Intelligent Tutoring Systems

Although, many have theorized about the link between cognition and affect and its potential importance in complex tasks such as problem solving and deep learning, this link has seldom been explicitly investigated during tutoring. Consequently, this paper investigates the relationship between learners’ cognitive and affective states during 50 tutoring sessions with expert human tutors. Association rule mining analyses revealed significant co-occurrence relationships between several of the cognitive measures (i.e., student answer types, question types, misconceptions, and metacomments) and the affective states of confusion, frustration, and anxiety, but not happiness. We also derived a number of association rules (Cognitive State → Affective State) from the co-occurrence relationships. We discuss the implications of our findings for theories that link affect and cognition during learning and for the development of affect-sensitive ITSs.

The purpose of this work is to enhance learner self-esteem while interacting with a tutoring system. Our approach is based on a subliminal priming technique that implicitly conditions learner self-esteem. An experimental study has been conducted to analyze the impact of this method on participants’ reported self-esteem on one hand and learning performance on the other hand. Furthermore, three physiological sensors were used to continuously monitor participants’ affective reactions, namely electroencephalogram, skin conductance and blood volume pulse sensors. The purpose was to measure the effect of our approach on both learner mental state and emotions. We then proposed a model that links learners’ physiological signals and priming conditions to learning results.

When our computers act in unexpected (and unhelpful) ways, we become frustrated with them. Were the computers human assistants, they would react by doing something to mitigate our frustration and increase their helpfulness. However, computers typically do not know we are frustrated. This paper presents research showing that user frustration can be detected with good accuracy (84%) using only two types of input data (head tilt and pupil dilation). We also show that reasonable accuracy (73%) can be achieved using only information about head tilt. We then propose how such technology could be employed to reduce learner frustration in adaptive tutoring applications.

The Hint Factory is an implementation of our novel method to automatically generate hints using past student data for a logic tutor. One disadvantage of the Hint Factory is the time needed to gather enough data on new problems in order to provide hints. In this paper we describe the use of expert sample solutions to “seed” the hint generation process. We show that just a few expert solutions give significant coverage (over 50%) for hints. This seeding method greatly speeds up the time needed to reliably generate hints. We discuss how this feature can be integrated into the Hint Factory and some potential pedagogical issues that the expert solutions introduce.

The traditional, well established approach to finding out what works in education research is to run a randomized controlled trial (RCT) using a standard pretest and posttest design. RCTs have been used in the intelligent tutoring community for decades to determine which questions and tutorial feedback work best. Practically speaking, however, ITS creators need to make decisions on what content to deploy without the benefit of having run an RCT in advance. Additionally, most log data produced by an ITS is not in a form that can easily be evaluated with traditional methods. As a result, there is much data produced by tutoring systems that we would like to learn from but are not. In prior work we introduced a potential solution to this problem: a Bayesian networks method that could analyze the log data of a tutoring system to determine which items were most effective for learning among a set of items of the same skill. The method was validated by way of simulations. In this work we further evaluate the method by applying it to real world data from 11 experiment datasets that investigate the effectiveness of various forms of tutorial help in a web based math tutoring system. The goal of the method was to determine which questions and tutorial strategies cause the most learning. We compared these results with a more traditional hypothesis testing analysis, adapted to our particular datasets. We analyzed experiments in mastery learning problem sets as well as experiments in problem sets that, even though they were not planned RCTs, took on the standard RCT form. We found that the tutorial help or item chosen by the Bayesian method as having the highest rate of learning agreed with the traditional analysis in 9 out of 11 of the experiments. The practical impact of this work is an abundance of knowledge about what works that can now be learned from the thousands of experimental designs intrinsic in datasets of tutoring systems that assign items in a random order.

Being able to argue with a student to convince her or him about the rationale of tutoring hints is an important component of pedagogy. In this paper we present an argumentation framework for implementing persuasive tutoring dialogues. The entire interaction between the student and the tutoring system is seen as an argumentation. The tutoring system and the student can settle conflicts arising during their argumentation by accepting, challenging, or questioning each other’s arguments or withdrawing their own arguments. Pedagogic strategies guide the tutoring system selecting arguments aimed at convincing the student. We illustrate this framework with a tutoring system for medical diagnosis using a normative expert model.

We explored the possibility of predicting the quality of student answers (error-ridden, vague, partially-correct, and correct) to tutor questions by examining their linguistic patterns in 50 tutoring sessions with expert human tutors. As an alternative to existing computational linguistic methods that focus on domain-dependent content words (e.g., velocity, RAM, speed) in interpreting a student’s response, we focused on function words (e.g., I, you, but) and domain-independent content words (e.g., think, because, guess). Proportional incidence of these word categories in over 6,000 student responses to tutor questions was automatically computed using Linguistic Inquiry and Word Count (LIWC), a computer program for analyzing text. Multiple regression analyses indicated that two parameter models consisting of pronouns (e.g., I, they, those) and discrepant terms (e.g., should, could, would) were effective in predicting the conceptual quality of student responses. Furthermore, the classification accuracy of discriminant functions derived from the domain-independent LIWC features competed with conventional domain-dependent assessment methods. We discuss the possibility of a composite assessment algorithm that focuses on both domain-dependent and domain-independent words for dialogue-based ITSs.

We have developed an artificial agent based on a computational model of peer learning we developed. That model shows that shifts in initiative are conducive to learning. The peer learning agent can collaborate with a human student via dialog and actions within a graphical workspace. This paper describes the architecture and implementation of the agent and the user study we conducted to evaluate the agent. Results show that the agent is able to encourage shifts in initiative in order to promote learning and that students learn using the agent.

Every learner is fundamentally different. However, few courses are delivered in a way that is tailored to the specific needs of each student. Delivery systems for adaptive educational hypermedia have been extensively researched and found promising. Still, authoring of adaptive courses remains a challenge. In prior research, we have built an adaptive hypermedia authoring system, MOT3.0. The main focus was on enhancing the type of functionality that allows the non-technical author, to efficiently and effectively use such a tool. Here we show how teachers can start from existing course material and transform it into an adaptive course, catering for various learners. We also show how this apparent simplicity still allows for building of flexible and complex adaptation, and describe an evaluation with course authors.

Design of pedagogically sound collaborative learning (CL) activities is a complex task, but necessary if the goal is to support learning. Through the design of CL scenarios, a designer can define structures that increase the chance for learning to occur. It means that the effectiveness of the collaboration depends on the transformation of the designer’s intentions into elements that will constitute the learning scenario. To support the creation of CL scenarios this paper presents an intelligent authoring tool that is equipped with the knowledge about different pedagogies and practices related to collaboration. Through the use of this information, the tool can provide intelligent guidance that support designers to create more effective CL scenarios. The results of an experiment suggest that our tool helps teachers to more easily introduce CL activities in classroom and creates favorable conditions for students to perform collaboration improving their overall learning performance throughout the year.

