BackHome
Refresh: OffView Personal Schedule
Paper Summary
Share...

Direct link:

Using Referencing Dialogue With a Teachable Social Robot to Balance Student Participation

Sat, April 15, 2:50 to 4:20pm CDT (2:50 to 4:20pm CDT), Hyatt Regency Chicago, Floor: West Tower - Ballroom Level, Regency C

Abstract

*Objective and Theoretical Framework*

Successful collaboration necessitates balanced participation with “a sufficient degree of turn taking” (Chi et al., 2014, p. 223), however this may not occur naturally. One potential solution to facilitate collaborative learning is student referencing, or using students’ names to influence participation (Komatsubara et al., 2014). Research has shown that referencing a less proficient collaborator improves the interaction over addressing a prompt to the whole group (Tegos et al., 2014). However, it may seem awkward or cause undue stress to over-reference that participant and exclude the other participant. Thus, in human-computer environments, different forms of artificial intelligence must be enabled to mimic adaptive scaffolding techniques. Under what contexts and with what frequency an AI should reference different parties in a collaboration is an open question.

Our work builds off previous work in which an individual middle-school student taught a NAO robot (i.e., Emma) how to solve math problems (Authors). This involves students talking to the robot in natural language, and the robot asking questions and providing explanations. In this project, we set out to investigate how a teachable robot can use dialogue to encourage students to participate equally in an interaction.

*Methods*

We expanded previous work for collaborative settings through a pilot study with undergraduate students (N = 28, 14 human-human-robot groups) to understand how students collaborate in these environments and potential directions for alterations. A key finding was that the dyads exhibited diverse patterns of participation (e.g., taking turns with different problems, or speakers with higher prior knowledge dominating the discussions). The patterns led to less balance of participation in some groups, which, as described above, influences learning outcomes.
As such, for our next iteration, we added student-referencing dialogue into Emma’s speech patterns. Emma has two main types of utterances: questions and statements. Since statements would not require a response (e.g., “cool”), we only added referencing to questions. Given the exploratory nature of this iteration, we randomly assigned student-referencing, which would allow us to analyze the specific in-the-moment effects of referencing and detect meaningful patterns that could lead to balanced participation. Each student had a 25% chance of being referenced at the beginning of a question, meaning that different students could be referenced within the same math problem.

*Results*

We tested the student-referencing with a small pilot group of 9th-grade students (N = 8, four groups). Although dominant personalities and differences in prior knowledge were quickly detectable in our preliminary analyses, it was clear that when Emma referenced certain students, a kind of “ownership” was established. Some students would direct others how to respond but still included them rather than just answering themselves without discussion. Others would try to convince the “owner” to see it their way. Such ownership increased turn-taking within a problem.

*Scholarly Significance*

Next steps involve using this data to determine when and how to include student-referencing to optimize student participation and learning using reinforcement learning (Park et al., 2019), which has scholarly implications for measuring collaborative learning.

Authors