This project seeks to achieve more efficient and accurate assessment of students' mathematical reasoning by developing improved speech recognition technology, calibrated to recognize children's speech, and integrating it with a computerized mathematics education environment (SimCalc) involving interactive representations and simulations. Specific questions of interest include: Q1. What are the most promising task-relevant mechanisms for constraining students? spoken responses in such a way that enables valid and reliable speech-based assessments despite the imperfect accuracy of automatic spoken language systems? Q2. What are the most efficient modifications to existing spoken language technologies that achieve acceptable performance with middle-school students engaged in spontaneous speech acts about mathematical models, representations, and simulations? Q3. Which aspects of the computer-based dynamic representation system must be integrated with student speech to provide a more complete representation of student knowledge of the mathematics underlying the model? Q4. Which features of the combined output of the computer-based environment and student speech allow the spoken language understanding engine to reliably assign rubric-based scores to student work?
Assessments will address five key aspects of fluency with mathematical models: comprehending, predicting, explaining, improving, and reflecting. A particularly unique focus of this work will be on combining inputs from the speech recognition engine with time-stamped information from a mathematics education environment to reliably score student responses according to a rubric. In addition to its application to education, this research on adolescents' spontaneous mathematical speech will drive advances in spoken language technology.
