Brian H. Ross - US grants

This project is being funded through the Learning and Intelligent Systems (LIS) Initiative. Concepts are essential for intelligent thought and action. The goal of the project is an integrated view of concept learning in humans and machines. The primary focus will be combining psychological experimentation with artificial intelligence modeling to examine the interaction of world knowledge and empirical information during concept learning. The representation of concepts consists of feature regularities observed in the instances and of features inferred from world knowledge. However, current theories focus on only one type of feature and do not consider how learning each might affect the other. Additional work will examine how the use of concepts (such as those used for problem solving) may affect learning, how prior knowledge may be restructured to accommodate new information, and how concepts may change with age and experience. Computational learning theory will be adapted to provide a mathematical characterization of the learning process. The view of concept learning that results from this work will be integrated in that it will (a) investigate and account for a wide variety of concept learning results that are often studied separately, and (b) pool the research strengths of psychology, artificial intelligence machine learning, and computational learning theory. The first goal will place greater constraints on theoretical accounts, suggest new possibilities, and help to decide among competing explanations. The second goal will lead to a theory that is psychologically and computationally plausible, yet sufficiently rigorous to be analyzed with the mathematical tools of computational learning theory. Such a theory will contribute to the generation of new knowledge by broadening the understanding of concept learning in each of the fields, and by promoting new research issues and approaches in each field through interdisciplinary work.

This project will develop and test psycholinguistic models of the relationship between first language fluency, second language competence and second language fluency. These models will be applied toward the automatic assessment of fluency in a second language. The project involves a unique collaboration between researchers with backgrounds in second-language pedagogy, testing methodology, linguistics, speech and language technology, and psychology, and is organized around a common set of data, namely oral presentations given by university students in third year Mandarin classes. These student performances are videotaped, transcribed and rated by trained raters who rate the students' fluency according to a custom-designed and validated testing procedure. The same students will be recruited at the beginning of the semester to participate in psycholinguistic experiments to measure their first language fluency, and related studies will be conducted during the course of the year. The results from expert rating of second-language fluency will be correlated with the psycholinguistic studies of first-language fluency. In parallel with this, the team will develop algorithms that will automatically assign scores to a student's second-language performance that will correlate with expert judgments. These algorithms will range from low-level signal processing methods to estimate such factors as syllable rate and pause duration, to Dynamic Bayesian Networks that combine information from large number of sources to improve the performance of Automatic Speech Recognition on the data. The results of this work will be both a better understanding of what it means to be fluent in a second language, as well as robust methods that will allow for objective automatic assessment of fluency.

A fundamental component of intelligent behavior is how people make inferences about novel objects, people, and situations. In category-based induction, people use their knowledge of categories to make such inferences about new items from those categories (Will that dog attack? Which treatment will help this patient?). The present research investigates how people make such predictions in a common situation, when the categories are uncertain. For example, if medical treatment is required before a final diagnosis is possible, how do people take account of the various possible diagnoses (categories)? Past research has shown people often make inductions poorly in such situations, focusing on the most likely category and ignoring other possibilities.

The proposed research examines the basis for these errors in explicit, conscious predictions and whether reasoning can be improved by including information from implicit, fast, unconscious predictions. Pilot work suggests that implicit predictions can be more accurate than explicit ones. The proposed research attempts to confirm this finding and explores the psychological mechanisms that underlie the improved reasoning.

This research will have two important contributions. First, it will help in understanding the differences and similarities between conscious and unconscious processes. This distinction is crucial in much research on mind and behavior, and the present research will help us understand why these differences exist. Second, the results may lead to specific proposals for ways to improve predictions in medical, personal, and economic decisions. When categories are uncertain, people often do not make good predictions, but decision aids that help include implicit, unconscious processes may improve performance.