SKATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT INGENIEURSWETENSCHAPPEN
DEPARTEMENT COMPUTERWETENSCHAPPEN
AFDELING INFORMATICA
Celestijnenlaan 200A — B-3001 Leuven
AN ARCHITECTURE AND FRAMEWORK FOR FLEXIBLE REUSE
OF LEARNING OBJECT COMPONENTS
Promotoren :
Prof. Dr. ir. E. DUVAL
Prof. Dr. H. OLIVIE´
Proefschrift voorgedragen tot
het behalen van het doctoraat
in de ingenieurswetenschappen
door
Katrien VERBERT
February 2008

SKATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT INGENIEURSWETENSCHAPPEN
DEPARTEMENT COMPUTERWETENSCHAPPEN
AFDELING INFORMATICA
Celestijnenlaan 200A — B-3001 Leuven
AN ARCHITECTURE AND FRAMEWORK FOR FLEXIBLE REUSE
OF LEARNING OBJECT COMPONENTS
Jury :
Prof. Dr. ir. D. Vandermeulen, voorzitter
Prof. Dr. ir. E. Duval, promotor
Prof. Dr. H. Olivie´, promotor
Prof. Dr. ir. Y. Berbers
Prof. Dr. ir. K. De Vlaminck
Prof. Dr. M.-F. Moens
Prof. Dr. T. Boyle (London Metropolitan University, UK)
Prof. Dr. D. Wiley (Utah State University, USA)
Proefschrift voorgedragen tot
het behalen van het doctoraat
in de ingenieurswetenschappen
door
Katrien VERBERT
U.D.C. 681.3∗H1, 681.3∗H3
February 2008

c©Katholieke Universiteit Leuven – Faculteit Ingenieurswetenschappen
Arenbergkasteel, B-3001 Heverlee (Belgium)
Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenigvuldigd
en/of openbaar gemaakt worden door middel van druk, fotocopie, microfilm,
elektronisch of op welke andere wijze ook zonder voorafgaande schriftelijke toe-
stemming van de uitgever.
All rights reserved. No part of the publication may be reproduced in any form
by print, photoprint, microfilm or any other means without written permission
from the publisher.
D/2008/7515/14
ISBN 978-90-5682-905-6

Preface
Barriers and enablers for the reusability of learning objects are important research
topics in the learning technology community. In various publications, it is argued
that reuse not only saves time and money, but also enhances the quality of digital
learning experiences, resulting in ecient, economic and eective learning.
It is commonly accepted that there is an inverse relationship between the size
of a learning object and its reusability. Fine-grained learning objects or learning
object components have the potential to be
exibly assembled into new learning
objects, whereas entire courses are often not suitable for use in a dierent context.
Many shared learning objects are coarse-grained and therefore dicult to reuse.
Typically, authors reuse parts of a learning object by copy-and-paste actions. This
method of reuse is possible in any authoring tool, but is limited in several ways:
the approach is non-scalable in terms of maintenance, tends to be error-prone, and
due to its inherent monotony, easily becomes both bothering and time consuming.
To support learning object reuse in a more methodological way, a component-
based reuse approach is investigated in this dissertation. A number of interrelated
fundamental research issues are investigated: (1) a learning object content model,
that identies dierent kinds of learning object components at dierent levels of
granularity; (2) a component architecture that enables structuring of composite
learning objects; and (3) the processes of aggregation and disassembly, to pro-
duce composite learning objects and to isolate their components, so as to enable
automatic reuse of learning objects that were originally produced as aggregates.
Interoperability aspects are strongly emphasized throughout this work, as the
interoperation of learning objects is critical in the creation of a global component
architecture for learning objects. The ultimate goal is a learning object economy
characterized by searchable repositories of reusable learning objects that can be
exchanged and reused across various learning systems.
The dissertation is organized as follows: Chapter 1 introduces the learning
object domain and presents challenges and issues impeding learning object reuse
on a global scale.
Chapter 2 presents the generic ALOCOM content model that denes learning
object granularity in a precise way. A number of learning object content models
i

ii
have been reviewed that dene learning objects and their components. Based
on a comparative analysis, the content models have been mapped to the generic
ALOCOM model to enable their interoperability.
Chapter 3 presents the RAMLET reference model for structuring of learning
objects. The reference model enables interoperability between dierent content
packaging specications that dene the structure of a collection of learning content.
A common nomenclature and conceptual model have been dened and crosswalks
among various content packaging specications that enable their interoperation.
Chapter 4 presents the ALOCOM decomposition and aggregation framework
for learning objects. The framework automates learning object reuse by enabling
on-the-
y access to learning object components contained in composite learning
objects. Prototypes of tools have been developed to validate the approach. Plug-
ins for Microsoft PowerPoint, Microsoft Word and the Reload Editor integrate
learning object reuse in the work
ow of authors.
Chapter 5 presents user and quality evaluations that validate the approach.
The goals of the evaluations were threefold: (i) to assess the eciency and ef-
fectiveness of the approach for reusing learning objects; (ii) to assess the sub-
jective acceptance of the ALOCOM plug-ins; (iii) to determine to which level of
granularity decomposition is relevant.
Finally, Chapter 6 concludes this dissertation with a summary of contributions,
a discussion on the possible impact of the research, and an exploration of the
potential it oers for future research.

Acknowledgement
The work on this dissertation has been extensive and sometimes challenging, but
in the rst place exciting, enlightening, and fun. Without help, support, and
encouragement from several persons, I would never have been able to nish this
work. It is now my great pleasure to take this opportunity to thank them.
First and foremost, I would like to thank my supervisor Prof. Dr. ir. Erik
Duval for his guidance and encouragement. His valuable input and constructive
feedback from the initial conception to the end of this work are highly appreciated.
I would like to thank my second supervisor Prof. Dr. Henk Olivie for his
support and thoughtful comments.
My sincere thanks go out to Prof. Dr. David Wiley, Prof. Dr. Tom Boyle,
Prof. Dr. Marie-Francine Moens, Prof. Dr. ir. Yolande Berbers and Prof. Dr. ir.
Karel De Vlaminck for their time and eort in evaluating this dissertation. I am
also grateful to Prof. Dr. ir. Dirk Vandermeulen for chairing the committee.
I would like to thank Jelena Jovanovic and Xavier Ochoa for a fruitful collab-
oration and as co-authors. I am grateful to the members of the RAMLET group
for the productive collaboration.
I would like to thank my former and current colleagues Joris Klerkx, Michael
Meire, Bram Vandeputte, Dr. Martin Wolpers, Stefaan Ternier, Xavier Ochoa,
Gonzalo Parra, Bram Luyten, Jehad Najjar, Riina Vuorikari, Nik Corthaut, Sten
Govaerts and Ben Bosman for creating a stimulating research environment.
I am grateful to Joris Klerkx and Bram Vandeputte for carefully reading parts
of my dissertation text and for providing many useful comments.
I am grateful to the many people who participated in the evaluations for their
eorts and their critical feedback.
I gratefully acknowledge the nancial support of the K.U.Leuven research coun-
cil through the BALO project and the Interdisciplinary Institute for Broadband
Technology (IBBT) through the Acknowledge project.
Finally, I would like to express my deepest gratitude to my husband Bert, my
brother Karel, my parents and many other relatives and friends for their continuous
encouragement and support.
iii

Contents
Contents v
List of Acronyms ix
List of Figures xi
List of Tables xiii
1 Introduction 1
1.1 Learning Object Denitions . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Learning Object Metadata . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Learning Object Repositories . . . . . . . . . . . . . . . . . . . . . 4
1.4 Standardization Eorts . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.1 IEEE LTSC . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.2 IMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 ADL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.4 ARIADNE . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Issues and challenges . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.1 Learning Object Granularity . . . . . . . . . . . . . . . . . 10
1.5.2 Learning Object Structure . . . . . . . . . . . . . . . . . . . 11
1.5.3 Learning Object Aggregation and Disassembly . . . . . . . 12
1.5.4 Learning Object Interoperability . . . . . . . . . . . . . . . 13
1.6 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 ALOCOM: a Generic Content Model for Learning Objects 15
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Structured Writing . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 IEEE LOM . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3 Ballstaedt . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4 The Component Display Theory . . . . . . . . . . . . . . . 19
v

vi CONTENTS
2.3 Overview of Learning Object Content Models . . . . . . . . . . . . 20
2.3.1 NETg Learning Object Model . . . . . . . . . . . . . . . . . 20
2.3.2 Learnativity Content Model . . . . . . . . . . . . . . . . . . 22
2.3.3 SCORM Content Model . . . . . . . . . . . . . . . . . . . . 23
2.3.4 Navy Content Model (NCOM) . . . . . . . . . . . . . . . . 25
2.3.5 Cisco RLO/RIO Model . . . . . . . . . . . . . . . . . . . . 26
2.3.6 dLCMS Component Model . . . . . . . . . . . . . . . . . . 28
2.3.7 New Economy Didactical Model . . . . . . . . . . . . . . . 30
2.3.8 Semantic Learning Model (SLM) . . . . . . . . . . . . . . . 32
2.3.9 PaKMaS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Comparative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Ontology-based Approach for Content Model Interoperability . . . 37
2.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.2 First Stage: Building the Global Ontology . . . . . . . . . . 37
2.5.3 Second Stage: Building Local Ontologies . . . . . . . . . . . 38
2.5.4 Third Stage: Dening mappings . . . . . . . . . . . . . . . 38
2.6 Abstract Learning Object Content Model (ALOCOM) . . . . . . . 38
2.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.2 Granularity Levels . . . . . . . . . . . . . . . . . . . . . . . 39
2.6.3 Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.6.4 Content Classications . . . . . . . . . . . . . . . . . . . . . 40
2.6.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.7 Local Ontologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7.2 The Cisco Ontology . . . . . . . . . . . . . . . . . . . . . . 44
2.8 Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.8.2 Learning Object Component Mappings . . . . . . . . . . . 46
2.8.3 Learning Object Mappings . . . . . . . . . . . . . . . . . . 46
2.8.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.9 Usage Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.10 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.11 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3 RAMLET: A Model for Structuring of Learning Object Compo-
nents 53
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.2 Use Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.3 Use case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.4 Use case 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Resource Aggregates . . . . . . . . . . . . . . . . . . . . . . . . . . 59

