| Abstract: | The rational analysis of memory (Anderson, 1990) proposes that memory's sensitivity to statistical structure in the environment enables it to optimally estimate the odds that a memory trace will be needed. We have analyzed sources of informational demand in the environment: speech to children and word usage in the front page headlines of the New York Times. In a previous paper (Anderson & Schooler, 1991) we have shown that factors that govern memory performance, including recency, also predict the odds that an item (e.g., a word) will be encountered. In the present paper we develop the theory to make precise predictions about how the odds of encountering an item now varies as a joint function of (1) the statistical associations between the item and elements of the current context and (2) how long it has been since the item was last encountered. The prediction was confirmed environmentally by analyses of the New York Times and speech to children. The corresponding behavioral prediction was tested, using a cued recall task in which the cues were either strongly associated or unassociated to the targets. In contrast to the environmental results, recall performance is more sensitive to the length of the retention interval in the presence of unassociated cues than in the presence of associated cues. Further modeling shows that incorporating estimates of the influence of non-retrieval processes (e.g., reading a word, deciding to respond, etc.) on overall performance reduces the discrepancy between the theoretical predictions and the observed data. |
