One-on-one tutoring is significantly more effective than traditional classroom instruction. In recent years, automated tutoring systems are approaching that level of effectiveness by engaging students in rich natural language dialogue that contributes to learning. A promising approach for further improving the effectiveness of tutorial dialogue systems is to model the differential effectiveness of tutorial strategies, identifying which dialogue moves or combinations of dialogue moves are associated with learning. It is also important to model the ways in which experienced tutors adapt to learner characteristics. This paper takes a corpus- based approach to these modeling tasks, presenting the results of a study in which task-oriented, textual tutorial dialogue was collected from remote one-on-one human tutoring sessions. The data reveal patterns of dialogue moves that are correlated with learning, and can directly inform the design of student-adaptive tutorial dialogue management systems.

Intelligent Tutoring Systems (ITSs) have begun to incorporate game-based components in an attempt to balance the learning benefits of ITSs with the motivational benefits of games. iSTART-ME (Motivationally Enhanced) is a new game-based learning environment that was developed on top of an existing ITS (iSTART). In a multi-session lab-based efficacy study with 125 high school students, those students with a low prior reading ability who were trained by a game-based tutoring system (iSTART-ME) or a traditional intelligent tutoring system (iSTART-Regular) performed significantly better on posttest measures than students assigned to a time-delayed control condition. Additionally, the low reading ability students who interacted with the game-based system had a tendency to gain more than students in the traditional ITS system.

Intelligent tutoring systems research aims to produce systems that meet or exceed the effectiveness of one-on-one expert human tutoring. Theory and empirical study suggest that affective states of the learner must be addressed to achieve this goal. While many affective measures can be utilized, posture offers the advantages of non-intrusiveness and ease of interpretation. This paper presents an accurate posture estimation algorithm applied to a computer-mediated tutoring corpus of depth recordings. Analyses of posture and session-level student reports of engagement and cognitive load identified significant patterns. The results indicate that disengagement and frustration may coincide with closer postural positions and more movement, while focused attention and less frustration occur with more distant, stable postural positions. It is hoped that this work will lead to intelligent tutoring systems that recognize a greater breadth of affective expression through channels of posture and gesture.

We evaluated an affect-sensitive version of AutoTutor, a dialogue based ITS that simulates human tutors. While the original AutoTutor is sensitive to students’ cognitive states, the affect-sensitive tutor (Supportive tutor) also responds to students’ affective states (boredom, confusion, and frustration) with empathetic, encouraging, and motivational dialogue moves that are accompanied by appropriate emotional expressions. We conducted an experiment that compared the Supportive and Regular (non-affective) tutors over two 30-minute learning sessions with respect to perceived effectiveness, fidelity of cognitive and emotional feedback, engagement, and enjoyment. The results indicated that, irrespective of tutor, students’ ratings of engagement, enjoyment, and perceived learning decreased across sessions, but these ratings were not correlated with actual learning gains. In contrast, students’ perceptions of how closely the computer tutors resembled human tutors increased across learning sessions, was related to the quality of tutor feedback, the increase was greater for the Supportive tutor, and was a powerful predictor of learning. Implications of our findings for the design of affect-sensitive ITSs are discussed.

Using natural language as a way for students to interact with an ITS has many advantages. However, creating the intelligence with which the tutor evaluates a student’s natural language input is challenging. We describe a system, ConceptGrid, that allows non-programmers to create the instruction for checking natural language input. Three tutor authors used the system to develop answer templates for conceptual-based questions in statistics. Results indicate ConceptGrid is a viable system for non-programmers to use to allow students to use natural language to interact with a tutor.

The current paradigm in student modeling has continued to show the power of its simplifying assumption of knowledge as a binary and monotonically increasing construct, the value of which directly causes the outcome of student answers to questions. Recent efforts have focused on optimizing the prediction accuracy of responses to questions using student models. Incorporating individual student parameter interactions has been an interpretable and principled approach which has improved the performance of this task, as demonstrated by its application in the 2010 KDD Cup challenge on Educational Data. Performance prediction, however, can have limited practical utility. The greatest utility of such student models can be their ability to model the tutor and the attributes of the tutor which are causing learning. Harnessing the same simplifying assumption of learning used in student modeling, we can turn this model on its head to effectively tease out the tutor attributes causing learning and begin to optimize the tutor model to benefit the student model.

This work presents an algorithm to provide a better represen- tation of space to artificially intelligent characters (i.e., agents or bots) in game and simulation environments by providing a more accurate breakdown of the traversable space present in the game environment. Such representations are generally constructed by decomposing the walkable space present in a game environment into a series of convex regions to form a data structure called a navigation mesh. We extend the basic concept of a navigation mesh by the introduction of an understanding of the textures that are attached to the underlying geometry creating what we refer to as a texture-aware navigation mesh. This does result in a more complex navigation mesh (more regions and a larger search space). However, since the textures of walkable geometry can be used to determine the appropriate traversal method for that terrain, a game character can determine valid paths for their traversal methods using just the navigation mesh (e.g., characters in cars can generate paths containing just roads or walking characters can create paths containing just sidewalks). We also present a use case that shows how such a system of texture aware naviga- tion meshes might benefit character path planning and search in virtual environments. In this use case, we examine a Real Time Strategy game style game environment, which shows it is possible to generate a navigation mesh such that each region is composed of a single terrain type.

We model the pursuit problem as a set of distributed agents communicating over a network subject to latency. Latency has serious deleterious effects on solving the pursuit problem. In this paper, we present a simple, yet effective way of dealing with latency that yields very good performance. Our method disperses predators within a region in which the prey may move that accounts for network latency.

This paper deals with planning domains that appear in computer games, especially when modeling intelligent virtual agents. Some of these domains contain only actions with no negative effects and are thus treated as easy from the planning perspective. We propose two new techniques to solve the problems in these planning domains, a heuristic search algorithm ANA* and a constraint-based planner RelaxPlan, and we compare them with the state-of-the-art planners, that were successful in IPC, using planning domains motivated by computer games.

In this paper, we outline a system for evaluating the performance of scientific research across a number of outcome metrics (e.g. publications, sales, new hires). Our system is designed to classify research performance into a number of metrics, evaluate each metric’s performance using only data on other metrics, and to cast predictions of future performance by metric. This study shows how data mining techniques can be used to provide a predictive analytic approach to the management of resources for scientific research.