List of Figures
1.1 Overview (based on [Downes, 2004]) . . . . . . . . . . . . . . . . . 3
1.2 The SCORM parts . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 A NETg course structure [Tan, 2002] . . . . . . . . . . . . . . . . . 21
2.2 UML representation of the NETg Learning Object Model . . . . . 21
2.3 UML representation of the Learnativity Content Model . . . . . . 23
2.4 UML representation of the SCORM Content Model . . . . . . . . . 24
2.5 UML representation of the Navy Content Model . . . . . . . . . . 25
2.6 The RLO and RIO structure [Barrit et al., 1999] . . . . . . . . . . 26
2.7 UML representation of the dLCMS Component Model . . . . . . . 28
2.8 UML representation of the New Economy Didactical Model . . . . 31
2.9 UML representation of the Semantic Learning Model . . . . . . . . 34
2.10 UML representation of the PaKMaS Model . . . . . . . . . . . . . 35
2.11 Ontology construction method [Bucella et al., 2003] . . . . . . . . . 38
2.12 The ALOCOM Aggregation Levels . . . . . . . . . . . . . . . . . . 40
2.13 The ALOCOM Model . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.14 UML representation of the Cisco RLO-RIO Model . . . . . . . . . 45
2.15 Learning Object Component Mappings . . . . . . . . . . . . . . . . 47
2.16 Learning Object Mappings . . . . . . . . . . . . . . . . . . . . . . . 49
2.17 Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.1 Use case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Use case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3 Use case 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 Conceptual description of elements in a manifest document . . . . 60
3.5 Conceptual description of elements in a METS document . . . . . 62
3.6 Conceptual description of elements in an MPEG-21 DID document 64
4.1 The ALOCOM Framework . . . . . . . . . . . . . . . . . . . . . . 81
4.2 The Decomposition Process . . . . . . . . . . . . . . . . . . . . . . 83
4.3 The ALOCOM plug-in for MS Powerpoint . . . . . . . . . . . . . . 90
xi

2 Introduction
learning object paradigm. Before going into the details of challenges and issues
impeding the approach, the learning object domain is brie
y introduced.
The chapter is organized as follows: Section 1.1 outlines learning object de-
nitions. The concept of learning object metadata and learning object repositories
is presented in Section 1.2 and Section 1.3. Important learning object standards
are outlined in Section 1.4. Section 1.5 summarizes challenges and issues that
are tackled in this dissertation. Finally, Section 1.6 provides an overview of the
subsequent chapters.
1.1 Learning Object Denitions
There are many denitions of learning objects. One of the rst denitions, dened
by the IEEE Learning Technology Standards Committee [IEEE, 2002], states that
a learning object is "any entity, digital or non-digital, which can be used, re-used
or referenced during technology supported learning".
David Wiley argues that this denition is too broad, because it "fails to exclude
any person, place, thing, or idea that has existed at anytime in the history of the
universe". He suggests a more rened denition as "any digital resource that can
be reused to support learning" [Wiley, 2002].
Other denitions focus on the components of the learning object: a learning
objective, a unit of instruction, and a unit of assessment [L'Allier, 2003]. The
Wisconsin Online Resource Center uses a time element in its denition and denes
learning objects as smaller units of learning, typically ranging from 2 minutes to
15 minutes [Chitwood et al., 2000].
Although there is no generally accepted denition, there is common consensus
that a learning object should be [Rehak and Mason, 2003]:
 Reusable - can be modied and versioned for dierent courses;
 Accessible - can be indexed and retrieved using metadata;
 Interoperable / portable - can operate across dierent hard/software;
 Durable - remains intact across upgrades of hard/software.
These characteristics are often referred to as the RAID principle. Similar charac-
teristics are dened by Downes [Downes, 2004], who argues that learning objects
are, or ought to be:
 sharable: may be produced centrally, but can be used in many dierent
courses;
 digital: can be distributed using the Internet;
 modular: capable of being combined with other resources;

4 Introduction
jects, and hence to facilitate their reusability.
The IEEE Learning Object Metadata (LOM) standard [IEEE, 2002] is the
primary standard for the description of learning objects [Wiley, 2007]. Relevant
attributes of learning objects to be described include: type of object, author,
owner, terms of distribution, format, and pedagogical attributes, such as teaching
or interaction style. The elements are organized into nine categories:
1. General: description of the learning object as a whole;
2. Lifecycle: the history and current state of the learning object;
3. Meta-Metadata: information about the metadata instance;
4. Technical: technical requirements and characteristics;
5. Educational: educational and pedagogical characteristics;
6. Rights: intellectual property rights and conditions of use;
7. Relation: the relationship with other learning objects;
8. Annotation: comments on the educational use of the learning object;
9. Classication: relation to a particular classication system.
Some e-learning initiatives use the Dublin Core metadata element set (DCMES)
[Weibel et al., 1998] for the description of learning objects. DCMES is an ISO
standard for metadata, intended for cross-domain resource description. The meta-
data standard includes two levels: Simple and Qualied. Simple Dublin Core
comprises fteen elements: title, creator, subject, description, publisher, contribu-
tor, date, type, format, identier, source, language, relation, coverage and rights.
Qualied Dublin Core includes three additional elements (audience, provenance
and rightsholder), and a group of element renements that make the meaning of
an element narrower or more specic.
The education working group [Weibel and Koch, 2000] of the Dublin Core
Metadata Initiative is developing education specic elements, element qualiers
and controlled vocabularies to be used with DCMES for describing educational
materials. Among others, the DC-Education proposal recommends the use of
three elements from the LOM metadata standard: Interactivity type, Interactivity
level and Typical learning time.
1.3 Learning Object Repositories
Learning objects are stored in databases called learning object repositories. There
are two major types of such repositories: those containing both learning objects

1.3 Learning Object Repositories 5
and learning object metadata, and those containing metadata only. In the latter
case, the learning objects themselves are located at an external location and the
repository is used as a tool to locate learning objects. These repositories are often
called referatories. In the former, the repository may be used to both locate and
deliver the learning object [Downes, 2004].
Two major models for learning object repositories exist. The most common
form is a centralized form in which learning object metadata are located on a single
server. An alternative model is distributed, in which learning object metadata are
contained in a number of connected servers.
The following are examples of learning object repositories:
 Multimedia Educational Resource for Learning and Online Teaching: Mer-
lot [Cafolla, 2002] is a centralized repository containing metadata only and
pointing to objects located at external locations. Merlot uses its own meta-
data format for the description of learning objects.
 Campus Alberta Repository of Educational Objects: CAREO [Friesen and
McGreal, 2002] is a centralized collection of learning objects intended for
educators in Alberta, Canada. CAREO contains metadata and provides
access to learning objects located on external web servers. CAREO uses the
LOM standard for the description of learning objects.
 Education Network Australia Online: EdNA Online [Adcock et al., 2000] is
an Australian centralized referatory. EdNA Online uses a metadata format
that is based on DCMES.
 ARIADNE Knowledge Pool System: The ARIADNE Knowledge Pool Sys-
tem (KPS) is a distributed repository of learning objects and associated
metadata [Duval et al., 2001]. ARIADNE uses the LOM standard for the
description of learning objects.
 Edutella: Edutella [Nejdl et al., 2002] is a distributed peer-to-peer repository,
containing metadata only. The referatory relies on an RDF binding of LOM.
Current research on learning object repositories focuses, amongst others, on
learning object discovery [Ochoa and Duval, 2006] [Orzechowski et al., 2007], in-
teroperability between repositories [Hatala et al., 2004] [Prause et al., 2007], and
long-term preservation [Lorie, 2001]. Research on long-term preservation is con-
cerned with archival of both content and programs that read the content, such
that they will still be readable somewhere in the future. Interoperability research
focuses on connecting and using learning objects located in heterogeneous and un-
aligned repositories. For instance, through the Simple Query Interface [Ternier and
Duval, 2006], a protocol for searching repositories, the ARIADNE KPS, Merlot
and EdNA Online are currently interconnected. Finally, research on discovery of
learning objects is concerned with user proling for more accurate learning object

6 Introduction
discovery [Orzechowski et al., 2007], enhanced search mechanisms [Zimmermann
et al., 2007], ranking and recommendation of learning objects [Ochoa and Du-
val, 2006] and information visualization techniques to enable
exible and ecient
access to learning object repositories [Klerkx et al., 2004].
1.4 Standardization Eorts
This section provides an overview of major organizations that contribute to the
development of e-learning standards: IEEE LTSC, IMS, ADL, and ARIADNE.
Important standardization eorts with respect to learning object reusability and
interoperability are brie
y described.
1.4.1 IEEE LTSC
Since 1996, the IEEE Learning Technology Standards Committee (IEEE LTSC)
[IEEE, 2002] develops internationally accredited technical standards, recommended
practices, and guides for learning technology. The LOM standard is the most
widely acknowledged IEEE LTSC specication. IMS, ADL and ARIADNE use
LOM elements and structures in their specications.
In addition to LOM, the LTSC is in the process of developing standards for a
variety of learning technology aspects:
 Digital Rights Expression Languages (DREL): A Digital Rights Expression
Language is a way of expressing and managing conditions and permissions
of learning objects. By standardizing such information, the rights assigned
by an author or publisher to a learning object may be preserved across a
variety of systems where the object may be used.
 Computer Managed Instruction (CMI): The Computer Managed Instruction
working group is developing a multi-part standard that covers, amongst oth-
ers: describing course content, organizing and sequencing individual lessons
in a single course, course management software, and communication between
CMI software and lessons.
 Reusable Competency Denition (RCD): This standard denes a data model
for describing, referencing, and sharing competency denitions, primarily in
the context of online and distributed learning.
Of interest in this dissertation is the CMI work, as it includes the development
of standards that will enable lessons, that are developed with dierent tools by
dierent people, to be brought together and used in a single course.

