Using Natural Language Processing to Inform Science Instruction

点击翻译按钮获取中文摘要

Often, middle school science classes do not benefit from participation of underrepresented students because of language and cultural barriers. This project takes advantage of language to help students form their own ideas and pursue deeper understanding in the science classroom. This work continues a partnership among the University of California, Berkeley, Educational Testing Service, and science teachers and paraprofessionals from six middle schools enrolling students from diverse racial, ethnic, and language groups whose cultural experiences may be neglected in science instruction. The partnership will conduct a comprehensive research program to develop and test technology that will empower students to use their ideas as a starting point for deepening science understanding. Researchers will use a technology that detects student ideas that go beyond a student's general knowledge level to adapt to a student's cultural and linguistic understandings of a science topic. The partnership leverages a web-based platform to implement adaptive guidance designed by teachers that feature dialog and peer interaction. Further, the platform features teacher tools that can detect when a student needs additional help and alert the teacher. Teachers using the technology will be able to track and respond to individual student ideas, especially from students who would not often participate because of language and cultural barriers. This project develops AI-based technology to help science teachers increase their impact on student science learning. The technology is aimed to provide accurate analysis of students' initial ideas and adaptive guidance that gets each student started on reconsidering their ideas and pursuing deeper understanding. Current methods in automated scoring primarily focus on detecting incorrect responses on test questions and estimating the overall knowledge level in a student explanation. This project leverages advances in natural language processing (NLP) to identify the specific ideas in student explanations for open-ended science questions. The investigators will conduct a comprehensive research program that pairs new NLP-based AI methods for analyzing student ideas with adaptive guidance that, in combination, will empower students to use their ideas as starting points for improving science understanding. To evaluate the idea detection process, the researchers will conduct studies that investigate the accuracy and impact of idea detection in classrooms. To evaluate the guidance, the researchers will conduct comparison studies that randomly assign students to conditions to identify the most promising adaptive guidance designs for detected ideas. All materials are customizable using open platform authoring tools. The Discovery Research PreK-12 program (DRK-12) seeks to significantly enhance the learning and teaching of science, technology, engineering and mathematics (STEM) by preK-12 students and teachers, through research and development of innovative resources, models and tools. Projects in the DRK-12 program build on fundamental research in STEM education and prior research and development efforts that provide theoretical and empirical justification for proposed projects.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.