Could you elaborate more on what your partner just said? Guiding dialogue in science education using a conversational agent
Adelson Dias De Araujo · 2024
Student dialogue is often not spontaneously productive. Teachers can provide guidance, but with multiple small groups to orchestrate, they find it challenging to scale effective guidance. Collaborative Conversational Agents (CCAs) seem to be a promising solution. This dissertation explores the use of Clair, the "Collaborative learning agent for interactive reasoning," in science learning tasks. Clair aims to facilitate productive student dialogue using "talk moves" based on the Academically Productive Talk (APT) framework. Clair's design incorporates learning analytics, machine learning, and a fuzzy rule-based system to trigger talk moves adaptively. Three classroom trials involved university student dyads and assessed Clair's effect on dialogue productivity. Findings indicated a significant improvement in two to three of these goals, suggesting Clair's potential to provide effective guidance. Still, the effects on students' perceived productivity and knowledge acquisition could not be detected. To conclude, the dissertation discusses key challenges for evaluating CCAs in education.