1.4 Standardization Eorts 7
1.4.2 IMS
In 1997, the National Learning Infrastructure Initiative of Educause [Oblinger,
2005] began a project to create a set of widely adopted standards for exchanging
college learning content. The specications published to date and ongoing projects
address requirements in a wider range of learning contexts, including K-12 schools
and corporate and government training. The acronym IMS originally stood for
Instructional Management Systems, but the full term is now rarely used.
The mission of IMS is to support the adoption and use of learning technology
worldwide by the development of open technical specications for interoperable
learning technology. These specications include:
 IMS Content Packaging: The IMS Content Packaging Specication provides
the functionality to describe and package learning objects, such as an in-
dividual course or a collection of courses, into interoperable, distributable
packages. Content Packaging addresses the description, structure, and loca-
tion of online learning objects and the denition of some particular content
types.
 IMS Digital Repository: The purpose of the IMS Digital Repositories spec-
ication is to provide recommendations for the interoperation of the most
common repository functions.
 IMS Learning Design: The IMS Learning Design specication supports the
use of a wide range of pedagogies in online learning, by providing a generic
and
exible language for expressing such pedagogies.
 IMS Question & Test Interoperability (QTI): The IMS QTI specication
describes a data model for the representation of question and test data and
corresponding results reports, enabling their exchange between authoring
tools, learning systems and assessment delivery systems.
 IMS Simple Sequencing: The IMS Simple Sequencing specication denes
a method for representing the intended behavior of an authored learning
experience. The specication incorporates rules that describe the branching
or
ow of instruction through content, according to the outcomes of learner
interactions with content.
The IMS Content Packaging specication is important in the context of this
dissertation, as the specication enables to describe the structure of a collection of
learning objects or learning object components into a coherent, structured, whole.
The IMS Simple Sequencing specication can be used for describing the intended
behavior of such a collection.

8 Introduction
1.4.3 ADL
The Advanced Distributed Learning (ADL) initiative [Looms and Christensen,
2002] was established in 1997 to develop a Department of Defense (DoD) strat-
egy for using learning and information technologies to modernize education and
training and to promote cooperation between government, industry and academia.
The most widely accepted ADL publication is the Sharable Content Object
Reference Model (SCORM) [SCORM, 2004]. The SCORM specication combines
elements of IEEE LTSC and IMS specications. The SCORM standard is com-
prised of four major elements (see Figure 1.2):
 Part 1 provides an overview, containing high-level conceptual information,
the history, current status and future directions of ADL and SCORM.
 Part 2, the SCORM Content Aggregation Model, describes content compo-
nents used in a learning object, how to package those components for ex-
change from system to system, how to describe those components to enable
search and discovery, and how to dene sequencing rules for the components.
 Part 3 outlines how to sequence and navigate learning objects. It describes
how SCORM-conformant content may be sequenced to the learner through
a set of learner-initiated or system-initiated navigation events.
 Part 4 covers the SCORM run-time environment. The purpose of the run-
time environment is to provide a means for interoperability between learning
content and Learning Management Systems (LMSs). It describes the LMS
requirements in managing the run-time environment, such as the content
launch process, standardized communication between content and LMSs,
and standardized data model elements used for passing information relevant
to the learner experience with the content.
Of particular interest is the SCORM Content Aggregation Model, as it addresses
the creation, discovery and aggregation of reusable learning objects. The SCORM
CAM integrates:
 A Content Model, that denes content components at dierent levels of
granularity and how these components are aggregated into higher level units
of instruction. The model is described in detail in the next chapter;
 Content packaging, using the IMS Content Packaging specication;
 Metadata, using IEEE LOM; and
 Sequencing and Navigation, using the IMS Simple Sequencing specication.

10 Introduction
 Content models, that dene learning object components and how they can
be aggregated, so as to enable reuse and repurposing;
 Social information retrieval techniques for
exible access to large-scale col-
lections of content.
The content models research is presented in this dissertation. Automatic gen-
eration of metadata is required to realize reuse of learning object components, as
components have to be described to enable their retrieval. The automatic meta-
data generation framework is described in Chapter 4. Attention metadata enables
building user attention proles that represent actual interests of users based on
content they worked with. The use of such proles enables a personalized ranking
mechanism for nding learning object components.
1.5 Issues and challenges
Duval & Hodgins [Duval and Hodgins, 2003] have listed a number of interrelated
research issues that are important for enabling learning object reuse on a global
scale. This section outlines the issues and challenges that have been addressed in
this dissertation.
1.5.1 Learning Object Granularity
There is an inverse relationship between the size of a learning object and its
reusability [Wiley, 2002]. As the size of the learning object decreases, its po-
tential for reuse increases. Indeed, ne-grained learning objects or learning object
components, such as images, denitions or exercises, have the potential to be as-
sembled into new learning objects, whereas entire courses are often not suitable
for use in a dierent learning context. Size is thus an important factor for enabling
successful learning object reuse.
There is no agreement in the literature on how to dene the size of learning
objects. The LOM standard [IEEE, 2002] identies four dierent levels of learning
object aggregation or "functional granularity", from the nest grained, such as a
single image or other digital asset, to the largest level of a complete certicated
course:
1. The smallest level of aggregation, e.g. raw media data or fragments.
2. A collection of level 1 learning objects, e.g. a lesson.
3. A collection of level 2 learning objects, e.g. a course.
4. The largest level of granularity, e.g. a set of courses that lead to a certicate.

12 Introduction
The Resource Aggregation Model for Learning, Education and Training (RAM-
LET) has been developed to describe the structure of learning objects in a uniform
way. Interoperability is achieved by the denition of crosswalks among various con-
tent packaging specications. The model has been developed by the IEEE LTSC
CMI working group, in which the author has been involved.
The RAMLET model enables to assemble and structure ALOCOM components
into coherent learning objects. In addition, interoperability is achieved with many
content packaging specications, that facilitates sharing and reuse among systems.
1.5.3 Learning Object Aggregation and Disassembly
As indicated in Section 1.5.1, there is a broad consensus that smaller learning
objects are more easily reusable. However, the majority of shared learning ob-
jects are coarse-grained compositions that are dicult to reuse [Motelet, 2004].
Typically, authors reuse parts of the learning object by copy-and-paste actions.
This method of reuse is possible in any authoring tool, but is limited in several
ways: the approach is non-scalable in terms of maintenance, as each time content
is copied, a new place is created that needs to be maintained. In addition, the ap-
proach tends to be error-prone, and due to its inherent monotony, easily becomes
both bothering and time consuming.
The authors are in a much better position if on-the-
y access to learning ob-
ject components is provided, and their re-composition is made, at least partially,
automatic. The main idea in our view is that learning objects are created by
selecting learning object components from a repository. These learning objects
can then be assembled into a new learning object. This can be referred to as
authoring-by-aggregation [Duval and Hodgins, 2003].
To enable authoring-by-aggregation, support is needed for automatic decom-
position of learning objects, to extract components of learning objects that were
originally produced as aggregates. A possible approach employs a more reusability
prone format of learning objects that makes their structure explicit. An explicit
content structure allows to disaggregate a learning object into its constituent com-
ponents. Those components, enriched with ne-grained descriptions, and stored
in learning object repositories, can then be selected to create new learning objects.
There are a number of issues that need to be dealt with to realize the approach.
First of all, there is the question of how far it is useful to decompose learning ob-
jects into components. As pointed out by [Rockley, 2002], sentence fragments or
individual words may not be appropriate for reuse. However, single paragraphs
may constitute denitions, examples or exercises that are reusable. Secondly, the
transformation of semi-structured or unstructured learning objects into an explic-
itly structured format needs to be investigated. Thirdly, integration of assembling
learning object components into the work
ow of authors needs to be examined.
This dissertation investigates both decomposition and assembly of learning ob-
ject components. In addition, prototypes of tools have been developed that enable

1.5 Issues and challenges 13
to validate the approach. Plug-ins have been developed for Microsoft PowerPoint,
Microsoft Word and the Reload Editor [Milligan et al., 2005], a packaging tool for
composition of SCORM content packages, that enable authors to search and reuse
components from within the authoring tools.
User and quality evaluations have been conducted that validate the approach.
The goals of the evaluations were threefold: (i) to assess the eciency and ef-
fectiveness of the approach for reusing learning objects; (ii) to assess the sub-
jective acceptance of the ALOCOM plug-ins; (iii) to determine to which level of
granularity decomposition is relevant.
1.5.4 Learning Object Interoperability
In order to enable widespread reuse, interoperability issues are extremely impor-
tant [Duval and Hodgins, 2003]. Standardization work presented in Section 1.4
focuses on interoperability between learning objects and learning management sys-
tems and interoperability between learning object repositories. An important, and
currently somewhat neglected, kind of interoperability is interoperability between
learning objects [Duval and Hodgins, 2003]. Examples include:
 Content objects from dierent original creation/authoring tools working to-
gether when assembled together into a learning object.
 Learning objects being able to work properly when moved among systems
using dierent specications.
Interoperability is required at dierent levels:
 Learning object content: content dened according to dierent learning ob-
ject denitions should be able to interoperate.
 Learning object structure: learning objects structured and packaged accord-
ing to dierent content packaging specications should be able to interoper-
ate.
 Learning object output formats: learning objects stored in dierent applica-
tion specic formats should be able to interoperate.
The three kinds of interoperability are investigated in this dissertation. Inter-
operability of learning object content is described in Chapter 2. The ALOCOM
content model is described and mappings between content models that enable their
interoperation.
Interoperability of learning object structure is described in Chapter 3. The
RAMLET model is described and mappings to other content packaging specica-
tions that enable their interoperation.

