Formalization to Improve Lifelong Learning.

SYLVIE RANWEZ [1]

TORSTEN LEIDIG [2]

MICHEL CRAMPES [1]

Utilizing the World Wide Web (Web) for delivering lifelong training services to individuals or communities raises great interest in intelligent methods for adaptive course curriculum building. People notice that it is insufficient to make training materials accessible by way of the browser. To satisfy learner requirements, the learning service has to give a highly customized and up-to-date view. Moreover, training is an interactive and collaborative process. The flow of the curriculum has to follow some didactic principles. Learners need to be guided through the training material according to these principles and their own preferences. A conventional hypertext system cannot provide this, mainly because the pedagogic semantics of the material are not "known" by the system.

This article presents a new approach for adaptive, computer-supported learning services within our Lifelong Learning project. This approach is based on a formalization of pedagogical knowledge, about learning material and learning processes, as well as the domain knowledge and a learner model. The goal of our research is to integrate efficient and reusable pedagogical knowledge representations into an adaptive learning system (and authoring tools). The article concentrates on the Didactic Planning obtained by various Pedagogical Strategies. Therefore, we focus on the strategy formalization using an expressive knowledge representation based on a dedicated learning ontology using conceptual graphs. The power of our approach for designing a new kind of adaptive learning system is demonstrated through examples of pedagogical strategies applied to course material.

The Internet creates a great interest in education. It brings out many exciting features to support trainees and trainers such as distant learning, course delivery, learner-teacher communication, and communication between learners in distributed rooms. Learning as a lifelong and expansive process reinforces this interest, particularly in industries where large amounts of money are invested in employee training.

Hypermedia systems have been widely used to store and explore large amounts of related information. Hypermedia systems are also often used for learning. However, the hypermedia model is very focused on linking of information and has little support to express the semantics of an application domain. Therefore hypermedia systems are not very well suited for semantic-driven applications and adaptive learning style modelling.

One of the fundamental problems is how to traverse the structures with a good strategy to find the right information. Pure, navigational-based systems have been severely criticized in the domain of education because the learner is often left free to choose whatever link they feel is attractive (Lindard & Zeilinger, 1995). Scalability is not well supported, thus authoring, navigation, and maintenance in the hypermedia net is too complicated if the number of nodes and links grow too large.

The Web has become a very popular Hypermedia system. For several reasons, the Web has become a de-facto standard, and it is often used as an infrastructure for Intranet and Internet applications. However, we observed that little research has been done in making web-based learning more interactive in the sense of applying pedagogical or didactic guidance to the learner and allowing those systems to reuse pedagogical materials. Several required characteristics in the context of hypermedia and learning need a closer look:

Adaptation. An adaptive authoring tool allows teachers to apply several pedagogical strategies and to design their courses.

Navigational support. A support for learner navigation through the course material is necessary. The learner has to be supported by an adaptive interface that takes individual performance into account (what that learner already has done, preferences, results of some tests, etc.).

Multimedia. Each course is composed of course segments (CS) that have to be delivered by way of several media types (e.g., videos, picture, sounds, and text). A CS is an indivisible unit of information that is going to become a part of a course and should be used as a presentation for one or more knowledge entities. Each CS must be annotated to allow the system to retrieve the most relevant CSs that are going to be formed in the course.

Our previous approach to telelearning was based on an extended hyper-text model with typed nodes and links in a fixed network. In addition, we used predicate logic to control whether or not certain links are navigable in a certain situation. Although we could provide the learner with a lot of navigational help, the courses were still very fixed and inflexible. We tried to overcome this limitation by a generation of dedicated hypertext courses for a certain audience. We always felt that the hypertext approach may not be a good solution to tackle the learning problem and searched for a more flexible approach than hypertexts.

THE SIBYL APPROACH

The mythical Sibyls were wise women, prophetesses with great influence in times long past. Often named by location and title rather than birth name, they are the Persian, Erythrean, Delphic, Cumaean, and Libyan Sibyls. Usually, the Sibyls are displayed as partners to the prophets, both sources of inspiration and wells of knowledge.

Analogous to that mythical picture, our didactical Sibyls:

* Have knowledge about the didactical quality of learning material (wells of knowledge).

