Eshaa M. Alkhalifa - Eshaa Alkhalifa - Cognitively Informed Systems - Utilizing Practical Approaches To Enrich Information Presentation and Transfer (2006, Idea Group Publishing) - Libgen - Li

Cognitively
Informed Systems:
Utilizing Practical
Approaches to Enrich
Information Presentation
and Transfer
Eshaa M. Alkhalifa
University of Bahrain, Bahrain
IDEA GROUP PUBLISHING

Hershey • London • Melbourne • Singapore
Acquisitions Editor: Michelle Potter
Development Editor: Kristin Roth
Senior Managing Editor: Amanda Appicello
Managing Editor: Jennifer Neidig
Copy Editor: Joyce Li
Typesetter: Jennifer Neidig
Cover Design: Lisa Tosheff
Printed at: Integrated Book Technology
Published in the United States of America by

Idea Group Publishing (an imprint of Idea Group Inc.)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: [email protected]
Web site: http://www.idea-group.com
and in the United Kingdom by

Idea Group Publishing (an imprint of Idea Group Inc.)
3 Henrietta Street
Covent Garden
London WC2E 8LU
Tel: 44 20 7240 0856
Fax: 44 20 7379 0609
Web site: http://www.eurospanonline.com
Copyright © 2006 by Idea Group Inc. All rights reserved. No part of this book may be repro-
duced, stored or distributed in any form or by any means, electronic or mechanical, including
photocopying, without written permission from the publisher.
Product or company names used in this book are for identification purposes only. Inclusion of the
names of the products or companies does not indicate a claim of ownership by IGI of the
trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data
Cognitively informed systems : utilizing practical approaches to enrich information presentation

and transfer / Eshaa M. Alkhalifa, editor.
p. cm.
Summary: "This book identifies the main areas of cognitive science and for each area, how different
system designs benefit from the findings made in that area"--Provided by publisher.
Includes bibliographical references and index.
ISBN 1-59140-842-3 (hardcover) -- ISBN 1-59140-843-1 (softcover) -- ISBN 1-59140-844-X
(ebook)
1. Human-computer interaction. 2. User interfaces (Computer systems) 3. Cognition. 4.
Cognitive science. I. Alkhalifa, Eshaa M., 1966-
QA76.9.H85C442 2006
004'.019--dc22
2005027609
British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.
All work contributed to this book is new, previously-unpublished material. The views expressed in
this book are those of the authors, but not necessarily of the publisher.
Cognitively
Informed Systems:
Utilizing Practical Approaches to
Enrich Information Presentation
and Transfer
Table of Contents
Preface ................................................................................................. vii
Chapter I
Cognitively Informed Systems: Justifications and Foundations ......... 1
Eshaa M. Alkhalifa, University of Bahrain, Bahrain
SECTION I
PERCEPTION, MEMORY, AND RECALL
Chapter II
Hypervideo and Cognition: Designing Video-Based Hypermedia for
Individual Learning and Collaborative Knowledge Building ............. 26
Teresa Chambel, University of Lisbon, Portugal
Carmen Zahn, Knowledge Media Research Center, Germany
Matthias Finke, Computer Graphics Center, Germany
Chapter III
Assisting Cognitive Recall and Contextual Reuse by Creating a
Self-Describing, Shareable Multimedia Object ................................. 50
Michael Verhaart, Eastern Institute of Technology,
New Zealand
Kinshuk, Massey University, New Zealand
Chapter IV
Guidance in the Interface: Effects of Externalizing Information
During Problem Solving ....................................................................... 74
Christof van Nimwegen, Utrecht University, The Netherlands
Hermina Tabachneck-Schijf, Utrecht University, The Netherlands
Herre van Oostendorp, Utrecht University, The Netherlands
SECTION II
MENTAL REPRESENTATION OF CONCEPTS, METAPHOR, AND LANGUAGE
Chapter V
Bridging the Gap between Human Communications and
Distance-Learning Activities ............................................................ 102
Sébastien George, INSA, Lyon, France
Chapter VI
Toward Noninvasive Adaptation of Metaphors in Content ............. 117
Alexei Tretiakov, Massey University, New Zealand
Roland Kaschek, Massey University, New Zealand
Chapter VII
A User-Centered Approach to the Retrieval of Information in an
Adaptive Web Site ............................................................................. 142
Cristina Gena, Università di Torino, Italy
Liliana Ardissono, Università di Torino, Italy
SECTION III
MENTAL REASONING AND INTERACTION
Chapter VIII
Swarm-Based Wayfinding Support in Open and Distance
Learning ............................................................................................. 167
Colin Tattersall, Open University of The Netherlands,
The Netherlands
Jocelyn Manderveld, Open University of The Netherlands,
The Netherlands
Bert van den Berg, Open University of The Netherlands,
The Netherlands
René van Es, Open University of The Netherlands,
The Netherlands
José Janssen, Open University of The Netherlands,
The Netherlands
Rob Koper, Open University of The Netherlands,
The Netherlands
Chapter IX
From Engineer to Architecture? Designing for a Social Constructivist
Environment ....................................................................................... 185
Karen Lee, University of Dundee, UK
Chapter X
Enriching Computer Support for Constructionism ........................... 210
Meurig Beynon, University of Warwick, UK
Chris Roe, University of Warwick, UK
SECTION IV
INDIVIDUAL DIFFERENCES AND INDIVIDUALLY BASED SYSTEMS
Chapter XI
An Architecture for Developing Multiagent Educational
Applications for the Web ................................................................... 236
Tasos Triantis, University of Patras, Greece
Panayiotis Pintelas, University of Patras, Greece
Chapter XII
Impact of Individual Differences on Web Searching
Performance: Issues for Design and the Digital Divide .................. 261
Allison J. Morgan, Pennsylvania State University, USA
Eileen M. Trauth, Pennsylvania State University, USA
Chapter XIII
Using Bayesian Networks for Student Modeling ............................ 283
Chao-Lin Liu, National Chengchi University, Taiwan
SECTION V
A REAL-WORLD CASE STUDY
Chapter XIV
The Effect of Technology on Student Science Achievement ........... 312
June K. Hilton, Claremont High School, USA
About the Authors .............................................................................. 334
Index ................................................................................................ 342

vii
Preface
We dance around in a circle and suppose, while the secret sits in the middle
and knows. (Frost, 1971)
It is the laborious life of scientific research that requires us to continually pre-

sume, then find support for our presumptions, which either stand the test of
time or fall prey to falsifying evidence only to be replaced by new ideas and
discoveries. It is this scientific track as described by Popper (1963) that has
been followed by scientists taking them from the early days of inventions to
current-day knowledge, always following a specific track. First the ideas, then
testing, followed by the formation of theories, and after a science reaches a
stable state of affairs, the fruits of that knowledge are reaped in the form of
applications and inventions that people can benefit from.
I do not think there is any thrill that can go through the human heart like
that felt by the inventor as he sees some creation of the brain unfolding
to success.... Such emotions make a man forget food, sleep, friends, love,
everything. (Nikola Tesla, 1856-1943)
Cognitive science as a field has gone through a long period of discovery and
knowledge in its various forms and areas of focus. The findings have stabilized
over the years to be able to predict and advise on how best to interact with
viii
the human mind, even if on a limited scale. This book aims to bring together
the contributors whose respect for the human mind has led them to take the
fruits of this science into account while laboring to design systems that interact
with that form of magnificence.
Consequently, this book comes as an expected step forward along the natural
path of research that starts with pure theory and ends with concrete designs,
development, and assessment of tools. Yet, it is to date unprecedented as it
contains an organization of the efforts put forward by researchers and design-
ers of novel approaches into five main streams that feed into a larger river.
These streams are partitioned in a fashion that is adequate to the human cog-
nitive machine.
For the first of these, we may notice that all humans start their interactions
with the world by utilizing their various senses for perceiving, storing what
they perceive in their memory to recall it at a later date. The second investi-
gates the organization of stored information in human memory and the cues
that cause someone to recall them in addition to the symbols and analogies
that are formed between concepts and their names or concepts and other
concepts. The third studies mental reasoning, which is the path followed from
what is given to the cognitive system until it gets to the deductions it can make
based upon that and this process, if repeated, may cause an interaction with
the outside world. This interaction may be in the form of trial and error with
the outside world along the path of discovery learning. The fourth stream iden-
tifies the individual differences between cognitive characteristics where some
individuals may have certain strengths, weaknesses, preferences, or even feel
more comfortable interacting with a particular type of system. The fifth and
last stream includes real-life case studies that are tested in actual classrooms.
The chapters of this book are therefore representatives of each of these five
streams of research in this field in order to cover all possible directions within
cognitively informed systems. However, representing each of the five main
branches does not imply that these chapters will exhaust all the relevant ques-
tions, but instead only act as suggestions of the vast possibilities that may
follow along the five main branches.
The first of the streams represents work on perception, recall of images, the
effects of externalization of information, and how that interacts with recall
during problem solving.
Teresa Chambel et al. in Chapter II exhibit one such direction by altering the
classical display of video for learners and presenting them instead in a
hyperlinked fashion. The various movie clips can be navigated and learners
are allowed to go and review a particular scene or to seek further details
ix
through another such link. The system also allows collaboration within differ-
ent student groups if they can watch the same movie or reference the same
movie clip link. Such work allows researchers to identify how learning from a
movie takes place and what parts of the movie may influence learning the
most, consequently informing cognitive learning theory of the particularities
that are discovered through the application of this approach.
Verhaart and Kinshuk introduce in Chapter III a proposal of how video clips
could be stored in memory in multiple representations. Their work comple-
ments what is presented in Chapter II by exhibiting the practicality of the
approach. In this chapter, the representations are autogenerated and allow
the user of the system to retrieve the images in a manner that does not neces-
sitate recall of all details as in using thumbnails to facilitate the retrieval of
images. Thumbnails provide the necessary cues to stimulate the recall of a
desired object that the searcher forgot parts of its details such as when searching
for an image while having forgotten its name.
Chapter IV, on the other hand, as introduced by van Nimwegen et al., has
more of an interest in how externalizing rules that have to be followed during
problem solving influence what students recall at a later date. This work is
extremely informative in that it shows how computer users who follow clear
instructions are relieved of the cognitive load imposed by the task they are
doing, but at the same time accomplish the task without learning how they did
that. Students who were tested several months later revealed that the informa-
tion presented to them during the problem-solving process affects what goes
into their long-term memories from the learning process.
The second of the streams studies the mental representation of concepts,
metaphors, and language. The mental representation of concepts is extremely
difficult to study even in the field of cognitive science, so one solution is to
attempt to organize information in a way that seems most suited to it and to
evaluate how that organization aligns itself with human mental representation.
Another issue is that of how words are used to “mean” certain concepts or, in
a more general view, how analogies or comparisons of different concepts are
made. Along this track, three chapters are offered.
In Chapter V, George offers an intuitive organization of a forum that is used
for discussions where the links of the forum are places such that they are
accessible when a student goes to a relevant lesson. The aim is to show learn-
ers how to access what is discussed on that particular topic or relevant topics.
The goal is to identify how concepts are related to each other and, in turn, to
reflect that by enforcing the links between the various forum contributions to
allow a faster, simpler access to students who wish for further feedback or
x
discussion. Here, the main issue is that of relevancy of the learned materials
and how it relates to what is discussed.
Tretiakov and Kaschek, conversely, offer a point of view in Chapter VI that is
novel in that it offers an interface that may work with various tutorial systems
without requiring any alterations to the actual tutorial system. The interface
offers the ability to select various metaphors or concept names and to de-
scribe them through analogical comparisons. The aim is to explain to students
through a means that allows them a deeper understanding of the concepts that
would in turn allow them to make deductions based upon the analogies made.
Ardissono and Gena classify users of the system they built into two levels:
novice and expert users. In Chapter VII, they describe how they collected
historical data on associations between different information needs that fre-
quently occurred together and utilized the results to make suggestions to users
of either of the two knowledge levels adaptively through the system. The time
it takes users to access their particular informational needs was greatly re-
duced through utilizing this historical data of associations that are based on
previous requests from the system. Here, knowledge is drawn from past us-
age in the form of cognitive associations and then applied to the design of this
system.
The third of the streams is concerned with mental reasoning. Although some
may assume that mental reasoning takes place with analogies, the form dis-
cussed in Chapter VI presents the analogies and checks whether students
access them. It does not investigate the deductions made. However, the fol-
lowing chapters allow students to make choices, and based upon these choices
they attempt to analyze the choices that must be allowed for students and how
this can be achieved within the limits of current computer systems.
Chapter VIII, as presented by Tattersall et al., presents a system that studies
how students choose the path sequences they follow during learning. It then
draws a map of their choices and tries to deduce the most successful paths
that were followed. This is then used to suggest to students possible future
steps while allowing students the full choices of whether to follow these sug-
gestions. Notice that it is unlikely here that all paths will converge to a single
one, as learners are known to have cognitive differences, so this study is likely
to illuminate how differences may emerge in the sequences of lessons students
follow to achieve more efficient learning.
Chapter IX, by contrast, is presented by Lee, to focus on student collabora-
tion during constructivist study. She uses a blackboard setting to promote
scaffolding as represented in presenting partial information to students and
allowing them to discuss and attempt to arrive at conclusions based upon the
xi
information they could gather on their own. Here, there is no imposed struc-
ture to the knowledge presented, and instead learners must discover it on
their own, and the only structure imposed is that by the scaffolding offered by
moderators who participated in shaping the discussions and ensuring they are
going in the right direction.
However, Beynon and Roe, in Chapter X, offer a contrary view indicating a
serious flaw with current programming issues when it comes to applying
constructivist approaches. They give a very detailed presentation of the
constructivist approach in addition to explaining the core differences between
the current programming paradigm and the basic requirements of constructivism.
The difference seems to lie between giving learners a goal to accomplish,
where they have to draw a plan of how to reach it, and when they are given an
environment and no plan is possible other than discovery while their conclu-
sions emerge spontaneously. An interesting perspective here on cognition is
that this may relate to what is currently described as cognitive insight as repre-
sented as the moment of discovery.
The fourth stream is concerned with individual differences whether they exist
within different cognitive characteristics or whether they exist within genders
or whether they exist in the frequency and types of errors students may make.
Chapter XI, as presented by Triantis and Pintelas, describes a multiagent ar-
chitecture where a mobile agent represented by the user interface interacts
directly with the learner and seeks to provide whatever is needed by that
learner. This agent is able to accomplish this by interacting with other agents
that exist in the background with the aim of requesting tutorials from them
according to the learning needs of the student. This approach is similar to
providing a representative who provides custom designs to each individual
accessing the system, to reduce search time, in addition to providing a user-
friendly environment and reducing any anxieties that may exist prior to learn-
ing.
Morgan and Trauth, on the other hand, present in Chapter XII a detailed
analysis of the digital divide between those who have access to the Internet
and those who do not take advantage of the digital world. Their emphasis in
this particular chapter is on the differences within each gender rather than
between the two genders. Their particular focus is on the words used while
running Web searches on search engines. People’s individual selections of
words that are used to represent the concepts they are searching for may
differ and therefore impact their use of these search engines and possibly even
impact the use of the Internet as a whole. The choice itself is a cognitive one,
which is made based on what that particular individual deems as the most
xii
appropriate to describe the concept desired, but this may not correspond to
the same choice made by the designers of the various Web sites.
Liu, in Chapter XIII, studies through a Bayesian model the occurrence of
guessing in student responses in tests. The study concludes that the level of
similarity between the items in the test itself and those presented in the course
materials play a crucial role in affecting the amount of successful guesses that
students may make. If on the other hand the number of correct answers is
fixed in various trials, guessing is yet again affected. This particular chapter
identifies, the false positives that may emerge in tests and assists in isolating
these especially when evaluating tutoring systems or in student assessments.
The last stream is concerned with a real-world study of the effects of technol-
ogy on student science achievement. Although the previous chapters are all
informative, the whole book and all the ideas proposed have one goal in mind—
namely to benefit those who use the systems.
Hilton presents in Chapter XIV a study on 1,194 students in a high school in
California. One of the main conclusions made is that the use of technology
does affect learning. The main problems exist in how the technology is used,
as using it in one way may encourage the amount of learning made while using
it in another may hinder learning. This is a clear indicator that care must be
taken when designing tutorial system to take into account all cognitive charac-
teristics that may influence learning, because ignoring them may lead to undes-
ired consequences.
The main conclusion that one arrives at is that this book presents practical
work that is being done today to bring the findings of a mature science to the
world of reality. Various approaches build upon dispersed findings made in
perception, reasoning, memory, and so on, to present the world with cognitively
informed systems. The book comes as a natural consequence of the evolution
of science, and it is not likely to be the last that will be geared toward this
particular target. As systems grow more aware of the cognitive state, they are
more likely to better serve and be more aligned with what they are built to
achieve. The progress continues.
xiii
References
Frost, R. (1971). The road not taken. New York: Henry Holt.
Popper, K. (1963). Conjectures and refutations: The growth of scientific
knowledge. London: Routledge & Kegan Paul.
xiv
Acknowledgments
I would like to first and foremost, thank God for the moments of insight that
illuminated this path toward knowledge.
I would also like to thank each and every author and contributor to this work,
whom I constantly bugged with my frequent e-mails and nagging to keep the
process going. Without your patience and hard work we would not have come
to reap the fruits of our labor.
I also would like to thank those who took the time to review the chapters in
this book whether they were from the contributors or just reviewers whose
sole purpose was to participate in the success of this work. From the latter
group, may I thank in particular Ali Khan, Amjad Mahmood, Gamal Kassem,
Kirtsten Ellis, Kostadin Kratchar, Mansour Al Aali, and Mencar Corrado.
Final words of thanks go to my supervisor Professor Keith Stenning who
helped start off the initial sparks of insight that led to this work.
If the words that are said,

lead you to be impressed,
Then pray thee do not be led,
to believe they show,
Anything that I know,
For God taught me all I said.
(Alkhalifa)
Cognitively Informed Systems 1
Chapter I
Cognitively
Informed Systems:
Justifications and
Foundations
Eshaa M. Alkhalifa, University of Bahrain, Bahrain
Abstract
Cognitively informed systems as introduced by Alkhalifa (2005b) is a

perspective that encourages system designers to consider the findings of
cognitive science as informative to the design of their systems. This relies
on an underlying assumption that the presentation, interaction abilities,
as well as the system structure, are likely to achieve more efficient
communication if the design is aligned with the expectations of the human
cognitive machine. In other words, this perspective deals with issues such
as how to best present materials for the perceptual system to isolate the
required differences and focus on the correct points in the image, how to
offer sufficient interaction to enhance learning, or how to elicit different
levels of cognitive engagement with the system. This chapter offers a
survey of the main areas of the field and examples are given of how these
areas can inform particular aspects of future system design. A case study
is also presented as support to this perspective. The main conclusion that
Copyright © 2006, Idea Group Inc. Copying or distributing in print or electronic forms without written
permission of Idea Group Inc. is prohibited.
2 Alkhalifa
can be drawn is that this new perspective is not only practical but also
worthwhile.
Background
Paul Thagard (2004) describes cognitive science as the interdisciplinary study

of mind and intelligence. He indicates that it embraces philosophy, psychology,
artificial intelligence, neuroscience, linguistics, and anthropology. The first fruits
of this science emerged in the mid-1950s when researchers in several different
fields, including George Miller, John McCarthy, Marvin Minsky, and Allen
Newell, started to place the foundations of theories of mind. They started by
founding the field of artificial intelligence and in their endeavor to do so, started
to study the workings of the cognitive system, with the attempt of learning how
it works and modeling that onto a computer system.
The main foundation of the science is that thinking can be understood in terms
of representational structures of the knowledge in the mind, and that compu-
tational procedures may describe all the processing required on those struc-
tures that is necessary to make deductions (Thagard, 2004).
There are two standard computational approaches that are utilized in modeling
any system in cognitive science. The first is described as symbolicism and deals
with symbolic processing where each concept is given a symbol to represent it
and rules are utilized to make deductions based on the values of these symbols.
The second is known as connectionism, where neural networks are used to
represent the structure of the system where each neuron acts as a unit that
interacts with its inputs to produce its outputs (Willaford, 2004). According to
Andy Clark (1993), cognitive science, “sets out to explain the mechanisms
implicated in events which are recognizably psychological in nature, such as
reasoning, planning, and object recognition.”
Consequently, a conclusion that one may arrive at is that this science is rooted
in the philosophy of mind and branched out as a science attempting to
concentrate and analyze the workings of the human mind and/or brain and to
produce computerized models either through symbolic programming or through
connectionist modeling. One question that may arise is, Will this theoretical
science continue with its current target without any subtrack branching out? In
the case of artificial intelligence, expert systems branched out into the world and
found themselves a place in modern-day systems in online help, decision

support systems, and many such applications. The success of expert systems
justifies the target sought by the perspective suggested here in benefiting from
the other findings of that science during the early design phases of hypermedia
systems. However, in order to achieve that benefit, the main areas of possible
contribution have to be isolated, as a novice to the field may be confused by the
diverse directions of work that is currently pursued.
Foundations of
Cognitively Informed Systems
Cognitively informed systems is a term that represents all computerized systems

that carry within their design an assumption or finding made by the science that
seeks to comprehend the cognitive system with the intention of offering a
clearer channel of communication or interaction with the human user who is
likely to access it. The justification for this type of system lies in the fact that
interacting with a human being inherently implies that the human will utilize his
or her cognitive system during that interaction. As an example from tutoring
systems, one may consider that if the topic to be taught is mathematical series,
then the designer of the teaching system should be acquainted with concepts
such as cognitive load, which is defined as the amount of cognitive processing
required to perform an operation. For example, the cognitive load associated
with performing an addition of two numbers is less than that required for
learning how to add. Information of comparisons between different levels of
cognitive load, in addition to the peculiarities of the concept itself, all stem from
the field of cognitive science. Unfortunately, although work that is cognitively
informed exists, it is dispersed, in most cases ill organized, and rarely subjected
to controlled evaluation and assessment. The pioneers include Jonassen
(1991), van Jooligan (1999), as well as Albacete and VanLehn (2000a,
2000b).
Jonassen (1991), for example, advocates the constructivist approach to
learning where students are given several tools to relieve them from repetitive
computation or to externally represent text they are required to recall as is
usually done when writing on paper, in order to allow them to focus on the
learning task at hand. He adopts the assumption originally proposed by Lajoie
4 Alkhalifa
and Derry (1993; Lajoie, 1990) that computers fill the role of cognitive
extensions by performing tasks to support basic thinking requirements, such as
calculating or holding text in memory, which caused them to label computers as
“cognitive tools.” Jonassen’s (1991) central claim is that these tools are offered
to students to lower the cognitive load imposed during the learning process,
which in turn allows them to learn by experimentation and discovery. However,
no experimental evidence was presented to support these claims where
students achieved more learning with these designs. By contrast, van Nimwegen
et al. in Chapter IV of this book offer counter evidence to show that this
reduction does occur but also directly affects what is being learned because
students become dependent on the availability of that support.
Wouter van Jooligan (1999) takes the concept assumed by Jonassen a step
farther through proposing an environment that allows students to hypothesize
and pursue the consequences of their hypotheses. They presented two systems:
the first supports the hypothesis formation step by providing several windows
that help students form their hypotheses and the second provides a formatted
presentation of experiments already tested and their results in a structured
manner. They also added intelligent support to the system by providing
feedback to students to guide their hypothesis formation approach. This
approach supports scaffolding by guiding students toward their target or goal,
but yet again the work was lacking a proper comparative evaluation. However,
Karen Lee shows in Chapter IX that the discussions do end up much more
structured and reflect an increase in knowledge.
Albacete and VanLehn (2000a, 2000b) recognized the cognitive anomaly that
exists between the naive students’ ill-structured knowledge of conceptual
physics and the highly structured knowledge of experts in the field. Conse-
quently their presented system concentrates on teaching students how the
various concepts relate to each other. The evaluation of results exhibited no
significant differences between the learning outcomes of the control group when
compared to the learning outcomes of the experimental group. Albacete and
VanLehn (2000b) then utilized alternative means of analysis to highlight various
differences in learning between the groups. The first was through measuring the
effect size, as done by Bloom (1984), while the second was to compare results
to the nationwide score on a standardized test. The third was to consider how
much students who have different pretest scores learned when compared to
each other. Perhaps the problem that lies here is in the evaluation step of the
results rather than the design of the system because results seemed positive in
some testing environments and no difference emerged in other testing environ-
ments.
Main Areas of Focus in

Cognitively Informed Systems
One may notice that the focus here concentrates on specific areas in the design
of the computerized system. These include the design of the interface of the
system that will interact with the user; they include the logic behind the
operations of the system which will comprehend what the user’s wishes, report
the results desired by the user, or assess various user ability or styles, in addition
to the output mechanism, which includes the modes used to display information
to the user.
A formalization of the main areas of contribution to qualify a system to be
described as a “cognitively informed system” is as follows:
1. Perception and Recognition: The aim is to learn from various findings

along this track how the perceptual system of the person using the
hypermedia is likely to be affected by the presented materials. Questions
that may arise include: Is this likely to convey an implication other than that
intended by the designer? Or in the case of a medical tutoring system: How
can a student’s attention be attracted to a particular part of the image or
scan of a patient? Several theories exist in this field and the following is just
a small sample of some of the findings:
a. The visual system is very organized and it seems to perceive the world
as accurately as possible.
b. Helmhotz (1821–1894), Bruner (1957), Neisser (1967), and Gre-
gory (1972, 1980) all agree to assume that the stimuli is subjected to
a set of hypothesis or postprocessing. This implies that what is
perceived is in a sense interpreted and if this interpretation occurs in
an erroneous fashion, than errors occur.
c. There are basically two research assumptions that investigate how a
complex item is perceived. Marr and Nishihara (1978) argue that the
main axes of an object are utilized to recognize an object. Template
theorists argue that the object that is viewed is viewed as a whole, and
is then compared to several existing templates in memory to be able
to recognize what has been seen. For example, a chair may have a
template or form that any newly seen chair can be compared to.
6 Alkhalifa
Biederman (1987) follows the feature theorist view in that objects are
perceived as a complex object made up of parts. In this case, a chair
is broken up into its basic features, what is the shape of the back, what
are the shapes of the legs, and so forth.
d. Pattern recognition theories do not regard the context in which the
stimulus is presented as influential in the recognition process. They do
not regard the interrelationship between the various objects as
influential either.
e. Bruce and Young (1986) offer convincing evidence that when people
recognize stimuli that is extremely similar to one another as with
recognizing faces, then recognizing a familiar face occurs in a different
fashion from recognizing an unfamiliar face.
f. Bruce and Young (1986) also found that associating a name with a
face is very different from associating information about that person
with a face.
g. Gregory (1970, 1980) indicates that many of the classical visual
illusions occur because we impose the images we see in the three-
dimensional world that we live in onto a two-dimensional image or
representation and this causes an error in interpreting it.
These findings are only representatives of what may influence the design
of cognitively informed systems. In the field of teaching medicine, for
example, it is essential to identify how to highlight the important parts of
the image and to ensure that students acquire the ability to recognize them.
They will be examining images of high similarity and, as identified by Bruce
and Young (1986), may need to carefully review the images they seek to
be able to diagnose the problem by making themselves familiar will all of
its possible variations. Another issue is to avoid visual illusions that cause
the users of the system to be confused by the display.
2. Attention and Memory: Once the issues of perception are resolved, one
may wonder how to direct the system user’s attention to a particular point
of focus. Another goal is to encourage recall of whatever is being
displayed on the screen. It is not in the systems designer’s interest to use
a display plan that is highly likely to cause confusion or to lose the attention
of the user.
a. Theories of attention focus on limitations in the cognitive capacity to

attend to a particular input by explaining possible causes for these
limitations through various approaches (Broadbent, 1958; Deutsch
& Deutsch, 1963; Treisman, 1964).
b. Hampson (1989) identified that both focused and divided attention
are similar in some settings. One of these is when the system offers
different modalities because the difference facilitates the division of
attention. The reason proposed is that the processes that occur for
different modalities are distinct and therefore little interference occurs
between them.
c. Wickens (1984) concludes that tasks interfere with each based upon
the modality used (visual and auditory), the stages required of
processing of each task, and related memory encoding.
d. With respect to memory, a central division has been established
between short-term and long-term memory (James, 1890) where the
term “short-term memory” was eventually replaced by Baddeley and
Hitch (1974) to be “working memory.”
e. Atkinson and Shiffrin (1968) indicate that the working-memory
model is of relevance to activities such as mental arithmetic (Hitch,
1978), verbal reasoning (Hitch & Baddeley, 1976), and comprehen-
sion (Baddeley & Hitch, 1974) in addition to the task of recalling
things from memory.
f. One of the basic assumptions is that a learner who is acquiring a skill
has to recall the instructions as well as execute them by recalling the
given information. For example, someone learning how to drive has
to recall how to drive in addition to paying attention to the road and
other cars. Once this person acquires the skill of driving, recall is
reduced to the road situation because the driving task turns into a
motor activity.
g. Craik and Lockhart (1972) proposed a framework for memory
based upon a number of different levels of processing, ranging from
shallow or physical analysis of a stimulus to deep or semantic analysis.
Depth of analysis is defined as the meaning that is extracted from the
stimulus.
8 Alkhalifa
This is yet another sample that indicates the importance of attention when
displaying information on any screen. Multimedia systems find support
within this domain as they offer a medium that does allow multiple
representations without risking interference. Another issue is that of
memory, and here it is dependent on the main goals of the system and what
the user is expected to recall following its use.
3. Mental Representation of Concepts: Mental representation research
concentrates on how information is represented in memory and how
different similarities and relationships are stored. For example, if the word
“car” is stored as a word, then the visual image of car is stored as an image
or word considering the person would describe it as a car. Is the image
stored as it was seen or altered to a format that is similar to it?
a. The first competition was between the two basic approaches to

represent knowledge assumed a basis of this science: symbolism
versus distributed representations. Neither of the two groups could
offer any clear support that either one or the other is redundant
(Anderson, 1993; Baddeley, 1986; Marschark, Richman, Yuille, &
Hunt, 1987).
b. The second issue is the difference between written and graphical
representations. Kosslyn (1980, 1983) clarifies several basic differ-
ences between the two. The first is that linguistic representation is
made up of symbols represented in words made up of letters, while
pictorial representations have no obvious small components. The
second is that linguistic representation has words that stand for things
they represent while pictures show what they represent graphically
without symbols. The third is that words are organized according to
the rules of grammar and graphical representations are not organized
according to such rules.
c. Issues of interference that adversely affects recall rose within this
domain as is exemplified by the finding made by Baddeley, Grant,
Wight, and Thomson (1975). Subjects were informed of the loca-
tions of digits on a matrix verbally while they were visually tracking a
light moving along a circular track and they were then asked to
reproduce the matrix. Results showed that verbal messages that can
be easily visualized are adversely affected, while complex messages
that cannot be visualized remained unaffected.
Albacete and VanLehn (2000a) attempted to utilize the findings on the

structure of mental representation in physics. They based the teaching
strategy of the “conceptual helper” by comparing the unstructured mental
representation of students of conceptual physics as compared to the highly
structured mental representation of experts. The system, therefore, con-
centrated on helping students find the “links” that connect the domain
concepts to each other. They defined these links as associations that are
classically used in semantics to describe a relationship such as that
between the concepts “parrot” and “birds” because the first belongs to the
category of the second. By doing this, they assumed a symbolic mental
representation which is primarily propositional (see, e.g., Collins &
Quillian, 1969; Rumelhart & Ortony, 1977).
4. Natural-Language Comprehension and Generation: Communication
in an educational setting can only be achieved if both parties arrive at a
common interpretation of the written text in a hypermedia setting. Here the
idea is to give the computer system the ability to communicate with the user
through natural language rather than through imposing restrictions such as
allowing users to select words from drop-down menus or select names
from a list.
a. Comprehension of text or speech involves the analysis of the gram-

matical structure of the sentence (syntax), followed by an analysis of
the meaning (semantics) and an interpretation of the sentence in order
to comprehend the intended meaning. An example is the sentence, “I
will buy a toy train; what will you get him?” A possible context for this
sentence is that it is said about a boy who is having a birthday party.
b. Frazier and Rayner (1982) proposed a garden path model which
earned its name because it can “lead up the garden path” by
ambiguous sentences formed with correct grammar as in, “The horse
raced past the barn fell,” “When Fred eats, food gets thrown,” “Mary
gave the child the dog bit a band-aid,” and “I convinced her children
are noisy.” They used this to support the claim that meaning plays no
role in determining the initial syntactical or grammatical structure
considered by the reader.
c. There is a general agreement that inferences are drawn during
reading. The simplest form is anaphora in which a pronoun or a noun
in a text is to identify with a previously mentioned noun or phrase. For
10 Alkhalifa
example: “Fred sold John his lawn mower, and then he sold him his
garden hose.” It requires an inference to deduce that “he” refers to
Fred rather than to John.
d. Additionally, there is a great deal of work on story comprehension of
which one of the most successful theories was proposed by Kintsch
and van Dijk (1978) where they indicated that story processing
occurs at two levels: the micro structure where the details of the story
are considered at the level of propositions, and the macro structure
level where the edited version of the micro structure is formed. The
generalization that occurs is of particular interest to learning as some
students tend to overlook important details when they generalize
learned texts.
e. Text generation by contrast involves generating language in forms as
close to “natural languages” as possible and this is subject to various
theories. The goals are usually to guide subjects toward self-reflec-
tion and defending their own arguments. The setting usually involves
an intelligent tutoring system that generates the text according to
specific points it notices as in remediation of common errors. In short,
this domain is vast, as it incorporates all the findings made in the study
of “effective communication” and many findings could be used as a
guide.
The primary aim of utilizing these findings is to either generate language or

comprehend language in the most efficient fashion, but it can also be to
make language more comprehensible to the user of the system. Natural-
language generation is no longer a dream and can occur to a limited degree
of accuracy, but natural-language comprehension still faces problems.
5. Reasoning and Deduction: This area concentrates on the reasoning
procedure that humans follow when arriving at any deduction, and there
are two basic tracks that are followed: mental models theory and the
theory of interpretation.
a. Johnson-Laird and Byrne (1993) indicate that deductive reasoning is

a central intellectual ability which is essential: “in order to formulate
plans; to evaluate alternative actions; to determine the consequences
of assumptions and hypotheses; to interpret and formulate instruc-
tions, rules and general principles; to pursue arguments and negotia-
tions; to weigh evidence and to assess data; to decide between
competing theories and to solve problems.” It is these domains that

are affected by attaining an understanding of reasoning.
b. The mental models theory (Johnson-Laird & Byrne, 1991) assumes
that models are formed according to preset criteria such that “truth”
is reinforced. Students therefore dislike assuming false facts unless
they are explicitly stated. In the learning domain this implies that
students are likely to accept presented materials at face value rather
than question what if an exception emerges whenever instruction does
not include concrete examples.
c. The theory of interpretation (Stenning & van Lambalgen, 2004) is
based on the assumption that all tasks presented to students in natural
language are subject to a number of possible interpretations as
dictated by the semantics of the language. Following that, it is quite
possible for reasoning to occur in a logical fashion. This allows
different learners to associate different interpretations with the same
presentation materials if any ambiguities exist and the range of these
presentations can be predicted by the logic of the presented materi-
als. An example perhaps is the work done by Suthers, Weiner,
Connely, and Paolucci (1995), which attempts to impose a particular
“ideal” reasoning structure onto student reasoning.
Although system designers may wonder how this relates to their work as
it is a study of human error, the relationship is there. If a site presents a
product and then indicates its positive points, then its designer wishes the
customer or visitor to make the decision to buy the product. If an explicit
negative exists in the promotion, then this may be cue for the customer’s
cognitive system to decide against buying by raising the possibility of not
getting what is expected. Explicit negatives are one of the reasoning to
increase doubt in a rule as found by Johnson-Laird and Byrne (1993).
6. Cognitive Learner Differences: There are no guarantees that every
human is the same in their thought processes and cognitive styles. In fact,
differences have emerged between genders and even different types of
thinking.
a. Jonassen and Grabowski (1993) give a detailed account of basic

learner differences that are embedded into cognitive learning theory.
These include the differences between a visualizing learner who likes
12 Alkhalifa
to imagine concepts versus the verbalizing learner who likes to learn

through verbal communication.
Al Balooshi and Alkhalifa (2002) showed that cognitive differences of

being a visualizer or a verbalizer does not necessitate that the person does
not learn from the other type of representation. Students of both styles,
when presented with a multimedia representation of the two styles, found
the “extra” presentation reinforcing their learning rather than interfering
with it. Improvements of up to 40% were recorded. Ignorance of these
differences may result in an ineffective design of the educational system.
7. Cognition and Emotion: Freud (1915, 1943) argued that very threaten-
ing or anxiety-provoking material is repressed from gaining access to
conscious awareness and in turn cannot be remembered. Based upon this,
Gilligan and Bower (1984) indicate that recall is best when the mood of
the student at recall matches that at the time of learning. Eysenck (1992)
also argued that the main function of anxiety is to detect an environmental
threat and as a consequence anxiety may affect how widely focused a
student’s attention is. Images in a topic such as medicine should be
informed of these results.
This list contains all major areas of influence where cognitive science can inform
hypermedia system design and evaluation. However, the implementation of this
description into practice can only occur through a case study.
Case Study:
Multimedia Tutoring System
Cognitive information may be consciously made part of the system design and
therefore investigated during the evaluation phase of the system, or it may
inherently exist in a classical design without showing any effect until the
evaluation of that system isolates the aspects that cause the effect obtained.
Several book chapters were offered by Alkhalifa (2005a, 2005b, 2005c, in
press; Al Balooshi & Alkhalifa, 2002; Alkhalifa & Al Balooshi, 2003) in order
to offer support to the perspective presented here and the means of how the
theory can be implemented in actual design.
A concrete example is represented in multimedia educational systems (Alkhalifa,

2005a). The first step was to carefully review all design questions and to
consider which aspects of cognitive learning theory may be informative of the
desired design features. Once the system was built according to preset criteria,
it was subjected to rigorous testing followed by the evaluation stage.
Students were presented with either a classroom lecture in data structures or
a classroom lecture followed by the use of multimedia system or the use of the
multimedia system alone. Analysis of the results reflected that the system limited
their imagination abilities by suggesting examples depicted graphically so that
students did not suggest examples other than those displayed. Conversely, the
classroom lecture allowed them to imagine various different examples from life.
On the other hand, students learned procedures and how things happen from
the system more clearly than from a classroom lecture. Here evaluations
isolated specific aspects of the results that are of concern to the educator, such
as what type of materials is best taught through a particular system, and this
could only be done by offering a framework of evaluation that takes into
consideration cognitive differences between types of knowledge such as
isolating procedural from description knowledge.
Consequently, the work was accomplished in two stages: design and imple-
mentation followed by evaluation.
A Cognitively Informed Design of a Multimedia System
Alkhalifa (in press) presented a mapping of the various decisions that a designer
may need to take and the corresponding areas that may be of relevance to that
decision.
Amount or Complexity of the Media Offered
The decision made here is if more than one media is offered and at what level
of complexity. Issues that may be of relevance are as follows:
• Cognitive load
• Limited attention span
• Interference between different mental representations
14 Alkhalifa
The decision made was to investigate the effects of two media types in
particular, namely, animation and textual representation. The different modali-
ties are not expected to cause any interference except that they do describe the
same subject matter. To avoid dividing attention, full control of the animation
was given to the student to start it, stop it, and repeat it whenever desired.
Cognitive load was considered by allowing the student to take control of the
sequence of lessons as well as requiring only minimal interactivity.
How the screen is partitioned amongst the various media or objects displayed
This decision affects the number of objects or windows displayed at the same
time and the issues of relevance concerning the person’s interactions and
attention given to these windows.
• Perception and recognition

• Attention
The screen was partitioned into exactly two parts, the right-hand side had the
text and the left-hand side had the animation. Attention need not be divided as
the student can start the animation at any time, and colors were uniform for both
partitions.
Delivering Information in Parallel through the Multiple Windows
This decision affects the material displayed within the multiple windows and
how they affect the user of the system with respect to being complementary or
confusing.
• Redundancy could cause interference

• Limited working memory (cognitive-load issues)
• Limited attention span
• Learner difference
The redundancy did exist between the materials, but the representations
differed in that one was purely textual and the other was graphical. Differences
in cognitive styles also existed and were tested to identify any learning
differences between the two types of learners: visualizers and verbalizers.
Use of Colors
This decision affects colors used in the system and how their choice affects the
user of the system by attracting attention, for example:
• Affects attention focus

• Perception of edges to promote recall
Most of the animations were as simple as they could be by including only

outlines of the object representations to promote recall and attract attention.
Use of Animation
This decision affects the use of animation in the system to simulate changes or
transformation.
• Cognitive-load reduction
• Accommodates visualizer/verbalizer learners
Cognitive-load reduction is supposed to occur by actually showing the students

the procedure as it takes place in front of them step by step. Yet at the same
time, the full procedure was described in a textual form in the adjacent window.
Use of Interactivity
This decision affects the interactivity of the system and the choices it allows
users to take to control their use of the system.
• Cognitive-load reduction
• Raises the level of learning objectives
The level of interactivity in this setting was not high. Raising it may cause learners
to interact with the animation window and ignore the textual description, so the
decision to keep it low was taken.
16 Alkhalifa
Aural Media
This decision affects the use of aural cues or reading of materials.
• Speech perception issues such as accent and clarity

• Interference with other media
No aural media was allowed to avoid any external influences.
Verbal Presentation of Material
This decision affects whether textual material appears and natural language
issues such as ambiguity and so forth.
• Clarity of communication
• Accommodates verbal/serialist learners
Verbal descriptions were written in very simple English and sentences were as
clear as possible, especially since the users of the system speak English as a
second language.
Evaluation of Multimedia Systems
The consideration of the presented cognitive areas of relevance will necessitate

an adjustment of the classical evaluation framework to assess the effects of the
decisions made. A three-dimensional evaluation framework of educational
systems may therefore be presented as follows:
1st Dimension: System Architecture
This dimension is concerned with the system’s main modules, their program-
ming complexity, as well as their interactions. Evaluation within this dimension
should be performed in any or all of the following methods:
• Full description of system modules and complete check of interaction

• Expert survey of the system filled by experts or educators
• Student evaluations to consider their perspective of the system
• Architectural design must be based on cognitive science findings rather
than chance
• Everything else concerning the system design such as cost analysis and
portability
2nd Dimension: Educational Impact
This dimension is concerned with assessing the benefits that could be gained by
students when they use the system. Classically, these are done in pre- and
posttests and this is carried on in this framework with more attention given to
detail.
• Students grouped according to their mean grade in a quiz.

• Posttests are used to compare one group with system only and another
classroom only. A third group attends the classroom lecture with the class
group and does a pretest then uses the system before doing a posttest for
comparison with the other two.
• Questions in the pre-/posttests must be mapped to each other to test the
same types of knowledge, mainly consisting of declarative and procedural
knowledge.
• The tests should best be attempted with students who were never exposed
to this material previously to assess their learning rate.
3rd Dimension: Affective Measures
This dimension is mainly concerned with student opinions on the user friendli-
ness of the system and allows them to express any shortcomings in the system.
This could best be done through a survey where students are allowed to add
any comments they wish without restraints.
It should be apparent that this framework could be easily generalized to
evaluate any type of hypermedia system by necessitating the inclusion of
cognitive science findings in the design and then evaluating their effects in
18 Alkhalifa
isolation whenever possible to identify the strengths and weaknesses of that

system.
Results
First of all, student grades were analyzed using the analysis of variance
(ANOVA) test. In order to conduct a significance test, it is necessary to know
the sampling distribution of F given the significance level needed to investigate
the null hypothesis. It must be also mentioned that the range of variation of
averages is given by the standard deviation of the estimated means.
The ANOVA test did indeed show that there is a significant improvement in
group two between the first test which was taken after the lecture and the
second test which was taken after using the system. However, this was not
sufficient to be able to pinpoint the strengths of the system. Therefore, a more
detailed analysis was done of student performance in the individual questions
of test one and test two. Since the questions were mapped onto each other by
design, it was easy to identify significant changes in student grades in a particular
question type for students of group two who responded to similar questions
before and after the use of the system. For example, a highly significant
improvement with F=58 and p<.000 was observed in the question “Using an
example, explain the stack concept and its possible use?” which is an indication
that the use of the system did strongly impact the student understanding of the
concept of a “stack” in a functional manner.
Another point of view is to examine the scores by using the total average, which
is 10.639, which can be approximated to 10.5, which can be used as a border
line. The rest of the scores can then be divided around this line. It was noticed
that the average score of the third group was not high, yet 10 of scores were
above the border line while comparatively 6 scores were above it from the
second group and only 6 of group one, which took the class-only option. This
shows the results of the third group used the multimedia tutoring system alone
and the second group, which had both the classroom lecture and the tutoring
system exposure, to be close. It also underlines how much the second group
improved their test results after taking the CBI and in the same time showing that
the first group had not improved much only with the lecture learning.
These results indicate that the use of the system may introduce a “limiting” effect
that follows the initial introduction to the concepts (Al Balooshi & Alkhalifa,
2002). Classroom lectures introduce students to the concepts allowing them all
the freedom to select all types of applications, which is in some ways
overwhelming. The use of the system, on the other hand, produces a safe haven
to test their ideas and strongly pursue the examples they can imagine, which
helps them arrive at a solid procedural understanding of the concepts. It goes
without saying that such a conclusion would have been impossible to make if
the questions were not purposely set in the shown mapped fashion.
Additionally, students of groups two and three who were exposed to the system
were asked to fill out an evaluation form composed of a series of questions as
proposed by Caffarella (1987). They generally gave ratings of around 4 to 5
on a scale of 0 to 6 with the highest for “The use of graphics, sound, and color
contributes to the student’s achievement of the objectives” and “The user can
control the sequence of topics within the CBI program,” and the lowest score,
which was 3.559, for “The level of difficult is appropriate for you.” Therefore,
it seems that the students in general enjoyed learning through the system
although they found the level of difficulty of the concepts presented challenging.
In addition to all this, three peer experts filled out evaluation forms to rate the
system from an instructor’s point of view and they gave the system an average
rating of 5.33 on the same scale of 0 to 6.
Conclusion
The main conclusion made here is that there exists a science that may inform the
design of a computer system that interacts with the human’s cognitive system
and the suggestion made is to apply the findings to form a clearer channel of
communication. The claim that human minds work by representation and
computation is an empirical conjecture and therefore may not necessarily be a
correct assumption. In fact, John Searle (1992) has claimed that this approach
is fundamentally mistaken. Other challenges exist including, for example, that
human thought is affected by social interactions or that the mind is dynamic in
nature rather than computational. Thagard (1996) indicates that the science
exists and is currently expanding to include many different representations and
seeks to meet these challenges. Consequently, a conclusion that can be made
here is that the question that is posed for cognitively informed systems is not
whether the cognitive system is in fact computations, but rather what types of
interacts produce which outcomes. When the question is reformulated in this
20 Alkhalifa
way, the path of cognitively informed system differs significantly from that of
cognitive science in that it no longer seeks to fully comprehend what really goes
on, so as information about real-world cognitive reactions are monitored and
utilized in predicting reactions, thus influencing system design. By this alteration
of the research goal, the target becomes practical and assessable.
References
Albacete, P. L., & VanLehn, K. (2000a). The conceptual helper: An intelligent

tutoring system for teaching fundamental physics concepts. In Gauthier,
Frasson, & VanLehn (Eds.), Intelligent tutoring systems: 5th interna-
tional conference (pp. 564–573). Berlin: Springer.
Albacete, P.L., & VanLehn, K. (2000b). Evaluating the effectiveness of a
cognitive tutor for fundamental physics concepts. In L. R. Gleitman & A.
K. Joshi (Eds.), Proceedings of the 22nd Annual Meeting of the
Cognitive Science Society (pp. 25–30). Mahwah, NJ: Erlbaum.
Al Balooshi, F., & Alkhalifa, E. M. (2002). Multi-modality as a cognitive tool,
special issue: Innovations in learning technology. Journal of Interna-
tional Forum of Educational Technology and Society, 5(4), 49–55.
Alkhalifa, E. M. (2005a). Multimedia evaluations based on cognitive science
findings. Encyclopedia of information science and technology, Vol. IV
(pp. 2058–2062). Hershey, PA: Idea Group Reference.
Alkhalifa, E. M. (2005b). Effects of the cognitive level of thought on learning
complex material. Educational Technology & Society, IFETS, IEEE,
8(2), 40–53.
Alkhalifa, E. M. (2005c). Open student models: Mirror modeler. The encyclo-
pedia of distance learning, teaching, technologies, and applications.
Hershey, PA: Idea Group Reference.
Alkhalifa, E. M. (in press). Cognitively informed multimedia interface design.
The encyclopedia of human computer interaction. Hershey, PA: Idea
Group Reference.
Alkhalifa, E. M., & Al Balooshi, F. (2003). A 3-dimensional framework for
evaluating multimedia educational software. In F. Al Balooshi (Ed.),
Virtual education: Cases in learning & teaching technologies. Hershey,
PA: Idea Group Inc.
Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Lawrence Erlbaum.

Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system
and its control processes. In K. W. Spence & J. T. Spence (Eds.), The
psychology of learning and motivation (Vol. 2). London: Academic
Press.
Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press.
Baddeley, A. D., Grant, S., Wight, E., & Thomson, N. (1975). Imagery and
visual working memory. In P. M. A. Rabbitt & S. Dornic (Eds.),
Attention and performance (Vol. 5). London: Academic Press.
Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. H. Bower
(Ed.), The psychology of learning and motivation (Vol. 8). London:
Academic Press.
Biederman, I. (1987). Recognition by components: A theory of human image
understanding. Psychological Review, 94, 115–147.
Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group
instruction as effective as one-to-one tutoring. Educational Researcher,
13, 4–16.
Broadbent, D. E. (1958). Perception and communication. Oxford: Pergamon.
Bruce, V., & Young, A. W. (1986). Understanding face recognition. British
Journal of Psychology, 77, 305–327.
Bruner, J. S. (1957). On perceptual readiness. Psychological Review, 64,
123–152.
Caffarella, E. P. (1987). Evaluating the new generation of computer-based
instructional software. Educational Technology, 27(4), 19–24.
Clark, A. (1993). Associative engines: Connectionism, concepts, and
representational change. Cambridge, MA: MIT Press.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic
memory. Journal of Verbal Learning and Verbal Behavior, 8, 240–
248.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework
for memory research. Journal of Verbal Learning and Verbal Behav-
ior, 11, 671–684.
Deutsch, J. A., & Deutsch, D. (1963). Attention: Some theoretical consider-
ations. Psychological Review, 70, 80–90.
22 Alkhalifa
Frazier, L., & Rayner, K. (1982). Making and correcting errors during
sentence comprehension: Eye movements in the analysis of structurally
ambiguous sentences. Cognitive Psychology, 14, 178–210.
Freud, S. (1915). Repression. In Freud’s collected papers (Vol. 4). London:
Hogarth.
Freud, S. (1943). A general introduction to psychoanalysis. New York:
Garden City.
Gilligan, S. G., & Bower, G. H. (1984). Cognitive consequences of emotional
arousal. In C. Izard, J. Kagen, & R. Zajonc (Eds.), Emotions, cognition,
and behaviour). New York: Cambridge University Press.
Gregory, R. L. (1970). The intelligent eye. New York: McGraw Hill.
Gregory, R. L. (1972, June 23). Seeing as thinking. Times Literary Supple-
ment.
Gregory, R. L. (1980). Perceptions as hypotheses. Philosophical Transac-
tions of the Royal Society of London, Series B, 290, 181–197.
Hampson, P. J. (1989). Aspects of attention and cognitive science. The Irish
Journal of Psychology, 10, 261–275.
Helmholtz, H. von (1962). Treatise on physiological optics (vol. III). New
York: Dover. (Original work published 1866.)
Hitch, G. J. (1978). The role of short-term working memory in mental
arithmetic. Cognitive Psychology, 10, 302–323.
Hitch, G. J., & Baddeley, A. D. (1976). Verbal reasoning and working
memory. Quarterly Journal of Experimental Psychology, 28, 603–
621.
James, W. (1890). Principles of psychology. New York: Holt.
Johnson-Laird, P. N., & Byrne, R. M. J. (1991). Deduction. Hove, UK:
Lawrence Erlbaum.
Johnson-Laird, P. N., & Byrne, R. M. J. (1993). Models and deductive
rationality. In K. I. Manktelow & D. E. Over (Eds.), Rationality:
Psychological and philosophical perspectives. London: Routledge.
Jonassen, D. H. (1991). Objectivism vs. constructivism: Do we need a new
philosophical paradigm shift? Educational Technology: Research &
Development, 39(3).
Jonassen, D. H., & Grabowski, B. L. (1993). Handbook of individual
differences: Learning and instruction. Lawrence Erlbaum.
Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension
and production. Psychological Review, 85, 363–394.
Kosslyn, S. M. (1980). Image and mind. Cambridge, MA: Harvard Univer-
sity Press.
Kosslyn, S. M. (1983). Ghosts in the mind’s machine: Creating and using
images in the brain. New York: Norton.
Lajoie, S. P. (1990, April 16–20). Computer environments as cognitive
tools for enhancing mental models. Paper presented at the Annual
Meeting of the American Educational Research Association, Boston.
Lajoie, S., & Derry, S. (1993). Computers as cognitive tools. Hillsdale, NJ:
Lawrence Erlbaum.
Marr, D., & Nishihara, K. (1978). Representation and recognition of the
spatial organisation of three dimensional shapes. Philosophical Transac-
tions of the Royal Society (London), B200, 269–294.
Marschark, M., Richman, C. L., Yuille, J. C., & Hunt, R. R. (1987). The role
of imagery in memory: On shared and distinctive information. Psychologi-
cal Bulletin, 102, 28–41.
Neisser, U. (1967). Cognitive psychology. New York: Appleton-Century-
Crofts.
Rumelhart, D. E., & Ortony, A. (1977). The representation of knowledge in
memory. In R. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.),
Schooling and the acquisition of knowledge. Hillsdale, NJ: Lawrence
Erlbaum.
Searle, J. (1992). The rediscovery of the mind. Cambridge, MA: MIT Press.
Stenning, K., & van Lambalgen, M. (2004). A little logic goes a long way:
Basing experiment on semantic theory in the cognitive science of condi-
tional reasoning. Cognitive Science, 28(4).
Suthers, D., Weiner, A., Connely, J., & Paolucci, M. (1995). Belvedere:
Engaging students in critical discussion of science and public policy issues.
Proceedings of AI-Ed 95, the 7th World Conference on Artificial
Intelligence in Education (pp. 266–273).
Sweller, J. (1994). Cognitive load theory, learning difficulty and instructional
design. Learning and Instruction, 4, 295–312.
Thagard, P. (1996). Mind: Introduction to cognitive science. Cambridge,
MA: MIT Press.
24 Alkhalifa
Thagard, P. (2004). Cognitive science. Stanford Encyclopedia of Philoso-

phy. Retrieved from http://plato.stanford.edu/entries/cognitive-science/
Treisman, A. M. (1964). Verbal cues, language, and meaning in selective
attention. American Journal of Psychology, 77, 206–219.
Van Jooligen, W. (1999). Cognitive tools for discovery learning. Interna-
tional Journal of Artificial Intelligence, 10, 385–397.
Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman
& D. R. Davies (Eds.), Varieties of attention. London: Academic Press.
Willaford, W. (2004). Cognitive science. In C. Eliasmith (Ed.), Dictionary of
philosophy of mind. Retrieved from http://philosophy.uwaterloo.ca/
MindDict/
Section I
Perception,
Memory, and Recall
26 Chambel, Zahn, & Finke
Chapter II
Hypervideo
and Cognition:
Designing Video-Based
Hypermedia for Individual
Learning and Collaborative
Knowledge Building
Teresa Chambel, University of Lisbon, Portugal
Carmen Zahn, Knowledge Media Research Center, Germany
Matthias Finke, Computer Graphics Center, Germany
Abstract
This chapter discusses how advanced digital video technologies, such as

hypervideo, can be used to broaden the spectrum of meaningful learning
activities. Hypervideo is conceptualized as the true integration of video
into nonlinear information structures by means of spatio-temporal links.
Based on cognitive-psychological perspectives, the discussion focuses on
the way cognitive and socio-cognitive processes relate to the specific
characteristics of hyperlinked videos, and how they inform their design.
Then, with regard to technology, two approaches are introduced, providing
tools for knowledge building and interaction with nonlinear information
structures based on dynamic video information. Case studies and research
findings are presented and prospects for future research are outlined.
Hypervideo and Cognition 27
Introduction
New technologies do not only meet existing needs in terms of communication

and learning practice, but they can also redefine our educational culture by
enabling new learning experiences in resource-rich learning environments
(Beichner, 1994). For example, the advent of video technology, including both
analog and advanced digital video, has substantially altered some of our
traditional paradigms of educational practice in schools and higher education.
Film and video technologies can be used to enrich regular lessons and lectures
with dynamic visualizations of knowledge that foster a better understanding, to
depict concrete real-world problems or cases in authentic ways, or to conduct
video projects, a specific kind of media project where students engage in active
video production in a motivating and authentic collaborative task (Baake,
1999). However, by itself, video provides a limited support for reflection and
it is difficult to relate it to other materials and activities in learning environments.
Hypervideo technology, which refers to the integration of video in hypermedia
structures, can provide the additional means to augment video educational
capabilities, contributing to learning in several distinct ways: as a presentation
medium, it can support self-regulated cognitive processing of dynamic visual-
izations; as a nonlinear and interactive medium, it allows for interactive learning
as well as for reflective and elaborative knowledge building individually or in
group (Chambel 2003; Chambel & Guimarães, 2002; Guimarães, Chambel, &
Bidarra, 2000; Zahn, Barquero,& Schwan, 2004; Zahn & Finke, 2003; Zahn,
Schwan, & Barquero, 2002). These ideas, their underlying assumptions, and
the mechanisms for the design and realization of systems that support them in
learning contexts will be discussed in more detail in the following sections.
What is Hypervideo?
The term “hypervideo” reflects the idea of true integration of video in hypermedia
spaces, where it is not regarded as a mere illustration, but can also be structured
through links defined by spatial and temporal dimensions (Chambel, Correia,
& Guimarães, 2001; Chambel & Guimarães, 2002). Hypervideo structures
may also be defined as a combination of interactive video and hypertext, as they
consist of interconnected video scenes that may further be linked with addi-
tional information elements, such as text, photos, graphics, audio, or other

videos in the hypermedia space (Zahn et al., 2002).
The roots of hypervideo structures lie in the early days of hypertext, when Ted
Nelson extended his hypermedia model to include “branching movies” or
“hyperfilms” (Nelson, 1974). However, technology has been slow in bringing
these ideas to full realization (Chambel et al., 2001; Chambel & Guimarães,
2002). HyperCafe (Sawhney, Balcom, & Smith, 1996) is one of the earliest
hypervideos, featuring digital video and revisiting hypermedia concepts in this
scenario. Since then, different levels and types of video integration in hypermedia
have evolved (Zahn et al., 2002). For example, regarding the media types that
are involved in the hypervideo, we might differentiate between:
• Homogeneous hypervideo, where video is the only medium involved,

consisting of dynamic audiovisual information presented as a continuous
stream of moving pictures that can be navigated by the user;
• Heterogeneous hypervideo that integrates other media, providing fur-
ther and related information to the video, or having video illustrate and
complement it. For this broader perspective, the name of “video-based
hypermedia” or “hyperlinked video” (Chambel & Guimarães, 2002; Zahn
& Finke, 2003) is sometimes adopted.
We might also differentiate between different types of hypervideo with regard

to their structure and navigational options:
• Video nodes may be structured in a network-like hypervideo, where a

substantial number of short video scenes are linked together to be freely
navigated by the users, as is exemplified in “HyperCafe” (Sawhney et al.,
1996).
• A linear film may be divided into single scenes, according to different
themes that can be navigated as thematic paths in the hypervideo.
Depending on the theme specified, different sequences of the film’s scenes
can be arranged and selected by the users to be viewed in succession. By
following different thematic perspectives, viewers are assumed to develop
a more flexible mental representation of the structure and content of the
film. This concept is exemplified in a well-known hypervideo tutorial for
the interpretation of Orson Welles’ film Citizen Kane described by Spiro
and Jehng (1990).
• Another type of hypervideo can be described as a film supplemented by

multimedia “footnotes.” Basically, a “main” film is presented in its
original form (i.e., in linear sequence), but contains dynamic hyperlinks
attached to visual objects within the video that branch out to additional
information elements, such as a text, an image, or another video clip. After
having visited the link destination, the users get back to the main video and
may continue watching it as before. This type is very similar to hierarchical
hypertext.
Hypervideo shares with classical hypertexts the characteristic of being struc-

tured in nonlinear ways according to different patterns, offering the users
opportunities of taking different “routes” through learning materials and learning
processes. This cognitive dimension in the design and use of hypervideo is the
main focus of the next sections.
Learning with Hypervideo
An effective design of tools and environments that support learning requires the
understanding of human cognition and learning processes. This section presents
the main cognitive concepts relevant for discussing video and hypervideo as
supporting tools for learning.
Cognitive Modes, Learning Phases, and Learning Styles
Norman (1993) identifies two Cognitive Modes: the experiential mode

relates to a state in which we perceive and react to events in an effortless way,
it is about perception and motivation, and good for accretion of facts and tuning
of skills; the reflective mode relates to comparison and contrast, thought and
decision making, essential for restructuring of knowledge. Both are important
in human cognition, but they require different kinds of technological support.
In addition to cognitive modes, different Learning Phases have been identified
for the learning process. The classic learner-centered pedagogy model has
three phases:
1. Conceptualization of the subject and its domain;

2. Construction, where the learner actively engages with the subject, while
relating to his/her own knowledge framework;
3. Dialogue, where the learner expresses aspects of the emerging under-
standing and relates this to the understandings of fellow learners and
tutors.
Besides different cognitive modes and learning phases in individual learning,

people also develop different Learning Styles, or cognitive preferences, that
determine the ways of learning best suited to them. There are many theories,
models, and instruments to determine learning styles, but they are all essentially
based on the idea that individuals perceive, organize, or process information
differently (Chambel & Guimarães, 2005). Examples of these theories include
the VARK Perceptual Learning Styles (Fleming, 1995), distinguishing four
styles: visual, aural, read-write, and kinesthetic; the Kolb’s Learning Styles
Inventory (Kolb, 1984), identifying four styles: reflector, pragmatist, theorist,
and activist; and the Howard Gardner’s Theory on Multiple Intelligences
(Gardner, 1983), identifying eight intelligences: verbal-linguistic, logical-math-
ematical, visual-spatial, musical-rhythmic, bodily-kinesthetic, interpersonal,
intrapersonal, and naturalist.
This differentiation suggests a need for a flexible support of different styles. An
ideal learning environment would support all the learning styles, with the
flexibility to allow each learner to spend more time on his/her preferred style,
and induce the development of skills in nondominant styles.
It is important to note in this context that not only different individuals, but also
possible interactions between different individuals in learning groups might be
considered. The dialogue phase and some learning styles, like the one under-
lying the interpersonal intelligence, already refer to this interactive dimension,
but learning in groups involves more specific aspects that will be addressed in
the next section.
Learning in Groups: Group Learning
Although learning in the long run is always based on individual cognitive

processing, it is at the same time situated, process oriented, and related to
social activity (Salomon, 1993). Many theorists of educational psychology and
pedagogy therefore argue that intelligence is not an individual property but
distributed within socio-technical systems (Pea, 1993) and that most learning
occurs within a framework of knowledge communication and knowledge-
related cooperative and collaborative action (Salomon, 1993; Scardamalia,
2004). Such a framework can be provided, for example, by collaborative
problem-solving tasks including collaborative activities such as writing texts or
editing hypertext and multimedia (Beichner, 1994; Scardamalia, 2004; Stahl,
2002).
The educational value of such collaborative tasks may be seen on both a
motivational and a socio-cognitive level. On the motivational level, the
experience of solving a complex problem or designing any kind of product in
collaboration with others (peers, teachers, etc.) and thereby using a modern
and culturally extended technology (computers, software, authoring tools, and
video) can promote a feeling of importance (Carver, Lehrer, Connell, &
Erickson, 1992) and improve the self-conceptions of learners (Lehrer, 1993).
It may also particularly increase feelings of becoming a competent member of
a “community of practice” (e.g., Penuel, Korbak, Cole, & Jump, 1999).
On the socio-cognitive level, collaborative tasks serve as a setting where
individual knowledge interacts with group knowledge. Applying Salomon’s
(1993) spiral interaction model, we can assume that repeated interactions
between individual knowledge and group knowledge during discussions and
discourse steadily lead to higher levels of knowledge related to both individual
cognition and to the knowledge resources of the group. The basic argument
underlying such positive expectations derives from developmental psychology,
where individual cognitive development is generally assumed to be facilitated
most where it naturally occurs from the very beginning of life, that is, during the
social interaction with significant others and during peer interactions (e.g.,
Vygotsky, 1978). Or in terms of contemporary computer-supported collabo-
rative learning (CSCL) theory, individual knowledge develops best within
group knowledge processes involving both socio-cognitive processes and
cultural artifacts (Stahl, 2002).
Group knowledge is also referred to as “shared knowledge” or “common
ground” (e.g., Baker, Hansen, Joiner, & Traum, 1999). In contrast to individual
knowledge, group knowledge must be identified, negotiated upon, and ex-
pressed in the form of shared information during different phases in
collaborative knowledge acquisition. Group knowledge is developed by learn-
ers acting collaboratively on shared information such as texts, images, or even
dynamic videos or animations. Activity contexts for interactions between
learners should be provided for groups to develop this common ground and to
express their shared knowledge in a shared information environment.
After having outlined the general arguments in favor of flexible support for
learning, we will consider in the next section why and when using dynamic visual
materials and video, in particular, might be a good choice.
Video as a Cognitive Tool
There are a number of topics or problems that can hardly be understood

without using dynamic visual materials as a referential basis. Imagine, for
example, geography students exploring the formation of a thunderstorm
(Mayer, 2001), or a group of school children trying to understand Newton’s
laws in their physics class. In some learning situations, videos or animations are
not only a desirable but an important prerequisite for successful learning to take
place. From a cognitive perspective, audiovisual materials support learning:
• By “replacing” real experience, because of their authenticity and

realism, which evoke feelings of “observing real situations” (Schwan,
2000). Concrete real-world problems or cases can be depicted in
authentic ways and then related to more abstract knowledge and prob-
lem-solving skills. This is illustrated, for example, by the famous “Jasper
Woodbury Series,” a set of interactive videos developed by the Vanderbilt
University in the late 1980s and 1990s for complex mathematics problem
solving (Cogntion and Technology Group at Vanderbuilt University,
1997). Here, video is supposed to help to situate knowledge for the
purpose of “anchored instruction.” It could be shown in an experiment that
those groups of students who were asked (a) to pose their own subordi-
nate questions while working with the video and (b) to self-dependently
find the relevant information to answer these questions in a video episode,
outperformed other groups of students who just viewed the video episode
and received general text-based information on problem solving unrelated
to the video (Van Haneghan, Barron, Young, Williams, Vye, & Bransford,
1992).
• By visualizing dynamic processes, which might not be observable in
reality or which are hard to describe verbally (Park & Hopkins, 1993).
Empirical findings on learning with video media consistently show that
audiovisual presentation formats facilitate the comprehension and transfer
of knowledge, especially in those domains where dynamic processes and
concrete objects or complex systems need to be observable for a proper
understanding of the topic (for overview, see Wetzel, Radtke, & Stern,
1994; Park & Hopkins, 1993).
• By combining diverse symbol systems, such as pictures, texts, and
narration, into coherent media messages (Mayer, 2001). The specific
qualities of video presentations are supposed to support the construction
of rich mental representations and, by dual coding (Paivio, 1986; Mayer,
2001), improve the transfer of knowledge.
• Through the conducting of “video projects,” where learners engage in
active video production, relying on an idea sometimes described as
“learning by design” (Reinmann & Zumbach, 2001) or “project-based
learning” (Baake, 1999; Bereiter, 2002). Video is not only used to present
information or situate a problem to be solved; creating video artifacts is the
problem to be solved.
In accordance with these assumptions, empirical findings have consistently

shown that dynamic media facilitate the comprehension and transfer of knowl-
edge in individual learning (Park & Hopkins, 1993; Mayer, 2001). In specific
collaborative scenarios, video can also be considered supportive for cogni-
tive processing. For reasons similar to those in the case of individual learning,
video is helpful when meaningful collaboration depends on visual perceptions
of concrete objects, actions, or complex relations; and when knowledge is
created within networked groups, where learners do not meet in the same place
at the same time and, hence, cannot observe the same things in the same
situation. In these cases, video might support mutual understanding by acting as
a referential anchor for collaborative activities.
To summarize, video, as dynamic and figurative information combined with
verbal audio, forms a powerful means of communicating meaningful scenarios
rapidly and efficiently (LiestØl, 1999; Paivio, 1986). It can bring context to
topics and enhance the authenticity of a computer-based learning environment,
thus fostering what Norman (1993) called an experiential cognitive mode.
However, to allow reflection, a system must have a medium that affords adding,
modifying and manipulating representations, and performing comparisons. It
must also afford time for reflection, elaboration, and comparison processes.
Broadcast television and most videos are usually watched in an experiential
mode and cannot augment human reflection in this sense (Norman, 1993).
According to Eco (1979), the lack of communication and debate is also an
important drawback in television teaching capabilities. He claims that learning
should take place in a broader context where discussion could happen.
The Role of Hypervideo
Television and video could also be a powerful tool for reflection, if designed in
a way that would allow the viewer to select what to watch, to control the pace
of the information flow, to stop and make annotations, and to relate to other
materials or to other people’s points of view (Correia & Chambel, 1999;
Norman, 1993; Zahn et al., 2004).
Hypervideo technology can provide this kind of support to the different learning
modes, phases, and styles, and integrate an environment that allows the
communication and collaboration among learners, teachers, and other experts
on the subjects at hand (Chambel & Guimarães, 2002; Correia & Chambel,
1999; Guimarães et al., 2000; Zahn et al., 2002; Zahn & Finke, 2003).
Hypervideo might promote learning in its main modes and phases for different
individuals by providing and supporting:
1. Interactive access to rich audiovisual information, contextualized in

video-based hypermedia spaces;
2. Construction of knowledge in a sense of relating concepts and expand-
ing on them, using annotations and cognitive maps;
3. Communication, allowing the debate of ideas, the exchange and sharing
of information and knowledge, or the collaborative elaboration of the
previous hypermedia spaces.
By allowing the viewer to watch video in his/her natural experiential mode and
by inducing and supporting more active and reflective attitudes through control,
comparison, and annotations, hypervideo can support both cognitive modes
suggested by Norman (1993). Its ability to integrate heterogeneous media and
activities also allows the support of diverse learning styles and the interaction
between different learners.
However, all this power is accompanied by potential complexity and may not
be used to its best in all cases, especially for learners and authors with little
experience and background knowledge. Hence, design guidelines play an
essential role in the authoring of effective hypervideo spaces (Chambel &
Guimarães, 2002; Zahn et al., 2002).
Design for Learning with Hypervideo
Hypervideo shares with traditional hypermedia the potential of an increased

cognitive load that might also lead to disorientation (Conklin, 1987; Nielsen,
1995). But in the case of hypervideo, this kind of problem might be even more
pronounced. Due to its richness, video itself sometimes carries with it the risk
of overstraining the cognitive capacities of the learner; and the dynamic nature
of nodes and links may put time pressure on the users, when they are required
to make navigational decisions (Zahn et al., 2002). The integration of a dynamic
medium, like video, with static media, such as text and images, also raises
important rhetorical and aesthetic challenges to hypervideo, since they induce
different attitudes in the user (Chambel & Guimarães, 2002; LiestØl, 1994;
Sawhney et al., 1996). It is important to avoid discontinuity, especially when
navigating from a passive video watching to an active text-reading experience.
The effective design of hypervideo spaces may greatly benefit from following
guidelines that help to face the extra challenges video brings to the scenario, and
from an adequate support from the underlying systems that bring these
hyperspaces to life, as exemplified in the next two sections. Many of the
mechanisms adopted for hypermedia need to be extended for hypervideo in
order to accommodate its increased complexity, and should address the
provision of (Chambel & Guimarães, 2002)
• Control, to be able to navigate the videos and the hyperspace, for

example, extra mechanisms have to be available in order to provide users
with information about the existence (where, when, for how long) of links
on the video — “link awareness” (Chambel et al., 2001), a more complex
issue in hypervideo, since video changes in time;
• Consistency and coherence, in terms of structure, interface, and naviga-
tion to reduce cognitive load (Zahn et al., 2002);
• Context for orientation purposes, for example, through the synchroniza-
tion of video with navigation maps;
• Familiarity, for example, through the adoption of metaphors, such as
television, books, and traveling; and
• Continuity, especially when navigating between dynamic and static
media, for a sense of unity and coherence.
Thus, it is a basic necessity to think carefully about how to adapt technology to

the cognitive prerequisites of potential hypervideo users, thereby setting the
groundwork for effective and satisfying learning to take place. The design
approaches specified in the next sections encompass the following: (1) infor-
mation is mainly presented by audiovisual media, (2) knowledge can be created
collaboratively on the basis of video presentations, by means of both linking
information and annotating, and (3) the process of knowledge building is
reflected in resulting visualizations in the hypervideo structure.
Hypervideo in HTIMEL
To explore the use of hypervideo in learning environments, we developed the

