Hi Susan - thank you so much for watching our video! To answer your first question regarding how students develop math ideas, one of the goal's of the teachable robot is to help cement understanding through teaching - by teaching, learners may attend more to the problems, consider their own misconceptions, and elaborate on their own knowledge to construct explanations. In order to guide the learner's teaching, we provided the worked-out solutions to the problems the learners taught Nico and gave them time to review the solutions before they began teaching. Then, while they were teaching Nico, Nico would prompt for explanations from the learner and ask questions about the learner's approach to solving a particular problem. Nico would also elaborate, so for example, the learner might say "Now multiply by two" and Nico would respond with, "Oh okay. We multiply by two. Is that because we have twice as many bags?" Engaging in the activity of explaining and responding to Nico facilitated learning by encouraging learners to reflect on what they know. The participants interacted with Nico for 30 minutes in a single interaction. To measure learning, we utilized a pretest-posttest design with an A and B form of the test. The two forms were isomorphic and counterbalanced within condition (half of the participants in each condition received test A as the pretest and test B as the posttest, and vice versa). The tests consisted of 10 procedural and conceptual questions around ratios. We then calculated the normalized learning gain. The numbers reflected in the slide are the average normalized learning gains per condition (a social condition, where the robot spoke socially, an adaptive social condition where the robot adapted and spoke socially, and a control condition where the robot was neither social nor adaptive).

Hi Teruni - we appreciate your questions!  The robot responds via spoken dialogue that is generated via a dialogue system consisting of an automatic speech recognition engine (via the Web Speech API), a dialogue manager, a text-to-speech engine (built into the Nao robot), and feature extraction/transformation module. The dialogue manager utilizes a chat-bot based design to generate responses using artificial intelligent markup language. With this, we introduce some social dialogue responses such as praise and using the student's name. We also have the ability to adapt to paraverbal features of speech such as tone of voice, loudness and speaking rate. In the learning results presented, the robot adapted to the student's tone of voice, or pitch, modeling a human-dialogue phenomenon known as entrainment. With entrainment, people adapt to one another to become more similar over the course of a conversation. Nico modeled this behavior and we found that when Nico adapted on pitch and spoke socially, the participants had higher learning gains. You can find out more about the dialogue system and these results in these two papers...

Lubold, Nichola, et al. "Automated pitch convergence improves learning in a social, teachable robot for middle school mathematics." International Conference on Artificial Intelligence in Education. Springer, Cham, 2018.

Lubold, Nichola. Producing Acoustic-Prosodic Entrainment in a Robotic Learning Companion to Build Learner Rapport. Diss. Arizona State University, 2018.

Ooops missed your other questions! The role of the teacher involves helping Nico solve Nico's math problems - we provide a narrative to explain why Nico wants to answer the ratio-based problems that are posed (Nico is going camping with its friends). There is only one robot however so we had the robot in a separate room and each student came in, individually, to interact with it. 

Hi Beth - thank you for watching it! For now, we are purely exploring basic research questions around the design of teachable robots, how they should behave, and how we can use robots like Nico to explore open questions learning, such as the role of rapport in influencing learning between peer tutors and building self-efficacy. This last aspect of Nico can provide insight into immediate strategies for teachers in the classroom. In terms of Nico, there is only one robot so it is currently not deployable to all students in a classroom. Erin Walker, one of the presenters on this video, has explored using teachable robots with groups of students versus the one-on-one interaction we designed with Nico. This may have some promise for teachers being able to incorporate something like Nico in the near term. There are also explorations of whether virtual agents can have similar effects. Let me know if you have any additional questions about this.    

Hi Denise - thank you for your questions and I definitely see potential for Nico with students with special needs. We did not explore this with any of the students who interacted with Nico, but I know that there are groups who interested in this and have shown positive results with other kind of interactive robots. Definitely something to think about for future work.

We did not measure feelings of responsibility but we did collect measures of self-efficacy and self-reported rapport as well as asking students about their goals. We also coded for rapport-building dialogue on the part of the student. We are in the process of exploring the self-efficacy and goals. For self-reported rapport, we did not find that the learners differed significantly across conditions - in general, they all really liked the robot and reported fairly high self-reported rapport. However, when we looked at rapport-building dialogue (i.e. did they praise the robot, use Nico's name, etc.), we found that learners who taught a robot that used social dialogue and adapted to them used more rapport-building dialogue and that this was positively correlated with ratings of self-reported rapport.  This seems to indicate that the robot's behavior did influence how they felt (or at least behaved) towards to the robot. 

That's a great question!  We did not compare directly (i.e. we did not have a condition) where students taught one another versus teaching the robotic companion so we can only comment on what we observed from students teaching the robot versus other prior studies of human peer-to-peer tutoring, and it is particularly challenging as the robot behaves more consistently than human partners in human peer-to-peer tutoring. We did design the dialogue responses of the robot based on observations of students teaching one another from prior studies.

While not necessarily an answer to your question, one observation I can make is that different students had different responses to teaching the robot; for example students who were more comfortable with robots were more likely to accept social behavior from Nico. These findings are less likely to transfer to human-human peer tutoring. On the other hand, finding that higher rapport and higher learning were associated with social behaviors that were based on what we observe people do provides some insight into what might make peer-to-peer tutoring successful.    

Thank you!  That sounds like an awesome project; looking forward to hearing more about it.