An Instructional Component for

Dynamic Course Generation and Delivery

Carsten Ullrich

[email protected]

Abstract:
E-Learning offers the advantage of interactivity: an E-Learning system can adapt
the learning materials to suit the learner’s personality and his goals, and it can react to
the students interaction with the learning materials. Since the advent of the Web, lots of
research investigated the new possibilities to support learning offered by the means of
hypertext and the Internet. However, even though important results have been achieved
with respect to adaptive hypermedia, open student modeling, collaborative web-tools
to name but a few, current features of adaptivity fall behind compared to pre-web
instructional planning systems developed in the early nineties.
In this paper, I describe an instructional component for the E-Learning environ-
ment ACTIVE M ATH that marries work in instructional planning, on how to generate
learning materials that are best suited for an individual learner and his tasks, and recent
work for E-Learning in the Web, especially in knowledge representation, of learning
materials (IEEE LOM) and the Semantic Web, i.e., RDF and OWL.

1 Motivation
A good teacher reacts to the needs of his students. He possesses instructional knowledge:
he knows how to adapt his selection of content with respect to his student’s knowledge,
needs, capabilities, and goals. E-Learning has the potential to offer at last some of these
instructional capabilites.
In this article, I describe a component for E-Learning systems that possesses instructional
knowledge similar to a teacher and uses this knowledge to generate and deliver courses of
learning materials. With respect to the scope of this workshop, technology and applications
of XML-technologies, I especially highlight those parts of my work that illustrate how
XML-technologies can come into play in instructional planning. First, I briefly provide
some of the relevant background information on E-Learning.

2 Background
This section contains a succinct description of relevant research on which my work is
based. Due to page constraints, this section distinguishes itself mainly for its fragmentari-
ness.
Instructional Knowledge. Instructional design theories [Re83] describe how to design
teaching materials that are effective (how well is learned), efficient (effectivity divided
by used time), and appealing with respect to the learning goal. Instructional design is
relevant for adaptive E-Learning systems because it provides information about the general
instructional knowledge that systems should possess and about how this knowledge is
reflected in instructional strategies.
Instructional Planning. Numerous approaches exist to formalize instructional knowledge
and to use it for automatic teaching Peachy and McCalla [PM86] realized one of the first
approaches to combine instructional knowledge and artifical intelligence techniques by ap-
plying planning techniques to generate sequence of learning materials. Subsequent work
adds instructional goals (such as Bloom’s taxonomy of educational objectives [Bl56]), and
reactivity in an one-to-one problem solving tutoring situation [Wa90] and a course set-
ting [Va95]. Whereas these approaches mingle instructional and domain knowledge, van
Marcke developed a generic tutoring environment (GTE) [vM98]. His instructional tasks
(activities accomplished by an instructor during an instructional process) and methods
(methods that can be used to perform an instructional task) are to a great extent domain-
independent. However, GTE fails to have an explicit representation of pedagogical control
knowledge and therefore can not represent different instructional strategies. It was never
ported to the Web nor does it use hypertext, and does not provide an explicit representation
of instructional goals.
Adaptive Hypermedia. The last ten years have seen an increasing research making usable
the possibilities offered by the spread of the Internet and hypertext in order to support
learning. Several techniques for adaptive hypermedia have been developed (for an exhaus-
tive overview see [Br01]). Until recently, the adaptive generation of a complete courses
and the content of their pages, fell behind the capabilites of systems like GTE (but see
recent systems such as ACE [SO98] or WINDS [SKPK01]).
Knowledge Representation. Parallel to the research in adaptive hypermedia, considerable
effort has been spend investigating the description of learning materials. The part of this
research concerning the metadata of single learning objects was eventually integrated in
IEEE LOM [IE02]. Descriptions of learning materials collections led to IMS Content
Packaging [IM03a]. Educational modeling languages [IM03b, IM03c] add pedagogical
information to these collections, for instance, the pedagogical role they fulfill in a course.
All these knowledge representations have XML-bindings.
The ACTIVE M ATH learning environment. ACTIVE M ATH [MBA+ 01], a learning environ-
ment developed at the Saarland University and the German Research Center for Artificial
Intelligence (DFKI), integrates instructional planning, adaptive hypermedia, and semantic
knowledge representation. It uses instructional knowledge to generate a course adapted
to the individual learner and his goals from learning objects represented in the XML-
language OMD OC [Kh01]. OMD OC allows the encoding of different categories of items
at paragraph level (mathematical concepts such as definitions and theorems, and further
items such as examples, exercises, or elaborative texts) and their annotation with metadata.
ACTIVE M ATH uses pedagogical rules to dynamically assemble the learning objects to a
course. These rules define what pages for concepts should look like depending on the
knowledge and goals of the learner, e.g., what elements are to appear on a page and in
which order. The learner can explicitly state his learning goals by choosing among differ-
ent scenarios (e.g., overview, guided tour, or Polya-inspired proof presentation).
The current course generation suffers several drawbacks, some of which are interesting
with respect to the scope of this article. An explicit representation of instructional task or
methods is missing, thereby hampering the implementation of additional learning strate-
gies and the integration of third-party content. Furthermore, the complete course is gen-
erated at forehand, and reactivity is limited only. In the following section, I present those
parts of my thesis that alleviate these problems by marrying instructional planning tech-
niques and todays sophisticated knowledge representations and descriptions.