HTIMEL (HyperText Markup Language with Time Extensions) model and
language, as an extension to HTML and existing Web tools (Chambel et al.,
2001). The temporal dimension was considered to allow the addressing of
video in space and time, for the definition of link anchors, and to synchronize
media elements. New forms of integration, annotation, and navigation of video
in hypermedia were conceived, with a special concern to the support of
cognitive processes. These tools were used to create course material, mainly
in literature and mathematics, following our ideas about the way video should
be integrated and augmented to support learning (Chambel & Guimarães,
2002). Case studies were developed at the University of Lisbon, in collabora-
tion with the Portuguese Open University, in the context of literature distance
learning, and with the Center for Mathematics and Fundamental Applications
at the University of Lisbon, the California Institute of Technology, and the
Technical University of Berlin, for the communication and learning of math-
ematics.
Figure 1 illustrates one of our case studies. The information is organized around
text-based and video-based pages that closely relate to the book and video-
tapes that were originally used in the course. Through their integration in
hypervideo, relations among them could be fully captured, maintaining the
original context. In text-centered pages (Figure 1a), video is integrated as an
illustration that is played when the hypervideo links defined in the text are
followed. While keeping some familiarity with the book metaphor, the user has
an augmented experience of text reading.
Figure 1. Hypervideo example in HTIMEL
The user may focus on the videos through video-centered pages (Figures 1b,
c, d), where the whole video can be played. Indexes, synchronized with the
video, make the video structure explicit, provide for user orientation, and can
be used to navigate the video, handing control to the user. Different types of
indexes act as different views or maps of the video. For example, a table of
contents (Figure 1b) represents the video structure, whereas an image map
(Figures 1c, d) is a visual summary of the video; a cognitive map represents its
knowledge structure (Guimarães et al., 2000), while user annotations (Correia
& Chambel, 1999) capture a personal view of the video (Figures 2c, d), and
exercise maps (Chambel & Guimarães, 2002) relate activities and quests to
video content. In video-centered pages, while keeping some familiarity with the
TV/video metaphor, the user has an augmented experience of video watching.
Links can also be defined among different parts of the video, addressed in
space and time, allowing the user to navigate it through related information
(Figure 2a–b), or among portions of video and text (Figure 1d–a), comple-
Figure 2. HTIMEL (a–b) spatio-temporal link from video to video, and

navigation history; (c) annotations map; (d) insertion of an image
annotation to video
menting or contextualizing the information conveyed by each of them. Another

type of video linking is achieved through video montage. Authors can assemble
video segments to make personalized versions of their favorite movies; and a
teacher can make available portions of a video, in a particular order, to illustrate
concepts in a context where the original video would not be so efficient or
concise. Thus, it provides the means to construct hypervideo paths.
Navigation history is recorded and lets the user go backward and forward
along visited nodes in the hyperspace (Figure 2b–a). When following from
video centered to text-centered pages, video or audio may provide context and
smooth the transition from an experiential to a reflective medium. This was
adopted when accessing exercises from a video on the Story of Pi. When
arriving at the text-centered page, the portion of the video introducing the solid
the exercise refers to is played, promoting continuity in navigation.
Navigation can also be made through a timeline, positioned beneath the videos
in the presented examples. It provides direct access to any position on the
video, in a continuum, and in this sense it contrasts to video indexes that give
access to specific positions in a discrete space. HTIMEL timelines also play a
part in link awareness, representing the source and/or destination time interval,
when the cursor is over any link that involves the video in the context. Besides,
while the cursor is over the timeline, all the context changes, reflecting what
would happen if the video was to play from that position, providing for context
awareness. Other forms of link awareness mechanisms are discussed in
(Chambel, 2003; Chambel & Guimarães, 2002). Being developed for the
Web, these hypermedia spaces allow an easy integration of communication
mechanisms. Students may then create their own versions of hypervideo
documents, expand upon the original ones, through annotation, and share them
with colleagues and teachers.
From our studies and experiences, we conclude that students are more
motivated to watch the videos in this type of hypervideo than in traditional
settings, as the process becomes more flexible and engaging; and it is easier to
search for information and to capture the videos’ messages through the different
maps available. Video course material, as rich as it is, is better used in a
reflective mode if presented in a way where the content or knowledge structure
is made explicit. The full integration of video in hypermedia also allows the
capturing of important relations between video and other media, like text,
through contextualized explanations and illustrations that promote deeper
understandings of the different materials. Different learning styles are also
Figure 3. HTIMEL code for video with a spatio-temporal link

<DIV ClassName="HTIMEL_Video" ID="video1"
activeColor="#EAAB4D" inactiveColor="#993347"
spotColor="#FFFFFF">
<OBJECT CLASSID="CLSID:05589FA1-C356-11CE-BF01-00AA0055595A">
<PARAM NAME="FileName" VALUE="E:HPi.mpg">
</OBJECT>
<MAP name="map1" orgT1="226" orgT2="230"> 
<AREA shape="rect" coords="132, 131, 238, 242"
ClassName="HTIMEL_link" destT1="1455" dest="video1">

</MAP>
...
</DIV>
supported, through the integration of different media and perceptual modalities,

interactive, and navigational choices (Chambel & Guimarães, 2002; Guimarães
et al., 2000).
From a technical perspective, HTIMEL is an extension of HTML. New
elements and attributes were defined, and their functionality is supported by a
set of scripts that are generic, reusable, and almost transparent to the author of
the hyperdocument. The authoring process is done in a declarative way in
HTIMEL. The current prototype is based on Dynamic HTML and uses
VBScript language and ActiveMovie technology. Scripts deal with these
language extensions, acting as a browser extension, in order to make the
declarative authoring possible, without the need for plug-ins or the develop-
ment of a different browser. Figure 3 exemplifies the definition of a spatio-
temporal link from video to video, like the one presented in Figure 2a–b.
The adoption of a declarative format allows the externalization of documents’
behavior and provides for maximum portability and reuse of created docu-
ments. While programming-based approaches may offer some performance
advantages in the short run, a declarative approach provides wider access to
quality information with less author effort. It contributes to easier authorship,
either using a simple text editor or through automated production. Open and
flexible production frameworks that use video and audio processing tech-
niques, for segmentation and indexing, were conceived to automate the
authoring of hyperdocuments with markup languages like HTIMEL (Chambel,
2003). The image map presented in Figures 1c and d and the alignment of the
video with the text transcriptions of its audio, in text maps, were constructed
this way. From both cognitive and technical perspectives, HTIMEL has had
good results as a proof of concept approach.
Dynamic Information Spaces (DIS)
To explore the use of hypervideos in collaborative learning scenarios, the

Hypervideo System presented in the following paragraphs was developed.
Here, hypervideo is conceptualized as a complex dynamic information space
(DIS), where learners may add their own materials to selected video objects
and make specific annotations, while collaboratively expanding their knowl-
edge on the topic at hand. The general approach is aligned with considerations
of cognitive psychology and theories of CSCL (Stahl, 2002).
Thereby, the system provides specific facilities to jointly elaborate on video

materials and to change a hypervideo presentation according to the develop-
ment of knowledge present in any group. Accordingly, a hypervideo document
can be changed and extended as a basis to share knowledge and to commu-
nicate within a community. The underlying model of the dynamic information
space defines three categories of content types (annotated video, additional
information, and communication contribution) as separate nodes. Video se-
quences, which contain sensitive regions, are denoted as annotated videos.
Additional information, which is linked to objects in the annotated videos, can
be of any kind of multimedia content (text, pictures, graphics, etc.). Commu-
nication contributions are the outcome of group conversations and can be
linked either to video objects directly or to the associated additional informa-
tion units.
The Web-based HyperVideo System is based on a client-server architecture.
The DIS containing the content of the hypervideo is stored entirely at the server
side preventing the users from any form of data inconsistency. The clients are
allowed to extract (presentation mode) and to integrate (authoring mode)
content to and from the DIS on request. The Web-based graphical user
interface basically consists of a special video player that presents visually the
spatio-temporal hyperlinks besides the movie sequences within the video
display and offers functionalities in order to create new video annotations.
The cross-platform video player itself is written in JAVA using the Java Media
Framework for the purpose of manipulative video rendering. Newly created
video annotations are immediately transferred from the client to the server in
order to be instantly shareable by the community. The system concept allows
the adaptation due to different GUI layouts. Thereby different end-device types
can be supported. Figure 4 shows the overall system architecture concept of
the collaborative hypervideo system, as suggested by Finke and Balfanz
(2004).
The human-computer interface concept is based on a view model (see Figure
5). This model allocates for each node type (annotated video, additional
information, and communication) a separate view within the graphical user
interface. In addition, a fourth view is introduced that enables the disclosure of
the hypervideo structure in order to support user orientation within the
graphical user interface.
Each view provides its own user interface to access certain parts of the dynamic
information space (see Figure 5). By means of these interfaces, a user can
initialize so-called system events, for instance, the activation of a hypervideo
Figure 4. DIS: System architecture

Video Server Web Server
Java Application Server
Servlet Servlet Servlet Servlet
Video Navigation Com Info Annotation

Streaming Engine Engine Engine Engine
Server
Web
Content
Hypervideo
Communi- Information
Structur +
cation content
Video Metadata
View 1 View 3
Annotation
Engine
Video / Annotation Information
View Inforamtion
View
Java Applet Client
View
and
View 2
Communication
Navigation View
View HTML
HTML
Web Browser
Presenter Engine
Figure 5. DIS: Examples of the graphical user interface showing the

different views
Video Video
View View
Navigation Information Navigation Communication

View View View View
link, by clicking on an object in the video view. Each view offers a number of
different facilities:
• Video view: The video view presents video sequences containing video
annotations. By means of VCR functionalities, a user can control the
tempo of the presentation. The existence of a video annotation is an-
nounced by the visualization of its sensitive region within the video display.
Since it might be disturbing in some learning situations, the user is in charge
of initializing the visualization process of sensitive regions. Hypervideo
links can be activated by clicking on the corresponding sensitive region
with a mouse pointer. Furthermore, the video view is used by the user to
generate a sensitive region, which defines the anchor of a hypervideo link;
• Information view: Within this view, the information nodes are displayed,
which are linked to sensitive regions in the annotated video sequences.
The information view offers the definition of hypervideo links. Thereby,
more than one information node can be linked to a sensitive region in the
video view (multiple link strategy);
• Communication view: The communication view presents the group
conversation in the form of text-based dialogs. Users can reply on existing
conversation or start a new dialogue. Thereby, a dialogue is always
related to an object in the video or to a specific information node. This
provides us with the opportunity to combine conversations with associ-
ated subjects within the hypervideo. The advantage is that a user has a fast
access to specific conversations and does not have to go through the entire
communication contribution created by the community;
• Navigation view: Within this view, the disclosure of the hypervideo
structure is presented in order to support user orientation. The arrange-
ment of all node types within the structure is visualized in a text-based manner.
Users can browse the navigation view and activate hypervideo links, which
will lead to the presentation of the content in the associated views.
The development of the Web-based user interface based on the view model
was paralleled with experimental work on learning with hypervideo environ-
ments and case studies in the context of university teaching. The experimental
research was conducted at the University of Tuebingen in Germany, in
cooperation with the Computer Graphics Center at Darmstadt in Germany with
an earlier version of this HyperVideo System. Results revealed that hypervideos
provide a very successful mode of learning, also positively acknowledged by

the satisfaction of the 74 subjects. Results further suggest that slight variations
in design options concerning position and number of links to additional
information did not significantly influence navigation and successful knowledge
acquisition in individual learning, but individual strategies and navigation
behavior, including exploration activity, redundancy, and duration of links
activated, were significantly and positively correlated with knowledge acqui-
sition (Zahn et al., 2004).
Our current research projects focus on collaborative “learning by design” and
include an implementation study with two media psychology courses at the
Universities of Muenster in Germany and Linz in Austria (Stahl, Zahn, Schwan,
& Finke, in press). In this project, about 30 advanced psychology students
collaboratively designed hypervideo structures on “presentation techniques.”
Results indicate a great general satisfaction with the work of crafting hypervideos
with the system.
Conclusion and Perspectives
From a cognitive viewpoint, video can be considered a powerful referential

anchor, serving to stimulate and facilitate both individual and collaborative
processes of learning and knowledge building, particularly in specific domains
such as the natural sciences, which rely heavily on visual phenomena. Video
provides context information in an efficient way but has limited advantages to
reflective learning. Hypervideo additionally allows more control and the
composition of rich and flexible knowledge structures, corresponding to
enriched mental models. Learners may also express their internal knowledge
structures externally and share and discuss them with other students. Thus,
hypervideo provides a better support to reflection and learning, in accordance
with learners’ individual needs and styles, at different learning phases (Chambel
& Guimarães, 2002; Guimarães et al., 2000; Zahn et al., 2002).
From our experiences, we concluded that the ability to integrate video in rich
hypermedia spaces enables learners to create rich representations and pro-
motes deeper understandings. It improves both text and video understanding
due to the contextualized explanations made possible by the integration of both
materials. Important relations with video information can be captured. How-
ever, there are some main challenges, including the management of a potentially
high cognitive load in processing information based on hypervideo, and usually
significant investments involved in the production of multimedia and video-
based products. The definition of design guidelines and the development of
effective and flexible production frameworks and authoring tools can help to
meet these challenges. Some work has been done in this direction (Chambel et
al., 2001; Chambel & Guimarães, 2002; Zahn et al., 2002, 2004), but more
research needs to be done.
Our directions for future research, either from a cognitive and technical
perspective, include new developments in tools and design guidelines for
hypervideo in different environments and contexts of use. This process involves
the definition and evaluation of new mechanisms to support individual and
collaborative learning with video on the Web and also in interactive TV and
mobile environments. It also includes the interaction with video augmented
books, furthering our previous research on hypervideo and digital talking
books. In collaborative learning, for example, a project will investigate the
interactions of modern hypervideo technology with individual cognition and
teachers’ instructions while supporting group discussion. The topic at hand will
be “persuasive strategies of TV advertising,” for media education in secondary
schools, and the learning goals include visual and digital literacy skills. The main
focus will be put on students’ communication patterns, collaborative hypervideo
design processes, and the groups’ products. In summary, we intend to explore
a broader approach to the support of individual and collaborative learning,
inspired by field studies in different learning contexts that might raise specific
challenges and require differentiated support.
Our cognitive relation with video is not only experienced in formal learning.
Most of the considerations made and the approaches proposed also apply to
many other applications, including situations of informal learning and art works.
Because video has important communicative, entertaining, and artistic proper-
ties, hypervideo can be used to support new forms of expression in new media
(Url-nm) in many areas of culture and entertainment, for example, in interactive
cinema, electronic literature, and museums. These areas can benefit from many
of the mechanisms developed for learning environments, but they may as well
raise new challenges that will inspire new methodological and technological
developments.
References
Baake, D. (1999). Projekte als Formen der Medienarbeit [Projects as a form

of working with media]. In D. Baake, S. Kornblum, J. Lauffer, L. Mikos,
& G. A. Thiele (Eds.), Handbuch Medien: Medienkompetenz - Modelle
und Projekte (pp. 86–93). Bonn: Bundeszentrale für politische Bildung.
Baker, M., Hansen, T., Joiner, R., & Traum, D. (1999). The role of grounding
in collaborative learning tasks. In P. Dillenbourg (Ed.), Collaborative
learning—Cognitive and computational approaches (pp. 31–63).
Amsterdam; Lausanne; New York; Oxford; Shannon; Singapore; Tokyo:
Pergamon.
Beichner, R. (1994). Multimedia editing to promote science learning. Journal
of Educational Multimedia and Hypermedia, 3(1), 55–70.
Bereiter, C. (2002). Emergent versus presentational hypertext. In R. Bromme
& E. Stahl (Eds.), Writing as design: Conceptual and empirical
approaches (pp. 73–78). Amsterdam; Boston; London; New York;
Oxford; Paris; San Diego; San Francisco; Singapore; Sydney; Tokyo:
Pergamon/Elsevier Science.
Carver, S. M., Lehrer, R., Connell, T., & Erickson, J. (1992). Learning by
hypermedia design: Issues of assessment and implementation. Educa-
tional Psychologist, 27(3), 385–404.
Chambel, T. (2003). Video based hypermedia spaces for learning contexts.
Unpublished doctoral thesis, Faculty of Sciences, University of Lisbon,
Portugal.
Chambel, T., Correia, N., & Guimarães, N. (2001). Hypervideo on the Web:
Models and techniques for video integration. International Journal of
Computers & Applications, 23(2), 90–98.
Chambel, T., & Guimarães, N. (2002, June 11–15). Context perception in
video-based hypermedia spaces. Proceedings of ACM Hypertext ’02,
College Park, MD.
Chambel, T., & Guimarães, N. (2005). Learning styles and multiple intelli-
gences. Encyclopedia of Distance Learning (pp. 1237–1247). Hershey,
PA: Idea Group Reference.
Cognition and Technology Group at Vanderbilt (1997). The Jasper Project:
Lessons in curriculum, instruction, assessment, and professional
development. Mahwah, NJ: Lawrence Erlbaum.
Conklin, J. (1987). Hypertext: An introduction and survey. IEEE Computers,

20, 17–41.
Correia, N., & Chambel, T. (1999, November). Active video watching using
annotation. In Proceedings of ACM Multimedia ’99, Orlando, FL.
Eco, U. (1979, summer). Can television teach? Screen Education, 31, 15-24.
Finke, M., & Balfanz, D. (2004). A reference architecture supporting hypervideo
content for ITV and the Internet domain. Computers and Graphics,
28(2), 179–191.
Fleming, N. (1995). I’m different: Not dumb. Modes of presentation (VARK)
in the tertiary classroom. Proceedings of the 1995 HERDSA Confer-
ence (Vol. 18, pp. 308–313).
Gardner, H. (1983). Frames of mind: The theory of multiple intelligences.
New York: Basic Books.
Guimarães, N., Chambel, T., & Bidarra, J. (2000). From cognitive maps to
hypervideo: Supporting flexible and rich learner-centred environments.
IMEJ-Interactive Multimedia Electronic Journal of Computer-En-
hanced Learning, 2(2).
Kolb, D. (1984). Experiential learning: Experience as the source of
learning and development. Englewood Cliffs, NJ: Prentice-Hall.
Lehrer, R. (1993). Authors of knowledge: Patterns of hypermedia design. In
S. P. Lajoie & S. J. Derry (Eds.), Computers as cognitive tools (pp.
197–227). Hillsdale, NJ: Lawrence Erlbaum.
LiestØl, G. (1994). Aesthetic and rhetorical aspects of linking video in
hypermedia. Proceedings of ACM Hypertext ’94 (pp. 217–223).
LiestØl, G. (1999). Essays in rhetorics of hypermedia design. Unpublished
doctoral thesis, Department of Media & Communication, University of
Oslo, Norway.
Mayer, R. (2001). Multimedia learning. Cambridge, UK: Cambridge Uni-
versity Press.
Nelson, T. (1974). Branching presentational systems: Hypermedia. Dream
Machines, 44–45.
Nielsen, J. (1995). Multimedia and hypertext—The Internet and beyond.
San Diego, CA: Academic Press Inc.
Norman, D. (1993). Things that make us smart. Cambridge, MA: Addison
Wesley.
Paivio, A. (1986). Mental representation: A dual coding approach. Oxford:

Oxford University Press.
Park, O., & Hopkins, R. (1993). Instructional conditions for using dynamic
visual displays: A review. Instructional Science, 21, 427–448.
Pea, R. (1993). Practices of distributed intelligence and designs for education.
In G. Salomon (Ed.), Distributed cognitions: Psychological and edu-
cational considerations (pp. 47-87). Cambridge, UK: Cambridge Uni-
versity Press.
Penuel, W. R., Korbak, C., Cole, K. A., & Jump, O. (1999). Imagination,
production, and collaboration in project-based learning using multimedia.
Proceedings of the Computer Support for Collaborative Learning
(CSCL) 1999 Conference (pp. 445–453).
Reinmann, P., & Zumbach, J. (2001). Design, Diskurs und Reflexion als
zentrale Elemente virtueller Seminare [Design, discourse, and reflexion as
central elements of virtual seminars]. In F. Hesse & H. Friedrich (Eds.),
Partizipation und Interaktion im virtuellen Seminar (pp. 135–163).
Münster, Germany: Waxmann.
Salomon, G. (1993). No distribution without individuals’ cognition: A dynamic
interactional view. In G. Salomon, Distributed cognitions. Psychologi-
cal and educational considerations (pp. 111–138). New York: Cam-
bridge University Press.
Sawhney, N., Balcom, D., & Smith, I. (1996). HyperCafe: Narrative and
aesthetic properties of hypervideo. Proceedings of ACM Conference
on Hypertext’96 (pp. 1–10).
Scardamalia, M. (2004). CSILE/Knowledge Forum. Educational technol-
ogy: An encyclopedia. Santa Barbara, CA: ABC-CLIO.
Schwan, S. (2000). Video in Multimedia-Anwendungen.
Gestaltungsanforderungen aus Kognitions-psychologischer Sicht [ ]. In
G. Krampen & H. Zayer (Ed.), Psychologiedidaktik und Evaluation II:
Neue Medien und Psychologiedidaktik in der Haupt- und
Nebenfachaus-bildung. Bonn: Deutscher Psychologenverlag.
Spiro, R. J., & Jehng, J. C. (1990). Cognitive flexibility, random access
instruction, and hypertext: Theory and technology for non-linear and
multidimensional traversal of complex subject matter. In D. Nix & R. J.
Spiro (Eds.), Cognition, education and multimedia: Exploring ideas
in high technology (pp. 163–205). Hillsdale, NJ: Lawrence Erlbaum.
Stahl, E., Zahn, C., Schwan, S., & Finke, M. (in press). Knowledge acquisition
by designing hypervideos: Different roles of writing during courses of
“new media production.” In L. v. Waes, M. Leijten, & C. Neuwirth
(Eds.), Writing and digital media. Elsevier.
Stahl, G. (2002, January 7–11). Contributions to a theoretical framework for
CSCL. Proceedings of CSCL 2002, Boulder, CO.
Url-nm. New media. ArtLex Art Dictionary. Retreived from www.artlex.com/
ArtLex/n/newmedia.html
Van Haneghan, J. P., Barron, L., Young, M. F., Williams, S. M., Vye, N. J.,
& Bransford, J. D. (1992). The Jasper series: An experiment with new
ways to enhance mathematical thinking. In D. F. Halpern (Ed.), Enhanc-
ing thinking skills in the sciences and mathematics (pp. 15–38).
Hillsdale, NJ: Lawrence Erlbaum.
Vygotsky, L. S. (1978). Mind in society: The development of higher
psychological processes. Cambridge, MA: Harvard University Press.
(Published originally in Russian in 1930.)
Wetzel, C., Radtke, P., & Stern, H. (1994). Instructional effectiveness of
video media. Hillsdale, NJ: Lawrence Erlbaum.
Zahn, C., Barquero, B., & Schwan, S. (2004). Learning with hyperlinked
videos—Design criteria and efficient strategies of using audiovisual
hypermedia. Learning and Instruction, 14(3), 275-291.
Zahn, C., & Finke, M. (2003). Collaborative knowledge building based on
hyperlinked video. In B. Wasson, R. Baggetun, U. Hoppe, & S. Ludvigsen
(Eds.), Proceedings of the International Conference on Computer
Support for Collaborative Learning—CSCL 2003 (pp. 173–175).
Zahn, C., Schwan, S., & Barquero, B. (2002). Authoring hypervideos: Design
for learning and learning by design. In R. Bromme & E. Stahl (Eds.),
Writing hypertext and learning: Conceptual and empirical approaches
(pp. 153-176). Oxford: Elsevier Science.
50 Verhaart & Kinshuk
Chapter III
Assisting
Cognitive Recall and
Contextual Reuse
by Creating a Self-
Describing, Shareable
Multimedia Object
Michael Verhaart, Eastern Institue of Technology, New Zealand
Kinshuk, Massey University, New Zealand
Abstract
Digital media elements, or digital assets, are used to illustrate things such
as images, sounds, or events. As humans, we use many senses to assist our
cognitive processes, and providing multiple representations will enhance
our ability to store, recall, and synthesise the knowledge and information
contained in the digital asset. This chapter introduces a model for a
multimedia object, that allows multiple representations to be managed,
and includes a structured metadata file describing the asset that captures
the original context. Humans are capable of classifying and describing
Assisting Cognitive Recall 51
millions of such objects, but recalling context and content often blurs over
time. Computer systems provide us with a way to store electronic objects,
and with a variety of representations and sufficient metadata they can be
used to assist cognitive recall.
Introduction
With the large number of digital assets available on the Internet, it has become
common practice to use and reuse these elements in many different contexts.
A significant problem that occurs is that the original context and associated
metadata gets lost. When the digital assets, such as images, sounds, or videos,
are created, they have specific properties and they exist in a specific and
describable context. Electronic elements have derived properties, such as a
type and size, and can be automatically generated. If the element is part of a
group, then common properties could be described and would include such
things as the author, possibly the location, and maybe some contextual
information such as the event. This metadata could be created in a template and
automatically added to the description of an element. Annotated information
that describes the element can also be attached to the metadata.
Humans are capable of classifying and describing millions of such objects. For
example, for image, the scene information may be retained in short- or long-
term memory, and for most individuals, the details will blur over time. Computer
systems provide us with a way to store electronic objects, and with sufficient
metadata, they can be used to aid in classifying, managing, searching, and
reusing these objects in a variety of contexts whilst still retaining their original
context.
This chapter describes a model that allows a digital asset to be described in its
original context, captures ownership details and annotated metadata, and
allows for multiple representations. As humans have variable memory capaci-
ties, the representation of a digital asset as a multimedia object will assist
multiple cognitive processes by providing appropriate metadata and alternative
representations.
The proposed metadata model is based on commonly used standards and
utilises eXtensible Markup Language (XML). Standards included are the
Semantic Web’s Resource Definition Framework (RDF), the Dublin Core
(used by library systems), and vCard which is used to identify individuals.
Need and Importance of Context: Russian-Finnish

Border “Misplaced” Context
Consider the images shown in Figure 1, which were taken on a bus trip to the
Russian-Finnish border.
If a presentation were to be constructed by a person who was not on the trip,
and was searching for an image from a media repository, it would be more likely
that the image on the right would be chosen as the one from the actual Russian-
Finnish border. In fact, the image on the right was taken at a tourist stop where
a “fake” border was constructed, while the other was taken at the actual
border. This is a common problem when sourcing media from a huge repository
such as the Internet, where the context of what is being reused is lost.
Problem of Missing Context or Loss of Context
There is significant research being undertaken in the area of deriving context,

by major organisations such as IBM and Google (Fallows, 2004). For
example, Google (n.d.) provides an image search facility, where keywords are
used to locate an appropriate image. The keywords are generated from the
context in which the image was found. Unfortunately, this can provide many
negative hits, for example, on a page, there can be images of a logo, navigation
buttons, unrelated advertising, and so forth.
Figure 1. Photos taken during a visit to Russian-Finnish border
Need for Metadata
So how can this contextual information be added to a media element? Data that
describe the properties of an object are referred to as metadata, and are
commonly referred to as data about data. Many file types have a built-in
ability to add information about the element. For example, in Windows
Explorer, right clicking on a jpg file, selecting Properties, and then Summary
displays the editable information as shown in Figure 2.
The ability to attach with an object the data about that object gives us the ability
to describe the object’s context.
Need for Multiple Representations
In the example shown in Figure 2, an image can be represented in two ways:

as a photo and as a textual description. The usage will depend on the context,
and additional representations may be desirable for appropriate representation
in different situations. For example, if a gallery of images is to be displayed, the
image needs to be represented as a thumbnail. If the image is to be displayed
on screen, then it should be sized accordingly, that is, for a screen display of an
image from the Web, download of a 5 MegaPixel digital camera photo (say 2.5
MB) is not necessary; a 1 MegaPixel image (350K) will suffice (CardMedia,
2004).
Figure 2. jpg file metadata
Carrying this further, if a video media element is being displayed, four

representations are desirable: a thumbnail, a static image that is representative
of the video, the actual video, and a textual description of the video. When a
video is displayed on a Web page, the first frame is rendered on the screen. For
example, a video of a kiwi bird may start with a clip showing just a bush scene,
which progresses to the kiwi. Therefore, the first frame may not assist in
identifying the contents of the video and an alternative static image would be
desirable. The textual description is used for searching and for providing an
alternative representation for those with visual impairment.
Therefore, a desirable feature for working with electronic media is the ability
to manage multiple representations. This allows for an appropriate format to be
displayed to an end user depending on the required context.
Creating Shareable
Digital Assets and the Need to
Retain Contextual Information
Commonly, the term digital asset is used to describe a media element. This can
be an unstructured media element, such as text, an image, an animation, a
sound, or a video, or a structured media element, such as a word-processing
document (typically contains text and images though it is possible to include
animations, sound, and video), a spreadsheet, a database, and so on (AOL
Time Warner Group, 2003). The form in which a digital asset is presented
should not only retain the appropriate contextual information but also present
it in the most appropriate form.
What Contextual Information is Required?
A title is probably the most significant piece of contextual information, followed

by the creator of the digital asset. This information can often yield a significant
amount of implicit knowledge (Dienes & Perner, 1999); for example, an image
of a forest with small upright silver birches may indicate to someone a Finnish
landscape. Referencing information, such as the source, publisher, and, on the
Web, universal resource locator (URL), would be important for digital assets.
Situational annotations, where the media element is described, are also an

important part of the attached contextual information. For example, the image
of the Finnish forest described previously would have no specific or contextual
meaning to anyone who could not identify the scene information such as the
trees in Finland.
User annotations (annotations made by someone other than the author) also
enable contextual information to be attached. Consider the case of a painting
by a famous painter such as Picasso. Annotations added by an expert on
Picasso’s paintings would greatly improve the contextual information of the
image’s content.
Multiple Representations are Required for Different

Contexts
An electronic image may have four basic representations: thumbnail, computer

display, printable (most often the original), and textual. Experience of the
authors indicates that computer display may even require two types of
presentations: one that allows for the image to be displayed on a Web page but
fits correctly on a printed page (an image width of 595 pixels), another that
enables the image and text to be displayed side by side on the screen in a
balanced way (an image of width 295 pixels).
How can a digital asset be represented with multiple files? The first option is to
embed multiple files in one single file, for example, the Portable Network
Graphic (PNG) format, Adobe’s Portable Document Format (PDF) file, and
Microsoft’s Document file (DOC) format. The second option is to put the files
into a wrapper. Many learning object repositories use the ZIP file format to
keep the files together. Substantial research in representing multimedia objects
is being done by the Motion Picture Experts Group developing MPEG-21
(Bormans & Hill, 2002), and by the organisations and individuals involved in
the aceMedia (2004) project. The third option is to keep the files separate and
use a naming convention to keep the files together. Microsoft has adopted this
in part to save Web pages in Internet Explorer where the main file is saved along
with the creation of a folder of the same name containing the included files. A
disadvantage of this approach is that files can become separated, while its
advantage is that no specialist software is required to view the files. The first
option would require considerable design and implementation skills and the
widespread adoption of the standard, and the second option is fundamentally
Figure 3. Multimedia Object (MMO) model
an extension of the third, that is the files are placed in a wrapper. Hence, this
chapter investigates a model conforming to the third option.
After evaluating several prototypes, a workable model has evolved, which
consists of a file that contains the textual metadata for the digital asset and
associated files in their original formats. In order to describe this collection of
files, the term MultiMedia Object (MMO) has been coined (Verhaart,
Jamieson, & Kinshuk, 2004), and is illustrated in Figure 3.
To enable the metadata and to describe the information about the associated
manifest of files, a markup language has been developed. This language is
needed to
• conform to existing standards;

• manage derived information, such as file name, type, and size; and
• manage annotated information, including the contextual information
such as situational data, creator information, bibliographic information, and

additional annotation information that may be added later when the digital asset
is reused.
Figure 4. MMO showing MVML file and associated files
Figure 5. MVML file manifest
The term Media Vocabulary Markup Language (MVML) is used to describe

the metadata language. As the files are to remain in their original form, a way
is needed to maintain their grouping. This is achieved by requiring files to be
named the same as the core digital asset but with different extensions (e.g., the
file could be called 2004Verhaart_FinlandRoad.jpg, and the associated MVML
file would be 2004Verhaart_FinlandRoad.mvml). Where multiple files of the
same type exist a ^ is used to extend the filename. For example, a thumbnail of
the file is called 2004Verhaart_FinlandRoad^.jpg, and an image formatted for
the screen 2004Verhaart_FinlandRoad^w595.jpg.
Figure 4 shows a conceptual diagram of an MMO showing the MVML file and
associated files.
Figure 5 shows how a single digital asset could be represented by a manifest
of related files and the MVML metadata file.
Managing Referencing Information Such as Rights and

Copyright
Media elements can come from two sources: the creator or external. The goal
of the MMO is to create a shareable object; therefore, whether it has been
created or copied, rights and copyright information need to be captured.
Objects can be created in many ways; for example, they may be hand
(computer) crafted where an authoring tool (such as a drawing program) or a
capturing device (such as a digital camera) is used. The rights and copyright
metadata will fundamentally be the same; it would contain the creator’s details
and copyright notice.
In the case of borrowed objects, this becomes less clear. With the Web as a
significant tool for information gathering and rearranging, many media elements
lose the rights and copyright information because this information is mostly not
stored with the element but is placed on the Web page. The ability to create an
associated metadata file with each media element can go some way in keeping
the rights and copyright information.
The Need for Standards and to Adopt Existing Standards
In order to make the sharing of media elements possible, a set of standards

needs to be developed. For example, if an image from an external Web site is
to be used, the associated metadata can be browsed to extract content,
context, and rights information.
Libraries have been working with these issues over many years, and Dublin
Core (DCMI, 2004) is the most common standard for digital works. The
publishing industry has developed the Publishing Requirements for Industry
Standard Metadata (PRISM) (Daniel, Hansen, & Pope, 2003).
A major initiative in progress on the Internet is the development of a framework
of technologies for the Semantic Web (Ford, 2004; Knorr, 2002; W3C,
2004a). Baker (2002) describes the Semantic Web as one “that will allow
machines to easily understand and work with the words and information
humans stuff into e-mails, documents and databases. Many time-killing clerical
tasks that today require a live person at the keyboard could be automated.”
Two important standards that are part of the Semantic Web are the eXtensible
Markup Language (XML) (W3C, 2004b) and the Resource Definition Frame-
work (RDF) (Miller, Swick, & Brickley, 2004). XML allows data to be
represented in a common and structured way and is a suitable container to
describe the metadata. The RDF uniquely identifies a resource, whether
electronic or not. Kuchling (2002) states that “RDF is part of a W3C effort
called the Semantic Web, aimed at building a web of information that’s easy for
machines to read.”
User Interface Design Considerations
The success of Hypertext Markup Language (HTML) is commonly attributed

to the fact that it has a clearly defined and manageable structure. This
characteristic is also essential in the design of a user interface for MVML. The
three design parameters of creation, viewing, and sharing all need to be
carefully considered for acceptable use of MMO.
When designing MMO/MVML creation software, it is important that it can run
in stand-alone mode as well as in an Internet mode, as there are many users not
connected permanently to the Internet. The application should be capable of
autogenerating the standard associated files; for example, when the MVML file
for an image is created, a thumbnail should also be created.
Viewing an MMO/MVML file can be generalised but it should be possible to
annotate the MMO. The amount of change required in existing media players
should be minimal, although at present a Web browser would be the most
obvious choice because it can cope with the greatest number of different media
types. An advantage of the MMO with its multiple associated files is that
cognitive loading can be decreased for each media element. For example, work
on exploration space control (Kashihara, Kinshuk, Opperman, Rashev, &
Simm, 2000) emphasises the display of optimal amounts of information to be
displayed on screen to reduce cognitive overload. A system utilising the MMO
would then be able to display an appropriate media object from the manifest
and provide a link to the additional related objects in the manifest.
Multimedia Objects (MMOs) and

Multimedia Vocabulary Markup
Language (MVML)
What Metadata Needs to be Attached to a Digital Asset?
There are two kinds of metadata that need to be attached to a digital asset:
derived and annotated (Goldfarb & Prescod, 2002). Derived data can be
extracted from the electronic file itself, for example, the name of the file or its
physical size. There are many advantages in capturing this information implicitly
rather than getting the application to retrieve this information each time it is
needed. For example, a thumbnail of an image with a description and its
physical size can be quickly displayed to allow the user the choice of accepting
the download or not.
Annotated metadata is that which is added by the user. In order for an MMO
to have an existence, it must have the mandatory field “title.” This is consistent
with database schemas where a primary key is defined or in the physical world
where a person is given some sort of identifier, for example, a name. An initial
list of annotated metadata attributes for an MMO includes title (mandatory),
creator, subject keywords, rights, and context (e.g., an image should be
described as if the viewer was blind).
Existing Metadata Standards
There are many groups working collaboratively designing common metadata

standards. The ability to describe an object in a structured manner is an
important way in which objects can be classified and organised. For example,
in a database, metadata about a student, employee, or client are used to clearly
identify them. Defining standards are essential for developing shareable ob-
jects.
There has been much work in defining objects used in learning. McGreal and
Roberts (2001) discuss the levels of granularity and define the simplest level as
being the information object or component (e.g., simple text or a photograph).
It can be argued that any object can be used for learning, but it is the authors’
contention that in a learning object definition this should be explicit rather than
implicit. Learning object specifications that are relevant include IEEE LOM
(IEEE LTSC, 2002), DCMES (Dublin Core Metadata Element Set) (DCMI,
2004), and SCORM (Shareable Content Object Reference Model) (ADL,
2004). These standards are then used in metadata schema implementations,
such as CanCore (Friesen, Fisher, & Roberts, 2004) or UK-LOM Core
(Campbell et al., 2004).
Before deciding which existing metadata standards or implementations can be
adapted or modified, core requirements of the MVML-based MMO need to
be analysed. The fundamental metadata schemas that need to be considered
are Dublin Core, vCard, and the Resource Definition Framework.
Librarians have needed metadata to describe the publications they manage, and
the Dublin Core metadata schema has been developed to classify electronic
material. “The Dublin Core has been described as the most broadly based
metadata specification. It coexists comfortably with the other metadata sets
and all its elements are optional and syntax independent. Like the other
specifications, it can be tagged in HTML, raw XML or RDF/XML. It consists
of only fifteen fields such as Title, Description, Creator, Subject, etc.” (McGreal
& Roberts, 2001).
This has been significantly extended by the Metadata Encoding and Transmis-
sion Standard (METS) (Library of Congress, 2004). METS is a standard for
encoding descriptive, administrative, and structural metadata regarding objects
within a digital library. It is expressed using XML and includes the Dublin Core
standard.
vCard is a specification that is used when authors need to be identified, as in
the IEEE Learning Object Metadata Specification (IEEE LTSC, 2002) for
identifying a creator (vCard, 1996). It is commonly used in e-mail systems as
an electronic business card. The vCard standard was developed prior to the
emergence of XML as a common wrapper for describing metadata. Iannella
(2001) has developed an XML-based schema for vCard, and it is used in the
metadata of the MMO.
As mentioned earlier, the next generation of the World Wide Web is the
Semantic Web, and the RDF is a way to uniquely identify a resource, whether
electronic or not. Complying with this standard is an important part of the
MMO.
MVML a Markup Language to Describe a Digital Asset
As mentioned earlier, the ability to create reusable knowledge depends on

developing standards. The core standards adopted in MVML are the Dublin
Core, vCard, and the use of XML and RDF.
Figure 6. Driving in Finland
Figure 7. MVML file for MMO driving in Finland

<? xml version="1.0"?>
<rdf:RDF
xmlns:rdf = "http://www.w3.org/1999/ 02/ 22-rd f-syntax-ns#"
xmlns:dc = "http://purl.org/dc/elements/1.1/"
xmlns:vCard = "http://www.w3.org/2001/vcard-rdf/ 3.0#"
xmlns:mv ml = "http://is-research.massey.ac.nz/verhaart/1.0" >
This section contains meta-metadata used by all files in the manifest, such as a title, creator’s
name and details (in vCard format), creation date, keyword list (subject), and context.
<rdf:Description rdf:about = "2004Verhaart_Finland Road.mv ml" >

<dc:t itle>Dri ving on the Right-Hand Side of the Road in Finl and</dc:t itle>
<mv ml:creators>
<dc:c reator rdf:parseType="Resource">
<vCard:FN> M ichael Verhaart</v Card:FN>
<vCard:N rdf:parseType="Resource">
<vCard :Family>Verhaart</v Card:Family ><v Card :Given>Michael</vCard:Given>
</vCa rd:N>
<vCard:EMAIL rdf:parseType="Resource">
<rdf:value>[email protected]</rdf:va lue>
<rdf:type rdf:resource="http://www.w3.org/2001/vcard-rdf/3.0# internet"/>
</vCa rd:EMAIL>
</dc:creator>
</ mv ml:creators>
<dc:date>2004-12-27</dc:date>
<dc:source rdf:parseType="Resource">
<mv ml:URI>
<mv ml:root>http://is-research.massey.ac.nz/verhaart/</mvml:root>
<mvml:path>me_ med ia/2004Finland</ mv ml:path>
</ mv ml:URI>
</dc:source>
<dc:subject>Finland, Joensuu, Taxi, Lights</dc:subject>
<dc:rights>Copyright 2004 M H Verhaart. All Rights Reserved</dc:rights>
<mv ml:context>Fi nland, Joensuu, Fri. 3 Sept photo through windscreen of taxi to airport. PDA on
dashboar d, truck appr oaching with lights on. Black car wi th trailer and pl anks in front. All cars have lights
on during day.</ mv ml:context>
</rdf:Description>
This section contains a manifest of the associated files with their attributes.
Figure 7. cont.
<mvml:manifest>
<rdf:file rdf:about = "2004Verhaart_FinlandRoad.jpg" >
<mvml:derived>
<width>1656</ width><he ight>1242</height><size>462810</size>
</ mv ml:derived>
<mv ml:annotated>
<type>image</type><subtype>basic</subtype><use>original</use>
</ mv ml:annotated>
</rdf:file>
<rdf: file rdf:about = "2004Verhaart_FinlandRoad^.jpg" >
<mv ml:derived>
<width>94</width><height>70</height><size>3072</size>
</ mv ml:derived>
<mv ml:annotated>
<type>image</type><subtype>basic</subtype><use>thumbnail</use>
</ mv ml:annotated>
</rdf: file >
<rdf: file rdf:about = "2004Verhaart_FinlandRoad^w595.jpg" >
<mv ml:derived>
<width>595</width><height>446</height><size>47791</size>
</ mv ml:derived>
<mv ml:annotated>
<type>image</type><subtype>basic</subtype><use>display</use>
</ mv ml:annotated>
</rdf: file >
</ mv ml:man ifest>
This section would contain the annotations.
<mv ml:annotations>
</ mv ml:annotations>
</rdf:RDF>
Figures 6 and 7 illustrate how an MMO would be represented: an image (Figure

6) and the actual MVML file (Figure 7). The manifest of files was illustrated
earlier in Figure 5.
Figure 7 illustrates how the MVML file is structured into the overall metadata
(commonly referred to as the meta-metadata) and the manifest. The meta-
metadata (in the <rdf:Description> tag) contains the overall description of the
MMO, such as the name of the MVML file and author details. The manifest
contains a list of the associated files and their properties. The annotations allow
users to add comments directly to the MMO.
Derived Metadata
Reviewing Figure 7 reveals that many of the values can be generated by

computer software. For example, the date when the MVML file was created
(<dc:date>) and the details of the files in the manifest (such as width, height, and
size of images). Therefore, it should be possible to create much of the metadata
automatically.
The bold parts in Figure 7 illustrate annotated metadata, which are those
values that relate specifically to the image being coded (title, subject, and
context). Other values such as the creator’s name could be inherited (as
discussed in the following section). This shows that the actual amount of
metadata that would need to be entered by the user is actually quite manage-
able. In MVML, title is a mandatory field, though an interesting observation is
that many metadata schemas, such as Dublin Core and LOM, do not have any
mandatory fields.
When creating metadata for a group of MMOs, some of the data may be
common. For example, if a collection of images were being codified, it is likely
that they would all have the same creator. The ability to create a template with
this common information would greatly speed up the process of creating the
MVML file. This can be referred to as inherited metadata.
The MVML Template
A significant problem that exists is that metadata creation requires manual input
by an individual or group. Unless there are real benefits, this will not occur
(Phillips, 2000). Further, the addition of metadata is often time consuming, with
much of the metadata being repeated. In order to facilitate rapid entry of the
metadata, a template with an embedded coding language has been developed.
This coding language allows the template’s author to specify which entries can
be automatically created by the system (the derived data), which data need to
be added by the user (such as the title), and which data can be inherited from
other data entered by the user (such as full name from given and family name).
Figure 8. MVML template: Example of folder generic template

<? xml version="1.0"?>
<rdf:RDF
xmlns:rdf = "http://www.w3.org/1999/ 02/ 22-rd f-syntax-ns#"
xmlns:dc = "http://purl.org/dc/elements/1.1/"
xmlns:vCard = "http://www.w3.org/2001/vcard-rdf/ 3.0#"
xmlns:mv ml = "http://is-research.massey.ac.nz/verhaart/1.0" >
The next section shows how the title (a * for mandatory) and derived data, &M(mvml file,)
and &D (date) ) are defined. The creator’s information is already entered.
<rdf:Description rdf:about = "^&M^" >

<dc:t itle>^Title[*]||^</dc:title>
<mv ml:creators>
<dc:c reator rdf:parseType="Resource">
<vCard:FN> M ichael Verhaart</v Card:FN>
<vCard:N rdf:parseType="Resource">
<vCard :Family>Verhaart</v Card:Family >
<vCard :Given>M ichael</vCa rd:Given>
</vCa rd:N>
<vCard:EMAIL rdf:parseType="Resource">
<rdf:value>[email protected]</rdf:va lue>
<rdf:type rdf:resource="http://www.w3.org/2001/vcard-rdf/3.0# internet"/>
</vCa rd:EMAIL>
</dc:creator>
</ mv ml:creators>
<dc:date>^&D^</dc:date>
<dc:source rdf:parseType="Resource">
<mv ml:URI>
<mv ml:root>http://is-research.massey.ac.nz/verhaart/</mvml:root>
<mvml:path>me_ med ia/2004Finland</ mv ml:path>
</ mv ml:URI>
</dc:source>
<dc:subject> shows the three-part pattern, the title (Subject), a default value (Finland, ICALT
2004), and a help prompt that would be used for automatic form generation.
Figure 8. cont.
<dc:subject>^Subject|Finland, ICA LT 2004|Enter keywords separated with a comma ^</dc:subject>
<dc:rights>^Copyright|Copyright 2004 M H Verhaart. All Rights Reserved^</dc:rights>
<mv ml:context>^Context |Finland Plane (Sat 21 Aug-Sun 22 Aug)Su mmer School(Mon 23-Aug-Fri 27
Aug)/Conference (Mon 30 Aug - Wed 1 Sep)/ Technology Park(Thu 2 Sep),Plane(Fri 3 Sep-Sun 5 Sep)
2004|Describe the resource as if to a b lind person^</mv ml:context >
</rdf:Description>
<mv ml:manifest>
</ mv ml:man ifest>
<mv ml:annotations>
</ mv ml:annotations>
</rdf:RDF>
The MVML template is made up of two parts: the folder generic template
(FGT) and the MVML manifest media-specific template. FGT is stored, as the
name implies, in the folder with related files. For example, if a collection of
images is about a particular country, it would make sense that there would be
common metadata. Figure 8 gives an example FGT.
This template illustrates how common metadata is entered, such as creator’s
name, and how data to be captured is specified. Since an MMO can be made
up of many associated media files, a separate MVML manifest media-specific
template has been created and is kept in a common folder accessible by the
entire system. Figure 9 shows a portion of this template for images. Equivalent
models have been created for text, animation, audio, video, and application,
and further extensions are possible.
Possible Use Cases for MMOs
The ability to add structured metadata that describes a media object in a

common and consistent way, irrespective of the media type, will allow many
applications to be developed. The separation of the media and their metadata
from a specific application will allow resources to be easily shared between
applications. The following sections look at possible applications where the
MMO/MVML model would be useful.
Figure 9. MVML template: Example of manifest media-specific template

<model>
<type>image</type>
<subtype>{basic|gif|png|jpg|jpeg|}</subtype>
<temp late>
<rdf: file rdf:about = "^&fn^" >
<mv ml:derived>
<width>^&dw^</width><height>^&dh^</height>
<size>^&ds^</size >
</ mv ml:derived>
<mv ml:annotated>
<type>image</type><subtype>^&as^</subtype>
<use>^choice[orig inal|icon|thu mbnail|display|prn]||Use type^</use>
</ mv ml:annotated>
</rdf: file >
</template>
</ model>
Content Management Systems
In content management systems (CMS), there is a need to show a representa-

tion of the media, be it image, animation, audio, video, or structured media
appropriate to the destination device (e.g., screen or printer) or in the correct
context (e.g., on a Web page). Reusing images often means redescribing the
media and most often the original context and creation information is lost. This
is amplified further when media is shared with other people. Figure 10 displays
a prototype CMS that has been developed to explore many of these problems.
The prototype demonstrates the use of basic structure of MMO.
Business Card and Contacts Organiser
Using MMOs to manage business cards and contacts provides opportunities

to assist in the recall of that person. To assist cognition, humans will use multiple
references. For example, it is not uncommon for a recall to be triggered when
hearing a voice. As the MMO is capable of managing multiple representations
of an object, this will enable multiple cognition channels to be stimulated to aid
recall. Prototype in Figure 11 displays a business-card reader where the actual
business card is scanned, the vCard description attached (in the MVML file),
Figure 10. Displaying a small fragment in a content management system
Figure 11. Business-card reader based on MMO/MVML
and other media representations associated, such as a video. Note the

occurrence of multiple business cards, where one is in English, the other in
Chinese.
Bibliographic Reference Organiser
MMO/MVMLs can be utilised in the management of electronic bibliographic

artifacts, such as journal papers, books, and book chapters. While online
libraries such as the Association of Computing Machinery (ACM, 2004)
provide citation information, most online sources require references to be
manually constructed. The MVML mechanism can be used to generate, say, an
American Psychological Association (APA) style reference (APA, 2004). For
example, if an electronic object (such as a pdf, html, or doc file) is encapsulated
in an MMO, the MVML description can be used to automatically create the
APA referencing style. Further, the ability of annotation allows comments to be
added to the object that will help identify the salient parts relevant to the
research context. A prototype has been developed to test the ability to produce
an APA citation style by producing an XML file that can be imported into
EndNote (n.d.).
Future Directions
There are many opportunities to use MVML-based MMOs and although much
of the groundwork has been achieved in defining a workable language and
model, there is still much work to be done. This includes evaluating existing
metadata implementations to assess whether they will be able to accommodate
the requirements of the MVML-based MMO.
The original CMS prototype, based on MMO, is being further developed, and
will be used to test the usability and practicability of the approach. Initial work
in this area, some of which has been described in the chapter, is encouraging.
As the CMS has evolved, it has begun moving in a new direction. In a test
environment it is evident that instructors in a blended environment customise the
content to suit individual needs rather than use unchanged shareable content.
From this, an individualised environment is being developed, where an instruc-
tor can place content and solicit feedback in a personal context. This has been
coined as “Virtual Me” and is currently a work in progress.
Conclusion
This chapter presents a model that can be used to describe digital assets in such
a way that they can retain their original context, be used efficiently in search
routines, and be easily shared. The context is attached to the asset via an
additional and related file, keeping the original digital asset intact and unaltered.
As the file is separated from the objects in the manifest, this will allow for current
and future formats to be catered for. Objects can be represented in a variety
of formats to allow the appropriate type to be delivered depending on the
context it is to be used in.
Cognitive recall is greatly enhanced when multiple representations are pre-
sented. For example, an image and an audio clip greatly assist in recalling a
contact. Our memories may be triggered in many ways, so multiple represen-
tations can greatly assist in reinforcing the acquisition knowledge. Different
situations may require different representations. For example, a set of textual
steps can be useful in recalling a process but a video of the process is more
effective when learning the process.
The variety of possible applications illustrated in this chapter demonstrates how
the MVML-based MMO can be used to provide supporting details to assist
in cognitive processes. Allowing annotations to be added to the object can
assist at the metacognitive level, and indeed can provide a level of dialogue
when multiple users are accessing the digital asset.
Acknowledgments
The authors would like to thank Professor Nian-Shing Chen for his support.
References
aceMedia. (2004). Project description. Retrieved December 15, 2004, from

www.acemedia.org/aceMedia/files/document/aceMedia Brochure v1.pdf
ADL. (2004). SCORM 2004 2nd edition overview. Retrieved January 18,
2005, from www.adlnet.org
American Psychological Association (APA). (2004). Retrieved January 5,

2005, from www.apastyle.org/
AOL Time Warner Group. (2003). What is a digital asset? Retrieved
December 2, 2004, from http://common.ziffdavisinternet.com/download/
0/1895/p057.pdf
Association for Computing Machinery (ACM), Digital Library. (2004). Re-
trieved January 5, 2005, from http://portal.acm.org/dl.cfm
Baker, D. (2002). Tech visionaries push the Semantic Web. Retrieved April
22, 2002, from www.sfgate.com/cgi-bin/article.cgi?file=/chronicle/archive/
2002/04/16/BU155164.DTL
Bormans, J., & Hill, K. (Eds.). (2002). MPEG-21 overview v.5. Retrieved
January 6, 2005, from www.chiariglione.org/mpeg/standards/mpeg-21/
mpeg-21.htm
Campbell, L., Baker, P., Duncan, C., Graham, G., Johnson, P., Nobel, H., et
al. (2004). UK LOM Core v 0.2 (2nd draft). Retrieved May 6, 2004,
from www.cetis.ac.uk/profiles/uklomcore
CardMedia. (2004). Photos stored per memory card. Retrieved December
23, 2004, from www.card-media.co.uk/memory+info.htm?page=ms
Daniel, R., Hansen, D., & Pope, C. (Eds.). (2003). PRISM: Publishing
Requirements for Industry Standard Metadata version 1.2h. Re-
trieved November 21, 2003, from http://prismstandard.org/
DCMI. (2004). Dublin Core Metadata Initiative. Retrieved December 11,
2004, from http://dublincore.org/
Dienes, Z., & Perner, J. (1999). A theory of implicit and explicit knowledge.
Retrieved September 2, 2003, from www.bbsonline.org/documents/a/
00/00/04/57/bbs00000457-00/bbs.dienes.html
EndNote. (n.d.). Retrieved August 4, 2004, from www.endnote.com
Fallows, J. (2004, December 26). TECHNO FILES; At I.B.M., That Google
Thing Is So Yesterday. New York Times, Section 3, p. 7.
Ford, P. (2004). WWW2004 Semantic Web Roundup. Retrieved June 2,
2004, from www.xml.com/pub/a/2004/05/26/www2004.html
Friesen, N., Fisher, S., & Roberts, A. (2004). CanCore Guidelines, for the
implementation of learning object metadata (IEEE 1484.12.1-2002)
VERSION 2.0. Retrieved May 19, 2004, from www.cancore.ca/
indexen.html
Goldfarb, C. F., & Prescod, P. (2002). XML handbook (4th ed.). Englewood
Cliffs, NJ: Prentice Hall.
Google. (n.d.). Retrieved December 21, 2004, from www.google.com
Iannella, R. (2001). Representing vCard objects in RDF/XML. Retrieved
September 24, 2003, from www.w3.org/TR/2001/NOTE-vcard-rdf-
20010222/
IEEE LTSC. (2002). 1484.12.1 IEEE standard for learning object,
metadata. Retrieved January 14, 2005, from http://ltsc.ieee.org/wg12/
Kashihara, A., Kinshuk, Oppermann, R., Rashev, R., & Simm, H. (2000). A
cognitive load reduction approach to exploratory learning and its applica-
tion to an interactive simulation-based learning system. Journal of
Educational Multimedia and Hypermedia, 9(3), 253–276.
Knorr, E. (2002). The Semantic Web gets real. Retrieved December 26,
2002, from http://techupdate.zdnet.com/techupdate/stories/main/
0,14179,2901665,00.html
Kuchling, A. (2002). Rough notes on RDF. Retrieved March 2, 2004, from
www.amk.ca/xml/rdf-notes.html
Library of Congress. (2004). METS: An overview & tutorial. Retrieved
January 6, 2005, from www.loc.gov/standards/mets/
METSOverview.v2.html
McGreal, R., & Roberts, T. (2001). A primer on metadata standards: From
Dublin Core to IEEE LOM. Retrieved November 14, 2001, from http:/
/vu.cs.sfu.ca/vu/tlnce/cgi-bin/VG/VF_dspmsg.cgi?ci=130&mi=1
Miller, E., Swick, R., & Brickley, D. (2004). Resource description frame-
work (RDF). Retrieved December 11, 2004, from www.w3.org/RDF/
Phillips, L. (2000). RDF—The resource definition framework. Using XML:
Special Edition, Que., p. 494.
vcard: The Electronic Business Card Version 2.1. (1996). Retrieved June 16,
2002, from www.imc.org/pdi/pdiproddev.html
Verhaart, M., Jamieson, J., & Kinshuk. (2004). Collecting, organizing and
managing non-contextualised data, by using MVML to develop a human-
computer interface. Lecture Notes in Computer Science, 3101, 511–
520.
W3C. (2004a). Semantic Web. Retrieved December 11, 2004, from
www.w3.org/2001/sw/
W3C. (2004b). Extensible Markup Language (XML). Retrieved December

11, 2004, from www.w3.org/2001/XML/
74 van Nimwegen, Tabachneck-Schijf, & van Oostendorp
Chapter IV
Guidance in
the Interface:
Effects of Externalizing
Information During
Problem Solving
Christof van Nimwegen, Utrecht University, The Netherlands
Hermina Tabachneck-Schijf, Utrecht University, The Netherlands
Herre van Oostendorp, Utrecht University, The Netherlands
Abstract
How can we design technology that suits human cognitive needs? In this
chapter, we review research on the effects of externalizing information on
the interface versus requiring people to internalize it. We discuss the
advantages and disadvantages of externalizing information. Further, we
discuss some of our own research investigating how externalizing or not
externalizing information in program interfaces influences problem-
solving performance. In general, externalization provides information
relevant to immediate task execution visibly or audibly in the interface.
Thus, remembering certain task-related knowledge becomes unnecessary,
Guidance in the Interface 75
which relieves working memory. Examples are visual feedback aids such
as “graying out” nonapplicable menu items. On the contrary, when
certain needed task-related information is not externalized on the interface,
it needs to be internalized, stored in working memory and long-term
memory. In many task situations, having the user acquire more knowledge
of the structure of the task or its underlying rules is desirable. We
examined the hypothesis that while externalization will yield better
performance during initial learning, internalization will yield a better
performance later. We furthermore expected internalization to result in
better knowledge, and expected it to provoke less trial-and-error behavior.
We conducted an experiment where we compared an interface with
certain information externalized versus not externalizing it, and measured
performance and knowledge. In a second session 8 months later, we
investigated what was left of the participants’ knowledge and skills, and
presented them with a transfer task. The results showed that requiring
internalization can yield advantages over having all information
immediately at hand. This shows that using cognitive findings to enhance
the effectiveness of software (especially software with specific purposes)
can make a valuable contribution to the field of human-computer
interaction.
Introduction
Humans interact with information in the world around them by taking it in,
processing it, and outputting reactions. To process information, they use
cognitive skills such as thinking, learning, reasoning, recognizing, and recalling,
as well as metacognitive skills, which entail thinking about cognition (for
instance, planning, strategizing, or choosing between reasoning or calculation
types). Cognitive science studies these domains of human thought. Much
research in this field is done through the analysis of subject reactions to
presented information. This makes cognitive science a source of knowledge
that could—and does—guide interface and system designers toward a more
effective presentation of information in computer systems. We believe that
utilizing cognitive findings to enhance the effectiveness of software can make a
valuable contribution. Increasingly humans exchange information with the aid of
computers, for instance, in education, entertainment, office tasks, information
search, e-mail, and many other domains. Advances in computer and multimedia
technology ensure that the format of this information is increasingly diverse
using multiple media. Moreover, programs can have hundreds of functions.
However, progression becomes difficult with this complexity of choices and
representations. Harnessing this complexity to make it manageable for humans
gave rise to the domain of “usability.” Soon, among other things, the importance
of minimizing user memory load became apparent. This resulted in recommen-
dations to simplify the interface, restricting available options to those needed to
carry out the task action at hand, and to keep options visible on the interface
so users could interact on the basis of recognition rather than recall (Nielsen,
1994). In other words, the aim was just-in-time delivery of just the right
information, obviating the need for memorization and extensive search in
memory.
Our research does not aim at uncovering more principles that make systems
even more usable, intuitive, or appealing. It goes beyond plain usability and
focuses on how to shape interfaces that induce a user to learn cognitive and
metacognitive skills, and thereby learn about the domain underlying the
interface. We would like to find patterns of human behavior occurring with
computer use, to find out what kind of behavior certain decisions in interface
design provoke, not only during interaction, but also after delays and in transfer
situations. In this, we feel that one continually has to consider the real purpose
of the system. If a system ought to teach material to students or children, or
needs to make sure that users do not mindlessly follow interface cues because
the task to perform is of a certain crucial nature, then we should know what it
is about an interface design that induces people to think and learn. In this
chapter, the focus is on internalization and externalization of information,
and how this may lead to different behavior on the user’s part. In the following
sections, we explain the different terms used in this context. After this, we will
look at the pros and cons of externalizing and internalizing information, and
some effects of varying interface elements on learning and metacognitive
processes. In the next sections we discuss an experiment on users’ behavior
that two interface styles (internalization and externalization) provoke, and more
specifically, the amount of planning and learning from the users’ side. In the
concluding section, we discuss our findings and lay out our future plans.
Externalization vs. Internalization

