Engaging learning groups using Social Interaction Strategies

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 677–680, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Engaging learning groups using Social Interaction Strategies Rohit Kumar Carolyn P. Rosé Language Technologies Institute Carnegie Mellon University, Pittsburgh, PA, 15213 rohitk@cs.cmu.edu cprose@cs.cmu.edu Abstract Conversational Agents have been shown to be effective tutors in a wide range of educational domains. However, these agents are often ig- nored and abused in collaborative learning scenarios involving multiple students. In our work presented here, we design and evaluate interaction strategies motivated from prior re- search in small group communication. We will discuss how such strategies can be im- plemented in agents. As a first step towards evaluating agents that can interact socially, we report results showing that human tutors em- ploying these strategies are able to cover more concepts with the students besides being rated as better integrated, likeable and friendlier. 1 Introduction Conversational Agents (CAs) are autonomous in- terfaces that interact with users via spoken or writ- ten conversation. One of the applications of CAs is tutoring. Various research groups have developed tutoring agents in domains like reading, algebra, geometry, calculus, physics, computer literacy, programming, foreign languages, research methods and thermodynamics. Many of the evaluations show that CAs can be effective tutors (Arnott et. al., 2008; Kumar et. al., 2007; Graesser et. al., 2005). Most systems that use CAs as tutors have been built for learning scenarios involving one student. Evaluation of learning technologies involving stu- dents working in groups with interactive agents has shown that learners are helped both by learning as a group and receiving tutorials from agents (Kumar et. al., 2007). However, some previous studies have reported that students learning in groups ig- nore the tutor’s messages, unlike the case where students are individually tutored. Groups are more likely to abuse tutors than individual students. We reason that the presence of other students in collaborative learning scenarios causes the agents to compete for the attention of the students. Since the agents are not adept at performing social inter- active behavior, which makes up the bulk of for- mative communication in a group, they are quickly pushed to the periphery of the group. Research on small group communication has identified twelve interaction categories that are commonly observed in small groups (Bales, 1950). These categories are broadly classified into task and social-emotional categories. Content presented by most current CAs mostly classifies under the task categories. In section 2, we will list the con- versational strategies motivated from the three pos- itive social-emotional interaction categories. Thereafter, the implementation and evaluation of a CA that interleaves these social interaction strate- gies while executing a task plan will be described. 2 Social Interaction Strategies Balesian methodology (Bales, 1950) identifies three positive social-emotional interaction catego- ries: showing solidarity, showing tension release and agreeing. Participants contribute turns of these categories to address the problems of re- integration, tension release and decision respec- tively. We have mapped these categories to …

Content truncated to 500 words