implications for the other areas of research.