Scaffolding and Assessing Students’ Explanations in Game Dialog With Hidden Markov Modeling and Computer Adaptive Testing Techniques

• In Event: Advances in Assessing Game-Based Learning

Abstract

Objectives
This paper presents a model for scaffolding and assessing students’ explanations in game dialog with hidden Markov modeling and computer adaptive testing techniques. The goal focuses on explicit articulation of connections between intuitive understandings and disciplinary concepts.

Theoretical Framework
Research on self-explanation by Chi and others provides clarity into the value of explanation for learning (e.g., Chi, Bassok, Lewis, Reimann, & Glaser 1989; Roy & Chi, 2005; Chi & VanLehn, in press). This emphasis on explanation is mirrored in research on science education, often accompanied with prediction (White & Fredericksen, 1998, 2000; Mazur, 1996; Scott, Asoko & Driver, 1991; Champagne, Klopfer & Gunstone, 1982; Kearney, 2004).

Methods
Our intention involves integrating explanation generation into the game narrative, in a way that preserves narrative space (Salen & Zimmerman, 2003), allows for identity construction and agency (Gee, 2004; Pelletier, 2008), and respects learners’ definitions and aims regarding the essence of play (Huizinga, 1980; Caillois, 1961). All of these are important elements of games and play, which, it is hypothesized, can be disrupted by assessment (Shute, Reiber & Van Eck, 2011).
For the purposes of our model, we will assume that there is a “core” game that engages the player in applying science concepts to navigate or work toward an established goal. Popular examples might build on ideas from Crayon Physics, CellCraft, Switchball, Gravitee, and Angry Birds. One of the primary goals of the core game is to facilitate developing an intuitive understanding of the science concepts through game play (similar to learning by doing).



We then create a parallel explanation game that is interwoven between levels of the core game where the player teaches or helps one or more computer-controlled characters solve specific targeted challenges in the core game. To structure the explanation dynamic in a meaningful, appealing, and engaging way, players have the opportunity to explain and justify their strategies and the concepts underlying those strategies to NPCs in order to (a) convince the NPCs to adopt these solutions and (b) help the NPCs successfully overcome similar focused challenges. The challenges faced by the characters will often be presented as contrasting cases tied to common misconceptions (Bransford & Schwartz, 1999; Schwartz & Martin, 2004).
By scaffolding players in these processes, the model not only supports learning, but also provides excellent opportunities to assess players’ learning and adjust scaffolding and challenge difficulty through the application of hidden Markov modeling (e.g., Rabiner, 1989; Li & Biswas, 2002) and computer-adaptive testing techniques (e.g., Luecht, 1996; Segall, 1996; van der Linden and Glas, 2010).


Scholarly Significance
Current goals for science literacy that focus on students’ ability to engage in extended problem-solving that involves exploration, explanation, application of integrated conceptual knowledge to rich and realistic contexts (AAAS, 1993; NRC, 1996; NRC, in preparation) and the broader 21st century skills recognized as critical for all citizens (NRC, in press). This paper presents a model for operationalizing, supporting, and assessing students’ progress and proficiency in alignment with these science proficiency goals.

Authors

• Douglas B. Clark, Vanderbilt University

• Mario Manuel Martinez-Garza, Vanderbilt University

• Gautam Biswas, Vanderbilt University

• Richard M. Luecht, University of North Carolina - Greensboro

• Pratim Sengupta, Vanderbilt University