9510585 AHN How do people learn new concepts? Previous studies in categorization fall into two camps, similarity-based and theory- based views. The similarity-based view assumes that natural categories in the world look coherent to us because members in the same categories are similar to each other whereas the members in different categories are dissimilar to each other. The theory-based view rose in reaction to the similarity-based view, which tended to neglect the role of existing background knowledge. Many researchers have shown that background knowledge or context drastically changes similarities among object. For example, a person from Maine and a person from Florida might look very different from each other if they are compared in Washington D.C. But if they are compared in Tokyo, they would look very similar to each other because the feature, "being an American" would become salient in the context of Tokyo. One's background knowledge also plays a crucial role in categorization. For example, students learn to classify a whale as a mammal because of their background knowledge on mammals, although a whale is perceptually more similar to fish. This research will focus on the structural or syntactic aspect of causal background knowledge. That is, how does the status of features in causal structure determine various similarity functions and categorization processes? Suppose a patient displays typical symptoms of Disease X except for one symptom. The idea is that the impact of this missing symptom in further diagnosis or categorization depends on the causal status of the symptom in the clinician's background knowledge. The long-term goal of this project is to develop a computational model of categorization incorporating causal knowledge. Developing a computational model is very important for both theoretical and practical reasons. On the theoretical side, a computational model will allow researchers to generate precise predictions to be tested t hrough experiments. On the practical side, a precisely defined computational model can serve as a basis for tutoring systems which can diagnose students' background knowledge and tailor instruction for each individual. Teachers can also be aware of biases that students might display in learning new concepts because of their initial background knowledge. This kind of computational model can also be used for developing expert systems which can spontaneously learn new concepts and serve as aids for human experts. This Research Planning Grant (RPG) will allow Ahn to collect pilot data on categorization and similarity judgment tasks which manipulate causal structures, and (2) to test existing computational models on these tasks. The aim of the RPG is to come up with a coherent framework for approaching the issues discussed above, so that the output of this preliminary research can be used to develop a full proposal for the regular funding competition. ***