* Are "asked" by the learner to give relevant learning contents (inspirations).

Our central goal is to use a conceptual model of the didactical structure of the learning material to give an educated choice of alternatives to the learner of how to navigate, based on the learner's previous activities and knowledge. A raw illustration of the possible learner interface used by our first prototype is given in Figure 1. The central part of the screen is occupied by the learning content. Other parts are generated by the Sibyl system and give information about the didactical nature of the displayed content as well as educated hints about other relevant instructions (at the moment only the generated references to other instructions are shown). In order to accomplish this, a formalization of didactical knowledge must be the first step.

Knowledge Representation using Conceptual Graphs

We used several models to express the learning knowledge. First there is the knowledge that concerns the content of the course domain. Second there is the pedagogical and didactic structure of the learning material, which can be expressed with a vocabulary that is independent of the taught domain. Finally, we have supplemental knowledge about the learner, which is subsumed to the learner model.

The central idea of Sibyl is the use of CGs for the representation of IP, people, and learning structures as a common knowledge representation. The theory of CGs (Sowa, 1984) is often used for knowledge representation in the domain of natural language comprehension. CGs describe the semantics of a text and its meaning. However, CGs can be applied to several domains. Additionally, because of its non-ambiguity and ease of use, this formalism is used in several applications such as knowledge acquisition, information retrieval, or reasoning on the conceptual knowledge.

In Sibyl, we use CGs as a uniform representation for describing domain knowledge, competencies, and learning strategies. All the learning resources are available through a large number of rather small CGs. The CGs are implicitly linked to each other through the generic and domain specific ontology. So, our learning system basically becomes a general machine that performs queries and operations submitted by humans or agents into the concept database. Also, it gives a multimedia presentation of the retrieved knowledge entities to the user. Finding the required knowledge sources is done by matching the concept structures with instructional and domain specific rules.

Our approach provides some important advantages in comparison to conventional hypertext systems. The most important advantage is scalability. The knowledge provider (author) does not have to know the overall structure of the hypertext network, which usually becomes very large and complex in real applications. Instead, the author only needs to know the basic ontology and has to describe the single knowledge entities in terms of that ontology.

Adaptation to the needs of the individual learner and instructional navigational support is possible by the design of the matching rules. Also, simulation and collaboration become quite easy because the domain-specific concepts are part of the stored information and can be the subject of queries.

Using the uniform conceptual graphs for all the models leads to a very high degree of integration and flexibility. Further, conceptual graphs are relatively easy to understand for non-technicians (at least in their graphical display form), and can serve as an ideal communication means between pedagogic experts and computer scientists.

Modeling Pedagogic Knowledge

Our model is composed of six parts.

* Information concerns the analysis of the content of a piece of knowledge. a unit of information can be a classification, a fact, a process, a principle, or a structure. One can also see this as the cognitive nature of the knowledge.

* Instruction refers to the form that a teacher (human or computerized) gives to information in order to transfer it to the learner. This form makes the information more accessible and understandable, and it helps in the structure of the course. An instruction can be an annotation, an introduction, a conclusion, a summary, a definition, an explanation, an example, an exercise, a hypothesis, etc. For example, information can concern Beethoven's birthday (fact). To make the student replace this person in historical and professional context, we can say that Beethoven is an example of a sonata composer. We applied Merrill's (1991) transactional instructions to maintain a model of the learner's state.

* Learner State concerns all the knowledge used to model the learner during a student's curriculum, which includes status in the curriculum (nodes visited, answers given to some tests, preferences concerning the media, the strategy, etc.) as well as personal characteristics (diplomas, professional experiences, age, etc.).

* System functionality is comprised of all characteristics and functionality of the learning system. In these parts, we can find the definitions of all actions and functions of the system, including interaction between the users and the system.

* Learning goals and objectives explains learner knowledge or ability if a certain instructional unit was successfully completed. The learning objectives have to be specified in a very formal way because they are used as an anchor for finding the appropriate instructions.

* Media concerns the different media types and usage for the instructions. Several media types can be used to represent a piece of instruction. The most useful type depends on the cognitive nature of the knowledge and the learner's preferences. For example, a table or an animated flowchart might best represent a procedure.