Information: Related Fields
Visualization of task-specific information, thus minimizing user memory load, is

often called “externalizing” the information (Zhang & Norman, 1994).
Externalization of information can be contrasted with internalization of informa-
tion, whereby certain task-related information is less directly available and
needs to be internalized (inferred and stored in memory).
Early work of Simon (1975) examining advanced chess skills and strategies to
solve the Tower of Hanoi puzzle had noticed the interdependence of “external
memory” (perception, cueing recognition memory), and internal memory
(recall memory). Norman (1988) argued for the need of a similar analysis for
human-computer interaction. Tabachneck-Schijf, Leonardo, and Simon (1997)
created a model in which distributed cognition was reified. For an example of
distributed cognition, showing the interaction between internal and external
memory in creating a Microsoft PowerPoint page, see Tabachneck-Schijf
(2004). The question is, how much information should be internalized and how
much should be externalized for an optimal task execution? Some research
shows that the more information is externalized, the easier executing a task
becomes (e.g., Zhang & Norman, 1994). Other research shows that external-
izing all needed information seduces the user into mindlessly following the
interface cues and learning or planning little (e.g., Mayes, Draper, McGregor,
& Oatley, 1988). Yet other research shows that giving users incentives to plan
and learn (i.e., internalizing information) pays off in better performance (e.g.,
O’Hara & Payne, 1999). Internalizing and externalizing information are related
to, but not the same as, internal and external memory, internal and external
cognition, and plan-based and display-based problem solving. All relate to
having information in the head, be it in long-term or in working memory, versus
information that is available via (direct) perception. Simon and colleagues refer
to external memory as all information that can be made available for
perception (e.g., Greeno & Simon, 1974). In practice, this comes down to all
information that, within the task execution, can be made available for direct
perception and working memory. Internal memory is all the information within
long-term memory that can be made available to working memory.
External cognition is a term coined by Scaife and Rogers (1996) and refers
to the interaction between internal and external representations (the latter being
“external memory”) when performing cognitive tasks (e.g., learning). Scaife
and Rogers provide a useful overview on the use of graphical representations

in educational environments and outline a theoretical framework. They concep-
tualize how different kinds of graphical representations (e.g., diagrams, anima-
tion, and multimedia) are used during cognitive activities. The framework
presents a set of core properties, of which the central one is computational off-
loading (the extent to which different external representations reduce the
amount of cognitive effort required for cognitive activities). This property was
also mentioned by Greeno and Simon (1974) and Larkin (1989).
Plan-based and display-based problem solving relate to which source is
used to draw information from. In plan-based problem solving, one uses
detailed problem-solving strategies from (internal) long-term memory. Dis-
play-based makes little use of learned knowledge but relies on (external)
interface information (Larkin, 1989; O’Hara & Payne, 1999).
Externalizing and internalizing information, finally, refer to making informa-
tion available on external memory versus not making it available on external
memory, thus requiring recording it on internal memory. We feel that studying
the relationship between internalizing and externalizing information in order to
find effects of interface information on behavior and problem solving is an
important first step.
Pros and Cons of

Externalizing Information
In many instances, and for many purposes, externalizing information and

thereby reducing working memory load is indeed a fine idea. Externalizing
interfaces “take the user by the hand,” limit choices, and provide feedback.
What information should a computer system offer? If one is working with a
dynamic visual display on a problem-solving task, problem states are directly
visible and elements that are normally stored in working memory can now be
read from the display. A way to facilitate reduction of memory load is by visually
showing what possible steps can be taken to go from one state to the other in
the problem space (for a problem space, see Figure 4). It prevents users from
internalizing a great deal of information. Other ways to reduce memory load and
support explorative behavior are, for instance, guidelines such as “keep the
number of possible operations small,” “make the possible operations distin-
guishable,” “make clear what the consequences of every action will be,” “make
the effects of actions visible once they have been executed,” or “make actions
easily undoable to make it safe to experiment” (Lewis & Polson, 1990; Van
Oostendorp & De Mul, 1999).
Systems such as cash dispensers, automated airline check-in systems, and so
forth should be designed in such a way that users can simply and easily follow
the interface cues. Other examples are wizards and help options. Graying-out
menu items that cannot be used at that moment is also an example of
externalizing information. For example, in Word, you cannot select “paste”
from the “edit” tab in the menu when nothing is copied or cut first. “Paste” is
shown in gray, indicating that the command exists, but that using it is impossible
at the moment. In various lists of user interface guidelines similar issues, such
as “visibility status” and “visual feedback” are mentioned.
Interfaces that externalize do minimize working memory load. For instance,
Larkin (1989) considered the role played by differences in external displays in
cognitive problem solving, finding that making information visible, enabling
“display-based problem solving,” helped people recover from interruptions in
work. As mentioned, Zhang and Norman (1994) showed that the more
information was externalized, the easier solving a problem became.
Figure 1. Examples of externalization by graying-out items
Van Joolingen (1999) studied the idea of learners actively engaging in order to
learn about the properties of what is underlying a certain domain. He mentioned
the reduction of memory load in the context of research on the role of learner
support in discovery learning. Discovery learning assumes that learners take an
active role and construct their own knowledge base by interacting with a
domain, and inferring rules from the results of these interactions. Active
involvement of the learner would result in a better base of knowledge in the
learner as opposed to more traditional ways of learning. The question is, of
course, how to accomplish this behavior from the users’ side. He addressed the
issue of supporting discovery learning by means of cognitive tools. Cognitive
tools were defined as instruments that support cognitive processes by relieving
working memory and which can aid the learner to direct the process, to perform
a part of it, or to externalize part of processes and memory.
However, externalization of information also appears to have disadvantages.
There is some evidence that during computer-based problem solving, with
most information externalized users simply “follow the yellow brick road.”
Users search the interface for cues and reliably find them, perform an action that
alters the interface, thus enabling another search for cues and so, applying
simple perception-based means-ends analysis, eventually complete the task. If
the task appears to be completed correctly, then it follows that users will not
be triggered to plan, learn, or look for underlying rules or reasons, and also, that
people can only carry out a sequence of actions if the interface supplies the
needed cues. In a study comparing memory of users for a sequence of actions,
researchers found just that (Mayes et al., 1988). Users using a command-style
interface (e.g., LaTeX) that requires internalization could recall a sequence,
while users using a direct-manipulation interface (e.g., Microsoft Word) with
most cues externalized could not. Also, external cues can easily lead subjects
to incorrect actions. Tabachneck (1992) found that when the needed informa-
tion was externalized on a graph so that simple perception yielded the correct
answer, subjects gave the correct answer. However, in different representa-
tional formats simple perception yielded an incorrect answer; here subjects also
followed the interface cue and gave the incorrect answer. In summary,
externalizing information does decrease memory load, but drawbacks are that
people are likely to perform continual trial-and-error recognition-based search
and not learn or plan. Also, the external information may lead to the wrong
answer and thus to errors. Task execution via externalized information may also
be very resistant to speeding up. We feel that there are many tasks for which,
during task execution, internalization of information is desirable. Examples
are tasks with the purpose to acquire knowledge or skills, be they cognitive or
metacognitive; tasks one does often, and consequently speeding up and
reducing errors is desirable (such as booking a trip for a travel agent), or tasks
where the cost of errors is high (e.g., controlling nuclear power plant or flying
an airplane). When looking more specifically at problem-solving tasks, plan-
ning and learning are thought to be essential factors. It is therefore desirable to
do more research on how interface design, internalizing, and externalizing
information can stimulate people to do more internalization. Not externalizing
all information may stimulate self-directed search, more active planning, and
learning.
Varying Elements and

(Meta)Cognitive Processes
O’Hara and Payne (1998) and Trudel and Payne (1996) found that when
people are induced to plan, too strong a reliance on external information leads
to negative effects regarding planning and transfer of skills. Inducing people to
do plan-based problem solving would lead, they hypothesized, to internaliza-
tion of metacognitive and cognitive knowledge, in turn improving task perfor-
mance and easier recovery from errors (see also Payne, Howes, & Reader,
2001). O’Hara and Payne’s (1998) experiments showed that plan-based
activity did lead to shorter solution routes and easier recovery from errors,
while a display-based strategy involved more steps because of more (trial-and-
error) searching. O’Hara and Payne used an interesting methodology to get
people to plan more. In one of two conditions, they made the interface harder
to use by imposing delays on operators. They reasoned that subjects plan more
when there is something at stake: here, reducing costly delays. A similar
observation was made by Svendsen (1991). Using the Towers of Hanoi
problem, a high-cost interface yielded improved understanding of problems. In
the studies of Payne and colleagues, the amount of information externalized was
not varied. We felt that not externalizing all information will similarly stimulate
students to self-initiate a search for the missing information to plan and to learn.
In the first session of our experiment, discussed next, we researched whether
varying the amount of externalized information would have an effect on
performance, planning, and learning while solving a problem. In the second
session we assessed what was left after 8 months of what subjects had learned
in the first session and whether this still had an effect on task performance.
Details follow.
Experiment Session 1
Hypotheses
1. Externalization will initially yield better task performance than internaliza-

tion. Knowledge is yet to be acquired, so guidance by indicating the
legality of moves will help externalization subjects.
2. Internalization yields better task performance later, especially after a
severe distraction.
After a while, subjects will have learned to solve the puzzle. Internalization
subjects had no guidance and had to acquire the solving skill by them-
selves. They will have stored the rules in long-term memory more solidly,
and have the needed information more readily available and thus perform
better later, especially after an interruption erased working memory.
Because of the guiding nature of the interface, externalization subjects will
plan and think less than the internalization subjects, therefore work more
on the basis of trial and error and consequently display worse perfor-
mance.
3. Internalization yields better knowledge. Not having externalized informa-
tion available will motivate a subject to start planning on the basis of self-
acquired rules. After the experiment, we expect the explicit knowledge of
rules to be memorized better by internalization subjects. There may also
be a difference in implicit procedural knowledge, but we expect it to be
smaller since both groups will have played the puzzle a similar number of
times.
An experiment was conducted in which subjects solved a puzzle on a PC nine

times in two conditions: internalized and externalized. In terms of performance,
we expect the outcomes to take the following course:
Figure 2. Hypotheses
• Phase 1: Three trials. Subjects do not have any needed knowledge

available yet, externalization offers more interface cues, and will result in
better performance.
• Phase 2: Three trials. Performance in the two conditions will be equal by
this time, because internalization subjects acquired internal knowledge
that compensates for the interface cues.
• Interruption: This task is sufficiently demanding that all of the working
memory is erased.
• Phase 3: Three trials, after the interrupting distraction task. Internalization
subjects perform better because of better internalized knowledge and a
more elaborate plan.
Materials
Our problem-solving task, “Balls & Boxes” (B&B), is an isomorph of the

classic puzzle “Missionaries and Cannibals” (M&C). Five missionaries and five
cannibals are standing on a riverbank, and all have to reach the other bank by
boat. Constraints are that the boat only holds three creatures, and the minimum
to sail is one, because the boat cannot move by itself. Furthermore, the
cannibals can never outnumber the missionaries at any place (except when
there are zero missionaries), or the missionaries will be eaten. Our B&B
problem (Figure 3) uses exactly the same problem space (Figure 4), but is more
abstract.
We made the puzzle more abstract to reduce the many common knowledge
elements in the classic M&C: cannibals eat people, boats cross rivers, boats
cannot sail empty, and so forth. Using boxes, blue and yellow balls, and a dish
Figure 3. The Balls & Boxes puzzle
instead, we offer a semantically less rich problem, avoiding the rules to be

learned and remembered too easily.
The rules of B&B were as follows:
1. Balls should be transported using the dish.

2. You can transport at most three balls at a time.
3. To move, the dish must contain at least one ball.
4. The dish has to reach the boxes in turn.
5. There can never be more blue than yellow balls in the dish (except when
there are zero yellow balls).
6. There can never be more blue than yellow balls left in either of the boxes
(except when there are zero yellow balls).
Rules 5 and 6 are the most problematic rules, making no semantic sense
whatsoever. In our puzzle, it was possible to consult all the rules, but having
access to these does not imply that one knows how to solve the problem, and
surely not how to solve it in the most efficient manner. Below we depict the
formal problem space of the puzzle (Figure 4). The numbers indicate the
number of blue and yellow balls in the left and the right box, and the “*” indicates
on which side the dish is (see the legend of Figure 4). The shortest path from
“A” to “N” takes 11 moves. There are also states of the puzzles that we call
“dead-end states,” such as states D, E, G, J, K, and M. These states are not
on a direct solution path, and force backtracking.
The interface controls of the puzzle were straightforward (Figures 5 and 6). To
get balls into or out of the dish, blue or yellow up-arrows or down-arrows had
to be clicked. To move the pink dish horizontally, one had to click a pink arrow
(left or right). After the dish was moved, the balls automatically dropped into
Figure 4. Problem space of the Balls & Boxes problem
the box. To consult the rules one clicked on the rules tab. The rules screen
overlaid the puzzle screen. The puzzle was designed in two versions:
1. Internalized: All arrows were always colored (Figure 5) and clickable,

providing no information about the legality of moves (performing illegal
moves is possible). One could click all buttons at all times; however, illegal
moves would be carried out only partially. For instance, if one wanted to
transport the dish empty from left to right and clicked the right pink arrow,
the empty dish would move to the right. Then, an error notification would
pop up saying, “This is not possible.” By clicking “OK” in the dialogue
box, the program would undo the move.
2. Externalized: The arrows are colored (and thus clickable) only when an
action is legal. The arrows are grayed-out (unclickable, as in Figure 6)
when a move is illegal. For instance, in Figure 6, moving the dish empty
from left to right is illegal (violating rule 3), externalized by graying-out
both pink arrows. It is now only legal to move balls up into the dish.
Figure 5. Internalized version: No information whether an action is

possible (no gray-out)
P in k arro w s R u les-tab
Y ello w arro w s/b alls B lu e arro w s/b alls
Figure 6. Externalized version: Information whether an action is possible

(grayed-out)
Greyed out buttons
Subjects and Design
Thirty subjects of age 19–28 (mean age 21.8), experienced with PCs, were
recruited at Utrecht University. Subjects were paid • 5. The puzzle, a Java-
applet, ran on a Pentium IV 2.0 GHz PC with a 17" monitor. Our design has
one random-assignment between-subjects independent variable: internaliza-
tion versus externalization. Among the dependent variables were the following:
• Performance measures (logged)
• Number of puzzles subjects solved per phase (maximum three per

phase)
• Speed: time needed to solve the puzzles
• Reaching dead-end states. These are puzzle states that are off a direct
solution path, indicating trial-and-error search (states D, E, G, J, K,
and M in Figure 4)
• Knowledge test: After the puzzles, we measured implicit and explicit

knowledge of the rules and shortest-path solutions of the problem, using
a paper questionnaire.
• Attitudes: Likert-scale questions concerning perceived amount of plan-
ning, feeling lost during interaction, and other such measures.
Procedure
The experiment was conducted in the usability lab at the Center for Content and
Knowledge Engineering, Utrecht University. We informed the subjects of the
course of the experiment, and gave a brief oral explanation of the interface and
a short demonstration. The experiment consisted of nine puzzle trials, divided
into three equal phases, and a 10-minute distraction task between phase 2 and
3. The maximum time for each trial was set at 7 minutes. Slightly different
starting situations of the puzzle were used to avoid subjects simply repeating
actions (states A, B, and C in Figure 4). Also, in the second phase, the playing
direction of the puzzle was reversed to right to left. In the third phase, the playing
direction was set to left to right again. After the last trial, subjects filled out a
knowledge test (score 0–8) consisting of four multiple choice and four open
questions with screenshots of puzzle situations. They had to judge and explain
whether and why certain actions were possible (implicit knowledge), and recall
the rules (explicit knowledge). Subjects also rated how clear the rules were for
solving the problem.
Results Experiment, Session 1
Number of Puzzles Solved per Phase
Subjects could attempt to solve three puzzles per phase. A MANOVA

(analysis of variance) showed a significant main effect of phase on the average
number of puzzles solved, F(2,56) = 53.74; p < .001 (Figure 7) (pointer to
figure). The number of puzzles solved (out of three) improved in later phases,
indicating a learning effect. Post-hoc comparisons showed that only the
difference between performance in phase 1 and phase 2 was significant (p <
.05) (Figure 8) (pointer to figure). Although the graph suggests an overall
better performance for internalization (M = 6.93, SD = 2.05 vs. M = 6.33, SD
= 3.09), this did not reach significance. There were no interaction effects.
Time Needed for the Puzzle Trials
A MANOVA showed a main effect also for phase on average solving time,
F(2,42) = 35.16; p < .001. The time subjects needed to solve puzzles lessened
in later phases, also a learning effect. Post-hoc comparisons showed that all
subjects needed significantly more time in phase 1 than in phase 2, and also in
phase 2 than phase 3.
Figure 7. Puzzle trials solved per phase, per version
3 .0
2 .5
2 .0
T ria ls s o lv e d p e r p h a s e
1 .5
1 .0 Inte rfa c e s tyle
.5 Inte rna liza tio n
0 .0 E xte rna liza tio n

1 2 3
P ha s e
Figure 8. Time needed per solved trial, per version
A v e ra g e tim e (s e c o n d s ) p e r s o lv e d tria l
300
250
200
150
100 In te rfa c e s tyle
50 In te rna liza tio n
0 E xte rna li za ti o n
1 2 3
P ha s e
The graph suggests that internalization subjects need some more time on
average, but ANOVA showed that this was just a trend (F(1,21) = 2.79, p =
0.11). Looking closer, it was only in the beginning, in phase 1, that internaliza-
tion subjects needed more time (M = 238.77, SD = 62.20) to solve the puzzles
than externalization subjects (M = 186.77, SD = 67.08). This difference was
significant, t(21) = 1.94, p = 0.03 (one-sided). After this first phase, the
differences were minuscule and no longer significant.
Dead-End States
Below is the number of times per phase subjects reached dead-end states.
Figure 9. Dead-end states per phase, per version
14
N u m b e r o f d e a d -e n d -s ta te s re a c h e d
12
10
6
Inte rfa c e s tyle
4
Inte rna liza tio n

2
0 E xte rna liza tio n

1 2 3
P ha s e
A MANOVA showed an interesting nearly significant main effect of interface

style (Figure 9). Overall, externalization subjects reached more dead-end
states (F(1,28) = 3.58; p = 0.06). In addition, there is a trend for an interaction
effect of phase and version (F(2,56) = 2.11; p = 0.13). Internalization subjects
significantly improved from phase 1 to 2 (M = 11.4, SD= 4.70 and M = 3.4,
SD = 3.18), t(14)=5.80, p = 0.00, one sided). This is an improvement of eight
states. They also improved from phase 2 to 3 by almost two states, nearly
reaching floor level (M = 3.4, SD= 3.18 and M = 1.47, SD = 2.07),
t(14)=1.96, p=0.04, one sided. Externalization subjects improved by 6.2
states from phase 1 to 2 (M = 12.67, SD= 6.91 and M = 6.47, SD = 7.11),
t(14)=2.74, p = 0.02, one sided. But after the interruption, it was different.
From phase 2 to 3, they did not further improve. On the contrary, in phase 3
they reached more dead-end states than before, although not significantly so.
Here in phase 3 the difference between internalization (M = 1.47, SD = 0.53)
and externalization (M = 7.73, SD = 10.64) was significant, t(28)= “2.24, p =
0.04. Externalization subjects reached more than six more dead-end states in
phase 3 than internalization subjects.
Knowledge Test
Before measuring knowledge of the rules and states, we confirmed that there
was no difference in the number of times subjects in both conditions consulted
the rules. The knowledge test that subjects received after the experiment
consisted of several procedural knowledge questions and one explicit knowl-
edge question. The Balls & Boxes puzzle contained more than 30 legal states
(Figure 4). There were seven procedural implicit knowledge questions about
seven of those states (open and multiple choice) in which subjects were visually
presented with a puzzle state. They had to judge whether certain moves led to
the solution, and explain why. The scores on those seven questions were high
for both internalization (M = 6.3, SD = 0.96) and externalization subjects (M
= 5.8, SD = 1.20), considering that the maximum score was seven. There was
a trend that internalization on average scored higher than externalization on the
procedural knowledge questions (t(28) = 1.17, p = 0.12).
To test the explicit rule knowledge, the possibilities were limited. The puzzle
had only few rules, some of which (rules 1 to 4) were extremely easy to grasp
and remember. Rules 5 and 6 are more crucial and define the difficulty of the
puzzle. They can be merged to just one rule: “Blue balls can never be in the
majority at any place, except when there are only blue balls (zero yellow balls).”
Subjects were asked about this rule with a multiple-choice question. All
internalization subjects answered this question correctly, whereas of the
externalization subjects only 60% answered it correctly. This difference was
significant (Phi = -.50, p < 0.01). Furthermore, subjects had to estimate the
clarity of the rules. They rated the question “the given rules were clear enough
to solve the problem” (score range 1–5). Internalization subjects found the
rules clearer than the subjects from the externalization condition (M = 4.13, SD
= 0.52 and M = 3.53, SD = 1.25), t(28)=1.72, p = .04 (one-sided).
Summary Experiment Session 1
Our first hypothesis stating that initially externalization yields better perfor-
mance was only partly supported in terms of the time subjects needed.
Internalization subjects did take more time in the first phase. Other measures
were in the expected direction, but not significantly so. Unexpectedly,
externalization yielded no better performance in terms of puzzles solved in the
beginning, as we expected.
The second hypothesis stating that internalization yields better performance in
a later stage was partly supported. We saw that the time subjects needed
stabilized, and the same was true for the number of trials solved per phase.
However, we also looked at performance in a more delicate manner, not in
terms of time or correctness, but at how, via what route, subjects reached their
goal. We introduced the measure “dead-end states” to inform us as to how
subjects behaved, in terms of the insight they had, the “smartness” of their route.
We assumed that internalization subjects do some smarter, more elaborate
planning, while externalization subjects are expected to solve more by trial and
error and on the basis of interface cues. It showed that internalization subjects
performed better after the interruption and reached those dead-end problem
states less often in all three phases (indicating less “lostness”). Furthermore,
there was also the trend-like interaction that after the interruption, internaliza-
tion subjects kept improving, while externalization subjects fell back, reaching
more dead-end states than they did before. This confirms our expectation that
after the interruption, internalization subjects would continue to work on the
basis of the plan-based strategy as they did before. Externalization subjects, on
the other hand, perform no better after the interruption. They fell back to
depending on the interface, having a less elaborate plan.
The third hypothesis in which we expected that internalization would result in

having better knowledge was supported. We also assumed that internalization
subjects, who could rely less on interface information, had to build a stronger,
more elaborate plan. Though when testing implicit knowledge both groups
scored equally high, with a trend advantage for internalization subjects, when
asked about explicit knowledge of the crucial rule that defines the difficulty of
the puzzle there was a significant difference. All the internalization subjects
could correctly answer this question, whereas only 60% of the externalization
subjects could—in spite of having the rules readily available for consultation
and consulting them as often as internalization subjects. Furthermore, there was
also the tendency that internalization subjects rated the clarity of the rules
higher. This is intriguing, because in the externalization version of the puzzle
subjects had interface feedback and were able to consult the rules. Internal-
ization subjects, who only had the rules and no interface help found the rules
clearer. We carefully interpret the latter two findings as indicators of better
understanding in the internalized condition.
Experiment Session 2
We were curious to see how stable this better knowledge provoked by

internalization was, and therefore decided to conduct a second session after a
delay of 8 months. We invited the same subjects of experiment 1; 14 of the 30
were available. There were two reasons for this rerun. First, to see whether the
better knowledge measured in the internalization subjects had endured, we
asked subjects to solve B&B five times (experiment 1 showed three to four
puzzles suffice for all subjects to be able to solve the puzzle within the allotted
time). Second, to see whether the better knowledge might result in better
performance on a transfer task, we also confronted subjects with a transfer
problem. Transfer problems require subjects to apply acquired skill on a
different task of the same nature. To be able to measure differences in
performance between the two initial groups (internalization and externalization)
on the same material, we presented all subjects with the same material this time,
one interface style, namely, externalization. Note that the internalization sub-
jects had to make a change in interface style.
Hypotheses
1. Because internalization subjects still have better memory of the rules and
solution, they will perform better. We expect the internalization subjects
to solve the puzzle faster initially because of their better knowledge of the
rules. They will be faster in recollecting knowledge needed to solve the
puzzle. After a while, we expect the two groups to perform more or less
equally on this puzzle, like in session 1.
2. Internalization subjects will perform better on the transfer tasks. After five
times of B&B, we expect the two groups to perform at the same level. But
when confronted with transfer tasks (that have similarities but also a few
differences), we expect internalization subjects to perform better, again
because they possess better knowledge of the rules.
Materials
To test knowledge retention, all subjects first solved the B&B puzzle in the
externalized version five times. To test transfer performance, we used another
puzzle of the same M&C “family,” but with varying characteristics. We first
used a quite literal version of M&C, which was further away in terms of transfer.
Balls & Boxes
The only difference between this and the previous B&B was that the rules were
not consultable anymore, so we could obtain a more pure measure of retention.
All subjects received the same externalized version. This meant that attempting
illegal moves (moves that violate rules) was not possible, as the externalized
interface only allowed legal moves.
Missionaries and Cannibals
The second problem was a transfer task, ironically the original version of the
M&C. It was a game that literally shows missionaries, cannibals, a river, and
a boat. The solution algorithm to this problem was the same as B&B, but there
were some difficulties that did not exist before. The most important one is that
Figure 10. The initial state of the Missionaries & Cannibals puzzle
there were not five of each entity as before, but three, and the playing direction
was always from right to left (Figure 10). Although it might seem that one can
solve the three-creature problem faster, this is not so—the shortest path to the
solution is also 11 moves. Also, unlike in B&B, in this game attempting illegal
moves does not lead to an error message, but subjects would “die,” as in a
computer game, and the game would start over. Errors thus had more severe
consequences, and subjects needed to exercise caution. Subjects had to solve
the problem as many times as they could in 8 minutes.
Subjects and Procedure
Fourteen of the 30 subjects from session 1 were willing to participate a second

time. Luckily, seven of them had worked with internalization, and seven with
externalization in session 1. We confirmed that these 14 subjects were subjects
who performed “normally” (no extreme scores) in session 1. The experiment
consisted of two parts:
1. B&B (five times). The maximum time for each trial was set at 7 minutes.
Slightly different starting situations of the puzzle were used to avoid
subjects simply repeating actions.
2. M&C (8 minutes).
After completing the experiment, subjects received a • 5 reward.
Results Experiment Session 2
• Balls & Boxes. Encouraging results were found concerning solving the
puzzle correctly again the first time. After not having seen the puzzle for
8 months, it took the internalization subjects only half the time used by the
externalization subjects to solve the first B&B puzzle (M = 432 sec, SD
= 314 and M = 778 sec, SD = 397). This difference was significant, t(12)=
–1.81, p = .05. There were no further significant differences between the
two groups. After the first puzzle, as expected, all subjects solved the
remaining four trials of B&B puzzle equally well, just as 8 months ago.
• Missionaries & Cannibals. This puzzle was a bit further in transfer. The
graphical characteristics of this puzzle differed considerably. Still the
algorithm to solve it in itself was the same, but the number of creatures to
transport was different, and also the maximum number of creatures
allowed in the boat. Although the shortest solution path is the same length
as B&B, the problem does have a smaller problem space. The same basic
concept had to be applied to a situation that differed at face value, and also
underneath. Out of our 14 subjects, 10 managed to solve this puzzle one
or more times. Just as in the B&B puzzle, internalization subjects solved
it the first time faster (M = 176 seconds, SD = 72.5 vs. M = 302.8, SD
= 202.7), though it was just a trend, t(8) = –1.44, p < 0.10 (one-sided).
Moreover, internalization subjects managed to solve the puzzle three
times more often (M = 4.83, SD = 1.94) in the 8 minutes than externalization
subjects (M = 2.5, SD = 1.73). This was significant, t(8)= 1.94, p < .05
(one-sided).
Summary Experiment Session 2
Eight months is a long time, but still the interface style subjects received in
session 1 appeared to make a difference. Subjects were not told at the first
session that they would be asked to play the puzzle again, and when we
approached them, we explicitly told them it was not the same experiment they
had participated in. Upon being confronted with the puzzle again, the subjects
had to recollect the rules and solution strategy from their long-term memory.
The two groups both worked with an externalized version this time and had
exactly the same information. We looked at how long it would take subjects to
remember how to do this correctly. It showed that in solving the first B&B
again, internalization subjects were indeed significantly faster, almost twice as
fast, in fact. This is all the more remarkable given that these subjects received,
to them, an unfamiliar interface—externalized rather than internalized. After
that first success, the performance of both groups equalized. This is coherent
with the results from the first session, where stabilization in performance also
occurred. The above finding supports our first hypothesis of experiment 2
stating that internalization subjects would still have a better memory of the rules
and solutions of the puzzle.
The M&C puzzle that subjects received was the same for everyone. It can be
regarded as an internalization version (no direct interface clues were available,
all controls were available all the time). Also here there were some interesting
results. The subjects that worked with the internalization version of B&B
managed to solve this puzzle three times more often than externalization
subjects, and this was significant. Furthermore, just like in B&B, internalization
subjects needed less time to solve the problem the first time, although this was
just a trend. It supports the second hypothesis stating that internalization
subjects will perform better on a transfer task. In summary, we demonstrated
that the interface style subjects worked with 8 months ago still appeared to be
of influence months later, both in solving the same puzzle again as in solving a
far transfer puzzle. We take it as encouraging support for the better memory of
knowledge provoked by the interface style subjects worked with.
General Discussion on the Two Sessions
We analyzed the influences that externalizing certain information on the

interface, thereby making it visible, or not externalizing it has on performance
and knowledge acquisition in a problem-solving task. Furthermore, we looked
at the long-term effects after 8 months, and at the issue of transfer. Both in
session 1 and in session 2, the delayed session, NOT externalizing information
led to enduring advantages in performance and knowledge, while externalizing
information merely led to a fleeting advantage at the very start of session 1.
According to Zhang (1997), the more information is externalized, the easier
performing a task becomes. We hypothesize, however, that being led by the
interface also incurs a cost, namely, a lessening of metacognitive activity such

as planning, and consequently, less learning. In such situations as where learning
is required, transfer is at stake, or the task is prone to interruptions, planning,
learning, and in general, getting to know more about the structure of the task
at hand are desirable outcomes. Surprisingly, performance measures such as
time needed to solve the puzzles and number of correctly solved puzzles were
not much influenced by the interface style—if anything, the internalization group
was at an advantage. Zhang’s (1997) prediction was not confirmed. We feel
that the attention of users in the externalization group was immediately taken by
the guiding nature of the interface, and as they were not confronted with actual
mistakes (one could not make illegal moves, only inefficient legal moves were
allowed), they simply kept on solving without feeling a need to figure the
problem out more. Subsequently, less active learning took place. This idea of
attention taken by an interface fits with Carroll and Rosson’s (1987) paradox
of the active user—users of computer systems are so consumed with immediate
productivity that they are not motivated to take time to learn better ways of
accomplishing a task.
Contrary to the externalization group, the internalization group was confronted
with errors for which an explanation was not immediately available. These
subjects incurred a cost when making errors: a dialog box popped up which
they had to click away, and their attempted move was reversed. Though both
groups consulted the rules sheets equally often, this cost probably contributed
to motivating the internalization subjects to study the rules better in order to
avoid incurring the cost. We found that the internalization group possessed
better explicit knowledge of the rules and engaged in more planful problem
solving. Applying more metacognition to avoid a cost concurs with the findings
of O’Hara and Payne (1998, 1999).
During session 2 we found that the effects found in session 1 were still present.
Again, only the internalization subjects from months ago showed advantages.
This group was both faster in solving the same problem again, and also faster
in figuring out the solution to a transfer task. It is remarkable that the influence
of working with one interface or the other has effects even after such a long time.
In general, it was surprising for us to see that externalization showed even less
advantage than we thought, almost none, to be precise. We expected
externalization at least to be of help in the beginning when users were not
familiar with the system and the problem. It was, but only in time taken on the
first few trials—and this was just a trend that was not confirmed in the number
of puzzles solved correctly. This very small and fleeting advantage did not
concur with Zhang’s (1997) findings. Please note that our type of externalization
parallels Zhang’s more physical type: both restrict the number of illegal moves
that can be made, but do not alter the problem space or give cues as to what
the shortest path is through the problem space.
We find the results so far encouraging in that lessening the amount of external-
ized knowledge apparently can encourage cognitive and metacognitive behav-
ior. Moreover, one can be very precise about which information not to
externalize—in other words, this way the information to be learned can be
manipulated. Therefore, we feel that the issues of manipulating the amount of
externalization in interfaces deserve more attention. In the context of current
technology and the widespread use of computer systems in virtually all
domains, we are convinced that understanding how the system’s users will
perceive, reason, and interact with the system can be crucial. The implications
of the findings so far and the ones in the future can be valuable for development
of applications where active engagement and learning from users is the aim.
Examples are systems to teach material to students or children, or situations
where it is important that users are not “kept” stupid by the software. In certain
situations, users of a system need to understand the underlying structure/rules
of the system because the tasks are of a crucial nature. To accomplish this we
have to look beyond plain usability and common sense and better scrutinize the
interaction between internal memory and cognition on the one hand and
external memory and recognition and perception on the other hand. We believe
this can, for certain types of systems, enhance the effectiveness in achieving
their targets.
Future Trends
There are still many challenges in human-computer interaction and many issues
that need to be explored. Understanding how users will react to interface
information (on the basis of cognitive findings) is one important issue in attuning
software to its purpose, thereby allowing it to achieve its goal. In the future, we
will further investigate this issue by exploring behavior in different types of more
realistic planning-related tasks. As a more realistic planning task, we think of,
for example, spreadsheet or drawing applications where actions are less
repetitive, more complex, and could be part of a real job. We are currently
designing such an environment.
References
Carroll, J. M., & Rosson, M. B. (1987). The paradox of the active user. In J.
M. Carroll (Ed.), Interfacing thought: Cognitive aspects of human-
computer interaction (pp. 80-111). Cambridge, MA: MIT Press.
Greeno, J. G., & Simon, H. A. (1974). Processes of sequence production.
Psychological Review, 81, 87–98.
Larkin, J. H. (1989). Display based problem solving. In D. Klahr & K.
Kotovsky (Eds.), Complex information processing: The impact of
Herbert A. Simon (pp. 319-341). Hillsdale, NY: Lawrence Erlbaum.
Lewis, C. H., & Polson, P. G. (1990). Theory-based design for easily learned
interfaces. Special issue: Foundations of human-computer interac-
tion. Human Computer Interactions, 5, 191–220.
Mayes, J. T., Draper, S. W., McGregor, A. M., & Oatley, K. (1988).
Information flow in a user interface: The effect of experience and context
on the recall of MacWrite screens. Proceedings of the Fourth Confer-
ence of the British Computer Society on People and Computers, IV
(pp. 275–289).
Nielsen, J. (1994). Usability engineering. San Francisco: Morgan Kaufmann.
Norman, D. A. (1988). The psychology of everyday things. New York:
Basic Books.
O’Hara, K., & Payne, S. J. (1998). The effects of operator implementation
cost on planfulness of problem solving and learning. Cognitive Psychol-
ogy, 35, 34–70.
O’Hara, K. P., & Payne, S. J. (1999). Planning the user interface: The effects
of lockout time and error recovery cost. International Journal of
Human-Computer Studies, 50, 41–59.
Payne, S. J., Howes, A., & Reader, W. R. (2001). Adaptively distributing
cognition: A decision-making perspective on human-computer interac-
tion. Behaviour & Information Technology, 20(5), 339–346.
Scaife, M., & Rogers, Y. (1996). External cognition: How do graphical
representations work? International Journal of Human-Computer
Studies, 45, 185–213.
Simon, H. A. (1975). The functional equivalence of problem solving skills.
Cognitive Psychology, 7, 268–288.
Svendsen, G. (1991). The influence of interface style on problem solving.

International Journal of Man-Machine Studies, 35, 379–397.
Tabachneck-Schijf, H. J. M. (1992). Computational differences in mental
representations. Effects of mode of data presentation on reasoning
and understanding. Doctoral dissertation, Department of Psychology,
Carnegie Mellon University, Pittsburgh, PA.
Tabachneck-Schijf, H. J. M. (2004). How the format of information deter-
mines our processing: An IPP/CIP perspective. In H. van Oostendorp, L.
Breure, & A. Dillon (Eds.), Creation, use and deployment of digital
information (pp. 123–144). Mahwah, NJ: Lawrence Erlbaum.
Tabachneck-Schijf, H. J. M., Leonardo, A. M., & Simon, H. A. (1997).
CaMeRa: A computational model of multiple representations. Cognitive
Science, 21(2), 305–350.
Trudel, C. I., & Payne, S. J. (1996). Self-monitoring during exploration of an
interactive device. International Journal of Human-Computer Stud-
ies, 45, 723–747.
Van Joolingen, W. R. (1999). Cognitive tools for discovery learning. Interna-
tional Journal of Artificial Intelligence in Education, 10, 385–397.
Van Oostendorp, H., & De Mul, S. (1999). Learning by exploration: Thinking
aloud while exploring an information system. Instructional Science, 27,
269–284.
Zhang, J. (1997). The nature of external representations in problem solving.
Cognitive Science, 21(2),179–217.
Zhang, J., & Norman, D. (1994). Representations in distributed cognitive
tasks. Cognitive Science, 18, 87–122.
Section II
Mental Representation
of Concepts, Metaphor,
and Language
102 George
Chapter V
Bridging the Gap

between Human
Communications and
Distance-Learning
Activities
Sébastien George, INSA, Lyon, France
Abstract
This chapter introduces context-aware computer-mediated communication

for distance learning systems. It argues that linking deeply communication
to learning activities offers an interesting approach to develop the
efficiency of systems in facilitating and increasing discussions between
learners. To make this link, the author bases his work on various theories,
such as communication theories, situated cognition theory, and activity
theory. This theoretical study leads to research issues concerning a
contextual forum model. The description of the computing implementation
of this model aims at giving researchers some possible uses and
Human Communications and Distance-Learning Activities 103
recommendations in dealing with context-sensitive communication tools.

Finally, the chapter mentions futures trends and suggests emerging
research opportunities within the field of communication services that are
able to adapt dynamically to the user’s activity.
Introduction
In a distance-learning context, the emergence of learners’ communities has a

favourable impact on learning conditions. Indeed, in a socioconstructivist
approach (Doise & Mugny, 1984), interactions between learners play a
dynamic role to individual learning. However, distance discussion tools are not
always really suitable for the emergence of learners’ communities. Some works
have highlighted that these phenomena are too rare in distance-learning
environments (Gommer & Visser, 2001; Hotte & Pierre, 2002).
Forum tools currently used in online educational platforms are mostly unspecific
to educational situations (George & Hotte, 2003). The discussion activities are
not linked to the learning activities; consequently, this does not encourage the
learners to use them for communication. Current distance-learning systems do
not respect human communication process that is an “in-context” process
(Jakobson, 1960). Our main idea is then to make communication more
immediate during learning activities. The aim of this chapter is to describe the
conception of forum models and tools that are specific to distance-learning
systems. The research question lies in determining how to link discussion
activities to learning activities by the mean of well-suited computer tools. On the
whole, the work concerns the design of human communication systems that
attempt to respect human thought process. We totally agree with the paradigm
of “cognitively informed systems,” which defines systems that utilize, as a basis
for their design, some form of cognitive findings to enhance the effectiveness of
the systems in achieving their targets. For the design of our communication
system, some results from communication theories, situated cognition theory,
and activity theory are used to develop the efficiency of the system in facilitating
and increasing discussions between learners. These theoretical considerations
guide the system design toward a more effective presentation of discussions.
The chapter concerns the design of forum models and tools which aim at
promoting text-based asynchronous discussions during learning activities that
are not collaborative a priori. During individual learning activities, to only
104 George
provide usual communication tools is not always sufficient to create interactions

between learners and favour the construction of collective knowledge. Usual
communication tools could be appropriate if a collective learning activity is set
up, during a project-based learning, for example (George & Leroux, 2001).
Nevertheless, in distance education, all learning activities cannot be collabora-
tive, and the approach presented in the chapter aims at encouraging interactions
during individual activities that do not commit learners to a forced collaboration.
We propose a forum model, named CONFOR (CONtextual FORum), which
is based on two special features: contextual view and structuring. The contex-
tual view of the forum, always visible, allows the learner to focus on pertinent
discussions, that is, on messages that correspond to his/her activity.
Contextualization is common in annotation systems but not in forum tools.
Adding this feature to forums, the intention is to closely link communications to
learning activities. To provide this contextual view, the discussions need to be
structured. We suggest in this research two means of structuring: according to
(1) the content structure of a course and to (2) the cognitive structure of a
course.
This chapter starts with a discussion about some theories we rely on. This first
part leads to the research issues. Then, two ways to structure contextual forums
are detailed before proposing an approach to integrate them. Finally, we give
some results of an experiment and we mention lines of future trends.
Theories, Background, and

Research Issues
This section presents a literature review to support our position. We conclude

this section by stating our research issues.
Theories of Communication
Numerous theories of communication have been developed. The aim of this

part is not to make a complete state of the art of these theories but to underline
some aspects especially of interesting for the research.
According to the model of Jakobson (1960), all acts of communication, be they
written or oral, are contingent on six constituent elements: Addresser, Mes-
sage, Context, Contact, Code, and Addressee. The “context” element particu-
larly interests us because it defines the frame of the message reference, that is,
the ground on which a communication can occur. This referential function
emphasizes that communication is always dealing with something contextual.
Indeed, “When humans talk with humans, they are able to use implicit situational
information, or context, to increase the conversational bandwidth. Unfortu-
nately, this ability to convey ideas does not transfer well to humans interacting
with computers” (Abowd & Dey, 1999, p. 1). This notion of context is rarely
explicit during mediated communications, except by the addresser of the
message. One of our goals is to contextualize communication in e-learning
situations, that is, to link discussions to context of discussions.
Situated Action Theory
Works on communication theories can be complete with some other works

dealing with situated actions (Lave, 1988; Suchman, 1987). The term “situated
action” underlies the idea that each action closely depends on material and
social circumstances in which it occurs. Situated action theory introduces an
interesting idea: action is not the execution of a ready-conceived plan, but it is
the user’s adaptation to the context. Applying situated action theory to
computer mediated communication, Mantovani (1996) concludes that users
are social actors with their own aims and autonomy in situations, and it is
technology that must adapt to them. In this sense, “the most effective way of
clarifying the meaning of messages is to relate them to a shared context” (Riva,
2001, p. 217).
Thus, by extension, communication is a situated activity (Lambert, 1992). The
situated actions theory pointed out the fact that communications should occur
during the action, at the time when the user needs it. So if a user can’t
communicate easily in action, s/he will not do it. Our work aims at facilitating
communication in action.
Activity Theory
Another theory can throw light on the issues: the activity theory (Leontiev,
1978). This theory, based on the initial ideas of Vygotsky, emphasises how
knowledge is “socially constructed.” In this way, activities are integrated in a
social matrix composed by persons. Basing on this theory, Engeström (1987)
106 George
has suggested a framework for a collective activity. His popular representation

of activity theory adds to the triangle subject-object-artifact (Vygotsky, 1978)
another level containing rules, community, and division of labor. Thus, the first
relational triangle—the subject who achieves an object by using instruments
(e.g., computer tools)—is then widened to show that a subject is not lonely but
is a part of a community.
The activity theory is not only useful to describe humans’ behaviour during
collective activities but also makes it possible to conceptualise collective
learning. Thus, these principles can offer a framework to analyse and conceive
educational environments (Lewis, 1997; Roschelle, 1998). The next section
will come back to the utility of this approach to better understand communica-
tion activities during individual learning activities.
Research Issues
We can use the representation of human activity of Engeström (1987) to

illustrate the two activities that interest us in this research: individual learning
activity and discussion activity (Figure 1).
Figure 1. Linking learning activity and discussion activity through activity

theory
We remind the reader that we focus our work on learning activities that are a
priori defined to be individual but with possible and desirable communications
between learners. Figure 1 shows that, currently in this situation, learning
activities and discussion activities are not linked. On the one hand, the learner
uses different tools to see learning objects in order to do a particular learning
activity (small triangle on the left part of Figure 1). On the other hand, the learner
can use communication tools (as forum) to discuss with other learners of the
course (big triangle on the right part of Figure 1). In our case, the community
does not have imposed communication rules or recommendation to divide the
labour. Within this configuration, the only link between learning and discussion
activities is the learner. To see the emergence of a community, it supposes that
learners have the capacity to link themselves the two activities. We have
symbolized by a question mark the point that seems important to us and that is
currently missing. It could be seen as a synchronisation point between learning
and discussion activities. In other words, we aim at studying communication
artefacts as integrated and inseparable components in human learning activity.
How to merge the two triangles together?
Starting from these considerations, the work consists in finding a solution to
deeply integrate communication into distance learners’ activities. Learning
activities are taken in a broad sense, such as reading an electronic document,
doing an exercise, or using a simulation. On the one hand, we want to make
communication more situated in distance-learning system and, on the other
hand, we aim at defining more explicitly the context of each discussion.
We focus on one kind of communication tools in distance-learning systems:
forums, which are tools for asynchronous communications. The research
question lies in determining a model of contextual forum and to develop
computer tools based on this model. Currently, we have studied two means to
contextualise discussions, which are described in the following section.
Work on Contextual Forum
In this section, we discuss two kinds of contextual forums structuring studied

in our work. The first one is based on educational scenarios and the second one
is a knowledge-based forum structure. We propose then a solution to integrate
these two approaches. Finally, we mention current results of our research on
contextual forums.
108 George
Figure 2. Screen shot of CONFOR
Contextual Forum Based on Educational Scenarios
In the first approach of forum structuring, each forum thread is linked to an item
of the content navigation of an online course. For that, educational content may
be sequenced for the learner: “the branching and flow of that content may be
described by a predefined set of activities, typically defined at design time”
(Dodds, 2003, p. 13). In the model we propose (George & Hotte, 2003) each
root message of the forum is a reference to a learning activity. Thus, a reference
could be, for example, the title of a course chapter or the number of an exercise.
The forum is then hierarchically structured according to learning activities, by
reference to the course structure. According to contextualization seen above,
the opening of an educational object leads to the opening of a forum partial view
corresponding to the activity in progress. With these references, the goal is to
focus learners’ exchanges on learning objects.
The interface of the CONFOR tool is shown in Figure 2. The upper part of the
window contains a learning activity of an online course. Under this course is the
contextual view of the forum, which is automatically updated depending on the
upper part. For instance, in Figure 2, a learner carries out the activity 2.2 of the
module 2 of his/her course and s/he sees, at the same time, the messages of the
forum that correspond to this activity (messages under the reference “activity
2.2”). This contextual view of the forum is a part of a unique global forum. This
should be noted that this global forum can be displayed in a global view (to see
the entire tree of messages). In both views—contextual or global—the left part
of the forum displays the list of the message titles and the references names.
When the user clicks on a message title, the content of the message is displayed
on the right part of CONFOR. The forum can be resized or put in an “always
on top” window.
In order to provide the contextual display of the forum, we have to define
references in connection with the online course structure. In this model,
references contained in the forum are linked to the learning activities structure.
So references are dependent on the educational scenario designed by the
author of the course. The question is then to determine how to add these
references before the course starts. A solution is to give the possibility of
manually inserting references. For that, we propose a designer interface in
CONFOR allowing this manual definition of references. Each reference is
defined by a name and a link. The name will appear in the forum and the link
is the reference to the educational object or to the learning activity (e.g., an
URL). Moreover, references are linked together in a hierarchical manner.
Nevertheless, this manual definition of references can become a hard work if
the course is large. For this reason, we also suggest an automatic procedure to
add references in the forum. For instance, an automatic import procedure has
been done for educational scenarios described with SCORM (Sharable
Content Object Reference Model).
Knowledge-Based Contextual Forum
From the previous work, an issue emerges: it would be a good idea to propose
a different structuring of forum, by defining references in connection with
knowledge dealt in online courses. From a first experiment of CONFOR (cf.
the last part of this section), we observed that two messages could be situated
in two different threads even though these messages dealt with the same content
or with the same knowledge. So the goal became to design a structuring model
based on knowledge representation while keeping the contextual view of the
forum.
We studied various taxonomies which make it possible to describe knowledge
elements approached in learning documents. In particular, taxonomies used by
110 George
libraries have been examined. Among those, we chose the Dewey Decimal
Classification (DDC) because it is flexible, simple to use and allows a
classification of knowledge sufficiently fine for our work. The DDC is a
knowledge organisation method which is used worldwide. It is universal
because it uses numerical indexes to classify documents and, thanks to its
international standardized notation, it is alphabet and language independent.
We chose to use the DDC in conjunction with the Learning Object Metadata
(LOM) (IEEE, 2002). In online education, the metadata are used to describe
courses and learning objects. They include a number of descriptors which are
defined according to a standard, so these courses and learning objects are more
easily accessible and usable: interoperable, reusable, long lasting, adaptable
(Downes, 2001). LOM contains nine categories in order to describe educa-
tional resources, but according to our objectives, the ninth category, named
“classification,” is the category that particularly interests us for contextual
forum. This field ensures the classification and indexing of educational objects
according to knowledge taxonomies such as the DDC.
However, providing metadata is not always easy work for some authors of
learning objects. From their point of view, this work requiring a literature study,
which is not always in their field of competences, is tiresome and nonproduc-
tive. We believe that it will be one of the major problems for the development
of educational objects. However, we take as a working hypothesis that to use
CONFOR, each learning object will be well documented and described with
LOM. If this work is not done by authors, information specialists could do it.
Then we suggest a model of knowledge-based forum. In this model, the topics
are organised according to a structure defined by the knowledge tackled in a
course. With the attribute “classification” of the LOM description of each
educational object, the knowledge elements being consulted at a time can be
identified. Therefore, a forum function can show in a contextual way all the
topics corresponding to these knowledge elements (George, 2004). The
learner may then consult, share, and interact with other learners about the
knowledge of the course. An advantage of this mechanism is that two students
who work on two different learning objects will be able to meet on the same
thread to discuss a common knowledge item.
Toward an Integrated Approach of Contextual Forum
Our current research concerns the integration of the two models presented
above. Actually, using a singular approach has some limitations. In the first
Figure 3. An integrated model of contextual forum
approach, contextual forum based on educational scenarios, some messages

could be situated in different threads even though these messages dealt with the
same content. In the second approach, knowledge-based contextual forum,
general discussions about learning activities have no place in the knowledge
structure.
The idea of integration consists of showing the learners a discussion thread
corresponding to the current activity (e.g., to discuss about the organisation
inside the course) and also the discussion threads corresponding to knowledge
at stake at a time (in order to discuss the content). Figure 2 represents a model
that takes into account these two levels of contextual discussions.
In this model, an educational object—or a resource—is referenced as an
object of an educational scenario (in the upper part of Figure 3) and this object
also deals with several knowledge elements described in its metadata (in the
lower part of Figure 3). Knowledge elements could be defined by an ontology
of a particular field or by a taxonomy such as Dewey (DDC). Always in this
model, each circle is then a discussion topic inside the forum. So when a learner
opens an educational object, the contextual forum displays automatically the
activity topic and all the knowledge topics linked to the resource.
112 George
Current Results
An assessment has been carried out at the Tele-university of Quebec to

evaluate the contextual forum. CONFOR was assessed within an introductory
course on “training in the workplace” offered to students registered in an
undergraduate certificate program in business-oriented computer sciences.
The CONFOR version used for this experiment was the first version, with the
forum structure only based on educational scenarios. Nevertheless, this experi-
ment provides interesting information.
The experiment lasted 8 months and about 70 students took the course. During
the evaluation, two different tutors supervised the students. The goal of the
assessment was to study the use of CONFOR. More specifically, we wanted
to test the utility and usability of the forum contextual display. For the purposes
of the assessment, we used questionnaires, interviews, and regular observa-
tions coupled with computer traces analysis. It should be pointed out that the
course used to assess CONFOR was newly offered by the Tele-university. For
this reason, we will not directly compare the use of CONFOR with that of other
forums used at the Tele-university, too many parameters being different.
The questionnaire responses indicate that the students are quite appreciative of
the reference-based structure of the forum. Similarly, even if a global view of
forum was provided, they favour the forum’s contextual display (4.5 more
messages opened in the contextual view than in the global view). Furthermore,
the contextual view favours the sending of messages (7.5 more messages sent
in the contextual view than in the global view). The results also indicate that
CONFOR helps students in finding messages relevant to their activities, that is,
messages useful for the learning activity they are engaged in. Finally, students
found that the forum fostered the organisation of discussions. Since forums are
also an important tool for e-learning tutors, we conducted semistructured
interviews with them to obtain information on their use of CONFOR. Tutors
found the interface simple and intuitive to use. Concerning utility, tutors
appreciated having the forum and the course on the same page. They appre-
ciated the ease of locating new messages, which facilitated their monitoring
activities.
At this point, we can thus conclude that contextualisation of discussions for
learning activities is appreciated. Users also seem to appreciate the fact that
communication and learning are integrated into a single space. Having access
to the opinions of others, as they carry out their learning activities, motivates
students to locate discussions that help them to understand and to build their
knowledge. From this point of view, we can contend that this kind of forum has
a positive effect on learning.
Conclusion and Future Trends
There is an increasing need for context-aware services and applications that are
able to adapt dynamically to the user’s activity. Dey (2000) states that “A
system is context-aware if it uses context to provide relevant information and/
or services to the user, where relevancy depends on the user’s task” (p. 6). In
other words, context-sensitive applications are those that respond to changes
in their environment. Some research studies have been done in this direction,
for example, in the field of mobile and ubiquitous computing. The general aim
is to make information more relevant to the situation in which it is being used.
Mechanisms to provide context-sensitive help are common examples.
We introduce in this chapter a new idea that consists of defining a context-
driven support communication. Our research aims at proposing specific forum
models and tools for online education. The work led to the idea of contextual
display of forum messages. We suggest two versions of contextualisation. The
first one is based on a forum structuring according to online course structures.
Some results of an experiment led us to study another forum structuring, by
taking into account the cognitive structure of a course. The result is a discussion
tool that displays to the learner an “activity topic” and several “knowledge
topics” linked to the learning resource that is open. Currently, the context is
limited to the activity in progress and to concepts studied at a time. We wish
to extend the notion of context, taking into account more parameters such as
a learner’s history or learner’s goals. Then, we could use this information to
better adapt displayed topics to each user. For example, carrying out the same
activity, two learners would see different and specific discussions topics
according to their past actions and to their personal characteristics.
We can also mention some limits of contextual communication tools. As pointed
by Dimitracopoulou and Petrou (2004) and take up by Gao, Baylor, and Shen
(2005), “the problem with the systems that contain embedded communication
tools is that discussions are usually fragmented by artefacts, which causes
learners to lose a holistic view of the discussion and the relationships between
different aspects of the artefact” (p. 76). In our case, we try to reduce this effect
by also providing a global view of forum discussions. We believe that in
114 George
developing contextual communication tools, it should be suitable to give several

ways to enter and to read the discussions.
Furthermore, we only study at this time a context-sensitive asynchronous
system but we want to extend the mechanism to synchronous discussions
(“contextual chat”). A future trend will be to no longer see forum or chat tools
simply as a communication tools, but also as tools helping to put users in touch
with others. These kinds of context-aware communication tools will bring users
together depending on their interests, motives, or needs.
Finally, we do not believe in completely generic context-sensitive applications.
Context-gathering mechanisms could not be totally generic. In the case
presented in this chapter, the context-gathering mechanism is adapted to the e-
learning situation even if the global model is generic. We could easily adapt it
to a computer-supported cooperative work system, for example, but the
sensors would not be the same. Context-driven support communication will be
really pertinent only if situations are well defined and if users’ activities are
circumscribed.
References
Abowd, G. D., & Dey, A. K. (1999). Towards a better understanding of

context and context-awareness. In Panel at the 1st International
Symposium on Handheld and Ubiquitous Computing, HUC’99,
Karlsruhe, Germany (pp. 304-307). Berlin: Springer.
Dey, A. K. (2000). Providing architectural support for building context-
aware applications. Unpublished Computer Science thesis, Georgia
Institute of Technology, Atlanta.
Dimitracopoulou, A., & Petrou, A. (2004). Advanced collaborative distance
learning systems for young students: Design issues and current trends on
new cognitive and metacognitive tools. THEMES in Education, Interna-
tional Journal.
Dodds, P. (2003, October 22). The SCORM Content Aggregation Model,
version 1.3. Report of the Advanced Distributed Learning Initiative.
Alexandria, VA: ADL.
Doise, W., & Mugny, G. (1984). The social development of the intellect.
New York: Pergamon Press.
Downes, S. (2001). Learning objects: Resources for distance education

worldwide. International Review of Research in Open and Distance
Learning, 2(1). Retrieved May 10, 2005, from www.irrodl.org/content/
v2.1/downes.html
Engeström, Y. (1987). Learning by expanding: An activity-theoretical
approach to developmental research. Helsinki: Orienta-Konsultit.
Gao, H., Baylor, A. L., & Shen, E. (2005). Designer support for online
collaboration and knowledge construction. Educational Technology &
Society, 8(1), 69-79.
George, S. (2004). Contextualizing discussions in distance learning systems.
Proceedings of the 4th IEEE International Conference on Advanced
Learning Technologies (ICALT 2004), Joensuu, Finland.
George, S., & Hotte, R. (2003). A contextual forum for online learning.
Proceedings of the International Conference on Open and Online
Learning (ICOOL 2003), University of Mauritius, Reduit.
George, S., & Leroux, P. (2001). Project-based learning as a basis for a
CSCL environment: An example in educational robotics. Proceedings of
the First European Conference on Computer-Supported Collabora-
tive Learning (Euro-CSCL 2001), Maastricht, The Netherlands.
Gommer, L., & Visser, G. (2001). Implementation of a digital learning
environment: The real results. Proceedings of the World Conference on
the WWW and Internet (WebNet 2001), Orlando, FL.
Hotte, R., & Pierre, S. (2002). Leadership and conflict management support
in a cooperative telelearning environment. The International a Journal
of E-learning (IJEL), 1(2), 46–59.
IEEE. (2002, June 15). Draft standard for learning object metadata.
Institute of Electrical and Electronics Engineers, Inc. Ne w York.
Jakobson, R. (1960). Closing statement: Linguistics and poetics. In T. A.
Sebeok (Ed.), Style in Language (pp. 350–377). Cambridge, MA: MIT
Press.
Lambert, B. L. (1992). A connectionist model of message design. Unpub-
lished Philosophy in Speech Communication thesis, University of Illinois,
Urbana-Champaign.
Lave, J. (1988). Cognition in practice. Cambridge, UK: Cambridge Univer-
sity Press.
116 George
Leontiev, A. N. (1978). Activity, consciousness, and personality. Englewood

Cliffs, NJ: Prentice-Hall.
Lewis, R. (1997). An activity theory framework to explore distributed commu-
nities. Journal of Computer Assisted Learning, 13(4), 210–218.
Mantovani, G. (1996). Social context in HCI: A new framework for mental
models, cooperation and communication. Cognitive Science, 20, 237–
296.
Riva, G. (2001). Communicating in CMC: Making order out of miscommuni-
cation. In L. Anolli, R. Ciceri, & G. Riva (Eds.), Say not to say: New
perspectives on miscommunication (pp. 203–233). Amsterdam: IOS
Press.
Roschelle, J. (1998). Activity theory: A foundation for designing learning
technology? Journal of the Learning Sciences, 7(2), 241–255.
Suchman, L. A. (1987). Plans and situated actions. Cambridge, UK:
Cambridge University Press.
Vygotsky, L. S. (1978). Mind in society: The development of highest
psychological processes. Cambridge, MA: Harvard University Press.
Noninvasive Adaptation of Metaphors 117
Chapter VI
Toward Noninvasive
Adaptation of
Metaphors in Content
Alexei Tretiakov, Massey University, New Zealand
Roland Kaschek, Massey University, New Zealand
Abstract
Web information systems (WIS) can be considered as media. These media

implement a tool language that enables access to content. Accessing that
content aids users in achieving their goals. The language in which that
content is given nowadays is often natural language or very close to it.
Consequently the content involves metaphors. As the Web is open to
virtually everyone, the users of WIS are likely to differ from each other
with respect to ethnicity, language, gender, age, culture, education,
preferences, physical or mental handicaps, and so forth. Consequently
users are likely to respond differently to the metaphors occurring in the
content. This chapter, therefore, proposes an approach to adapting the
employed metaphors to user types for improving the value that the WIS
offers to its users. This is expected to result in both increased user
118 Tretiakov & Kaschek
acceptance and number of business transactions. Therefore, an increased

return on investment for the WIS is expected as well. We propose a
conceptual model for user type and context-aware mapping of concepts
in a target domain to concepts in a source domain. The respective
mapping is used for modeling metaphors. We formalize that mapping in
terms of the Higher-Order Entity-Relationship Modeling (HERM) language
and in the Web Ontology Language (OWL). The conceptual model we
provide can be used as a basis for hot generation of content representation
such that the metaphors occurring in the content are adapted to the types
of the users interacting with the WIS. As a step toward implementation, we
formulate a high-level architecture that enables us to noninvasively adapt
the metaphors in the WIS content to the types of the users. We report our
experiences regarding exploration of the feasibility of the architecture.
These experiences result from implementing a prototype that shows how
metaphors—in a noninvasive manner, that is, without changing its code
or content—can be added to the contents of an already-existing WIS. The
chapter is completed by presenting the results of formal user evaluation
which demonstrate the user acceptance of the respective metaphors.
Introduction
According to Hirschheim, Klein, and Lyytinen (1995, p. 11), Langefors

defined the term “information system” from a functional perspective as “a
technologically implemented medium for the purpose of recording, storing, and
disseminating linguistic expressions as well as for the supporting of inference
making.” According to Kaschek, Schewe, Wallace, and Matthews (2004), a
Web information system (WIS) is understood as an information system that is
integrated in the Web and uses it as a resource. Integration of an information
system into the Web makes it open in the sense that virtually everyone can
access it. The extent to which system developers know the system users at
development time is therefore larger with respect to WIS than it is with respect
to information systems in general. User typing and adaptation to users, thus,
must be taken more serious in case of WIS than in case of traditional
information systems.
We slightly modify a classification of the languages used to interact with a WIS
we found in Wallace and Matthews (2002) as “tool language,” “domain
language,” and “metaphor language.” These languages are used for explaining
to the user the semantics of the WIS functionality, the domain regarding which
the WIS can aid in problem solving, and the metaphors used in the WIS user
interface or the domain. In any of these three languages metaphors might be
used. We target the metaphor use in the domain language, as that is the language
in which the content will be expressed. Metaphor use with respect to WIS has
been discussed by Nielsen (2000, p. 190). Nielsen assesses the metaphor use
in several WIS examples as negative. His assessment, however, does not apply
to our work, as he discusses metaphors as tool in WIS interface design, which
means that a metaphor is chosen as guideline for both the overall system design
and the system interface. Nielsen specifically mentions that such a metaphor
might cover a part of the WIS functionality well, but might be poorly cover other
parts. That remark does not apply to using metaphors as understanding aid
applied to content, as several metaphors can be used easily. It is clear,
however, that a sensible use of metaphors must be targeted and is not trivial to
achieve. Note that other sources, such as Thalheim and Düsterhöft (2000),
recommend using metaphors as design guideline. That makes the findings
inconclusive that we know of regarding using metaphors as a design tool.
That inconclusive state of the literature should not make one forget how
important metaphors are for modern computing. In a Microsoft Windows-
based environment, for example, one makes use of metaphors that are
connected to the drag-and-drop style of handling files in case of copying or
deleting them. Some of the basic concepts of modern computing such as “file”
are or were metaphorical (see Murray Hopper, 1981, p. 16). The functional
view definition of the term “information system” insofar as system implementa-
tion is concerned is based on a metaphor. A universe of discourse is repre-
sented as a particular part of a state of a computer. Such a state is a pattern of
electrical current, magnetization, capability to polarize light, or similar. That
representation can be manipulated according to rules that may depend on the
universe of discourse. After respective manipulations, the pattern part is
interpreted in terms of the universe of discourse and notified to the user in terms
of suitable linguistic expressions. What actually happens “inside” a computer is
beyond of what can be understood by most of the users. However, experts in
the fields mentioned understand these processes to such an extent that they can
even put in place automated procedures for translating linguistic expressions
into the patterns mentioned and vice versa. A broad conceptualization of
metaphor as “understanding one domain in terms of another one” here is
superposed by a division of labor.
The chapter is organized such that next, we provide more discussion on

metaphors. Then, we discuss our conceptual model of a user-type depending
mapping of concepts in a source domain into a target domain. After that, we
discuss our architecture of a noninvasive add-on for adaptation of metaphors
to user types. We then discuss an OWL schema for such an adaptive concept
mapping. The paper concludes with a case study, a conclusions section, and
our references.
More on Metaphors
While historically metaphor was the topic of literary studies and rhetoric,
recently, conceptual metaphors started to play an important role in the fields of
cognitive sciences (in particular, cognitive linguistics; see, e.g., Janda, 2000;
Evans & Green, 2005) and artificial intelligence (Nehaniv, 1999). In addition,
in the second half of the 20th century, metaphors became a standard topic in
philosophy of science (Bailer-Jones, 2002). The Oxford English Dictionary
Online defines metaphor as a “figure of speech in which a name or descriptive
word or phrase is transferred to an object or action different from, but
analogous to, that to which it is literally applicable; an instance of this, a
metaphorical expression.” Similarly, more general and a bit more fuzzy defines
(O’Grady, Archibald, Aronoff, & Rees-Miller, 2001) metaphor as “the
understanding of one concept in terms of another.” A more recent text, that is,
Kövecses (2002), that is wholly dedicated to metaphors agrees with that and
proceeds with defining “[a] conceptual metaphor consists of two conceptual
domains, in which one domain is understood in terms of the other. A conceptual
domain is any coherent organization of experience.... The conceptual domain
from which we draw metaphorical expressions …” (i.e., words or other
linguistic expressions) “is called source domain, while the conceptual domain
that is understood that way is called target domain” (p. X).
Effective use of metaphors thus constitutes reusing knowledge regarding a
particular source domain. Metaphors are known as an effective cognitive tool
for improving end-user experience by facilitating the creation of mental models
of the system (Comstock & Duane, 1996). For a metaphor to be effective, two
conditions need to be met. First, its recipient needs to be capable of identifying
the metaphor. Second, the recipient needs to be sufficiently familiar with the
source domain, so that s/he can effectively draw from it, using the knowledge
and, more importantly, the intuition gained for concepts in the source domain
to infer the structure of the target domain. One may utilize everyday life as the
source domain. However, it is not uncommon to rely on the user’s prior
knowledge of a specialist area. For example, for the target domain of
economics, it is common to use the following metaphors sourced in the domain
of physical sciences: chain reaction (chain reaction of bankruptcies), pendulum
(economic cycles as a pendulum), diffusion (technology diffusion), equilibrium
(equilibrium of supplies and requests), and so forth.
Metaphors utilizing a source domain unfamiliar to the user are likely to result in
confusion or even create misconceptions. For example, the “smiling cat”
(maneki-neko) metaphor occasionally used in Japanese e-business WIS is
unlikely to be effective for a North American user, who might form an
inadequate mental model (e.g., of a “pet shop” rather than of “a small
business”). Clearly, to leverage the power of metaphors fully, these need to fit
to the targeted recipient’s characteristics, such as culture and background
knowledge.
Two different approaches might be considered for having effective metaphors
in place. First, given the projected audience of a Web-based system the
metaphors optimally fitting it might be identified and hard coded as part of the
interface provided by the system. This is the approach used in classical human-
computer interaction (Carroll, 2003). Second, metaphors may be added to the
Web-based system content dynamically at execution time. We focus on the
latter approach, as it offers the flexibility required for supporting the increas-
ingly multiage, multibackground, and multicultural WIS user community.
The view on metaphor adopted in this chapter is very close to the one taken by
most researchers in the emerging field of cognitive linguistics. In particular, we
stress the importance of the conceptual substrate formed by the situational and
cultural context in which metaphors are applied. This leads us to the necessity
to explicitly take into account the user type and the wider context when using
metaphorical expressions in WIS contents. In our view, learning to use a WIS
entails the same processes as learning new words and expressions in a
language. Cognitive linguists believe that for understanding language structure
and dynamics, one should consider the cognitive processes underlying and
determining language use (Evans & Green, 2005). That view according to Pena
Cervel (2003, p. 27) is adopted from Lakoff and Johnson who in 1980 wrote,
“metaphor is typically viewed as characteristic of language alone, a matter of
words rather than thought or action.… We have found, on the contrary, that
metaphor is pervasive in everyday life.… Our ordinary conceptual system, in

terms of which we both think and act, is fundamentally metaphorical in
nature.…” The latter point is also made by O’Grady, Archibald, Aronoff, and
Rees-Miller (2001, p. 255): “Many people think of metaphor as a literary
device reserved for the use of authors and poets. In fact, however, it has a
prominent place in the conceptual system shared by all human beings.”
According to Kuhn (1993, p. 539), “Metaphor plays an essential role in
establishing links between scientific language and the world. Those links are
not, however, given once and for all.… The earth was like Mars (and was thus
a planet) after Copernicus, but the two were in different natural families before.”
We have to observe that in some essential aspects, the use of metaphorical
expressions in WIS in general and in e-learning WIS in particular is similar to
their use in model building in natural sciences. As in case of scientific models
(Bailer-Jones, 2002), there are clear quality criteria for WIS in general and in
e-learning in particular. The best, the most suitable metaphorical expressions
are the ones that would allow users to achieve their goals (such as attaining goals
in learning and training) as fast as possible, with the minimal effort. On the other
hand, we acknowledge that while in case of scientific models the adequate
description of phenomena under study is the only applicable criterion, in case
of WIS (and e-learning) apart from achieving immediate goals, secondary
considerations, such as brand building or motivating lifelong study, may be of
importance. In terms of these secondary considerations, metaphorical expres-
sions applied in WIS could be more akin to the ones applied in poetry than to
the ones applied in natural sciences.
The ongoing attempts to build human-like mechanisms (robots) and software
agents (e.g., avatars) have led to attempts to formalize the formation and the use
of metaphorical expressions, for example, in terms of algebraic structures
(Goguen, 1999). However, so far there is no generally established approach to
such formalization. In this chapter, we opt to formalize and to implement the
metaphor use as a straightforward mapping, thus following very closely the
mainstream views established in cognitive linguistics and in philosophy of
science. For similar reasons, we do not apply the theory of blends, suggesting
that metaphorical expression are formed in “blend domains” formed by mixing
source and target domains (Veale, 1999); we do not believe that the additional
complexity is justifiable for the purpose of building adaptive WIS interfaces,
although it does provide additional insight in certain linguistic phenomena.
Matching Target Domain Concepts to

