Project Title: Collaborative Research: A Collaborative Dialogue Agent for Peer Learning Interactions
This project is developing a computational model of effective episodes within peer learning dialogues and embodying that model in KSC-PaL, an artificial agent that can collaborate with a human partner via natural-language dialogues. (KSC stands for "Knowledge Sharing and Construction.") KSC-PaL is a flexible agent, insofar as it can assume various roles-in particular, novice peer and expert peer or "coach." As such, it could move educational applications from the paradigm of the computer as teacher to the computer as a peer who dynamically shifts from novice to expert, as occurs naturally during human interactions.
Although KSC-PaL is being developed to help Computer Science (CS) students develop abstract, generalizable knowledge about data structures and algorithms, there is the potential to scale up the dialogue agent architecture for other domains. By first using KSC-PaL as a peer in CS education, the investigators hope to increase retention in undergraduate CS programs, especially among female students. Peer collaboration on programming tasks in classroom and laboratory settings has been shown to benefit both male and female students.
Experimentation during the project is focused on establishing that the peer agent impacts learning positively, above and beyond a simulated coach or non-interactive control, and on isolating the effectiveness of particular features of peer-student interactions, such as different negotiation strategies. Results of these experiments will be disseminated in conference papers and journal articles and made available via the web.