Instead of constructing a new ontology from scratch, we used a part of the learning ontology given by Mizoguchi, Sinitsa, and Ikeda (1996a) as a base of our model. We made some simplifications to adapt it to our needs, and added special elements according to pedagogical strategies.

Figure 2 shows the overall classification of concepts of the pedagogical ontology. Note that the individual types have more subtypes, which are omitted for simplicity reasons. In the following pages, we give some examples on how the conceptual knowledge is expressed with conceptual graphs. Some examples are given using our prototypical course about the sonata music form. Our ontology is not complete now because it reflects some ideas, which can be handled by our prototype system. We see a lot of room for extensions and improvements, and we also recognize that such a learning ontology can never be complete. The state-of-the-art of pedagogic science will change over time, and the pedagogic, multimedia distance education is still a relatively new area.

LEARNING GOALS

An important prerequisite of a good course is an in-depth analysis of the coarse and fine-grained learning goals and objectives. Usually, the learning goals are defined on paper in an early design phase, which is not supported by tools (if they are defined anyway). In a later design stage, the learning goals are only present in the designers' heads and are never used by the final tutoring system. However, in an intelligent tutoring system, some explicit and detailed learning goals can provide a good foundation for reasoning about appropriate learning material. Learning goals have a general schematic form in our model, which is outlined in Figure 3.

Note that the concepts are not simple concepts, rather they denote a set of possible structures defined by closures, much like rules of a grammar. Circumstances for examples could expand to a situation such as "playing a simulation game," which is shown in Figure 4.

A possible action (ACT) could then be that the student is showing a desired behaviour, which can also be expressed as conceptual graph. The pedagogic ontology has to provide enough expressiveness to allow a wide spectrum of graphs. On the other hand, it should be small enough to be easy to learn and manage.

INFORMATION TYPES

Also, the information types are modeled with CGs. We only give an example of structures in Figure 5. Structures are things that are composed of several parts. Sometimes, these parts are in a specified order. A learning strategy which deals with parts can propose a top-down or bottom-up approach to teach how things are composed.

In our sonata course, a sonata is a structure, composed of several parts such as exposition, development, recapitulation, and coda. Each of the parts are also structures because they are composed of different subparts. Note that the exposition and the recapitulation are usually related to the same theme in a sonata.

PEDAGOGICAL STRATEGIES

Pedagogical Strategies (PS) are those heuristics that a teacher uses to direct the navigation of the learners through a set of pedagogical resources (exposure, exercises, simulations, etc.). All teachers use PS. They are often implicit, sometimes even for the teachers who make use of them. As a result of the application of the PS, the learners can only see pedagogical plans and local pedagogical interactions. Figure 6 shows a strategy rule, which says that if an explanation and an example refer to the same topic, then the presentation of the explanation precedes the presentation of the example.

In traditional education, where the teacher and the learners meet in a real classroom, the teacher permanently reconsiders the pedagogical plan according to the difficulties encountered by the learners. The teacher spots these difficulties in different ways such as observation of the learners' attention, the results of a test, learner's questions, etc. In Computer-Based Learning (CBL), there is no teacher to supervise the learning process, and the learners must be guided by some pedagogical knowledge-based system. Such expertise has always been an important area of investigation in the Intelligent Tutoring System (ITS) community. With the intense spreading of the usage of the Internet in schools, companies, and homes, and with the problem of optimizing the pathways of learners, the model of this expertise is becoming more acute.

Conducting the Parts

In order to provide the learner with an adaptive navigational aid, we have to solve four major problems:

1. Collect all information from the learner and to build the learner model.

2. Retrieve the relevant data in the KB for a special course. This retrieval Consists of selecting the CSs that will form the course, with respect to the Learner State, the Pedagogical Strategy chosen, and author's directives.

3. Order the CS retrieved above in a coherent way and to build a course. In other words, build a Didactic Plan (DP) constituted with the CS ordered with respect to learner's state and pedagogical strategies. An example is detailed in Figure 7.

4. Compose the didactic plan and present it in a web browser, i.e., to automatically build a page having a title, the content of the current CS and hyperlinks on following CSs (Figure 1 is a view of the learner interface).