Source Domain Concepts
Following Kaschek, Schewe, Wallace, and Matthews (2004), we conceptu-

alize a WIS such that it creates an information space ant that this space contains
information objects some of which might be connected by means of traversable
links. According to the Longman Dictionary of Contemporary English, the
context of something is “the situation, events, or information that are related to
…” that “something, and help you to understand it better.” In a simplifying view,
the context of an information object (i.e., an information chunk on a Web page)
is understood as the information objects connected to it. More advanced
aspects of context in WIS have been discussed in Bineman-Zdanovicz,
Kaschek, Thalheim, and Schewe (2004).
The HERM-diagram in Figure 11 formalizes the requirements for user-type–
aware mapping of target domain concepts to source domain concepts. Regard-
ing the HERM concepts and notation, see, for example, Thalheim (2000). We
limit ourselves to metaphors obtained by relating a concept in the target domain
to a concept in the source domain, and do not consider complex metaphors
realized by coordinated use of multiple user interface elements.
The data required for creating a metaphor are tied together by an instance of
the relationship type “Grounds.” The context in which the target domain
concept is situated is taken into account by including the “Target domain
concept context” (abbreviated as X) entity type. Obviously, w could include
the context concept into making metaphor adaptation depending of the user’s
virtual location in the information space, that is, the Web page currently visited.
However, one could introduce other interpretation of what the target domain
context is, the X entity type serving as an extension point for further adapting
metaphor mapping. The “Explanation” (E) entity type accounts for an explicit
description of how the target concept relates to the source concept used to
clarify it. The “User type” (U) is accounting for a group of users with similar
culture and background knowledge. For a recent paper on user typing, see, for
example, Kaschek, Schewe, Thalheim, Tschiedel, and Kuss (2004).
The HERM-diagram in Figure 11 formalizes the requirements for user-type–
aware mapping of target domain concepts to source domain concepts. Regard-
ing the HERM concepts and notation, see, for example, Thalheim (2000). We
limit ourselves to metaphors obtained by relating a concept in the target domain
to a concept in the source domain, and do not consider complex metaphors

realized by coordinated use of multiple user interface elements.
The data required for creating a metaphor are tied together by an instance of
the relationship type “Grounds.” The context in which the target domain
concept is situated is taken into account by including the “Target domain
concept context” (abbreviated as X) entity type. Obviously, w could include
the context concept into making metaphor adaptation depending of the user’s
virtual location in the information space, that is, the Web page currently visited.
However, one could introduce other interpretation of what the target domain
context is, the X entity type serving as an extension point for further adapting
metaphor mapping. The “Explanation” (E) entity type accounts for an explicit
description of how the target concept relates to the source concept used to
clarify it. The “User type” (U) is accounting for a group of users with similar
culture and background knowledge. For a recent paper on user typing, see, for
example, Kaschek, Schewe, Thalheim, Tschiedel, and Kuss (2004).
Lookup constraints annotating the diagram are essential for understanding its
semantics (see Thalheim, 2000, for a detailed explanation of lookup constraints
notation). Here we describe in plain English the restrictions imposed by the
lookup constraints in Figure 1:
Figure 1. Matching target domain concepts to source domain concepts
Explanation
(E) a) look T , X , U , S (0, m)
Source Target
b) look S , U , T (0,1)
domain domain
concept Grounds concept
(S) (T)
c) look T ,U , S , E (0,1)
Target
d) look E , T , U , S (1,1)
User domain
type concept
(U) context
e) look T , S , U (0, m)
(X)
Figure 2. Substitutability of source domain concepts
From Target
Source domain
domain Substitutes concept
concept To (T)
(S)
Target
User domain
type concept
(U) conteXt
(X)
look T , X , L , From, To (0, m)
a. For a given user type, target domain concept, in its context, may or may
not be representable by a suitable source domain concept (hence, a
suitable source domain concept is not always available). Furthermore,
more than one suitable source domain concept may be available.
b. For a given user type, a source domain concept may illustrate no more than
one target domain concept. Hence, source domain concepts utilized in
metaphors can be unambiguously traced back to the target domain
concepts they represent.
c. There can be no more than one explanation for a given combination of user
type, target domain concept, and source domain concept.
d. Each explanation relates to one, and only one, combination of user type,
target domain concept, and source domain concept.
e. A given combination of a target domain concept and a source domain
concept may occur several times, with different user types.
Note that the only role of context (of a target domain concept) is to differentiate
between source domain concepts matching the target domain concept. We
envisage applications that would ignore the context or use it as a refinement step
(providing a source domain concept matching the context if one is available,
otherwise ignoring the context and attempting a match just to the target domain
concept and to the user type). On the other hand, a situation when no
“Grounds” instances are available in a given context can be used to indicate the
context in which the use of a metaphor is not desirable.
It should be noted that source domain concepts matching a given combination

of a target domain concept, its context, and a user type when more than one
such concept is available, are not differentiated in Figure 1, and can be
substituted for each other. This is reflected in Figure 2, which is not indepen-
dent, but is a direct consequence of Figure 1 (with lookup constraints taken into
account). Substitutable concepts may be differentiated by the system to take
into account individual user preferences, although the data model in Figure 1
would have to be enhanced to form a basis for such differentiation.
A given pair of source and target domain concepts can be related to several
instances of an Explanation, distinguished by the user type. For example, for a
user type involving users very familiar with the source domain, the explanation
can be concise. On the other hand, for a user type involving users not familiar
with the source domain, the explanation may cover some aspects of the source
domain (such an explanation is warranted if the metaphor in question is a very
common one, so that it almost constitutes part of the target domain knowledge).
Alternatively, for a user type involving users with no knowledge of the source
domain, it may be the case that no “Grounds” instances are provided, so that
for them, the metaphor is not available, and thus, they are spared from
confusion and conceptual overload. Finally, note that Figure 1 allows chaining
of metaphors. That enables a source domain concept of a metaphor to be
considered as the target domain concept of another metaphor.
An Architecture Enabling
Metaphor Adaptation
In human-computer interaction, traditionally metaphors are considered as

design tool in the sense that they guide the overall architecture and design. The
drawback of this approach is that to add metaphor support to an existing
system, it may be necessary to redevelop the system. This may be unfeasible
since one might not have the access required. Also, if metaphors form an
integral part of a Web-based system, it increases the complexity, and thus
increases the costs of both development and maintenance. Here, we propose
an alternative, noninvasive approach focusing on WIS content. That approach
relies on adding adaptive metaphor support as a separate architectural layer by
using the decorator design pattern (Gamma, Helm, Johnson, & Vlissides,
Figure 3. Adding metaphors to base service
1 : request +
metaphors 2 : request
:Metaphor-
:Base service
enabler
4 : response + 3 : response
metaphors
Figure 4. Dependencies in the noninvasive architecture
Metaphor
Base service
enabler
1995). Henceforth, we occasionally refer to this layer as “metaphor support

layer.”
In the basic architecture we identify two main components (see Figure 3). We
call the component providing the basic user service a “base service compo-
nent.” A “Base service” component could be a legacy system that is usable on
its own. Another component, the “Metaphor enabler,” intercepts the user’s
communication with the system providing the base services, and performs data
and control transformations outlined in the previous section. The metaphor
enabler may need to access session context and user-type information in the
base service component or it may provide session and user-type management
of its own.
While the metaphor enabler depends on the base service component, the base
service component is unaware of the metaphor enabler, as illustrated in Figure
4. Thus, the proposed architecture is a noninvasive approach to adding
metaphor support to WIS. The services implemented by the WIS do not need
to cooperate in any way to allow metaphor adaptation.
The risks associated with the architecture stem from the fact that the metaphor
enabler depends rather strongly on the base service. Depending on the
implementation approach taken, changes in the base service may need to be
matched by changes in the metaphor enabler. Therefore, care should be taken
when deciding on the implementation approach to control the degree of
brittleness arising from the dependency.
An OWL Schema for

Adaptive Concept Mapping
The Web Ontology Language (OWL) is a language designed to formally

describe the meaning of terminology used in Web documents (World Wide
Web Consortium, 2004). OWL serves (along with RDF) as an underlying
technology of the Semantic Web (World Wide Web Consortium, 2005). It is
intended for use when the information contained in documents needs to be
processed by applications, as opposed to situations where the content only
needs to be presented to humans. Clearly, the purpose of OWL is highly
relevant to implementing metaphor mapping in context of the WWW.
An OWL schema implementing the conceptual data model of Figure 1 is
provided in the Figure 5. The “Grounds” relationship type is represented by the
AdaptiveRepresentationInContext OWL class. All entity types are repre-
sented as OWL classes with similar names. Instances of
AdaptiveRepresentationInContext are connected to the instances of classes
representing the entity types by using the OWL properties formalism. We
named properties to ensure readability: an instance of the TargetConcept is
represented by an instance of the AdaptiveRepresentationInContext class.
AdaptiveRepresentationInContext is given by an instance of SourceConcept,
whenUserTypeIs an instance of UserType, and whenConteXtIs an instance of
a ConteXt.
One should note, however, that OWL is not sufficiently expressive for
formalizing all of the constraints implied by the HERM diagram in Figure 1
including the lookup constraints. Therefore, we chose not to use OWL
cardinality constraints, and to assume that the constraints are to be enforced by
the application, using a mechanism external to OWL.
An OWL-based implementation would be highly suitable to support a situation
in which some of the class and property instances are available over the
Internet, with URI serving as object IDs in OWL corresponding to actual
physical locations. In particular, we envisage that the vocabulary for concepts
(both in target and in source domains), and the vocabulary for user types, could
be reused this way. On the other hand, the vocabulary relating to context is
likely to be system specific and to have little potential for reuse. As instances
of AdaptiveRepresentationInContext and of all OWL properties using
AdaptiveRepresentationInContext class as the domain are context dependent,
their reuse potential is also limited. A better reuse potential may be realized by
Figure 5. An OWL schema implementing the adaptive concept mapping

for metaphor enabling
<?xml version="1.0" encoding="UTF-8"?>

<rdf:RDF
xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
xmlns:owl="http://www.w3.org/2002/07/owl#"
xmlns:xsd="http://www.w3.org/2000/01/XMLSchema#"
xml:base="http://isrc.massey.ac.nz/adaptifity/metaphors"
>
<owl:Class rdf:ID="SourceConcept"/>
<owl:Class rdf:ID="TargetConcept"/>
<owl:Class rdf:ID="UserType"/>
<owl:Class rdf:ID="ConteXt"/>
<owl:Class rdf:ID="Explanation"/>
<owl:Class rdf:ID="AdaptiveRepresentationInContext"/>
<owl:ObjectProperty rdf:ID="forTargetConcept">
<rdfs:domain rdf:resource="#AdaptiveRepresentationInContext"/>
<rdfs:range rdf:resource="#TargetConcept"/>
</owl:ObjectProperty>
<owl:ObjectProperty rdf:ID="givenBy">
<rdfs:range rdf:resource="#SourceConcept"/>
<owl:ObjectProperty rdf:ID="whenUserTypeIs">
<rdfs:range rdf:resource="#UserType"/>
<owl:ObjectProperty rdf:ID="whenConteXtIs">
<rdfs:range rdf:resource="#ConteXt"/>
Figure 5.cont.
<owl:ObjectProperty rdf:ID="explanation">
<rdfs:range rdf:resource="#Explanation"/>
<owl:ObjectProperty rdf:ID="representedBy">
<rdfs:inverseOf rdf:resource="#forTargetConcept"/>
</rdf:RDF>
refactoring the schema, exposing functional dependencies implied by lookup

constraints (b) and (c). We leave it, however, as a topic for further research.
Case Study: Metaphor Enabling

an E-Learning System
The purpose of the study we are reporting about in this section is to demonstrate
the feasibility of noninvasive metaphor adaptation and to gather experience
regarding user acceptance of metaphor adaptation.
Metaphor Enabler
The architecture outlined in Figure 3 was implemented by using the PAW

(Proactive Web Filter) open-source filtering HTTP proxy based on the Brazil
Framework provided as an open-source project by Sun Microsystems (Brazil
Project, 2005; Proactive Web Filter Project, 2005). The PAW proxy offers
the capability to plug in custom filters (which have to comply with filter APIs
defined as part of the Web application development framework defined by the
Brazil Project). At present (in version 2.7), the PAW proxy comes with a
RegExp filter, allowing us to match Web page data included in HTTP responses
as a string against a set of regular expressions, and to replace matches by
Figure 6. Physical implementation of the architecture outlined in Figure 3
PAW proxy
with Target
RegExp filter WIS
(metaphor enabler) (base service)
Request
Request
Browser Media Converter
SD
_5VDC.
__ __
+
1A
RX
UP LINK TX
HTML
Adapted HTML
LINK PWR LINK
arbitrary strings. To fully implement the conceptual model given in Figure 1, one
would have to develop a custom filter, combining the capability to match and
replace strings with access to session and context data (such as current user
identity and user-type assignment, the URL of the page being retrieved, etc.).
For the purposes of initial evaluation, we opted to limit ourselves to relying on
the functionality of the already-available RegExp filter, so that the concept
mapping was not context sensitive: rules defined for string replacement were
defined for all Web pages of the target WIS (the “base service”) in a uniform
way. While it would be easy to introduce a form of user-type adaptivity by
running several instances of the proxy simultaneously, one for each user type,
and by instructing users belonging to different user types to use the appropriate
proxy settings, the number and the variety of users available for the initial
evaluation was not sufficient to validate the value of adaptivity to user types.
Hence, we defined the transformations necessary to add metaphor support for
a single proxy instance, supporting a single user type, and assuming a uniform
context for target domain concepts.
To make sure that the addition of a metaphor support layer does not diminish
the user experience under any circumstances (even for users who do not
consider the metaphor support to be useful), rather than replacing the occur-
rences of target domain concepts by source domain concepts, we transformed
the Web page HTML in such a way that each target domain concept would
became an anchor that, when activated by the user, would display a pop-up
showing the corresponding source domain concept (see Figure 7). The RegExp
filter was configured to add JavaScript allowing us to achieve this effect on all
Figure 7. A metaphor-enabled Web page, with the source concept (pipe

work) shown for the target concept graph, after the user clicked on the
source concept highlighted as an anchor by the metaphor support layer
Highlighted in blue
HTML pages served via the PAW proxy. Thus, from a user perspective, the
metaphor support layer we introduced was “transparent”: users could see
through it without taking note of it, if they wished to.
As seen in Figure 7, words corresponding to target domain concepts were
highlighted in blue to make them appear similar to HTML links in default HTML
styling, but they were not underlined (as HTML links are), so that they could
be distinguished from HTML links. By adopting a styling of target domain
concepts similar to HTML links, we hoped to suggest to the user that when he/
she clicks on the target domain concept, some relevant information (the source
domain concept) will appear. We believe that highlighting target domain
concepts should not negatively affect the usability of the interface for a user who
opts not to use the metaphor support layer (and it could even improve it).
Target WIS Choice and Metaphor Design
As a target WIS for the metaphor support layer evaluation, we chose the
Introduction to Graph Theory and Euler Circuits and Paths online tutorials from
Graph Theory Tutorials by Chris K. Caldwell (Caldwell, 1995) (in the
following, we often refer to them collectively as “the tutorial”). The reasons for
this choice were as follows:
• The tutorials are highly interactive and involve static pages, dynamic
pages, and HTML forms, thus representing a full-featured WIS.
• The tutorials have a linear structure, which ensured that all test users
followed the same path, and had comparable exposure to the WIS
information.
• The tutorials include a large number of quizzes, which allowed us to
consider the success of the users in learning the subject.
• The topic covered by the tutorials is relevant for the background of the test
users we had available.
Since the test WIS was devoted to teaching graph theory, the target domain
contained various graph theory terms. As the source domain, we chose
plumbing with pipes standing for graph edges and pipe junctions for vertices,
on the assumption that all of the prospective users have a degree of everyday
experience relevant to this domain. We have built a system of metaphors based
on this representation of edges and vertices, with movement along edges and
vertices interpreted as “swimming” along “pipes.” In addition to textual
representation, we added the relevant drawings. Care was taken to make sure
that each target concept representation would stand on its own, and would not
require any further clarification via cross-referencing and so forth. All of the
metaphors employed formed a metaphor system organized around the common
high-level metaphor (of a pipe work and movement through it). Some of the
target concept representations we used are listed in Figure 8.
Evaluation Setup and Results
The evaluation was conducted with 11 users of similar background (most of

them postgraduate students employed as graduate assistants not majoring in
mathematics). The users were instructed to reconfigure their browsers to use
the PAW proxy imposing the metaphor support layer, and to complete the
tutorial with as few quiz resubmissions as possible. The tutorial was set up (by
its original creator) in such a way that in order to proceed the user needed to
answer all questions in each of the intervening quizzes correctly; if a quiz was
answered incorrectly, the user was prompted to redo it until all answers were
correct (see Figure 9).
There was no time limit, and the users were not asked to maximize the speed
with which questions are answered, so the number of quiz resubmissions was
Figure 8. Some of the metaphors used in the evaluation
the only quality criterion (the fewer resubmission, the better, with zero
resubmissions corresponding to all questions answered correctly at the first
attempt). The evaluation instructions we used can be viewed online at Tretiakov
and Kaschek (2005).
All HTTP requests were logged at the proxy. We attributed Web page and
metaphor invocations to user sessions via IP addresses, which were stored by
the proxy as part of each log entry. As IP addresses in our environment are
allocated dynamically, it was not possible to determine the identity of the user,
which ensured user anonymity. In addition to completing the tutorial, users were
asked to anonymously provide free-form comments on the usefulness of the
metaphor layer.
A Python script that was created for that purpose was used to filter out log
entries for each session and to represent the session as an animated replay of
user interactions with the tutorial and with the metaphor layer.
For each user, the animated replay was visually analyzed to make sure that any
spurious or irrelevant interactions are not taken into account. While most users
navigated along the tutorial linearly, only once, from beginning to end, some
completed only half of the tutorial on the first attempt; later, they resumed from
the beginning, and this time, completed it, or almost completed it. For such
users, we disregarded the part of the second attempt session duplicating the
first attempt, so that all data correspond to the first encounter of the user with
the particular part of the system’s hyperspace. A detailed summary of the data
is given in Table 1.
As is clear from Table 1, we divided all metaphor invocations into two
categories: the ones in the context of exploration, and the ones in the context
of fulfilling short-term goals. This was achieved as follows: we divided all pages
of the tutorial into three categories—exploration page, in which new concepts
were introduced; quiz page, showing questions to be answered by the user; and
feedback page, providing feedback for quiz submissions (see Figure 9). When
a metaphor was invoked from an exploration page for the first time, or from a
feedback page after a successful quiz submission, we assumed that it was for
the purpose of learning a concept, in the context of exploration, rather than for
the purpose of recollecting a definition (a short-term goal). On the other hand,
when a metaphor was invoked from a quiz page, or from a feedback page after
Figure 9. The structure of the tutorial
Information Information
Foreward
Tutorial Graph Theory Definition of

Introduction Introduction Graph
Quiz Quiz
Quiz Quiz
Graph Theory
Introduction [Success] Definition of
Quiz Graph Quiz
[Failure] Submit Submit [Failure]
Graph Theory Definition of

Introduction Graph Quiz
Quiz Feedback Feedback
Feedback
[Success]
Feedback
Succession of information,
quiz and feedback pages,
till the end of the tutorial is
reached
Table 1. User success (measured by the number of quiz resubmissions) and

the number of metaphor invocations. For users who did not complete the
tutorial, the number of resubmissions is not given (marked n/a).
Metaphor invocations Metaphor invocations

Number of quiz
User no. in context of in context of fulfilling
resubmissions
exploration short-term goals
1 11 9 2
2 1 2 1
3 1 8 8
4 n/a 6 1
5 9 2 0
6 6 0 4
7 12 4 8
8 8 11 0
9 n/a 4 0
10 8 6 1
11 4 9 3
an unsuccessful submission, we assumed that it was for the purpose of fulfilling

an immediate information need prompted by the necessity to solve a problem.
Most of the users made use of the metaphor layer feature, with the mean number
of metaphor invocations per user of about 8, which considering that there were
34 pages in the tutorial, constitutes about 1 invocation per 4 pages visited. As
one can see from Table 1, the number of metaphor invocations differed
significantly from user to user, the standard deviation is estimated as 4.6 (more
than half of the mean value of 8).
An unexpected and highly interesting outcome was that the users made heavy
use of metaphors in the context of exploration, with almost 70% of invocations
falling in that category. There were 45 metaphor instances available on
information pages, 29 on quiz pages, and a smaller number of metaphor
instances on feedback pages, with the user exposure to metaphors on feedback
pages depending on his/her answers in the corresponding quiz. While the
specific numbers are difficult to interpret, as the impact obviously depends not
just on the number but also on the quality and appropriateness of the metaphors
available, the numbers clearly indicate that users, overall, accept the services
provided by the metaphor layer, as they invoke them without an imposition.
Figure 10. The difference between the number me of metaphor invocations

in context of exploration to the number of metaphor invocations in
context of solving a problem mp, normalized by the total number of
invocations me + mp, separately for each user (with the user number shown
at the top of each bar).
1.5
5 8 9
1
1 4 10
11
0.5
me − m p 2
me + m p 6 7 3
0
1 2 3 4 5 6 7 8 9 10 11
-0.5
-1
-1.5
The difference of the number of metaphor invocations in the context of

exploration and the number of metaphor invocations in the context of solving
a problem, normalized by the total number of invocations, is plotted in Figure
10.
On the other hand, with the relatively small number of users participating in the
evaluation, we were not able to observe a clear correlation between metaphor
layer usage and user success measured by the number of quiz resubmissions
(the smaller, the more successful). As can be seen from Figure 11, there were
heavy and poor metaphor users at all levels of that success measure.
Figure 10 appears to suggest that while many of the users favor using the
metaphor layer in exploration context, others rely on it in problem-solving
context, suggesting that two user types could be defined, with metaphors
provided differently to each type depending on the context (e.g., users using
metaphors in exploration mode could be provided with richer, more complex
source concepts). However, it is a matter for further research to consider how
to use the adaptivity to user type in metaphor layer to alter user behavior (e.g.,
Figure 11. The total number of metaphor invocations as a function of the

number of quiz resubmissions nr (the fewer the number of resubmissions,
the more successful is the user). Users 4 and 9, who did not finish the
tutorial, are not included.
18
16 3
14
12
11 7
10
8 1
me + m p 8
6
10
4 6
2
2 5
0
0 5 10 15
to encourage long-term learning), rather than to just cater for facilitating the
existing usage patterns.
Out of the 10 users who did provide free-form comments, nine were very
positive both about the metaphor layer and about its implementation in support
of the tutorials used in the evaluation. One user was positive about the metaphor
layer in general, but for the implementation used in the evaluation did not
recognize the choice of the source domain and the choice of source domain
concepts as appropriate or useful. An evaluation involving a larger number of
users would reveal if there are indeed user types for whom a different choice
of the source domain would be more appropriate.
Conclusion
We have introduced a conceptual data model allowing us to adaptively map

target domain concepts to source domain concepts, for inclusion in metaphors.
The model allows defining a database for a metaphor enabler forming part of
a noninvasive architecture that we propose for adaptive metaphor enabling of
Web-based systems.
Since our approach is noninvasive, it can be applied to both legacy systems and
to systems under development. The architecture suggests decorating a Web-
based system with metaphors by adding a separate architectural layer, so that
the base service does not need to be changed.
In addition, we explored how Web Ontology Language (OWL) could support
a potential implementation by providing an OWL schema matching the concep-
tual model defined in HERM.
We implemented a limited prototype system and conducted a user evaluation.
The results of the evaluation indicate a good user acceptance of the metaphor
layer, with users tending to access metaphors in context of learning/exploration,
rather than limiting themselves to invocations in support of fulfilling short-term
goals.
References
Bailer-Jones, D. M. (2002). Models, metaphors and analogies. In The

Blackwell guide to the philosophy of science (pp. 108–127). Blackwell.
Bineman-Zdanovicz, A., Kaschek, R., Thalheim, B., & Schewe, K.-D.
(2004). Context-aware Web information systems. In Proceedings of
Conceptual Modelling 2004, First Asia-Pacific Conference on Con-
ceptual Modelling, 26(6), 37–48.
Brazil Project. (2005). Brazil Project objective. Retrieved February 28,
2005, from http://research.sun.com/brazil
Caldwell, C. K. (1995). Graph theory tutorials. Retrieved February 28,
2005, from www.utm.edu/departments/math/graph
Carroll, J. M. (2003). HCI models, theories, and frameworks: Toward a
multidisciplinary science. Morgan Kaufmann.
Comstock, E. M., & Duane, W. M. (1996, April). Embed user values in system
architecture: The declaration of system usability. Proceedings of the
SIGCHI Conference on Human Factors in Computing Systems,
Vancouver, British Columbia, Canada.
Evans, V., & Green, M. (2005). Cognitive linguistics: An introduction.
Edinburgh UK: Edinburgh University Press.
Gamma, G., Helm, R., Johnson, R., & Vlissides, J. (1995). Design patterns.
Addison-Wesley.
Goguen, J. (1999). An introduction to algebraic semiotics, with application to

user interface design. In Computation for Metaphors, Analogy, and
Agents (pp. 242–291). Springer.
Hirschheim, R., Klein, H., & Lyytinen, K. (1995). Information systems
development and data modeling. Cambridge University Press.
Janda, L. (2000, February). Cognitive linguistics. Proceedings of the special
workshop/forum on the Future of Slavic Linguistics in America,
SLING, Bloomington, IN.
Kaschek, R., Schewe, K.-D., Thalheim, B., Tschiedel, B., & Kuss, T. (2004,
August). Learner typing for electronic learning systems. Proceedings of
the International Conference on Advanced Learning Technologies,
Joensuu, Finland.
Kaschek, R., Schewe, K.-D., Wallace, C., & Matthews, C. (2004). Story
boarding for Web-based information systems. In D. Taniar & J. W.
Rahayu (Eds.), Web information systems. Hershey, PA: Idea Group.
Kövecses, Z. (2002). Metaphor: A practical introduction. Oxford Univer-
sity Press.
Kuhn, T. S. (1993). Metaphor in science. In A. Ortony (Ed.), Metaphor and
thought. Cambridge University Press.
Murray Hopper, G. (1981). Keynote address at the ACM SIGPLAN History
of Programming Languages conference, June 1–3 1978. In R. L. Wexelblat
(Ed.), History of programming languages (pp. 7–20). Academic
Press.
Nehaniv, C. (1999). Computation for metaphors, analogy, and agents. In
Computation for metaphors, analogy, and agents (pp. 1–10). Springer.
Nielsen, J. (2000). Designing Web usability. New Riders Publishing.
O’Grady, W., Archibald, J., Aronoff, M., & Rees-Miller, J. (2001). Contem-
porary linguistics. New York: Bedford/St. Martin’s.
Pena Cervel, M. S. (2003). Topology and cognition. LINCOM Europa.
Proactive Web Filter Project. (2005). What is PAW? Retrieved February 28,
2005, from http://paw-project.sourceforge.net
Thalheim, B. (2000). Entity-relationship modeling. Springer.
Thalheim, B., & Düsterhöft, A. (2000). The use of metaphorical structures for
Internet sites. Data and Knowledge Engineering, 35, 161–180.
Tretiakov, A., & Kaschek (2005). Metaphors layer evaluation—Instruc-

tions. Retrieved February 28, 2005, from http://is alt.massey.ac.nz/
metaphors/evaluations_instructions.htm
Veale, T. (1999). Pragmatic forces in metaphor use: The mechanics of blend
recruitment in visual metaphors. In Computation for metaphors, anal-
ogy, and agents (pp. 37–51). Springer.
Wallace, C., & Matthews, C. (2002). Communication: Key to success on the
Web. Proceedings of eCoMo. Springer.
World Wide Web Consortium. (2004, February 10). OWL Web Ontology
Language Reference. W3C Recommendation.
World Wide Web Consortium. (2005). Semantic Web. Retrieved March 3,
2005, from www.w3.org/2001/sw
142 Gena & Ardissono
Chapter VII
A User-Centered
Approach to the
Retrieval of
Information in an
Adaptive Web Site
Cristina Gena, Università di Torino, Italy
Liliana Ardissono, Università di Torino, Italy
Abstract
This chapter describes the user-centered design approach we adopted in

the development and evaluation of an adaptive Web site. The development
of usable Web sites, offering easy and efficient services to heterogeneous
users, is a hot topic and a challenging issue for adaptive hypermedia and
human-computer interaction. User-centered design promises to facilitate
this task by guiding system designers in making decisions, which take the
user’s needs in serious account. Within a recent project funded by the
Italian Public Administration, we developed a prototype information
Retrieval of Information 143
system supporting the online search of data about water resources. As the
system was targeted to different types of users, including generic citizens
and specialized technicians, we adopted a user-centered approach to
identify their information needs and interaction requirements. Moreover,
we applied query analysis techniques to identify further information needs
and speed up the data retrieval activity. In this chapter, we describe the
requirements analysis, the system design, and its evaluation.
Introduction
The development of a Web-based information system targeted to different

types of users challenges the Web designer because heterogeneous require-
ments, information needs, and operation modes have to be considered. As
pointed out by Nielsen (1999) and Norman and Draper (1986), the user’s
mental model and expectations have to be seriously taken into account to
prevent her/him from being frustrated and rejecting the services offered by a
Web site. Indeed, this issue is particularly relevant to Web sites offering task-
oriented services, because most target users utilize them out of their leisure
time, if not at work. Being under pressure, these users demand ease of use as
well as efficient support to the execution of activities.
The positive aspect of a technical Web site is, however, the fact that the users
can be precisely identified and modeled; moreover, their information needs,
representing strong requirements, can be elicited by means of a suitable domain
analysis. Therefore, utilities, such as data search and retrieval, can be devel-
oped to comply with different goals and backgrounds. Of course, users’
involvement and testing have to be carried out also in this case because they
support the development of effective and usable services (see Dumas &
Redish, 1999; Keppel, 1991).
In our recent work, we faced these issues in the development of ACQUA, a
prototype Web-based information system for the Italian Public Administration
presenting information about water resources (a demo is available at http://
acqua.di.unito.it). During the system design phase, we put in practice traditional
usability principles and adaptive hypermedia best practices and we derived
general guidelines for the development of usable Web-based systems for
technical users (see Brusilovsky, 1996, 2001; Fink, Kobsa, & Nill, 1999;
Maybury & Brusilovsky, 2002). The system described in the rest of this
chapter is targeted to two main classes of users:
• Generic users, such as the citizen, who want to be informed about the
general health state of rivers, lakes, and underground waters.
• Technical users, such as the public administration employees, who re-
trieve specific pieces of information for analysis purposes.
In this chapter we describe the requirements analysis, the design, and the
evaluation of ACQUA, focusing on the user-centered approach adopted in
the prototype design and development phases. We involved domain experts
and end users since the beginning of our work in order to assess the usefulness
and suitability of the functionality offered by the system, as well as of its user
interface. For further information about the system, see Gena and Ardissono
(2004).
The rest of this chapter is organized as follows: Section “Background” provides
an overview of the relevant user-centered design research. Section “The
ACQUA Project” presents our work. Specifically, Section “Application
Requirements” describes the interaction and user interface requirements that
emerged during the design phase; Section “Adaptive Features” presents the
adaptive features we developed for our system; Section “Association Rules”
describes the techniques supporting the personalized information search;
Section “Evaluation of ACQUA” presents the results of an evaluation we
carried out to test the system functionality with real users; and Section
“Comparison with Other Solutions” compares our proposal with some related
work. Finally, section “Future Trends” discusses some open technical issues
and suggests how to address them, and Section “Conclusion” concludes the
chapter.
Background
Several researchers suggested to address usability issues by developing

adaptive systems. For instance, Benyon (1993) proposed adaptivity as a
solution, because a single interface cannot be designed to meet the usability
requirements of all the groups of users of a system. However, it is possible to
prove that adaptivity enhances the usability of a system only if it can be shown
that, without the adaptive capability, the system performs less effectively.
Benyon identifies five interdependent activities to be considered when design-
ing an adaptive system:
1. Functional analysis, aimed at defining the main functions of the system.

2. Data analysis, concerned with understanding and representing the mean-
ing and structure of data in the application domain.
3. Task knowledge analysis, focused on the cognitive characteristics
required by the system users, such as the user’s mental model, cognitive
load, and the required search strategy.
4. User analysis, aimed at determining the scope of the user population to
which the system is targeted. This analysis concerns the identification of
the user attributes that are relevant for the application, such as required
intellectual capability, cognitive processing ability, and similar. The target
population is analyzed and classified according to the aspects of the
application derived from the above-mentioned points.
5. Environment analysis is aimed at identifying the characteristics of the
environment in which the system is going to operate.
Notice that these phases are similar to the steps followed during the require-
ments analysis phase of a generic software system (Preece, Rogers, Sharp, &
Benyon, 1994). Benyon underlined the fact that adaptive systems should
benefit more than other systems from a requirements analysis before starting
any kind of evaluation, because the development of these systems has to take
a high number of features into account. The recognition that an adaptive
capability may be desirable leads to an improved system analysis and design.
As a demonstration, he reported an example of an adaptive system develop-
ment, wherein he prototyped and evaluated the system with a number of users.
Several user characteristics were examined to determine their effects on the
interaction. Then, further task knowledge and functional analysis were carried
out.
Also Oppermann (1994) proposed a user-centered perspective and suggested
a design-evaluation-redesign approach. He noticed that the adaptive fea-
tures can be considered as the main part of a system and thus have to be
evaluated during every development phase. The problem is circular:
• A problem solvable by means of the adaptivity has to be identified.

• The user characteristics related to the problem have to be selected.
• Ways of inferring user characteristics from interaction behavior have to be
found.
• Adaptation techniques offering the right adaptive behavior have to be

designed.
This process requires a bootstrapping method: first some initial adaptive

behavior is implemented, then tested with users, revised, and tested again. The
reason is that it is hard to decide which particular adaptations should be
associated to specific user actions. Furthermore, the adaptations must be
potentially useful to the user. The necessity of an iterative process is due to the
fact that the real behavior of users in a given situation is hard to foresee;
therefore, some evidence can be shown only by monitoring the users’ activity.
From the iterative evaluation point of view, the design phases and their
evaluation have to be repeated until good results are reached.
Oppermann’s iterative process is very similar to the user-centered system
design approach originally phrased by Gould and Lewis (1983) and extended
by Norman and Draper (1986).
Dix, Finlay, Abowd, and Beale (1998) pointed out that the iterative design is
also a way to overcome the inherent problems of incomplete requirements
specification, as only a subset of the requirements for an interactive system can
be determined from the start of the project. The iterative evaluation process
requires empirical knowledge about the users’ behavior from the first develop-
ment phases. In the case of an adaptive system, prior knowledge about the real
users, the context of use, and domain experts facilitates the selection of the
relevant data for the user model, such as personal features, goals, plans, domain
knowledge, and context. Deep knowledge about users also offers a broad view
of the application goals and prevents the system designer from serious
mistakes, especially when dealing with innovative applications.
Petrelli, De Angeli, and Convertino (1999) proposed the user-centered
approach to user modeling as a way to move from designer questions to
guidelines by making the best use of empirical data; they advocated incremental
system design as a way to satisfy large sets of users. They reported that at the
early stage of development of a mobile device presenting contextual informa-
tion to museum visitors, they decided to revise some of their initial assumptions
about the user model. Indeed, they made this decision after having analyzed the
results of a questionnaire distributed to 250 visitors. For instance, they
discarded the former user modeling techniques based on stereotypes (because
the sociodemographic and personal data taken in consideration did not
characterize the users’ behavior in a satisfactory way) in favor of a socially
oriented and context-aware perspective. For instance, they noticed that people
do not like to visit museums on their own and prefer looking at paintings to
interacting with a device.
As discussed by Höök (2000), intelligent user interfaces may violate many
usability principles developed for direct manipulation systems. The main
problem is that these systems may violate many good principles, such as
enabling the user to control the system, making the system predictable (given
a certain input, the system always generates the same response), and making
the system transparent so that the user understands at least partially how it
works. In addition, most adaptive interface developers are more concerned
with defining inference strategies than with interface design. For Höök,
intelligent user interfaces sometimes require a new way of addressing usability,
different from the principles outlined for direct-manipulation systems. Instead
of measuring factors such as task completion time, number of errors, or number
of revisited nodes, other aspects have to be considered. For instance, “if the
system should do information filtering, then we must check whether subjects
find the most relevant information with the adaptive system and not necessarily
whether they find it fast. This is not to say that the traditional measurements are
always wrong—this of course depends upon the task that user and (adaptive)
system should solve together” (Höök, 2002, p. 12).
Finally, Palmquist and Kim (2000) investigated the effects of (field independent
and field dependent) cognitive style and online database search experience on
WWW search performance. They concluded that cognitive style significantly
influences the performance of novice searchers. In contrast, experienced
searchers display a common behavior: they usually do not get lost in Web pages
including many links, but they are able to choose useful navigation strategies.
Therefore, Palmquist and Kim suggested that novice users should benefit from
Web pages that have a simple design and few links providing information
necessary to perform analytic search.
The ACQUA Project
Application Requirements
In 2003, the Water Resources Division (Direzione Risorse Idriche) of the

Piedmont Region and the University of Torino started a project for the
development of ACQUA, a Web-based information system presenting data
about water resources derived from the monitoring activities on the territory.
The goal was to make information available on a Web site that describes the
Division and supports a search for data in real time, in order to limit the
distribution of information on a one-to-one basis via e-mail messages and paper
publications. The technicians of the Division guided us in the system develop-
ment by specifying application requirements and by sharing with us a repository
of e-mail messages they exchanged with users asking for information through-
out the years. The repository provided us with evidence about the users’
interested in water resources data, the inspected information, and the regulari-
ties in the search for data. Most questions were posed by the following:
• Employees of other Public Administrations, such as technicians and

researchers, who are often interested in environmental impact studies,
construction feasibility studies, and historical data.
• Technicians, such as companies working at the construction of bridges
and houses.
• Attorneys, who are typically interested in the examination of data con-
cerning specific regions, for example, as a consequence of an environmen-
tal disaster.
• Farmers, who wish to monitor the biochemical state of their fields.
• Students attending secondary school, university, and also doctoral pro-
grams. These users collect information for the preparation of reports
concerning, for example, historical changes in biological and chemical
composition of waters, or the evolution of the capacity and hydrometric
levels of rivers, and similar.
Following a user-centered approach, we developed the system by involving

domain experts and end users since the first design phases. After a requirements
analysis phase, we developed a number of mock-ups, which we discussed and
redesigned after several focus group sessions with the experts and the users
involved in the project. We decided to adopt a cooperative design approach
(Greenbaum & Kyng, 1991) in order to utilize the experience of domain experts
and technicians in the design of an effective user interface. We based the
development of our first prototype on the collected feedback. As the ACQUA
system is devoted to the Public Administration, we had to satisfy usability and
predictability requirements that imposed the design of a simple user interface.
Specifically, our interlocutors suggested the following:
• The interface should be usable and intuitive in order to satisfy user needs
and expectations. This first requirement should be followed in every
interface design project; however, Public Administrations have the man-
datory goal of satisfying all the citizens, thus usability is also intended as
a service for the collectivity.
• The system behavior should be highly predictable (Dix et al., 1998) to
support first-time visitors in their search task, but also to avoid frustrating
professional users who would regularly use it at work. Notice that the
predictability requirement has some subtle aspects: for instance, not only
the user should foresee what is going to happen next, but also what should
not be expected from the service. This is very important to prevent the
user from starting the exploration of paths that will not provide her/him
with the information, or the functionality (s)he is looking for.
• The system should provide the user with data that can be analyzed without
preprocessing. Therefore, search results should be presented in machine-
processable formats, in addition to the pictorial ones suitable for a
general-purpose presentation in Web pages.
• For the sake of accessibility, the pages of the user interface should be
optimized for standard browsers, without the need of special equipments
or software environments.
In order to maximize the usefulness of the information that can be retrieved from
the Web site, we decided to make the system generate the search results in
formats, such as MS Excel® tables and textual (TXT) files, directly supporting
the data analysis and interpretation at the user side. We also tried to address
efficiency in the retrieval of information by reconsidering the design of the
general Web site to be presented. We wanted to offer the right information the
user is looking for; thus, we decided to show the main search functions in the
home page of the Web site, and to move textual information, such as the pages
describing the Public Administration divisions, in secondary pages, which can
be reached by following hypertextual links.
Moreover, having analyzed the data about the users interested in water
resources, we identified two main targets to which the system should adapt. For
shortness, we denote these categories as novices and experts.
Figure 1. Searching quantitative data (continuous hydrometric and

chemical-physical parameters) about Po River in the Torino-Murazzi
observation point
• Novice users, such as students and generic citizens, visit the Web site on
an occasional basis and are not familiar with the content presented by the
information system.
• Expert users, such as technicians, farmers, and the personnel of other
Public Administrations, frequently visit the site and are familiar with the
domain-specific information provided by the system.
In order to take the interaction requirements of these users into account, we

defined two search functions:
(i) The simple search is a geographical search modality and guides the user
step by step in the retrieval of information;
(ii) The advanced search offers forms where the expert user may compose
the queries in single step. Figure 1 shows the user interface of the ACQUA
prototype supporting the advanced search; the menus enable the user to
specify the river (“Scegli il corso d’acqua”), observation point (“Scegli il
Figure 2. Portion of the page describing the Torino observation point on

Po River
punto di monitoraggio”), start date (“Data Inizio”), and end date (“Data
Fine”). Moreover, the user interface enables the user to select the
hydrometric and chemical-physical parameters to be inspected.
Thus, novice users may search for information in a friendly modality and the
eligible choices are restricted and presented along the path, while expert users
benefit from a faster search function.
As a matter of fact, the information about water resources exploited by the
system is unavoidably incomplete. For instance, some data are collected by
automatic stations, which have been set up at different times over the years and
sometimes are out of order. Moreover, unfortunately, data collected in manual
observation points have been stored in unstructured formats and the historical
series has been reconstructed only for the very recent past.
For the sake of predictability, the simple and advanced search functions
prevent the user from composing any queries that are incorrect, or are aimed
at searching for unavailable data. The idea is that, in both cases, the system
should only present the choices leading to available results. For instance, as
shown in Figure 1, the labels of the parameters, which are not available for the
Po River, are shaded and cannot be selected by the user to define a query.
Adaptive Features
The information about water resources concerns rivers, lakes, and under-
ground waters and includes the following:
• Descriptive data about resources and observation points: for example,

maps of the points, charts representing environmental changes, pictures,
documents, publications, and descriptions of the monitoring stations. For
instance, Figure 2 (“Caratteristiche della stazione di monitoraggio
TORINO” [“Features of the Torino monitoring station”]) shows the
coordinates and other information about the observation point on Po
River located in Torino, Parco Michelotti.
• Measurement parameters concerning physical dimensions and other
features, which characterize the environmental state of the resources.
These parameters are grouped in two main classes:
• Qualitative parameters, which are periodically measured: techni-

cians visit the observation points, collect data, and take samples for
laboratory tests.
• Quantitative parameters, which are monitored by automatic sta-
tions.
These stations carry out the measurements on a daily basis.

The ACQUA Web site is organized in four main sections, respectively devoted
to the presentation of qualitative and quantitative information about rivers,
information about lakes, and information about underground waters. The
system enables the user to retrieve data about water resources by performing
a simple or advanced search in all the sections of the site. Therefore, a large
amount of heterogeneous data is accessible, ranging from biological and
chemical data to capacity measurement and hydrometric levels (for details, see
Gena & Ardissono, 2004).
We noticed that, by performing queries aimed at selecting a large number of

data items, belonging to different categories, the results returned by the system
were complex and hard to present in an intuitive results table. However, as
shown by the repository of user requests we analyzed, users often need to
combine heterogeneous data to accomplish their goals. For example, in
construction feasibility studies, users are interested in qualitative and quantita-
tive parameters of rivers and underground waters, considering the historical
series. In order to keep the user interface simple and to guarantee that the
presented results are not confusing, we decided to limit the user’s freedom in
composing the queries: to retrieve very heterogeneous types of information, the
user must define more than one search query. For example, as shown in Figure
1, the ACQUA query interface enables the user to choose from different rivers,
observation points, years, and data types. Other categories, such as qualitative
and quantitative data about rivers, lakes, and underground waters are treated
as separate sections of the Web site and have their own query functions.
Unfortunately, although this approach enforces the clarity of the results, it
makes the search for multiple types of information a lengthy task. Therefore, a
compromise between clarity and efficiency must be found. In order to address
this issue, we extended the system with an intelligent search component,
which complements the user’s explicit queries with follow-up queries (Moore
& Mittal, 1996) frequently occurring together in navigation paths. When
possible, the system anticipates the user’s queries and makes the extended
search results available as personalized suggestions that can be downloaded on
demand. If the user is interested in the recommended information, (s)he can
retrieve it by clicking on the adaptive suggestion links, without performing any
further queries. At the same time, the system retrieves the extended results only
after the user clicks on a suggestion link in order to avoid precaching possibly
useless data.
For instance, Figure 3 shows the recommendations generated by the system in
the lower portion of the page (“Ti consigliamo anche i valori dei parametri
chimici e microbiologici” [“We also suggest results about chemical and micro-
biological parameters”]).
During different interaction sections, the same user may be interested in rather
different types of information; therefore, we decided to base the system’s
recommendations on the analysis of her/his navigation behavior, leaving the
management of a long-term user model apart. One immediate advantage is the
fact that the user can interact with the system in an anonymous way, without
signing up for the service. The follow-up queries are generated as follows: the
Figure 3. Annotated link for the suggested information and descriptions

of the monitoring stations
search queries performed by the user while (s)he browses the Web site are
monitored and association rules which suggest other, strictly related queries are
applied.
Each association rule has a condition part specifying constraints on the previous
navigation behavior, and an action part defining a relevant follow-up query to
be performed in order to retrieve complementary information. The rules we
defined in our current prototype are mutually exclusive and they are selected
and fired by applying a very tiny and efficient inference engine. This engine
would not be suitable to manage a large set of conflicting rules: a general-
purpose rule-based engine should be employed to that purpose. However, we
prefer to maintain a simple set of adaptation rules, and to avoid embedding
complex rule-based engines in order to keep the adaptive features as light-
weight as possible. In fact, the management of the interaction is subject to a
relevant overload due to the generation of results in multiple formats, which is
a main requirement for the information system. In this situation, minimalist but
efficient adaptation to the user is strongly preferred to flexible but complex one.
Association Rules
In order to define the association rules to be applied for anticipating the user’s
information needs, we analyzed a repository of requests, which real users
posed to the Water Resources Division over the years; the requests consisted
of e-mail messages and fax documents. As we noticed that different kinds of
information frequently occurred together in these requests, we decided to
analyze the frequency of co-occurrence in order to identify the regularities.
Specifically, we analyzed 97 requests and we selected a set of features
describing the requests in a systematic way. These features concerned rather
different aspects of the requests; thus, for clarity purposes, we grouped them
in subcategories. In the following, we report the subcategories we defined and
for each one we list some sample features:
• Kind of request: for example, environmental impact study, construction

feasibility studies, and lawyers’ studies.
• Request features: for example, information about one or more rivers,
about lakes or underground waters, about one or more observation points
on a river or lake.
• Kind of data: for example, qualitative or quantitative parameters, biologi-
cal and chemical data, hydrometric level, average daily capacity.
• Data features: for example, raw data, or elaborated data such as
medium, maximum, and minimum values during a time interval.
• User features: for example, research center, Public Administration,
technicians, farmers, and attorneys.
We computed the frequency with which the features co-occurred in the

requests: if the frequency exceeded a given threshold, the set of involved
features became a possible candidate for an association rule. Then we
compared the extracted associations with their original requests in order to
validate our findings with factual knowledge, and finally we asked the techni-
cians of the Water Resources Division if our conclusions were correct. After
this last check, we selected the correct associations and we encoded the rules
in the system.
For instance, a rule suggests to retrieve qualitative parameters about a water
resource if the user has asked for quantitative historical data for more than one
observation point on that resource, supposing that (s)he is looking for informa-
tion for a construction feasibility study. Another rule suggests retrieving the
environmental state indexes of a resource if the user has requested biological
and chemical data, under the hypothesis that (s)he is involved in an environmen-
tal impact study.
Evaluation of ACQUA
We first evaluated the ACQUA prototype in a usability test by involving

external users who were not cooperating at the project (see Dumas & Redish,
1999, for methodological details). The evaluation highlighted some usability
problems concerning the presentation of basic information, such as the choice
between simple and advanced search and the background color of the menus.
After having solved those problems, we tested the final prototype with real end
users representative of the users the Web site is devoted to. In particular, we
involved technicians working at the Water Resources Division in different fields
(rivers, lakes, underground rivers, etc.) and not collaborating to the design of
the project. We carried out both an experimental evaluation and a qualitative
session to assess the suitability of the adaptive features offered the system.
Subjects. We evaluated 10 potential users of the ACQUA system, four females

and six males, aged 30–50. All the users worked in the water resource area and
none of them was involved in the project.
Procedure. The subjects were split up in two groups (five subjects each) and
randomly assigned to one of the two groups. The experimental group had to
solve some tasks using the adaptive Web site, which applies the association
rules described in Section “Adaptive Features” to compute the results of
follow-up queries related to the users’ explicit queries. Instead, the control
group had to solve the tasks without adaptation.
Experimental tasks. Every subject had to solve seven tasks, each one
representing a real task the user can perform in the Web site. As suggested by
our correlation study, the tasks were strictly correlated and could be grouped
in three search activities the user often performs together. The first activity
conveyed the whole information useful to an environmental impact study. The
second one supported construction feasibility studies. The third activity sup-
ported lawyers’ studies and activities.
• In the control group, the users had to submit a new query for every task,
in order to obtain the requested results. The new queries were submitted
by filling in the query specification forms (see, e.g., Figure 1).
• In the experimental group, the users could obtain the extra information
related to the next task to be performed by clicking on an adaptive
suggestion link that supports the immediate retrieval of the suggested
information (see, e.g., Figure 3).
Experimental design. Single-factor (the adaptivity) between-subjects design.
Measures. The subjects’ navigation behavior was recorded by using Camtasia

Studio®. We measured the task completion time and then the subjects’
satisfaction, by means of a post-task walk-through.
Hypothesis. We hypothesized that the users working in the experimental group

could obtain better performance results than those of the control group.
Results. The ANOVA (analysis of variance) showed that the subjects of the
experimental group achieved the best performance results. In addition, we
calculated the effect size (treatment magnitude) and the power (sensitivity) as
suggested in Chin (2001). The effect size (w2) measures the strength, or the
magnitude, of the treatment effects in an experiment. In behavioral sciences,
small, medium, and large effects of w2 are 0.01, 0.06, and >0.15, respectively.
The power of an experiment (n’) is the ability to recognize treatment effects and
the power can be used for estimating the sample size. In social science, the
accepted value of the power is equal to 0.80, which means that the 80% of
repeated experiments will give the same results. In the following, we show a
summary of the results:
Task 2.
ANOVA: F(1.8) = 12.45 p<0.01; w2=0.53; n’=3.49
Task 3.
ANOVA: F(1.8) = 12.12 p<0.01; w2=0.53; n’=3.60
Task 5.
ANOVA: F(1.8) = 14.16 p<0.01; w2=0.57; n’=3.04
Task 7.
ANOVA: F(1.8) = 9.23 p<0.05; w2=0.45; n’=4.86
It should be noticed that all the results are significant and have a large estimate
of the magnitude of the treatment effect. In addition, by exploiting a power of
0.80 and the corresponding w2 for each task we could determine the requested
sampled size, which fits our sample size (n=5) (for details about statistics, see
Keppel, 1991).
Post-task walk-through. During any post-task walk-through, test subjects

are asked to think about the event and comment on their actions. Thus, after
each test we talked to the subjects to collect their impression and to discuss
their performance and the problems encountered during the test. In this session,
we also aimed at retrieving useful feedback for a qualitative evaluation of the
site. In fact, although our experimental evaluation reported significant results
supporting our hypothesis, the actual user behavior could be different. As
recently pointed out by Nielsen (2004), statistical analyses are often false,
misleading, and narrow; in contrast, insights and qualitative studies are not
affected by these problems because they strictly rely to the users’ observed
behavior and reactions.
In most cases, the interviewed users were satisfied with the site. Most of them
encountered some problems in the execution of the starting query of task 2, thus
we modified the interface form.
• All the users of the experimental group followed the adaptive suggestion
link provided by the system but they did not realize that it represented a
personalization feature. When we explained the adaptations, they noticed
the particularity of the suggestion (“We also recommend you ...”).
Anyway, they were attracted from the suggestions and they appreciated
the possibility of skipping the execution of a new query. The adaptive
suggestions were considered visible and not intrusive.
• The users of the control group reported similar considerations when we
described the adaptive features offered by the Web site. Even if they did
not receive any suggestions during the execution of tasks, they explored
the result pages in order to find a shortcut to proceed in the task execution.
After having followed some links, they went back to the previous query
page or to the home page by clicking on the “Back” button of the browser.
Both groups displayed a common behavior pattern: the users explored a results
page before starting a new search. Nevertheless, their behavior could be
influenced by the test condition, because tested users tend to pay a lot of
attention to their own actions and to the page design.
We conclude by admitting that although the test subjects were satisfied with the
adaptation features, only the real system usage can demonstrate our hypoth-
esis. However, both quantitative and qualitative test results are encouraging
and we think that the adaptations are correctly placed. After this test, we
presented the adaptive version of the Web site to the technicians of the Water
Resources Division collaborating on the project. They confirmed the correct-
ness of association rules we defined and they decided to replace the non-
adaptive version of the prototype system with the adaptive one.
Comparison with Other Solutions
The ACQUA system has a plain user interface, designed to meet simplicity,
usability, and predictability requirements, but it offers advanced interactive
features enabling the user to create a personal view of the information space.
Two search features, targeted to novice and expert users, are available, and the
search results are presented in both pictorial and machine-readable formats in
order to support direct data manipulation at the user side. Moreover, the
system analyzes the user’s queries to identify her/his information needs, and it
employs association rules to propose follow-up queries complementing the
search results with strictly related information. The follow-up queries are
applied on demand; thus, the user can ignore them if (s)he is not interested in
the additional data, and the system does not need to retrieve any uninteresting
information.
The advanced search features we presented differ from the related work in
various aspects. On the one hand, the inferences performed by our system are
simpler than the probabilistic ones applied in other automated assistants, such
as Lumière (Horvitz, Breese, Heckerman, Hovel, & Rommelse, 1998) and
ACE (Bunt & Conati, 2003), which exploit Bayesian networks to capture the
dependencies among the user actions. The point is that the user interacting with
the ACQUA system does not carry out a complex task requiring a problem-
solving activity. Therefore, lightweight rules associating contextually related
search queries are sufficient to predict the implicit information needs and to
complement the search for information accordingly. Our approach also differs
from the follow-up question answering techniques proposed by Moore and
Mittal (1996): in order to efficiently manage the query selection process, our
follow-up queries are precompiled in a set of association rules, instead of being
generated by a planner.
On the other hand, we apply query analysis techniques to identify regularities
in search patterns. This differs from the typical inferences carried out in
recommender systems, which reason about the features of the selected items
to identify the user’s priorities (see, e.g., Billsus & Pazzani, 1999), or about the
regularities in the selection of individual items (see, e.g., the work by Cotter &
Smyth, 2000; GroupLens, 2002).
Liu, Yu, and Meng (2002) propose other query analysis strategies for person-
alized Web search. However, instead of personalizing the proposed results,
their system supplies a small set of categories as a context for each query. The
system combines the user’s search history with a general user profile automati-
cally extracted from a category hierarchy to offer a personalized context for
disambiguating the proposed query results. In ACQUA, we do not manage
long-term user preferences because we noticed that, in different interaction
sections, the same users are interested in rather different types of information.
We thus decided to base the recommendations only on the analysis of the user’s
search behavior.
Future Trends
It is worth mentioning that the manual definition of the first set of association
rules supporting the user’s search task was a lengthy work and might not be
easily replicated to revise the rules along time. However, if the Water Re-
sources Division employs the ACQUA system as its official Web site, the log
files generated by the system will provide structured evidence about user
behavior (in addition to e-mails and faxes). Thus, data-mining techniques could
be exploited to automatically recognize usage patterns and revise the associa-
tion rules accordingly.
Indeed, we believe that these techniques can support the analysis of user
behavior in an effective way, but they still have to be coupled with human
analysis, in order to validate and interpret results: in several cases, these
techniques have generated some very interesting results, but also other irrel-
evant or hardly understandable findings, which have been discarded.
At any rate, Web usage mining techniques, derived from machine learning
methods such as knowledge discovery in data (KDD or data mining) can
contribute to automate the adaptation of Web-based systems to the users.
According to the scheme proposed by Pierrakos, Paliouras, Papatheodorou,
and Spyropoulos (2003), ACQUA can be classified as a Web personalization
system offering task performance support: this functionality involves the
execution of a particular action on behalf of the user. In our case, the system
generates queries and makes the results available as links to some files storing
them. This functionality is considered as the most advanced personalization
function and it is seldom offered by Web-based personalized services.
The most suitable data-mining technique, given the adaptive goals of the
ACQUA system, is the sequential pattern discovery, which is aimed at
identifying navigational patterns (event sequences) in the analyzed data (in our
case, Web usage data). This methodology supports the discovery of event
sequences that can be summarized as follows: “If event A, B, and C occur in
that order, then events D, E, and F always follow.” Two types of methods are
generally applied to discover sequential patterns: deterministic techniques,
which record the navigational behavior of the users and extract knowledge from
the analyzed data, and stochastic methods, which use the sequence of
already-visited Web pages to predict the behavior occurring in the next visits.
Once sequential patterns have been discovered, the extracted knowledge can
be automatically integrated in the personalization process, and the system
behavior adapted accordingly.
Conclusion
We presented our experience in the design and development of ACQUA, an

interactive prototype Web site for the Public Administration. The system
presents information about water resources and supports the user in the search
for generic information, as well as technical information about the rivers, lakes,
and underground waters.
The usability and functional requirements that emerged during the design of the
ACQUA system were very interesting and challenging, as they imposed the
development of functions supporting the efficient retrieval of data by means of
a simple user interface. We found out that the introduction of basic adaptivity
features, aimed at understanding the user’s information needs in detail, was
very helpful to meet these requirements.
We were asked to develop a system having a simple user interface, designed
to meet usability and predictability requirements. This fact limited our freedom
to add advanced interaction features, desirable in a Web site visited by
heterogeneous users; however, it challenged us to find a compromise between
functionality and simplicity. In order to address this issue, we developed two
interactive features enabling the user to create a personal view on the informa-
tion space:
• The system offers a simple and an advanced search functions targeted to

novice and expert users, respectively.
• Moreover, the system carries out a query analysis aimed at identifying the
user’s information needs, and applies association rules to extend the
user’s queries and complete the search results with data that is usually
retrieved together by end users.
Qualitative and quantitative evaluation results showed that the adaptive user
interface was more successful than the nonadaptive one. The reason was
probably the concrete help offered by the adaptive suggestions, which speed
up the execution of time-consuming search tasks. Moreover, the adaptive
features were not perceived as intrusive and the user was allowed to skip
useless suggestions. Furthermore, the system did not impose a previous
annoying and discouraging registration phase.
As discussed in Section “Future trends,” the adaptive features offered by the
ACQUA system could be improved by the integration of Web-usage mining
techniques aimed at discovering real usage patterns. In that way, the associa-
tion rules employed to identify the user’s implicit information needs could be
automatically updated along time. However, we believe that the rules we
manually defined provide a knowledge base that cannot be replaced with
automatically extracted rules. In principle, both kinds of rules could be
integrated in order to enhance the effectiveness of the system adaptations.
Acknowledgments
This work was funded by Regione Piemonte, Direzione Risorse Idriche. We

thank Giovanni Negro, Giuseppe Amadore, Silvia Grisello, Alessia Giannetta,
Maria Governa, Ezio Quinto, Matteo Demeo, and Vincenzo Pellegrino, who
assisted us during the system development and provided the domain-specific
knowledge.
References
Benyon, D. (1993). Adaptive systems: A solution to usability problems.

International Journal of User Modeling and User-Adapted Interac-
tion, 3, 65–87.
Billsus, D., & Pazzani, M. (1999). A personal news agent that talks, learns and
explains. In Proceedings of 3rd International Conference on Autono-
mous Agents (pp. 268–275).
Brusilovsky, P. (1996). Methods and techniques of adaptive hypermedia.
tion, 6(2–3), 87–129.
Brusilovsky, P. (2001). Adaptive hypermedia. International Journal of User
Modeling and User-Adapted Interaction, 11(1–2), 87–110.
Bunt, A., & Conati, C. (2003). Probabilistic student modelling to improve
exploratory behaviour. International Journal of User Modeling and
User-Adapted Interaction, 13(3), 269–309.
Chin, D. N., (2001). Empirical evaluation of user models and user-adapted
systems. International Journal of User Modeling and User-Adapted
Interaction, 11(1–2), 181–194.
Cotter, P., & Smyth, B. (2000). WAPing the Web: Content personalization for
WAP-enabled devices. Proceedings of International Conference on
Adaptive Hypermedia and Adaptive Web-Based Systems (pp. 98–108).
Dix, A., Finlay, J., Abowd, G., & Beale, R. (1998). Human computer
interaction (2nd ed.). Prentice Hall.
Dumas, J. S., & Redish, J. C. (1999). A practical guide to usability testing.