The plethora of different subfields in intelligent tutoring systems (ITS) are often difficult to integrate theoretically when analyzing how to design an intelligent tutor. Important principles of design are claimed by many subfields, including but not limited to: design, human-computer interaction, perceptual psychology, cognitive psychology, affective and motivation psychology, statistics, artificial intelligence, cognitive neuroscience, constructivist and situated cognition theories. Because these theories and methods sometimes address the same grain size and sometimes different grain sizes they may or may not conflict or be compatible and this has implications for ITS design. These issues of theoretical synthesis also have implications for the experimentation that is used by our various subfields to establish principles. Because our proposal allows the combination of multiple perspectives, it becomes apparent that the current “forward selection” method of theoretical progress might be limited. An alternative “backward elimination” experimental method is explained. Finally, we provide examples to illustrate how to build the bridges we propose.

This paper describes efforts to both promote and recognize student dialogue in free-entry text discussion within an inquiry-learning environment. First, we discuss collaborative tools that enable students to work together and how these tools can potentially focus student effort on subject matter. We then show how our tutor uses an expert knowledge base to recognize (with 88% success rate) when students are discussing content relevant to the problem and to correctly link (with 70% success) that content with an actual topic. Subsets of the data indicate that even better results are possible. This research provides solid support for the concept of using a knowledge base to recognize content in free-entry text discussion. The paper concludes by demonstrating how this content recognition can be used to support students engaged in problem-solving activities.

Learners engaged in CSCL macro-scripts are involved in self-organization activities. We present an exploratory study that suggests that Bardram’s theoretical model of collective work dynamics is a pertinent basis for both (1) designing interfaces providing a passive support that engages learners in an explicit organization activity, and (2) making learners’ organization more easily detectable and analyzable within a perspective of active support.

In this study, we describe a conversational agent designed to support collaborative learning interactions between pairs of students. We describe a study in which we independently manipulate the social capability and goal alignment of the agent in order to investigate the impact on student learning outcomes and student perceptions. Our results show a significant interaction effect between the two independent variables on student learning outcomes. While there are only a few perceived differences related to student satisfaction and tutor performance as evidenced in the questionnaire data, we observe significant differences in student conversational behavior, which offer tentative explanations for the learning outcomes we will investigate in subsequent work.

Virtual humans are animated, lifelike characters capable of free-speech and nonverbal interaction with human users. In this paper, we describe the development of two virtual human characters for teaching the skill of deception detection. An accompanying tutoring system provides solicited hints on what to ask during an interview and unsolicited feedback that identifies properties of truthful and deceptive statements uttered by the characters. We present the results of an experiment comparing use of virtual humans with tutoring against a no-interaction (baseline) condition and a didactic condition. The didactic group viewed a slide show consisting of recorded videos along with descriptions of properties of deception and truth-telling. Results revealed that both groups significantly outperformed the no-interaction control group in a binary decision task to identify truth or deception in video statements. No significant differences were found between the training conditions.

Narrative-centered learning environments offer significant potential for creating effective learning experiences in which students actively participate in engaging story-based problem solving. As the capabilities of narrative-centered learning environments expand, a key challenge is identifying experiential factors that contribute to the most effective story-based learning. To investigate the impact of students’ narrative experiences on learning outcomes, a Wizard of Oz (WOZ) study was conducted with middle school students interacting with a narrative-centered learning environment. Students’ experiences were examined using narrative profiles representing their type of story interaction. With narrative planning, tutorial planning, and natural language dialogue functionalities provided by wizards, the WOZ study revealed that in interactive story-based learning supported by beyond-state-of-the-art ITS capabilities, 1) students exhibit a range of learning outcomes, 2) students exhibit a range of narrative profiles, and 3) certain student narrative profiles are strongly associated with desirable learning outcomes. The study suggests design decisions for optimizing story-based learning.

A key promise of narrative-centered learning environments is the ability to make learning engaging. However, there is concern that learning and engagement may be at odds in these game-based learning environments and traditional learning systems. This view suggests that, on the one hand, students interacting with a game-based learning environment may be engaged but unlikely to learn, while on the other hand, traditional learning technologies may promote deep learning but provide limited engagement. This paper presents findings from a study with human participants that challenges the view that engagement and learning need be opposed. A study was conducted with 153 middle school students interacting with a narrative-centered learning environment. Rather than finding an oppositional relationship between learning and engagement, the study found a strong positive relationship between learning outcomes and increased engagement. Furthermore, the relationship between learning outcomes and engagement held even when controlling for students’ background knowledge and game-playing experience.

We implemented and evaluated a collaborative lecture module in an ITS that models the pedagogical and motivational tactics of expert human tutors. Inspired by the lecture delivery styles of the expert tutors, the collaborative lectures of the ITS were conversational and interactive, instead of a polished one-way information delivery from tutor to student. We hypothesized that the enhanced interactivity of the expert tutor lectures were linked to efforts to promote student engagement. This hypothesis was tested in an experiment that compared the collaborative lecture module (dialogue) to less interactive alternatives such as monologues and vicarious dialogues. The results indicated that students in the collaborative lecture condition reported more arousal (a key component of engagement) than the controls and that arousal was positively correlated with learning gains. We discuss the implications of our findings for ITSs that aspire to model expert human tutors.

The use of technology for instruction, and the enormous amount of information available for consumption, places a considerable burden on instructors who must learn to integrate appropriate student practices and learning assessment. The Pedagogical Workflows project is developing a novel workflow environment that supports efficient assessment of student learning through interactive generation and execution of various assessment workflows. We focus especially on how student discussion use can be combined with more traditional assessment data. In this paper, we present our initial assessment workflows, the initial feedback from instructors, and the user portal that is being developed for running the workflows. Inherent in the development of the workflows is an examination of what teachers think is important to learn about their students, a question that is central to every intelligent tutoring system. We anticipate that assessment workflows will become an important tool for instructors, researchers, and ITS development.

A vital goal of instruction is to enable learners to transfer acquired knowledge to appropriate future situations. For elementary school children in middle-high-SES schools, “explicit” instruction on the Control of Variables Strategy (CVS) has proven to be very effective at promoting transfer, even after time delays, when administered by human instructors [1], [2] and when administered by our computer tutor (“TED” for Training in Experimental Design). However, when the same instruction was delivered to students in low-SES schools, near—but especially far—transfer rates were lower. We discuss our findings of the predictors of transfer in this population, and an initial investigation assessing the causal status of one candidate factor for far transfer, understanding the logic of CVS. Finally, we discuss the potential implications of these findings for ways to adapt instruction to individual students.

In this paper we discuss the development of the Discussion Forum Analyzer Tool (DFAT) for the Moodle Learning Management System (LMS). This application was designed to enhance student engagement in course discussion forums by emphasizing the participation of fellow learners and by making discussions more efficient and pertinent to students. In addition, by mining user-submitted content and applying learning algorithms, DFAT “tags” replies/comments as relevant or non-relevant, and recommends other related discussions and readings.