14 Introduction
Finally, interoperability of learning object output formats is described in Chap-
ter 4, in the context of the (de-)composition framework for learning objects. Such
interoperability is a condition to realize the vision of an open, large-scale learning
object infrastructure with sucient critical mass [Duval and Hodgins, 2003].
1.6 Outline
This dissertation describes conceptual designs and prototypes of tools that have
been developed to validate the approach. Earlier versions of the chapters have
been published, in whole or in part, in recent years. Among the most important,
in the context of this dissertation, are: [Verbert and Duval, 2004] [Verbert et al.,
2004a] [Verbert et al., 2005] [Verbert et al., 2005a] [Jovanovic et al., 2005] [Verbert
et al., 2005b] [Verbert et al., 2006] [Verbert and Duval, 2007] [Verbert et al., 2008]
and [Verbert and Duval, 2008].
The remaining chapters are organized as follows: Chapter 2 tackles the issue
of learning object granularity. The generic ALOCOM content model is presented,
and content model mappings that enable the interoperation of learning content
denitions.
Chapter 3 presents the RAMLET model for content packaging specications
and mappings to other content packaging formats. Use cases illustrate the level of
interoperability that can be achieved.
Chapter 4 presents a decomposition and aggregation framework for learning
objects. The framework automates learning object reuse by enabling on-the-
y
access to learning object components contained in composite learning objects.
Chapter 5 presents user and quality evaluations that measure the impact of
the approach on eective and ecient content reuse.
Finally, Chapter 6 concludes this dissertation with a summary of contributions,
a discussion on the possible impact, and an exploration of the potential it oers
for future research.

16 ALOCOM: a Generic Content Model for Learning Objects
on the belief that we can create independent and self-contained learning content,
which may be used alone or dynamically assembled, to provide "just enough" or
"just-in-time" learning. On top of that, these learning components can be com-
bined to form longer educational interactions or can be reused in dierent learning
contexts [Tan, 2002].
However, there are many dierent content models and learning object deni-
tions across these models vary considerably. Some models dene learning objects
as lessons, while others relate learning objects to concepts, principles, facts, pro-
cedures or processes. The heterogeneity of denitions is a barrier for learning
content reuse on a global scale, as it is unclear whether content can be reused or
repurposed in a dierent context.
In order to address heterogeneity problems, we have developed an abstract
learning object content model (ALOCOM) for content model interoperability. Ex-
isting content models have been investigated and mapped to the generic ALOCOM
model. Mappings have been implemented according to the method introduced
in [Bucella et al., 2003]. The method has three main stages:
 building a global ontology that covers existing content models,
 building local ontologies for each content model, and
 dening mappings between the ontologies.
Mappings can enable share and reuse of learning objects across repositories. Learn-
ing object components stored in a SCORM [SCORM, 2004] repository can, for
instance, be identied and potentially repurposed in the context of a Cisco [Barrit
et al., 1999] or NETg [L'Allier, 2003] learning system.
To facilitate the description and comparison of learning object content mod-
els, we rst brie
y introduce content classication schemes that are used by the
investigated content models for dening granularity levels. In Section 2.3, the
content models that were included in the investigation are presented and Section
2.4 presents a comparative analysis. The method used for implementing mappings
is described in Section 2.5. The global ALOCOM content model is presented in
Section 2.6, local content model ontologies in Section 2.7 and mappings in Section
2.8. Use cases are described in Section 2.9 and related work is discussed in Section
2.10. Finally, conclusions are drawn in Section 2.11.
2.2 Background
Learning object content models dene dierent levels of content components, the
properties of these components, such as granularity, and how these components
can be aggregated [Schluep, 2005]. In order to dene granularity levels, dierent
classication schemes are used by current content models, such as the Structured

2.2 Background 17
Writing methodology developed by Robert Horn [Horn, 1993], the classication of
Ballstaedt [Ballstaedt, 1997], classications dened in LOM [IEEE, 2002] and the
component display theory [Merrill, 1983]. The classications are brie
y introduced
in this section.
2.2.1 Structured Writing
The Structured Writing method of Robert Horn [Horn, 1998] was developed for
instructional developers and business writers to prepare clear and concise training
manuals, proposals, reports and memos. The methodology should enable man-
agers, sales people, oce personnel, and technicians to learn new products, ser-
vices, and operating procedures rapidly and precisely.
In the methodology, a paragraph is replaced by an information block, a chunk of
information that is organized around a single subject, containing one clear purpose.
Horn dened 200 types of information blocks, including: analogy, block diagram,
checklist, classication list, classication table, classication tree, comment, cycle
chart, decision table, denition, notation, objectives, outlines, parts-function table,
parts table, prerequisites to course, procedure table, purpose, rule, synonym, and
theorem.
In addition, a set of content analysis categories and question types were dened
based on seven information types [Horn, 1993]:
1. Concept: A "concept" describes an abstract or generic idea generalized from
particular instances. A concept is used for teaching a group of objects,
symbols, ideas, or events which are designated by a single word or term,
share a common feature and vary on irrelevant features [Barrit et al., 1999].
2. Fact: A "fact" provides information based on real occurrences; it describes
an event or something that holds without being a general rule [Ullrich, 2005].
3. Classication: A "classication" is a sorting of items into categories. An
example is "overview of technologies within medical imaging" [Ceusters and
Bouquet, 2000].
4. Structure: A "structure" is a physical object or something that can be di-
vided into parts and has boundaries. A typical example is "the anatomy of
the human brain" [Ceusters and Bouquet, 2000].
5. Principle: A "principle" is a basic generalization that is accepted as true and
that can be used as a basis for reasoning or conduct [Ullrich, 2005].
6. Procedure: A "procedure" consists of a specied sequence of steps or formal
instructions to achieve a goal. Typical examples are "Euclid's algorithm" or
"instructions to operate a machine" [Barrit et al., 1999] [Ullrich, 2005].

2.2 Background 19
non-textual representations. Textual representations are categorized into oral and
written texts.
Written texts are further divided into the following categories:
1. Expository texts: these texts contain factual representations of the subject
matter to be taught. Such texts may contain denitions and explanations.
2. Narrative texts: narrative texts are subjective descriptions of personal expe-
riences related to some subject matter.
3. Instructions: instructions provide a detailed description of how to perform a
procedure step-by-step.
4. Supplementary didactic texts: these texts are didactically motivated ele-
ments that support the learning process, classied as: learning objectives,
advanced organizers, summaries, examples, excursions, glossaries and self-
assessments.
Non-textual representations of learning content include: charts, tables, dia-
grams, gures, icons, and maps.
In contrast to Structured Writing and IEEE LOM, the research is not widely
disseminated: only the dLCMS content model [Schluep, 2005] refers to the clas-
sication. The classication might be useful in dening a global content model
for learning objects, though, as it is specically targeted at classifying learning
content.
2.2.4 The Component Display Theory
The Component Display Theory of David Merrill [Merrill, 1983] classies learning
along two dimensions: content and performance (see Table 2.1). Four types of
content (concept, fact, principle, procedure) are crossed with four types of learning
performance (remember generality, remember instance, use, nd). The content
types are contained in the Structured Writing classication.
In addition, the theory species four primary presentation forms: rules (exposi-
tory presentation of a generality), examples (expository presentation of instances),
recall (inquisitory generality) and practice (inquisitory instance). Secondary pre-
sentation forms include: prerequisites, objectives, helps, mnemonics, and feedback.
The theory species that instruction is more eective to the extent that it
contains all necessary primary and secondary forms. Thus, a complete lesson
would consist of an objective, followed by a combination of rules, examples, recall,
practice, feedback, helps and mnemonics appropriate to the subject matter and
learning task. The theory suggests that for a given objective and learner, there

20 ALOCOM: a Generic Content Model for Learning Objects
Facts Concepts Procedures Principles
Find Dene a class,
or set of ob-
jects or events
Derive, create
a procedure or
technique for
achieving a
goal
Discover cause
and eect
relations
Use Classify new
examples
Perform the
procedure
Solve a Problem
Make an infer-
ence
Remember
Generality
Remember
the facts
Remember the
denition
Remember the
steps
Remember the
guidelines
Remember
Instance
examples examples examples examples
Table 2.1: Content-Performance Matrix
is a unique combination of presentation forms, that results in the most eective
learning experience [Merrill, 1983].
The Component Display Theory provides the foundation for Ruth Clark's per-
formance matrix [Clark, 1989]. Clark's performance matrix, along with Merrill's
Component Display Theory, can help designers classify instructional outcomes and
are developed in some content models, such as the Cisco [Barrit et al., 1999] and
Learnativity [Wagner, 2002] models.
2.3 Overview of Learning Object Content Models
In this section, nine content models are presented that were included in the inves-
tigation. Models dened by some of the major players in the e-learning eld are
presented rst, followed by models that were developed for academic purposes.
2.3.1 NETg Learning Object Model
NETg [L'Allier, 2003], the National Education Training Group, is a worldwide
leader in blended learning solutions. In NETg, a course is structured as a matrix
(Figure 2.1) divided into three major components: units (the vertical), lessons
(the horizontal) and topics (the cells) [Tan, 2002].
Each unit, lesson and topic in this structure is dened, in part, by its relation-
ship to the other components.
1. Course: Made up of units
2. Unit: Made up of lessons

