SBIR Phase I: Virtual Learning Assistants for Constructed Response Assessment

This SBIR Phase I project focuses on creating scalable Virtual Learning Assistant (VLA) technology for constructed response assessment. The best pedagogies responsible for improving learning outcomes generally involve (i) constructed response assessments and (ii) one-to-one tutoring. Students learn the best when they are given an opportunity to construct answers in their own words (instead of selecting from multiple choices) and when they receive immediate guidance and coaching in a one-to-one conversation with a human tutor. However, the costs and time associated with the constructed response assessment and one-to-one tutoring are significant, making them very difficult to scale. The proposed project will apply the most advanced technologies such as Artificial Intelligence and Natural Language Processing to solve both these problems. Students will benefit from the interactive formative assessment that engages them in a natural language conversation. This innovation is applicable across the grade levels in K-12, higher education, and adult learning and across the subjects areas such as English language arts, STEM and humanities. It will facilitate implementation of more rigorous academic standards and make online education more effective. This innovation will improve students' learning outcomes, save teachers' time and reduce the cost of delivering high quality engaging education on a large scale. This project will create a new type of virtual assistant technology that is exclusively focused on education. The proposed Virtual Learning Assistant (VLA) will advance the conversational AI technology to create pedagogically rich learning and assessment environments for any topic in a content area. The VLA is uniquely distinct from general purpose virtual assistants in its ability to evaluate an answer instead of merely serving information. This project will investigate and create various algorithms for processing natural language input arising in an educational setting across different subjects or topics. The resulting web based product will allow teachers to create new high quality assessment items with minimal input and assign them to their students. When a student answers a question, the VLA will analyze it instantly for linguistic syntax and semantics using statistical and deterministic knowledge representations. The VLA will generate not only a numerical score reflecting the accuracy of the answer, but also a qualitative feedback that will guide the student towards conceptual mastery of the topic. As part of this Phase I research, a pilot study will be conducted involving teachers and students to study the efficacy of the VLA and to verify its usability and feasibility.