Norwood, NJ: Ablex.
Fink, J., Kobsa, A., & Nill, A. (1999). Adaptable and adaptive information for
all users, including disabled and elderly people. New Review of Hypermedia
and Multimedia, 4, 163–188.
Gena, C., & Ardissono, L. (2004). Intelligent support to the retrieval of
information about hydric resources. Proceedings of 3rd International
Conference on Adaptive Hypermedia and Adaptive Web-Based Sys-
tems (pp. 126–135).
Gould, J. D., & Lewis, C. (1983). Designing for usability: Key principles and
what designers think. Proceedings of CHI ’83 (pp. 50–53).
Greenbaum, J., & Kyng, M. (1991). Design at work: Cooperative design of
computer systems. Hillsdale, NJ: Lawrence Erlbaum.
GroupLens. (2005). GroupLens Research. Retrieved November 2, 2005,
from www.GroupLens.org
Höök, K. (2000). Steps to take before IUIs become real. Journal of
Interacting With Computers, 12(4), 409–426.
Horvitz, E., Breese J., Heckerman D., Hovel D., & Rommelse, K. (1998). The
Lumière project: Bayesian user modeling for inferring the goals and needs
of software users. Proceedings of 14th Conference on Uncertainty in
Artificial Intelligence, San Francisco.
Keppel, G. (1991). Design and analysis: A researcher’s handbook.
Englewood Cliffs, NJ: Prentice-Hall.
Liu, F., Yu, C., & Meng, W. (2002). Personalized Web search by mapping
user query to categories. Proceedings of the 2002 ACM Conference on
Information and Knowledge Management, McLean, VA.
Maybury, M., & Brusilovsky, P. (Eds.). (2002). The adaptive Web. Commu-
nications of the ACM, 45.
Moore, J. D., & Mittal, V. O. (1996). Dynamically generated follow-up
questions. IEEE Computer, 29(7), 75–86.
Nielsen, J. (1999). Web usability. Indianapolis, IN: New Riders Publishing.
Nielsen, J. (2004). Risks of quantitative studies. Retrieved March 9, 2004,
from www.useit.com/alertbox/20040301.html
Norman, D. A., & Draper, S. W. (1986). User centered system design: New
perspective on HCI. Hillsdale, NJ: Lawrence Erlbaum.
Oppermann, R. (1994). Adaptively supported adaptivity. International Jour-

nal of Human-Computer Studies, 40, 455–472.
Palmquist, R. A., & Kim, K. S. (2000). Cognitive style and on-line database
search experience as predictors of Web search performance. Journal of
the American Society for Information Science, 51(6), 558–566.
Petrelli, D., De Angeli, A., & Convertino, G. (1999). A user centered
approach to user modeling. Proceedings of the 7th International Con-
ference on User Modeling (pp. 255–264).
Pierrakos, D., Paliouras, G., Papatheodorou, C., & Spyropoulos, C. D.
(2003). Web usage mining as a tool for personalization: A survey.
tion, 13(4), 311–372.
Preece, J., Rogers, Y., Sharp, H., & Benyon, D. (1994). Human-computer
interaction. Addison Wesley.
Section III
Mental Reasoning
and Interaction
Swarm-Based Wayfinding Support 167
Chapter VIII
Swarm-Based
Wayfinding Support
in Open and
Distance Learning
Colin Tattersall, Open University of The Netherlands, The Netherlands
Jocelyn Manderveld, Open University of The Netherlands, The Netherlands
Bert van den Berg, Open University of The Netherlands, The Netherlands
René van Es, Open University of The Netherlands, The Netherlands
José Janssen, Open University of The Netherlands, The Netherlands
Rob Koper, Open University of The Netherlands, The Netherlands
Abstract
Open and distance learning (ODL) gives learners freedom of time, place,
and pace of study, putting learner self-direction centre stage. However,
increased responsibility should not come at the price of overburdening or
abandonment of learners as they progress along their learning journey.
This paper introduces an approach to wayfinding support for distance
learners based on self-organisation theory. It describes an architecture
168 Tattersall, Manderveld, van den Berg, van Es, Janssen, & Koper
that supports the recording, processing, and presentation of collective

learner behaviour designed to create a feedback loop informing learners
of successful paths towards the attainment of learning goals. The approach
is presented as an alternative to methods of achieving adaptation in
hypermedia-based learning environments which involve learner modelling.
Introduction
Open and distance learning (ODL) gives learners freedom of time, place, and
pace of study, putting learner self-direction centre stage. Brockett and Hiemstra
(1991, p. 24) define learner self-direction as the learner’s assumption of
“primary responsibility for and control over decisions about planning, imple-
menting and evaluating the learning experience,” and Hiemstra (1994) notes
learners’ preference to take on responsibility for their own learning. However,
taking on new responsibilities is not without its challenges. Brookfield (1985,
p. 7) notes that although self-directed learning “has connotations of autonomy,
independence and isolation,” investigations have highlighted that “adults would
like more, rather than less, assistance in their learning pursuits.” Similarly,
Candy (1991) writes that self-directed learners are often challenged to assume
certain responsibilities, and that when deciding how to approach learning tasks,
the self-directed learner is “confronted with the problem of how to find a way
into and through a body of knowledge that is unknown at the outset. Without
the benefit of any explicit guidance, a self-directed learner is obliged to map out
a course of inquiry that seems appropriate, but that may involve a certain
amount of difficulty and disappointment that could have been averted” (p. 283).
Candy’s description calls to mind the image of the distance learner as navigator,
charting a course through educational waters, following Darken and Silbert’s
(1993) definition of navigation as the “process of determining a path to be
travelled by any object through any environment” (p. 157). In subsequent
work, Darken and Peterson (2002) use the term “wayfinding” to refer more
specifically to the navigator’s decision-making process. We use the term
“educational wayfinding” to describe the cognitive, decision-making process
carried out by self-directed learners as they assume responsibility for choosing
and sequencing their learning events. The wayfinding decisions with which
learners are faced arise from the freedom offered to them by learning providers
on their way to the attainment of particular goals. In some highly constrained
situations, both the choice of learning events and their ordering may be fixed by
a learning provider. More likely, learners may be permitted to select and order
modules, perhaps to accumulate credit points towards a certificate. In this
context, we note Yorke’s (1999) warning that “as the unitization of curricula
spreads through higher education, so there is a need for greater guidance for
students to navigate their way through the schemes” (p. n/a). This provides the
background to this chapter: difficulties in the educational wayfinding process
can lead to learners not reaching their goals, or taking unduly long to do so. The
rationale for our work is that self-directed learners can benefit from support in
the educational wayfinding process, and we describe a new approach to
supporting the educational wayfinding process which has the potential to
address the drawbacks of existing approaches found in the literature. We
examine a number of alternatives to the provision of such support, and
introduce our approach to issue, which builds on self-organisation theory.
Approaches to Wayfinding
Support in ODL
There are a number of approaches to wayfinding support used today in ODL

in addition to those identified in ODL research but not yet widely implemented.
The first approach involves fixing routes through materials in advance of their
delivery, creating curricula or content plans to be followed by learners thereby
reducing navigational choices. However, this preplanning limits the possibilities
for learner self-direction, and Evans (1994) notes that the didactic models used
in open and distance education are often “founded on highly didactic models
which provide [the students] with little control over their own learning … and
the students are left with little option but to adhere to the curriculum” (p. 68).
This observation suggests the need for a flexible, adaptive approach to
wayfinding support.
Such flexibility can be realised through so-called “learner support services”
(Simpson, 2000). Although capable of providing highly individualised advice,
learner support services do not come without a price. Costs are likely to be
variable with student numbers and be exacerbated by the less predictable,
demand-driven nature of ODL.
A significant amount of research has explored the creation of educational
hypermedia systems (De Bra, 2002) as part of the Adaptive Hypermedia
research area (Brusilovsky, 2001; Cristea & De Bra, 2002). Thus activity
continues the research line established in the 1980s in the area of intelligent
tutoring systems, and seeks to “build a model of the goals, preferences and
knowledge of the individual user and use this through the interaction for
adaptation of the hypertext to the needs of the user” (De Bra, Brusilovsky, &
Houben, 1999, p. 58).
User models are representations of a world outside the computational environ-
ment and may contain wrong, outdated, or inadequate information (Fischer,
2001). A case in point is cited by Kilfoil et al. (2003)—the digital video
recorder automatically recording programs it assumes its owner will like, yet
based on an inappropriate assumption regarding the owner’s lifestyle. As De
Bra notes (2000, p. 76), “bad guidance is worse than no guidance.” Self
(1987), writing over 15 years ago, noted the absence of a theory of learning
which might be used to maintain learner models. In a later article, Self (1988,
p. 20) describes the scope of the student modelling problem—“from compu-
tational questions, to representational issues, through plan recognition, mental
models, episodic memory to individual differences—to encompass, it would
seem, almost all of cognitive science.” Concerns on the practical application of
user modelling continue to be raised (Atif, Benlamri, & Berri, 2003; Kay, 2001;
Strachan, Anderson, Sneesby, & Evans, 1997).
This raises a research question for cognitively informed systems: Is there an
alternative approach to wayfinding guidance in ODL which might provide a
cost-effective solution yet which does not rely upon learner modelling?
Self-Organisation and Wayfinding
The previous section reviewed three sources of wayfinding support—course

designers, attempting to predict efficient paths; learner support services,
providing flexible advice but at a price; and adaptive hypermedia systems,
still challenged to prove their practical application. A fourth source can be
found in the social context of learning, a point noted by Brookfield (1985) when
he states that the “successful self-directed learners … place their learning within
a social setting in which the advice, information, and skill modelling provided
by other learners are crucial conditions for successful learning” (p. 9). This
observation finds echoes in the information navigation literature, where the term
social navigation (Höök & Benyon, 2003) has been coined to describe
research reflecting the fact that “navigation is a social and frequently a

collaborative process” (Dieberger, 2003, p. 294). This point is also made by
Forsberg et al. (1998) who state that “most information navigation in the real
world is performed through talking to other people” (p. 13). However, we need
to question whether learners would be prepared to support their peers in
advice-giving dialogues given the pressures on their time.
In fact, social navigation does not always involve direct interaction. The field
has been divided into two areas of research. The first, direct social navigation,
sees actors as “co-present and in direct contact with one another” (Dieberger,
Höök, Svensson, & Lönnqvist, 2001, p. 107). In contrast, indirect social
navigation exploits traces of interactions left by others (Shipman et al., 1996;
Wexelblat, 1999). Applications of indirect social navigation can be found in the
educational literature (Shipman, Furuta, Brenner, Chung, & Hsieh, 2000;
Zeiliger, Reggers, Baldewyns, & Jans, 1997), although the focus has tended to
fall on teachers or students predefining trails through information space for
others to follow later. This approach brings with it a certain cost to the ODL
learner who may not be disposed to investing time and effort to create a trail
for unknown learners coming along later. The ideal approach would avoid
anyone precreating wayfinding guides and have them somehow “emerge” so
that learning processes, as it were, spontaneously acquire (sequential) struc-
tures or organisations. This is the language of self-organisation—”the sponta-
neous formation of well-organised structures, patterns or behaviours, from
random initial conditions” (Soraya Kouadri et al., 2003, p. 62). Indeed the
“acquiring of spatial, temporal or functional structure” is seen as the essence of
self-organisation by Hadeli et al. (2003, p. 54), and is echoed by Heylighen and
Gershenson (2003)—“a self-organizing system not only regulates or adapts its
behavior, it creates its own organization. In that respect it differs fundamentally
from our present systems, which are created by their designer” (p. 73).
Bonabeau, Dorigo, and Theraulaz (1999) give ant foraging trails as an example
of the spatio-temporal structures which emerge as a result of self-organisation.
The ability of ants to find efficient (i.e., short) routes between nests and food
sources suggests an approach to cost-effective, flexible, and implementable
wayfinding support. Paths identified by ants are not preplanned but emerge,
spontaneously, as a result of indirect communication between members of an
ant colony—a form of indirect social navigation. Dorigo and Di Caro (1999)
describe how ants deposit a chemical substance known as pheromone which
can be sensed by other ants. When a navigational decision has to be made, such
as taking a left branch or a right one, ants make a probabilistic choice based on
the amount of pheromone they smell on the branches. Initially, in the absence
of deposited pheromone, each of the branches is chosen with equal probability.
However, if one branch leads to food faster than the other, ants on their way
back will select the shorter branch due to the presence of the pheromone they
deposited on the forward journey. More pheromone is deposited, leading to
more ants selecting the shortest path, and so on, creating a feedback loop which
leads ants along efficient paths to their destination. This process of indirect
communication exploited by members of ant colonies is known as “stigmergy.”
In their overview article, Theraulaz and Bonabeau (1999) state, “The basic
principle of stigmergy is extremely simple: Traces left and modifications made
by individuals in their environment may feed back on them…. Individuals do
interact to achieve coordination, but they interact indirectly, so that each insect
taken separately does not seem to be involved in coordinated, collective
behavior” (p. 111). Stigmergy, self-organisation and ant colony algorithms are
the subject of much interest in the computer science community for optimisation
and routing problems (Di Caro & Dorigo, 1998; Dorigo, Bonabeau, &
Theraulaz, 2000; Schoonderwoerd, Holland, Bruten, & Rothkrantz, 1996).
The application of stigmergy is also being explored in the e-learning domain
(Dron, 2002; Dron, Boyne, & Mitchell, 2001), albeit not in the area of
wayfinding support.
In the educational arena, efficient paths are not defined in terms of distance but
rather time. We can imagine learners’ interactions with learning resources being
recorded automatically as they progress through a body of knowledge. The
time stamping of these interactions allows sequences to be identified which can
be processed and aggregated to derive a given “pheromone strength” favouring
paths which are faster to complete. This information can be fed back to other
learners, providing a new source of navigational guidance indicating “good”
ways through the body of knowledge—a self-organising, stigmergic approach
to wayfinding support. Such an approach seems to provide an answer to ODL
needs in this area. It is cost-effective, since trail creation occurs unnoticed as
a side effect of learner interaction with e-learning systems, it is flexible, able to
emerge from and adapt to different circumstances, and holds the prospect of
being implementable, since its adaptivity (Cristea & De Bra, 2002) does not
depend upon learner modelling. Indeed, such an approach abstracts entirely
from the characteristics of individual learners, relying instead on the collective
behaviour of the swarm of learners to identify efficient paths.
The next section introduces an architecture which supports the feeding back of
collective learner behaviour to support learners in reaching their educational
goals efficiently.
A Software Architecture
for Wayfinding Support
Our work on ODL wayfinding support is being carried out within the context
of a larger research and development programme, designed to help the creation
of flexible learning facilities that meet the needs of learners at various levels of
competence throughout their lives. We refer to these network facilities for
lifelong learners as “Learning Networks” (LNs) (Koper et al., 2004). LNs
support seamless, ubiquitous access to learning facilities at work, at home, and
in schools and universities. LNs consist of learning events, called activity nodes
(ANs), in a given domain. An AN can be anything that is available to support
learning, such as a course, a workshop, a conference, a lesson, an Internet
learning resource, and so forth. Providers and learners can create new ANs,
can adapt existing ANs, or can delete ANs. An LN typically represents a large
and ever-changing set of ANs that provide learning opportunities for lifelong
learners (“actors”) from different providers, at different levels of expertise
within the specific disciplinary domain.
Wayfinding support in LNs relies on the following concepts:
• The learner’s goal is a description of the level of competence a learner

wants to achieve (e.g., the bachelor’s or master’s level in a particular
discipline).
• A route is a plan to reach a goal, described as a series of selections and/
or sequences of ANs. ODL providers offer programmes with curricula
(i.e., routes) by which individuals can reach their goals.
• A learning track is the sequence of ANs successfully completed by a
learner.
• The learner’s position is the set of ANs that have actually been completed
(i.e., the learning track) together with those which can be considered as
completed, perhaps as a result of exemptions arising from previous study
or work experience.
Position and goal equate to “you are here” and “there’s where I want to be,”
respectively, and the wayfinding guidance, which is fed back, concerns
effective ways of getting from here to there, based on the behaviour of previous
learners along the available route(s).
The architecture we propose combines elements which record, collect, pro-

cess, and present collective learner behaviour. Andersson et al. (2002) use the
phrase emergent interaction systems to describe systems which “consist of
an environment in which a number of individual actors share some experience/
phenomenon. Data originating from the actors and their behaviour is collected,
transformed and fed back into the environment. The defining requirement of
emergent interaction is that this feedback has some noticeable and interesting
effect on the behaviour of the individuals and the collective—that something
‘emerges’ in the interactions between the individuals, the collective, and the
shared phenomenon as a result of introducing the feedback mechanism” (p. 1).
The “something that emerges” in our situation are paths through bodies of
knowledge, rather like well-worn footpaths in forests. Our initial focus is on
efficient paths, that is, those which minimise the time taken to reach a certain
goal from a particular position. Subsequent research will investigate attractive
paths, those rated highest by other learners, in line with work on recommender
systems (Herlocker, Konstan, Terveen, & Riedl, 2004; Resnick & Varian,
1997).
Central to the approach are logs of learner information indicating what learners
did and when. The use of Internet technologies in e-learning has brought with
it an increase in the level of standardisation of transmission protocols and data,
and logging information is no exception. The World Wide Web Consortium has
defined common and extended log file formats (World Wide Web Consortium,
1996) and a whole area of research is now dedicated to the processing and
analysis of these files for various purposes, known as Web usage mining (Punin,
Krishnamoorthy, & Zaki, 2001; Spiliopoulou, Pohle, & Faulstich, 1999;
Srivastava, Cooley, Deshpande, & Tan, 2000). The techniques have also been
applied in education (Sheard, Ceddia, & Hurst, 2003; Zaïane, 2001).
However, the events which are registered in these logs are extremely low level.
This complicates their analysis, making it difficult to know which users are
interacting (since only IP addresses are logged) and what they are doing (since
only cryptic uniform resource locators [URLs] are logged). Oberle et al.
(2003) note that “an interpretation of URLs in terms of user behaviour, interests
and intentions is not always straightforward … web usage analysis is not
interested in patterns of URLs but rather in patterns of application events” (p.
155). The route to solving this problem taken by Oberle et al. is to enhance the
logs with additional information drawn from a formal ontology. However, the
characteristics of our domain suggest a different type of log is more appropri-
ate, one incorporating a higher level of application event and which records not
Figure 1. A software architecture for wayfinding support for learners
Set goal
Retrieve route to
LN Functionality Goal system goal To Do List Calculator
Retrieve learner
TO DO list
Store interactions
Set exemptions Retrieve learner Recommender

LN Interaction Log Positioner
position
Retrieve Best
Retrieve Next
completed AN
data Transition Matrix
Retrieve learning
track for learner Calculator Best Next
Learning Track Calculator
Calculator
Retrieve all Retrieve
learning tracks Transition Matrix
Retrieve last Row
completed AN for
learner
only which learner did what, but also whether or not this was successful (e.g.,
by including the results of an assessment).
Such a level of logging is available in the learner records data store described
in the IEEE Draft Standard for Learning Technology—Learning Technol-
ogy Systems Architecture (IEEE, 2001). This data store, specifically de-
signed to cater for the nomadic nature of lifelong learners, is defined as a
repository of “learner information, such as performance, preference, and other
types of information. The learner records may store/retrieve information about
the past (e.g., historical learner records), but may also hold information about
the present (e.g., current assessments for suspending and resuming sessions)
and the future (e.g., pedagogy, learner, or employer objectives)” (p. 18).
With these notions in place, the elements of an architecture for self-organising
wayfinding support for learners can be introduced (Figure 1).
Learners interact with the LN Functionality available in a learning network
(Koper et al., 2004). Part of the functionality available allows learners to select
from a list of the learning goals in a learning network (the goal system), and
thereby also identify the route to the goal. Learner interaction is stored in an LN
interaction log (i.e., a learner record store, as described above), including
information on the learner, the AN, a time stamp, and an indication of
Figure 2. A transition matrix showing learner transitions from ANs (rows)

to other ANs (columns)
A B C D E
{} 1 3 2 4 5
A 4 2 5 1
B 2 2 1 3
C 3 4 1 2
D 4 2 4 5
E 1 2 5 3
performance (e.g., pass or fail). This information can be processed to create

sequences of ANs successfully completed by learners (done by the Learning
Track Calculator—see Mobasher (2004) for an examination of the tech-
niques involved). Using information on the tracks of all learners, a transition
matrix (Deshpande & Karypis, 2004) can be calculated (by the Transition
Matrix Calculator) over pairs of ANs, indicating, for each from node, how
many learners have successfully progressed to the following to node (see
Figure 2).
The Positioner deals with the maintenance of the ANs which have been
completed by learners, or can be considered as having been completed. The
former is straightforward to calculate, since it is the learning track for a given
learner. The latter, referred to as the Recognition of Prior Learning or Prior
Learning Assessment (Breier, 2005; Starr-Glass, 2002), is considerably more
complex (see Van Bruggen et al. (2004) for an examination of approaches to
this problem).
The To Do List Calculator maintains the difference between the requirements
expressed in the route associated with the learner’s goal, and his/her current
position. Using the transition matrix and the Learner’s To Do list, the Best Next
Calculator selects an AN to recommend based on the progress of the swarm
of other learners. The algorithm used to select the AN from the candidates is
that described by Koper (2005). Using the transition matrix shown in Figure 2,
if we imagine a learner having just completed the AN labelled “A” and en route
to a goal which requires A, B, C, D, and E to be successfully completed, a list
is first drawn up of all the transitions made from A by all previous learners (i.e.,
four from A to B, two from A to C, five from A to D, and one from A to E):
Figure 3. The prototype Recommender component
[B, B, B, B, C, C, D, D, D, D, D, E]
The recommendation is identified by drawing one item randomly from this list.
The result is that the most frequently followed path has a higher probability of
being selected (in this case A to D), although, to prevent suboptimal conver-
gence to this path, there is a chance that the other paths (A to B, A to C, and
A to E) will be selected. The use of randomness in the procedure follows the
ingredients for self-organisation described by Bonabeau et al. (1999).
The final component in the architecture is the Recommender, which pulls
together the various pieces of information to present a coherent picture to the
learner, including information on the learner’s goal, position, to do list, and the
recommendation itself. Figure 3 shows a prototype of the recommender,
implemented in the open source virtual learning environment, Moodle
(Dougiamas, 2004).
Discussion
This chapter has described the rationale behind our research into self-organising
wayfinding support, and described an architecture for its provision. Our
approach is designed to adapt support for decisions on the sequencing of
learning events not on the basis of a model of the individual learner but using
information on the collective behaviour of the swarm of other learners.
We are currently carrying out experiments to measure the actual value of the
approach using two groups of learners. One group of learners will receive
feedback on how others have progressed to their shared target, the other group
will be left to their own devices, and we will compare the numbers of learners
who manage to reach the goal in a given time period. The results of the
experiment will be used to determine whether to adopt the approach on a larger
scale in our institution.
In conclusion, our work is intended to open a new source of information to help
learners in deciding how to progress towards their learning goals using a
feedback loop on how others with shared positions and goals have fared. The
envisaged feedback loop has an advisory character, not intending to push all
learners down a single path as quickly as possible but rather to allow learners
to make informed choices concerning steps on their learning journey.
Acknowledgments
The authors would like to thank Francis Brouns and Hub Kurvers of the
Educational Technology Expertise Centre; Wim Waterink and Catherine
Bolman of the Psychology Department; and Pierre Höppener of the Marketing
department, all of the Open University of the Netherlands for their contributions
to this work.
References
Andersson, N., Broberg, A., Bränberg, A., Janlert, L.-E., Jonsson, E.,
Holmlund, K., et al. (2002). Emergent interaction—A pre-study.
Umeå, Sweden: Department of Computing Science, Umeå University.
Atif, Y., Benlamri, R., & Berri, J. (2003). Learning objects based framework
for self-adaptive learning. Education and Information Technologies,
8(3), 345–368.
Bonabeau, E., Dorigo, M., & Theraulaz, G. (1999). Introduction. In M. D. G.
T. E. Bonabeau (Ed.), Swarm intelligence. Oxford: Oxford University
Press.
Breier, M. (2005). A disciplinary-specific approach to the recognition of prior
informal experience in adult pedagogy: ‘rpl’ as opposed to ‘RPL’.
Studies in Continuing Education, 27(1), 51–66.
Brockett, R. G., & Hiemstra, R. (1991). Self-direction in adult learning:
Perspectives on theory, research, and practice. London: Routledge.
Brookfield, S. (1985). Self-directed learning: A critical review of research. In
S. Brookfield (Ed.), Self-directed learning: From theory to practice
(Vol. 25, pp. 5-16). San Francisco: Jossey-Bass.
Brusilovsky, P. (2001). Adaptive hypermedia. User Modeling and User-
Adapted Interaction, 11, 87–110.
Candy, P. C. (1991). Self-direction for lifelong learning. San Francisco:
Jossey-Bass.
Cristea, A., & De Bra, P. (2002). ODL education environments based on
adaptivity and adaptability. Paper presented at the World Conference
on E-Learning in Corp., Govt., Health., & Higher Ed. (ELEARN),
Montreal.
Darken, R. P., & Peterson, B. (2002). Spatial orientation, wayfinding, and
representation. In K. M. Stanney (Ed.), Handbook of virtual environ-
ments: Design, implementation, and applications. Mahwah, NJ:
Lawrence Erlbaum.
Darken, R. P., & Silbert, J. L. (1993). A toolset for navigation in virtual
environments. Paper presented at the ACM User Interface Software &
Technology, New York.
De Bra, P. (2000). Pros and cons of adaptive hypermedia in Web-based

education. CyberPsychology & Behavior, 3(1), 71–77.
De Bra, P. (2002). The adaptive Web—Adaptive educational hypermedia on
the Web. Communications of the Association for Computing Machin-
ery, 45(5), 60–61.
De Bra, P., Brusilovsky, P., & Houben, G.-J. (1999). Adaptive hypermedia:
From systems to framework. ACM Computing Surveys, 31(4).
Deshpande, M., & Karypis, G. (2004). Selective Markov models for predict-
ing Web page accesses. ACM Transactions on Internet Technology
(TOIT), 4(2), 163–184.
Di Caro, G., & Dorigo, M. (1998). AntNet: Distributed stigmergic control for
communications networks. Journal of Artificial Intelligence Research,
9, 317–365.
Dieberger, A. (2003). Social connotations of space in the design for virtual
communities and social navigation. In K. H. D. Benyon (Ed.), Designing
information spaces: The social navigation approach. London: Springer.
Dieberger, A., Höök, K., Svensson, M., & Lönnqvist, P. (2001). Social
navigation research agenda. Paper presented at the CHI ’01, Seattle,
WA.
Dorigo, M., Bonabeau, E., & Theraulaz, G. (2000). Ant algorithms and
stigmergy. Future Generation Computer Systems, 16, 851–871.
Dorigo, M., & Di Caro, G. (1999). The ant colony optimization meta-heuristic.
In D. Corne, M. Dorigo, & F. Glover (Eds.), New ideas in optimization
(pp. 11–32). London: McGraw-Hill.
Dougiamas, M. (2004). Moodle. Retrieved December 12, 2004, from http:/
/moodle.org/
Dron, J. (2002). Achieving self-organization in network-based learning
environments. Unpublished Ph.D. thesis, University of Brighton, UK.
Dron, J., Boyne, C., & Mitchell, R. (2001). Footpaths in the stuff swamp.
Paper presented at the Proceedings of WebNet 2001, Orlando, FL.
Evans, T. (1994). Understanding learners in open and distance education.
London: Kogan Page.
Fischer, G. (2001). User modeling in human-computer interaction. User
Modeling and User-Adapted Interaction (UMUAI), 11(1/2), 65–86.
Forsberg, M., Höök, K., & Svensson, M. (1998). Design principles for social
navigation tools. Proceedings of the 4th ERCIM Workshop on User
Interfaces for All, Special Theme “Towards an Accessible Web,”
Stockholm, Sweden.
Hadeli, P. V., Zamfirescu, C. B., Van Brussel, H., Saint Germain, B., Holvoet,
T., & Steegmans, E. (2003). Self-organising in multi-agent coordina-
tion and control using stigmergy. Paper presented at the First Work-
shop on Self-Organising Engineering Applications (ESOA 2003),
Melbourne, Australia.
Herlocker, J. L., Konstan, J. A., Terveen, L. G., & Riedl, J. T. (2004).
Evaluating collaborative filtering recommender systems. ACM Transac-
tions on Information Systems (TOIS), 22(1), 5–53.
Heylighen, F., & Gershenson, C. (2003). The meaning of self-organization
in computing. IEEE Intelligent Systems, Section Trends & Controver-
sies—Self-Organization and Information Systems, May/June.
Hiemstra, R. (1994). Helping learners take responsibility for self-directed
activities. In R. Hiemstra & R. G. Brockett (Eds.), Overcoming resis-
tance to self-direction in adult learning (Vol. 64, pp. 81–87). San
Francisco: Jossey-Bass.
Höök, K., & Benyon, D. (2003). Designing information spaces: The social
navigation approach. London: Springer.
IEEE. (2001). IEEE Draft Standard for Learning Technology - Learning
Technology Systems Architecture. Retrieved March 26, 2004, from
ltsc.ieee.org/wg1/files/IEEE_1484_01_D08_LTSA.pdf
Kay, J. (2001). Learner control. User Modeling and User-Adapted Interac-
tion, 11(1–2), 111–127.
Kilfoil, M., Ghorbani, A., Xing, W., Lei, Z., Lu, J., Zhang, J., et al. (2003).
Toward an adaptive Web: The state of the art and science. Paper
presented at the 1st Annual Conference on Communication Networks &
Services Research (CNSR 2003), Moncton, Canada.
Koper, E. J. R. (2005). Increasing learner retention in a simulated learning
network using indirect social interaction. Journal of Artificial Societies
and Social Simulation (forthcoming).
Koper, E. J. R., Giesbers, B., Van Rosmalen, P., Sloep, P., Van Bruggen, J.,
Tattersall, C., et al. (2004). A design model for lifelong learning
networks. Interactive Learning Environments (in press).
Mobasher, B. (2004). Web usage mining and personalization. In M. P. Singh

(Ed.), Practical handbook of Internet computing. Baton Rouge:
Chapman & Hall/CRC Press.
Oberle, D., Berendt, B., Hotho, A., & Gonzalez, J. (2003). Conceptual user
tracking. In J. S. E. M. Ruiz & P. S. Szczepaniak (Eds.), Web intelli-
gence, First International Atlantic Web Intelligence Conference
(AWIC 2003) (pp. 155-164). Berlin: Springer.
Punin, J. R., Krishnamoorthy, M. S., & Zaki, M. J. (2001). Web usage mining:
Languages and algorithms. In M. Schwaiger & O. Opitz (Eds.), Explor-
atory Data Analysis in Empirical Research, Studies in Classification,
Data Analysis, and Knowledge Organization (pp. 266-281). Heidel-
berg: Springer-Verlag.
Resnick, P., & Varian, H. R. (1997). Recommender systems. Communica-
tions of the ACM, 40(3), 56–58.
Schoonderwoerd, R., Holland, O. E., Bruten, J. L., & Rothkrantz, L. J. M.
(1996). Ant-based load balancing in telecommunications networks.
Adaptive Behavior, (2), 169–207.
Self, J. (1987). User modelling in open learning systems. In J. Whiting & D. A.
Bell (Eds.), Tutoring and monitoring facilities for European open
learning (pp. 219-237). Amsterdam: Elsevier.
Self, J. (1990). Bypassing the intractable problem of student modelling. In C.
Gauthier & C. Frasson (Eds.), Intelligent tutoring systems: At the
crossroads of artificial intelligence and education (pp. 18-24).
Norwood, NJ: Ablex.
Sheard, J., Ceddia, J., & Hurst, J. (2003). Inferring student learning behaviour
from Website interaction: A usage analysis. Education and Information
Technologies, 8(3), 245–266.
Shipman, F., Furuta, R., Brenner, D., Chung, C., & Hsieh, H. (2000). Guided
paths through Web-based collections: Design, experiences, and adapta-
tions. Journal of the American Society of Information Sciences, 51(3),
260–272.
Shipman, F., Marshall, C., Furuta, R., Brenner, D., Wei, H., & Kumar, V.
(1996). Creating educational guided paths over the world-wide web.
Paper presented at the ED-TELECOM, Boston.
Simpson, O. (2000). Supporting students in online, open and distance
learning. London: Kogan Page.
Soraya Kouadri, M., Rana, O. F., Foukia, N., Hassas, S., Giavanna Di Marzo,
S., Van Aart, C., et al. (2003). Self-organizing applications: A survey.
Paper presented at the Engineering Self-Organising Applications, First
International Workshop, ESOA 2003, Melbourne, Australia.
Spiliopoulou, M., Pohle, C., & Faulstich, L. C. (1999). Improving the
effectiveness of a Web site with Web usage mining. Paper presented
at the Proceedings of the KDD Workshop WEBKDD’99, San Diego,
CA.
Srivastava, J., Cooley, R., Deshpande, M., & Tan, P. (2000). Web usage
mining: Discovery and applications of usage patterns from Web data.
SIGKDD Explorations, 1(2), 12-23.
Starr-Glass, D. (2002). Metaphor and totem: Exploring and evaluating prior
experiential learning. Assessment & Evaluation in Higher Education,
27(3), 221-231.
Strachan, L., Anderson, J., Sneesby, M., & Evans, M. (1997). Pragmatic user
modelling in a commercial software system. Paper presented at the User
Modeling: Proceedings of the Sixth International Conference, UM97,
Sardinia.
Theraulaz, G., & Bonabeau, E. (1999). A brief history of stigmergy. Artificial
Life, 5(2), 97–116.
Van Bruggen, J., Sloep, P., Van Rosmalen, P., Brouns, F., Vogten, H., Koper,
R., et al. (2004). Latent semantic analysis as a tool for learner positioning
in learning networks for lifelong learning. British Journal of Educational
Technology, 35(6), 729–738.
Wexelblat, A. I. P. (1999). Footprints: History-rich tools for information
foraging. Paper presented at the Proceedings of Conference on Human
Factors in Computing Systems (CHI ’99), Pittsburgh, PA.
World Wide Web Consortium. (1996). Extended log file format. Retrieved
March 26, 2004, from www.w3.org/TR/WD-logfile
Yorke, M. (1999). Leaving early. Undergraduate non-completion in
higher education. London: Falmer Press.
Zaïane, O. R. (2001). Web usage mining for a better Web-based learning
environment. Paper presented at the Proceedings of the Conference on
Advanced Technology for Education, Banff, Alberta, Canada.
Zeiliger, R., Reggers, T., Baldewyns, L., & Jans, V. (1997). Facilitating Web
navigation: Integrated tools for active and cooperative learners. Paper
presented at the Proceedings of the 5th International Conference on
Computers in Education, ICCE ’97, Kuching, Sarawak, Malaysia.
Designing for a Social Constructivist Environment 185
Chapter IX
From Engineer
to Architecture?
Designing for a Social
Constructivist Environment
Karen Lee, University of Dundee, UK
Abstract
This chapter examines the design requirements of a social constructivist

virtual learning environment. It uses the example of teaching expertise to
practitioners to demonstrate the practical application of the cognitive
theories underpinning a community of practice, with the objective of
providing an insight into some of the complex issues involved in creating
such an environment. It is argued that the analogy of designer as architect
is most appropriate, reflecting a move from task-centred courseware to
learner-centred situated environments.
186 Lee
Introduction
The focus of this chapter is on the marriage of instructional approaches, based

on the theory of the cognitive process involved in learning, with appropriate
technology, to facilitate teaching and learning. Specifically, this is examined in
the design of a social constructivist virtual learning environment (VLE).
From a teaching perspective, Lynch, back in 1945 (in Ertmer & Newby, 1993)
called for employing an engineering analogy as an aid for translating theory into
practice, providing a bridge between learning research and educational
practice. More recently, Hung (2001) puts forward the idea of teachers as
“pedagogical engineers” with the “responsibility to plan a lesson(s) with the
most relevant instructional approaches and technologies at his or her disposal”
(p. 286).
From the technologist’s perspective, Bednar, Cunningham, Duffy, and Perry
(1995) contend that the field of instructional systems technology prides itself on
being an eclectic field, “Dewey’s proverbial linking science between theories
of the behavioural and cognitive sciences and instructional practice” (p. 100).
The primary strategy for providing this link has been to collect concepts and
strategies suggested by the theories and make them available to the practitio-
ners.
More recently, Goodyear (2000) highlights the “emerging responsibility of
educational technologists to move beyond the design of instructional tasks to
the requirement to become architects and creators of virtual learning environ-
ments reflective of real world activities” (p. 1) and to move toward student-
centred educational, technology-mediated learning. The designer as architect
creates online learning spaces that afford and nurture learning but do not
determine it.
This chapter will recap the drivers toward a social constructivist learning
environment by examining the change in cognitive theories underpinning learn-
ing, and the changes in the higher education sector. Teaching expert profes-
sional practice to distance-learning students at university level will be used as
an example of cognitive strategies that might be used, giving examples of how
these have been employed in a managed learning environment in the author’s
institution. The issues involved in creating and supporting such a system are then
discussed.
Background
Educational theory has shown a recent shift from an acquisition metaphor to a

constructivist view, and from a focus on the individual to an emphasis on the
social context of learning (Mayes, 2001) which argues that learning does not
occur in isolation, but requires emotional and personal support from others,
context, feedback, and reflection (Alexander & Boud, 2002). To highlight the
resulting changes at a practical level, a generalisation of the main groups of
theories will be presented.
Behaviourism
In behaviourism, learning is seen as a change in observable behaviour. Key

concepts include stimulus response, and conditioning and reinforcement,
through the work of Thorndike, Watson, Pavlov, and Skinner (Bartlett, Burton,
& Peim, 2001). Learning is the correct response to an environmental stimulus.
A principle is that behaviour that is positively reinforced will recur. This type of
learning may be effective in some situations, such as the learning of a manual skill
or rote repetition of facts, but ignores the cognitive processes involved. Early
computer-aided instruction (CAI) was based on behaviourist principles.
Cognitivism
In contrast, from recognition of these limitations, cognitive learning theory

attempts to explain the cognitive process involved in learning such as knowing,
understanding, remembering, and problem solving. Concurrent progress in
computing resulted in an information-processing model of learning, with
memory taking a central role. Incoming information is attended to, encoded,
and stored for subsequent retrieval. Information is organized in schema (mental
maps) developed from previous knowledge and experience, similar to frames
of knowledge within artificial intelligence (Bartlett et al., 2001). CAI supplying
appropriate responses to student input is based on a recognition of the
cognitive processes being employed.
188 Lee
Constructivism
Constructivists believe that individuals actively construct knowledge through

problem solving, utilizing prior knowledge, experience, and beliefs. Knowl-
edge stored is therefore not fixed but changes. Learning is interpreting the
environment from an individual’s perspective and creating meaning from
experience. Here computer “instruction” is a process of supporting that
construction rather than communicating chunks of information (Lefoe, 1998).
Social Constructivism
Social constructivism recognises individual construction of knowledge but

believes that the interpretation of knowledge is dependent on the cultural and
social context through which the knowledge was constructed, with interaction
with other people, mediated by language, crucial in cognitive development
(Hung, 2001). Learning is not just an individual process but is influenced by
interaction with others.
In practical terms, these can be viewed as a hierarchy of complexity in terms
of cognition, the computer environment and tutoring processes involved. Edgar
(1995) uses the analogy that the “PC is to Piaget as WWW is to Vygotsky” (p.
1). There has been much debate about whether these theories are mutually
exclusive (Goodyear, 2000). Ertmer and Newby (1993) believe that the key
question for instructional designers is which theory is the most effective in
fostering mastery of specific tasks by specific learners, which is “not to suggest
that one should work without a theory, rather that one should be able to
intelligently choose the appropriate methods for achieving optimal instructional
outcomes in that situation” (p. 67). In contrast, Bednar et al. (1995) believe that
“abstracting concepts and strategies from the theoretical position that spawned
them strips them of their meaning,” arguing that instructional design requires to
be based on a theory of learning and/or cognition, with effective design only
possible where the designer has a “reflexive awareness” of, and deliberately
applies, the theoretical basis underpinning the design (p. 102). Hung (2001)
proposes that for a balanced perspective, a distinction should be observed
between cognitive paradigms and instructional approaches. Thus, for example,
reservations surrounding a behaviourist paradigm should not cause us to
discard CAI as an instructional approach.
It is widely accepted that different approaches are required at different stages

in the complex process of learning. Ertmer and Newby (1993) describe a
continuum of knowledge from knowing what (behavioural), to knowing how
(cognitive), then reflection in action (constructivist). Similarly, different tasks
require different levels of cognitive processing, with tasks requiring high levels
of processing best learned by social constructivist strategies such as situated
learning, cognitive apprenticeship, and social negotiation.
Jonassen, Mayes, and McAleese (1993) have linked these to the stage of the
learner. At the first phase of introductory learning the learner has little prior or
transferable knowledge and is in the early stages of schema development. At
this stage behaviourist and cognitivist instructional design to achieve predeter-
mined learning outcomes are effective. At the second phase of advanced
knowledge acquisition, as in a university setting, constructivist approaches are
appropriate to support complex problem solving. The final phase is the
development of expertise, which I will argue requires a social constructivist
approach.
In summary, the significant, practical differences from a design perspective are
that while objectivist (behaviourist and cognitivist) approaches transmit knowl-
edge to achieve prescribed outcomes, constructivist approaches facilitate learn-
ers playing an active role in discovering the environment rather than reacting to
it, the learning outcomes of which may not be predicable (Jonassen et al., 1993).
This requires that the systems approach is modified to accommodate constructivist
values (Schwier, 1995). However, Jonassen (1994) reassures us that “constructivist
environments are not the unregulated, unsupported, anarchic, sink-or-swim,
open-discovery learning cesspools that many fear” (p. 35).
The remainder of the chapter examines what a constructivist environment might
look like, using an example from the author’s institution.
Expertise in a Social Constructivist

Virtual Environment
What is Expertise?
The module in question is for nurses aiming to become expert practitioners.

Benner (1984) applied the Dreyfus model of skill acquisition to nursing
190 Lee
practice, defining an expert as someone who no longer relies on analytical

principles such as rules and guidelines when assessing a situation and acting
appropriately, having a deep and intuitive grasp of the whole situation from
extensive experience. An expert nurse practitioner is thus required to develop
higher levels of judgement, discretion, and decision making in his/her clinical
practice. As nursing is situational, much of the knowledge needed for effective
practice is embedded in practice itself. This presented a challenge in converting
from a face-to-face to distance mode of delivery. However, it has been
observed that distance learning provides a unique context in which to “infuse”
constructivist principles as it is expected that learners will behave as “self-
motivated, self-directed, interactive, collaborative participants in their learning
experiences” (Tam, 2000, p. 1).
A social constructivist virtual environment is most suitable to facilitate this,
offering an approach to learning which matches the affordances of the technol-
ogy, by emphasizing student centredness and by enabling learning to be related
to context and to practice (Oliver & McLoughlin, 1999) with a focus on
knowledge construction not reproduction, reflective practice, context- and
content-dependent knowledge construction, collaborative construction of
knowledge through social negotiation, and multiple representations of reality
(Jonassen et al., 1993) and tasks authentic to the practice field in which the
knowledge and skills are embedded (Mayes, 2001).
The virtual learning environment available within the author’s institution was
Blackboard (version 5). Blackboard was used primarily to provide resources
and allow computer-mediated communication (CMC).
Cognitive Underpinnings of the Approach Used
Garrison (2002) contends that the connectivity and asynchronicity provided by

CMC is congruent with the reflective, metacognitive, self-directional, and
collaborative dimensions of the deep learning required; “there is every reason
to believe that asynchronous online learning can be extremely effective in
supporting higher-order learning and creating quality cognitive presence online”
(p. 2). He further argues that CMC has the potential to significantly enhance
learning environments and outcomes, based on its ability to support virtual
communities of inquiry. Palloff and Pratt (1999) agree, observing that “commu-
nity” and “communicate” have the same root “communicare” which means “to
share.” Through this communication of language, the members of a community
“learn to carve out the world in similar ways; they develop similar anticipations
about external reality” (Hung, 2001, p. 282).
Shaffer and Anundsen (1993, in Palloff & Pratt, 1999) define community as a
dynamic whole that emerges when a group of people share common practices,
are interdependent, make decisions jointly, identify themselves with something
larger than the sum of their individual relationships, and make a long-term
commitment to well-being (their own, one another’s, and the group’s). Most
appropriate to the author’s context are “communities of learning” and “commu-
nities of practice.”
The “learning community” is the vehicle through which learning in a cohort of
students occurs online. Members depend on each other to achieve the learning
outcomes of the course (Palloff & Pratt, 1999). Communities of practice are
“groups of people who share a concern, a set of problems, or a passion about
a topic, and who deepen their knowledge and expertise by interacting on an
ongoing basis” (Wenger, 2004, p. 1). McConnell (2002) distinguishes the two
in that communities of practice do not set out to focus on their learning, whereas
learning communities deliberately do so, and look reflectively at their learning
processes. In contrast, a community of practice is a social phenomenon where
learning is the lived experience of making meaning in a group situation. Wenger
(2004) observes that in the real world we learn most in the process of becoming
part of a community and contributing to what it is doing.
Within a community of practice, two theories of cognition are important—
situated cognition and cognitive apprenticeship.
Situated Cognition
Learning in a community of practice is predominately by situated cognition

(Lave & Wenger, 1991), which asserts that knowledge and understanding are
a product of the learning situation and the learning activity, being embedded in
that context (Miao, 2004), with an active learner and the environment part of
a mutually constructed whole (Hung, 2001).
Lave and Wenger’s work on situated learning shows the importance of
legitimate peripheral participation for enabling effective learning in many
(perhaps most) contexts, with learners initially learning from observing but
gradually gaining knowledge and skill, as well as the values and thought patterns
of the experienced practitioners. In a community of practice, the community
models the role and language of a practitioner, and the individual’s identity is
192 Lee
shaped by participating within the community (Wenger, 1998). This is a form

of cognitive apprenticeship.
Cognitive Apprenticeship
Cognitive apprenticeship (Brown, Collins, & Duguid, 1989) is a situated model

of teaching where the teachers’ thinking, problem-solving, and decision-
making processes are made visible to the student to facilitate their learning of
these cognitive and metacognitive skills. Techniques identified by Brown,
Collins, and Duguid are modelling, coaching, scaffolding, articulation, reflec-
tion, and exploration. Jonassen et al. (1993) contend that this is an appropriate
model for providing real-world experiences. The combination of the two
approaches is important and synergistic. Apprenticeship implies working in a
situation in which the knowledge/skills learned are to be applied. Thus, “the
situated context structures knowledge appropriate to its use, modelling expert
performance and making tacit knowledge explicit” (Jonassen et al., 1993, p. ).
In addition, students are able to reflect on their performance in relation to the
expert and other students.
A similar theory that supports the value of working with others and of a
scaffolded model is that of Vygotsky’s zone of proximal development. This is
defined as the difference between independent problem-solving ability and
potential ability under guidance (Hung & Chen, 2001). Learners can only reach
higher cognitive levels if appropriate scaffolding is provided by more competent
peers or the teacher (Faggiano, Roselli, & Plantamura, 2004). Mentis, Ryba,
and Annan (2001) introduce the concept of a “collective zone of proximal
development,” where learners form an intellectual collective where there is both
participation and mutual guidance between peers, resulting in individual learn-
ing for all participants as well as group learning. There is also integration of
learning with the workplace: McConnell (2002) from analysis of CMC
transcripts observes that the meaning discerned in discussions is taken back to
the workplace, applied, and tested, thus developing the learner’s professional
practice. Knowledge and insights gained from this are then taken back to the
group and used in its projects. This results in a weaving together of work,
theory, and practice that “becomes almost natural.”
In summary, situated learning is based on social interaction where learners learn
together as a community of practice where the community embodies certain
beliefs and behaviours, based on the assumption that learning is a direct function
of the activity, context, and culture in which it takes place.
How did we Attempt to do this on Blackboard?
Students were trained nurses working toward a degree in the field of infection
prevention and control, in a variety of acute hospital, community, and public
health settings. As stated previously, Blackboard was used primarily to provide
resources and allow CMC in the practice context. Resources included key
professional documents and guidelines required for practice. Students were
asked to apply resource material to practice scenarios on the discussion
boards, drawing on their work experiences. Other activities included, for
example, identifying good and bad practice from photographs of clinical
settings, and seminar PowerPoint presentations on current topics in the field.
Each cohort comprised approximately 20 students, predominately from across
the United Kingdom.
Collaborative learning included group problem solving of practice incidents on
the discussion board including PBL (problem-based learning) moderated by
tutors who were specialist practitioners and visiting experts in the field. In PBL,
students were presented with a practice scenario such as a transcript of a phone
call. Students individually identified their learning needs and then asked
questions on the discussion board and responded to those of others. This raised
awareness of different approaches in different practice settings and countries,
causing students to reflect on the assumptions underlying their own practice,
and come to a group consensus on the actions they would take. Learning from
previous students’ experiences was facilitated by the use of frequently asked
questions, footprints (messages left for those following behind), and examples
of good work.
Discussion
What are Some of the Issues Involved in Designing a

System Such as This?
Technological Infrastructure
Turoff (1995) observes that the sophistication and flexibility of software

designed to support distance education varies widely. The author has discussed
194 Lee
the use of a VLE, defined as “learning management software systems that

synthesise the functionality of computer-mediated communications software
and online methods of delivering course materials” (Britain & Liber, 1999, p.
2). Issues of access, cost, and organizational issues, including novelty of use
and speed of implementation usually result in teachers requiring to use a
commercially designed VLE (Mayadas, Bourne, & Moore, 2003). The aim of
commercial VLEs is to enable them to be used by instructors with little
knowledge of HTML and Web navigation, but these constraints have a
negative impact on flexibility (Storey, Phillips, Maczewski, & Wang, 2002). As
can be seen by the examples, in addition to asynchronous discussion boards,
students are required to access and read documents, give PowerPoint presen-
tations, and engage in group work. Hung (2001) has identified five types of
tools: “individual instructive tools,” “informative tools” (materials and re-
sources), “individual constructive tools” (generic purpose tools, e.g., word
processing), “social communicative tools” (e-mails, video conferencing), and
“social constructive tools” (e.g., document sharing), noting that in the course of
a problem-solving activity, learners could require all of these, thus an environ-
ment is needed in which these can all be integrated. Blackboard was very
limited in its ability to do this, primarily acting as a CMC board and resource
repository. For example, word processed documents could be posted only as
attachments and opened with students’ own software. PowerPoint presenta-
tions required to be e-mailed to instructors, published as Web pages then
uploaded into Blackboard. The environment was supplied segregated into
different sections such as “course information,” “assessment,” “learning mate-
rials,” and “discussion board,” with very limited customisation available. This
led to a tendency for fragmentation of learning activities, particularly evident
where discussion took place away from the associated directions and re-
sources. As observed by Klemm and Snell (1996), the threaded discussion
system provided does not aid team building and effective group working.
Facilities for group work were limited to separate group areas with a discussion
board and student file share, but no ability to add learning materials, or for
student-shared editing of work or resources. A basic virtual classroom was
available for synchronous discussion but required at sometimes lengthy down-
load of a Java Plug-in, which again caused problems at a distance, and was
blocked by some hospital firewalls. In addition to known accessibility prob-
lems, this raised issues of equity.
Learner Factors
Biggs (1999) notes that there have been extraordinary changes in the structure,
function, and financing of universities within the past decade. These include
childhood to adult to lifelong learning, teacher-centred to student-centred, the
few to the many, theoretical to practical/vocational, and of course face-to-face
to distance, meaning that the traditional view of a course and student group is
no longer valid (Jarvis, Holford, & Griffin, 1998). Where previously students
in the author’s department were school leavers, many are now mature adults,
and increasingly studying part time at a distance. The learners therefore tend to
have an expectation of knowledge being delivered to them. Many also lack
study skills such as self-direction and time management, and have not been
exposed to group work. IT skills are often also poor. As found by previous
authors, students had a “considerable degree of anxiety” (Moore & Kearsley,
1996, p. 155) requiring both academic and nonacademic support (Simpson,
2002). There was therefore a need to scaffold IT and self-directed study skills.
Initial activities required to be easy to find with clear and comprehensive
instructions. At the beginning of the module, additional resources provided
needed to be simple to use, for example, a direct Web link to a passage of
interest rather than the need to search or navigate an external site. There was
a tendency for students to attribute any problems, such as nonfunctioning Web
links, to their perceived incompetence rather than the technology. This further
increased their anxiety, requiring reassurance, and where possible warnings
such as the need for Acrobat Reader, or that a document may take a few
minutes to download. Nevertheless, all students were able to use the basic
functionality of Blackboard competently within a few weeks. However, an
attempt to use an electronic portfolio facility to allow students to share their
work and enable peer assessment was unsuccessful due to the high stress levels
generated in students and high workload for staff in trying to guide students at
a distance through the complex sequence of technical procedures required.
CMC, Collaboration, and Social Negotiation
Using asynchronous discussion forums requires that a new student learn not
only about the medium and how to use it practically, as discussed above, but
also the rules of a new type of discourse (Kaye, 1989). Review of the literature
by Bullen (1998) suggests many of the attributes of CMC, vis-à-vis place and
time independence, many-to-many, and text-based communication, can pro-
196 Lee
mote or discourage participation in CMC. Similarly, Hammond (2002) notes

its attributes may be two edged, for example, permanency allows reflection but
means comments cannot be undone. McDonald (2002) suggests potential
negative impacts include a lack of responsiveness, hesitancy in committing
ideas/feelings to print, and misunderstandings as a result of the lack of
nonverbal communication. However, she points out that while these are
deterrents for some participants, in reality, research has found that more
students respond, responses are longer and more complex, and interactions are
increased compared to face-to-face. In a study by Bullen (1998), shy students
reported CMC as liberating, and personality conflicts and distractions have
been noted to be less intense (Klemm, 1995); however, others experience
feelings of detachment, and anxiety in terms of being unsure who the student is
communicating with, when to expect a response, and what type of response it
will be (Bullen, 1998). Hammond (2000) suggests there is a high threshold for
many to cross to participate. Its value is also dependent on the knowledge and
experience students can bring to the discussion, thus it is particularly suited to
professional, adult learning (Kaye, 1989). Harasim (1989) notes that CMC is
a generic tool, and to be successful in collaborative learning is difficult and
requires a tremendous amount of effort by the tutor. The more supports or
“scaffolds” built in to support the learner, the more likely it is to succeed.
The scaffolding model we employed to structure activities to facilitate collabo-
rative and deep learning was Salmon’s (2000) five-step model, which moves
from socialization and “using the technology” activities, with a lot of teacher/
moderator support, working up to knowledge construction, and ultimately
emancipation, thus scaffolding students in progressively acquiring the required
skills at each level. All material was written in a friendly tone, addressed to the
student. The first activity was to introduce themselves to their fellow students,
thus requiring no preparation and with no right or wrong answers, both to help
students get over the first hurdle, and to begin to get to know each other.
Optional face-to-face tutorials were provided and students who had met each
other at these reported that it was easier to converse on Blackboard. However,
where students were local enough to attend several, there was a tendency to
reduced participation on Blackboard perhaps due to less of a feeling of
isolation. The ability for students to add a photograph or biographical details
to Blackboard might assist socialisation for students unable to have any face-
to-face contact. Photographs and contact details of the tutors are provided on
Blackboard. In addition, information was provided on netiquette and good
practice was modelled by the tutors.
To enable students to learn from each other’s experiences, differentials in type

and amount of experience are required; however, it was necessary to design
and sequence activities so that everyone, even those new to the job, felt they
had something useful to contribute. It was important to time activities to keep
the momentum going so that students could contribute each time they logged in,
without causing information overload. To give everyone an opportunity to
contribute, students were initially restricted to giving only one part of the answer
to early activities to prevent a student logging on to find everything they wanted
to say had already been said. Later in the module where activities required deep
learning responses, there was occasionally a delay with students appearing not
to want to be the first to respond. This was overcome by starting the discussion
with an experience-based question then moving on to the more theoretical
issues. Moderators of the discussions thus had a major role to play in initiating,
shaping, clarifying, explaining, weaving, and summarising discussions and
bringing them to a conclusion, in a climate that encourages and role models
negotiation and critical analysis (Kaye, 2003; Garrison, 2002). A lot of tutor
input and encouragement was required initially, but by the end of the module,
the tutor became largely superfluous, with students helping, answering, and
challenging each other. Later activities aimed at fostering social negotiation
skills as well as deep learning with students required to articulate their practice,
reflect on differences, examine the underlying theory, and either defend their
position or agree a change or compromise.
Examples of Activities
Some examples of students’ activities are given below.
1. An early, simple, initially individual activity, limited to giving only part of the
answer, but one where the total problems identified would be greater than any
one student could achieve on his/her own:
“What is one point that you could make about the attached photo that
demonstrates an aspect of decontamination practice? Maybe it is a good
practice that you can identify, or a poor practice that could be improved.”
198 Lee
2. A slightly later, structured activity with resources and detailed guidance

provided, but where the students are required to begin to work as a team.
“What surveillance you might do as a result of the scenario…. Using the

suggested steps in the paper by Lee and Baker to help focus your thoughts
(activities 3.19 and 3.20), what surveillance will you undertake? If there
is more than one approach suggested, try to come to an agreement as a
team.”
3. A late, unstructured activity to generate debate, once students are familiar

with each other.
“What did you think of the article? Do you think it has implications for
infection control? Do you see health promotion as part of your role?”
4. This activity occurs at the end of the module as a group versus group activity.
The scenario begins with one infectious patient and evolves depending on the
decisions made by the students, leading to very complex situations, requiring
critical application of theory to practice and social negotiation:
“By today, you have eight wards closed, including three in the surgical
block, but all your control measures are in place and you think you might
be getting on top of the outbreak. However, as you all anticipated, the
local press are now making inquiries. How do you convince the public and
hospital workers that the hospital is safe? What are some of the wider
management issues that the outbreak team might take into consideration
in deciding whether to close the hospital?”
Community
Ryan (2001) also suggests “a skilful facilitator can create a community of

learners which can replicate the social learning atmosphere needed by most
learners” (p. 89), and can reduce the lack of support that leads to student
withdrawal. Student factors are also important, for example, social presence,
defined as the “ability of learners to project themselves socially and affectively
into a community of enquiry” (Rourke, Anderson, Garrison, & Archer, 2001,

p. 3).
Is it possible, therefore, to design community into a learning environment?
Conrad (2002) likens the attempt of course designers to build community
through structure to an “arranged marriage,” where learners are pushed not
pulled into community; however, online literature suggests students will not
collaborate unless it is built into the course (Vonderwell, 2003). Similarly,
Conrad herself proposed from her study that participation in online learning
activities exists before community that it contributes to community, that it is the
vehicle for maintaining community and that it eventually becomes the measure
of the health of the community. Whilst agreeing that communities need their own
dynamic and it might not be possible to create them artificially, Goodyear
(2000) argues that we can lay out the appropriate environment and frameworks
within which they can function effectively.
Did it Work?
Palloff and Pratt (1999) provide indications that an online community has been
forming based on the premise that “socially constructed meaning is the hallmark
of a constructivist classroom in which an active learning process is taking place”
(p. 32)—the ability to collaborate and create knowledge and meaning commu-
nally.
These indications that an online community has been forming are as follows:
• Active interaction involving both course content and personal communi-

cation
• Collaborative learning evidenced by comments directed primarily student
to student rather than student to instructor
• Socially constructed meaning evidenced by agreement or questioning,
with the intent to achieve agreement on issues of meaning
• Sharing of resources among students
• Expressions of support and encouragement exchanged between students,
as well as a willingness to critically evaluate the work of others. (p. 32)
200 Lee
The (abbreviated) examples below, taken from the discussion board of a PBL
scenario surrounding meningitis, demonstrate that these criteria were met, with
formation of community and social construction of knowledge:
“Hi, Karen, If meningitis was suspected and as benzylpenicillin was given,

confirming that this was suspected, why was the patient admitted to an
open ward? All hospital staff should be aware that the patient must be
isolated due to the risk of airborne transmission.”
“The Trust I work in follows a 48-hour rule of isolation including

antibiotic treatment during this period. Therefore, I would feel that the
patient being nursed in an open bay prior to 48 hrs of treatment could pose
a potential risk to others. Immunosuppressed patients, visitors, and staff
could all be affected.”
“Meningitis is spread by respiratory droplets, nasal or oral secretions. I

believe from my conversations with HPA (health protection agency) that
it is not as easy as is commonly thought to pass on to others. One method
is by kissing which would not be taking place (I hope!) between the patient
and healthcare staff, or patient to other patients. Meningitis does not
survive for a long period outside the body. Therefore, with the above in
mind, although we also isolate until the patient has received 24 hours of
antibiotics, perhaps this is not necessary, and it begs the question that if
the isolation is so important what about the ambulance drivers who may
have brought the patient in (not specifically thinking about this question
now)?”
“I feel that there is perhaps an increased risk of others being susceptible

within a hospital as patients are immunocompromised, and perhaps this
is the reason we isolate, but I have a feeling that it is because it has a feel-
good factor to isolate. It is certainly something I will think about and look
into, but also one of the symptoms of meningitis can be photophobia and
therefore the patient may prefer a side room.”
“Hello, everyone, I suppose a risk assessment should be made with regard

to isolation, there is evidence to show that patients isolated for the
purposes of infection control are less likely to receive care from staff (all
grades), so the morbidity may increase. If this pt. is very sick it may be
unsafe to isolate.”
“Ok, I’ll defend my position! I have found a reference which states that on
admission to hospital all patients with known or suspected meningitis
should be isolated in a single room, for at least 24 hours following the start
of antibiotic therapy. Damani N (2003) I will continue to look for further
evidence to support my previous statement of 48 hours.”
“The HPA states that the meningococcal bacteria is not highly infectious
and rarely develops into meningococcal disease. However, if infection
does occur it can spread rapidly and is fatal in about 10% of cases.
Statistics demonstrate that 1 in 8 people who do recover can experience
long-term effects, including headaches, joint stiffness, epileptic fits,
deafness, and learning difficulties. I suggest that the risk of mortality and
the evidence of morbidity following this infection would be factors
indicating that isolation is advisable!”
“Hi, Pam, I would agree with you that it seems sensible to isolate
especially as although it may be relatively difficult to transmit, the
consequences of having it transmitted to you can be enormous, I was just
playing devil’s advocate really and trying to work through the process.”
“Mel, thanks, have enjoyed the lively debate. Still looking for evidence!
Pam”
“The HPA site has an article describing an outbreak of meningococcal

disease among university students which I have found interesting and is
useful comparison to this scenario; Communicable Disease and Public
Health Vol 2 No 3 September 1999.”
“It certainly made me think about the practice we carry out at the Trust
I work in.”
202 Lee
Table 1. Number of messages at each SOLO level

SOLO Le vel Number of Messages
Forum 1 Forum 2
Prestructural 0 0
Unistructural 0 0
Multistructural 3 2
Relational 6 4
Extended Abstract 18 13
Depth of Learning
Looking at the depth of learning achieved, Beer et al. (2002) examined a similar
context to the author (use of PBL amongst occupational therapists), using the
SOLO (Structure of the Observed Learning Outcome) taxonomy to examine
transcripts for deep learning in group work. Thought process in messages
relevant to the learning outcomes were coded as prestructural, unistructural,
multistructural, relational, and extended abstract (Biggs, 1999). Examining two
discussion forums using this methodology (Table 1) demonstrated that the
majority of messages were extended abstract, indicating deep learning. There
were no messages at prestructural or unistructural levels. This is a function of
the ability of the students, plus the fact that limited information is given at the
start of PBL requiring students to think laterally and explore surrounding issues,
as shown in the example above.
Student Comments
Comments on the student end of module evaluation questionnaires reflected

similar themes:
“I like to have contact and reassurance that what I am doing/thinking is

along the right lines or completely off the wall. Blackboard has given me
that communication.”
“It was all so new to me, I mean looking at everything from an ICN’s point
of view. I felt I couldn’t contribute much to the discussion but I did enjoy
reading what everyone had to say. I feel more confident about using
Blackboard now and I’m sure I will give my opinion more in the next
module.”
“Blackboard has been a great support throughout the last few modules.
Being able to see other peoples’ comments has helped me learn and
sometimes look at things differently.”
“Although I didn’t often have much to say I’ve enjoyed reading the
messages and have picked up on things that I hadn’t thought of, or was
just about to ask myself.”
“Thank you all for your comments, your support, your good humour and
help during the modules we’ve used blackboard for. I’ll miss visiting the
site when it finally closes.”
Conclusion
The examples above show a successful social constructivist virtual learning

environment in action, with deep learning evidenced by individual and commu-
nity knowledge construction through collaboration and social negotiation of
meaning. Improvements in the technology would have further facilitated this,
such as provision of good group-working tools and transparent technology,
requiring minimal skills and support. Research suggests that when tools are
difficult to navigate or frustrating to the user, they can have a negative effect on
learning, as we found (Storey et al., 2002). In addition, tools to improve
socialisation would be beneficial, such as synchronous chat and student
resumes.
As previously discussed, the principle differences in designing for a constructivist
environment are the focus on process not product, outcomes that are flexible
not predetermined, and the use of open-ended rather than closed systems. To
return to the analogy of engineer to architect, there is still a need to engineer;
interactions need to be planned and managed, including those that happen
spontaneously in classroom settings (Hirumi, 2002), and good design of
activities and scaffolding of skills (such as IT, lifelong learning, group working,
204 Lee
and critical thinking) is essential. However, the essential difference is that the
product cannot be predetermined—each student will construct his/her own
knowledge based on the experiences he/she brings and his/her engagement
with the task and community. Furthermore, the knowledge generated by each
cohort of students will be different—we are merely architects of the environ-
ment.
Jonassen (1994) suggests that a constructivist design process should be
concerned with designing environments, which support the construction of
knowledge, a meaningful authentic context, collaboration among learners and
with the teacher, be designated the three C’s (context, construction, and
collaboration). Whilst these still hold true, I would like to extend these to the
three S’s—situated, social, and student centred. The shift required is in using
technology to create environments that support authentic activity rather than
creating courseware in which learner tasks are prescribed (Goodyear, 2000).
Well-designed tasks based on cognitive theory are still required, but as a
resource for situated learning activity. The second difference is that a “space”
becomes a “place” in which a learning community can develop (Goodyear,
2000).
In conclusion, in the author’s experience, the albeit limited technology of the
Blackboard VLE has been effective in enabling distance learning to be situated
in a practice context, and providing the virtual space for a community of
practice in which students could collaborate in authentic tasks, facilitating the
application of theory to practice, and the development of metacognitive skills
required for specialist nurse practitioners. The architecture required for this is
an environment emotionally and intellectually conducive to learning, which
facilitates the cognitive processes involved, such as conflict or disagreement,
explanation, internalization, appropriation, shared cognitive load, mutual regu-
lation, and social grounding (Dillenbourg, 1999, in Goodyear, 2002). Ulti-
mately however, whether this occurs depends on a complex interaction of
learner, group, tutor, activity, and environmental factors. Feenberg (1989)
observes that organising groups in the “written world’ demands an unusual
insight into group processes as well as an awareness of the technical features
of communications systems; “failures and breakdowns occur at the social level
far more often than at the strictly technical level” (p. 8). “The quality of
learnplaces, the conviviality of learning communities, the richness of learning
activities: these are key. But we may be mistaken in believing we can work on
these things directly” (Goodyear, 2000, p. 15). However, good architecture
can nurture them.
References
Alexander, S., & Boud, D. (2002). Learners still learn from experience when
online. In J. Stephenson (Ed.), Teaching and learning online (pp. 3-15).
London: Kogan Page.
Bartlett, S., Burton, D., & Peim, N. (2001). Introduction to educational
studies. London: Paul Chapman Publishing.
Bednar, A. K., Cunningham, D., Duffy, T. M., & Perry, J. D. (1995). Theory
into practice: How do we link? In G. Anglin (Ed.), Instructional technol-
ogy: Past present and future (pp. 100–112). Englewood, CO: Libraries
Unlimited.
Beer, M., Green, S., Armitt, G., Sixsmith, A., Van Bruggen, J., Daniels, R., et
al. (2002, April). Developing communities of interest in a European
internet school. Proceedings of the Networked Learning Conference,
University of Sheffield, UK.
Benner, P. A. (1984). From novice to expert: Excellence and power in
clinical nursing practice. Menlo Park, CA: Addison-Wesley.
Biggs, J. (1999). Teaching for quality learning at university. Buckingham,
UK: Open University Press.
Britain, S., & Liber, O. (1999). A framework for pedagogical evaluation of
virtual learning environments. Bangor, UK: University of Wales Press.
Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the
culture of learning. Educational Researcher, 18(1), 32–42.
Bullen, M. (1998). Participation and critical thinking in online university
distance education. Journal of Distance Education, 13(2), 1–32.
Conrad, D. (2002). Deep into the hearts of learners: Insights into the nature of
online community. Journal of Distance Education, 17(1), 1–14.
Edgar, R. (1995). PC is to Piaget as WWW is to Vygotsky. SIGGRAPH ’95,
Los Angeles, CA. Retrieved January 5, 2005, from www.iconceptual.com/
siggraph.html
Ertmer, P. A., & Newby, T. J. (1993). Behaviourism, cognitivism,
constructivism: Comparing critical features from an instructional design
perspective. Performance Improvement Quarterly, 6(4), 50–72.
Faggiano, E., Roselli, T., & Plantamura, V. L. (2004). Networking technolo-
gies to foster mathematical metacognitive processes. Proceedings of the
206 Lee
4th IEEE International Conference on Advanced Learning Technolo-

gies, California (pp. 410–415). IEEE.
Feenberg, A. (1989). The written world: On the theory and practice of
computer conferencing. In R. Mason & A. Kaye (Eds.), Mindweave:
Communication, computers and distance education (pp. 3–21). Ox-
ford: Pergamon Press.
Garrison, D. R. (2002). Cognitive presence for effective asynchronous
online learning: The role of reflective inquiry, self-direction and
metacognition. Retrieved January 5, 2005, from http://sln.suny.edu/sln/
public/original.nsf/dd93a8da0b7ccce0852567b00054e2b6/755285ff
b5847a4385256c3c006246ea/$FILE/Learning%20Effectiveness
%20paper%20-%20Garrison.doc
Goodyear, P. (2000). Environments for lifelong learning. Ergonomics, archi-
tecture and educational design. In J. M. Spector & T. Anderson (Eds.),
Integrated and holistic perspectives on learning, instruction & tech-
nology: Understanding complexity (pp. 1–18). Dordrecht, The Neth-
erlands: Kluwer Academic Publishers.
Goodyear, P. (2002). Psychological foundations for networked learning. In C.
Steeples & C. Jones (Eds.), Networked learning: Perspectives and
issues (pp. 49–76). London: Springer-Verlag.
Hammond, M. (2000). Communication within on-line forums: The opportuni-
ties, the constraints and the value of a communicative approach. Comput-
ers and Education, 35, 251–262.
Hammond, M. (2002, April). Discourses on collaborative networked learning.
Proceedings of the Third International Conference on Networked
Learning, Sheffield University, UK.
Harasim, L. (1989). On-line education: A new domain. In R. Mason & A.
Kaye (Eds.), Mindweave: Communication, computers and distance
education (pp. 50–62). Oxford: Pergamon.
Hirumi, A. (2002, January / March). The design and sequencing of e-learning
interactions: A grounded approach. International Journal on E-Learn-
ing, 19–27.
Hung, D. (2001). Theories of learning and computer-mediated instructional
technologies. Education Media International, 38(4), 281–284.
Hung, D. W. L., & Chen, D.-T. (2001). Situated cognition, Vygotskian
thought and learning from the communities of practice perspective:
Implications for the design of Web-based e-learning. Education Media