Our work takes as a starting point McCalla’s proposed ecological approach for the design of peer-based intelligent tutoring systems and proposes: (i) to develop an algorithm for selecting appropriate content (learning objects) to present to a student, based on previous learning experiences of like-minded students (ii) to build on this research by also having students leaving explicit annotations on learning objects to convey refinements of their understanding to subsequent students; the challenge is to intelligently match students to those annotations that will be most beneficial for their tutoring (iii) to develop methods for intelligently extracting learning objects from a repository of knowledge, in a manner that may be customized to the needs of specific students (iv) to apply our work to the specific application of assisting health care workers via peer-based intelligent tutoring, primarily for homecare environments.

Technological support for personalized learning has the potential to

transform the educational system

in the United States. There is a growing interest in educational games and their potential for motivating learners. Techniques from the educational data mining and intelligent tutoring systems communities can be leveraged to better understand, design, and evaluate educational games for both learning effectiveness and learner engagement. This work explores the use of intelligent feedback in games as well as the potential pitfalls; it concludes with a proposed study designed to explore the differences between intelligent tutoring systems and educational video games.

Despite the growing number of ITSs for teaching argumentation, few tutors actually debate the student. An

intelligent debater

allows the student to practice argumentation and provides the motivation to analyze evidence. Here I describe an

intelligent debater

used in Policy World to argue with students about policy recommendations. The

intelligent debater

forces the student to recommend a policy intervention, to describe how the intervention affects the desired policy outcomes, and to provide evidence. The debater then attacks infeasible recommendations, implausible mechanisms, and weak evidence. Key aspects of the intelligent debater algorithm are presented for those interested in using intelligent debaters in science, law, and history.

Multiple External Representations (MERs) have been used successfully in instructional activities, including fractions. However, students often have difficulties making the connections between the MERs spontaneously. We argue that interactive fraction representations may help students in discovering relevant features and relating the MERs to one another. Support for guiding student interaction is provided by example-tracing tutors.

The graph-as-picture misconception (GAPm) is a commonly reported error, but basic questions about its prevalence and degree have not been addressed. An interactive educational diagrammatic system was designed to assess and to help students overcome GAPm. Three activities were administered to students (N=48) using touch screen technology: a diagram/picture decision task, a multiple choice questionnaire, and an interactive racing car game in which the child moves a car along a track and a speed/distance graph is plotted concurrently alongside. Preliminary results demonstrate the utility of the diagram/picture discrimination task for detecting GAPm and providing rich information about children’s knowledge of different representational forms.

Over the years of a university degree, students face many challenges, including: selecting elective subjects, gaining a sense of their own progress, understanding the reasons that they are required to do particular learning tasks, and deciding the time to devote to different assignments. These are important for student engagement and success. We aim to create a new way to help students address these challenges, based on a curriculum mapping system. This will aid students making more informed elective decisions based on personalized progress reports and knowledge gap analysis. Our system will capture whole degree programs, mapping the detailed degree requirements (accreditation, learning objectives, attributes) to the individual subjects, the assessments and each student’s actual performance on assessment tasks. We will evaluate this in three stages: qualitative studies of students’ interaction with the system based on a fictional student’s personalized progress report and knowledge gaps; qualitative studies based on actual profiles; and a field trial which tracks interaction and makes use of questionnaires.

Previous work suggests that student dispositional attributes have an effect on help-seeking behaviors. We report on studies attempting to apply this work to a computer-supported collaborative environment that suggest that dispositional attributes of motivation orientation and self-efficacy may be more easily examined in the context of task choices that follow failure events as part of an experimental design. Two studies that have lead to this conclusion are discussed, as well as plans for a recovery-from failure study in preparation.

We propose a data visualization tool that offers insights into the way students solve procedural domain problems. The tool uses nodes and edges to represent states and actions which students have generated using an intelligent tutoring system or computer aided instruction tool, ultimately showing the way a student has solved a problem. We use the example of logic tutor data and suggest two methods of evaluation for ensuring the tool is effective at aiding educators to better understand student learning.

Step-based ITS have been proven successful in well-defined domains, but their success is mitigated by their cost. Different approaches have been investigated to reduce these efforts; one of them is a framework that eases the development of tutors for a given class of task domains. In this paper, we explain how a domain is modeled with the ASTUS framework and we discuss why an authoring language is a promising technique to improve its usability.

Educational data mining methods are being used to automatically generate hints to students in intelligent tutoring systems. Using these methods, we hope to create a system that can give individualized instruction. By analyzing time snapshot data from exams in an introductory programming course, we will write a program to construct state graphs for each student’s performance, eventually resulting in a Markov decision process that represents different approaches to writing the target program, and providing feedback to students. Once this system is sufficiently tested and refined, it will then be applied to subsequent semesters students in the programming course.

Effective feedback is necessary to support expertise development in clinical reasoning. Technology-rich environments (TREs) often use expert models as one means of providing this feedback. A review of empirical studies showed 3 different types of expert models in TREs: outcome models, process models, and dynamic models. This paper presents examples of each of these models and discusses their implications for the future design of feedback mechanisms to support clinical reasoning development through self-assessment.

This paper presents preliminary results on a generalizability study that was carried out to evaluate the robustness of a knowledge extraction algorithm.

ITS authoring frameworks are useful to reduce the efforts needed to create tutors, but the resulting tutors are usually more limited than domain-specific ones. The sophistication of the pedagogical behaviors they can produce depends on the knowledge components available to model a task domain. In this paper, we present research focused on integrating sophisticated domain-independent pedagogical behaviors in an ITS framework.

We are developing a general argumentation framework for implementing tutoring feedback in the form of persuasive dialogues. The objective is to have an intelligent tutoring system capable of arguing with the student to convince him of the rationale of the feedback provided to him. The application domain is that of medical diagnosis skill learning.

We want to create coordinate geometry learning environment with game –like properties, that is, elements of games that are engaging such as cover story, graphical representation, and animated feedback. This paper proposes that adding game-like properties to a computer tutor results in more student engagement and interest in the material. However, in addition to taking instructional time away, adding such properties imposes new limitations and difficulties in constructing content. Therefore, we have taken a measured and minimalist approach to making the original environment more game-like making a balance between stimulation and overload.

We compare three learning conditions on 57 middle school students’ short- and long-term retention at applying the control of variables strategy. Collapsing over time, direct instruction with reification yielded more robust learning than either direct instruction without reification or discovery learning conditions as measured by skill at constructing unconfounded experiments.

This study attempted to determine if it is possible to create an automatic affect detector using a combination of semantic and keystroke data. While the resulting models attained detection accuracies comparable with other studies, their reliabilities were not ideal. One model however shows that interaction logs may have a potential as a detector for confusion.

We study how student behaviors associated with engagement differ across different school settings. We present a study to investigate the variation in gaming the system and off-task behavior in schools in the USA and Philippines, using quantitative field observations on students using the same Cognitive Tutor lesson on scatterplots. We find that students in the Philippines go off-task significantly less but game the system significantly more than our sample of students in the USA. This study suggests that ITS designed for different settings or used in different settings will need to emphasize adaptation to different disengaged behaviors.