It should be noticed that the DP is given to the learner under an understandable form for the learner: only the hyperlinks that are advised or that can be chosen are given to the learner. By this way the didactic strategy is transparent. Moreover, this DP is evolutive due to the learner's current state through the courses (node visited, answers given to some tests, etc.).

TECHNICAL VIEW

Architecture

We consider that the resources and services described above are either available on, or referenced by, a learning server that gives the users access through a dynamic, interactive web interface. According to the situation in the process, the system coordinates the flow of knowledge. As a consequence, the overall architecture of Sibyl consists of the Sibyl Server, web browsers (which may be enhanced by plug-ins and Applets in order to provide new media types), and additional services for multimedia collaboration and billing.

In our architecture, access to the conceptual structures is made through the CG Engine, which provides mechanisms for creating, changing, and retrieving conceptual structures. The different databases are filled according to the generic ontologies for domain-specific processes and learning.

The various operational aspects within Sibyl are implemented by groups of cooperating agents. The general structure of the technical architecture of a web-based system for adaptive course curriculum is given in Figure 8. The CG engine is a module that can answer some queries in retrieving data in the KB (using the graph manipulation function of CGs (Sowa, 1984), including the maximal join and various match-and-query algorithms).

In the following, we recapitulate the process of the server, and explain how the data are linked in order to produce a course with respect to a given pedagogical strategy. First, a general description is given in Figure 9. Because the user interface is recognized by the web browser, all operations are done on the server that maintains a current state for each learner. If a user clicks on a specific button, the document will be loaded and the didactic plan will be updated through a series of GC operations and both will finally be sent to the browser.

In a top-down strategy, the URL that refers to an explanation of a topic must precede the URL that refers to an example of the same topic. Therefore, when the system deals with the learning goal that specifies that the student must understand the concept sonata, it selects two CSs from the database: first, an explanation of what a sonata is, and second, a recorded example of a sonata. Due to the top-down strategy, the system builds a constraint that specifies that the explanation precedes the example. This means that the system will not advise the sonata-sound until the sonata.html has been visited. By applying the whole set of rules that correspond to a given strategy, the system has a set of relevant CSs that are linked with special constraints. Thus, it can propose a DP. Figure 9 gives a general example of the sonata

USER INTERFACE

According to the learner model, the system builds a didactic plan and displays it with additional navigation hints. At any time, the learner can visit any node desired, although the system indicates the best way. This leads to a visible indication that the learner is no longer on track.

The general form of this interface is given in Figure 1. The hyperlinks given in the left portion represent the nodes that can be visited. They have several colors to indicate which nodes are advised and which ones have been visited. In future versions, a friendlier mode of presentation has to be found.

To make our system accessible, even for the authors that are not specialists in computer science, we started to integrate a simple authoring tool in our system. The authors capture their course content through defined patterns. These data are translated in conceptual graphs and integrated in the knowledge base of the taught domain. However, we must admit that authoring and didactical annotation are not easy, and we think that special training is necessary to obtain good results.

RELATED WORK

Intelligent Tutoring Systems

The first quality of a system called intelligent is its facility of adaptation (Mizoguchi, Ikeda, & Sinista, 1997). This adaptivity takes the evolution of the world in which the system is and the possible events (new selection of the user(s), results of tests, change of priorities...) into account.

John Eklund (1996) insists on the cooperation between his system (based on WEST's ontology) and the users (learners or teachers). This tool allows several students and/or the teacher, to communicate. Significant attention is given to the state of the student. The learner's level is established by tests, and progression through the system is memorized (the nodes that have been visited).

We then come to the problem of retrieving data in a hypertext database. Cleary (1996) proposed four methods to do this, and a way to evaluate those methods. The system uses typed links in hyper-graphs, and imitates the general discussion with an expert (query/answer's system).

The SHOE project is interesting for us because its authors wanted to create a kind of annotation method for HTML pages (Luke, Spector, Rager, & Hendler, 1997). They say that indexing the html pages with semantic (using ontology) allows for better efficiency in solving problemss and optimizing the retrieval of data.