3 An Instructional Component for Dynamic Course Generation and

Delivery
A goal of my thesis is to realize a generic, domain independent instructional component
that generates courses of learning materials and supports the student while interacting
with the course. As already mentioned, I restrict myself to those parts that employ XML-
technologies. Other aspects of my work are described briefly, if necessary.
I focus on the following questions: How can generic instructional planning be realized?
How can the generated course be made usable for third parties? How can the student be
supported while interacting with the course? How can an uncomplicated integration of
third-party content be achieved?

3.1 Instructional Planning

The goal of an instructional session is that the student achieves his learning goals. The
instructional planner selects which learning objects to present and in which order.
The course generation happens the following way: The learner (or a teacher) specifies
his learning goal, such as teachApplication(addition). The instructional planner treats
this goal as an instructional task. Other examples of instructional tasks are summarize,
provideMotivation. In general, an instructional task can be fulfilled by several instruc-
tional methods by decomposing the task in sub-tasks. For instance, the task to arouse the
students interest can be fulfilled by the method EasyExamplePresentation or RealWorld-
ProblemPresentation. Instructional methods decompose tasks into subtasks until primi-
tive, non-decomposable tasks are reached. These primitive tasks correspond to actions
that act on the course itself, for instance they choose that exercise number 15 is shown as
the second element on page 3.
The choice of methods depends on the instructional strategy. A strategy comprises a selec-
tion of tasks and methods, and a rating. For instance, a learning-by-doing strategy prefers
methods that foster active problem solving.
Figure 1 contains an instructional plan that teaches about the mathematical concept group.
The task teachTopic(GROUP) is fulfilled using the instructional method ProblemBased-
Training. Different choices would have been possible, for instance ClassicDidacticAp-
proach. In this case however, the planner opted for a problem-based learning strategy
because Eva, the current user, learns best using concrete problems (a fact represented in
her user model). The planner decides to present the necessary prerequisites, together with
a test (not shown in the figure), followed by the definition of group (the learning object
with the id group def ) and the problem (group problem1).

Task
teachTopic(GROUP)

Method ProblemBasedTraining

teachPrereq(GROUP) provideProblem(GROUP)

PrereqTraining SupportedProblemProvision

showMissingPrereq(GROUP)
showConcept(GROUP)
showProblem(GROUP)
... remindKnowPrereq(GROUP)
... ElementDisplay(group_problem1)
ElementDisplay(group_def)

