Conference Reports: Weiying Li at AERA 2023
We're so pleased to have been able to support our amazing students this year in sharing their incredible research at the leading conferences in their field. Below Weiying Li shares her experience at the AERA 2023 Conference on “Interrogating Consequential Education Research in Pursuit of Truth.”
The 2023 American Educational Research Association Annual Meeting is under the theme of "Interrogating Consequential Education Research in Pursuit of Truth". I reported my own research on "Responding to students’ science ideas in a Natural Language Processing based Adaptive Dialogue" in a poster session. Working with Prof. Marcia Linn, Prof. Libby Gerard, Allison Bradford, and Brian Riordan and Kenneth Steime from Educational Testing Services, we explored how Natural Language Processing dialogs that are designed following Knowledge Integration pedagogy elicit rich student ideas about photosynthesis and cellular respiration. We tested the dialog in 7th grade middle school classrooms with 162 students. The dialog asks students to explain how animals get energy from the sun to survive. Students receive adaptive guidance based on their response, followed by generic guidance asking about their uncertainties. We found that the adaptive guidance helped students link normative ideas with evidence as well as generate non-normative ideas that needed further attention. After the dialog, most students distinguished among all the ideas elicited and significantly improved their science explanations. Findings suggest that adaptive dialogs are a promising tool to scaffold science sense-making.
I also got chances to have conversations with other researchers within the edutech field and how they address critical issues there. In the symposium of Navigating the dilemmas, tensions, and contradictions of teaching critical issues in the pursuit of “truth”, the panelists discussed the complexities, tensions, and contradictions that arise when educators attempt to address “truth” and examine critical issues in the technology settings in the classroom. From another poster by Ha Nguyen from Utah State University, she found that most NLP models’ responses to the intersectional identities in climate change were overly sympathetic or uncertain, spread stereotypes, or ignored the speakers’ identities. As researchers we need to debias the models by human validation and participatory design