Goodkovsky (1997) proposes a predefined plan that must be followed in the design of ITS. This plan, which is supposed to be generic, adopts a top-down approach. In Schoening (1997) the course is divided into small parts called Domain Modules (DM). This can remind us of our CSs. He also uses Pedagogic Ontology to support his model. This approach is close to ours.

The necessity of taking into account what the student already knows is discussed by Woods and Warren (1995, 1996). The courses are stored in an electronic book and the retrieval of data is done by key words. This seems to be too fixed and not as efficient as a representation based on semantics.

Murray's (1996) approach of the problem has been very useful for us. His system called EON is an authoring tool that wants to remain as generic as possible for giving course content to several ITSs. In this article, the knowledge representation, the structure and complexity, and the modularity of the course content is discussed. A more general view is given (Murray, 1998) where a very good state of the art is given. The analysis given in this article is close to our ideas. The necessity of having several models, the domain model, the student model, and the learning domain, is explained. For those representations of knowledge, Tom Murray uses the correspondent ontology.

In all those tools, a common drawback appears. The representation of what the system knows is limited. A lot of methods have been tried (matrices, frame language, objects...), but none of them explicitly treat the semantic in the Knowledge Representation (KR). Thus, those systems are only accessible by specialists, and the understanding of the database is quite difficult.

Ontology

To build ontology is a difficult job: One must collect information about a domain, classify it, formalize it, etc. Gruninger and Fox (1995, 1996) propose a tool, Toronto Virtual Enterprise (TOVE), in which models of companies can be represented by ontologies, and then the system can answer some questions. New methodologies can help us in the conception of such tools. The idea of a Pedagogic Ontology (PO), used as a base supporting the dialog between authors and software designers, is promising. In this field, good contributions can be found in Mizoguchi, et al.(1996 and 1997).

Martin (1995) explains how his tool uses the WEST ontology to allow users to build their own systems. Having a well-defined ontology concerning a topic has several advantages. An ontology creates a standard, and thus allows a clear definition of what is really needed, and a comparison between several systems concerning the same domain, such as education. Through speaking the same language, communication is improved. An ontology also allows us to reuse modules between several applications. Reusability is an efficient approach for designing, particularly in LES domains, where tools are mostly built from scratch for the time being. Finally, an available ontology simplifies analysis and synthesis of a particular piece of knowledge.

Other Topics

Adaptivity principles have been proposed by Michel Crampes (1995). The GODART system can make adapted multimedia compositions related to a particular narration, in other words, present the narration with respect to the user preferences and the media chosen by the user. This project is based on conceptual graphs and a multi-agent architecture.

These techniques have been reused for adaptive narrative abstraction in the KARINA project (Crampes, Veuillez, & Ranwez, 1998). A narrative is composed of Narrative Segments (NS). The more relevant are selected to appear in the summary and then the summary is displayed. This composition is done due to users' preferences.

Besides the basic reference to Conceptual Graphs (CG) (Sowa, 1984), we adopted some simplifications proposed by Mugnier and Chein (1996) and paid a particular attention to the work of a French team in Montpellier that has build a general platform for CG: CoGITo (Haemmerle, 1995). In particular, this team has exposed and solved a lot of problems concerning how to give query to the system, how to find the correct answer. (Carbonneill & Haemmerle, 1994) give general notions about CGs and analyze the correct answer that can be given to several queries. It explains in which way CGs are useful in the retrieval of data in a knowledge base: in particular it shows that the same treatment can be applied on data's CGs (graph containing knowledge information) and on nested CGs (more general graphs containing information on structure).

A closer look at a tool called PEIRCE (Ellis & Callaghan, 1997) also helped us develop and improve our CG engine. To understand how it is possible to reason with graphs, we can refer to Sowa's explanations given in (Sowa, 1993).

SUMMARY

In this article, we have presented our proposal for a new and innovative Learning System. Guided by a pedagogical ontology, our system uses conceptual graphs to express semantics for both the domain model and pedagogic model. Therefore, the data retrieval for courses is improved. Moreover, pedagogical strategies can be applied on retrieved data by organizing the final display of courses. Our system is adaptive with respect to strategies and preferences given by the author and the learner's state.

