William K. Estes - US grants

The proposed research focuses on two general issues: (1) how are simple, natural concepts combined to form complex ones, and (2) what processes govern the course of learning novel categories. With regard to (1), our previous work led to a model of how people combine concepts like red and fruit into adjective-noun conjunctions like red fruit. The basic idea is that the noun concept is represented by default values on each of a set of attributes, and that the adjective modifies the value on one of the attributes. This model is among the first to offer a detailed account of how prototype concepts can be combined to form new prototypes. The model has implications for many areas of psychology and our proposed research will test some of these implications and extend the model to new domains. Some studies will evaluate the model's claims about real-time processes during categorization, while others will try to bridge the gap between categorization and psycholinguistics by showing that the categorization processes we have described in our model are involved in language understanding. Still other studies will extend our analyses to conjunctions involving verbs, as well as explore the relation between combining concepts and decision making. The research on category learning considers how experience with exemplars leads to the development of categories. The proposed work will concentrate on distinguishing similarity-based from likelihood-based models of category learning. Some of the experimental tasks simulate the problem of assigning patients to disease categories on the basis of symptom patterns, the task being accomplished either by judging the similarity of new to old exemplars of the categories or by computing the likelihoods of the categories given the symptoms. Correlations between symptoms will be introduced so as to enable differential predictions from similarity and likelihood based models. Our prior research on concepts turned up important implications for psychiatric diagnosis, as well as for personality categories and social stereotypes. We expect the current work to continue to have implications for diagnosis, stereotypes, and aspects of mental illness, particularly since some of the work on combining concepts involves personality concepts, while some of the research on category learning uses diagnosis as the experimental task.

This work involves the continued development and application of models for recognition memory. The first phase included formulation of the architecture and basic processing modules of a recognition model based on general theories of classification and memory; the model has been formulated with special attention to usefulness in increasing the informativeness of measures extracted from recognition data. A major published application reported a study of how performance on a test of recognition of an item or event depends on past frequencies of encounters with the same or similar stimuli and on presence or absence of positive or negative payoffs for correct responses on previous tests. The major result reported was that performance in recognition is not simply a matter of matching perceived with remembered stimulus patterns, but is influenced by many of the same conditions that control goal-directed behavior. At this juncture, a new set of benchmark experiments has been completed and is ready for model-based analysis. Several of the experiments were addressed to the problem of how prior familiarity with stimuli such as words or faces affects their recognition in a new situation; others deal with ways of enhancing the precision of measurement of the way factors such as stimulus frequency, stimulus exposure duration, and retention interval affect performance in situations that simulate eyewitness identification. Results from this series of experiments are ready for model-based analyses, which will be very computationally demanding, because all models and model versions examined will be fitted to the data of individual subjects, typically 30 to 80 in number per experiment. In all of this work, there will be major attention to advancing methodology for dealing with artifacts of averaging data and to communicating theoretical results in forms available to potential users in education, medicine, and technology.