A key factor in the successful involvement of teachers in the development of intelligent tutoring systems (ITS) is a development paradigm that accommodates teachers’ skills and goals. In this paradigm, a mental model that is meaningful from the teacher’s perspective must be created for each task. We have focused on supporting teachers throughout the process of developing, deploying and analyzing Adaptive Tutorials that use Virtual Slides (ATuVS), which were created to assist learning of microscopic morphology. This was facilitated by the Virtual Apparatus Framework (VAF) – an ITS architecture that enables development of online learning activities analogous to real-world laboratory activities. VAF allows us to develop authoring tools that follow a well-established pedagogical process, which teachers can easily work with. In order to evaluate the effectiveness of VAF as a teacher-oriented design paradigm, we introduce the concept of “pedagogical ownership”. We argue that mainstream adoption of ITS in general, and ATuVS in particular, is only possible if teachers can assert pedagogical ownership over them.

The deteriorating patient activity (DPA) is a low-fidelity educational simulation that prepares medical students to effectively approach emergency situations. This paper outlines how we have captured and represented expertise in rapidly changing emergency situations to develop a dynamic and adaptive computerized model of the DPA to enhance medical teaching and learning.

DynaLearn is an Interactive Learning Environment that facilitates a constructive approach to developing a

conceptual

understanding of how systems work. The software can be put in different interactive modes facilitating alternative learning experiences, and as such provides a toolkit for educational research.

Inspectable (open) learner models have been in use for some time now. We present views of the learner model that have become more common, such as skill meters and concept maps; and introduce developments in less common interfaces for open learner models including the use of animation, audio and haptic feedback, and user-constructed learner model views.

This paper presents an empirical learning study using a prototype system designed to provide fully automatic, domain-independent conceptual personalization algorithms. The prototype system, the

C

ustomized

L

earning Serv

i

ce for

C

oncept

K

nowledge (CLICK), was implemented as an adaptive essay writing environment in a scientific domain. Results demonstrate that conceptual personalization promotes deep metacognitive strategies during online learning, and these strategies correlate with deep domain understanding.

This paper proposes a novel application of text categorization for two types questions asked in a micro-blogging supported classroom, namely relevant and irrelevant questions. Empirical results and analysis show that utilizing the correlation between questions and available lecture materials in a lecture along with personalization and question text leads to significantly higher categorization accuracy than i) using personalization along with question text and ii) using question text alone.

Successful individuals appear to have developed efficient method for reducing and coping with stress and anxiety. There is some evidence that emotions of this kind are correlated negatively with performance. Furthermore, it is important that new Intelligent Tutoring Systems (ITS) involve emotional coping strategies in order to improve the learner’s performance. In this paper, we show that problem-focused coping strategies are more efficient than emotion-focused ones for inducing positive emotions.

This paper presents results from an empirical study conducted with a novel subliminal teaching technique aimed at enhancing learners performance in an ITS. We replicated previous findings with the same technique but in a 2D environment. Non intrusive physiological sensors were used to record affective and cerebral responses. A brain analysis technique called Event-Related Potential (ERP), known to describe and confirm cognitive functions in the brain, provided strong evidence that subliminal cues and miscues were cognitively processed even though reported as not seen. The obtained results showed that only subliminal cues, not miscues, could significantly increase learner performance and intuition in a logic-based problem solving task.

This paper studies the influence of learner’s affective states on the EEG-mental engagement index during a problem solving task. The electrical activity of the human brain, known as electroencephalography or EEG was registered according to an acquisition protocol in a learning environment specifically constructed for emotional elicitation. Data was gathered from 35 healthy subjects using 8 biosensors and two video cameras. The effect of learners’ emotional states on the engagement index was analyzed as well as their impact on response time variability.

Increasing both user enjoyment and persistence, or engagement, is a challenge in ITS development. The current study investigates engagement in ITSs through the implementation of games and game-based elements. To investigate this possibility, we use both a physical version and a computerized version of the same educational board game. We compare user experience and reading strategy data for the implementation of two games – iSTART: The Board Game and MiBoard. We discuss game design implications for virtual environments and ITSs.

This study investigated players’ motivation during serious game play. It is based on a theoretical model of motivation (John Keller’s ARCS model of motivation) and EEG measures. Statistical analysis showed a significant increase of motivation during the game. Moreover, results of power spectral analysis showed EEG waves patterns correlated with increase of motivation during different parts of serious game play.

We describe a methodology for simulating student behavior to predict the effects of skill-learning parameter changes on system behavior. Validation against data collected after the changes were made shows that accurate predictions can be made despite a different cohort of students. Furthermore, deviations from the predictions may help explain unexpected effects of other changes made to the tutoring system.

Intelligent Tutoring Systems can be very expensive and complex to design, build and maintain. We explore the feasibility of adding automatic personalised feedback to an existing online learning system, by mining the student data collected by the system. This work was carried out on a web site in which students are taught programming basics in Python. Using 2008 and live 2009 data, the 2009 system generated hints to help students in topic areas they were found to be struggling with. We found that students who used the hinting system achieved significantly better results (26% higher marks) than those who did not, and stayed active on the site longer. A qualitative survey also revealed positive feedback from the students.

This paper explores automatically detecting student zoning out while performing a spoken learning task. Standard supervised machine learning techniques were used to create classification models, built on prosodic and lexical features. Our results suggest these features create models that can outperform a Bag of Words baseline.

Dynamic assessment (DA) has been advocated as an interactive approach to conducting assessments to students in the learning systems. Sternberg and others proposed to give students tests to see how much assistance it takes a student to learn a topic; and to use as a measure of their learning gain. To researchers in the ITS community, it comes as no surprise that measuring how much assistance a student needs to complete a task successfully is probably a good indicator of this lack of knowledge. However, a cautionary note is that conducting DA takes more time than simply administering regular test items to students. In this paper, we report a study analyzing 40-minutes data of totally 1,392 students from two school years. The result suggests that for the purpose of assessing student performance, it is more efficient for students to take DA than just having practice items.

One key component of creating an intelligent tutoring system is forming a model that monitors student behavior. Researchers in machine learning area have been using automatic/semi-automatic techniques to search for cognitive models. One of the semi-automatic approaches is learning factor analysis, which involves human making hypothesis and identifying difficulty factors in the related items. In this paper, we propose a hybrid approach in which we leverage findings from our previous educational data mining work to aid the search for a better cognitive model and thus, improve the efficiency of LFA. Preliminary results suggest that our approach can lead to significantly better fitted cognitive models fast.

In a tutoring system based on an exploratory environment, it is also important to provide direct guidance to students. We endowed iList, our linked list tutor, with the ability to generate proactive feedback using a procedural knowledge model automatically constructed from the interaction of previous students with the system. We compared the new version of iList with its predecessors and human tutors. Our evaluation shows that iList is effective in helping students learn.