26 ALOCOM: a Generic Content Model for Learning Objects
4. An Asset is a single text element or a single media element (e.g. an assess-
ment object, a video, and other data elements).
A terminal objective is a major objective for a topic or task, describing the over-
all learning outcome. An enabling objective supports a terminal objective. Such
an objective describes specic behaviors (single activities) that must be achieved.
The Navy Content Model uses SCORM as its foundation. Table 2.2 presents
the relationship between the SCORM and NCOM hierarchy. NCOM correlates a
single enabling objective to a SCO and a single terminal objective to a SCORM
Activity. ELO and TLO content is thus more restrictive.
SCORM NCOM
Content aggregation Learning Object Aggregation
Activity Terminal Learning Object (TLO)
Sharable Content Object (SCO) Enabling Learning Object (ELO)
Asset (with metadata) Asset
Table 2.2: Relationship between the SCORM and NCOM hierarchy (Source: [Con-
key, 2006])
2.3.5 Cisco RLO/RIO Model
Cisco Systems, Inc. [Barrit et al., 1999] has also adopted an object-based strategy
for developing and delivering learning content. As illustrated in 2.6, Cisco denes
"Lessons" as Reusable Learning Objects (RLOs) and "Topics", of the lesson, as
Reusable Information Objects (RIOs).
Figure 2.6: The RLO and RIO structure [Barrit et al., 1999]

2.3 Overview of Learning Object Content Models 27
RIOs relate to a single learning objective and contain content, practice, and
assessment items. Cisco further classies each RIO as a concept, fact, procedure,
process, or principle. Content items are classied as a denition, example, review,
next steps, analogy, topology illustration, block diagrams, additional resources,
cycle charts, instructor notes, introduction, principle statement, illustration, im-
portance, outline, fact list, objectives, non-example, table, job-based scenario, pre-
requisites, guideline, procedure table, decision table, demonstration, staged table,
or combined table.
To build a lesson or RLO, ve to nine RIOs are grouped together with an
overview and summary (see Figure 2.6). For RIO types, and RLO Overviews and
Summaries, guidelines are provided that describe which content items are required
and which may be used optionally (see Table 2.3).
RLO-RIO type Content Items
RLO Overview Introduction (r), importance (r), objectives (r), prerequi-
sites (r), scenario (o), outline (r)
RLO Summary Review (r), next steps (o), additional resources (o)
Concept RIO Introduction (r), facts (o), denition (r), example (r),
non-example (o), analogy (o), instructor notes (o)
Fact RIO Introduction (r), facts (r), instructor notes (o)
Procedure RIO Introduction (r), facts (o), procedure table (r), decision
table (r), combined table (r), demonstration (o), instruc-
tor notes (o)
Process RIO Introduction (r), facts (o), staged table (r), block dia-
grams (r), cycle charts (r), instructor notes (o)
Principle RIO Introduction (r), facts (o), principle statement (o), guide-
lines (r), example (r), non-example (o), analogy (o), in-
structor notes (o)
Table 2.3: Overview of content items to be used for RIO types, RLO Overview
and RLO Summary (Source: [Schluep, 2005]); (r)=required, (o)=optional
A RIO can function as an independent learning component that can be called
up by a learner who needs a specic piece of information. Such RIOs can be com-
bined together to build custom RLOs that meet the needs of individual learners.
RLOs can be sequenced to create a course on a particular subject [Tan, 2002].
The Cisco model is grounded in the learning object thinking of David Merrill
[Merrill, 1983] and Ruth Clark [Clark, 1989]. RIO and RLO classications and
guidelines for their construction are based on the Structured Writing methodology
developed by Robert Horn [Horn, 1998].
A RIO correlates to a NETg topic, a SCORM SCO, and an NCOM ELO.
Content items relate to NCOM and SCORM assets and both raw data & media
elements and information objects in Learnativity. An RLO correlates to a Lear-

2.3 Overview of Learning Object Content Models 29
2. Content elements are dened as small, modular pieces of learning content,
which: (1) serve as basic building blocks of learning content, (2) can be ag-
gregated into larger, didactically sound learning units, (3) are self-contained,
(4) are based on a single didactic content type, (5) are reusable in multiple
instructional contexts, and (6) may contain assets. Examples include exer-
cises, experiments, questionnaires and summaries.
3. A learning unit is dened as an aggregation of content elements, which is
presented to the learner. Typically, a learning unit serves as an online lesson
and may be used to teach several learning objectives. A learning unit pro-
vides a way to dene a chapter-like, hierarchical structure of nodes. Each
node will be associated to a content element through reference. The content
elements are not copied into the learning unit, but are referenced by links.
The component model does not dene any further levels for the aggregation
of learning units.
The dLCMS model denes a set of Content Elements categories that are related
to Gagne's Nine Instructional Events (see Table 2.4).
Instructional Event Related Didactic Content Type
Gaining Attention Example13, problem statement2
Informing learners of the objective Learning objective1
Stimulating recall of prior learning Advanced organizer1
Presenting the stimulus Expository text1, denition3, narrative
text12, instruction1
Providing learner guidance Example13, excursion1, glossary1, litera-
ture, experiment2
Eliciting performance Exercise123, self-assessment2, simulation2
Providing feedback (Feedback of self-assessment and simula-
tions)
Assessing performance Questionnaire2
Enhancing retention and transfer Summary1
Table 2.4: Classication of didactic content types and their possible relations to
Gagne's Nine Instructional Events [Schluep, 2005]
The content categories are based on the classication of Ballstaedt1 [Ballstaedt,
1997], the vocabulary of the LOM Learning Resource Type2 [IEEE, 2002] and
ContentModule types of LMML3 (see Section 2.3.9) [Su et al., 2000]. Literature
is added to the classication.
The dLCMS model provides a well-dened hierarchy of learning object content:
Assets are assembled into Content Elements and Content Elements are assembled
into Learning Units. Learning units may be of any size and may be used for

30 ALOCOM: a Generic Content Model for Learning Objects
multiple learning objectives. dLCMS does not dene a learning object level that
relates to a single learning objective.
The model has been developed for academic purposes. A prototype demon-
strates how to handle and process modular learning content that is compliant
to the dLCMS model. The implementation supports learning object authoring,
storage, assembly and linking, and publishing and export functionalities.
2.3.7 New Economy Didactical Model
Another content model developed for academic environments is the New Economy
didactical model [Loser et al., 2002], developed in the context of the New Economy
research project, which is supported by the German Federal Ministry for Educa-
tion and Research. The aim of the project is the creation of new curricula and
the development of interactive multimedia-based material for online and blended
learning MBA studies. The project partners belong to 7 German universities and
research institutes.
The model denes eight component types, as shown in Figure 2.8:
1. An Information Object is dened as a small learning object, without complex
logical structures, which sums up physical media (picture, video, text) to
didactically appropriate units.
2. A Learning Component is dened as a small learning object, that combines
a small number of information objects, in order to form one of the following
features: motivation, basic knowledge or theory, example, exercise, reference,
further material, open question, problem, and virtual laboratory.
3. A Learning Module is dened as a logical structure with a didactic aim,
consisting of individual Learning Components. A Learning Module is related
to a Cisco RLO or lesson.
4. A Learning Unit is dened as a structure designed to mediate complex con-
tent. A Learning Unit combines Learning Modules and Learning Compo-
nents. An example is a case study containing three learning modules, com-
bined with a virtual laboratory.
5. A Course combines Learning Modules and Learning Units and can be part
of a Curriculum.
6. A Curriculum is a composition of Courses and Learning Units according to
one or more academic specications.
7. A Learning Path is a structure consisting of Learning Modules and Learning
Units, that can be individually adjusted to the learner.

2.3 Overview of Learning Object Content Models 31
Figure 2.8: UML representation of the New Economy Didactical Model
8. A Sequence is dened as a result of individual research within dierent repos-
itories, in order to extend personal knowledge. It is part of the informal, but
organized, learning procedure.
In addition, the following characteristics are dened (see Table 2.5):
1. Number of combined elements: describes the number of individual elements,
such as video clips, pictures, or texts, that are combined.
2. Type of the combined objects: describes types of the learning objects, which
can be combined, in order to form this learning object.

32 ALOCOM: a Generic Content Model for Learning Objects
3. Relationship logical structure/contents: describes the portion of logical struc-
tures in relation to content wise elements.
4. Possible didactical learning model: manufactures the connection between the
learning object and learning theory.
5. Reusability in other learning objects: describes the possibility of reuse within
other learning objects.
6. Reusability in other contexts: describes the possibility of the use of learning
objects in other domains.
From a content perspective, six aggregation levels are dened. Learning Path
and Sequence are pure structural elements. According to the authors of the model,
an Information Object correlates to a Learnativity Information Object, a Learning
Component to a Cisco RIO and a Learning Module to a Cisco RLO.
The dened characteristics derive from the work of David Wiley [Wiley, 2002].
The classications are based on the didactical concept of problem based learning.
The New Economy project is the conceptual design and implementation of a
multimedia-based curriculum for online classes regarding new economy in the elds
of economics, media and communications, as well as computer sciences. Modules
of the curriculum are available for workshops and for distance learning. Integration
of the program into regular lectures is intended.
2.3.8 Semantic Learning Model (SLM)
The Semantic Learning Model is aimed at supporting decomposition of learning
objects and has been developed for academic purposes [Fernandes et al., 2005].
The model is illustrated in Figure 2.9 and denes 6 categories:
1. The lowest granularity level is an Asset. Assets can be pictures, illustrations,
diagrams, audio and video les, animations, and text fragments.
2. Pedagogical information is dened as "a group of assets that express the
same meaning". An example is a gure associated with a comment.
3. A pedagogical entity is dened as "a pedagogical information component, as-
sociated with a pedagogical role". Four roles are dened: concept, argument,
solved problem and simple text.
4. A pedagogical context is dened as "a semantic structure (or network) in
which pedagogical entities are grouped".
5. A pedagogical document contains a pedagogical context, associated with
prerequisites.