Figure 1: Instructional Plan

3.2 Course Data Structure

A generated course needs to be delivered to the learner. A course does not solely consist of
a structured sequence of learning materials, but ideally it contains additional information,
e.g., on the instructional goal of a step, on time constraints, on reactions when time con-
straints are violated or goals are met. Furthermore, the generated course should be usable
in different learning environments, so that content can not only be imported as seen in the
previous section, but also exported.
A recent standard meets these requirements. Using IMS Simple Sequencing [IM03c], an
author specifies the structure (e.g., sections and subsections) of a collection of learning
materials and additionally provides information on how to guide the learner through this
structure. Conditional rules can make, for instance, a learner skip a section if her knowl-
edge exceeds a given threshold.
Therefore, the result of the instructional planning will be an IMS SS structure generated
from the instructional plan. Figure 2 provides the IMS SS structure from the plan of
Figure 1. The first page to be presented contains the prerequisites, followed by an exercise
that tests whether Eva indeed possesses the necessary knowledge, and finally the new
concept and the problem.
However, using IMS SS leads to several problems. First, the complete course needs to
be generated at forehand but some assumptions underlying the planning can turn out to
be wrong. Second, reactions as specified in IMS SS depend solely on interactions of the
learner with the content. Other properties of the user, such as his field of study, are not
taken into account. Third, interactivity in IMS SS is limited. As it relies on a fixed IMS
CP, it is not suited for dynamically adding or deleting content.
Whereas IMS SS seems an appropriate data structure for some dynamic aspects, we need
a component that allows for real interactivity and reactivity that overcomes the limits of
programmed instruction,
 Course1
Sequencing Information
Content Packaging

Page1 Page2 Page3

structure_def associative_def inverse_def exc1 group_def group_problem1

Figure 2: Simple Sequencing Datastructure

3.3 Delivery and Monitoring

The role of the delivery and monitoring component is to watch the actions of the user and
to react, if necessary. Possible reactions consist of re-planning parts of the course and
adding or removing parts of the plan. Reactions can occur, if assumptions underlying the
plan generation turn out to be wrong (e.g., knowledge about concepts, motivational as-
pects), or, if the learner explicitly utters requests such as inserting some content or asking
for the next best learning step. More specifically, by having explicitly represent an instruc-
tional vocabulary, there is no reason why the learner himself should not have access to this
vocabulary and sends his own instructional goals to the planner.
With respect to the above example, if Eva fails to show her knowledge about the prerequi-
sites, the monitoring component reacts by re-planning and inserting new, remedying nodes.
To a limited extent, this information can be represented in IMS SS. However, this would
require to plan ahead every possible action of the student and every sensible reaction of
the system. This is not feasible in a tutorial setting as envisaged in this thesis.

3.4 Integrating Content

Writing learning materials is expensive and therefore the reuse of existing learning mate-
rials is an important goal. Separating structure from content, as possible with XML was
a first, important step towards reusability. However, today a myriad of knowledge rep-
resentations for learning materials exist. IEEE LOM [IE02] provides at least a common
representation of metadata, but is insufficient for automatic usage of learning materials as
envisaged in this thesis. So the challenge remains how to integrate content not written in a
specific knowledge representation and not constrain unnecessarily content sources.
The idea is to define a general, abstract ontology of instructional objects and to provide
the possibility of defining mappings from third-party content to these objects.
Luckily, the Semantic Web faces a similar challenge. Lots of data exists in the Web, so
how can this data be described, how can it be used for reasoning? Research done on
semantic nets and reasoning led to the formulation of the resource description framework
RDF [Wo99] and more recently to the Web ontology language OWL [Wo02].
In the context of this paper, RDF is used to assign a semantic to a learning object. While the
XML-binding of LOM prescribes the structure and only unofficially defines the semantics
of the elements, RDF makes the semantics explicit. Furthermore, it allows to extend the
metadata and to make the extensions accessible.
RDF descriptions refer to a defined vocabulary, in this case an instructional vocabulary.
The vocabulary is described in OWL. More specifically, OWL is used for defining the
instructional objects (definition, example, etc.) themselves, and the relations between these
objects. Thus, it defines an generic ontology of learning objects. By using the OWL
ontology mapping facilities, third-party content providers can use the representation to
define a mapping from their vocabulary to the generic instructional objects. This provides
a mechanism for planning with third-party content.
Figure 3 provides an example of the instructional vocabulary used by the planner (middle
part) and mappings from two different content knowledge representations (OMD OC and
a fictitious physic knowledge representation). By using these mappings, the instructional
planner can easily apply its instructional knowledge to a different domain.