Classical approaches for supporting tutoring services face several limitations for ill-defined domains. To overcome these limitations, we argue for the utilization of hybrid approaches for supporting tutoring services. In this paper, we describe a hybrid model that combines an expert system, the model-tracing paradigm, and a data mining approach in an ITS for learning to operate a robotic arm. The result is tutoring services that exceed what was possible to offer with each individual approach for this domain.

This paper examines the problem of modeling when students are engaged in “gaming the system.” We propose and partially validate an approach that uses a hidden Markov model, as is used in knowledge tracing, to estimate whether the student is gaming on the basis of observable actions. By doing so, we provide a common modeling approach that is applicable to gaming, or other constructs such as off task behavior. We find that our initial approach gave promising results, with parameter estimates that are plausible, and also exposed some weaknesses in our initial attempt. Specifically, that relying solely on response time is probably insufficient to construct a strong model of gaming.

This paper deals with a system which helps the designer of on-line learning to specify pedagogical scenario and more particularly on the guidelines wished for a learning activity given. The first idea is to propose for a designer, who is neophyte or expert, an adapted assistance for the description of a pedagogical scenario from a model of tutoring organization. The second is to export the pedagogical scenario in the standardized form of his choice depending of training platform used. In this paper, we describe the cooperative system EdiScenE, based on Tutoring organisation model, which includes author and help tools for the development of pedagogical scenario.

Operation Aries!

is a computer environment that helps students learn about scientific methods and inquiry. The system has several components designed to optimize learning and motivation, such as game features, animated agents, natural language communication, trialogues among agents, an eBook, multimedia, and formative assessment. The present focus is on a Case Study learning module that involves critiquing reports of scientific findings in news media that have flawed scientific methodology. After the human student lists the methodological flaws of a Case Study in natural language, a teacher agent and a peer agent hold a trialogue with the student that evaluates each listed flaw and that uncovers additional flaws that that student missed.

This paper presents a case-based reasoning (CBR) approach to provide adaptive support in microworlds. Interaction in microworlds is complex and unstructured, making the analysis of student behaviour difficult and the provision of computer-based feedback challenging. Our approach starts with the ellicitation of expected solutions to microworld tasks (both valid and common mistakes) to generate a case base. This is used to evaluate the actions of students and provide adapted feedback.

We present a distance tutoring system that allows a tutor to provide instruction via an animated avatar. The system captures pen-gestures of real tutor, generates 3D behaviors automatically, and animates a virtual tutor on the remote side in near real-tim The uniqueness of the system comes from the pen-gesture interface. We have done a study to test this interface. The system can effectively recognize and animate different types of gestures. Gesturing on the tablet and gesturing on the board were then compared. The results show that users can easily adopt to tablet, and pen-gesture on the tablet naturally. They were able to use pen-tablet interface effectively after a short instructional period.

There is interplay between brainwaves and learning. To describe and understand part of this complex interaction, this paper proffers a new learner model called the LBD Model (or Learning and Brainwaves Dominances Model). Twenty-three participants were recruited to validate this Model. Results show distinct instances of the LBD Model regarding three situations of learning: positive learning, unconscious learning and unlearning.

Intelligent Tutoring Systems (ITS) are increasingly used for distance learning around the world. However, most systems present the learning content regardless of the learner’s cultural background. This paper presents a resource personalization technique for cultural adaptation within ITSs. The approach is based on a collaborative filtering technique using an implicit cultural profile, which is automatically updated using the learner’s interactions with the system.

In this paper, an interactive learning environment for problem-changing exercise where a learner is required to solve and change the problem has been described. Activity to make a new problem from the original one and to compare their solutions is promising to promote a learner to be aware of the relation between the problems. For knowledge-rich problems, for examples word problems in arithmetic, mathematics or physics, this awareness is very important to master the use of solution methods. In order to realize such exercise in physics, we have already developed a prototype of computer-based learning environment that allows a learner to change a problem and can also diagnose the problem change and give feedback for the learner.

In the mathematics domain of decimals, students have common and persistent misconceptions. These misconceptions have been identified, studied, and published by many researchers, spanning over 80 years of time. However, no paper discusses and brings together

all

of the identified misconceptions. This paper presents an initial taxonomy of decimal misconceptions, summarizing the results of past work. We also discuss the potential use and benefits of such a taxonomy in supporting the development of intelligent tutors that use erroneous examples as a learning tool for middle-school math students.

iSTART is an Intelligent Tutoring System designed to improve students’ reading comprehension skills. iSTART was the main component in a long term experiment (across a full academic year) with 389 students who completed a pretest, interacted with iSTART for 6 months, and then completed a posttest. A new extended practice module was implemented, which provided students with repeated practice across a variety of texts. Analyses found improvement in performance for all students, and indicate that students’ initial self-explanation abilities may differ, but these abilities improve and converge as a function of practice.

One should expect the unexpected when deploying intelligent tutoring systems. This paper describes a case study in collecting, warehousing, and analyzing field usage data from two language and culture learning environments, to understand what happened when they were deployed. A data warehousing system, Hoahu, was used to process the raw data and transform it into a relational database to facilitate queries and analysis. The system also supported data annotation by subject matter experts to facilitate comparison of automated assessments against human raters. Errors and inconsistencies in the data were identified and corrected. The resulting data warehouse has proven valuable for understanding the trajectory of learning over extended periods of time and analyzing the strengths and weaknesses of complex interactive subsystems such as spoken dialog systems.

Authoring the domain knowledge of an intelligent tutoring system (ITS) is a well-known problem, and an often-mentioned approach is to use authors who are domain experts. Unfortunately, this approach requires that potential authors learn to write and debug knowledge written in a formal knowledge representation language. If authors were able to use natural language to represent knowledge it would allow them to add and update knowledge far more easily. In this paper, the design of such an authoring system, ‘Natural-K’ is presented. Natural-K is an authoring system in which domain authors including non-programmers are able to add problem statements and background knowledge such as commonsense, in natural language.

As online discussion boards become a popular medium for collaborative problem solving, we would like to understand patterns of group interactions that lead to collaborative learning and better performance. In this paper, we present an approach for assessing collaboration in online discussion, by profiling student-group participation. We use a modularity function to compute optimal discussion group partitions and then examine usage patterns with respect to high-versus low-participating students, and high- versus low-performing students as measured by grades. We apply the profiling technique to a discussion board of an undergraduate computer science course with three semesters of discussion data, comprising 142 users and 1620 messages. Several patterns are identified, and in particular, we show that high achievers tend to act as ‘bridges’, engaging in more diverse discussions with a wider group of peers.

While there is acknowledgement of the importance of metacognition in education, some researchers indicate that the domain of metacognition lacks coherence. In order to overcome this issue, it is necessary that each researcher explains his own approach by using his own or other people’s framework of metacognition. We propose a new framework for metacognition which explains what types of metacognition-driven learning occur so to enable regulation of cognitive activities. With this framework, it becomes possible not only to identify what types of metacognitive activity a computer system supports but also to propose new functions that support the various types of metacognitive activities.

