Although artificial intelligence has been successfully introduced to enhance Education through technologies in the past few years, major challenges still remain. One of them is how to represent the knowledge of intelligent systems. To represent the knowledge of systems to support collaborative learning is particularly challenging because it is based on various learning theories and given the complexity of group learning. The main objective of this work is to introduce an ontological infrastructure on which we can build well-grounded theoretical knowledge based on learning theories and to show how we can use it to develop programs to support intelligent guidance for an effective design of group activities.
One goal of our project is to create a dialogue agent that can behave as a student peer and collaborate with a human student to explain or diagnose data structures programs. A human peer must be able to initiate something the agent may not be expecting and the agent must be able to respond to it as a peer would. This paper describes the work we have completed to extend an existing natural language tutorial dialogue system to provide such a capability and the work that remains.
Recent learning theories stress that learning does not necessarily translate solely into knowledge gains: rather, it can be measured in terms of increased participation and interaction of the individual with their group or community. We explore this in conjunction with exploitation of an existing professional tool, which we have enhanced with mechanisms to extract useful information from learner interaction traces and then make these available in a form that is meaningful for teachers and learners. This broad approach offers huge promise at a far lower cost than that needed to build pure intelligent learning systems. We present our learning environment used in a course where small teams of students work on software development projects and learn about working in groups. We report the success of the approach in terms of the different levels of interaction across two cohorts, the first using just the professional tool and the second using our enhanced visualisations in conjunction with it.
This work surveys the opinions of teachers and other education professionals regarding the potential for Open Learner Models (OLM) in UK schools. We describe the aims of OLM, and of current UK initiatives involving formative assessment and promotion of metacognitive skills. We conclude that UK education professionals appreciate a synergy of these approaches, and that OLM-based systems could be valuable in achieving educational aims in schools.
This paper describes new techniques for assessing pedagogical discourse via threaded discussions that are based on an analysis of speech acts and course topics. The context is an undergraduate computer science course. We use speech act analysis to assess the effect of instructor participation on student participation and to create thread profiles that help identify threads characterized by agreement, debate, and unresolved issues, such as threads that may have unanswered questions. For the first time, we integrate an analysis of course topics to help identify discussions of particular interest or confusion.
This paper presents an evaluation of McFeSPA (Metacognitive
In term of metacognitive scaffolding, we mean that McFeSPA provides more opportunity for fostering reflective thinking by the feedback giver about giving feedback.
Feedback Scaffolding System for Pedagogical Apprenticeship), a system to help inexperienced teaching assistants (TAs)
Inexperienced TAs are taken to include novice teachers, novice tutors, and novice lecturers
who lack training in how to provide quality feedback to help them improve their feedback skills while marking programming assignments. The system exploits techniques drawn from artificial intelligence, cognitive psychology and theories of education. The results of this study indicate that McFeSPA could help TAs who mark programming assignments think about/reflect on their feedback giving to provide quality feedback to students in general.
We conducted a within-subjects study of the effect on student learning of using open student model in the form of taxonomic concept map. We found a significant improvement in learning on two questions related to the topic of the tutor, and no pre-post improvement on questions unrelated to the topic of the tutor. These preliminary results suggest that students may learn concepts indirectly by viewing the open student model presented as taxonomic concept map after each problem.
We evaluated the effect of providing error-flagging as support for error detection, but not error correction while the student is solving a problem. We found that providing error-flagging in addition to demand feedback during practice learning was no more effective than providing only demand feedback when the tutor did not explicitly mention that errors were being flagged. On the other hand, explaining and providing error-flagging without demand feedback during pre- and post-tests resulted in significantly better scores on pre- and post-tests even though error-flagging did not provide any error-correction support.
We propose “paragraph development schemata” in English to help learners learn good paragraph structures for English composition. Each schema is expressed by standardized and fine grain sized terminology. Owing to the schemata, learners will be able to know what paragraph structures exist, recognize how each structure is different, and when a learner selects one schema she can be aware of the lack of necessary information by herself.
Rules have been showed to be appropriate representations to model tutoring and can be easily applied to intelligent tutoring systems. We applied a machine learning technique, Classification based on Associations, to automatically learn tutorial rules from annotated tutoring dialogues of a human expert tutor. The rules we learn concern the tutor's attitude, the domain concepts to focus on, and the tutor moves. These rules have very good accuracy. They will be incorporated in the feedback generator of an Intelligent Tutoring System.
This paper proposes the notion of problem templates (PTs), a concept based on theories of memory and expertise. These mental constructs allow experts to quickly recognise problem states, almost instantaneously retrieve potentially vast amounts of domain-specific information, and seemingly effortlessly implement strategies to solve the problem. We investigate the role of problem templates in intelligent tutoring systems. An evaluation study was conducted in which PTs were created and implemented in SQL-Tutor. PTs were used to model students and make pedagogical decisions. Students using templates showed high levels of learning within short periods of time. Further evaluation studies are required to investigate the extent and detail of its effect on learning.
In a multi-user, real-time, and situation-based learning environment, the availability of enough and appropriate situations is crucial for success. In order to improve effectiveness and efficiency of learning, we develop a new type of pedagogical agent: situation creator. Such an agent intentionally creates specific situations in the shared virtual driving place according to users' performance information. We conduct a pilot evaluation and found that the situation creators significantly increase the number of situations that a learner can expect to encounter while using the system.