Internation, 38(1), 4–12.
Jarvis, P., Holford, J., & Griffin, C. (1998). The theory and practice of
learning. London: Kogan Page.
Jonassen, D. H. (1994). Thinking technology. Toward a constructivist design
model. Educational Technology, 34(4), 34–37.
Jonassen, D., Mayes, R., & McAleese, R. (1993). A manifesto for a
constructivist approach to technology in higher education. In T. Duffy, D.
Jonassen, & J. Lowyck (Eds.), Designing constructivist learning
environments (pp. 231–247). Heidelberg, Germany: Springer-Verlag.
Kaye, A. (1989). Computer-mediated communication and distance educa-
tion. In R. Mason & A. Kaye (Eds.), Mindweave: Communication,
computers and distance education (pp. 3–21). Oxford: Pergamon
Press.
Klemm, W. R. (1995). Computer conferencing as a cooperative learning
environment. Retrieved January 5, 2005, from http://www.cvm.tamu.edu/
wklemm/cl.html
Klemm, W. R. & Snell, J. R. (1996). Enriching computer-mediated group
learning by coupling constructivism with collaborative learning. Journal
of Instructional Science and Technology, 1(2), 1–11.
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral
participation. Cambridge University Press.
Lefoe, G. (1998). Creating constructivist learning environments on the
Web: The challenge in higher education. Wollongong, Australia:
Ascilite.
Mayadas, F., Bourne, J., & Moore, J. (2003). We have the technology. Sloan-
C View, 2(1), 1–2.
Mayes, T. (2001). Learning technology and learning relationships. In J.
Stephenson (Ed.), Teaching and learning online (pp. 16–26). London:
Kogan Page.
McConnell, D. (2002, April). Negotiation, identity and knowledge in e-
learning communities. Proceedings of the Networked Learning Con-
ference, University of Sheffield, UK.
McDonald, J. (2002). Is “as good as face-to-face” as good as it gets? Journal
of Asynchronous Learning Networks, 6(2), 10–23.
208 Lee
Mentis, M., Ryba, K., & Annan, J. (2001). Creating authentic on-line
communities of professional practice. Paper presented at the Austra-
lian Association for Research in Education Conference, Freemantle.
Retrieved January 5, 2005, from www.aare.edu.au/01pap/men01511.htm
Miao, Y. (2004). Supporting situated learning for virtual communities of
practice: Representation and management of situated knowledge. Pro-
ceedings of the 4 th IEEE International Conference on Advanced
Learning Technologies, California (pp. 490–494). IEEE.
Moore, M. G., & Kearsley, G. (Eds.). (1996). The distance education student.
In Distance education: A systems view. Belmont, CA: Wadsworth
Publishing.
Oliver, R., & McLoughlin, C. (1999). Curriculum and learning resources issues
arising from the use of Web-based course support systems. Interna-
tional Journal of Educational Telecommunications, 5(4), 419–436.
Palloff, R. M., & Pratt, K. (1999). Building learning communities in
cyberspace. San Francisco: Jossey-Bass.
Rourke, L., Anderson, T., Garrison, D. R., & Archer, W. (2001). Assessing
social presence in asynchronous text-based computer conferencing.
Journal of Distance Education, 14(2), 1–15.
Ryan, Y. (2001). The provision of learner support services online. In G. Farrel
(Ed.), The changing faces of virtual education, the commonwealth of
learning (pp. 71–94). Vancouver, Canada: COL.
Salmon, G. (2000). E-moderating: The key to teaching and learning
online. London: Kogan Page.
Schwier, R. A. (1995). Issues in emerging interactive technologies. In G.
Anglin (Ed.), Instructional technology: Past, present, and future (pp.
119–130). Englewood, CO: Libraries Unlimited.
Simpson, O. (2002). Supporting students in online, open and distance
learning (2nd ed.). London: Kogan Page.
Storey, M.-A., Phillips, B., Maczewski, M., & Wang, M. (2002). Evaluating
the usability of Web-based learning tools. Educational Technology and
Society, 5(3), 91–100.
Tam, M. (2000). Constructivism, instructional design and technology: Implica-
tions for transforming distance learning. Educational Technology and
Society, 3(2), 50–60.
Turoff, M. (1995). Designing a virtual classroom. International Conference

on Computer Assisted Instruction, Taiwan. Retrieved January 5, 2005,
from www.njit.edu/old/njIT/Department/CCCC/VC/Papers/Design.html
Vonderwell, S. (2003). An examination of asynchronous communication
experiences and perspectives of students in an online course: a case study.
Internet and Higher Education, 6, 77–90.
Wenger, E. (1998). Communities of practice; Learning, meaning and
identity. Cambridge University Press.
Wenger, E. (2004). Communities of practice: A brief introduction. Re-
trieved January 5, 2005, from www.ewenger.com/theory/
communities_of_practice _intro.htm
210 Beynon & Roe
Chapter X
Enriching Computer
Support for
Constructionism
Meurig Beynon, University of Warwick, UK
Chris Roe, University of Warwick, UK
Abstract
The dominant emphasis in current e-learning practice is instructionist in

character. This is surprising when we consider that the benefits of
constructionism as a learning paradigm are so widely recognised.
Moreover, though the constructionist philosophy can be seen as applying
to activities that are not necessarily computer based (such as bricolage
and concept mapping), its modern application in educational technology
has been closely linked with computer use. In particular, Papert’s work on
mathematical education through Logo programming has both informed
the original concept of constructionism and been a major influence over
subsequent computer-based constructionist developments. This chapter
questions whether — despite these precedents — traditional computer
programming paradigms are well suited for the constructionist educational
Enriching Computer Support for Constructionism 211
agenda. It argues that other approaches to computer model building, such

as those based on spreadsheet principles, are in fact much better aligned
to the objectives of constructionism. Building on this basis, it proposes
that more effective computer support for the constructionist perspective
can potentially be offered by Empirical Modelling (EM). Adopting this
approach demands a reappraisal of the relationship between the formal
and the informal with relevance for education, mathematics, and
computing.
Introduction
The development of e-learning environments has been driven by the needs of

universities, where the lecture is the dominant teaching method. The Internet
can serve as an efficient mode of effecting lecture delivery. In some situations,
the delivery of factual information is entirely appropriate. On the other hand,
educationalists recognise the importance of interaction, and constructionists
(Harel & Papert, 1991) go further to propose principles that facilitate active
learning. In this chapter, we reflect on two complementary experiences: of
developing microworlds as part of an e-museum, and of experimenting with a
novel approach to building computer-based models for educational applica-
tions—that of Empirical Modelling (EM). We draw on these experiences to
raise fundamental questions about how computer support for constructionism
can best be provided. We identify respects in which conventional thinking
about general-purpose computer programming conflicts with the aspirations of
constructionism. This leads us to propose EM as a promising alternative
foundation for constructionist practices. Though we introduce a simple illustra-
tive example to inform the discussion, a full demonstration of the practical
usefulness of EM is outside the scope of this chapter. In any case, without a
broader infrastructure for our principles and tools, a proper empirical evalua-
tion of EM in an educational context would be premature at this stage. And
although there are indications that EM potentially has useful educational
application across a wide range of subjects (see, e.g., McCarty, Beynon, &
Russ, 2005; Roe, 2003), our attention in this chapter is confined to mathemati-
cal education. This is in keeping with Papert’s original motivation for introduc-
ing constructionism, and informs a brief discussion of the deeper foundational
issues that are topical in understanding mathematical concepts.
212 Beynon & Roe
Principles of Constructionism
Throughout its history, educational software has progressed through many

phases, mirroring the development of cognitive theories of learning. Early
instructional software was dominated by programs that reflected a behaviourist
outlook—the inspiration for much “drill and kill” software—where the com-
puter acts as a replacement teacher, simply asking questions and gauging
learning from the pupil’s responses. Over recent years, educational software
has tended to reflect a constructivist approach that acknowledges the active
role played by the learner in the making of meaning.
A more specific form of constructivism began to emerge in the 1960s when a
team, headed by Papert and Feurzeig, was developing the Logo programming
language at MIT. This early work was primarily concerned with programming
and problem solving in the context of mathematical education (see Papert,
Watt, diSessa, & Weir, 1979; Watt, 1979). In particular, it advanced the
radical notion that children need to play with and use mathematical concepts
within a supportive computer-based environment before being introduced to
formal work with those concepts (Papert, 1971, p. 18):
When mathematizing familiar processes is a fluent, natural and enjoyable

activity, then is the time to talk about mathematizing mathematical
structures, as in a good pure course on modern algebra.
These initial ideas led Papert (1980) to a vision for mathematical education that
was subsequently elaborated into a new paradigm for teaching and learning
mathematics—the constructionist approach (Harel & Papert, 1991). Follow-
ing (Roe, Pratt, & Jones, 2005), we shall discuss model building with reference
to six underlying features of constructionist learning distilled from the literature:
1. Quasi Concrete Objects—Turkle and Papert (1991): The computer

allows formal ideas to be accessed in a concrete way, through developing
iconic representation of abstract mathematical ideas that can be manipu-
lated directly by the user.
2. Integrating the Informal and the Formal—diSessa (1988): Incorpo-
rating formal representations of mathematical objects in models in differ-
ent ways may help a child to make connections between the various
formalisations and their informal use.
3. Using Before Knowing—the Power Principle, Papert (1996): In

everyday life, we typically use tools for particular purposes. Through such
use, we learn about the effectiveness of a tool, its limitations, how well it
serves its purpose, and may sometimes gain some insight into how it
works. In schools, mathematics is a subject where, in general, you learn
how to generate the object before you use it. In practice, the task of object
generation proves too difficult, especially when disconnected from pur-
pose. The computer makes it possible to invert the activity of learning
mathematics so that use precedes generation.
4. Dynamic Expression: When Papert proposed the turtle as a tool for
constructing a dynamic notion of angle (and, of course, much else), he
acknowledged that the computer offers a medium which—unlike paper
and pencil—can incorporate dynamic representations of the world. He
suggests that systems that are expressive of dynamic and interactive
aspects of the world are more engaging to learn than static and abstract
formalisms.
5. Building: Constructionism is based on the tenet that encouraging the
building of artefacts is a particularly felicitous way of teaching mathemat-
ics. Pratt (2000) has demonstrated how this approach generalises to
related activities such as mending.
6. Purpose and Utility: The microworld approach can encourage purpose-
ful activity through the building and modification of artefacts. Through such
interaction, the usefulness of the relevant abstractions is appreciated and
their limitations are gradually discriminated.
The e-Muse project illustrates an application of constructionist principles. The

broad aim of this project was to develop an Internet museum. The output of the
project comprised exhibits relating to the ancient Olympic Games, explanatory
material related to the exhibits, and interactive activities to engage visitors.
These resources were developed with both museum and schools environments
in mind.
From the first, the project engaged with two tensions. In developing a virtual
museum that bridged museum and school environments, there was an underly-
ing cultural conflict. Whereas museologists were concerned primarily with
accuracy and appropriate presentation, classroom practitioners’ foremost
concern was promoting interaction and engagement. The second tension
stemmed from the first. The primary interest of the museologists—in common
214 Beynon & Roe
with that of the designers of so-called e-learning environments—was in the

efficient delivery of accurate materials, whereas that of the educationalists—in
common with that of classroom practitioners—was on stimulating learning and
engagement.
In e-Muse, two microworlds were developed. These provided an interactive
experience for e-museum visitors. They were based on the throwing events of
the Olympics and targeted at children of 10 years old and upward. Their
development reflected a constructionist perspective: it made use of the Imagine
Logo programming environment (Kalas & Blaho, 2000), and was guided by
observation and feedback obtained at a local school, where prototypes were
trialled with schoolchildren.
One of the primary findings of the e-Muse project was that the realisation of
constructionist ideals for learning was obstructed to some considerable degree
by the lack of a facility for children to build or modify models themselves (Roe
et al., 2005). Though the Imagine Logo environment offered many features to
assist the developers in empowering the user to interact with models in
imaginative ways, the task of adapting models in response to children’s
perceived and expressed needs was typically too technical to be undertaken by
any participant other than the primary developer. For this reason, the e-Muse
environments afforded less flexibility and openness in interaction than was
ideally envisaged.
The predominant delivery model for e-learning exhibits similar restrictions and
lack of flexibility in interaction, perhaps to an even more marked degree. As
Bannan Ritland, Dabbagh, and Murphy (2002) observe, designers of e-
learning environments typically structure content in a particular sequence for
delivery to the learner. This leads them to remark (p. 12) that
...there are alternative theoretical foundations other than a traditional

instructional system design perspective that can be applied to learning
object systems based on constructivist philosophy of learning. To the best
of our knowledge, a learning object system based in theoretical approaches
steeped in constructivism has not yet been developed.
Of course, it is not self-evident that the level of interaction advocated by a

constructivist philosophy is achievable. Indeed, Ehrmann (2000) has argued
that the idea of attaining interactive courseware that can give full support to
constructionist principles is a mirage. He claims that this is due to the high human
costs needed to achieve appropriate levels of interactivity. We shall argue that

whatever ultimate limits may have to be set on the aspirations of computer
support for constructionism, the adoption of an alternative approach to
computer-based modelling offers the potential for far greater levels of interactivity
in e-learning environments. In the next section, we provide the background and
motivation for this argument by considering the relationship between conven-
tional programming and constructionism, both in its historical context, and in
relation to the findings of the e-Muse project.
Constructionism and Programming
Of particular significance in the constructionist tradition are environments that

support “active learning,” in which learners are actively involved in building their
own public artefacts. The emphasis in active learning is on the mental processes
that occur during the construction of the artefact, not on the quality of the final
product. The situated and public nature of the construction activity is also
identified as important. For instance, in developing his vision for construction-
ism, Papert stresses that the active building of knowledge structures in the head
often happens especially felicitously when it is supported by construction of a
more public sort “in the world” (Papert & Harel, 1991). There are many
reasons why active learning is seen as particularly beneficial: learners can
pursue their particular interests, can see a tangible result with potential
application and relevance, and are motivated to communicate their understand-
ing to others.
The advent of computer technology for learning has opened up new avenues for
developing concrete models in the form of interactive computer based artefacts.
To meet the requirements for (computer-mediated) active learning, it must be
possible for ordinary computer users to construct such artefacts, so that
meaningful learning of a domain can proceed in tandem with the construction
of the interactive artefact. In her study of end user programming, Nardi (1993)
claims that those who are not computer specialists can create personally
meaningful computer models if the programming environment eliminates much
of the accidental computational complexity. By way of example, she cites end
users creating computer-aided design models, Logo programs and spread-
sheet models.
216 Beynon & Roe
In our view, the issues surrounding computer support for active learning have
yet to be adequately addressed. Ever since Papert first developed the Logo
environment, there has been some ambiguity about the relationship between
computer programming and the educational objectives of constructionism. Is
computer programming to be viewed as an activity that—of itself—serves the
educational objectives of the constructionist agenda, as the Logo environment
might suggest? Or is computer programming simply the means to set up
environments for model making using techniques that are not—or at any rate
are not perceived as—computer programming? In practice, the distinction
between “learning about computer programming” and “learning about a domain
independent of computer programming” is not always clearly respected in
computer based environments that support active learning. What is more,
educationalists and computer scientists alike seem relatively insensitive to the
potential implications of adopting different perspectives and approaches to
constructing computer models.
In this chapter, we argue that there are highly significant distinctions to be made
between the different perspectives we can adopt on providing computer
support for active learning. In particular, there is a fundamental conceptual
distinction to be made between using spreadsheet principles and other pro-
gramming paradigms that focus on programs as recipes for performing goal-
directed transformations. Our thesis is that programming paradigms rooted in
the classical view of computation (which embrace the full range of traditional
programming idioms including procedural, object-oriented, and declarative
approaches) are not well suited to providing support for the constructionist
learning agenda. On this basis, we propose an alternative framework that builds
on the principles for spreadsheet engineering identified by Grossman (2002).
The remainder of the chapter is in two principal sections: the first discusses the
relationship between classical computer programming and constructionism; the
second briefly introduces and illustrates the alternative perspective afforded by
Empirical Modelling.
The relationship between computer programming and constructionism is con-
ceptually complex. Papert’s aspiration for the use of Logo is that constructing
a program should be a valuable learning experience in which a pupil becomes
familiar with geometric concepts and with strategies for problem solving and
design (Papert, 1993). There is an implicit assumption that the process of
program construction is well-aligned to useful domain learning and to con-
structionist principles, but there are potentially problematic issues to be
considered:
• Extraneous activity: Much of the learning associated with model build-

ing is computer programming specific: it is concerned with manipulating
programming language commands, procedures, and parameters rather
than with developing knowledge of geometric concepts or abstract
thinking strategies;
• Planning rather than exploration: Classical programming is not con-
ceived as an iterative experimental process: programmers are encouraged
to plan and preconceive their application rather than to develop a model
in an open-ended fashion where its significance can emerge during the
development.
Extraneous activity in computer-supported domain learning is a problem for

which many different remedies have been proposed. Soloway (1993) raises
“the heretical question: Should all students learn to program?” and advocates
the use of domain-specific, scaffolded, computer-aided design environments
as an appropriate substitute for “those pesky semi-colons.” The educational
experts who responded to his question were positive about the importance of
learning to program, and about the valuable—if not essential—contribution it
can make to broader domain learning. The diversity of opinions about how to
teach programming so that it does not obstruct domain learning highlights such
issues as how best to provide programming interfaces for end users; whether
to use graphical front ends; whether to use object-oriented principles or
recursion, linked lists, and trees.
The significance of being able to treat computer programming as an exploratory
activity, rather than a planned activity, is likewise well-recognised. Ben-Ari
(2001) discusses how computer programming, as practised, contains the
element of redesign in response to interaction with the partially developed
programme that is characteristic of bricolage (Levi Strauss, 1968). This is
endorsed by Fred Brooks’s observation (1995) that programmers see their
work as a craft where they wrestle with incompletely understood meaning, and
by proposals for software development based on techniques such as “eXtreme
Programming” (Beck, 2000).
The thesis of this chapter is that a proper appreciation of the problem of
providing computer support for constructionism can only be gained through
looking at a deeper issue than the flavour of programming paradigm, the
interfaces for the end user, or the method of software development. There is a
profound ontological distinction between an artefact that is developed in active
learning and a computer program. To interpret computer support for construc-
218 Beynon & Roe
tionism effectively it is necessary to shift attention from the concept of computer

program that is endorsed by the classical theory of computation, and focus
instead upon the way in which the programmed computer itself serves as a
physical artefact. This is best appreciated by comparing the thought processes
that accompany contemplation of the artefact in active learning with those
associated with developing a computer program.
In active learning, the artefact under development is a source of experience.
Throughout its development, the learner is invited to project possible interpre-
tations and applications on to the artefact as it evolves. The learner asks such
questions as “What can I do with this now?” and “How can this particular kind
of interaction with the artefact now be interpreted?”. Insofar as some reliable
interactions with the artefact are familiar to the learner, it implicitly embodies
knowledge. At the same time, since many of the plausible interactions contem-
plated may be as yet unexplored, the artefact in some respects embodies the
learner’s ignorance. The educational qualities of interaction with the artefact
mirror those exhibited in an informal exposition of a proof. Such an exposition
is mediated by artefacts, so that the reader can be invited to anticipate the next
step, and introduced to the situations in which false inferences can be drawn or
unsuccessful strategies adopted.
By contrast, developing a program is understood (from the perspective of the
classical theory of computation) with reference to assertions of the form “this
is what the program is intended for; these are the kinds of interaction that it
admits; these are the ways in which responses to this interaction are to be
interpreted.” It is, of course, the case that in any complex programming task,
essential knowledge of the domain is developed through experimental activity
involving artefacts (as represented by use cases, UML diagrams [Jacobson et
al., 1992], and prototypes of various kinds). But while this domain knowledge
plays a fundamental role in programming, it is primarily directed at the intended
functionality and interpretation of the program. For this reason, the artefacts
developed in framing requirements serve only for reference purposes once the
program implementation begins. A computer program resembles a formal
proof in that it follows an abstract pattern of steps whose meaning is entirely
contingent upon adhering to a preconceived recipe that is invoked in the
correct—fastidiously crafted—context.
The above discussion suggests that the conventional perspective on computer
programming is unhelpful in understanding how to give computer support to
constructionism. This is not to deny the practical value of computer based
environments that have already been developed for active learning, but to
observe that they ideally demand a conceptual framework quite different from
that offered by classical computer science. With the possible exception of
domains in which learning is primarily concerned with understanding pro-
cesses, it is in general inappropriate to think of a learning artefact as a computer
program. For reasons to be briefly explained and illustrated in the following
sections, we prefer to characterise computer-based artefacts for active learn-
ing as construals.
Our proposal to discard the notion of program in favour of “construal” is in the
first instance significant only as a metalevel shift in perspective. In practice,
spreadsheets already provide examples of such construals. It is also likely that
in asserting that “we need to fundamentally rethink how we introduce program-
ming to students,” “we require new types of programming tools,” and “we need
new programming paradigms,” Resnick and Papert (Soloway, 1993) have in
mind a much broader notion of “program” than the classical view of compu-
tation supports. Nevertheless, making the explicit distinction between pro-
grams and construals liberates a radically different view of what computer
support for constructionism entails, and lays the foundation for a better
understanding with implications for theory and practice. For instance, it can
help to identify more effective principles and tools for building learning
artefacts, and may help to explain practical developments, such as the success
of spreadsheets and the relative lack of popularity of programming as a learning
tool for the non-specialist (cf. Nardi, 1993), and the emergence and subse-
quent disappearance of Logo from the UK National Curriculum (cf. Noss &
Hoyles, 1996).
Constructionism and
Empirical Modelling
Our description of a learning artefact as a construal borrows from the work of

David Gooding, a philosopher of science. Gooding (1990) used the term to
describe the physical artefacts and procedures for interaction, observation, and
interpretation that Faraday developed to embody his understanding of electro-
magnetic phenomena, as it evolved through practical experiment and commu-
nication with other experimental scientists. In that context, experiment has a
significance beyond the popular understanding of the scientific method (as in
Sannella, 1997): “One develops a theory that explains some aspect of reality,
220 Beynon & Roe
and then conducts experiments in order to provide evidence that the theory is
right or demonstrate that it is wrong.”). Although Faraday’s experiments did
eventually underpin Maxwell’s mathematical theory, they initially had a far more
primitive role. For instance, they served to distinguish transient effects from
significant observables, and to relate Faraday’s personal construals of a
phenomenon to those of others who had typically employed different modes of
observation and identified different concepts and terminology. Such experi-
ments were not conducted post-theory to “explain some aspect of reality”, but
rather to establish pre-theory what should be deemed to be an aspect of reality.
A construal is typically much more primitive than a program. It is built in an
open-ended fashion with a situation rather than a specific purpose in mind. The
conventions for interacting with it and interpreting these interactions are quite
informal and fluid. In general, whether a particular interaction has an interpre-
tation can only be appreciated by consulting the immediate experience it offers
and recognising this as corresponding to an experience of the referent. A
possible construal for the electromagnetic phenomenon associated with a wire
coil, such as Faraday himself first developed as a physical artefact, and we
might now realise on a computer, would depict the relationship between the
direction and strength of the electric current and the disposition and density of
the associated magnetic field. A primitive interaction with such a construal
would involve observing the impact of changing the current on the strength of
the magnetic field in both the computer model and its referent. The relationship
between current and field would be perceived as a direct correspondence
between dependencies in the model and its referent. In this context, the
counterpart of a program would be a much more sophisticated construction—
such as a model of an electric motor—that has some autonomous reliable
behaviour that cannot be experienced through being present in just one
situation.
Empirical Modelling (EM) describes the characteristics of a construal (cf. a
spreadsheet) with reference to three key concepts: observables, dependen-
cies, and agency. An observable is a feature of the situation or domain that we
are modelling to which we can attach an identity (cf. a spreadsheet cell). The
main requirement of an observable is that it has a current value or status (cf. the
value of a spreadsheet cell). A dependency is a relationship amongst observables
that expresses how they are indivisibly linked in change (cf. the definition of a
cell). Unlike constraints, which express persistent relationships between values
in a closed world, dependencies express the modeller’s current expectation
about how a change in one variable will affect the value of another in an open-
ended exploratory environment. Observables and dependencies together

determine the current state of an EM model. An agent is an entity in the domain
being modelled that is perceived as capable of initiating state change. In
developing an EM model, our perspective on agency within the domain evolves
with our construal.
Developing a construal in EM is a voyage of discovery, a creative activity that
is quite unlike conventional programming, where the emphasis is on represent-
ing well-understood behaviours. An EM model is empirically established
(informed by experience and subject to modification in the light of future
experience) and experimentally mediated (our experience with it guides its
evolution). A construal must be testable beyond the limits of the expected range
of interactions with it. In specifying a conventional program, the modeller has
to preconceive its behaviour, thereby restricting the exploratory interactions
that can be undertaken. In contrast, EM model construction privileges experi-
mental interaction. Interactions can take account of the changing real world
situation, can probe unknown aspects of a referent, and may even be nonsen-
sical in the world.
The potential implications of adopting an EM perspective on computer support
for constructionism will be briefly illustrated using a simple example. A beam
detector for the unit circle is a set of points that intercepts all lines crossing that
circle. Eppstein (1998) describes a beam detector constructed by taking a
regular hexagon ABCDEF that circumscribes the unit circle, joining the points
ABDE using a Steiner tree, and dropping line segments from the two vertices
C and F on to the nearest side of the quadrilateral ABDE. The length of such
a detector is 2/√3 + 4 = 5.1547.... Eppstein observes that this is nonoptimal and
conjectures that nonregular hexagons can be used to reduce this length.
A teacher wishing to exploit Eppstein’s beam detector as an aid to active
learning might consider many issues:
• Motivating the search for a detector of optimal length: To this end,

Ian Stewart (2004) devises a detective story, recasting the problem as
digging trenches of minimal size that are guaranteed to detect a drainage
pipe in the neighbourhood of a statue. To exploit this interpretation, it
might be helpful to construct a virtual reality model;
• Situating the problem within computational geometry: Eppstein’s
construction is an application for Steiner trees. This motivates making a
model that incorporates and builds on a method of Steiner tree construc-
tion. For further investigation, this model could be extended to display
222 Beynon & Roe
critical lines that pass through just one of the five straight line segments of
the given beam detector;
• Using the beam detector to illustrate school geometry: Modelling the
detector is an exercise in geometric construction that helps students to
learn about tangency, trigonometric relationships, perpendicular lines,
and so forth;
• Using the detector as a case study for modelling tools: Students could
make a geometric model of the detector using a special purpose tool such
as Cabri Geometry, or study it as an optimisation problem using a
spreadsheet.
Issues of presentation are also relevant. The teacher might wish to present
Eppstein’s construction of the beam detector using an interactive whiteboard,
to distribute instances of the construction to the pupils for them to experiment
and compete to find the best solution, and to monitor and to display the details
of the detector of smallest total length encountered so far concurrently in real
time (e.g., as might be done in a sporting event).
If we regard these potential applications as specifications for independent
programming exercises to be addressed, there is a prohibitive overhead. Model
building directed at capturing the different functional requirements involved in
developing a virtual reality environment, setting up a spreadsheet, or emulating
CABRI, cannot exploit abstraction above the level of a general purpose
programming language. By building a construal, on the other hand, it is possible
to build an integrated family of models adapted for each of these different
purposes.
Screenshots and extracts from variants of an EM model of Eppstein’s beam
detector are shown in Figures 1 and 2. A full account of the principles behind
the construction of the model and its variants is beyond the scope of this
chapter. The details of the models can be inspected and exercised more closely
by accessing the beamdetectorRoe2004 directory of the EM repository at
http://empublic.dcs.warwick.ac.uk/projects/. Other models from the reposi-
tory illustrate that the features needed to make the extensions of the Beam
Detector model envisaged above are broadly within the scope of the current
EM tools. The following brief discussion will highlight some of the most salient
points about the development of the Beam Detector models.
The original source for the Beam Detector model was developed by the first
author at the suggestion of a colleague who was studying beam detection as an
abstract optimisation problem in computational geometry. The model was
Figure 1. The original Beam Detector model and some of its script
definitions
constructed over a period of 2 or 3 days, and involved some 3 or 4 hours

development. The listings in Figure 1 illustrate the kind of activity that was
involved—the creation of a script of definitions to record the key observables
(such as points, lines, and labels) and dependencies (such as relationships of
incidence and perpendicularity) in Eppstein’s construction. The script creation
was an incremental process, so that the definitions in Listing 1(a) were devised
first, and those in Listing 1(b) were developed subsequently (see
beamdetectorRoe2004 for all the subscripts involved in building up the entire
model stage by stage, from which the listings in Figure 1 have been extracted).
Figure 1 illustrates some general characteristics of EM. The script of definitions
for the model evolves in conjunction with the visual artefact as the modeller’s
224 Beynon & Roe
understanding is clarified. At any stage of the development, there is a script and

an associated visualisation that records such observables and dependencies as
the model builder has so far explored. Some experimental interaction, guided
by the visualisation, is typically involved in each step of the incremental
construction. In introducing the definitions in Listing 1(b), for instance, some
experimentation was used to establish and confirm how the principal value of
the angle denoted by the acos function was being selected. The overall
character of EM in this context is consonant with the way in which Edwards and
Hansom (1989) identify modelling as an iterative process comprising under-
standing the particular phenomena to be modelled, identifying the key variables
and explicitly defining the relationships amongst the variables.
As Figure 1 also illustrates, a script of definitions serves as both a description
of the model in its current state of development, and a record of the interaction
that has led to this current state. This dual role lends to the script the rather loose
and messy quality that is characteristic of much of learning activity. The
evolution of the models reflects different stages in the understanding, for which
different configurations and visualisations of observables are generally appro-
priate. In Listing 1(b), for instance, the observable SE in the model, which refers
to the geometric location of the point labelled “L” in the diagram, has been
redefined. This redefinition has been made in order to ensure that whenever the
points N and S are relocated, the angle NSL is 120 o, as is appropriate in a
Steiner Tree.
The attentive reader will note that in Figure 1(c), which depicts one of the
possible states of the basic Beam Detector model in (EMRepository:
beamdetectorRoe2004), the angle NSL is no longer 120o. This is because, at
a subsequent stage in the model building, the definition of SE was restored to
that in Listing 1(a). The explanation for this is that—without loss of generality—
any location of the points N and S leads to a beam detector configuration that
is congruent to one in which the line NS is vertical. On this basis, a full
exploration of the design space of optimal beam detectors on Eppstein’s
pattern can be carried out without needing to reorient the line NS at any stage.
With this fact in mind, there is a useful educational purpose in neither
constraining angle NSL as in Listing 1(b), nor preventing NS from taking up a
nonvertical orientation. Allowing a student to displace N and S arbitrarily then
supplies experimental evidence that deviating from a Steiner Tree can only
make the length of the beam detector suboptimal.
The above discussion illustrates the complexity of the issues involved in building
models to support learning, and in particular the subtle role played by placing
Figure 2. An extension of the basic EM model of a Beam Detector
constraints on interaction. The virtue of the basic Beam Detector model as an

interactive artefact for experiential learning is that, unlike a conventional
program, it is not developed with closed learning objectives and ease of use in
mind. However, we may extend the basic model with a view to making it less
open-ended and more self-explanatory, so as to give greater prominence to
specific targets for the learning. Figure 2 is an extension of the Beam Detector
model, carried out by the second author at a much later date, that incorporates
features to assist the learner. In this model, points and lines can be manipulated
dynamically, rather than merely relocated in a discrete fashion, so that experi-
ence of a different quality is brought to bear in trying to minimise the length of
the beam detector. The geometric components of the beam detector have been
highlighted and a representation of the beam itself has also been added, as have
spreadsheets and textboxes to display significant values. This process of
extension has precisely the same character as the creation of the original model,
and exploits reuse of other EM models.
226 Beynon & Roe
Perspectives on Constructionism
In the previous section, we have compared and contrasted the support for
constructionist principles afforded by EM and by conventional programming.
Our illustrative example, the Beam Detector model, like the e-Muse microworlds,
relates to implementing a constructionist approach to mathematics education.
In this section, we review our findings from a broader perspective, briefly
discussing how they relate to topical perceptions of education, mathematics,
and computing. In this way, it becomes apparent that some of the specific
tensions between learner, teacher, and developer perspectives alluded to in the
previous section are symptomatic of more profound conflicts in thinking about
mathematics, education, and computing, both interdisciplinary and
intradisciplinary. When trying to bring these disciplines together, these conflicts
are not merely unresolved—they are to a large extent unacknowledged.
From an educational perspective, model-building by computer is an activity
that superficially appears best aligned to teaching mathematics, or a mathemati-
cal science. There are a number of plausible reasons for this. Computer support
for constructionism has its historical roots in mathematics education. Program-
ming computers is perceived as primarily a logical exercise in framing se-
quences of action that, like formal mathematics, requires great precision and
abstract thought. The kind of model-building with computers that has most
educational credibility is model-building that is based on mathematical theory,
as when Newton’s Laws of Motion are implemented in an e-Muse microworld.
Orthodox computer science thinking endorses this attitude to computer use in
education only partially. Having regard to the still unresolved problems of the
“software crisis,” building software from a theory is perceived by many as the
only way forward for computing (see, e.g., Turski & Maibaum, 1987). In this
context, the issue of which computer programming techniques are to be
commended, and which deprecated, is a matter of great controversy. It is
widely recognised that what is actually involved in instructing the computer by
way of programming is far less significant than how these instructions are linked
to the key observables of the domain in which the program operates.
The difference in viewpoint between educationalist and computer scientist can
call the educational value of computer supported constructionism into question.
For instance, in rule-based programming in the context of a microworld
(Goldstein, Noss, Kalas, & Pratt, 2001), the educationalist sees value in
engaging a bright pupil in discussion of whether a particular rule should be
attached to one object or another. The computer scientist, by contrast,

recognises the kind of uncomfortable pragmatic decision that is typically
encountered in thinking about applying programming paradigms; decisions for
which the lack of principled grounds for judgement underline the very discon-
nection of program from domain understanding that computer science seeks to
avoid.
In fact, the formal computer scientist’s dream of building software from theories
is far from being realised in practical computing. On the contrary, as critics such
as Brian Cantwell Smith (2002) have argued, theoretical computer science is
most ill suited to accounting for contemporary computing practice. What is
even more discomforting for computer science as it is presently understood is
that—whilst much practice remains unconvincing and incoherent—some as-
pects of practice deliver results unanticipated and unexplained by computa-
tional theory. In particular, classical thinking about computation has little
relevance for one of the most widely used and powerful techniques for
computer based model building—modelling with spreadsheets. It is indicative
of how far practical experience has outstripped theoretical explanation in
computing that Baker and Sugden (2003) conclude their extensive review of
the applications of spreadsheets in education by observing: “There is no longer
a need to question the potential for spreadsheets to enhance the quality and
experience of learning that is offered to students.”
It is against the background of such highly confused and contradictory visions
for making sense of the relationships between mathematics, computing, and
education that EM has been conceived. In EM, the aspiration is to develop
principles and tools that can support computer-based model-building that is
intimately connected with domain learning. The precedents for EM are drawn
not from traditional computer programming or formal mathematics, but from
other disciplines where practical activities have a well-developed role, such as
laboratory sciences, engineering design, the humanities, and the arts. In these
domains, practice also takes mature forms, more difficult to formalise than
mathematical model-building, but established on far sounder conceptual foun-
dations than computer programming. Consider for instance, the “scientific
method,” architectural design, and musical analysis.
A key observation is that although the association of mathematics education
with computer programming and with constructionism is very natural, it is also
potentially misleading. Because both mathematics and conventional computer
programming operate with precise and abstract concepts, traditional computer
programming can offer good support for mathematical model building in some
228 Beynon & Roe
respects. In the Beam Detector model, many of the functional relationships that
feature in definitions use simple mathematical operators whose implementation
requires relatively straightforward procedural code. However, constructionism
is not essentially about precise and abstract concepts; on the contrary, it is
motivated by the desire to engage with pre-articulate experience and tacit
knowledge that is made accessible only through exposure to situations. Whilst
the educator is able to envisage imaginative ways of introducing mathematics
into the world of experience (cf. the e-Muse microworlds), classical computer
programming—with its roots in logic and abstract computation—is a reluctant
fellow traveller.
In keeping with the features of constructionism identified earlier in the chapter,
EM typically entails blending the formal and informal. As the discussion of the
Beam Detector has shown, EM can support this integration of the formal and
the informal without in any way compromising its own integrity. In this respect,
it resembles the “scientific method,” which is fundamentally concerned with
interaction in the world, yet (in the context of the school science laboratory, if
not necessarily in its more authentic setting of the research laboratory) is
typically exercised in conjunction with abstract theoretical understanding. EM
is also distinguished from mathematics and from computer programming as they
are conventionally conceived, in that its characteristic movement is from the
informal to the formal (cf. Beynon, Rungrattanaubol, & Sinclair, 2000), rather
than from the formal to the informal.
The adoption of an EM perspective on computer based model building involves
a switch of priorities where pre- and post-theory understanding is concerned.
In the context of mathematics education, this is consistent with the “revaluation
of the concrete” to which Turkle and Papert (1991) refer in their discussion of
constructionist practices. Such a shift in perspective also has a philosophical
aspect concerned with whether we take a Platonistic or intuitionist view of the
foundations of mathematics (Goodman, 1994). Where the Platonist is merely
setting formal ideas in the context of concrete experiences in order to make
them more accessible (cf. the discussion of characteristic features of construc-
tionism earlier in the chapter), the intuitionist regards their very meaningfulness
as contingent at some level upon experience. Of the “two versions of
constructivism” in the foundations of mathematics alluded to by Goodman
(1994), EM seems best aligned with what Goodman describes as “Among
philosophers the most influential contemporary version of constructivism ... the
intuitionism of Michael Dummett (1977).” Dummett’s intuitionism follows “an
essentially Wittgensteinian philosophy of language: to understand mathematics
is primarily to understand mathematical speech, the meaning of which must be

constituted by its use.” More specifically, Dummett advocates a version of
phenomenalism with respect to which “The phenomenalist ... must interpret the
sentence ‘the book is on the table’ by explaining what sense experiences would
justify the assertion of the sentence.” Also relevant in this context is Goodman’s
observation (1994) that, according to the mathematical empiricism of Lakatos,
“mathematics is expounded in an order almost the reverse of that in which it is
discovered.”
Conclusion
This chapter has argued that established thinking about the nature of computer
programming and its relation to formal mathematics is obstructive where
enhancing computer support for constructionism is concerned. In our view, the
use of EM to build construals can better approach the ideal of integrating the
roles of learner, teacher, and developer to which constructionism aspires. To
fully understand the prospects and implications for EM in this respect requires
a more mature and coherent understanding of the relationship between math-
ematics, education, and computing than we have at present.
Shifting the emphasis away from mathematical model building based on pre-
existing theory echoes the philosophy of engineering developed by Vincenti
(1990). Vincenti characterises engineering as a species distinct from applied
science, where there is a role for blind variation—interaction “without
complete or adequate guidance”—potentially leading to discovery. When
seeking to support personal engagement and creativity in learning, the motiva-
tion for a perspective of model-building in which there is no preconceived and
fixed framework for interpretation is clear. Modelling activity that enables us to
manage cognitive conflict and construct new meanings to resolve such tensions
is an essential foundation of constructionist learning. It will not prove easy to
gain full acceptance for such an approach to modelling, as it superficially
appears to encourage just such practices—experiment without abstract specifi-
cation, exploration without preconception—as are deprecated in conventional
software development. Helpful precedents are to be found in existing modelling
tools that exploit dependency, such as spreadsheets and engineering design
packages. Effective model building to support learning demands something
conceptually much more radical than merely adding dependency to the arsenal
230 Beynon & Roe
of conventional programming techniques, however—in this connection, EM

principles are vital in helping to discriminate between emerging understanding
and incoherence in interaction.
Where e-learning is concerned, the general application of constructionist
principles will require a model-building approach that can be adapted to a much
broader range of disciplines. The range of topics addressed in the EM
repository already indicates that EM has much wider potential application than
has been illustrated in this chapter. Some preliminary thinking about how EM
might be applied in modelling for the humanities is further described in
McCarty, Beynon, and Russ (2005).
An interesting development in the use of the Web for learning is that adopted
by the WebLabs project (WebLabs). The WebLabs portal has been designed
to encourage children to share their projects, written in ToonTalk (Kahn,
1996), with other children both local and remote. Such sharing involves posting
a project onto the Web site, commenting directly on other peoples’ projects,
running projects directly on the Web, and downloading them to allow repro-
gramming in ToonTalk.
The WebLabs project clearly highlights the enormous potential for collabora-
tive e learning within a constructionist framework, but the arguments advanced
in this chapter suggest that it will be exceptionally difficult to deliver to this
potential with the chosen programming paradigm. In the longer term, we believe
that EM will prove far more effective at meeting the challenge of implementing
the kind of interaction envisaged in the WebLabs project. For this purpose, we
would favour interaction through multiuser—potentially concurrent—redefini-
tion in scripts, such as has been illustrated in distributed EM models, and is
commended for collaborative Web-based modelling in Cartwright, Adzhiev,
Pasko, Goto, and Kunii (2005).
Acknowledgments
We are indebted to Dave Pratt and Ian Jones for their reflections on the e-Muse
project and to three anonymous referees for their helpful comments.
References
Baker, J., & Sugden, S. (2003). Spreadsheets in education: The first 25 years.
Spreadsheets in Education e-Journal, 1(1), 18–43.
Bannan-Ritland, B., Dabbagh, N., & Murphy, K. (2002). Learning object
systems as constructivist learning environments: Related assumptions,
theories and applications. In D. A.Wiley (Ed.), The instructional use of
learning objects: Online version. Retrieved November 14, 2005, from
http://reusability.org/read/chapters/bannan-ritland.doc
Beck, K. (2000). eXtreme programming explained. Boston: Addison Wesley.
Ben-Ari, M. (2001). Constructivism in computer science education. Journal
of Computers in Mathematics and Science Teaching, 20(1), 45–73.
Beynon, W. M. (1997). Empirical modelling for educational technology.
Proceedings of Cognitive Technology 1997 (pp. 54–68).
Beynon, W. M., Rungrattanaubol, J., & Sinclair, J. E. (2000). Formal
specification from an observation-oriented perspective. Journal of Uni-
versal Computer Science, 6(4), 407–421.
Brooks Jr., F. P. (1995). The mythical man-month: Essays on software
engineering (anniversary edition). Boston: Addison-Wesley.
CABRI geometry. www.cabri.com/web/nsite/html/home.html
Cantwell-Smith, B. (2002). The foundations of computing. In M. Scheutz
(Ed.), Computationalism: New directions (pp. 25–58). Cambridge,
MA: MIT Press.
Cartwright, R., Adzhiev, V., Pasko, A., Goto, Y., Kunii, T. L. (2005). Web-
based shape modelling with HyperFun. IEEE Computer Graphics and
Applications, 25(2), 60–69.
diSessa, A. (1988). Knowledge in pieces. In G. Forman & P. Pufall (Eds.),
Constructivism in the computer age (pp. 49–70). Hillsdale, NJ:
Lawrence Erlbaum.
Dummett, M. E. (1977). Elements of intuitionism. Oxford: Clarendon Press.
Edwards, D., & Hansom, M. (1989). Guide to mathematical modelling.
Boca Raton, FL: CRC Press.
Ehrmann, S. (2000). Technology & revolution in education: Ending the cycle
of failure. Liberal Education, Fall, 40–49.
232 Beynon & Roe
Empirical Modelling: Model Repository. http://empublic.dcs.warwick.ac.uk/

projects
Empirical Modelling Web site. www.dcs.warwick.ac.uk/modelling
e-Muse Project: e-learning for museum and schools environments. EC e-
learning initiative: 2002-4084/001-001 EDU-ELEARN. http://emuse.cti.gr
Eppstein, D. (1998). The geometry junkyard. www.ics.uci.edu/~eppstein/
junkyard/beam
Goldstein, R., Noss, R., Kalas, I., & Pratt, D. (2001). Building rules.
Proceedings of the 4th International Conference of Cognitive Tech-
nology CT2001 (pp. 267–281).
Gooding, D. (1990). Experiment and the making of meaning. Dordrecht:
Kluwer Academic Publishers.
Goodman, N. D. (1994). Some current positions in the philosophy of math-
ematics. In I. Grattan-Guinness (Ed.), A companion encyclopedia of the
history and philosophy of the mathematical sciences (pp. 680–686).
London: Routledge.
Grossman, T. A. (2002). Spreadsheet engineering: A research framework.
Proceedings EUSPRIG 2002 (pp. 23–34).
Harel, I., & Papert, S. (Eds.). (1991). Constructionism. Norwood, NJ:
Ablex.
Jacobson, I., Christeron, M., Jonson, P., & Overgaard, G. (1992). Object-
oriented software engineering: A use-case driven approach. Boston:
Addison-Wesley.
Kalas, I., & Blaho, A. (2000). Imagine … New generation of Logo: Program-
mable pictures. Proceedings of WCC2000 (pp. 427–430).
Kahn, K. (1996). ToonTalk™—An animated programming environment for
children. Journal of Visual Languages and Computing, 7, 197–217.
Lakatos, I. (1976). Proofs and refutations. In J. Worrall & E. Zahar (Eds.),
The logic of mathematical discovery. Cambridge University Press.
Levi-Strauss, C. (1968). The savage mind. Chicago: University of Chicago
Press.
McCarty, W., Beynon, W. M., & Russ, S. B. (2005, June). Human computing:
Modelling with meaning. ACH/ALLC Conference Abstracts, Victoria,
BC, Canada (pp. 138–144).
Nardi, B. (1993). A small matter of programming: Perspectives on end

user computing. Cambridge, MA: MIT Press.
Noss, R., & Hoyles, C. (1996). Windows on mathematical meanings:
Learning cultures and computers. Dordrecht, The Netherlands: Kluwer.
Papert, S. (1971, July). Teaching children to be mathematicians vs. teaching
about mathematics. MIT/AI Memo No 249, Logo memo No 4.
Papert, S. (1980). Mindstorms: Children, computers and powerful ideas.
New York: Basic Books.
Papert, S. (1993). The children’s machine. New York: Basic Books.
Papert, S. (1996). An exploration in the space of mathematics educations.
International Journal of Computers for Mathematical Learning,
1(1), 95–123.
Papert, S., & Harel, I. (1991). Situating constructionism. In I. Harel & S.
Papert (Eds.), Constructionism: Research reports and essays (pp. 1–
11). Norwood, NJ: Ablex.
Papert, S., Watt, D., diSessa, A., & Weir, S. (1979). Final report of the
Brookline Logo Project: Parts 1 and 11 (Logo Memos Nos. 53 and
54). Cambridge, MA: MIT Artificial Intelligence Laboratory.
Pratt, D. (2000), Making sense of the total of two dice. Journal for Research
in Mathematics Education, 31(5), 602–625.
Roe, C. (2003). Computers for learning: An empirical modelling perspec-
tive. Unpublished PhD thesis, Department of Computer Science, Univer-
sity of Warwick, Warwick, UK.
Roe, C., Pratt, D., & Jones, I. (2005, February). Putting the learning back into
e-learning. Proceedings of CERME4, Sant Felix de Guíxols,Spain.
Sannella, D. (1997). What does the future hold for theoretical computer
science? Proceedings of the 7th International Conference on Theory
and Practice of Software Development (TAPSOFT’97), 1214 (pp.
15–19).
Soloway, E. (1993). Should we teach students to program? Log On Educa-
tion, 36(1), 21–24.
Stewart, I. (2004). The great drain robbery. In Math hysteria—Fun and
games with mathematics. Oxford University Press.
Turkle, S., & Papert, S. (1991). Epistemological pluralism: Styles and voices
within the computer culture. In I. Harel & S. Papert (Eds.), Construc-
tionism (pp. 161–191). Norwood, NJ: Ablex.
234 Beynon & Roe
Turski, W. M., & Maibaum, T. S. E. (1987). The specification of computer

programs. Wokington, UK: Addison-Wesley.
Vincenti, W. (1990). What engineers know and how they know it: Analyti-
cal studies from aeronautical history. Baltimore: Johns Hopkins Uni-
versity Press.
Watt, D. (1979). Final report of the Brookline Logo Project Part III: Profiles
of individual student’s work. Logo Memo No. 54, MIT, 4.10–4.17.
WebLabs Project Web site. www.weblabs.eu.com
Section IV
Individual Differences
and Individually
Based Systems
236 Triantis & Pintelas
Chapter XI
An Architecture
for Developing
Multiagent Educational
Applications
for the Web
Tasos Triantis, University of Patras, Greece
Panayiotis Pintelas, University of Patras, Greece
Abstract
New applications in training and education are emerging daily trying to

meet the requirements of distance learners. Network-based or World
Wide Web (WWW)-based intelligent tutoring systems (ITSs) are expected
to meet most of these requirements. In this context, software agents seem
to be a promising distributed software technology that can be used to
implement WWW ITSs. In this chapter, we present a multiagent approach
for constructing an educational application for distance learning. The
proposed architecture exploits the assumption that each teaching subject
Multiagent Educational Applications 237
can be regarded as the synthesis of elementary pieces of knowledge, each

of which can be presented by an independent expert. Moreover, in order
to better support individualized learning, a mobile agent is placed at the
disposal of each distance learner.
Introduction
The increasing Internet penetration rate and the potential of World Wide Web
(WWW) could be used as an enabling technological framework for the
development of courseware applications, which could meet the requirements
of distance learners. Currently, popular Web-based courseware applications
provide location independent learning; however, they exploit disadvantages
such as (a) nonrenewable educational material without the interfering of the
trainer, (b) nonadaptable to the different levels of trainees’ needs, and (c)
nonuse of existing information on the WWW. Most of these shortcomings are
not new; they had already been encountered again in the past and led to the
development of intelligent tutoring systems (ITSs) (Sherman, Singley, &
Fairweather, 1999). On the contrary, network-based or WWW-based ITSs
are expected to meet most of the requirements of the distance learners, allowing
the emulation of a human tutor in the sense that an “intelligent” tutor can know
what to teach, how to teach it, and are able to find out certain teaching-related
information about the learner being taught. Most of the network-based ITSs
provide server-based access to the courseware. Such an approach, although
having benefits in terms of development, maintenance, and access control, is
lacking in flexibility and scalability.
In the light of these, an approach based on agent technology seems to be a
promising distributed software technology that can be used to implement
flexible, truly distributed WWW ITSs. In this chapter, an agent is considered
as a resource-bounded autonomous software entity, which is capable of
collaborating with peers and of exhibiting reactive and proactive behaviour
(Jennings & Wooldridge, 1998). Then, a multiagent system (MAS) is consid-
ered as a loosely coupled network of communicating and cooperating agents
possibly situated on distributed machines, which provide complementary
services (Sycara, 1998). In brief, the main features of the proposed architec-
ture are the possible distribution of teaching knowledge among different
network nodes and the support of end-user mobility. The proposed architec-
ture uses a network of communicating agents and exploits the assumption that
each teaching subject can be regarded as the synthesis of elementary pieces (or
aspects) of knowledge, each of which can be presented by an independent
expert. Moreover, in order to better support individualized learning, a mobile
agent is placed at the disposal of each distance learner.
A brief survey of WWW ITS architectures and agent-based systems on the
Web is presented in section 3. Then, in section 4, this chapter proposes a new
software architecture for distributed courseware applications, which is based
on the multiagent paradigm. Two applications of the proposed architecture are
described in section 5, while some implementation issues are described in
section 6. The advantages of the proposed architecture as well as future work
are discussed in the last section of the chapter.
Issues Related to
Distance-Learning Courseware
When designing a courseware to be used by distance or network-based

learners, designers should take into account issues related to technology and to
the special learning situations.
Technology-related issues include the following:
• Availability of training service: In contrast to stand-alone learners,

network-based learners may suffer from network unavailability or break-
down.
• Robustness: The software should guarantee a minimum level of educa-
tional services available under all conditions.
• Quality of service (QoS): This issue is, in fact, a complex one and can be
analyzed in constituent issues, such as
• Accessibility, which may depend on whether the learner logs via a

LAN or WAN (Internet), or uses a dial-up connection to the
educational service provider. Different educational software designs
need to be developed for different platforms of learner access.
• Performance: This may prove of paramount importance if large files of

dynamic data have to be transferred across the network. Again, modular-
ity of content and stream-based transmission may improve the situation.
• Response time: Learners are aware that they are using interactive
software, so the latter should behave as one. Research has shown that
once a learner starts a 1-hour lesson, there should be a probability higher
than 95% that he/she eventually gets through the lesson (Rindos, Vouk,
Woolet, Hines, & Lester, 1995).
Issues that stem from the special nature of network-based learning include the
following:
• Management of learner profiles: In order to achieve personalized

tutoring, the educational software has to maintain learner profiles.
• Communication and collaboration: Communication initiated by the
service provider towards the distance learners may increase their confi-
dence to the process. Collaboration may be required between a group of
learners, between learners and tutors or even involve the participation of
experts (André, Rist, van Mulken, Klesen, & Baldes, 2000).
• Knowledge management: Knowledge management includes (a) access
to distributed information databases over WANs, (b) updating the
information that is presented to the trainee, (c) filtering a large amount of
information according to current teaching procedure’s needs, and (d)
dynamically composing a teaching material for each learner through
interaction with him/her (Andriessen & Sandberg, 1999).
• Mobility: Contemporary learners tend to use a variety of network access
devices and locations. Educational software should be able to adapt the
content presentation depending on the access point and device used.
• Instructional strategy: Because learners differ in many respects, an
educational software system should track these differences and adopt the
instructional strategy. Currently in the literature exist four kinds of instruc-
tional strategies (Clark, 1998): (a) Receptive, in which the learner has the
minimum control of his/her own learning; (b) Directive, in which the learner
responds to several instructional events, in order the system to offer a
personalized and immediate feedback; (c) Guided discovery, in which the
system only depends on the answers given by the learner to accomplish
its tutoring actions; and (d) Exploratory, in which, complete freedom is

given to the learner to navigate in the instructional system.
In addition to these, there exist a few more issues related to the educational
content:
• Interactive and dynamic content: Distance learners may easily be

distracted from the process by events that take place in the environment.
That is why the educational software should engage the learner in a highly
interactive learning process.
• Property rights: Since everything distributed via the Internet can be
publicly accessed, educational service providers have to ensure that (a)
they have the right to distribute educational content and (b) only subscrib-
ers to the service access the content.
• Certification: Subscribers to educational services need to be assured of
the high quality of the educational content and the training process.
State of the Art
New applications in training and education are emerging daily trying to meet the
requirements of lifelong or distance learners, such as (a) constant updating of
knowledge, (b) accessing distributed information databases over the network,
(c) managing (filtering) huge amount of information, (d) providing anytime and
anywhere learning, (e) providing learning to people of different ages, (f) taking
into account learners’ needs, and (g) exploiting adaptation to different didactic
strategies. Existing applications come under four categories:
Network-Based Tutoring Systems
A large number of high-quality network-based tutoring systems have been

developed recently. Most of the systems exploit the capabilities of a traditional
computer-assisted training (CAT) system and support the management of a
virtual classroom, the ability of anytime and anywhere learning, an (typically)
unlimited number of learners, and the trainer with a set of tools for easily
developing, deploying, and preserving courses. Such systems usually run on a

Web server and can be accessed via an Internet browser and make use of
mature (but sometimes expensive) network technologies. Examples of the
previously described systems are the Course Management System (former
eCourse) by eCollege and the Learning Management System (former Learning
Space) by Lotus. The main disadvantage of the previous systems is that they
do not support multiple instructional strategies and they are not adaptive to user
needs.
Network or WWW-Based Intelligent Tutoring Systems

(ITSs)
As stand-alone ITSs do, network-based ITSs can be used to act as the

learner’s private tutor, while the human trainer or tutor is then free to focus on
more complex and individualised learner needs. This requires the representa-
tion of a domain expert’s knowledge (called the Domain Module), an
instructor’s knowledge (called the Instructional or Tutoring Module), the
learner that is being taught (called the Learner Module) and the way of learner-
computer interaction (Interface Module). Figure 1 displays two popular
WWW ITS architectures.
In the first one, the learner uses an Internet browser with a downloadable Java
applet or pure HTML script (representing the Interface module of the ITS) to
interact with the ITS whose Domain, Student, and Tutor modules reside on a
Web server. An example of this architecture is the AlgeBrain equation-solving
tutor (Sherman et al., 1999), which provides an environment for practicing
Figure 1. Web-enabled ITS architectures
W e b S e rv e r
D a ta
D o m a in
S tu d e n t M o d e l
M o d u le
W eb T u to r i n g D o m a in
S tu d e n t M o d e l
S e rv e r M o d u le M o d u le
T u to r i n g
W eb M o d u le
UI = Java
B ro w s e r
A p p le t /
H M T L -C G I
In t e r fa c e M o d u le
W e b B ro w s e r
algebraic skills learned in a separate instructional setting. In the second

architecture, most of the modules of the ITS reside on a downloadable Java
applet and only some data needed for student modelling reside on the Web
server. Representative examples of the previous architecture are ELM-ART
(Weber & Brusilovsky, 2001) and CITS (Abdel Razek, Frasson, & Kaltenbach,
2003).
Multiagent-Based ITSs
Agent-based courseware, as an evolution of the traditional ITSs, is built upon

previous research on intelligent tutoring systems. Furthermore, it has to deal
with many of the same concerns that agents in general must address, such us
managing complexity, exhibiting robust behaviour in rich, unpredictable envi-
ronments, coordinating their behaviour with that of other agents, and managing
their own behaviour in a coherent fashion, arbitrating between alternative
actions and responding to a multitude of environmental stimuli.
In respect to the issues discussed in section 2, an MAS approach in building
educational systems ensures the following:
• Quality of service: agents are inherently distributed in nature, and as a

result MAS systems exploit (a) availability of the tutoring process to
learners that are LAN, WAN, or even mobile users, (b) high response
time, and (c) adaptability to network failure.
• Communication and collaboration: agents exchange messages only
when necessary. Such communication requires low network bandwidth
and provides the ability of supporting flexible collaboration among learn-
ers through exchanging messages.
• Knowledge management: distributed and mobile agents provide access
to distributed information databases over WANs, in order to update
information that is presented to the trainee and agents may embody
artificial intelligence techniques in order to filter that information according
to current teaching procedure’s needs.
• Availability, Robustness: the MAS approach ensures a high level of
workload scalability by sharing the workload through the pool of agents
(variable in number) distributed on different machines in order to optimise
the reply time and the workload of all servers.
Figure 2. MAS ITS functional architecture
Web Interface
Delivery Agent
Student Agent Domain Agent Tutor Agent
Domain of Application
From a software engineering point of view, MAS approach enhances perfor-

mance along the dimensions of computational efficiency, reliability, extensibil-
ity, robustness, maintainability, responsiveness, flexibility, and reuse. Cur-
rently, most of the multiagent courseware systems implement (some or all) the
modules of a traditional ITS architecture (Figure 2) as separate agents and the
whole tutoring system is delivered to the learner through a Web-based user
interface. Particularly, the Delivery Agent is responsible for monitoring and
transferring users’ events to the rest of agents. The Student Agent is respon-
sible for the creation and maintenance of a personalised student model. The
Domain Agent retrieves information about the domain, which is stored in a
database. The Tutor Agent is responsible for the metastrategies to be applied
in the tutoring process.
Typical examples of the above architecture (Figure 2) are JADEF (Silveira &
Vicari, 2002) and Alvarez and Fernandez Castro (2003).
Mobile Agent-Based ITSs
Mobile agents constitute an entirely different approach from Web-based

approaches, for example, applets, which are programs downloaded as the
result of a user action, then executed from beginning to end on one host. They
exploit several advantages that make them suitable for personalized and
distance learning, such as the following:
1. Distance learning: A mobile user interface agent may move through the
network to the learner’s place and interact with him/her.
2. Availability of the courseware to a large number of learners (typically
there is no upper limit) by producing only one educational application,
which is a result of the following notions: (a) each mobile user interface
agent that is attached to each learner is a clone of a generic one and (b)
the rest of the agents are common for all learners.
3. Reduction of network traffic: The information transmitted over the
network is minimized, which has strong cost implications for devices
connected by public wireless networks.
4. Support for disconnected operation: A mobile agent is still operational
even though the client that it is situated on is disconnected from the
network.
5. Security: It is possible now to deploy a mobile-agent system that
adequately protects either a machine against malicious agents or the
tutoring system itself and educational materials’ copyrights through au-
thentication of the user.
A representative example is Macro Cell (Deng, Shih, Huang, Liao, Wang, &
Hsu, 2002) in which a mobile agent becomes the user interface of the ITS,
replacing traditional Web approaches.
Proposed Educational
Application Architecture
In order to support knowledge management for self-learners and possibly

mobile learners, a software system has to provide (a) distribution of content:
the use of Internet as a library of knowledge and information enables isolated
learning in which the learner gains the required knowledge through the uncon-
ditional synthesis of distributed information, which sometimes may be contra-
dictory; (b) distribution of expertise: traditional classroom practice has
shown that learners gain the appropriate knowledge through a guided compo-
sition of distributed, different but sometimes limited or incomplete expertise; (c)
support for distributed learners: the software system has to support learners
who form virtual, distributed classes, the members of which change dynamically
depending on learner profiles and requirements; and (d) dynamic content
synthesis: in order to meet the constantly changing requirements of the
contemporary educational settings, educational applications need to dynami-
cally synthesize educational content and strategy based on the distribution of
expertise.
Web ITSs and MAS ITSs lack the capability of dynamic synthesis and
distribution of expertise because in each case the domain module of the tutoring
system resides either on the domain module of the ITS or on a single agent that
embodies the domain module of the MAS ITS. Moreover, in a Web ITS, the
learner, usually via a Web browser, accesses the tutoring system, which is
situated on a single-server machine over the network (client-server implemen-
tation framework). As a result, such tutoring systems do not support manage-
ment of distributed resources of knowledge, content, and information as well
as dynamic synthesis. In particular, the adoption of client-server framework,
allows the management of educational resources local to the server machine
only. In any case, the author of the courseware is charged with the development
of the appropriate content to be presented to the learner.
The architecture for distributed intelligent courseware proposed in this section
implements a network of communicating agents. By distributing both domain
knowledge and expertise into several agents, it allows the efficient management
of distributed information resources and the dynamic synthesis of the infor-
mation that is presented to the learner. In particular, each educational
application may consist of one instructional agent and many domain expert
agents, the configuration of which dynamically changes during the teaching
process (Figure 3).
The instructional agent undertakes the responsibility to control the informa-
tion that is presented to the learner based on a specific instructional strategy,
which is embodied in the instructional agent’s knowledge base. This instruc-
tional strategy is implemented as a set of teaching plans; each plan is a sequence
of teaching steps, each of which can be carried out by a team of expert agents.
Note that the instructional agent is only aware of the specialty of the expert
agents required per plan step; it then dynamically configures agent teams, based
on the set of agents that are active within the system at a particular instance of
time.
A domain expert agent is responsible for providing information and educa-
tional material to the learner about its particular domain of knowledge. A
domain expert contains a set of plans for recalling and presenting information
Figure 3. The proposed educational application architecture
User_1 User_2 User_k
Mobile Agent_1 Mobile Agent_2 ... Mobile Agent_k
Instructional Agent
Communication channel
Domain Domain Domain