38 ALOCOM: a Generic Content Model for Learning Objects
Figure 2.11: Ontology construction method [Bucella et al., 2003]
2.5.3 Second Stage: Building Local Ontologies
In this stage, an independent analysis of each content model is made, without tak-
ing the other content models into account. An ontology is created for each content
model, dening its own classes and relationships according to the specication of
the model. Section 2.7 illustrates the development of local ontologies.
2.5.4 Third Stage: Dening mappings
In this stage, mappings (and relationships) are dened between the classes dened
in the global ontology and classes dened in the local ontologies. This stage must
solve heterogeneity problems by making connections between the two stages. Such
mappings are presented in Section 2.8.
2.6 Abstract Learning Object Content Model (ALO-
COM)
2.6.1 Introduction
A global content model should dene the dierent granularity levels that are
present in current content models and their interrelationships. We have devel-
oped such a model in the ontology language OWL [Bechhofer et al., 2004], as we
use ontologies as a means to implement content model mappings.

2.6 Abstract Learning Object Content Model (ALOCOM) 41
 Video
 Audio
None of the content models dene a complete classication for content fragments.
Instead, the component types are used as examples.
Content Objects
Content models use (part of) the following classication schemes to dene content
objects:
 the vocabulary of the Learning Resource Type in IEEE LOM [IEEE, 2002],
 the classication of Ballstaedt [Ballstaedt, 1997], and
 Structured Writing [Horn, 1998]
Cisco, dLCMS, PaKMaS, New Economy and Learnativity use part of this Struc-
tured Writing classication. Cisco uses a subset that contains 29 of the 200 dened
component types: overview, summary, denition, example, review, analogy, topol-
ogy illustration, next steps, block diagrams, additional resources, cycle charts, in-
structor notes, introduction, principle statement, illustration, importance, outline,
fact list, objectives, non-example, table, job-based scenario, prerequisites, guide-
line, procedure table, decision table, demonstration, staged table, and combined
table. The New Economy model uses examples, references, and further material
that relates to additional resources. dLCMS uses objectives, summaries, examples
and denitions, and PaKMaS uses denitions, remarks, examples, and illustra-
tions. Learnativity uses the classication to exemplify information objects, but no
precise specication of component types suitable for dening learning content is
provided.
The IEEE Learning Object Metadata Standard [IEEE, 2002] denes a vocabu-
lary for Learning Resources Types that is partially used by the dLCMS, PaKMaS,
Learnativity and New Economy content models. Exercises, simulations, question-
naires, narrative text, experiments, problem statements and self-assessments are
used by the dLCMS content model. Exercises can be found in PaKMaS and ex-
ercises and simulations can be found in the Learnativity model. Finally, problem
statements, simulations and exercises can be found in the New Economy model.
The classication of Ballsteadt is used by the dLCMS model. The New Econ-
omy content model uses the term theory or basic knowledge to denote advanced
organizers. Finally, the following concepts are used by the dLCMS and/or New
Economy content models that are not represented in the Structured Writing, Ball-
staedt or LOM Learning Resource Type classications: motivation, open question,
paragraph, and literature. The content object classication dened by the global
content model represents the union of these used concepts, and is shown in Figure
2.13. The combination of these elements is brie
y discussed in Section 2.6.5.

42 ALOCOM: a Generic Content Model for Learning Objects
Figure 2.13: The ALOCOM Model

44 ALOCOM: a Generic Content Model for Learning Objects
2.7 Local Ontologies
2.7.1 Introduction
In the second stage of the method, local ontologies are dened for each content
model, representing concepts and relationships dened by the model. The local
ontology of the Cisco model is detailed in this section. Other local ontologies are
dened analogously. Their UML representations can be found in Section 2.3.
2.7.2 The Cisco Ontology
The Cisco ontology denes Cisco components and their interrelationships. An ex-
cerpt of the UML representation of the ontology is shown in Figure 2.14. Concepts,
concept hierarchies and aggregation relationships are represented in the UML dia-
gram, and in other UML diagrams presented throughout this chapter. Constraints
imposed on content components are presented in the axiom set below and are ex-
pressed in rst order logic. The constraints indicate, for instance, that a Cisco
concept should contain an introduction, denition and example, and may contain
a fact list, non-example, analogy and instructor notes. Cardinality constraints are
included in the UML diagrams.
AO = f(8x)Concept(x) ^ (8y)haspart(x; y)! Introduction(y) _ fact list(y) _ definition(y)
_example(y) _ non example(y) _ analogy(y) _ instructor note(y);
(8x)Overview(x) ^ (8y)haspart(x; y)! Introduction(y) _ importance(y)
_objectives(y) _ prerequisites(y) _ scenario(y) _ outline(y);
(8x)Summary(x) ^ (8y)haspart(x; y)! Review(y) _ next steps(y) _ additional resources(y);
(8x)RLO(x) ^ (8y)haspart(x; y)! Overview(y) _ Summary(y) _ RIO(y);
(99y)haspart(x; y) ^ RIO(y) ^ RLO(x); (94y)haspart(x; y) ^ RIO(y) ^ RLO(x);
(8x)Concept(x)! (9y)Introduction(y) ^ (9z)Definition(z) ^ (9w)Example(w):::g
2.8 Mappings
2.8.1 Introduction
In the last step, ontology mappings are dened between the global ALOCOM
model and local content model ontologies.
Ontology mappings are often dened as: "Given two ontologies A and B, map-
ping one ontology with another means that for each concept in ontology A, we
try to nd a corresponding concept, which has the same or similar semantics, in
ontology B and vice versa" [Su, 2006].

2.8 Mappings 47
Figure 2.15: Learning Object Component Mappings

50 ALOCOM: a Generic Content Model for Learning Objects
 The denition of content components in SLM is rather fuzzy: 6 aggregation
levels are dened, but only 4 can aggregate more than one content com-
ponent. Furthermore, it is unclear what is meant by the denition of a
pedagogical context, i.e. "a semantic structure in which pedagogical entities
are grouped". The lack of precise denitions and examples were a bottle-
neck in the analysis of the content model. For mapping the SLM content
model, we solely relied on its relationship to the Learnativity content model,
as proposed by the authors.
 As indicated in Section 2.8.3, the denition of New Economy Learning Com-
ponents is somewhat contradictory. The authors dene Learning Compo-
nents as motivations, theories, examples, exercises, references, further mate-
rial, open questions, problems, and virtual laboratories. In addition, they re-
late the component type to Cisco RIOs, that constitute concepts, facts, prin-
ciples, processes and procedures. To resolve the inconsistency, we mapped
the component type to the union of ALOCOM Content Objects and ALO-
COM Single Objective LOs.
2.9 Usage Scenario
Implementing content model mappings is useful in several ways. First of all, share
and reuse of learning object components is enabled across systems. For instance, an
LMS using SCORM content can be aligned with a Cisco repository at the content
level (see use case 1 in Figure 2.17). Equivalent components can be identied and
potentially repurposed within dierent contexts.
Figure 2.17: Use Cases

Chapter 3
RAMLET: A Model for
Structuring of Learning
Object Components
3.1 Introduction
The previous chapter has presented a content model for learning objects and their
components and is an important step towards supporting
exible reuse of learning
object components that can be aggregated to create new learning objects. In
order to realize the full potential of the approach, it is necessary to develop an
architecture that enables describing the structure of such aggregations.
Resource aggregation is the process of gathering resources and describing their
structure, so that the resulting aggregate can be used for transmission, storage,
and delivery to users [RAMLET, 2005]. The resource aggregate species how the
resources t together into a coherent, structured, whole. In addition, learning
object components comprising the resource aggregate can be structured in more
than one way.
Dierent communities, such as the multimedia, library, technical documenta-
tion, and learning technology community, have created their own specications
and standards for resource aggregates. Examples include the Metadata Encoding
and Transmission Standard (METS) [Cundi, 2004], an initiative of the Digital
Library Federation [Greenstein, 2002]; the IMS Content Packaging (IMS-CP) spec-
ication [IMS CP, 2004], that is predominantly used in the educational domain;
and the MPEG-21 Digital Item Declaration (MPEG-21 DID) [Bekaert, 2006], an
ISO standard for the audio-visual content industry. OASIS OpenDocument [Du-
rusau et al., 2007] and the W3C Synchronized Multimedia Integration Language
53

3.2 Use Cases 55
 RAMLET Core to MPEG-21 DID mapping: version 0.9.835
The chapter is organized as follows: Section 3.2 presents use cases of the RAM-
LET model. Section 3.3 outlines the resource aggregation formats that were used
in the development of RAMLET. Section 3.4 presents the RAMLET model and
the mappings. Finally, the use cases are revisited to clarify the RAMLET trans-
formation process and the level of interoperability that can be achieved, followed
by some concluding remarks.
3.2 Use Cases
3.2.1 Introduction
This section presents use cases that illustrate the need of a common reference
model for structuring of learning object components. The rst use case illustrates
exchange and reuse of resource aggregates among systems using dierent speci-
cations. Use case 2 illustrates the relation between RAMLET and ALOCOM.
ALOCOM components are assembled in a structured RAMLET aggregate and
exported to various resource aggregation formats. Use case 3 illustrates how the
use of RAMLET can enable interoperability of systems that use their own internal
format for resource aggregates.
3.2.2 Use Case 1
The use case addresses exchange and reuse of resource aggregates among systems
using dierent specications. For example, a system using METS might import re-
source aggregates that use IMS CP, and MPEG-21 DID, and create a new resource
aggregate.
Usage scenario
A content author in a university is developing a new resource aggregate and wishes
to include resources from dierent sources, including learning resources, reference
materials, and research data. The author searches for appropriate materials and
retrieves each resource to an authoring system. The resources are exported from
their repositories in resource aggregation formats specic to their respective reposi-
tories. The authoring system interprets the incoming resource aggregation formats
and converts them to its native format.
The author then creates the new resource aggregate, including the imported
resources, and makes the new resource aggregate available to the local learning
management system (LMS) or run-time system (RTS). The new resource aggregate
5http://www.ieeeltsc.org/working-groups/wg11CMI/ramlet/Pub/RAMLET-
MPEG21mapping.owl/view

56 RAMLET: A Model for Structuring of Learning Object Components
is in the resource aggregation format used by the authoring system. Figure 3.1
illustrates this scenario.
Figure 3.1: Use case 1
Use case summary: retrieve-interpret-aggregate-deploy
This use case addresses retrieving resource aggregates from diverse resource repos-
itories that provide resource aggregates in dierent resource aggregation formats.
The retrieved resource aggregates are interpreted, and converted into a single for-
mat that can be used by an authoring system and then aggregated into a new
resource aggregate. The new resource aggregate can be deployed by an RTS that
is limited to a single resource aggregation format.
3.2.3 Use case 2
This use case illustrates the relation between ALOCOM and RAMLET. The use
case addresses retrieving content from diverse repositories. The retrieved learning
objects are disaggregated into ALOCOM components and made available for reuse.
Authoring tools can connect to the repository, enabling on-the-
y aggregation of
relevant components from within authoring tools. RAMLET enables to describe
the structure and to deploy the resource aggregate to an LMS that is limited to a
single resource aggregation format.