Our previous work has demonstrated that the Extensible Problem Specific Tutor (xPST) framework lowers the bar for non- programmers to author model tracing intelligent tutoring systems (ITSs) on top of existing software and websites. In this work we extend xPST to enable authoring of tutors in 3D games. This process differs substantially from authoring tutors for traditional GUI software in terms of the inherent domain complexity involved, different types of feedback required and interactions generated by various entities apart from the student. A tutor for a village evacuation task has been constructed in order to demonstrate the capabilities of using the extended xPST system to create a game-based tutor.

Accelerated future learning, in which learning proceeds more effectively and more rapidly because of prior learning, is considered to be one of the most interesting measures of robust learning. A growing body of studies have demonstrated that some instructional treatments lead to accelerated future learning. However, little study has focused on under- standing the learning mechanisms that yield accelerated future learning. In this paper, we present a computational model that demonstrates accelerated future learning through the use of machine learning techniques for feature recognition. In order to understand the behavior of the proposed model, we conducted a controlled simulation study with four alternative versions of the model to investigate how both better prior knowledge learning and better learning strategies might independently yield accelerated future learning. We measured the learning outcomes of the models by rate of learning and the fit to the pattern of errors made by real students. We found out that both stronger prior knowledge and a better learning strategy can speed up the learning process. Some model variations generate human-like error patterns, but others learn to avoid errors more quickly than students.

Curriculum Sequencing is one of the most interesting challenges in learning environments, such as Intelligent Tutoring Systems and e-learning. The goal is to automatically produce personalized sequences of didactic materials or activities, on the basis of each individual student’s model. In this paper we present the extension of the

LS-Lab

framework, supporting an automated and flexible comparison of the outputs coming from a variety of Curriculum Sequencing algorithms over the same student models. The main aim of

LS-Lab

is to provide researchers or teachers with a ready-to-use and possibly extensible environment, supporting a reasonably low-cost experimentation of several sequencing algorithms. The system accepts a student model as input, together with the selection of the algorithms to be used and a given learning material; then the algorithms are applied, the resulting courses are shown to the user, and some metrics computed over the selected characteristics are presented, for the user’s appraisal.

General-purpose ontologies (e.g. WordNet) are convenient, but they are not always scientifically valid. We draw on techniques from semantic class learning to improve the scientific validity of WordNet’s physics forces hyponym (IS-A) hierarchy for use in an intelligent tutoring system. We demonstrate the promise of a web-based approach which gathers web statistics used to relabel the forces as scientifically valid or scientifically invalid. Our results greatly improve the F1 for predicting scientific invalidity, with small improvements in F1 for predicting scientific validity and in overall accuracy compared to the WordNet baseline.

The ability to argue is essential in many aspects of life, but traditional face-to-face tutoring approaches do not scale up well. A solution for this dilemma may be computer-supported argumentation (CSA). The evaluation of CSA approaches in different domains has led to mixed results. To gain insights into the challenges and future prospects of CSA we conducted a survey among teachers, researchers, and system developers. Our investigation points to optimism regarding the potential success and importance of CSA.

We are interested in cognitive diagnosis systems able to understand the learners’ work in environments involving them in various problem solving modalities. We are designing a diagnosis model taking into account various factors. In this paper we are interested in the problem solving modality factor and we present an experiment for analyzing the impact of it on the diagnosis.

This paper presents empirical results about the behavior effect of two different hinting strategies applied on exercises within an ITS: having some penalty on the scoring for viewing hints or not having any effect on the scoring; and hints directly available or only available as a result to an incorrect attempt. We analyze the students’ behavior differences when these hinting techniques changed, taking into account the type and difficulty of the presented exercises.

Prior work supports the pedagogical value of conceptual maps for offering students an overview of a topic as well as the connections between subtopics. In this poster we describe a system that uses automated topic modeling technology to map the topics and sub-topics in a collection of documents. An interactive graphical representation allows users to explore this topic analysis, using it as an interface for browsing a collection of documents. We present a small user study evaluating the usability of the interactive map.

Previous research using concept maps as domain models in intelligent tutoring systems has demonstrated their power and flexibility. However, these concept maps must still be authored by a domain expert, creating a development bottleneck. We present a new, streamlined methodology for automatically extracting concept maps from textbooks using term extraction, semantic parsing, and relation classification.

In this paper we present a theoretical framework describing an original method for the design of intelligent tutoring environments, building upon the notion of shared attention. This framework is defined as the

Levels of Interaction

(LoI) approach. It is applicable to applications where the learner/player is immersed in a 3D virtual environment, interacting and exchanging knowledge with an adaptive crowd of conversational agents.

T-algebra is an interactive learning environment for elementary algebra. The main program of T-algebra enables visualizing information about a particular student in tables, indicating solution times, the numbers of errors (in 20 categories) and hint usage for each task. Additional software for teachers allows examination of solution results in group views.

Mily’s World

is a learning environment for coordinate geometry that has game-like properties, that is, elements of games that are engaging such as cover story, graphical representation, and animated feedback. This paper proposes that adding game-like properties to a computer tutor results in more student engagement and interest in the content material. We have taken a measured and minimalist approach to making the original environment more game-like by making a balance between stimulation and overload. We received mixed result in our experiment with sixty six students.

Literature shows that Intelligent Tutoring Systems (ITS) are growing in acceptance and popularity because they increase performances of students, leverage cognitive development, but also significantly reduce time to acquire knowledge and competencies. We present an ITS offering the opportunity of evaluating various metacognitive indicators and able to share this information with other learning tools. Our online tutor is based on an existing ITS authoring tool that we extended to support metacognition and share learners’ profiles and activities into a standardized, distributed and open tracking repository.

This paper proposes a new approach for incorporating intelligent learning environments (ILEs) into K-12 classrooms that tightly integrates interactions between the students, the classroom teacher, and the ILE. The ILE’s ability for continual, fine-grained monitoring and analysis of students’ learning activities supports the teacher’s ability to more effectively guide student learning.

In this study, we compared our computer tutor (“TED” for Training in Experimental Design) to a teacher-guided control lesson also targeting experimental design but incorporating hands-on learning. Students in both groups showed significant gains in ability to design unconfounded experiments. TED instruction was significantly more efficient than the control lesson. When the teacher’s ratings of student ability were co-varied, students in the TED condition significantly out-performed control students on both immediate and delayed far transfer assessments taken three weeks after instruction. Students in both groups also reported a preference for physical over virtual materials.

The development of intelligent tutoring systems (ITS) capable of supporting training experiences for geographically-distributed team members in shared virtual simulation environment presents considerable challenges. Even human tutors face challenges in developing team cohesion, coordinating roles and assessing contributions. Just as a human tutor might assess collective performance, a team ITS must be capable of passively assessing the trainees’ readiness to learn and evaluating their progress toward team objectives. Passive sensing methods offer the opportunity for the ITS to understand the team’s cognitive and emotional state without interfering with the learning process. It also helps determine their any interventions needed to optimize performance. This article reviews challenges and hypothesizes functions for computer-based distributed team tutors.