Until now, we have not needed nested graphs, but we assume that they will be useful in some cases, for example during the rule treatment. Yet, we only consider very simple rules such as "IF (condition) THEN (conclusion)". It may also be useful to have rules with the following form: "IF (condition) THEN [IF (condition) THEN (conclusion)]".

We will now focus on two points: to refine our Pedagogical Ontology, especially concerning the Pedagogical Strategies with the help of specialists from the Bielfeld University, and improve the Didactic Plan building (and the navigation support that ensues). As already seen, the DP is composed of URLs that are ordered in a special way due to special strategies. For building this plan and displaying the links related to these URLs, we must find a more expressive form for the learner.

Finally, we think it may be more efficient to have multi-agent architecture for the building and the display of the course. This architecture can look like the one used in the GODART project described by Crampes (1995).

(1.) Laboratoire de Genie Informatique et Ingenierie de Production (LG12P)

EERIE-EMA

Parc Scientifique G. BESSE

30 035 NIMES Cedex 1, FRANCE 

[e-mail], [e-mail]

http://www.eerie.fr/LGI2P/lgi2p.htm

(2.) European Applied Research Center

SAP AG - CEC Karlsruhe

Vincenz Prie[beta]nitz Str. 1

76131 KARLSRUHE Germany

[e-mail]

References

Carbonneill, B. & Haemmerle, O. (1994). Standardizing and interfacing relational databases using conceptual graphs. In W.M. Tepfenhart, J.P. Dick, & J.F. Sowa, Proceedings of the 2nd International conference on conceptual structures: Current practices. College Park, MD, USA, LNAI 835, Springer, p. 311-330.

Cleary, C. (1996). Practical methods for automatically generating typed links. Proceedings of the Seventh ACM Conference on Hypertext. Washington, DC.

Crampes, M. (1995). Composition multimedia dans un contexte narratif Unpublished doctoral dissertation, University Montpellier II.

Crampes, M., Veuillez, J-P., & Ranwez, S. (1998). From Adaptive Narrative Abstraction to Coherent Hypermedia Navigation.

Eklund, J. (1996). Knowledge-based navigation support in hypermedia courseware using WEST[R]. Paper proposed for Australian Educational Computing, 11(2).

Ellis, G. & Callaghan, S. (1997). Peirce user manual. Peirce Holding International Pty. Ltd., Australia. [Online]. Available: http://www.phi-net.com/ DownloadsTable.html

Goodkovsky, V.A. (1997). Intelligent tutor: Top-down approach to ITS design. [Online]. Available: http://www.manta.ieee.org/p1484/goodkov/goodkov.html

Gruninger, M. & Fox, M.S. (1995). Methodology for the design and evaluation of ontologies. Workshop on Basic Ontological Issues in Knowledge Sharing, IJCAI, Montreal.

Gruninger, M. (1995). Designing and evaluating generic ontologies. Proceedings of the Workshop on Ontological Engineering. European Conference on Artificial Intelligence, Budapest, pp. 53-65.

Haemmerle, O. (1995). Implementation of multi-agent systems using conceptual graphs for knowledge and message representation: The CoGITo plat-form. Proceedings of the 3rd International Conference on Conceptual Structures. ICCS'95, Santa Cruz, CA, USA.

Linard, M., & Zeilinger, R. (1995). Designing a navigational support for an educational software. Blumental, Gornostaev, & Unger (Eds.), Lecture Notes in Computer Science N[degrees] 1015, 5th International Conference EWHCI'95, Moscow, Russia, pp 63-78, Springer, Berlin.

Luke, S., Spector L., Rager, D., & Hendler, J. (1997). Ontology-based web agents. Proceedings of First International Conference on Autonomous Agents_AA-97. [Online]. Available: http://www.cs.umd.edu/projects/plus/SHOE/spec.html

Martin, P. (1995). Using the WordNet Concept catalog and a relation hierarchy for knowledge acquisition. Proceedings of Peirce '95, 4th International Workshop on Peirce, University of California, Santa Cruz, CA, USA, pp. 36-47.

Merrill, D.M. (1991). An introduction to instructional transaktion theory, Utah State University Report.