3.3 Resource Aggregates 59
The delivery system is able to produce an internal representation of the resource
aggregate and to render the resources.
3.3 Resource Aggregates
3.3.1 Introduction
This section brie
y outlines the resource aggregation formats that were used in
the development process of RAMLET: IMS CP [IMS CP, 2004], METS [Cundi,
2004], MPEG-21 DID [Bekaert, 2006] and Atom [Sayre, 2005]. The structures and
properties of the resource aggregation formats are described. Simple examples
are included to illustrate the dierent constructs. Note that these examples are
incomplete and much simpler than real world applications.
3.3.2 IMS Content Packaging
IMS Content Packaging (IMS CP) is a specication that enables learning resources
to be transported between educational environments [IMS CP, 2004]. The IMS
Content Packaging specication was developed by the IMS Global Consortium and
plays a central role in the learning technology community.
An IMS Content Package contains two major components: an XML document,
called manifest le, that describes the content structure and associated resources
of the package, and the content making up the content package. The manifest le
is composed of four sections:
1. Metadata: Data describing the content package as a whole. Metadata are
usually included using IEEE LOM [IEEE, 2002], though strictly speaking it
can rely on other schemas.
2. Organizations: Contains the content structure or organization of the learning
resources making up a stand-alone unit or units of instruction.
3. Resources: Denes the learning resources bundled in the content package.
4. (sub)Manifest(s): Describes any logically nested units of instruction, which
can be treated as stand-alone units.
The organizations section describes zero, one, or multiple organizations of the
resource aggregate, as illustrated in Figure 3.4. Multiple organizations can pro-
vide learners with a variety of alternative structures for content. Each organi-
zation within the organizations section species how resources t together into a
hierarchically-arranged sequence of items. These items point to a resource in the
resource section, that lists the resources that together comprise the content of the
resource aggregate. An example is presented Listing 3.1.

60 RAMLET: A Model for Structuring of Learning Object Components
Figure 3.4: Conceptual description of elements in a manifest document
Listing 3.1: A simple manifest document
<mani fe s t>
<o r g a n i z a t i o n s d e f a u l t=" l e a r n i n g s e q ">
<o r g a n i z a t i on i d e n t i f i e r=" l e a r n i n g s e q " s t r u c t u r e=" h i e r a r c h i c a l ">
< t i t l e>Summer P i c tu r e s</ t i t l e>
<item i d e n t i f i e r=" item1 " . . . i d e n t i f i e r r e f=" re s1 ">
< t i t l e>Loch Katr ine</ t i t l e>
</ item>
<item i d e n t i f i e r=" item2 " . . . i d e n t i f i e r r e f=" re s2 ">
< t i t l e>Ben Ledi</ t i t l e>
</ item>
</ o r g a n i z a t i o n>
</ o r g a n i z a t i o n s>
<r e s o u r c e s>
<r e s ou r c e i d e n t i f i e r=" re s1 " type=" webcontent " h r e f=" f i v e . html">
< f i l e h r e f=" f i v e . html" />
< f i l e h r e f="supp/ r e l oadhe lp . c s s " />
</ r e sou r c e>
</ r e s o u r c e s>
</ mani f e s t>

3.3 Resource Aggregates 61
3.3.3 METS
The Metadata Encoding and Transmission Schema (METS) is a standard for en-
coding descriptive, administrative, and structural metadata regarding resources
within a digital library [Cundi, 2004]. The standard is developed as an initiative
of the Digital Library Federation [Greenstein, 2002].
A METS document consists of seven sections [Cundi, 2004], as illustrated in
Figure 3.5:
1. METS Header - The METS Header contains metadata describing the METS
document itself, including information such as the creator, editor, etc.
2. Descriptive Metadata - The descriptive metadata section may point to de-
scriptive metadata external to the METS document, or contain internally
embedded descriptive metadata, or both.
3. Administrative Metadata - The administrative metadata section, also em-
bedded or external to the METS document, provides information regarding
how the les were created and stored, intellectual property rights, metadata
regarding the original source object from which the resource aggregate de-
rives, and information regarding the provenance of the les comprising the
resource aggregate. Such metadata records modications that have been
made to a resource during its life cycle.
4. File Section - The le section lists all les containing content that comprise
the resource aggregate. <le> elements may be grouped within <leGrp>
elements, that can be used to organize individual le elements into sets. File
groups can, for instance, store multiple versions of the les.
5. Structural Map - The structural map is the heart of a METS document. As
in IMS CP, it outlines a hierarchical structure for the resource aggregate,
and links the elements of that structure to content les and metadata that
pertain to each element. Multiple structures of content can be specied.
6. Structural Links - The Structural Links section of METS allows METS cre-
ators to record the existence of hyperlinks between nodes in the hierarchy
outlined in the Structural Map.
7. Behavior - A behavior section can be used to associate executable behav-
iors with content in the METS object. Such sections link resources with
applications or programming code that are used to render or display the
resource.
The similarity between METS and IMS CP is great. Both resource aggregation
formats specify how resources t together into a hierarchically structured whole,

3.3 Resource Aggregates 65
a high bandwidth internet connection and a dial-up connection. Based on the
selection made by a user, the conditions attached to an entity of a digital item may
be fullled and the entity may become available. Dependent on the nature of the
conditions, the entity could contain a high resolution datastream or a compressed
le, respectively [Bekaert, 2006]. Such conditional behavior is not available in
METS or IMS CP.
Listing 3.3: A simple MPEG-21 DID document
<DIDL xmlns=" urn:mpeg:mpeg21:2002:01DIDLNS">
<Container>
<Desc r ip to r>
<Statement mimeType=" text / p l a i n ">
Album t i t l e
</ Statement>
</ Desc r ip to r>
<Item id=" track01 ">
<Desc r ip to r>
<Statement mimeType=" text / p l a i n ">
Song t i t l e 01
</ Statement>
</ Desc r ip to r>
<Component>
<Resource r e f=" track01 .mp3" mimeType=" audio /mpeg"/>
</Component>
</Item>
<Item id=" track02 ">
<Desc r ip to r>
<Statement mimeType=" text / p l a i n ">
Song t i t l e 02
</ Statement>
</ Desc r ip to r>
<Component>
<Resource r e f=" track02 .mp3" mimeType=" audio /mpeg"/>
</Component>
</Item>
</ Container>
</DIDL>
Listing 3.3 presents an example of an MPEG-21 DID document. The DIDL ele-
ment is the root element of the aggregation. A container element corresponds to
an IMS CP organization and METS structMap element and describes the structure
of the resource aggregate. In the example, the container element groups two item
elements, track01 and track02. MPEG-21 items corresponds to IMS CP items and
METS div elements. These items bind components to descriptor elements. Such
components, then, correspond to IMS CP Resource and METS le elements.

3.4 RAMLET 67
METS, MPEG-21 DID and IMS CP allow metadata to be expressed using
external standards. The communities that use the resource aggregation standards
are likely to favor the use of certain metadata standards. The learning community,
for instance, strongly favors the use of LOM metadata in IMS CP documents.
Libraries using METS are likely to favor MODS [Gartner, 2003], Dublin Core
[Weibel et al., 1998], or MARCXML [De Carvalho et al., 2004] for describing
content and MIX [ANSI/NISO Z39.87, 2006] or textMD [McDonough 2007] for
capturing technical metadata [Yee and Beaubien, 2004].
Parsing out MODS and MIX metadata into LOM categories and elements,
and vice versa, is beyond the scope of RAMLET. The RAMLET reference model
focuses on the structures of the various resource aggregation specications by ana-
lyzing and describing the properties of common structural concepts. Such reference
model will enable to transform a resource aggregate of one specication into the
structure of another, enabling to move resources aggregates between systems, and
thereby allowing easier reuse and sharing. Such transformation produces valid re-
source aggregates, as the resource aggregation formats do not restrict the format
of metadata. Metadata translation, or extraction of embedded content, will need
to be considered in some cases. For instance, an editor for IMS Content Packages
might not be able to display Dublin Core metadata correctly. As we use RAMLET
for structuring and exchange of learning object aggregations, such translation is
less important in the context of this dissertation.
3.4 RAMLET
3.4.1 Introduction
The RAMLET model denes a common nomenclature and an ontology that can be
used to represent dierent resource aggregation formats and specications. Inter-
operability is achieved by mappings, that have been dened between the RAMLET
model and other resource aggregation formats.
The model has been developed using the integration method presented in the
previous chapter. Dierent resource aggregation formats, such as IMS CP, METS,
MPEG-21 DID, were analyzed and a global ontology was constructed that covers
the specications. Local ontologies were developed for the resource aggregation
formats and nally mappings were dened between the global RAMLET model
and local resource aggregation ontologies.
This section presents the RAMLET model and the mappings that are currently
available. Denitions of RAMLET terms, abbreviations and acronyms can be
found in Appendix B.