Building effective learning environments is an art that can only be perfected by a great deal of explorations involving the environments' audience: the learners. This paper focuses on taking into account the learners' spatial ability into the development of Intelligent Tutoring Systems. We modified ERM-Tutor, a constraint-based tutor that teaches logical database design, to provide not only textual feedback messages, but also messages containing combinations of text and pictures, in accordance with the multimedia theory of learning . Results of a preliminary study performed show a promising indication for further explorations. We plan to use these results as the basis for another evaluation study in early 2007.
Brent Morgan, Roby Coles, Josh Brittingham, Barry Gholson
620 - 622
This experiment sought to generalize to middle- and high-school students the learning gains observed through deep-level reasoning questions among college students in vicarious environments. 342 eighth through eleventh graders were presented with either content alone, content preceded by deep-level reasoning questions, or an interactive session with a virtual tutor. Students exposed to the deep-level reasoning questions outperformed the interactive and content-only groups, suggesting that deep-level reasoning questions do indeed aid in knowledge representation amongst middle- and high-school students.
Roger Nkambou, Engelbert Mephu Nguifo, Olivier Couturier, Philippe Fournier-Viger
623 - 625
In this paper, we present an approach to capture problem-solving knowledge using a promising technique of data and knowledge discovery based on a combination of sequential pattern mining and association rules discovery.
Zachary A. Pardos, Mingyu Feng, Neil T. Heffernan, Cristina Linquist-Heffernan
626 - 628
Two modelling methods were used to answer the research question of how accurate various grained 1, 5, 39 and 106 skill models are at assessing student knowledge in the ASSISTment online tutoring system and predicting student performance on a state math test. One method is mixed-effects statistical modelling. The other uses a Bayesian networks machine learning approach. We compare the prediction results to identify benefits and drawbacks of either method and to find out if the two results agree. We report that both methods showed compelling similarity which support the use of fine grained skill models.
It is common between teachers of mathematics to assess the difficulty of the problems they gave to students. The evaluation is mostly based on subjective appreciation with no explicit consideration of difficulty factors. However, in a computerized environment for mathematics instruction it would be highly beneficial to dispose of such explicit factors and automated assessment of problem difficulty. In the present paper a set of factors for sequence problems in analysis are proposed. For the automated detection of these factors a set of rules was defined. Knowledge representation and implementation details are described.
Willow is an automatic and adaptive free-text scoring system. Its main purpose is to provide formative assessment to the students so that they can get more practise and feedback before their exams. It also keeps track of the terms that students use in their answers to automatically generate each particular student conceptual model and the group conceptual model. In both cases, the conceptual model can be defined as a set of concepts and their relationships that each student keeps in his or her mind about an area of knowledge. The conceptual model can be represented by COMOV as a concept map so that instructors can track students' concepts acquisition. In this paper, we present the new possibility and its consequences of showing the concept maps (from their individual and group conceptual models) not only to instructors but also to students. Moreover, to allow students to see their evolution, as they study and get more training, through their concept maps. The preliminary results of an experiment carried out with a group of students show how they like to be able to inspect their generated learning models.
Game-based learning software shows tremendous potential in its ability to motivate and engage learners. However, when giving pedagogical feedback in a game context, it is important to do so in a way that does not break a student's engagement and is not prone to abuse. This paper describes in-game help in the Tactical Language and Culture Training System, a tool designed to help learners rapidly acquire basic communicative skills in a foreign language and culture.
The design phase on the life cycle of the eLearning process is currently one of the main bottle necks in adaptive systems. To facilitate this process and reduce the design effort our approach focuses on providing dynamic assistance to some of the author's tasks that strongly depend on data coming from users' interactions. In particular, ADAPTAPlan project is developing a system where the learning route of a student is dynamically built from the combination of user modelling and scheduling techniques making a pervasive use of educational standards (IMS). The author is requested to provide simple information about the course structure, pedagogy and restrictions and the system utilize this information together with the user model to generate a personalize IMS-LD course flow suited to that learner. Since the output is given in a standardized format, the course can be run in any standard based learning environment. This approach will be tested on a real course, “How to teach online”, within the UNED's program for ongoing education and open courses.
Benjamin Shih, Kenneth Koedinger, Richard Scheines
644 - 646
Complex student models often include key parameters critical to their behavior and effectiveness. For example, one meta-cognitive model of student help-seeking in intelligent tutors includes 15 rules and 10 parameters. We explore whether or not this model can be improved both in accuracy and generalization by using a variety of techniques to select and tune parameters. We show that such techniques are important by demonstrating that the normal method of fitting parameters on an initial data set generalizes poorly to new test data sets. We then show that stepwise regression can improve generalization, but at a cost to initial performance. Finally, we show that causal search algorithms can yield simpler models that perform comparably on test data, but without the loss in training set performance. The resulting help-seeking model is easier to understand and classifies a more realistic number of student actions as help-seeking errors.
Hilary Smith, Rose Luckin, Danaë Stanton Fraser, Lisa Williams, Andreas Duenser, Eva Hornecker, Adrian Woolard, Brendan Lancaster
647 - 649
This poster introduces two pilot studies on children's story reading and retelling using interactive media that move beyond familiar keyboard and mouse interaction. These studies formed a subset of investigations into the potential for Augmented Reality (AR) story books to engage and motivate story reading using familiar and easy to set-up technologies. We contrasted AR with two other media and report initial findings that point towards the memorability and potential for engagement of these different media, concluding with our further research questions.
We have developed environments that use teaching as a metacognitive, reflective, and iterative process to help middle school students learn about complex processes. We demonstrate that metacognition and self-regulation are crucial in developing effective learners and preparing them for future learning tasks. As evidence, we discuss the impact of metacognitive support on students' learning performance and behavior patterns that promote better learning through self-monitoring.