Meta-tutoring applies the basic policies of interactive tutoring to get students to adopt effective meta-cognitive strategies. Unfortunately, when the meta-tutor is removed, students often revert to using ineffective strategies. This paper is an early report on the progress of the Affective Meta-Tutoring (AMT) project, which will use an affective learning companion to motivate students to more permanently adopt effective meta-cognitive strategies.

Intelligent Tutoring Systems are software applications capable of complementing and enhancing the learning process by providing direct customized instruction and feedback to students in various disciplines. Although Intelligent Tutoring Systems could differ widely in their attached knowledge bases and user interfaces (including interaction mechanisms), their behaviors are quite similar. Therefore, it must be possible to establish a common software model for them. A common software model is a step forward to move these systems from proof-of-concepts and academic research tools to widely available tools in schools and homes. The work reported here addresses: (1) the use of Design Patterns to create an object-oriented software model for Intelligent Tutoring Systems; (2) our experience using this model in a three-year development project and its impact on facets such as creating a common language among stakeholders, supporting an incremental development, and adjustment to a highly shifting development team; and (3) the qualities achieved and trade-offs made.

While high interactivity has been one of the main characteristics of one-on-one human tutoring, a great deal of controversy surrounds the issue of whether interactivity is indeed the key feature of tutorial dialogue that impacts students' learning results. There are two commonly held hypotheses regarding the issue: a widely-believed monotonic interactivity hypothesis and a better supported interaction plateau hypothesis. The former hypothesis predicts increasing in interactivity causes an increase in learning while the latter states that increasing interactivity yields increasing learning until it hits a plateau, and further increases in interactivity do not cause noticeably increase in learning. In this study, we proposed the tactical interaction hypothesis which predicts beyond a certain level of interactivity, further increases in interactivity do not cause increase in learning unless they are guided by effective tutorial tactics. Overall our results support this hypothesis. However, finding effective tactics is not easy. This paper sheds some light on how to apply Reinforcement Learning to derive effective tutorial tactics.

Self-explaining is a beneficial learning strategy for studying worked-out examples because it either supplies missing information through the generation of inferences or because it provides a mechanism for repairing flawed mental models. Although self-explanation is generated with the purpose of helping the individual, is it also helpful to produce explanations in a collaborative setting? Can individuals help each other infer missing information or repair their flawed mental models collaboratively? To find out, we coded the dialog from dyads collaboratively studying examples and contrasted it with individuals studying examples alone. The results suggest that dyads were more likely to attempt to reconcile the examples with their attempted solutions, and avoid shallow processing of examples through paraphrasing.

Traditional methods of evaluating student programs are not always appropriate for assessment of different instructional interventions. They tend to focus on the final product rather than on the process that led to it. This paper presents intention-based scoring (IBS), an approach to measuring programming ability that looks at intermediate programs produced over the course of an implementation rather than just the one at the end. The intent is to assess a student's ability to produce algorithmically correct code on the first attempt at achieving each program goal. In other words, the goal is to answer question "How close was the student to being initially correct?" but not to speak to a student's debugging skills or ability to ultimately produce a working program. To produce an IBS, it is necessary to inspect a student's online protocol, which is simply the collection of all programs submitted to a compiler. IBS involves a three-phase process of (1) identification of the subset of all programs in a protocol that represent the initial attempts at achieving programming goals, (2) bug identification, and (3) rubric-based scoring. We conclude with an example application of IBS in the evaluation of a tutoring system for beginning programmers and also show how an IBS can be broken down by the underlying bug categories to reveal more subtle differences.

Coached program planning is a dialogue-based style of tutoring aimed at helping novices during the early stages of program writing. The intent is to help novices understand and solve problems in their own words through the construction of natural-language style pseudocode as the first step in solving a programming problem. We have designed an environment supporting coached program planning and have used it in a human-to-human, computer-mediated evaluation of 16 novice programmers enrolled in a pre-CS1 programming course at the University of Pittsburgh. The results show that students who underwent coached program planning, compared to those who did not, were more prolific with comments in their programs, committed fewer structural mistakes, and exhibited less erratic programming behavior during their implementation. The dialogues collected from this experiment followed a clear 4-step pattern. Starting with this observation, we are developing a dialogue-based intelligent tutoring system called the Pseudocode Tutor to support coached program planning.

We present an adaptive interface designed to provide tailored support for the understanding of written instructional material. The interface relies on a user model based on a Bayesian network, that assesses users' understanding as users read the instructional material and try to understand it by generating explanations to themselves. The user model's assessment is used by the interface to generate tailored scaffolding of further user's explanations that can improve the user's comprehension. After illustrating how the Bayesian user model assesses understanding from the user's explanations and from latency data on the user's attention, we discuss initial results on the effectiveness of the interface's adaptive interventions.