3.4 RAMLET 69
Table 3.1: RAMLET CORE Draft August 9.
Class
Num-
ber
Class Name Annotation Associated
superclasses
2.3 aggregationFormat-
Version
Identifying string for an aggrega-
tion's dening format or prole ver-
sion.
descriptorObject
2.4 aggregationType Category for the resource being ag-
gregated.
descriptorObject
2.5 alternateID Contains an identifying string by
which the associated element compo-
nent is also known in another system.
descriptorObject
2.6 checksum A value that can be used to check the
integrity of the associated resource.
descriptorObject
2.7 checksumType Information about the algorithm by
which the associated checksum was
calculated.
descriptorObject
2.8 creationDate Element that contains the date when
an associated element was created.
descriptorObject
2.9 descriptive Container for information to support
nding, identifying, selecting and ob-
taining the resource.
descriptorObject
2.10 encodingType An element that indicates the
method by which the associated
resource has been serialized within
the aggregation.
descriptorObject
2.11 leSize Indicates the size of a component
le referenced in the aggregation in-
stance.
descriptorObject
2.12 generatingTool Identies the tool used to make the
aggregation instance.
descriptorObject
2.13 humanLanguage Indicates the human language in
which a resource is rendered.
descriptorObject
2.14 intendedUse Indicates the function an associated
set of digital resources is intended to
have in the aggregation.
descriptorObject
2.15 mdTypeIndicator Categorizes the metadata contained
in an element belonging to an aggre-
gation format.
descriptorObject
2.16 mimeType Indicates the electronic media format
of a component le referenced in the
aggregation instance.
descriptorObject
2.17 modicationDate Contains the last date of modica-
tion of associated entities.
descriptorObject

3.4 RAMLET 71
Table 3.1: RAMLET CORE Draft August 9.
Class
Num-
ber
Class Name Annotation Associated
superclasses
2.24 status Indicates the state of the associated
element.
descriptorObject
2.25 structureNodeType Indicates a category that describes a
component of an aggregation struc-
ture
descriptorObject
2.26 technical Information about the technical for-
mat of the described digital resource.
descriptorObject
2.27 textType Indicates the type of text used in an
associated container node, e.g., plain
text or rtf.
descriptorObject
2.28 wholeAggregation Information about the resource ag-
gregation instance itself, rather than
the resources it aggregates.
descriptorObject
2.28.1 identiferType Identies the identier scheme of an
associated identier.
wholeAggregation;
descriptorObject
3.0 aggregateID Identier for the aggregation. owl:Class
4.0 elementID Element that provides a local identi-
er.
owl:Class
4.1 nodeID Identier for a structural node. elementID
5.0 digitalResource The digital resource(s) that an aggre-
gation format instance aggregates.
owl:Class
5.1 digitalResource-
Fragment
A component of a digital resource
which is an entity specically ad-
dressed within the structure of an ag-
gregation instance.
digitalResource
6.0 staticStructure The structural relations between en-
tities described in a staticStructure
always hold true regardless of the
state of the aggregation.
owl:Class
6.1 staticStructureType Categorization of the nature of the
structure describing the relations be-
tween entities within an aggregation
instance.
staticStructure
7.0 staticStructureSet Contains collection(s) of static struc-
tures.
owl:Class
8.0 dynamicStructure Relations between entities in a dy-
namicStructure depend on factors
that are only true for one state, e.g.,
as calculated at runtime.
owl:Class
8.1 dynamicStructure-
Type
Categorizes the nature of a dynamic
structure.
dynamicStructure

3.5 Discussion 75
Table 3.2: Resource aggregation specication mappings
RAMLET IMS CP METS MPEG-21 Atom
2.28.1 identiferType IdType
3.0 aggregateID objID DIDL-
DocumentId
id
4.0 elementID Identier Id Identier;
Target
4.1 nodeID contentIDs
5.0 digitalResource File le Resource entry
5.1 digitalResource-
Fragment
area Fragment;
Anchor
6.0 staticStructure Organization structMap Container
6.1 staticStructure-
Type
structMapType
7.0 staticStructureSet Organizations
8.0 dynamicStructure behavior
8.1 dynamicStruc-
tureType
btype
8.2 dynamicStruc-
tureID
ChoiceId;
SelectId
8.3 assertion Assertion
8.4 condition Condition
8.5 choice Choice
8.6 selection Selection
8.7 maxSelections maxSelections
8.8 minSelections minSelections
8.9 defaultSelection default
8.10 require require
8.11 except except
9.0 dynamicStruc-
tureSet
behaviorSec
10.0 structureNode Item div;
fptr
Item
3.5 Discussion
The RAMLET model adequately covers the IMS CP, METS, MPEG-21 and Atom
formats. However, translating resource aggregates into the structure of another
format will not always be lossless. For instance, the transformation of an IMS CP
document into an Atom feed can only preserve content and (part of the) metadata.
Structural relationships between content les will be lost.
In this section, the use cases presented in Section 3.2 are brie
y revisited,
required transformations are outlined and information losses are described.

76 RAMLET: A Model for Structuring of Learning Object Components
3.5.1 Use Case 1
The use case addressed exchange and reuse of resource aggregates among systems
using dierent specications. For example, a system using METS might import
resource aggregates that use IMS CP, and MPEG-21 DID, and create a new re-
source aggregate.
Transformation summary: The import of an IMS CP and MPEG-21 DID resource
aggregate proceeds in two steps. In the rst step, the IMS CP and MPEG-21 DID
aggregates are transformed into a representation compliant with the RAMLET
model. The RAMLET resource aggregate is then transformed into METS.
3.5.2 Use Case 2
The use case addressed retrieving content from diverse repositories. The re-
trieved learning objects are disaggregated into ALOCOM components that are
re-assembled into new learning objects. RAMLET enables to deploy the resource
aggregate to an LMS that is limited to a single resource aggregation format.
Transformation summary: The structure of assembled ALOCOM components is
described in a RAMLET resource aggregate and is then transformed into an IMS
CP, MPEG-21 DID, METS or Atom aggregate.
3.5.3 Use Case 3
An LMS creates a resource aggregate just in time and will import, store, and make
available resource aggregates from systems using dierent specications. For ex-
ample, a system using its own internal format might import resource aggregates
that use IMS CP, METS, MPEG-21 DID and Atom.
Transformation summary: Atom, IMS CP, METS and MPEG-21 DID resource
aggregates are transformed to RAMLET. The RAMLET resource aggregate is
then transformed into the internal representation format of the LMS. To enable
such interoperability, a mapping between the internal format and RAMLET needs
to be implemented.
3.5.4 Information Losses
As described above, transformations between resource aggregate formats proceed
in two steps: rst, the resource aggregates are transformed into a representation
compliant with the RAMLET model. Second, the RAMLET resource aggregate
is transformed into another aggregation format, such as MPEG-21 DID, METS,
IMS CP and Atom.

Chapter 4
An Aggregation and
Disassembly Framework for
Learning Objects
4.1 Introduction
In the previous chapters, the ALOCOM and RAMLET models have been pre-
sented that enable the interoperation of learning content and structure available
in dierent e-learning formats. Such interoperability is important for supporting
content reuse on a global scale, as learning objects can be exchanged and reused
across various learning systems.
However, the majority of content available on the World Wide Web is stored in
unstructured or semi-structured formats, such as Microsoft Word, PowerPoint or
HTML formats. To enable their reuse, we have developed a framework that trans-
forms such formats into a representation compliant with the ALOCOM model.
In this transformation process, the framework decomposes learning objects and
provides direct access to content components, enabling their automatic reuse in
new learning objects.
There are a number of issues that need to be dealt with to realize the approach:
1. First of all, there is the question of how far it is useful to decompose learn-
ing objects into components. As pointed out by [Rockley, 2002], sentence
fragments or individual words may not be appropriate for reuse, as their
added value for reuse is questionable. Complete sections or chapters on the
other hand may be too coarse-grained for use in a dierent context. The
issue of nding an appropriate granularity level for decomposition is further
described in Section 4.2.2.
79

4.2 The ALOCOM Framework 81
Figure 4.1: The ALOCOM Framework
text fragments. Text documents are decomposed into sections and subsec-
tions, and each section is further decomposed into paragraphs, images, tables,
diagrams, etc. The current implementation of this module supports the ap-
proach for PowerPoint presentations and Wikipedia pages. Components are
extracted, preview thumbnails are generated and results are stored through
the AdvancedContentInserter (see Section 4.2.2).

4.2 The ALOCOM Framework 83
Figure 4.2: The Decomposition Process
The current implementation of this module automates decomposition of Mi-
crosoft PowerPoint presentations and Wikipedia pages. The latter are decomposed
on-the-
y at the client side (see Section 4.3). Decomposition of presentations is
performed on the server. The module is implemented as a .Net web service and
uses the PowerPoint API [Khor and Leonard, 2005] to retrieve content and struc-
ture from a presentation.
The decomposition method iterates over the slides and slide shapes of a Power-
Point presentation object. Each slide is stored in the PowerPoint format to enable
lossless reuse. Images are extracted and stored in their original format and text
fragments and tables are stored in an XML format containing their content and
structure. For slides, an XML representation is generated to enable their reuse
in other applications and for detecting reuse between slides (see Section 4.2.3).
Finally, preview thumbnails are generated for each component, using built-in ex-
port functions of the PowerPoint API. These thumbnails are used in the search
interface of client applications (see Section 4.3).
In the next step, the generated components are sent to the AdvancedCon-
tentInserter for storage and indexation.
4.2.3 AdvancedContentInserter
The AdvancedContentInserter is part of a Java web service that relies on the ARI-
ADNE Knowledge Pool System [Duval et al., 2001] for storage of learning object
components. The module automates reuse detection for individual components