“

Share, Remix, Reuse - Legally

”, the tagline for creative commons, cogently captures the ethos of peer production. Through the rapid growth of open educational resources (OER), peer production has begun to play a major role in how we teach and learn. OER are teaching and learning resources that reside in the public domain or have been released under licensing schemes that allow their free use or customization by others. They encompass a multiplicity of media types, including lesson plans, animations, videos, scientific data, etc. OER can be created by scientific institutions, by university faculty, by K-12 teachers, or by learners. Here, we focus on K-12 teachers engaging in peer-production for instructional purposes.

This paper describes Annie, a domain-independent intelligent tutor that can be “plugged-in” to digital games to guide learners using the core mechanics of the game.

We present students with common errors of others in the context of an intelligent tutoring system (ITS). We conducted two studies with students of different curriculum levels to measure the effects of learning through such erroneous examples. We report that erroneous examples with additional support can assist lower curriculum level students develop better meta-cognitive skills.

We present a feasibility study of an intelligent tutoring system Peoplia in which a socially intelligent tutoring agent uses common instructional methods that are augmented by social features to help students learn. Peoplia features pseudo-tutor assessments, free-text answering, personalized question generation, and adaptive question selection. It allows students to work both individually and collaboratively while the tutoring friend monitors their social behavior and motivates them by socially relevant interventions.

Recent research has emphasized the importance of reflection for students in an intelligent learning environment. This study tries to investigate whether agent prompts, acting as scaffolding, can promote students’ reflection when they act as tutor through teaching the agent tutee in a learning-by-teaching environment. Two types of agent prompts are contrasted in this research, both from the perspective of a tutee, differing in their specificity. Reflective prompts are content-independent tutee questions, aiming at fostering students’ general reflection on metacognitive strategies and beliefs. Interactive prompts, on the other hand, are content-dependent tutee questions that encourage students’ specific reflection on domain-related and task-specific skills and articulation of their explanatory responses. The result indicates that designers on intelligent learning environment should concentrate on fostering students to reflect on their metacognitive strategies and beliefs, and allow students to take responsibility for directing their own learning autonomy.

In this paper, we use supervised machine learning to automatically identify the problem localization of peer-review feedback. Using five features extracted via Natural Language Processing techniques, the learned model significantly outperforms a standard baseline. Our work suggests that it is feasible for future tutoring systems to generate assessments regarding the use of localization in student peer reviews.

Problem solving using a programming language such as C++ is a complex multi-step task. AlgoTutor trains introductory computer science students to start the problem solving process with algorithm design. The system then helps students learn how to trace an algorithm with an execution trace visualization tool. The final step of implementing an algorithm in code is accomplished with the ProgramPad portion of AlgoTutor. In ProgramPad students can see the connection between the algorithm and the converted C++ program. We have shown that some of the difficult programming concepts such as function parameters and array concepts can be addressed using ProgramPad. An additional benefit of AlgoTutor is that students can practice algorithm development skills and algorithm implementation skills in a web-based, interactive environment. The system provides online feedback for their algorithm so that students can work at their own pace at a convenient time.

Assessment-based educational games can produce useful information to guide student instruction. This paper describes an approach for integrating components of video games with those of adaptive technologies and assessment into the design of educational games. Three examples in the areas of English language learning and mathematics are also presented.

The Cognitive Tutor Authoring Tools (CTAT) [1] are a suite of software programs meant to make the creation of web-based ITS practical for non-programmers. CTAT supports a relatively novel type of tutors, called

example-tracing

tutors, which use examples of problem-solving approaches to assess and guide students as they practice solving problems. CTAT employs a programming-by-demonstration paradigm that relies on creating examples of how problems are to be solved, rather than defining general rules or constraints that characterize solutions or solution processes. The result is an

editable behavior graph

that contains the tutor’s intelligence about how to react to student actions and what hints to give for next steps. Although relatively easy to build, example-tracing tutors support the key behaviors identified by VanLehn [2] as characteristic of ITS. Data from over 26 research studies using CTAT indicate that these tools lower the cost of ITS development by a factor of 4-8.

Although educational games can be engaging and motivating for students, the game aspects tend to take up time that could have been used for instruction, and they add to the cognitive load of the students. Therefore, instead of completely integrating educational content into a game framework, we instead choose to start with a computer tutor and selectively incorporate those features of games that are engaging but do not overly detract from learning. Following this approach, we created a learning environment for coordinate geometry with

game-like properties

. We define game-like properties as the elements of games such as visual representation, cover story, and animated feedback, which are responsible for their engaging nature. While there have been many attempts to use game-like environment in tutors (e.g. Wayang Outpost), our approach differs by using an iterative process in a space of ITS and games, within the framework of stimulation and overload.

In this interactive demo we will present the Adaptive eLearning Platform – a web-based solution for the development, deployment and analysis of Adaptive Tutorials. Adaptive Tutorials (AT) are online instructional activities that exhibit three levels of adaptivity: students experience adaptive feedback with remediation targeted to their intrinsic misconceptions, while their activities are also sequenced adaptively based on performance. The third level of adaptivity is content adaptation through analysis and reflection.

In previous work, we developed SlideTutor - an Intelligent Tutoring System that teaches visual classification problem-solving in Pathology, and shown that use of the system is associated with rapid learning gains. Development of the SlideTutor system and content has required many years of effort by system developers, knowledge engineers and domain experts. Both cases and domain ontologies must be manually created and validated. The scope of medical knowledge that must be covered is extremely large. The further development of tutoring systems for visual classification in medical domains (including our own) requires software that reduces this high development burden. Towards this goal, we sought to create a generic framework for developing visual classification tutoring systems in medical fields such as Pathology or Radiology. In this interactive event, we present the first and most difficult step towards such a generic visual classification ITS authoring system – the component for creating and validating cases and domain ontologies.

This demonstration will present a novel community environment ‘AWESOME Dissertation Environment (ADE)’ which uses semantic wikis to implement the pedagogical approach of ‘social scaffolding’. ADE was developed within an interdisciplinary UK research project called AWESOME (Academic Writing Empowered by Social Online Mediated Environments) which involved the universities of Leeds, Coventry and Bangor. The environment was instantiated in several domains: Education, Fashion and Design, Philosophy and Religious Studies, and an Academic Writing Centre. Following both the encouraging feedback from the trial instantiations and the challenges faced in deploying the environment in practice, we conducted a second stage of the project which aimed at adapting the ADE to dissertation writing in computing. Following the lessons learnt from the first stage, we now performed a systematic approach to tailor the existing community environment to meet dissertation writing needs in a specific domain and in a particular educational practice.

This demonstration will show how a tutor can detect the content of collaborative behavior and offer relevant domain level interventions. Rashi is a domain independent intelligent tutor providing students with practice using inquiry skills. When working on human biology, students interact with a virtual sick patient whom they must successfully diagnose. Rashi supports students as they create hypotheses and collect data to support and refute these hypotheses. In order to increase the efficacy of Rashi, we incorporated collaborative tools that support group efforts by supporting students as they dynamically share experiences and work together to reach a diagnosis. In addition to this, Rashi contains an intelligent agent that examines collaborative efforts and automatically detects the expert knowledge students are working with. Visitors to this demo will first explore these collaborative tools in detail. Two people will collaborate about a diagnosis and the intelligent agent will examine their collaborative activity and compare it with an expert knowledge base, to determine what domain content is relevant to their activities. The tutor will provide interventions to the visitors that leverage this content recognition.