OMDoc Instructional Vocabulary Physics

OM:Def Item PhysicConcept

owl:equivalentClass

rdfs:subClassOf

OM:Exc Concept ... Exercise Experiment

OM:Exc OM:Exc
Definition Theorem Problem Application MCQ
V2.1

Type=Pro Type=Appl

 





Figure 3: Different ontologies with mappings

4 Conclusion and Further Work

My work tackles the problems of generating courses of learning materials from different
sources adapted to an individual learner and his tasks and delivering the course in a re-
active manner. XML-technologies (RDF and OWL) will be applied for representing the
instructional knowledge and to allow inclusion of third-party content. Existing learning
standards with an XML-binding will be used for the representation of the course itself. To
which degree XML can further serve as a means remains to be investigated. It may turn
out to be supportive when offering the instructional planner as a Web service, so that other
learning environments have a standardized way of access.
To my knowledge, no other system has made explicit the instructional knowledge to such
an extend as described in this work. Several systems offer some degree of adaptivity, some
allow to adapt the rules that govern the adaptivity. However, no system I know of explicitly
defines an instructional vocabulary as done in GTE and none offers the flexibility for an
instructor to define several strategies by referring to such a vocabulary.
Most work still needs to be done. First steps will be to investigate the instructional vo-
cabulary of objects, tasks, and methods. Evaluations need to be undertaken to determine
whether and to which degree adaptivity indeed supports learning.

References
[Bl56] Bloom, B. (Hrsg.): Taxonomy of educational objectives: The classification of educa-
tional goals: Handbook I, cognitive domain. Longmans, Green. New York, Toronto.
1956.
[Br01] Brusilovsky, P.: Adaptive hypermedia. User Modeling and User Adapted Interaction.
11(1/2):87–110. 2001.
[IE02] IEEE. 1484.12.1-2002 IEEE standard for learning object metadata. 2002.
[IM03a] IMS Global Learning Consortium. IMS content packaging information model. June
2003.
[IM03b] IMS Global Learning Consortium. IMS learning design specification. February 2003.
[IM03c] IMS Global Learning Consortium. IMS simple sequencing specification. March 2003.
[Kh01] Kohlhase, M.: OMD OC: Towards an internet standard for mathematical knowledge.
In: Lozano, E. R. (Hrsg.), Proceedings of AISC’2000. LNAI. Springer Verlag. 2001.
[MBA+ 01] Melis, E., Büdenbender, J., Andres, E., Frischauf, A., Goguadze, G., Libbrecht, P.,
Pollet, M., und Ullrich, C.: Activemath: A generic and adaptive web-based learning
environment. Artificial Intelligence and Education. 12(4). 2001.
[PM86] Peachy, D. R. und McCalla, G. I.: Using planning techniques in intelligent tutoring
systems. International Journal of Man-Machine Studies. 24:77–98. 1986.
[Re83] Reigeluth, C. M. (Hrsg.): Instructional Design Theories and Models: An Overview on
their Current Status. volume I. Lawrence Erlbaum Associates. Hillsdale, NJ. 1983.
[SKPK01] Specht, M., Kravcik, M., Pesin, L., und Klemke, R.: Authoring adaptive educational
hypermedia in WINDS. In: Henze, N. (Hrsg.), Proc. of the ABIS 2001 Workshop. 2001.
[SO98] Specht, M. und Oppermann, R.: ACE - adaptive courseware environment. The New
Review of Hypermedia and Multimedia. 4:141–162. 1998.
[Va95] Vassileva, J.: Dynamic courseware generation: at the cross point of cal, its and author-
ing. In: Proceedings of ICCE’95, Singapore. S. 290–297. December 1995.
[vM98] van Marcke, K.: GTE: An epistemological approach to instructional modeling. In-
structional Science. 26:147–191. 1998.
[Wa90] Wasson, B. J.: Determining the Focus of Instruction: Content planning for intelligent
tutoring systems. PhD thesis. Department of Computational Science, University of
Saskatchewan. 1990. Research Report 90-5.
[Wo99] World Wide Web Consortium. Resource description framework (RDF) model and syn-
tax specification. February 1999.
[Wo02] World Wide Web Consortium. Web ontology language (OWL) abstract syntax and
semantics. November 2002.