Agent_1 Agent_2 ... Agent_k
Communication channel : message passing

: event passing
Search Search
Agent_1 ... Agent_k
Local & Remote

Libraries
modules to the requesting agent. To this end, it is aware of the information

resources, which are associated with a specific domain of the educational
subject that the tutoring system deals with. In general, each domain expert agent
can contribute a part of knowledge or an approach in different parts or cases
of the subject; then a team of agents can collectively teach a subject that is
beyond the individual capabilities and knowledge of each of the domain expert
agents. The synthesis and the management of the distributed expert knowledge
are the responsibility of the instructional agent.
A pool of search agents undertakes the responsibility of searching for
educational material through local or remote information resources on behalf of
each domain expert agent, who filters the results of the search according to its
domain knowledge. During each step of the teaching process, based on the
instructional strategy, the instructional agent forms a team of domain expert
agents (based on the specialty and availability of each). Vertical communication
between the instructional agent and the expert agents includes (a) the request
from the instructional agent to each expert agent to execute one of its plans, (b)
the acceptance or rejection of the request by the expert agent, and (c) the
response of the expert agent to the instructional agent with the information
modules that, according to the expert agent, provide a better match to the
request of the instructional agent.
Moreover, the members of each team communicate horizontally by exchanging
messages in order to best meet the tutoring objectives. Such communication
may involve (a) one expert agent requesting another expert agent to execute
one of its plans, (b) one expert agent requesting another expert agent to act as
its replacement for the execution of a plan, (c) the acceptance or rejection of
the request by the expert agent, (d) the response of an expert agent to the
request of another expert agent with the information modules that, according to
then called expert agent, provide a better match to the request of the calling
expert agent. Then, each domain expert agent serves as a source of information
(possibly synthesizing and) providing the appropriate educational material.
In order to implement this teaching practice, two fundamental issues have to be
resolved: (a) Subject experts are available: They are capable of working
independently, of sensing the educational context, and of cooperating with the
teacher and provide the knowledge bits he/she asks from each of them, and (b)
the teaching material is structured in a way that it can be efficiently communi-
cated. Moreover, it includes a description of its content that can be used by the
teacher to synthesize a broader course.
Each learner accesses the application remotely through a mobile agent. As
long as the learner registers himself/herself once, through a Web interface, this
agent will follow him/her wherever he/she wants to start/continue the educa-
tional process. The mobile agent is the user interface of the tutoring application,
hence its responsibility is to display the educational material to the learner as
well as to track learners’ actions and to inform instructional agent about them.
Applications: A Use Case
The proposed architecture has been initially designed for and applied to the
development of medical courseware. In general, medical education involves the
transfer of voluminous knowledge that is constantly growing. The successful
application of medical skills calls for a synthesis of different knowledge sources
which sometimes offer contradictory and incomplete information. Traditional
medical education adopts a system-oriented approach: each system of the

human body is examined from several perspectives of medical interest.
We approached the CBME problem in a manner that is common among
physicians: each specialist in a medical subject contributes his/her knowledge
on the case at hand, in order to solve as a team a problem that is beyond the
individual capabilities and knowledge of each of them. Each developed
application is a decentralized system, composed of “medical agents,” each of
which implements a medical specialist system. A medical specialist system can
be regarded as a representation of a medical expert: it contains the knowledge
and is aware of the information resources, which are associated with a specific
domain of medical science. In the case of CBME, the domain that the systems
are aware of becomes a cognitive domain that can be used to describe (teach)
a part of a generic medical subject. In essence, each specialist system can
contribute a part of knowledge or an approach in different parts or cases of the
subject. Each aspect of the medical subject is taught by a specialized medical
agent, who is responsible both for acquiring and for delivering the information
pertinent to the subject, according to his specialty.
The main idea underlying our approach is similar to the one behind Huang,
Jennings, and Fox (1995). Although this system deals with healthcare manage-
ment and patient treatment, it also uses a multiagent architecture to deal with the
distributed, multidisciplinary and uncertain nature of medical knowledge. Our
design focuses mainly on issues concerning the implementation and deployment
of the multiagent system, as well as issues regarding the retrieval and synthesis
of medical knowledge with respect to concrete instructional plans.
In general, an organization populated by agents provides a framework for agent
interactions through the definition of roles, behaviour expectation, and authority
relations. Sycara (1998) points out four kinds of organizations: the Hierarchy,
where the decision control belongs to a single agent and interaction takes place
through vertical communication; the Community of Experts, where a group of
specialists interact through rules of order and behaviour; the Market, where a
group of agents interact through bidding and contractual mechanisms; and the
Scientific Community where solutions to problems are published for testing
and refinement.
In the proposal, we adopt a hybrid organization that combines features from the
first two approaches. In order to collect and synthesize medical knowledge, a
group of medical specialist agents is formed according to the Community of
Experts organization. However, even though we permit vertical and horizontal
communication we do not adopt the decision control to be based on bidding
and contractual mechanisms. On the contrary, decision control belongs to a

single agent, as in the Hierarchy organization.
An application based on the proposed architecture (a) consists of an instruc-
tional agent and a set of domain expert agents and (b) contains a logically
integrated (but probably physically distributed) database of educational mate-
rial. This is partitioned into self-contained modules called Learning Units (LUs).
An LU is a hypermedia construct of multimedia material and is described by
attributes that represent both static information used to identify the unit (i.e.,
title, type, location, size, etc.) and dynamic information that describes the
behaviour of the unit (i.e., pedagogical prerequisites and objectives, display
constraints, etc.).
The development of an educational application involves the design of (a) the
instructional strategy that the educational system should adopt during a teaching
procedure; (b) the domain knowledge for each domain agent that populates the
multiagent system; (c) the interrelations among the agents; and (d) the educa-
tional material that will be presented to the learner.
Using this system, and as a form of evaluating the approach, we have developed
two educational applications: Dermatology Tutor (Zaharakis, Kameas, &
Nikiforidis, 1998) for teaching psoriasis, and OPD Tutor (Triantis, Kameas, &
Nikiforidis, 2000) for teaching the treatment of “outlet” impingement in
orthopaedics.
Dermatology Tutor
As mentioned above, Dermatology Tutor implements the proposed architec-

ture (Figure 4) in order to teach psoriasis. Each medical discipline of the subject
(e.g., physiology, anatomy, immunology, microbiology, etc.) is taught by a
specialized medical agent, who is responsible both for acquiring and for
delivering the information pertinent to the subject, according to its specialty.
As the proposed architecture is composed for tutoring purposes, only the
Dermatology Agent can act as an instructor; the instructional strategy neces-
sary for the tutoring process is contained in this agent’s plan presented in
Table 1. Each step of this plan constitutes a lower-level plan, which can be
further analysed (Table 2).
Based on its plans, the Dermatology Agent can form and manage teams or
communicate with the other agents via the communication channel. The domain
Figure 4. Dermatology Tutor architecture
User_1 User_2 User_k
Mobile Agent_1 Mobile Agent_2 ... Mobile Agent_k
Dermatology Agent
Histology Physiology Biology Biochemistry Microbiology Radiology Immunology

Agent Agent Agent Agent Agent Agent Agent
Search Domain : message passing

Agent_1 ... Agent_k : event passing
Local & Remote

Libraries
agents populating the society can be members of a team formed by the

Dermatology Agent; they can also communicate with each other via the
communication channel.
For example, if the learner wishes to learn about the etiology and pathogenesis
of psoriasis, the Dermatology Agent will try to teach the specific topic based
on the corresponding plan (Table 1). According to this plan, it needs to form
a team with Biology, Biochemistry, Immunology, and Histology medical agents.
In particular, in order to fulfil its first partial goal (“Disorder of protein
expression in keratinocytes”), the Dermatology Agent requests, through mes-
sage communication, the action from Biochemistry, Biology, and Histology
agents. As the Biochemistry, Biology, and Histology agents have a correspond-
ing plan about the requested action, they execute the plan and return the
appropriate educational material to the Dermatology Agent. The Dermatology
Agent controls the flow of the educational material that is presented to the
learner according to the learner’s events. The educational material is presented
Table 1. Dermatology Agent plan for teaching psoriasis

Plan na me : Teach psoriasis
Preconditions:
Plan body: Definition and Epidemiology
Clin ical View
Histological View
Etiology and Pathogenesis
Treatment
Table 2. Etiology and pathogenesis plan of the Dermatology Agent
Plan name: Etiology and Pathogenesis

Preconditions: Associated Agents
Plan body: Disorder of p rotein expression in Biology, Biochemistry,
keratinocytes Histology
Antigenic stimulat ion and immunology Immunology
activation
Proteases and intense mitotic activity Biology, Biochemistry
Metabolic disorders Biology
Histopathologic changes of psoriasis Histology
to the learner via the mobile agent. Finally, each domain agent uses a search
agent in order to retrieve the appropriate LU from the database. Moreover,
each search agent retrieves LUs based on a set of keywords that was given (in
a form of message) by a domain agent.
OPD Tutor
As mentioned above, OPD Tutor implements the proposed architecture in

order to teach treatment of “outlet” impingement in orthopaedics. The case at
hand requires the contribution of the following medical experts: Orthopedist,
Anatomist, Kinesipathist, Radiologist, Pathoanatomist, Pathophysiologist,
Laboratorian, and Physiatrist, each of which is a medical expert in orthopedics,
anatomy, kinesipathy, actinology, pathoanatomy, pathophysiology, labora-
tory, and physiatry, respectively. Bearing in mind that this architecture is used
for tutoring purposes, only the Orthopaedist Agent can act as an instructor; all
the instructional strategies necessary for the tutoring process of “outlet impinge-
ment” are contained in this agent’s plans.
Low-Level Architecture
In order to achieve modular, open, and dynamically changing behaviour of the

system, the proposed architecture uses an agent called Facilitator (Genesereth
& Ketcpel, 1994). Facilitator is not involved in the teaching procedure. This is
the reason why Facilitator is not shown in the high-level architecture of the
educational system. Facilitator is responsible for maintaining an accurate,
complete, and timely list of the services that a registered agent can provide.
When each agent is started, its first responsibility is to register itself and its
services to Facilitator. Each agent that participates in the organization is
implemented as an individual, autonomous program without either knowing the
capabilities or the existence of the other agents, except Facilitator. All the
communication that takes place among the agents passes through the Facilita-
tor. For example, if the Dermatology Agent wishes to send a message a to the
Radiologist Agent, he will first send a message b to the Facilitator asking for it
to forward the message a to the Radiologist Agent. The answer of the
Radiologist Agent will pass through the Facilitator, too.
Allowing point-to-point communication only between an agent and the Facili-
tator, provides to the educational system with desirable properties such as (a)
Modularity: each agent can be implemented by different people and different
programming tools with the restriction of knowing the communication means
(explained later) between it and the Facilitator; (b) Reusability: it permits an
agent to be a member of more than one educational systems; (c) Adaptability:
it is easy to change the instructional strategy of the educational system simply
by changing the instructional agent without programming and recompiling the
whole system; (d) Openness: it is possible to replace any medical agent “on the
fly” (dynamically and transparently), while the system is in use; (e) Plug ‘n’
play: by registering his capabilities to Facilitator, each agent enters the tutoring
processing upon instantiation; (f) Scalability: considering that a Facilitator
controls one educational domain, the system can easily be scaled up to support
multidomain education by adding one or more multiagent systems, each with its
own Facilitator, which interact through a “common” Facilitator of a “higher”
level.
The main disadvantage of point-to-point communication only between an agent

and the Facilitator is the unavoidable delay of message transportation from one
agent to another. The delay increases if more than one domain participates in
the tutoring architecture, since communication may pass through the Facilitator
of two domains. Nevertheless, the above-mentioned advantages and the speed
of contemporary computers make up for these delays. The above advantages
of the selected low-level architecture fulfilled the goals of this framework.
Implementation Issues
In our work we consider an agent to be a modular and autonomous executable

program. His implementation comprises the development of three separate
parts: the Message Structure, the Agent Core, and the Communication
Module. The Agent Core (Figure 5) is the essential part of each agent. It
comprises the knowledge base and the reason-about mechanisms of each
agent. Its structure is based on the well-known BDI logical framework (Cohen
& Levesque, 1990) where each agent is viewed as having the three mental
attitudes of belief, desire, and intention (BDI). In addition every agent contains
a plan library, an inference engine, and a communication module. Plans
describe the possible ways that an agent can bring about an intention. The
Inference engine is the mechanism that handles and updates the agent’s mental
state (beliefs, desires, and intentions), selects, executes, and rejects plans
according to the mental state and processes the incoming message that transfers
the observations of the agent’s environment.
In our framework, we implement the Agent Core using the PROLOG program-
ming language because we believe that the built in backtracking mechanism as
well as the data structures that it supports, makes it suitable programming
language for nondeterministic management of knowledge and information.
The communication module is responsible for message transportation. It
consists of two separate parts: the integration module and the transport
mechanism. The integration module is responsible for the integration between
the Agent Core and the transport mechanism. In this way, we support the ability
to choose the appropriate transport mechanism according to the current
implementation circumstances without the need of adjustment of the whole
agent’s implementation.
Figure 5. The agent structure
A g en t C o re B eli efs
In fe re n c e E ng i n e
A gent
C o m m u n ic a t io n M o d u le
D esire s
In t e r g r a t io n M o d u l e
T r a n sp o r t Inte ntion s
M e ch a n is m
(C O R B A)
P lan s
Message
W o r ld
Agent Core
The transport mechanism is the physical means for message exchange. Several
transport mechanisms have been proposed in agent literature. In our frame-
work, considering each agent as a client/server entity, the transport mechanism
has been implemented with CORBA for two reasons: (a) FIPA addresses a
minimal set of requirements on the communication protocol for building open,
distributed systems (FIPA, 1999), which are fulfilled by CORBA, and (b)
CORBA offers services, such as naming services, trader services, access
control services, event services, which make it suitable to easily implement
interagent communication mechanisms.
The Message Structure
A message is an individual unit (building block) of interagent communication

that is based on the speech act theory (Searle, 1969), where every intended
communication action from an agent changes the world in an analogous way to
that of a physical action. A message as defined by FIPA in (FIPA, 1999) and
implemented in our framework has the following structure: Communicative
Act(Sender, Receiver, Content(), in-reply-to, reply-with, Language,
Ontology). The communicative act (CA) implies the action that an agent
performs. The parameters of the action described by the elements of the
message. In our framework, we make use of a subset of FIPA’s CAs, which
are common for all the agents that populate the society of the medical
educational system. Each CA implies that the content of the message contains
a specific set of actions. In this case the cancel, request, and refuse CA implies
that the receiver will perform a specific action. The inform CA implies that the
receiver will update his knowledge base and especially his beliefs about specific
things of the environment. The failure and not-understood CA implies that the
receiver will update his knowledge base and will act in order to deal with the
error handling.
The comprehension and the execution of the action depends on what actions
the recipient can perform. These actions are defined by the Content element,
which has precisely defined syntax, and semantics (FIPA, 1999). In our
framework, we make use of the PROLOG language structure and syntax, in
order to define the Content element. As a consequence, each action is a
PROLOG data structure. The Content is not the same for all the agents, but
depends on the actions that each agent performs. In this case, the actions are
divided into those supported by Facilitator and those supported by either of the
medical agents or the Instructional Agent.
Actions Supported by Facilitator
Register: An agent registers its services to the Facilitator in order to publicise

some or all of them to other agents. There is no intended future commitment or
obligation, on the part of the registering agent implied in the act of registering.
For example, an agent can refuse a request for a service, which is advertised
through the Facilitator. There is a commitment on behalf of the Facilitator to
honestly broker the information it holds.
Search: A search action implements a request from an agent, who wants as a

result, a specific list of agent names that involve specific services. The
Facilitator is responsible for searching his database of the registered and active
agents and to provide the list of agent names. The answer to a search request
is a result information.
Modify: This action involves changing an agent’s registration fields. The

content of a modify message will replace the information which is currently
registered for that agent.
Deregister: An agent deregisters in order to remove any record of his services

from a domain. The deregister action has the consequence that there is no
longer a commitment on behalf of the Facilitator to broker information relating
to that agent.
Forward: An agent can ask the Facilitator to forward a message to a

destination agent. In the case that the sender does not know the name of the
receiver then Facilitator has to search among the registered agents with the
service as a keyword and forward the message to one of the agents (randomly)
that provides the same service. For example, when the Orthopaedist Agent
needs someone with knowledge of Anatomy, it requests the Facilitator to
forward the message to a registered agent who provides Anatomy services.
Actions Supported by the Rest of Agents
Teach: The teach action implies that the recipient will undertake the responsi-
bility of finding the appropriate LU for the specific plan name and return the
LU’s location to the sender.
Location: The location action returns to the sender the appropriate LU’s
location according to the requested plan name. The location action is an answer
to the teach one. The location action updates the sender’s knowledge base.
Result: The result action is an answer to the requested action search. The
Facilitator returns to the sender a list of agents who can provide the requested
service. The result action updates the sender’s knowledge base.
Conclusion
The proposed architecture implements distributed intelligent courseware by

using features of MAS. It approaches an educational subject as a set of
distributed knowledge and expertise modules, and delivers educational content
to the learner through a mobile user interface agent. There are several
advantages that emerge from the proposed architecture:
1. Reusability. A domain expert participates in more than one tutoring

system. Thus a domain expert can be part of the teaching procedure of
more than one subject. In addition, an instructional agent of a particular
tutoring system can be a domain expert to another application.
2. Adaptivity. The instructional strategy can be adopted according to the
needs of the instructional process since it is easy to replace the existing
instructional strategy by replacing the instructional agent.
3. Scalability. Each instructional agent defines a different tutoring applica-
tion. Thus, the developer can scale up the proposed architecture to
support multiple subjects by adding the appropriate instructional agent as
well as the missing domain agents.
4. Rapid prototyping. The modular structure provides a variety of stable
intermediate forms that are essential for the rapid development of complex
systems.
5. Maintenance. Individual domain agents or organisational groupings can
be developed in relative isolation and then added into the system in an
incremental manner.
6. Distribution of expertise, as a result of the proposed educational
architecture, which consequently provides quality control independently
from the educational domain, by a number of experts (domain expert
agents).
7. Network fault tolerance through the existence of more than one expert
agent that exploits the same expertise. In the case that a domain expert
comes out of order, the request can be addressed to another domain
expert that shares the same knowledge with the previous one.
8. Communication decongestion: agents exchange messages among them-
selves only when necessary, without keeping a constantly open connec-
tion among themselves, as the traditional client/server approach requires.
Future Work
After designing and implementing educational systems for medical education,

we are now in the process of applying the described architecture to alternative
educational subjects. Currently, the research team is designing an application
for teaching aspects of educational software evaluation. Evaluating educational

multimedia software is a difficult subject matter, since it ought to cover an
extensive range of topics. It should be based on a predefined set of specifica-
tions and predictable results arising from the use of the evaluated product. This
application deals with the process of evaluating educational multimedia soft-
ware and in particular it considers the subjects of the evaluation factors and the
guidelines needed for selecting and evaluating educational software.
The embodiment of student modelling techniques to the proposed application
may be achieved (a) either by implementing a separate student modelling agent,
which will interact with the instructional agent in order to adapt the instructional
strategy to the learner’s profile, or (b) by enclosing these techniques to
instructional agent plan library.
Finally, in the context of a research project the authors have already proceeded
to the implementation of an integrated authoring environment for developing
multiagent distance learning systems based on the proposed architecture. The
authoring system, which is currently under evaluation, allows the trainer (author
user) to specify and describe the instructional strategy of the system as well as
the plans of the participating agents via a graphical formal model. Based on this
model, the authoring system produces automatically the educational multiagent
system, applicable to any educational domain, which is as complex, large, and
dynamic as any medical domain.
References
Abdel Razek, M., Frasson, C., & Kaltenbach, M. (2003). A context-based

information agent for supporting education on the Web. Proceedings of
the 2003 International Conference on Computational Science and Its
Applications, 2667 (pp. 170–179).
Alvarez, A., & Fernandez Castro, I. (2003). An open adaptive and multi
subject educational system for the Web. Proceedings of the 11th
International Conference on Artificial Intelligence in Education,
Sydney, Australia.
André, E., Rist, T., van Mulken, S., Klesen, M., & Baldes, S. (2000). The
automated design of believable dialogues for animated presentation
teams. In Cassell et al. (Eds.), Embodied conversational agents (pp.
220–255). Cambridge, MA: MIT Press.
Andriessen, J., & Sandberg, J. (1999). Where is education heading and how
about AI? International Journal of Artificial Intelligence in Educa-
tion, 10, 130–150.
Clark, R. (1998). Building expertise: Cognitive methodology training and
performance improvement. International Society for Performance
Improvement, Washington, DC.
Cohen, P., & Levesque, H. (1990). Intention is choice with commitment.
Artificial Intelligence, 42, 213–261.
Deng, L., Shih, T., Huang, T., Liao, Y., Wang Y., & Hsu, H. (2002). A
distributed mobile agent framework for maintaining persistent distance
education. Journal of Information Science and Engineering, 18, 489–
506.
e-College: Course Management System. www.ecollege.com/products/
collateral2.learn
Foundation for Intelligent Physical Agents (FIPA). (1999). Agent communi-
cation language specification. Retrieved December 30, 2004, from
www.fipa.org/specs/fipa00003/
Genesereth, R., & Ketcpel, P. (1994). Software agents. Communications of
ACM, 37(7), 48–53.
Huang, J., Jennings, R., & Fox, J. (1995). An agent architecture for distributed
medical care. In M. Wooldridge & N. Jennings (Eds.), Intelligent agents
(pp. 219–232). Springer-Verlag.
Jennings, N., & Wooldridge, M. (1998). Applications of agent technology. In
M. Wooldridge & N. Jennings (Eds.), Agent technology: Foundations,
applications, and markets. Springer-Verlag.
Lotus: Learning Management System. www-306.ibm.com/software/lotus/sw-
bycategory/subcategory/SWC00.html
Rindos, A., Vouk, M., Woolet, S., Hines J., & Lester, J. (1995). ATM
technology enabling educational applications across the North Carolina
information highway. Proceedings of Telecom 95, Geneva, Switzerland.
Searle, J. (1969). Speech acts: An essay in the philosophy of language.
Cambridge, UK: Cambridge University Press.
Sherman, A., Singley, M., & Fairweather, P. (1999). Deploying intelligent
tutors on the Web: An architecture and an example. International
Journal of Artificial Intelligence in Education, 10, 183–197.
Silveira, R., & Vicari, R. (2002). Developing distributed intelligent learning

environment with JADE–Java Agents for Distance Education framework.
Proceedings of 6th International Conference, Intelligent Tutoring
Systems, 2363 (pp. 105–118).
Sycara, K. (1998). Multiagent systems. AI Magazine, 19(2), 78–92.
Triantis, A., Kameas, A., & Nikiforidis, G. (2000). On the implementation of
a multi-agent system for computer based medical education. Proceed-
ings of 2nd Panhellenic Conference with International Participation
in Information and Communication Technologies in Education,
Patra, Greece.
Weber, G., & Brusilovsky, P. (2001). ELMART: An adaptive versatile system
for Web based instruction. International Journal of Artificial Intelli-
gence in Education, 12.
Zaharakis, I., Kameas, A., & Nikiforidis, G. (1998). A multiagent architecture
for teaching dermatology. Medical Informatics, 23(4), 289–357.
Impact of Individual Differences 261
Chapter XII
Impact of Individual
Differences on
Web Searching
Performance:
Issues for Design and the
Digital Divide
Allison J. Morgan, Pennsylvania State University, USA
Eileen M. Trauth, Pennsylvania State University, USA
Abstract
This chapter will encourage the consideration of the role of individual

differences in determining Web behavior and performance, which could
inform and improve the development of search engines. Currently, users
of search engines may experience differences in their level of success in
searching for information. This difference could be realized through
search success or search strategies. However, there is currently no
definitive explanation regarding the characteristics that influence
differences in search engine use and behavior. This chapter will serve as
262 Morgan & Trauth
an introduction to and explore the phenomena of online Web searching

and the potential role of individual differences in investigating this
situation. An overview of the literature will be detailed as well as issues
regarding how individual differences can be incorporated into this type
of research. This chapter will support the notion that individual usage
and performance with Web search engines is influenced by a collection of
factors, more specifically, individual differences.
Introduction
The Internet is a revolutionary technology that is changing the way that people
create and search for information, as well as communicate within the global
society. The importance of this technology has been widely documented in
terms of access to information, resources, organizations, virtual communities,
and networks of people.
Currently, Internet search engines provide users with their primary source for
locating content on the Web. They also assist in the information searching
process and navigation of the Internet using keywords and search terms.
The majority of Internet search engines are straightforward technologies that
do not include a great deal of capabilities for personal customization. The
developers of Web search engines make an effort to add advanced searching
features such as Boolean operators and character strings to make searching
more specific to the desired subject. In addition, a combination of skills with
the technology, knowledge of the subject matter, and knowledge of online
search processes can be beneficial in searching for information on the Internet.
However, the experiences of users of search engines may differ in their level
of success in searching for information. There are two ways in which the
performance of individuals may vary with regard to search engines. Primarily,
there is a difference in being able to find what one is searching for. In addition,
there are differences in strategies that are used to search or the way in which
people go about finding the information they seek.
So the question remains as to why some users are more effective searchers than
others. Another question is whether the experience that a user may have with
a Web search engine impacts his/her attitude toward Internet technology as a
whole? The characteristics that can be attributed to one searcher performing
more successfully than another are presently not well defined. It may possibly
be that the design of these systems provides advantage to certain types of users
rather than others. There may also be other factors. However, individual
differences may play a role in the Web searching skills of users and their
performance with Web search engines.
Individual differences could include a range of characteristics including gender,
race, age, socioeconomic status, education, income, cognitive style, personal-
ity, self-efficacy, and so forth. So an individual user’s skill and performance
with a Web search engine can be motivated or influenced by a number of these
factors. Our goal is to investigate how these individual differences play out in
the performance of users of search engines, and what, if any, impact they have
on the successful use of that technology. Although it is important to study the
range of individual differences, our focus will be mainly on demographic
differences.
In addition, we would like to frame the evaluation of this research in terms of
social exclusion and the digital divide. Greater numbers of people are exposed
to the Internet on a daily basis, but more research is needed to address the
motivations and barriers to effective use of the Internet. So the study of the
impact of individual differences in Web searching performance may also
provide additional information as to why a disparity exists among users and
nonusers of Internet technology.
Studying the role of individual differences with Web searching behavior is
important on many levels for many different populations. On the societal level,
Internet technology is increasingly becoming a part of everyday life as well as
a requirement for many jobs. Access to the Internet has been studied in various
dimensions to determine its effect on people’s use of the technology (Katz &
Rice, 2002). Access in this situation can be defined as a function of resources.
Research has shown that physical access can be a barrier to use of the
technology, but placing the Internet in publicly accessible places such as
schools, libraries, and community centers, has not yet worked to close the
digital divide. Therefore, it is important to continue to research the types of
barriers that may be deterring users from participating in Internet technology.
This should be carried out in an effort to determine what measures can be
implemented to address the problem. If the Internet or its components can be
improved to be a more inclusive and usable technology, then the resources and
benefits of the Internet could have a more far-reaching effect. Ideally, the Web
is a technology that serves all people, not just the privileged in society (Introna
& Nissenbaum, 2000).
264 Morgan & Trauth
The purpose of this chapter is to encourage the consideration of the role of

individual differences in determining Web behavior and performance, which
could inform and improve the design of search engines. This chapter will serve
as an introduction to the literature and basis for understanding the importance
of individual differences in relation to human information behavior. The chapter
will review current explanations of the digital divide to evaluate the facilitators
and inhibitors of Internet use. Next, we will explore the user behavior literature
to search for explanations of user behavior with search engines and information
retrieval. Then we will explore the role of individual differences in information
systems. We will then investigate what current research into individual differ-
ences explains, and introduce the individual differences theory by Trauth
(Morgan, Quesenberry, & Trauth, 2004; Quesenberry, Morgan, & Trauth,
2004; Trauth, Quesenberry, & Morgan, 2004; Trauth, Quesenberry, & Yeo,
2005). Last, we will outline an agenda for researching individual differences in
Web search engines and develop implications for design that integrate the
concept of individual differences.
This study will contribute to the research community as well as to society by
helping to identify the factors influencing different users’ Web search skills and
identifying behavior that could provide developers with recommendations to
build better, more usable systems. In order to make technology valuable to the
greater population of users it is necessary to understand the ways in which
information technology can be beneficial to diverse users, and the underlying
social judgments that people make when using technology.
Literature Review
In an effort to better understand this situation, this chapter will conceptually

explore the role of Web search engines in relation to Internet use. By
additionally evaluating the literature on human information behavior, the digital
divide, and individual differences, we will show the need for the study of
individual differences with Web searching behavior.
Web Search Engines and the Internet
If the Internet is an important technology in society today, Web search engines

can be identified as the gatekeeper to them. According to Introna and
Nissenbaum (2000), “search engines constitute a powerful source of access

and accessibility within the web.” Although many companies, organizations,
and individuals make a concerted effort to publicize their Web site URLs,
search engines still provide the primary source of locating content on the
Internet. The ability to search successfully has an impact on whether an
individual will be able to locate a Web site that contains the information that he/
she seeks. However, an individual user’s skill and performance with a Web
search engine can be motivated or influenced by a number of factors. These
factors could include two types of individual differences. One level of individual
differences takes in to account demographic and social characteristics, includ-
ing age, race, gender, socioeconomic status, and so forth. The other level of
individual differences takes in to account other characteristics, including
education level, socioeconomic status, and geographic location.
Current Web search engines are built upon the principles of traditional
information retrieval (IR) systems. Much of the functionality of these IR systems
is based on a notion that users are able to represent their information need in
the form of text. Web search engines follow in this tradition. The argument is
made that much of the current complexity in search engines is a result of the size
of the Web. Methods that originate from private collection IR systems are not
scalable enough to meet the needs of a collection with the magnitude of the
Internet. So traditional techniques have had to be adjusted and enhanced to
operate effectively with the Web. In addition, Jansen and Pooch (2001) assert
that information seeking in traditional information environments holds very
important differences to Internet-based Web searching. Studies show that the
advanced features of search engines are not as beneficial as they are described
to be. Also, the indication of these advanced search operators is that they are
useful only to those who are familiar with the structures and inner-workings of
IR systems.
Introna and Nissenbaum (2000) discuss how the structure and politics of Web
search engines affect users. The authors argue that Web search engines
emphasize certain Web sites while making others virtually invisible. In addition,
they state that many of the Web sites that receive visibility are popular and
mainstream commercial Web sites that are maintained by large economic
powers. These large organizations help to illuminate their own presence to the
detriment of other small entities. Those neglected Web sites are then further
alienated due to lack of traffic and eventually are completely taken down, which
narrows the options and content availability on the Internet to Web users. The
authors argue that access in its truest sense means a “comprehensive mecha-
266 Morgan & Trauth
nism for finding and being found” (p. 30). They conclude that search engines are
politically influenced through technical means such as crawlers, indexers, and
ranking algorithms.
Arasu, Cho, Garcia-Molina, Paepcke, and Raghavan (2001) provide a com-
prehensive overview of Web search engine design and its challenges. A large
number of Web search engines utilize IR techniques and algorithms. The
difference in the size of the audience for Web search engines versus private
collection IR systems makes the design of these systems even more complex
and important. The use of new techniques and methods to meet the challenge
of scalability for Web search engines improves chances for improving rel-
evance in the retrieval of information. Research suggests that the Web structure
mimics that of a “‘bow tie” in which roughly 28% of the pages constitute the
core of the bow tie and 22% make up the loop which connects also to the core
but cannot be reached from it” (p. 3).
The impact of query operators on Web search engines was researched by
Eastman and Jansen (2003) to determine if they result in improved searching
capabilities. Presently there are roughly 32,000 search engines in existence
which 71% of Web users visit to locate information on the Web (p. 389). In
Web search engines, advanced phrase search capabilities and Boolean opera-
tors are used rarely. However, there is a perception that the use of these
conventions increases the effectiveness of Web searching by “increasing the
total number of retrieved documents, increasing the number of relevant
documents retrieved, or improving the ranking of relevant documents” (p.
384). A number of the major Web search engines even recommends use of
advanced operators to improve search on their Web sites. Research suggests
that Web users do not utilize advanced searching capabilities for reasons that
include laziness and search engine design flaws. The results of Jansen’s
research show that, in fact, advanced operators in Web searching actually do
not increase or decrease the precision of the search. Additionally, these types
of advanced functions are useless without a working knowledge of IR systems
because integration of these conventions is not intuitive to the user.
Belkin, Cool, Stein, and Theil (1995) discuss the design of interactive IR
systems. Traditional IR systems have focused on “representation of texts and
queries, and on comparison of these representations” (p. 379). The inclusion
of interactivity into the IR systems has placed new focus on the activities of the
user, including formulation of queries and retrieval response. The authors
present a theoretical basis that supports these types of activities occurring
within the context of information-seeking strategies. Their model characterizes
information retrieval distinctly as human-computer interaction impacted by

information seeking strategies. This situation represents a dialogue between the
user and the system, and should be designed and developed based on that
notion. The structure of the resulting dialogue can be framed within different
information-seeking strategies, and use of case-based reasoning can provide
a model for patterns of interaction. Last, the integration of “information-seeking
strategies, dialogue structures, scripts, and cases can be used in a system design
that uses the advantages of each to ameliorate the disadvantages of each” (p.
394).
Searching is an important process to Internet use. People have different
experiences of success and failure in regard to information that they seek on
Web search engines. In addition, because it is possible that a person’s
experience with a search engine may affect his/her overall level of interaction
with the technology, it is important to identify if individual differences of users
affect their performance with Web search engines. If so, implications for the
design of Web search engines and the higher-order effect for the digital divide
should also be identified.
User Behavior
Users of technology interact with it in a number of ways, especially in their

search for information. When looking to solve a problem or answer a question,
users find different methods of formulating that question in the context of their
own knowledge. Based on their knowledge, users look for a way to find the
information they need in the environment of the information system. However,
the arrangement of information systems is not always based on the way that
different individuals organize their search for information. This makes the task
of information searching very complex. In addition, the experience that a user
may have with a Web search engine may also have an impact on their attitude
toward Internet technology as a whole.
Bates (1989) describes a way in which to model online and information systems
searching called “berry picking.” Through this method, she challenges the
conventional information retrieval model because she believes that it does not
adequately represent the way in which users search for information. Bates
argues that users begin with a piece of a larger topic and move iteratively
through sources where the query continually evolves. More specifically, berry
picking is characterized as the process that happens when users gather more
268 Morgan & Trauth
and more information at each stage of searching, thereby satisfying the query
by a series of selections, not the single retrieval set. Using a model such as berry
picking that involves the wide horizon of user behaviors may allow for better
design of information retrieval systems. Last, the author notes that users with
the widest range of search strategies will be the ones with the greatest
information retrieval power.
Borgman (1996) discusses the difficulty that users encounter with online
catalogs. She cites the main reason for that difficulty being the “design that does
not incorporate sufficient understanding of searching behavior” (p. 493).
Important to her argument is the notion that online catalogs are being designed
with skilled workers in mind instead of an average end user. The process of
searching is described as one that happens over time through a combination of
information technologies and resources, where the user works iteratively to
explore his/her inquiry. This process is very complex because searchers have
to interpret their information need based on their own knowledge of the
problem and understanding of the system. The perspective of this article
supports the notion that “information retrieval is a difficult problem because it
requires information that [one] does not yet have” (p. 494). Borgman illustrates
that for online catalog searching, three levels of knowledge are necessary:
conceptual knowledge of the information retrieval process, semantic knowl-
edge of how to implement a query in a given system, and technical skills in
executing the query.
The study by Gauch and Smith (1993) demonstrates the use of an expert system
that automatically reformulates queries to improve search results. The authors
conducted this study based on the problem of user unfamiliarity with search
tactics of online retrieval systems. The authors found that most users, even
experienced ones, reformulate their queries incorrectly. Their expert system is
a knowledge-based assistant that operates as a front end to information
retrieval systems.
Belkin, Oddy, and Brooks (1982) discuss a different approach to developing
IR systems based on anomalous states of knowledge (ASK). The authors’
premise is that the information needs of users cannot be easily specified
because they are not facts in themselves, but rather a means by which to find
the resolution to a problem. Traditional IR systems utilize a method known as
“best match” through which a system responds to a query based on the text
whose representation most closely matches it. However, this technique re-
quires that users be able to coherently express their information need in a simple
text form, which may or may not completely represent their understanding of
the problem. The authors’ hypothesis for the ASK system is that information
needs result from anomalies in the “user’s state of knowledge regarding a topic
or situation” (p. 62). Based on this notion, IR should use a process that helps
to understand the users ASK rather than require them to specify their own
request for information.
Saracevic and Kantor (1988a, 1988b) performed a study that aimed to
characterize the components of information seeking and retrieving. The authors
stress that human decisions and human-system interactions are the most
significant factors in the processes involving searching for and retrieval of
information. They also emphasize that the “key to the future of information
systems and searching processes … lies not in increased sophistication of
technology, but in increased understanding of human involvement with informa-
tion” (p. 162).
The human search for information is a complex process which is further
complicated by the inclusion of technology. Information systems in the form of
online catalogs and information retrieval systems appear to require the user to
search within the constraints of the system itself, rather than allowing the user
to specify his/her own criteria. Bates’ (1989) method of berry picking provides
a model for the iterative process of human information searching. However, this
model also highlights the difficulty that is encountered by users in trying to
specify their information needs as being able to search for information often
involves having some knowledge of the very topic on which the inquiry is based.
Beyond differences in user behavior, there also exists literature that discusses
the differences with individuals who utilize technology and those who do not.
Research on the digital divide highlights reasons for the disparity among users
of technology and addresses implications for its existence.
The Digital Divide
In 2000, the percentage of households in the United States that reported having
Internet access was 41.5% according to the National Telecommunications and
Information Administrations’ report entitled Falling Through the Net. The
disparity in the amount of people participating and not participating with the
Internet has been labeled the “digital divide.” In addressing the digital divide,
Kvasny and Truex (2000) state that as new technology is deployed, classes of
users are empowered while others lose power. Explanations for the gap have
been associated with the concept of unequal access.
270 Morgan & Trauth
Unequal access is largely correlated with technological access. Technological

access is the traditional idea of access that corresponds to the physical
availability of appropriate computing equipment or the ability to attain the
hardware, software, and connectivity necessary to actually use Internet tech-
nology. An additionally important notion regarding the digital divide is that of
social access. Social access is more concerned with the skills, knowledge, and
perceived benefit to engage in using the Internet. Social access has been
defined by Kling (1998) as the mix of professional knowledge and technical
skills that augment professional practices and social life. A variety of external
factors may cause individuals to use the Internet less, such as the need for
technical support and a connection to a network of other online users. Kling
asserts that the Internet is a medium that can be used socially by ordinary people
to accomplish a variety of tasks.
The digital divide is a phenomenon that represents the gap between those who
participate with technology and those who do not. Traditionally, the argument
is framed in terms of the “haves” and the “have-nots” as it relates to physical
access. Many different types of demographics are used to examine the use of
technology and much of the early research points to access in order to solve the
problem of the digital divide. However, access, though a first step in lessening
the divide, is not the end all for eliminating the problem. The digital divide is
largely a representative example of the inequality and the power struggle that
exist in our society. The technology that is being developed in present times
serves to further advantage those who already have privileged status and further
exacerbates the inequalities that exist.
The National Telecommunications and Information Administration (NTIA)
released a report in 1995 entitled Falling Through the Net. This report was
one of the seminal works to introduce the notion of the information “haves” and
“have-nots.” This survey contrasted the goal of “universal service” that
corresponds with the idea that all Americans should have the ability to access
the telephone service, with the levels of penetration of the Internet across the
nation.
Hoffman and Novak (1998) give a comprehensive look at the digital divide and
the importance of race as it relates to computer usage. This article examines the
differences between African American and White computer and Internet users
at different demographic levels to determine how to better address the divide.
Many key findings were addressed in this research based on a quantitative
analysis of the Nielsen Internet Demographic Study. The authors state that
household income is directly related to computer ownership in the home
regardless of race. Also, the level of education that a person has correlates to
access to computers at work regardless of race. However, race did become
a factor in other areas of the study. African Americans are less likely to own a
home computer than Whites at every level of education, and are more likely
than Whites to have access to a computer at work when income is taken into
account. Additionally, the authors say that both income and education affect
computer access and subsequent Web use. This article concludes that African
American students require multiple points of access to participate in computer
technology. Education is also determined to be the most important factor in
computer usage. This notion leads the authors to conclude that better educa-
tional opportunities for African Americans will result in better participation of
all Americans with technology. The article concludes with the notion that if
access to computer technology is ensured, then computer usage will follow.
Katz and Aspden (1997) report on the motivations and barriers to Internet
usage based on a national telephone survey administered in 1995. The results
of the survey showed evidence of a digital divide due to the wealthy and highly
educated status of Internet users as opposed to large numbers of ethnic
minorities who were largely unaware of Internet technology. The authors report
that Internet interest is stimulated greatly by social and professional networks
and that these also provide a means of support for users. Internet users are
motivated to participate for social-personal development, but nonusers believe
the value of the Internet to be in business/commercial use. The survey also
determined barriers that people overall felt it difficult to get over when starting
with the Internet, even those with technical experience. In addition, other
barriers to use include the cost of Internet service and the complexity of use.
In an article that explores the digital divide as a phenomenon regarding more
than demographics, Adams (2001) discusses the divide as a “complex web of
interconnected issues, with its base in structural and social inequalities that have
long been present in this country and world” (p. 6). The author’s argument is
that by allowing quantitative representations of the digital divide to explain their
existence, the actual questions of “who” and “why” the divide is affecting is
going unanswered. The presence of discrimination against minorities and the
underprivileged has increased the level of difficulty in obtaining both the social
and economic means by which to access and benefit from the Internet.
According to Adams (2001), the benefits of using computers and the Internet
are “access to job training and openings, the ability to strengthen social
networks through email, productivity benefits, participation in democratic and
political life, and access to an increasing pool of raw information and knowl-
272 Morgan & Trauth
edge” (p. 6). Also as awareness about the digital divide grows, we should not
assume that all individuals interact with technology on the same level. In order
to function within the digital community, there is an underlying knowledge that
who the user is, where he/she comes from, and his/her exposure to technology
will affect the way he/she will interact with different technologies.
Kraut, Lundmark, Patterson, Kiesler, Mukopahyay, and Scherlis (1998)
studied the effect of the Internet on social and psychological behavior. This
longitudinal research looked at whether the Internet increased or decreased
social involvement. Decreased social involvement is thought to lead to
disenfranchisement by citizens, thus allowing crime and political unrest to enter
in to the community. The results of the conducted research support the notion
that the Internet adversely affects “social involvement and psychological well-
being” (p. 11). The study affirms that an increase in the use of the Internet is
connected to a minor decline in social involvement and a more prevalent feeling
of loneliness. The measure of social involvement came from observations of
family communication and the size of the person’s social network.
DiMaggio and Hargattai (2001) discuss different types of prevalent inequality
present in the information age. The authors argue that the digital divide
represents more than a binary argument regarding those with and without the
ability to access technology. The dimension in which digital inequality exists
consists of five areas. Inequality of technical apparatus relates to availability of
adequate computing equipment, software, and means of connectivity to the
Internet. Inequality in the autonomy of use regards the location of Internet
access and its affect on the users’ ability to use the Internet for their desired
activities. Inequality in skill refers to the ability of users to utilize the Internet in
an effective way and for the desired outcome. Inequality of available social
support includes the other people that the users may interact with to help and
encourage them in the use of the medium. Inequality in variety of use encom-
passes the influencing factors such as education and income on the understand-
ing of the opportunities for use of the Internet.
Hargattai (2003) advances digital divide research by placing a focus on the
differences in the online skill of users. By doing so, the author expands the scope
of the digital divide past the “haves” and “have-nots.” By exploring the
differences among users in ability to perform online tasks, there will be a better
understanding of an additional potential barrier to equity in the use of technol-
ogy. The results of the study show that variation exists among the users of the
Internet in their ability to find content. Age showed to be an important factor
in the research as younger users navigated online content with a much greater
use than older users. The amount of time a user spends on the Web is also
connected to better navigation of content. This study supports the notion that
physical access alone cannot ensure participation in Internet technology.
The digital divide is a situation that is very complex and requires an in-depth
analysis of the many ways in which it is perceived. Not only are there constraints
in obtaining the necessary physical elements online, but there are also con-
straints in social elements. Users’ skills and ability with the Internet also play a
role in their level of participation. It is then important to understand the factors
that influence the individual user’s skill level and behavior with technology. The
theory of individual differences can provide a useful lens in investigating this
situation.
Individual Differences
The role of individual differences in information systems research is a growing

but very important area. Individual differences have been applied in several
disciplines and have been characterized in a variety of categories. Research in
individual differences has been largely carried out to better understand the
behavior of people in organizations or with technical applications. These types
of studies are carried out at the individual level of analysis as opposed to the
group or organizational levels. By studying the individual and his/her behavior
or interaction with different types of technology, researchers can make recom-
mendations for the improvement of systems based on different user character-
istics. Many measures of individual differences have been utilized in research,
such as personality, cognitive style, cognitive complexity, need for cognition,
problem-solving ability, demographic differences, self-esteem, anxiety, and
perceived usefulness.
Two very important measures of individual differences that will be used in the
evaluation of users in the study are cognitive complexity and perceived
usefulness. Cognitive complexity relates to the degree of differentiation with
which an individual views the world (Ford, Miller, & Moss, 2001). Perceived
usefulness is the degree to which a person believes that using a particular system
would enhance the task he/she is trying to complete. These measures are
important characteristics of users especially in relation to searching behavior.
This is due to cognitive complexity being a measure of a user’s perceptual
difference or frame of reference to activity. Users approach the task of
searching for information with a worldview that influences their search behav-
274 Morgan & Trauth
ior. Perceived usefulness takes into account the value of systems in accomplish-
ing their goals.
A recent study conducted in the area of individual differences and Internet
searching was completed by Ford, Miller, and Moss in 2001. The authors
recognize that increased access to the Internet is placing an emphasis on
information-seeking skills to a more diverse body of users. This study utilized
a positivist approach and gathered mainly quantitative statistical data. In
addition, the authors chose not to use hypothesis testing but instead to aim for
data exploration by way of factor and regression analyses to identify relation-
ships between factors of individual differences and retrieval effectiveness. The
dimensions of individual differences being researched are age, gender, cogni-
tive style, levels of prior experience, Internet perceptions, and study ap-
proaches. The authors found that effectiveness of retrieval was linked to males,
low cognitive complexity, an imager cognitive style, and several Internet
perceptions and study approaches that correlated to low self-esteem of users.
In addition, the theory of individual differences has been largely applied in the
context of gender and cognitive/information processing and personality. Trauth
is presently working toward a theory of individual differences with respect to
gender in the information technology (IT) workforce (Morgan et al., 2004;
Quesenberry et al., 2004; Trauth et al., 2004, 2005). This theory seeks to
address underlying reasons for the underrepresentation of women in the IT
industry. This emergent theory focuses on the differences within instead of
between genders and acknowledges that women respond in a variety of ways
to external and internal influences. This theory also supports the notion that
“both gender and IT are socially constructed at the individual level … [and that]
… women as individuals, experience a range of different socio-cultural
influences which shape their inclinations to participate in the IT profession in a
variety of individual ways” (Trauth et al., 2004, p. 115).
Internet users are a diverse group of individuals. No one type of user can be
identified as the benchmark for developing universal systems. Many differences
exist among types of users such as race, education level, socioeconomic status,
cognitive style, nationality, and so forth. The theory of individual differences
reflects and acknowledges this diversity.
The world is a diverse place, so as Internet use becomes increasingly integrated
into daily life, the more emphasis will be placed on usability of the technology.
Accordingly, it is important that the variety of people accessing the Web be able
to effectively use search engines. Importantly, the research shows that models
of human information behavior are not always implemented into systems that
are built to assist searchers in finding content. Therefore, it is possible that there
is a mismatch in the system design of Web search engines with the user model
of searching, or there could be other important factors that influence an
individual’s ability to find information and search successfully. So what is
currently not addressed in the literature is the effect of individual differences on
use and performance with Internet technology. Additionally, the connection is
not evident about the role of search engines in inhibiting or facilitating continued
use of the Internet.
Issues
Studying user behavior and the Internet can be accomplished in many ways.
The implication of doing so serves a number of purposes. First, by evaluating
the way that users interact with Internet technology, we can understand its utility
and how it is able to serve user needs. We can also observe the effect that
human interaction with technology has on patterns of use, learning, and
performance. Based on these ideas, the study of individual differences is critical
to identifying influences of these different types of user responses and behavior.
The theory of individual differences takes into account that people makes
choices in their environment based upon personal experience and external
influences. In the study of gender and IT, it has been used to illustrate that all
people of the same gender do not necessarily exhibit a single response in
behavior or choice in the IT workforce (Morgan et al., 2004; Quesenberry et
al., 2004; Trauth et al., 2004, 2005). This theory is an appropriate basis for
research on Web search skill and use, clearly because all users of this
technology do not exhibit a single type of behavior nor do they perform at the
same level. So investigation into individual differences can provide an explana-
tion of the influences on user behavior and performance that could inform the
design of improved systems. By extending and empirically testing the theory of
individual differences, much can be learned about users in a way that could
significantly affect performance issues by developing systems better suited to
their behavior or learning style.
Current research into user behavior would benefit from the exploration of this
theory because it would provide insight into how an individual’s collective
experience affects his/her performance with technology. Instead of viewing
users as one dimensional beings, the study of users from the perspective of
276 Morgan & Trauth
individual differences takes into account demographics and other personal

attributes. By using individual differences as a guiding theory, the study of the
user could provide a more complete view of individuals’ embodied behavior,
not just a slice from their observable actions.
The theoretical stance of individual differences supports the notion that people
are unique individuals that will behave and interact with the environment based
upon that fact. This theory rejects notions of behavior based upon static
biological characteristics. For example, the thought that women and men, as
groups, behave differently in respect to technology, would not hold with respect
to this theory. Based upon prior research conducted by Trauth et al. (2004,
2005; Morgan et al., 2004; Quesenberry et al., 2004), it is documented that
women display numerous types of behavior in respect to their positions in the
IT workforce. Based upon this theory, generalizations about men and women
do not adequately represent the diversity of the experiences that occurs within
or between each group.
The study of individual differences is important because the further develop-
ment of technology and more specifically, the Internet should be based upon the
actual study of user interactions. Using this theory as a basis for research into
user behavior supports a notion of user diversity. It accounts for important
characteristics that may influence how a user interacts with his/her environment.
In addition, it highlights an important area of interest for those engaged in the
development of systems. It illustrates that a number of attributes about a user
may influence their interaction with IT, and therefore it is necessary to gather
rich information about the users for whom the system is being developed. It may
also argue that increased levels of user personalization in systems can have a
positive impact on user behavior and interaction with technology.
Solutions and Recommendations
In an effort to encourage the incorporation of the theory of individual differ-

ences into the study of user behavior and Internet technology, the following
research program outlines our approach to the study of the phenomenon. One
goal of this research agenda is to add to digital divide research by investigating
another possible facilitator or inhibitor of technology use. In addition, because
there is evidence of a disconnect between current models of user searching
behavior and actual search engines, we seek to inform design based on the
perspective of individual differences. Therefore, research into individual be-

havior with technology has the possibility to enlighten the development process
of search engines. We also seek to support and extend individual differences
research and theory by documenting its role in human information behavior.
Research Agenda
The individual differences theory will be the theoretical basis for this research.
In addition, we will also work to expand the theory into the realm of Web
searching and information retrieval. We believe that Web searching is largely
influenced by individual differences. In addition, the theory is applicable to this
study due to the possibility of sociocultural influences affecting people’s
perception of Internet technology and their attitudes toward Web searching.
The Internet has recently been described as being essential to society in a
variety of ways (Hoffman, Novak, & Venkatesh, 2004). It is a vehicle for the
creation and searching of information as well as being a mechanism for
communication. So in the wake of this “Internet age,” it is important to
understand how the technology can be more usable and beneficial to all people.
Currently, due to the inequity of people with the ability to participate in Internet
technology, a digital divide exists among users and nonusers. Although a great
deal of research has been carried out to address causes for the problem, there
has not been a definitive reason identified. In an effort to continue articulation
into the problem of the digital divide, more research must delve into factors that
influence the motivation or inhibition of Internet use.
We seek to address this need by researching the role of individual differences
in online searching skill and performance. Research has identified that “knowl-
edge of more fundamental and enduring factors that can help us improve
people’s internet retrieval in deep and lasting as opposed to relatively super-
ficial and fleeting ways” is necessary (Ford et al., 2001, p. 1049). In addition,
the investigation into user diversity with interactive systems has also been
deemed important to the research community (Borgman, 1987).
The significance of this type of study would provide insight into how users
actually search for information and the influences, experiences, and perceptions
that shape their attitudes and behavior with Web search engines. This research
seeks to further support and expand the theory of individual differences being
developed by Trauth. This study would identify that the theory of individual
278 Morgan & Trauth
differences is applicable to both genders and is relevant to explaining both the

underrepresentation of women in IT and differences in Web search skills. In
addition, the results of the study could provide critical design information for
creators and administrators of Web search engines. In the event that the results
do show a connection between individual differences and online Web searching
skill, a case can be supported for more concentrated research effort into the
personalization and customization of Web search engines. Also, this research
may move the research community closer to finding a theory or framework that
links the concept of individual differences to information retrieval. This would
be an important contribution because without a guiding theory, the subject area
will “continue to proceed as in the past using a shotgun approach” (Saracevic,
1991, p. 85). Another important contribution of this research would be the
articulation of an additional obstacle to user participation with the Internet,
which also supports further explanation of the problem of the digital divide.
Future Trends
This chapter supports the mission of this book by drawing attention to the
importance of understanding individual characteristics with respect to technol-
ogy. We support the notion that systems should be developed based on
cognitive-, behavioral-, and performance-based user models. Due to the
increased interactivity and sophistication of systems, understanding users at a
superficial level will be insufficient to design effective usable systems. Research
into individual differences emphasizes user-based design and development
methodologies by stressing the uniqueness of individual users.
Future research into this area could explore a variety of areas. The study of
individual differences could be expanded from a micro level look at Web search
engine use and performance to overall use of technology. Also, this study could
be replicated in different international locations to explore how individual
differences may be affected by national culture or environment. The theory of
individual differences could also be used to evaluate other user behavior in
relation to Internet technology in order to develop a rich user model for the
development of other Web-based systems.
Conclusion
Users of technology interact with it in a number of ways, especially in their

search for information. When looking to solve a problem or answer a question,
users find different methods of formulating that question in the context of their
own knowledge. Based on their knowledge, users look for a way to find the
information they need in the environment of the Internet. However, the
arrangement of information systems is not always based on the way that
different individuals organize their search for information. This makes the task
of information searching very complex. However, we know that users identify
a way to find information in their environment. To better understand the factors
that influence different levels of performance with the Internet, the study of
individual differences can provide critical insight into users. By incorporating
the theory of individual differences into research of online searching skill and
performance, we can see what impact that situation has on use of the Internet.
This research program looks to extend research on individual differences and
information retrieval. In addition, by exploring patterns of usage and perfor-
mance, we can also address the impact of search engine use on overall
technology participation. This may well help to address and provide an
additional barrier or motivator to Internet use.
References
Adams, A. R. (2001). Introduction: Beyond numbers and demographics:

Experience-near explorations of the digital divide. Computers and
Society, 31(3).
Arasu, A., Cho, J., Garcia-Molina, H., Paepcke, A., & Raghavan, S. (2001).
Searching the Web. ACM Transactions on Internet Technology, 1(1),
2–43.
Bates, M. J. (1989). The design of browsing and berrypicking techniques for
the online search interface. Online Review, 13, 407–424.
Belkin, N., Cool, C., Stein, A., & Theil, U. (1995). Cases, scripts, and
information seeking strategies: On the design of interactive information
retrieval systems. Expert Systems With Applications, 9(3), 379–395.
280 Morgan & Trauth
Belkin, N., Oddy, R., & Brooks, H. (1982). Ask for information retrieval, part
1. Journal of Documentation, 38(2), 61–71, 145–164.
Borgman, C. (1996). Why are online catalogs still hard to use? Journal of the
American Society for Information Science, 47(7), 493–503.
DiMaggio, P., & Hargittai, E. (2001). From the digital divide to digital
inequality: Studying the Internet use as penetration increases (Work-
ing Paper Series, Number 15). Princeton, NJ: Princeton University
Center for Arts and Cultural Policy Studies.
Eastman, C. M., & Jansen, B. J. (2003). Coverage, ranking, and relevance: A
study of the impact of query operators on search engine results. ACM
Transactions on Information Systems, 21(4), 383–411.
Ford, N., Miller, D., & Moss, N. (2001). The role of individual differences in
Internet searching: An empirical study. Journal of the American Society
for Information Science.
Gauch, S., & Smith, J. (1993). An expert system for automatic query
reformulation. Journal of the American Society for Information Sci-
ence, 44(3), 124–136.
Hargittai, E. (2003). Second-level digital divide: Differences in people’s online
skills. First Monday, 7(4).
Hoffman, D., Novak, T., & Venkatesh, A. (2004). Has the Internet become
indispensable? Communications of the ACM, 47(7), 37–42.
Hoffman, D. L., & Novak, T. P. (1998). Bridging the racial divide on the
Internet. Science, 280, 390–391.
Introna, L., & Nissenbaum, H. (2000). Shaping the Web: Why the politics of
search engines matters. The Information Society, 16(3), 1–17.
Jansen, B. J., & Pooch, U. (2001). A review of Web searching studies and a
framework for future research. Journal of the American Society for
Information Science and Technology, 52(3), 235–246.
Katz, J., & Aspden, P. (1997). Motivations for and barriers to Internet usage:
Results of a national public opinion survey. Internet Research, 7(3).
Katz, J., & Rice, R. (2002). Social consequences of Internet use. Cam-
bridge, MA: MIT Press.
Kling, R. (1998). Technological and social access on computing, informa-
tion and communication technologies (A White Paper for the Presi-
dential Advisory Committee on High-Performance Computing and Com-
munications, Information Technology, and the Next Generation Internet).
Kraut, R., Lundmark, V., Patterson, M., Kiesler, S., Mukopahyay, T., &
Scherlis, W. (1998). Internet paradox: A social technology that reduces
social involvement and psychological well-being? American Psycholo-
gist, 53(9), 1017–1031.
Kvasny, L., & Truex, D. (2000). Information technology and the cultural
reproduction of social order: A research program. In R. Baskerville, J.
Stage, & J. DeGross (Eds.), Organizational and social perspectives on
information technology (pp. 277–294). New York: Kluwer Academic
Publishers.
Morgan, A. J., Quesenberry, J. L., & Trauth, E. M. (2004, August). Exploring
social networks in the IT workforce: Experiences with the Boys’ Club.
Proceedings of the Tenth Annual Americas Conference on Informa-
tion Systems, New York.
National Telecommunications & Information Administration (NTIA). (1995).
Falling through the Net: A survey of the “have nots” in rural and
urban America.
National Telecommunications & Information Administration (NTIA). (2000).
Falling through the Net: Toward digital inclusion.
Quesenberry, J. L., Morgan, A. J., & Trauth, E. M. (2004, May). Understand-
ing the “mommy tracks”: A framework for analyzing work-family issues
in the IT workforce. Proceedings of the Information Resources Man-
agement Association Conference, New Orleans, LA.
Saracevic, T. (1991). Individual differences in organizing, searching, and
retrieving information. Proceedings of the American Society for Infor-
mation Science, 28, 82–86.
Saracevic, T., & Kantor, P. (1988a). A study of information seeking and
retrieving. II. Users, questions and effectiveness. Journal of the Ameri-
can Society for Information Science, 39(3), 177–196.
Saracevic, T., & Kantor, P. (1988b). A study of information seeking and
retrieving. III. Searchers, searches, and overlap. Journal of the Ameri-
can Society for Information Science, 39(3), 197–216.
Trauth, E. M., Quesenberry, J. L. & Morgan, A. J. (2004). Understanding the
under representation of women in IT: Toward a theory of individual
differences. Proceedings of the ACM SIGMIS Computer Personnel
Research Conference, Tucson, AZ.
282 Morgan & Trauth
Trauth, E. M., Quesenberry, J. L., & Yeo, B. (2005). The influence of

environmental context on women in the IT workforce. Proceedings of
the ACM SIGMIS Conference, Atlanta, GA.
Bayesian Networks 283
Chapter XIII
Using
Bayesian Networks for
Student Modeling
Chao-Lin Liu, National Chengchi University, Taiwan
Abstract
This chapter purveys an account of Bayesian networks-related technologies

for modeling students in intelligent tutoring systems. Uncertainty exists
ubiquitously when we infer students’ internal status, for example, learning
needs and emotion, from their external behavior, for example, responses
to test items and explorative actions. Bayesian networks offer a
mathematically sound mechanism for representing and reasoning about
students under uncertainty. This chapter consists of five sections, and
commences with a brief overview of intelligent tutoring systems,
emphasizing the needs for uncertain reasoning. A succinct survey of
Bayesian networks for student modeling is provided in Bayesian Networks,
and we go through an example of applying Bayesian networks and mutual
284 Liu
information to item selection in computerized adaptive testing in

Applications to Student Models. We then touch upon influence diagrams
and dynamic Bayesian networks for educational applications in More
Graphical Models, and wrap up the chapter with an outlook and discussion
for this research direction.
Computer-Assisted Learning
In the past couple of decades, both the research literature and the real world
have seen flourishing studies and applications of computer-assisted learning.
(We use computer-assisted learning to refer to computer-assisted instruc-
tion as well.) The increasing capabilities and decreasing prices of personal
computers have created an affordable environment for individualized com-
puter-assisted learning. The explosive expansion of the Internet not only
provides a rich source of information but also nourishes the studies and
applications of Web-based learning systems. To give a few examples, Conati,
Gertner, and VanLehn (2002) have studied computer-assisted learning of
Newtonian Physics; Mislevy and Gitomer (1996) have investigated the tech-
niques for computer-assisted learning of the troubleshooting of hydraulics
systems in aircraft; Mitrovic, Martin, and Mayo (2002) have designed systems
for teaching the SQL database language; Horvitz, Breese, Heckerman, Hovel,
and Rommelse (1998) look into the possibilities of assisting users of Microsoft
Excel with software agents; Anderson et al. have developed a system for
learning LISP programming (Anderson, Boyle, Corbett, & Lewis, 1990) and
high school mathematics (Anderson, Douglass, & Qin, 2004); Virvou, Maras,
and Tsiriga (2000) construct systems for assisting the learning of the passive
voice in English; and Brusilovsky et al. have discussed issues such as course
sequencing (Brusilovsky & Vassileva, 2003) for Web-based education
(Brusilovsky, Schwarz, & Weber, 1996).
Student Modeling for Computer-Assisted Learning
In order to build the software infrastructure that supports computer-assisted

learning, researchers are employing techniques for modeling important partici-
pants in the learning process. Major participants in learning activities include
students, instructors, and the targeted learning/teaching objects. Researchers

have studied issues ranging from course material preparation to the eventual
course material delivery to students, and have taken diverse approaches to
modeling and decomposing the whole learning process. For instance, Virvou
and Moundridou (2001) consider student and instructor models in their design
of authoring tools; Mislevy, Steingerg, and Almond (2003) set up student, task,
and evidence models for the assessment task; and Brusilovsky et al. (1996)
places emphasis on making intelligent tutoring systems (ITSs) available on the
Web.
Despite the differences in the ITS architectures, a significant portion of the
literature focuses on student modeling. This is barely surprising. Students are
the major, if not the most important, users of any computer-assisted learning
systems, so inferring about the students for choosing appropriate learning
activities is a key issue for system designers. Since learning is a cognitive
process, modeling students’ learning process will have to explicitly or implicitly
construct some artificially assembled images of the students’ cognition process
(Anderson et al., 1990). For instance, Anderson et al. (1990, 1992, 2004)
base much of their work on theories of cognition, and by working on computer-
assisted learning systems, they examine and refine such cognitive theories as
ACT* and ACT-R at the same time. They show how students apply and learn
knowledge with performance and learning models, respectively. Their
systems then apply the model and knowledge tracing techniques in monitoring
students’ progresses in the problem domains. Similarly, Brusilovsky (1994)
proposes overlay student, error, and genetic models for demonstrating
different aspects of student performance. Overlay student models record
students’ competence in different areas of the learning targets, error models
provide an explanation of why students make mistakes, and genetic models
capture student performances in executing procedural knowledge.
Besides comparing student models based on how theories of cognition are
employed, we find previous approaches differ in how they link students’
external performance with their internal state of knowledge. On the one hand,
we can assume that there is a one-to-one deterministic relationship, and build
student models with either propositional or first-order logics. From here, it is
natural to treat the problems of diagnosing students’ deficiency in knowledge
as the problems of identifying buggy designs in electronic systems (de Kleer &
Williams, 1987). For instance, Kono, Ikeda, and Mizoguchi (1994) build their
work on de Kleer’s truth maintenance system (de Kleer, 1986). Such logic-
based approaches face the challenges that students may not perform consis-
286 Liu
tently over time and under different contexts, and researchers have to employ
such techniques as nonmonotonic reasoning in their systems. On the other hand,
we can accept that inconsistency is part of nature, and adopt reasoning
mechanisms that explicitly capture and reason about the uncertainties in
applications.
Sources of Uncertainties in Modeling Students
Uncertainties in student modeling come from several sources. The most

commonly cited are slips and lucky guesses (cf. Martin & VanLehn, 1995;
Millán & Pérez-de-la-Cruz, 2002; Mislevy et al., 1996; Reye, 2004). Let us
assume that students are either competent or incompetent in some concepts
and that we rely on test items to gauge students’ competences. We can identify
students who fail in test items that the students should be able to respond to
correctly. Analogously, we can identify situations when students respond to test
items correctly by guesswork, although the students do not have sufficient
knowledge to know the answers. Researchers call these phenomena slips and
lucky guesses, respectively. An interesting source of such uncertainties comes
from the nature of human behavior. After students have learned the correct
concepts, they might commit the same errors again. Although such a recurrence
of errors is less common in machines, it is quite conceivable in human behavior
(Sison & Shimura, 1998). As a consequence, we cannot tell for sure whether
students are really competent in a particular concept, given that they may
respond to test items correctly or incorrectly.
More generally, uncertainties exist when we are interested in knowing factors
that are not directly observable. For instance, even if the relationship between
being competent in some concepts and responding correctly to test items is
deterministic, we might not be able to exactly judge the students’ competences
in some situations. When there are multiple ways to solve the test items, we
cannot tell for sure which way the students have applied to solve the problems
(Martin & VanLehn, 1995).
More challenges come about in computer-assisted learning environments. In an
open learning environment, where students are allowed to explore topics of
their choices, it is hard to tell the real purposes of students’ activities (Bunt &
Conati, 2003). In an interactive environment, it is possible that students make
deliberate errors to see how the machines will react (Reye, 2004). Hence, we
may not want to jump to a specific conclusion after observing students’
mistakes.
The uncertain relationships between students’ responses to test items and

students’ competence in learning targets make student modeling a challenging
task. Had the relationship been deterministic, we could infer that a student is
incompetent in a concept when we observe that s/he failed to answer a test item
correctly. This would allow us to employ logics-based methods for student
modeling. In fact, the relationships are not deterministic, and logic-based
reasoning is not appropriate. Therefore, we would have to resort to other
approaches that consider uncertain relationships between the observed and
unobservable variables. In the remainder of this chapter, we introduce Baye-
sian networks for this task of reasoning under uncertainty.
Bayesian Networks
The aforementioned types of uncertainties prompt researchers to employ

appropriate techniques for student modeling. More than 10 years ago, re-
searchers began to employ Bayesian networks (Pearl, 1988) for student
modeling (cf. Villano, 1992).
Capturing and Reasoning About the Uncertainties
Bayesian networks (BNs) are directed acyclic graphs that consist of nodes and
arcs (Pearl, 1988). Nodes represent random variables, and arcs qualitatively
denote direct dependence relationships between the connected nodes. Each
node can take on its value from a set of possible values, and has an associated
conditional probability table that quantitatively specifies the degrees of mutual
dependence between the connected nodes. Nodes at the tails of arcs are parent
nodes of the nodes that are pointed by arrows. A node without parent nodes
should have a table of prior probabilities. A Bayesian network indirectly
encodes the joint probability distribution of the random variables in the
network, so we can compute any conditional probabilities of interest given a
Bayesian network that is constructed for the target application.
We illustrate the basic concepts of Bayesian networks with the examples
discussed in Liu (2005). Let C be the random variable that encodes the degree
of the mastery of a concept, and X the random variable representing the
outcomes of using an item for testing the mastery of C. Assume that variables
288 Liu
Figure 1. Some very basic Bayesian networks for assessing students’

competence in concepts
C A B C
C X
(a) ( b) Y X (c ) Y X
for both concepts and items are dichotomous. A variable for a concept takes
the value of either good or bad, and a variable for the response to an item takes
the value of either correct or incorrect. For simplicity of notation, we use a
small letter of the variable to denote the “positive” value of a random variable,
and a small letter with a bar to denote the “negative” value of the variable. For
instance, Pr( x | c ) denotes Pr( X = correct | C = bad). We use the symbol PR and
capital names of random variables to denote the probability values of all
possible combinations of the values of the involved random variables. For
instance, PR( X | C ) denotes {Pr( x | c), Pr( x | c), Pr( x | c ), Pr( x | c )}.
We can use the very simple Bayesian network shown in Figure 1(a) to represent
that X is a test item for assessing students’ competence in C. Node C will
include a table that contains the prior distribution for the random variable C, and
node X will have a table that contains the probability distribution for X
conditional on the values of C. The contents of PR( X | C ) will capture the
probabilities of making slips and lucky guesses when students respond to X. Let
Y denote the response to another test item. The network shown in Figure 1(b)
suggests that we have an extra item for measuring the competence in C. Again,
the contents of PR(Y | C ) contain the probabilities of making slips and lucky
guesses when students respond to Y. The network shown in Figure 1(c)
expands Figure 1(a) in another direction. We add a node for concept B that is
directly related to the competence in C, and yet another node for concept A that
is directly related to the competence in B. The contents of PR(C | B) dictate how
the competence in B will influence the competence in C
A Bayesian network indirectly specifies the joint probability distribution of the
random variables, so we can compute any conditional probabilities that involve
variables in the network. The very simple network shown in Figure 1(a)
indirectly encodes the joint distribution of C and X, and we can compute the
conditional probabilities of interest, for example, PR(C | X ). Researchers have
invented different algorithms for computing conditional probabilities with