Mizoguchi, R., Sinitsa, K., & Ikeda, M. (1996). Task ontology design for intelligent educational/training systems. Position Paper for ITS'96 Workshop on Architectures and Methods for Designing Cost-Effective and Reusable ITSs, Montreal. [Online]. Available: http://advlearn.lrdc.pitt.edu/its-arch/papers/mizoguchi.html

Mizoguchi, R., Sinitsa, K., & Ikeda, M. (1996b). Knowledge engineering of educational systems for authoring system design (A preliminary results of task ontology design). Presented at EAIED, Lisbon. [Online]. Available: http://www.cbl.leeds.ac.uk/[sim]euroaied/papers/Mizoguchi/

Mizoguchi, R., Ikeda, M., & Sinitsa, K. (1997). Roles of shared ontology in AI-ED research (Intelligence, conceptualization, standardization, and reusability). Proceedings of AI-ED 97, pp. 537-544, Kobe, Japan. [Online]. Available: http://www.ei.sanken.osaka-u.ac.jp/ieee/Them.paper.html

Mugnier, M.L., & Chein, M. (1996). Representer des connaissances et raisonner avec des graphes. Revue d'intelligence artificielle, 10(1), pp 7-56.

Murray, T. (1996). Special purpose ontologies and the representation of pedagogical knowledge. ICLS'96, "talk only" paper submission. [Online]. Available: http://www.cs.umass.edu/[sim]tmurray/papers/ontology/ontology.html

Murray, T. (1998). Authoring knowledge based tutors: Tools for content, instructional strategy, student model, and interface design. Journal of the Learning Sciences, 7(1), 5-64. [Online]. Available: http://www.cs.umass.edu/[sim]tmurray/papers/JLSEon/JLS96.html

Schoening, J.R. (1997). A case and strategy for developing standardized educational domain modules and ontologies. P1484 Working and Study Groups--Developing Technical Standards for Learning Technology. [Online]. Available: http://www.manta.ieee.org/p1484/ontol.htm

Sowa, J.F. (1984). Conceptual structures: Information processing in mind and machine. Addison-Wesley.

Sowa, J.F., Mineau, G.W., & Mineau, B. (Eds.) (1993). Relating diagrams to logic. Lecture notes in AI N 699, Conceptual graphs for knowledge representation. Proceedings ICCS'93 , Quebec City, Canada, Springer-Verlag.

Woods, P.J., & Warren, J.R. (1995). Generating tutoring systems with versatile teaching strategies. Proceedings, Workshop on Authoring Shells for Intelligent Tutoring Systems: AI-ED-95. [Online]. Available: http://www.cis.unisa.edu.au/staff/woods.p.j/aiedwl.html

Woods, P.J., & Warren, J.R. (1996). Adapting teaching strategies in intelligent tutoring systems. ITS '96 Workshop on Architectures and Methods for Designing Cost-effective and Re-useable ITSs. [Online]. Available: http://www.cis.unisa.edu.au/staff/woods.p.j./atsits.html

Classification of concepts

Information

Classification

Concept

Context

Fact

Principle

Procedure

Process

Structure Instruction

Annotation

Conclusion

Counter-Example

Definition

Example

Exercise

Fill-in-the-blank

Matching-exercise

Multiple-choice-test

Explanation

Hypothesis

Introduction

Overview

Problem

Reference

Lexicographic 

Bibliography

Annex 

Repetition

Solution

Summary

Learning Goals

+Who

Learner

Group of students

Author

+Circumstances

+Act

+Accuracy

Domain indep. Capability Tutoring

Domain dep. Capability

Solve a problem

Modify/combine schemas 

Creating a new problem

Media

Image

Movie

Picture

Diagram

Animation

Simulation

Sound

Table

Text

Learners state

Phase in learning process

L_Concepts mastered

Degree of mastery

Navigational history

Node visited

Curriculum

Diplomas

Age

Professional experiences

Mental state

Preferences

System's functionality Communication

Visio-conference

One-to-one discussions Group discussion

Evaluation

Accuracy

Feasibility

Control

Action

Give Exercise

Guide

Explain

Interaction between the system and the learner

Mode of interaction

Menu

Formal test

Hyperlink

Content type Answer

Explanation

Question

Control/Sequencing/Protocol

Learner-driven

System-driven