This demonstration provides evidence that expert knowledge bases are a plausible development option for intelligent tutoring systems because they can leverage content recognition to provide more useful feedback. In Rashi, some collaborative content is recognized when students manually match discussion items to expert knowledge. However, the greatest impact comes when the tutor recognizes participants’ content by matching words and phrases in the chat conversation. Experiments show that the tutor can recognize this content correctly with more than 70% accuracy. Thus, it can provide interventions that suggest what direction students might take if they reached an impasse.

This demonstration provides evidence that complicated NLP techniques are not always necessary; a tutor can understand domain level student activity and provide useful interventions using a well-built expert knowledge base. In addition, we show that even though the lack of more complicated techniques may lead to some error in content recognition, we can provide unique forms of feedback that are not detrimental to students when the content is incorrectly recognized, but is significantly helpful when it is correctly recognized.

We present our computer-based learning environment, Science Assistments (http://users.wpi.edu/ sci_assistments/; NSF-DRL # 0733286; NSF-DGE #0742503; U.S. Dept of Ed. # R305A090170), for Physics, Life Science, and Earth Science that scaffolds middle school students’ scientific process skills, namely, hypothesisgeneration, design of experiments, data collection, data interpretation, and warranting claims with evidence. Our project builds on prior development by the investigators of the Math Assistments project (http://www.assistment.org/). Specifically, we utilized the existing authoring functionality of the Math Assistments system and extended the logging functionality in order to capture students’ fine-grained actions within interactive microworlds. In addition, we developed a suite of inquiry tools to support students’ inquiry in terms of the five skills mentioned above. Together, the logging functionality and the inquiry tools provide the basis for adaptive scaffolding of students’ inquiry in real time. By reacting to students’ inquiry strategies in real time, we hypothesize that it will be possible to positively affect both students’ science process skills, shown by more goal directed inquiry and more systematic experimentation, measured through log files, as well as students’ content learning, as measured by prepost test gains. We plan to test our adaptive scaffolding in a series of randomized controlled studies in our four partner schools; the demographics of these students represent a wide range of SES and ethnic backgrounds, and thus, our data should generalize well. Goal outcomes include empirical data regarding the efficacy of our system at improving students’ science learning, namely, inquiry skills and content learning, across several dependent measures in each content domain.

Mixing worked-out examples with problem solving has been shown to be an effective blend of educational activities [1]. Given the positive impact on learning, some Intelligent Tutoring Systems attempt to incorporate worked-out examples into their learning environments. The approach taken by Carnegie Learning’s Cognitive Tutor, called

Interactive Examples

, was created to solve two design challenges.

Two of the most important characteristics to support learning of geometric concepts are (a) the possibility of allowing students to interact with geometric objects and (b) the capability of guiding these students during the process of interaction by helping them to discover interesting properties and to make “plausible” conjectures. Available interactive/dynamic geometry software allow for rich interaction between geometric objects and users, but with no computational support for guiding students. This work presents an on-going effort to design an interactive geometry software, referred to iGeom, that have both interactive geometric objects that students can fully manipulate and the capability of helping students by providing personalized guidance and feedback when needed. iGeom have been developed in Java, freely available on internet (http://www.matematica.br/igeom), and can be easily incorporated into other programs such as learning managements systems or ITSs. To provide guidance capabilities we have been using CTAT (http://ctat.pact.cs.cmu.edu/), an authoring tool that facilitates fast development of cognitive tutors.

There is a need for software that supports learners in actively dealing with theoretical concepts by having them create models and perform concept prediction and explanation (e.g. [3,4,5]). DynaLearn seeks to address this by developing a domain independent Interactive Learning Environment (ILE) based on Qualitative Reasoning (QR) [1]. The QR vocabulary fits the nature of

conceptual

knowledge, and the explicit representation of these notions in the software provides the handles to support an automated communicative interaction that actually discusses and provides feedback at the

conceptual

level.

The effect of tutor learning has been studied in various contexts, providing ample evidence to suggest that students learn when they teach others. Yet, the cognitive and social factors that facilitate or inhibit tutor learning are still not well understood. One factor that prohibited research progress in this area is that studying the tutor learning effect could often be done only at the cost of tutees’ learning. To address this problem, we built an on-line learning environment where students learn by teaching a computer agent, called

SimStudent

, rather than their peers [1].

Step-based ITS have been proven successful for well-defined domains, particularly in well-defined tasks, but their success is mitigated by their cost. Typically, the main factor behind the cost is the efforts needed to model the task domain. Different approaches have been investigated to reduce these efforts: Model-Tracing Tutors (e.g. Cognitive Tutors, Andes), Constraint-Based Tutors (e.g. SQL-Tutor, ASPIRE) and Example-Tracing Tutors (e.g. CTAT’s, ASSISTment).

Project LISTEN’s Reading Tutor listens to children read aloud, and helps them learn to read, as illustrated on the Videos page of our website. This Interactive Event encompasses both this basic interaction and new extensions we are developing.

Carnegie Learning’s Cognitive Tutors for mathematics have been the subject of a wide variety of research [3,4] and are the most widely deployed Intelligent Tutoring Systems. Currently, over 560,000 students and 2,700 schools in all 50 United States are using them. Many ITS researchers understand these tutors from printed works but have not had the opportunity to use the tutors in a hands-on fashion.

Geometric proof has long been a topic of study within Intelligent Tutoring Systems [3,4]. Proof is interesting because it supports a variety of solutions and strategies. However, implementation is challenging and such tutors have not been widely deployed.

ISLET is a multiplayer role-playing game whose goal is to help players develop intercultural communication skills for French-speaking Sub-Saharan Africa. Its vision is to utilize multiplayer gameplay to create the most compelling language learning environment available.

The Pittsburgh Science of Learning Center’s DataShop is an open data repository and set of associated visualization and analysis tools. DataShop has data from thousands of students deriving from interactions with on-line course materials and intelligent tutoring systems. The data is fine-grained, with student actions recorded roughly every 20 seconds, and it is longitudinal, spanning semester or yearlong courses. Currently over 188 datasets are stored including over 42 million student actions and over 150,000 student hours of data. Most student actions are “coded” meaning they are not only graded as correct or incorrect, but are categorized in terms of the hypothesized competencies or knowledge components needed to perform that action.

This interactive event presents a pressure sensitive chair constructed out of Wii Fit game controller boards. During this event, we will demonstrate how to arrange the boards to detect seat and back pressure, configure a PC to receive a bluetooth datastream from the boards, and modify the power supply of the boards to increase uptime and reliability. We claim that the pressure sensitive chair so constructed is highly suitable for recovering posture information from a student interacting with an ITS.