Bayesian networks (Jensen, 2001), and there are available commercial and
free software tools (http://www.cs.ubc.ca/~murphyk/Bayes/bnsoft.html).
However, before we explore other aspects of applying Bayesian networks for
student modeling, we must keep in mind that Cooper (1990) has proven that
computing exact probability distributions with Bayesian networks is an NP
hard task. Moreover, Dagum and Luby (1993) have proven that computing
approximate probabilities with Bayesian networks are also NP hard. Conse-
quently, computational costs can become a design concern for systems that
require complex networks (e.g., Martin & VanLehn, 1995; Mayo & Mitrovic,
2000), and Collins, Greer, and Huang (1996) and Reye (2004) discuss the
applications of special classes of Bayesian networks for student modeling.
A Running Example
Liu (2005) applies the network shown in Figure 2 for adaptive item selection
in student classification. The group node is multiple valued, and represents the
types of student groups. Nodes named cX represent basic concepts, and nodes
named dY represent composite concepts. Values of nodes that represent
concepts are either good or bad. All nodes named iZ represent responses to
test items, and the responses can be either correct or incorrect. For instance,
dAB is a composite concept whose prerequisites include cA and cB, and iA1,
iA2, and iA3 represent students’ responses to test items that are designed for
evaluating students’ competence in cA.
Links from concept nodes, for example, nodes named cX and dY, to item
nodes, for example, those named iZ, indicate the competence levels in
concepts influence the probability of answering correctly to test items. Links
from basic concepts to composite concepts indicate that understanding com-
Figure 2. A Bayesian network designed for item selection (Liu, 2005)

iB 1
iA 1 group iB 2
iC 1
iA 2 iB 3
iC 2
iA 3 cA cB cC iC 3
dA B d BC dA C d A BC
iA B 1 iA B2 iA B3 iA C 1 iA C 2 iA C 3
iBC 1 iB C 2 iB C 3 iA B C 1 iA B C 2 iA B C 3
290 Liu
posite concepts requires the knowledge of basic concepts. Links from the
group node to nodes that represent basic concepts indicate that different
student types may have different abilities in the basic concepts. Links from the
group node to the nodes that represent composite concepts indicate that
different student types may have different abilities in integrating basic concepts
to form knowledge about the composite concept.
The contents of the conditional probability tables quantify the strength of the
dependence between nodes that are directly connected. For instance, Pr(ia1 | ca)
and Pr(ia1 | ca) are, respectively, the chances of making slips and lucky guesses
for the test item iA1. Pr(cb | g i ) is the probability that the i-th type of student is
good at concept cB, and Pr(dabc | ca, cb, cc, g i ) is the probability that the i-th type
of student who is also good at cA, cB, and cC can actually learn the composite
concept dABC.
Given such a Bayesian network, we can conduct predictive and diagnostic
inferences. We predict the chances of correctly responding to a particular test
item given that we know a student’s type. In contrast, we compute the
probability distribution over the possible student types given the correctness of
students’ responses to some test items, and may choose to guess the students’
types based on the probability distribution over group. In fact, taking advan-
tage of the Bayesian network, we can compute the conditional mutual informa-
tion between any unadministered test items with the group node for selecting
the most effective test items which, hopefully, allow us to determine a student’s
type with as few test items as possible (Liu, 2005).
Learning Bayesian Networks
Obtaining appropriate Bayesian networks for the target problems is a challenge

that one must cope with. Researchers such as Heckerman (1999) have done
quite a lot of work in learning Bayesian networks from data, and Chickering
(1996) shows that learning Bayesian networks is NP complete. Practically, the
task of learning a Bayesian network has been split into two stages: parameter
learning and structure learning (Neapolitan, 2004), although learning good
structures and their parameters are not independent.
Based on domain knowledge in education and psychology, experts can directly
provide structures for the problem domains. For more complex applications,
one provides building blocks that model subparts of the whole problem, and the
building blocks will be used to dynamically construct a complete network on

the fly. This is the so-called knowledge-based model construction approach
(Breese, Goldman, & Wellman, 1994). Mislevy, Almond, Yan, and Steinberg
(1999) apply Markov Chain Monte Carlo methods to obtain parameters for a
given network. Mayo and Mitrovic (2001) initialize Bayesian networks with
data collected from a population of students, and adapt the conditional
probabilities for individual students with algorithms discussed in Heckerman
(1999). It is also possible that system designers employ heuristics for choosing
appropriate parameters from a specific set of choices (Conati & VanLehn,
2001).
Relatively, choosing a structure for the problem domain is less straightforward
than fitting the joint probability distribution encoded in a Bayesian network to
the observed data. On the one hand, we would like to have networks that reflect
the dependence relationships among all variables involved in the applications.
On the other hand, we also have to consider the computational costs of
evaluating Bayesian networks and whether we really can obtain sufficient
training data that will be used to train the networks for their conditional
probability tables. Mayo and Mitrovic (2001) classify different channels of
obtaining student models into three types: expert-centric, efficiency-centric,
and data-centric. In the expert-centric approach (e.g., Conati et al., 2002;
Mislevy & Gitomer, 1996), experts create student models based on domain
knowledge, but have to employ heuristics for obtaining the required conditional
probability tables. In the efficiency-centric approach (e.g., Reye, 2004; Mayo
& Mitrovic, 2000), people employ relatively simple networks to track stu-
dents’ competence over time. In the data-centric approach, both structures
and conditional probability tables are induced from training data.
Millán and Pérez-de-la-Cruz (2002) discuss directions of arcs for aggregation
relationships and evidence–knowledge relationships. Different choices of arc
directions imply different relationships among the variables, and will also
influence the difficulties confronted in model construction. Assume a meaningful
aggregation, denoted C , of a set of items, denoted P1, P2, ..., Pn. How do we
model the relationships between the basic items and the aggregated cluster? (For
ITSs, the basic items could represent the prerequisites of an advanced course.)
Conati et al. (2002) employ the structure shown in Figure 3(a) for a network that
mimics the structure of the domain knowledge, but Mislevy and Gitomer (1996)
adopt deductive reasoning and the structure shown in Figure 3(b).
Analogously, for inferring students’ competence in concepts, assume that we
have three test items, I1, I2, and I3 and two concepts, C1 and C2. If responding
292 Liu
Figure 3. Alternative structures for nodes of an advanced course and its

prerequisites
P1 P2 ... Pn C
C P1 P2 ... Pn
(a) (b)
correctly to and respectively, requires the knowledge of only both and and
only What is the structure that we should use to model this relationship? In this
case, the structure shown in Figure 4(b) will be more natural and appropriate,
and many projects adopt this choice (e.g., Guzmán & Conejo, 2004; Millán &
Pérez-de-la-Cruz, 2002). In practice, if a student responds incorrectly to this
would decrease the possibility that this student will respond to and correctly.
This intuition is implied by the structure shown in Figure 4(b) but not so by the
one shown in Figure 4(a). When we have no direct evidence about both and
nodes and remain independent of in Figure 4(a), due to d-separation (Pearl,
1988).
Despite the advances in computational technologies, it appears that it is very
challenging, if not impossible, to create a computer program that fully automati-
cally infers a convincing structure for a specialty domain. As pointed out by an
anonymous reviewer for this chapter, “there can be wide disagreement on what
constitutes a prerequisite for concept.” Consider systems that rely on models
that are provided by human experts on one extreme of the spectrum and
systems that rely on models that are fully automatically learned from data on the
Figure 4. Alternative structures for nodes representing concepts and test

items
C1 C2
I1 I2 I3
C1 C2 I1 I2 I3
(a) (b)
other extreme. A middle ground between these two extremes is to allow experts
to provide fragments of a complete network and let computer programs to
dynamically construct the complete network at run time. Before machines can
fully assist the task of model construction, an appropriate intervention of
experts may be required. Conati et al. (2002) apply Bayesian networks as the
infrastructure for assisting real students to learn Newtonian Physics. They
employ a set of fragments of network for domain-general knowledge and
another set for task-specific knowledge. At run time, the tutoring system
employs fragments from these two sets to compose networks that are tailored
for individual students.
A Note on the Semantics of Arc Directions
Semantics of arc directions is the key issue in our discussion of the structures
shown in Figures 3 and 4. It is tempting to interpret directions in Bayesian
networks as causal directions. This is evident in that many, including the
network shown in Figure 2, choose to implement their systems with structures
that are similar to those shown in Figures 3(a) and 4(b). In addition, typical
textbooks on artificial intelligence also advise that constructing Bayesian
networks based on the causal directions among random variables are more
likely to lead to concise networks (Russell & Norvig, 2003).
Despite these facts, arc directions in Bayesian networks do not have to follow
causal relationships among propositions which are represented by the nodes.
More technically speaking, Bayesian networks are independency maps rather
than dependency maps, meaning that lacking a direct link between nodes in
Bayesian networks only implies conditional independence and the directly
connected nodes might in fact be independent (Pearl, 1988). Roughly speak-
ing, links in Bayesian networks are more about conditional correlations
between random variables. Hence the choices between the networks shown in
Figure 3(a) and 3(b) can be subjective, and both can work for the applications
as long as the underlying distributions fit the real data.
Let C represent the competence in a particular concept, and X and Y be two
test items for assessing the mastery in C. We can apply any of the network
structures shown in Figure 5 to carry the joint probability distribution of these
three random variables. In fact, it is possible that we reverse directions of arcs
in Bayesian networks without changing the underlying joint probability distri-
butions, and we can transform the leftmost network shown in Figure 5 gradually
294 Liu
Figure 5. Alternative structures for concept C and test items X and Y
C C C
Y X Y X Y X
to the middle network and to the rightmost network with Shachter’s algorithm
(1986). The directions of arcs do not necessarily carry causal implications in
these alternative networks.
Arcs of Bayesian networks as they are defined in most literature (e.g., Jensen,
2001; Pearl, 1988), encode correlations among random variables. Correla-
tions do not necessarily imply causal relationships, although many researchers
have interpreted arcs in Bayesian networks as causal links. Many researchers
have been working on causal Bayesian networks, and readers who are
interested in causal inference are referred to Pearl (1995), Cooper (1999), and
Lauritzen (2001).
Applications of Bayesian Networks

to Student Models
Research on student modeling is an interesting topic in cognition. Good student

models not only reveal how things are learned but also guide ITSs to help
people to learn (Anderson et al., 1990). Course sequencing aims at providing
individual students with appropriate course material at the right timing
(Brusilovsky & Vassileva, 2003). A well-designed ITS should consider
students’ interests, competence, and learning styles, for presenting course
materials of appropriate challenge levels with suitable formats, for example,
textual, audio, or video. Conati et al. (2002) employ Bayesian networks for
both accessing students’ competence and recognizing students’ learning plans
in helping students to learn physics. More generally, an ITS should choose the
best tutorial actions in order to maximize the teaching effects (Murray &
VanLehn, 2000).
Computerized adaptive testing (CAT) is another component of ITSs, and relies
on good student models to choose the test items that will help us to assess
students’ competence in an efficient and effective manner. While researchers

have laid out a solid foundation for CAT (van der Linden & Glas, 2000;
Wainer, 2000), some recent reports look into details about how test items
should be selected for assessment. Guzmán and Conejo (2004) emphasize the
consideration of content-balanced tests and polytomous response models, and
their work has been implemented in a Web-based tool for adaptive testing
(Conejo, Guzmán, Millán, Trella, Pérez-de-la-Cruz, & Ríos, 2004). Collins et
al. (1996) and Millán and Pérez-de-la-Cruz (2002) propose different criteria
that are chosen based on utilities of test items. In this section, we explore the
possibility of applying mutual information (cf. Cover & Thomas, 1991) for
student classification with the help of Bayesian network-based student models
(Liu, 2005).
Mutual Information-Based Adaptive Item Selection
Consider the running example shown in Figure 2 and the competence patterns
shown in Table 1. For the basic concepts, denoted by cX, we use “1” and “0”
to represent the situation of whether typical students of a particular type are
competent in that concept. For the composite concepts, denoted dY, we use
“1” and “0” to represent the situation of whether typical students of a particular
type are competent in integrating basic concepts for that composite concept.
For instance, students of the third type are competent in cB, but cannot integrate
knowledge in cA and cB into dAB. As is explained in detail in Liu (2005), “1”
and “0” here do not intend to introduce deterministic relationships. When
generating Bayesian networks in experimental studies, some limited degrees of
uncertainty will be introduced, and we employ the concept of Noisy-And nodes
for the composite nodes.
Given the network shown in Figure 2, we can compute the mutual information
between a test item and the group node, and choose the item that has the
highest mutual information with the group node as the next test item. Let
MI(X;C) denote the mutual information between random variables X and C.
We choose from the item pool the test item I that maximizes MI(I;group) in the
following MI-ADAPT procedure.
296 Liu
Table 1. Competence patterns for different types of students
student types cA cB cC dAB dBC dAC dABC

1 1 1 1 1 1 1 1
2 1 1 1 1 1 0 0
3 1 1 1 0 0 1 1
4 1 1 0 1 0 0 0
5 0 1 1 0 1 0 0
6 1 0 1 0 0 1 0
7 1 1 1 0 0 0 0
8 1 1 0 0 0 0 0
9 0 0 1 0 0 0 0
MI-ADAPT: Procedure for Adaptive Student Assessment
1. Select and administer the item that has the largest mutual information with
group.
2. Select the most probable type in group as the student’s type, based on
the posterior probability distribution over group, updated for the results
of administering the selected items.
3. Stop the classification task, if every item has been administered; otherwise
continue.
4. Compute the mutual information between each available item and group,
given the results of administering previous items.
5. Select and administer the item that has the largest condition mutual
information with group, and return to step 2.
We employ simulated students for evaluating the effectiveness of MI-ADAPT,

similarly to previous work (e.g., Millán & Pérez-de-la-Cruz, 2002; VanLehn,
Ohlsson, & Nason, 1994). It is clearly not necessary to exhaust all test items
in the item pool in assessing a student at step 3. In a practical system, we can
and should stop the assessment when a high confidence on the student’s truth
type is reached. Liu exhausts all test items here just for observing the complete
performance profile in his small-scaled experiments. For convenience, we
follow VanLehn et al. (1994) and call the simulated students simulee hence-
forth.
In this simulation-based study, we assume the availability of the Bayesian
network for modeling the problem domain, and consider chances of students’
making lucky guesses and slips when assigning conditional probability tables for
the given network. Employing a random number generator, we utilize this
Bayesian network to create simulees. In each different experiment, 20,000
simulees were created. Half of the simulees were used to train another Bayesian
network that eventually was used to classify the remaining half of the simulees.
The structure of the learned Bayesian network was the same as the network that
was used to create simulees, but the conditional probability tables were
obtained from the training data with Lauritzen’s algorithm (1995) that was
implemented in Hugin (www.hugin.dk).
Chart (a) in Figure 6 shows the experimental results when we used the network
shown in Figure 2 in the simulation. Keys for the curves indicate the setups of
the experiments. Bn indicates that we applied the Bayesian network to
compute probability distribution of the group node for student classification.
Mi indicates that we relied on mutual information for item selection. “79”
indicates that we have seven concepts and nine different types of students, as
shown in Figure 2 and Table 1, respectively. If the second to the left-most digit
is 2, there were at most 20% of the chances of making lucky guesses and slips.
If the second to the left-most digit is 5, there were at most 5% of the chances
of making lucky guesses and slips. The left-most digit indicates the chances of
students deviating from the stereotypical behavior of a particular student type,
and was used to reflect the fuzziness in simulee creation. This left-most digit can
be either “2” or “5,” implying, respectively, 0.2 and 0.05 in probability of
deviation. The horizontal axis shows the number of administered test items, and
the vertical axis shows the percentage of correctly classified simulees among
the 10,000 test cases. It is easy to see that it became harder and harder to
correctly classify simulees when the students’ behavior became more and more
uncertain, that is, by our increasing the bounds of making slips and luck guesses.
Although using exact mutual information gives us a good guidance for selecting
test items, we recall that exactly evaluating Bayesian networks is NP hard. As
a consequence, we may have to wait a long time to compute the mutual
information between each test item with the group node for a realistic large item
pool. This may not be satisfactory when responsiveness and computational
298 Liu
Figure 6. Experimental results of using mutual information for student

classification
1 1
0.9 0.9
0.8
0.8
0.7
0.6 0.7
accuracy
accuracy
0.5 0.6
0.4 BnM i7955 0.5 BnMi7725
0.3 BnM i7925 0.4 BnHMi7725
0.2
BnM i7922 0.3 BnCont7725
0.1
0 0.2
1 6 11 16 21 1 6 11 16 21
administered items
administered items
(a) (b)
efficiency of ITSs become major concerns. Applying Theorem 1 and Corollary

1 listed below, we can design a heuristic for comparing dichotomous items
without requiring the computation of exact mutual information. Let C denote a
concept and X be a test item. The test item that has a larger Pr( x | c ) − Pr( x | c )
may have larger mutual information with the concept C, therefore might shed
more light on the classes of the simulees.
Theorem 1. (Liu, 2005) For a fixed Pr(c) when Pr( x | c) ≥ Pr( x | c ), MI(X;C) is
a monotonically increasing function of Pr( x | c ) for a fixed Pr( x | c ), and a
monotonically decreasing function of Pr( x | c ) for a fixed Pr( x | c ) .
Corollary 1. (Liu, 2005) Let C be the parent concept of items X and Y. We

have if MI(X;C) ≥ MI(Y;C) if Pr( x | c ) ≥ Pr( y | c ) ≥ Pr( y | c ) ≥ Pr( x | c ).
Again, we employed a simulation-based approach to evaluate the effectiveness

of different item selection strategies, and chart (b) in Figure 6 shows the results.
In all three experiments, we used the probability distribution of the group node
to guess simulees’ types. There were seven student types after we removed the
eighth and ninth types from Table 1 for the experiments, there were at most 0.2
in chances of making lucky guesses and slips, and there were at most 0.05 in
chances of simulees’ deviating from the stereotypical behavior of the simulees’
types. The middle part of the keys for the curves indicates how test items were
selected. Mi means that we chose test items with exact mutual information,
HMi means that we chose test items with the heuristic that is designed based
on Theorem 1 and Corollary 1, and Cont means that we chose test items purely
based on content balancing. Note that we did not have experts to manually
provide the relative importance of the concepts when attempting to meet the
demands of content balancing as Guzmán and Conejo (2004) did. Instead, we
randomly chose an item for a concept, and checked that the number of tested
items was balanced as much as possible among the seven concepts. The curves
in chart (b) are quite encouraging. Clearly HMi does not perform as well as Mi,
but it already provided quite a good performance profile.
The charts in Figure 6 and other charts reported in Liu (2005) indicate that a
good strategy for choosing test items will help us to identify types of students
earlier. Hence the results of this study (Liu, 2005) and Tsiriga and Virvou
(2004) are useful for detecting examinees’ initial level in computerized adaptive
testing.
A Comparison with Item Response Theory
Item response theory (IRT, cf. van der Linden & Hambleton, 1997) is such a
dominant theory for educational assessment that we have to compare the model
shown in Figure 2 with IRT models in more detail. There are three IRT models,
each including a different number of factors in the model. The three-parameter
model considers item discrimination ai item difficulty bi and the guess parameter
ci. The model prescribes that a simulee with competence θ will respond to item
I correctly with the probability provided in (1), where ki is a constant for
normalization.
1 − ci
(1) Pr(i | θ ) = ci +
1 + e ki ai (θ −bi )
300 Liu
For grading simulees, it is common to assume that the probabilities of correct

responses to different items are independent given θ. Assume that
ℑ = {i1 , i2 ,..., it } is the set of items administered in the test. We estimate the
competence Θ of the simulee using the following formula:
(2) Θ= arg max θ Pr(θ | ℑ) = arg max θ Pr(ℑ | θ ) Pr(θ ) = arg max θ ∏ i j ∈ℑ
Pr(i j | θ ) Pr(θ )
It should be clear that formula (2) is a realization of the naive Bayes (NB)
models. Although formula (1) is significantly more complex than typical formula
used in NB models, there is no essential difference between NB models and
IRT models when we use (2) for grading simulees.
From this perspective, we can easily see that the model shown in Figure 2 is
more expressive than the IRT models. Given that we know a simulee’s type,
say g, the probability values of correctly responding to different items, for
example, I and J, remain dependent. More specifically, unlike IRT models, the
assumption for the equality in (3) is not required in Liu’s models. Moreover, the
equality will hold only if the parent concepts of the test items are independent
given the testee’s identity, which generally does not hold in Liu’s simulations
and in reality.
(3) Pr(i, j | g ) ? = Pr(i | g ) Pr( j | g )
Hence, there are three major differences between Liu’s and the IRT models.
First, students are classified into types not competence levels in our work,
although we may design a conversion mechanism between these two criteria.
Second, the responses to different test items may remain dependent given the
identity of the simulee in Liu’s models. Third, because Liu assumes that all
random variables are dichotomous, Liu’s simulations use only two parameters
for each item, which is not as expressive as the three-parameter IRT models.
The role of ci is undertaken by the parameter for controlling the bound of making
lucky guesses, and the roles of ai and bi are undertaken by the parameter for
controlling the bound of making slips. It should be clear that the MI-ADAPT
procedure allows more complex models than the dichotomous ones. Allowing
the variables that represent the mastery statuses of concepts to take more than
two possible values, we will acquire more expressive power to catch the
concepts of discrimination and difficulty of test items, thereby allowing the

models to compete even with polytomous IRT models. Paying for the gains in
expressiveness, it would become harder to compare test items purely based on
their parameters.
More Graphical Models
Besides Bayesian networks, there are other formalisms of graphical models

that are useful for student and instructor modeling. We briefly go over dynamic
Bayesian networks (cf. Russell & Norvig, 2003), influence diagrams (cf.
Jensen, 2001), and qualitative probabilistic networks (Wellman, 1990) in this
section.
The network shown in Figure 2 models students’ status at a particular moment.
Networks of this kind may not meet needs of all kinds. Assume the following
situations: (1) we administered item iA1 at time T, and a student responded
incorrectly, (2) we administered item iA1 again at time T+1, and the student
responded correctly, and (3) we administered item iA1 again at time T+2, and
the student responded incorrectly. If we infer the student’s competence only
based on his/her latest performance, we do not really have to care about the
whole history. If not, we may have to employ Bayesian networks in a more
complex way (cf. Mislevy & Gitomer, 1996).
Figure 7 shows a segment of a dynamic Bayesian network (cf. Russell &
Norvig, 2003). We duplicate the same network structure for each stage in the
network. A node that appears in each stage represents the status of a random
variable at that stage. It is not necessary to interpret the stages as different time
frames, and we can duplicate each frame after each tutorial action (e.g., Murray
& VanLehn, 2000; Reye, 2004). Using this approach, we will need different
nodes for iA1 for different time stages for coping with the aforementioned
example.
Figure 7. A (partial) dynamic Bayesian network
c o m pe te nc e t c o m pe te nc e t + 1 c o m pe te nc e t + 2
pe rf o rm anc e t pe r fo rm anc e t+ 1 pe r fo r m anc e t + 2
302 Liu
Figure 8. A (partial) multiple-stage influence diagram
c o m pe te nc e t c o m pe te nc e t+ 1 c o m pe te nc e t +2
pe rfo rm anc e t pe rfo rm anc e t+ 1 le arning e ffe c ts

pe rfo rm anc e t +2
tuto rial_ ac tio n t tuto rial_ ac tio n t+ 1 tuto rial_ ac tio n t +2
While dynamic Bayesian networks extend the applications of Bayesian net-

works into temporal reasoning, influence diagrams introduce decision nodes
and value nodes so that we can plan tutorial actions for maximizing learning
effects. The influence diagram shown in Figure 8 illustrates the basic idea. The
diagram models choices of tutorial actions at different time stages by square
nodes, and the eventual achieved affects in learning by the diamond node.
Murray and VanLehn (2000) employ influence diagrams for determining
optimal tutorial actions.
As we employ gradually more complex graphical models for modeling stu-
dents’ internal states and instructors’ actions, the computational models
capture the real world more faithfully. However, computational costs progres-
sively become a concern as well. This is probably why Millán and Pérez-de-
la-Cruz (2002) explicitly employ dynamic Bayesian networks, and later adopt
roll-up techniques (Russell & Norvig, 2003, in Millán et al., 2003).
In some particular application domains, we may use a more compact formalism
to capture the probabilistic relationships among random variables, and infer
qualitative relationships more efficiently. Take a look at the structures shown
in Figure 1 again. In practice, we have no reason to assume that the probability
of correctly responding to X would decrease when a particular student gets a
hand on C. Therefore, in Figure 1(a), we also have Pr( x | c) ≥ Pr( x | c ), and we
show this positive influence of C on X by marking the link between them with
the “+” symbol, following the tradition of Qualitative Probabilistic Networks
(QPNs) (Wellman, 1990). In a fully fledged QPN, random variables may also
have relationships of negatively influence (denoted by “-”) and ambiguity
(denoted by “?”). One marks the relationship between a concept and an item
by “-”, when understanding the concept hinders a student from answering the
item correctly. This may happen when students learn a new concept that is
similar to C in nature but might mislead students to respond to X incorrectly.
Figure 9. Corresponding qualitative probabilistic networks of the networks

shown in Figure 1
+ +
+ C + A B
+ + C +
C X
( a) (b ) Y X (c ) Y X
Assume that C has positive influences on X and Y, and that A and B have positive
influences on B and C, respectively. We can enhance the networks in Figure 1
with qualitative signs, as shown in Figure 9.
Wellman (1990) and Druzdzel and Henrion (1993) have proposed inference
algorithms for QPNs. Given a network such as the one shown in Figure 9(c),
we can infer that the competence in A positively influences the competence in
C. We can also infer that the competence in B positively influences the values
of X and Y, indicating that competence in B increases the probability of
responding to X and Y correctly. Conversely, the chance that a student is
competent in B is qualitatively increased when s/he respond to X or Y correctly.
Conclusion
In order to assist students toward efficient and interesting learning, ITSs must
find ways to infer internal aspects of students from their external behavior.
Whether we infer about students’ competence in concepts based on their
responses to test items or we infer about students’ learning plans based on their
explorative actions in an open environment, the problem is about conducting
inference under uncertainty. To meet this technical challenge, this chapter
provides background information for probabilistic reasoning using Bayesian
networks and their extensions.
The applications of Bayesian networks in ITSs have widened in recent years.
Breese and Ball (1998) apply dynamic Bayesian networks for modeling the
emotion and personality of agents, and Conati (2002) applies Bayesian
networks for assessing and monitoring the user’s emotion in educational games.
Readers who are interested in the effectiveness in modeling real students with
Bayesian networks should refer to their work. Wang and Liu (2004) attempt
304 Liu
Figure 10. Mapping learning processes: An alternative for the network

shown in Figure 1
g ro u p
cA cB cC
dA B d BC dA C d A BC
to map students’ learning processes of composite concepts. Based on stu-

dents’ responses to test items, they apply techniques of learning Bayesian
networks for judging the relative genuineness of the networks shown in Figures
2 and 10. (Figure 10 does not include nodes for test items for simplicity of the
network structure.) These networks differ in the parent nodes of dABC, which
indicate different ways of learning the composite concept dABC from relatively
more basic concepts. Zapata-Rivera and Greer (2004) propose a negotiated
assessment process based on inspectable Bayesian student models. With this
trend, we may see growing contributions of Bayesian networks to actively
assist human users (Kautz, Etzioni, Fox, & Weld, 2003) and even to the
theoretical aspect of cognitive systems (Hoffman & Woods, 2005).
Toward this end, one might become interested in a philosophic question.
Anderson et al. (1990) have mentioned that one of the main purposes of
developing intelligent tutoring systems is to offer human users a relatively
economical substitute to human tutors, while examining the validity of cognitive
theories. From this viewpoint, it may be interesting to ask whether human tutors
employ Bayesian reasoning in their diagnoses of human students. That is, are
people Bayesian (El-Gamal & Grether, 1995)?
Acknowledgments
This work is partially supported by Grants 93-2213-E-004-004 and 94-

2213-E-004-008 from the National Science Council, Taiwan. The author
would like to thank anonymous reviewers and the editor of this book for
precious feedback on the draft version of this chapter.
References
Anderson, J. R., Boyle, C. F., Corbett, A., & Lewis, M. W. (1990). Cognitive
modelling and intelligent tutoring. Artificial Intelligence, 42(1), 7–49.
Anderson, J. R., Corbett, A., Fincham, J., Hoffman, D., & Pelletier, R. (1992).
General principles for an intelligent tutoring architecture. In V. Shute & W.
Regian (Eds.), Cognitive approaches to automated instruction (pp.
81–106). Lawrence Erlbaum.
Anderson, J. R., Douglass, S., & Qin, Y. (2004). How should a theory of
learning and cognition inform instruction? In A. F. Healy (Ed.), Experi-
mental cognitive psychology and its applications (pp. 47–58). Wash-
ington, DC: American Psychological Association.
Breese, J., & Ball, G. (1998). Bayesian networks for modeling emotional state
and personality: Progress report. AAAI 1998 Fall Symposium Techni-
cal Report FS-98-03, 37–42.
Breese, J. S., Goldman, R. P., & Wellman, M. P. (1994). Introduction to the
special section on knowledge-based construction of probabilistic and
decision models. IEEE Transactions on Systems, Man, and Cybernet-
ics, 24, 1577–1579.
Brusilovsky, P. (1994). The construction and application of student models in
intelligent tutoring systems. Journal of Computer and Systems Sciences
International, 32(1), 70–89.
Brusilovsky, P., Schwarz, E. W., & Weber, G. (1996). ELM-ART: An
intelligent tutoring system on World Wide Web. Proceedings of the
Third International Conference on Intelligent Tutoring Systems (pp.
261–269).
Brusilovsky, P., & Vassileva, J. (2003). Course sequencing techniques for
large-scale Web-based education. International Journal of Continu-
ing Engineering Education and Lifelong Learning, 13(1–2), 75–94.
Bunt, A., & Conati, C. (2003). Probabilistic student modelling to improve
exploratory behaviour. User Modeling and User-Adapted Interaction,
13(3), 269–309.
Chickering, D. M. (1996). Learning Bayesian networks is NP-complete. In D.
Fisher & H.-J. Lenz (Eds.), Learning from data: Artificial intelligence
and statistics V (pp. 121–130). Springer Verlag.
306 Liu
Collins, J. A., Greer, J. E., & Huang, S. X. (1996). Adaptive assessment using
granularity hierarchies and Bayesian net. Proceedings of the Third
International Conference on Intelligent Tutoring Systems (pp. 569–
577).
Conati, C. (2002). Probabilistic assessment of user’s emotions during the
interaction with educational games. Applied Artificial Intelligence,
16(7–8), 555–575.
Conati, C., Gertner, A., & VanLehn, K. (2002). Using Bayesian networks to
manage uncertainty in student modeling. User Modeling and User-
Adapted Interaction, 12(4), 371–417.
Conati, C., & VanLehn, K. (2001). Providing adaptive support to the
understanding of instructional material. Proceedings of the Sixth Inter-
national Conference on Intelligent User Interfaces (pp. 41–47).
Conejo, R., Guzmán, E., Millán, E., Trella, M., Pérez-De-La-Cruz, J. L., &
Ríos, A. (2004). SIETTE: A Web-based tool for adaptive testing.
International Journal of Artificial Intelligence in Education, 14, 29–
61.
Cooper, G. F. (1990). The computational complexity of probabilistic inference
using Bayesian networks. Artificial Intelligence, 42(2–3), 393–405.
Cooper, G. F. (1999). An overview of the representation and discovery of
causal relationships using Bayesian networks. In C. Glymour & G. F.
Cooper (Eds.), Computation, causation, & discovery (pp. 3–62).
Cambridge, MA: MIT Press.
Cover, T. M., & Thomas J. A. (1991). Elements of information theory. John
Wiley & Sons.
Dagum, P., & Luby, M. (1993). Approximating probabilistic inference in
Bayesian belief networks is NP-hard. Artificial Intelligence, 60(1),
141–153.
de Kleer, J. (1986). An assumption-based TMS. Artificial Intelligence, 28,
127–162.
de Kleer, J., & Williams, B. C. (1987). Diagnosing multiple faults. Artificial
Intelligence, 32, 97–130.
Druzdzel, M. J., & Henrion, M. (1993). Efficient reasoning in qualitative
probabilistic networks. Proceedings of the Eleventh National Confer-
ence on Artificial Intelligence (pp. 548–553).
El-Gamal, M. A., & Grether, D. M. (1995). Are people Bayesian? Uncovering

behavioral strategies. Journal of the American Statistical Association,
90, 432, 1137–1145.
Guzmán, E., & Conejo, R. (2004). A model for student knowledge diagnosis
through adaptive testing. Proceedings of the Seventh International
Conference on Intelligent Tutoring Systems (pp. 12–21).
Heckerman, D. (1999). A tutorial on learning with Bayesian networks. In M.
I. Jordan (Ed.), Learning in graphical models (pp. 301–355). Cam-
bridge, MA: MIT Press.
Hoffman, R. R., & Woods, D. D. (2005). Toward a theory of complex and
cognitive systems. IEEE Intelligent Systems, 20(1), 76–79.
Horvitz, E., Breese, J., Heckerman, D., Hovel, D., & Rommelse, K. (1998).
The Lumière project: Bayesian user modeling for inferring the goals and
needs of software users. Proceedings of the Fourteenth Conference on
Uncertainty in Artificial Intelligence (pp. 256–265).
Jensen, F. V. (2001). Bayesian networks and decision graphs. Springer
Verlag.
Kautz, H., Etzioni, O., Fox, D., & Weld, D. (2003, March). Foundations of
assisted cognition systems. University of Washington, CSE Technical
Report.
Kono, Y., Ikeda, M., & Mizoguchi, R. (1994). THEMIS: A nonmonotonic
inductive student modeling system. International Journal of Artificial
Intelligence in Education, 5(3), 371–413.
Lauritzen, S. L. (1995). The EM algorithm for graphical association models
with missing data. Computational Statistics & Data Analysis, 19(2),
191–201.
Lauritzen, S. L. (2001). Causal inference from graphical models. In O. E.
Barndorff-Nielsen, D. R. Cox, & C. Klüppelberg (Eds.), Complex
stochastic systems (pp. 63–108). Chapman & Hall/CRC.
Liu, C.-L. (2005). Using mutual information for adaptive item comparison and
student assessment. Journal of Educational Technology & Society,
8(4), 100–119.
Martin, J., & VanLehn, K. (1995). Student assessment using Bayesian nets.
International Journal of Human-Computer Studies, 42(6), 575–
591.
308 Liu
Mayo, M., & Mitrovic, A. (2000). Using a probabilistic student model to

control problem difficult. Proceedings of the Fifth International Con-
ference on Intelligent Tutoring Systems (pp. 524–533).
Mayo, M., & Mitrovic, A. (2001). Optimising ITS behaviour with Bayesian
networks and decision theory. International Journal of Artificial
Intelligence in Education, 12, 124–153.
Millán, E., & Pérez-de-la-Cruz, J. L. (2002). A Bayesian diagnostic algorithm
for student modeling and its evaluation. User Modeling and User-
Adapted Interaction, 12(2–3), 281–330.
Millán, E., Pérez-de-la-Cruz, J. L., & García, F. (2003). Dynamic versus static
student models based on Bayesian networks: An empirical study. Pro-
ceedings of the Seventh International Conference on Knowledge-
Based Intelligent Information and Engineering Systems (pp. 1337–
1344).
Mislevy, R. J., Almond, R. G., Yan, D., & Steinberg, L. S. (1999). Bayes nets
in educational assessment: Where do the numbers come from? Proceed-
ings of the Fifteenth Conference on Uncertainty in Artificial Intelli-
gence (pp. 437–446).
Mislevy, R. J., & Gitomer, D. H. (1996). The role of probability-based
inference in an intelligent tutoring system. User Modeling and User-
Adapted Interaction, 5(4), 253–282.
Mislevy, R. J., Steingerg, L. S., & Almond, R. G. (2003). On the structure of
educational assessments. Measurement: Interdisciplinary Research
and Perspectives, 1(1), 3–62.
Mitrovic, A., Martin, B., & Mayo, M. (2002). Using evaluation to shape ITS
design: Results and experiences with SQL-Tutor. User Modeling and
User-Adapted Interaction, 12(2–3), 243–279.
Murray, R. C., & VanLehn, K. (2000). DT Tutor: A decision-theoretic,
dynamic approach for optimal selection of tutorial actions. Proceedings
of the Fifth International Conference on Intelligent Tutoring Sys-
tems (pp. 153–162).
Neapolitan, R. E. (2004). Learning Bayesian networks. Prentice Hall.
Pearl, J. (1988) Probabilistic reasoning in intelligent systems: Networks of
plausible inference. Morgan Kaufmann.
Pearl, J. (1995). Causal diagrams for empirical research. Biometrika, 82(4),
669–688.
Reye, J. (2004). Student modeling based on belief networks. International

Russell, S., & Norvig, P. (2003). Artificial intelligence: A modern approach
(2nd ed.). Prentice Hall.
Shachter, R. D. (1986). Evaluating influence diagrams. Operations Research,
34(6), 871–882.
Sison, R., & Shimura, M. (1998). Student modeling and machine learning.
International Journal of Artificial Intelligence in Education, 9, 128–
158.
Tsiriga, V., & Virvou, M. (2004). A framework for the initialization of student
models in Web-based intelligent tutoring systems. User Modeling and
User-Adapted Interaction, 14(4), 289–316.
VanLehn, K., Ohlsson, S., & Nason, R. (1994). Applications of simulated
stu-dents: An exploration. International Journal of Artificial Intelli-
gence in Education, 5(2), 135–175.
van der Linden, W. J., & Glas, C. A. W. (Eds.). (2000). Computerized
adaptive testing: Theory and practice. Kluwer.
van der Linden, W. J., & Hambleton, R. (1997). Handbook of modern item
response theory. Springer Verlag.
Villano, M. (1992). Probabilistic student models: Bayesian belief networks
and knowledge space theory. Proceedings of the Second International
Conference on Intelligent Tutoring Systems (pp. 491–498).
Virvou, M., Maras, D., & Tsiriga, V. (2000). Student modelling in an intelligent
tutoring system for the passive voice of English language. Journal of
Educational Technology & Society, 3(4), 139–150.
Virvou, M., & Moundridou, M. (2001). Adding an instructor modelling
component to the architecture of ITS authoring tools. International
Wainer, H. (Ed.). (2000). Computerized adaptive testing: A primer (2nd
ed.). Lawrence Erlbaum.
Wang, Y.-T., & Liu, C.-L. (2004). An exploration of mapping the learning
processes of composite concepts. Proceedings of the Ninth TAAI
Conference on Artificial Intelligence and Applications, CD-ROM.
(In Chinese.)
310 Liu
Wellman, M. P. (1990). Fundamental concepts of qualitative probabilistic

net-works. Artificial Intelligence, 44(3), 257–303.
Zapata-Rivera, J.-D., & Greer, J. E. (2004). Interacting with inspectable
Bayesian student models. International Journal of Artificial Intelli-
gence in Education, 14, 127–163.
Section V
A Real-World
Case Study
312 Hilton
Chapter XIV
The Effect of
Technology on Student
Science Achievement
June K. Hilton, Claremont High School, USA
Abstract
Empirical data from a California secondary school was analyzed to

determine the direct and indirect effects of technology on student science
achievement. Prior research indicates that technology has had minimal
effect on raising student achievement. Little empirical evidence exists
examining the effects of technology as a tool to improve student
achievement by developing higher-order thinking skills. Furthermore,
previous studies also have not focused on the manner in which the
technology is being used in the classroom to enhance teaching and
learning. As a result, the variables analyzed in this study include student
computer use, teacher use of computers for instruction, student attendance,
Student Science Achievement 313
teacher preparation, demographic factors, and final science course grades.

The method of quantitative analysis includes a path analysis using final
course grades as the ultimate endogenous variable. Key findings include
evidence that while technology training for teachers increased their use
of the computer for instruction, a student’s final science course grade did
not improve.
Introduction
Those inside and outside of education have debated technology use and its
effect on student achievement over recent years. With the advent of such
federal mandates as the No Child Left Behind Act of 2001, school accountabil-
ity for meeting state and national standards has become a major concern for not
only local school officials but also their state counterparts. As school funding
becomes more closely linked to performance on standardized tests, improving
student achievement through innovative means, such as technology integration
and use, is critical.
In the pages that follow, the steps for an empirical study of the effects of
technology on student science achievement are presented. The indirect fac-
tors—student access after school, teacher technology preparation and expe-
rience, and type of science course (physical vs. life)—may provide more
interesting results than the direct effects (teacher use and availability). Accord-
ing to Skinner (2002), “purchasing computers and improving Internet connec-
tions are just part of what it takes to make technology an integral part of
teaching and learning. Preparing teachers to use and integrate technology into
their work in meaningful ways remains a challenge” (p. 53). Means, Wagner,
Haertel, and Javitz concur, explaining, “Existing survey data provide a portrait
of the distribution of technology resources in U.S. schools and, to a lesser
extent, the distribution of different general categories of technology use (e.g.,
drill and practice on basic skills, games, CDROM reference materials, Internet
search). What the data cannot tell us is whether involvement with one or
multiple kinds of technology is having a long-term impact on the students who
use them” (Means, Wagner, Haertel, & Javitz, 2000, p. 2). The analysis of the
data collected will provide more insight into this area.
314 Hilton
Objectives
Previous research focused on the initial phase of technology implementation,

that is, increasing access and initial professional development, which centered
on computer basics. What is now needed is an in-depth look into the effects
technology has on student science achievement at a representative school via
a quantitative study that focuses on how students and teachers use computers
in the teaching and learning process. This analysis will add much-needed
information to the knowledge base regarding technology and student achieve-
ment.
Background
In December 2000, a national technology plan, eLearning: Putting a World-

Class Education at the Finger Tips of All Children (U.S. Department of
Education, 2000), was released which outlined five goals to improve how
technology is used in both elementary and secondary education. With the
approval of this plan, a race to connect every classroom to the Internet was
underway. According to Cuban, “techno-promoters across the board assumed
that increased availability in the classroom would lead to increased use.
Increased use, they further assumed, would then lead to efficient teaching and
better learning, which in turn, would yield able graduates who can compete in
the workplace” (Cuban, 2001, p. 18). Unfortunately, results from the National
Center for Education Statistics (NCES) National Assessment for Educational
Progress (NAEP) 2000 Results in Science and Mathematics (National Center
for Education Statistics, 2002) showed no significant change in student
achievement from 1996 despite an increase in computers in the classroom.
The question becomes, “Why is increased access to technology not creating a
rise in student achievement?” The answer to this question is not simple and must
focus on how computers are used for instruction as well as explore the indirect
effects technology has on students and their perceptions of school.
Technology and Student Achievement
Simply because the hardware and software are in place in a classroom does not
imply that it is either being used or being used effectively. Hedges,
Konstantopoulos, and Thoreson (2000) found that even though teachers have
had increasing access to computers for instruction, very few actually use them.
Although 84% of all public school teachers said personal computers were
available to them, approximately 60% indicated that they used them (U.S.
Bureau of the Census, 1998). Analysis of teacher data from the National
Education Longitudinal Study (NELS) showed that about half of eighth grade
math teachers have students who spend less than 10% of class time working on
computers (Hedges et al., 2000), while across subject matter, teachers
average only about 4% of all instructional time with computers (Cuban, 1993).
A survey of middle school math and science teachers in South Carolina (Dickey
& Kherlopian, 1987) also showed that although 70% of these teachers had
access to computers, almost half of those with access did not use them. Thus,
even though computer technology may be widely available, in general, it is
poorly integrated into the classroom curriculum and is underused (Hedges et
al., 2000).
An additional concern deals with the accountability states place on technology
use and integration in the curriculum. Research has shown that the effectiveness
of technology in improving student achievement depends on a match between
the goals of instruction, the characteristics of the learners, the design of the
software, the technology, and the implementation decisions made by teachers
(Padilla & Zalles, 2001, p. 14). Schacter agrees, stating, “Learning technology
is less effective or ineffective when the learning objectives are unclear and the
focus of the technology use is diffuse” (Schacter, 1999, p. 10). Furthermore,
in addition to consistency between what is taught and the goals of the
technology, alignment must also occur between what is taught and the tests used
to assess and monitor student achievement. If the technology supports higher-
order thinking skills and the assessment system evaluates only basic skills, the
benefits of the technology may not be evident (Padilla & Zalles, 2001, p. 15).
According to Park and Staresina, “All but six states include technology in state
academic standards, and of those six states, Georgia is currently drafting
technology standards, and Mississippi plans to have standards in place next
school year. But only three states—New York, North Carolina, and Utah—
actually test student knowledge of technology to see if the instruction is having
an impact” (Park & Staresina, 2004, p. 67).
316 Hilton
Another concern for educators is the manner in which computers are used in the
classroom and the effect it is having on student achievement. The CEO Forum
Year 4 Report (CEO Forum, 2001) found that while students frequently use
computers at school for research (96%) or to write papers (91%), their actual
use for learning new concepts or practicing new skills learned in class was
significantly lower (60% and 57%, respectively). Results from the 2003
National Assessment of Education Progress (NAEP) background questions
found that “an overwhelming majority of students had teachers who were using
computers for basic drill and practice or for math games. Very few were using
computers for higher-order-thinking tasks such as simulations” (Park &
Staresina, 2004, p. 67).
The effect this use or misuse has on achievement is contradictory. A review of
studies by the CEO Forum found that “technology can have the greatest impact
when integrated into the curriculum to achieve clear, measurable educational
objectives” (CEO Forum, 2001, p. 2). Middleton and Murray (1999)
conducted a study on the impact of technology on student reading and
mathematics achievement of fourth and fifth graders. Their study found that the
level of technology used by the classroom teacher affected student achieve-
ment. Analysis of data from the NAEP’s 2000 Science Assessment on types
of computer use by students revealed that students whose teachers used
computers for learning games (grade 4); simulations and models for data
analysis (grade 8); and data download, collection, and analysis, use of probes,
or exchanging information via the Internet (grade 12) scored higher on the
Science Assessment than those whose teachers did not use the computer
(National Center for Education Statistics, 2002).
In his report on the effects of technology on student mathematics achievement,
Harold Wenglinsky (1998) found that unfortunately, for all of the investment in
educational technology, there is a surprising lack of hard data on its effects.
Wenglinsky found the types of use to which computers are put varies greatly
between fourth and eighth grade. Among fourth graders, 54.5% have teachers
reporting learning games as the primary use; 35.9% report drill and practice;
7.5% report simulations and applications; and 2.1% report introducing new
topics. Thus, the activity traditionally thought of as teaching higher-order skills,
applying concepts or developing simulations to illustrate them, is rarely used.
Based on his analysis of the 1996 NAEP Mathematics Assessment, Wenglinsky
found that at in all grade-level studies (Grades 4, 8, and 12) teachers who are
knowledgeable in the use of computers are more likely to use them for higher-
order purposes. When computers are used to perform certain tasks, namely
applying higher-order concepts, and when teachers are proficient enough in

computer use to direct students toward productive uses more generally,
computers do seem to be associated with significant gains in mathematics
understanding and skills (Wenglinsky, 1998).
The following section will take an in-depth look, through empirical analysis, at
the manner in which the members of the science department at one school
attempt to implement technology into their classroom and the effect this
implementation had on student achievement. Through this analysis, the ability
to evaluate the effectiveness of technology in raising student achievement will
be discussed.
Methodology
Student Population Data
The data set consists of the 2002 final course grades, student and teacher
computer use, and demographic information from students in Grades 9–11 at
a high-poverty, comprehensive high school in Western Riverside County,
California. The school’s enrollment, during the 2001–2002 school year, was
2,752 of which 400 were 12th graders. The final sample size was 1,194 after
removing those students not taking science or enrolled in science outside the
Science Department (Special Education, Opportunity School, Home Study).
The school district supplied information on ethnicity, home/primary language,
student gender, 2001–2002 student attendance, final course grade, and the
name of their science teacher. Information regarding teaching experience,
teacher participation in technology training, and teacher gender was obtained
from the Assistant Principal for Curriculum and Instruction. Information
regarding student and teacher computer use was determined via computer lab
sign-in sheets.
Analysis Methods
This study utilized path analysis to investigate the direct and indirect effects of
technology as it relates to Student Science Achievement. Descriptive statistics
318 Hilton
Figure 1. Proposed path model of student achievement predictors
for key variables, results from the multiple regressions, calculations involving
error vectors, and decomposition tables for bivariate covariation are pre-
sented. The proposed path is presented in Figure 1.
Research Questions
This study investigated the direct and indirect effects of technology as they
relate to a student’s final grade in his/her respective science course. The
specific variables studied were categorized as demographics, teacher prepa-
ration/experience, teacher instructional practices, and student factors. The
ultimate endogenous variable studied was the student’s final course grade. The
issues about the relationships of these factors can be expressed in empirically
testable terms by the following questions:
1. Does technology significantly affect a student’s final grade in science?

2. Are the indirect effects of technology significant predictors of a student’s
final grade in science?
Results
Descriptive Statistics
Descriptive statistics for the variables are given in Table 1. Based on these
descriptive statistics, approximately 51% of the students were enrolled in a
Geophysical Science course (Sheltered, Regular, or Honors), 28% of the
students were enrolled in a Biology course (Regular, Honors, or Sheltered),
and 20% of the students were enrolled in either a Chemistry or Physics course
(Regular, Honors, or Advanced Placement). Almost two-thirds of the students
spoke English at home and as their primary language as designated by the Home
Language Survey given to all students when they first enroll in the district.
Almost 60% of the students were of Hispanic ethnicity while approximately
40% were of White ethnicity. Student gender was almost evenly distributed
between males (47.5%) and females (52.5%). Students demonstrated a 95%
attendance rate. The mean final course grade was slightly above a “C”. The
teachers at the high school averaged almost 10 years of experience, were
mostly male (70%), and used the computer lab for instruction approximately
5% of the available time. Of the 11 science department members, 5 (45%)
participated in the school sponsored technology training.
Pearson product moment correlations of the key variables are given in Table
2. There are significant positive correlations, identified in bold print, between
type of science course taught, either chemistry or physics (C/P C), and years
of teaching experience (YTE) (0.429) as well as between teacher use of
computer lab for instruction (TUL) and type of science course (BC) (0.415).
Significant negative correlations, also identified in bold print, exist between
teacher gender (TG) and years teaching experience (“0.356) indicating that
males in the department have more years of teaching experience. Other
interesting correlations exist between type of science course and final course
grade (FCG) (0.259), teacher gender and teacher use of the computer lab for
instruction (0.274), and % days present (DP) and final course grade (0.225).
320 Hilton
Table 1. Descriptive statistics (N = 1,194)
Vari able % Yes % No

Geophysical Course 51.0 49.00
Biology Course 28.9 71.10
Che mistry or Physics Course 20.1 79.90
English Ho me Language 66.2 33.80
English Primary Language 67.2 32.80
Hispanic Ethnicity 57.3 42.70
White Ethnicity 38.5 61.50
Teacher Participation in Technology Training 45.3 54.70
% Female % Male
Student Gender 52.5 47.50
Teacher Gender 29.5 70.50
Actual Years of Teaching Experience Percent

1 10.6
2 12.2
4 9.8
5 6.9
8 9.6
11 23.3
15 13.7
20 3.6
25 10.3
% Teacher Uses Co mputer Lab Percent

0 13.7
1.7 11.7
3.6 10.6
4.2 9.8
5 6.9
5.9 14.7
6.7 10.3
8.3 1.2
9.2 11.6
10.9 9.6
Table 1. Cont.
Final Course Grade Percent
A 18.3
B 27.1
C 25.7
D 17.4
F 10.2
Missing Data 1.2
% Days Student Present Percent

100% 5.1
95– 99 43.2
90– 94 18.7
85– 89 6.1
80– 84 2.0
75– 79 0.8
70– 74 0.4
65– 69 0.0
60– 64 0.1
55– 59 0.1
Less than 54% 0.0
Missing Data 24.2
This illustrates the following:
1. The final grade in Biology was lower than in Chemistry/Physics.

2. Female teachers used the lab more often than male teachers.
3. Students with a higher attendance percentage had higher final course
grades in their respective science classes.
Correlations (also highlighted in bold) between Ethnicity and Primary/Home

Language provide expected results.
Multivariate Path Analysis
After performing the necessary stepwise multiple regressions and bivariate

correlations, the estimated model indicates technology has significant direct
322 Hilton
Table 2. Pearson correlation matrices for key variables (N = 1,194)

TPT YTE TUL BC C/P C % DP
TPT 1.000 0.291 0.073 −0.142 0.060 −0.004

YTE 0.291 1.000 −0.012 −0.013 0.429 0.043
TUL 0.073 −0.012 1.000 0.415 0.184 −0.003

BC −0.142 −0.013 0.415 1.000 −0.320 0.015
C/P C 0.060 0.429 0.184 −0.320 1.000 0.048
% DP −0.004 0.043 −0.003 0.015 0.048 1.000
SG 0.050 0.049 0.018 −0.093 0.100 −0.089

HE 0.014 −0.114 −0.059 0.035 −0.065 0.010
WE −0.008 0.073 0.059 −0.011 0.019 −0.020

EHL 0.016 0.131 0.087 −0.014 0.040 −0.086
EPL 0.009 0.117 0.055 −0.062 0.057 −0.088
TG −0.124 −0.356 0.274 0.297 −0.191 −0.085
FCG 0.043 0.156 −0.018 −0.058 0.259 0.225
SG HE WE EHL EPL TG FCG
TPT 0.050 0.014 −0.008 0.016 0.009 −0.124 0.043
YTE 0.049 −0.114 0.073 0.131 0.117 −0.356 0.156
TUL 0.018 −0.059 0.059 0.087 0.055 0.274 −0.018

BC −0.093 0.035 −0.011 −0.014 −0.062 0.297 −0.058
C/P C 0.100 −0.065 0.019 0.040 0.057 −0.191 0.259
% DP −0.089 0.010 −0.020 −0.086 −0.088 −0.085 0.225

SG 1.000 0.050 ?0.064 −0.033 −0.033 −0.014 0.097
HE 0.050 1.000 ?0.917 −0.566 −0.546 0.120 −0.171

WE −0.064 −0.917 1.000 0.565 0.553 −0.123 0.151
EHL −0.033 −0.566 0.565 1.000 0.825 −0.176 0.089
EPL −0.033 −0.546 0.553 0.825 1.000 −0.178 0.128
TG −0.014 0.120 −0.123 −0.176 −0.178 1.000 −0.240

FCG 0.097 −0.171 0.151 0.089 0.128 −0.240 1.000
Legend for Table: TPT = Teacher Participation in Technology Training; % DP = %

Days Present; YTE = Years of Teaching Experience; SG = Student Gender; TUL =
Teacher Uses Computer Lab for Instruction; TG = Teacher Gender; BC = Biology
Course; HE = Hispanic Ethnicity; C/P C = Chemistry/Physics Course; WE = White
Ethnicity; FCG = Final Course Grade; EHL = English Home Language; EPL = English
Primary Language
and indirect effects on student achievement. The analysis of the data revealed
that the significant exogenous variables were the ethnicity variable of Hispanic
and Student Gender. With the exception of Home Language Spanish, Primary
Language Spanish, Student Use of the Computer Lab After School, and Type
of Science–Geophysical, all of the proposed endogenous variables remained
in the model to predict student science achievement as measured by final course
grade. Home Language English, Primary Language English, Teacher Participa-
tion in Technology Training, Teacher Gender, and Years Teacher Experience,
however, were not direct predictors of the ultimate endogenous variable. Table
3 illustrates significant predictors of final course grade. Tables 4–7 illustrate the
significant predictors of the other exogenous variables, which remained in the
regression equation.
The calculations of the error vectors as well as the decomposition table for
bivariate covariation are given below in Tables 8 and 9, respectively. The path
coefficients and error vectors for the estimated model are given in Figure 2.
The results of the estimated model indicate that the variables are moderate to
good predictors of student science achievement as measured by final course
grade. Two of the seven error vectors (Final Course Grade and Type of
Science – Biology) were between 0.85 and 0.95, indicating a moderate causal
relationship. The error vectors for the endogenous variables of Primary
Table 3. Prediction of final course grades for students in grades 9–11
Independent Variables Beta t Sig t

Type of Science – Chemistry/Physics 0.284 9.312 <.001
% Days Present 0.196 7.728 <.001
Ethnicity – Hispanic −0.169 −6.267 <.001
Teacher Uses Computer Lab −0.121 −3.824 <.001
for Instruction
Student Gender 0.103 3.787 <.001
Type of Science – Biology 0.097 2.931 0.003
R = .384
R2 = .147
Adj. R2 = .143
F = 34.115
Sig F < .001
N = 1194
324 Hilton
Table 4. Prediction of Biology Course for students in grades 9–11

Teacher Gender 0.296 10.734 <.001
Student Gender −0.089 −3.228 0.001
R = .310
R2 = .096
Adj. R2 = .095
F = 63.328
Sig F < .001
N = 1194
Table 5. Prediction of Chemistry/Physics Course for students in grades 9–11

Teacher Gender −.190 −6.713 <.001
Student Gender 0.098 3.449 0.001
R = .215
R2 = .046
Adj. R2 = .045
F = 28.815
Sig F < .001
N = 1194
Language – English (0.556) and Teacher Use of the Computer Lab for
Instruction (0.796) indicate a good relationship. The error vector values signify
moderate to low residuals for these variables and also correspond to the
moderate to good R2 values (> .10) indicating that these variables account for
as much as 69.1% of the dependent variable. The R2 value of 0.143 (Final
Course Grade) indicates a moderate relationship, accounting for 14.3% of the
dependent variable. The other R2 values of .367 (Teacher Use of the Computer
Lab for Instruction) and .691 (Primary Language – English) indicate moderate
to strong relationships accounting for 36.7% and 69.1%, respectively, of the
dependent variable.
Table 6. Prediction of Teacher Use of Computer Lab for Instruction for

students in grades 9–11

Type of Science – Biology 0.544 20.641 <.001
Type of Science – Chemistry/Physics 0.465 16.880 <.001
Teacher Years of Experience −0.209 −7.210 <.001
Teacher Participate in Technology Training 0.202 8.219 <.001
Teacher Gender 0.152 5.838 <.001
R = .608
R2 = .370
Adj. R2 = .367
F = 139.468
Sig F < .001
N = 1194
Table 7. Prediction of Primary Language English for students in grades

9–11
English Ho me Language 0.752 38.569 <.001
Ethnicity – White 0.128 6.581 <.001
R = .832
R2 = .692
Adj. R2 = .691
F = 1335.218
Sig F < .001
N = 1194
Analysis of the decomposition table reveals that one-third of the noncausal

values (6 of 18) were more than 0.10, indicating a difference between what the
model predicts and what the correlation describes. Two of the noncausal values
can be considered to show a moderate relationship (between 0.06 and 0.10)
while 10 noncausal values (< 0.05) showed a good relationship between the
model estimation and the correlation. Thus, it appears that the expected and
observed outcomes are virtually the same for about two-thirds of the relation-
ships.
326 Hilton
Table 8. Calculation of error vectors
Endogenous Variable R R2 e= 1- R2
Final Course Grade 0.384 0.143 0.926

% Days Present 0.112 0.011 0.994
Primary Language – English 0.832 0.691 0.556
Type of Science – Biology 0.310 0.095 0.951
Type of Science – Che mistry/Physics 0.215 0.045 0.977
Teacher Participation in 0.291 0.085 0.957
Technology Train ing
Teacher Use of Co mputer 0.608 0.367 0.796
Lab for Instruction
Figure 2. Estimated path model
Table 9. Decomposition table

FG/ EH FG/ DP FG/S G FG/B FG/ CP FG/ TUL DP/PLE PLE/ HLE PLE/ EW
Original −0.17 0.23 0.10 −0.06 0.26 −0.02 −0.09 0.83 0.55
Covariati on
Direct Effects −0.17 0.20 0.10 0.10 0.28 −0.12 −0.08 0.75 0.13
Indirect
Effects 0.00 0.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00
Total Effects −0.17 0.20 0.12 0.10 0.28 −0.12 −0.08 0.75 0.13
Non c ausal 0.00 0.03 −0.02 −0.16 −0.02 0.10 −0.01 0.08 0.42
B/SG B/TG CP/SG CP/TG TPT/ YTE TUL/B TUL/ CP TUL/ TPT TUL/ YT E
Original −0.09 0.30 0.10 −0.19 0.29 0.42 0.18 0.07 −0.01
Covariati on
Direct Effects −0.09 0.30 0.10 −0.19 0.29 0.54 0.47 0.20 −0.21
Indirect
Effects 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.06
Total Effects −0.09 0.30 0.10 −0.19 0.29 0.54 0.47 0.20 −0.15
Noncausal 0.00 0.00 0.00 0.00 0.00 −0.12 −0.29 −0.13 0.14
Legend for Table: TPT = Teacher Participation in Technology Training; DP = % Days

Present; YTE = Years of Teaching Experience; SG = Student Gender; TUL = Teacher
Use of the Computer Lab for Instruction; EH = Ethnicity – Hispanic; B = Biology Course
EW = Ethnicity – White; CP = Chemistry/Physics Course HLE = Home Language –
English; FG = Final Course Grade; TG = Teacher Gender; PLE = Primary Language
– English
Two of the 18 variables displayed indirect effects. These variables included

Student Gender and Years of Teaching Experience. This shows that the
exogenous variable of Student Gender not only affected Final Course Grade
directly but also indirectly through Type of Science – Biology or Chemistry/
Physics. Interestingly, while Years of Teaching Experience did not affect Final
328 Hilton
Course Grade directly, it did so indirectly through Teacher Use of the

Computer Lab for Instruction.
Conclusion
Interpretation and Recommendations
The results of this study indicate that teacher use of technology displayed both
direct and indirect effects on student achievement. These effects are detailed
below.
Teacher Use of the Computer Lab for Instruction
eacher use of the computer lab (beta = “.121) was a significant predictor of final
course grade. Although the negative sign indicates that students whose teachers
used the computer lab had lower grades, it does not necessarily imply that
teachers should forsake using the computer lab, as their students’ achievement
will not improve. It does indicate that teachers using the computer lab were
doing so ineffectively, that is, selecting instructional strategies that were not
maximizing student achievement. Cuban’s (2001) study of technology use in six
preschools, five kindergartens, two high schools, and one university in Califor-
nia supports this finding. Cuban found that in classrooms of teachers who
considered themselves serious or occasional users of technology, most stu-
dents’ use of computers was for completing assignments, playing games,
conducting Internet research, or exploring CD-ROMS to gather information
(p. 133). It had little to do with the primary instructional task set forth in the
teacher’s instructional objectives. This finding coincides with the activities
occurring in the computer lab at the study school. The Geophysical Science
teachers, who taught over 50% of the students rarely, if ever, used the computer
lab. The Biology teachers, who taught 28% of the students, used the computer
lab primarily for Internet research on such topics as the cell, mitosis, meiosis,
and ecology. The Chemistry teachers, however, utilized the computer lab to
develop more advanced skills through the use of a virtual chemistry lab
program, online simulations for the Gas Laws, balancing equations, and
chemical bonding. The Physics teacher utilized the computer lab in order for
students to understand, identify, and explain force pairs via the West Point
Bridge Builder Program and other online simulations. Unfortunately, the

number of students enrolled in Chemistry or Physics comprised only 20% of the
population. Thus, student activities in the computer lab did not coincide with
strategies to increase higher-order or critical-thinking skills
Teacher Participation in Technology Training
A closer look at the indirect effects of technology (teacher participation in

technology training) indicates that those teachers who did participate in training
were more likely to use the computer lab for instruction—positive beta. While
this may seem counterintuitive—teachers used the computer lab but their
students’ achievement did not increase—it supports the research that stresses
how the computer is used for instruction is crucial for student achievement.
While teachers who participate in technology training may feel more comfort-
able taking their students to the computer lab, it does not imply that they have
mastered the use of the computer in instruction to develop critical-thinking
skills. Smerdon et al. found that elementary teachers were more likely than
secondary teachers to instruct students to use computers or the Internet to
practice drills and to solve problems and analyze data (Smerdon et al., 2000,
p. 104). Other studies found that technology can enable the development of
critical-thinking skills when students use technology presentation and commu-
nication tools to present, publish, and share results (Cradler, McNabb,
Freeman, & Burchett, 2002, p. 48). Christmann and Badgett (1999) studied
the effects of computer-aided instruction (CAI) on student achievement in
science and found that computer simulations were valuable in learning science
as they allowed students to complete experiments that were either too
dangerous to perform in a real laboratory or unrealistic due to the expense of
the materials.
Effective Use
While the secondary school in this study addressed two important issues of
technology use in education—accessibility and professional development—it
did not address the issue of effective use. An interview with the Assistant
Principal for Curriculum Instruction (personal communication, April 2, 2003)
revealed that the type of training teachers received as part of their technology
training centered on basic computer literacy skills, not on integration into
330 Hilton
instruction via simulations, software programs, data collection, and data

analysis. This fact supports the finding that teachers who used the computer lab
were not necessarily improving the type of instruction their students received
in the content area. Instead, emphasis was placed on activities such as word-
processing reports or unstructured Internet research rather than on programs
that developed higher-order thinking skills or that supported classroom instruc-
tional objectives. As Wenglinsky reported, “It seems that for fourth and eighth
grades, computers are used for lower-order activities (i.e., drill and practice)”
(Wenglinsky, 1998, p. 22). The NCES, in its study of the 2000 NAEP Science
Assessment, found that of the two-thirds of the 12th-grade sample taking a
science course in their senior year, those who reported using computers to
collect data, download data, or analyze data had higher scores than students
who reported never doing so (National Center for Education Statistics, 2002,
p. 12). Clearly, effective use is the key to increased achievement.
Relationship Between Training and Use
The indirect effect of teachers who received training using the computer lab
more frequently is well documented in the literature. Andrew Trotter reports
that training seems to make a positive difference to those who received it,
particularly when it came to confidence levels, use of digital content, and
willingness to experiment (Trotter, 1999, p. 40). Fatemi points out that “A lack
of training is the most important obstacle inhibiting the use of digital content.
Teachers who received technology training in the past year are more likely than
teachers who had not to integrate technology into their classroom lessons and
are also more likely to use and rely on digital content for instruction, spend more
time trying out software and searching for Websites to use in class” (Fatemi,
1999, p. 7). Trotter also found that teachers who received training were more
likely to use software to enhance instruction in their classrooms, to rely on
software and the Internet in classroom instruction to a “very great” or
“moderate” extent, and to spend time trying out or teaching themselves about
software as well as searching the Internet for information and resources to use
in the classroom (Trotter, 1999, p. 40). Thus, teachers who received training
used the computer lab for instruction, although their attempts at integration did
not improve achievement, due in part to insufficient training. As Wenglinsky
states, “Technology does matter to academic achievement, with the important
caveat that whether it matters depends on how it is used” (Wenglinsky, 1998,
p. 32).
Implications and Future Trends
The study provides a piece (effective technology use) to a much larger puzzle
(increasing student achievement in science). It indicates that how technology is
used in the classroom is more interesting and important than how much
technology is in the classroom. It provides business and educational personnel
with information on how and where monies should be allocated in local, state,
and federal budgets. Teacher training on effective use of technology must
become a priority if this puzzle piece is to have a significant effect on student
achievement. This study also provides an analysis of ways that teachers and
students use technology at the beginning of the 21st century. Although further
research is needed in the form of a longitudinal study using the same analysis
model, this study offers one possible solution for improving the teaching and
learning process and, hence, increasing student achievement.
References
CEO Forum. (2001). Year 4 Report: Key building blocks for student
achievement in the 21st century. Washington, DC: CEO Forum on
Education and Technology.
Christmann, E., & Badgett, J. (1999). A comparative analysis of the effects of
computer-assisted instruction on student achievement in differing science
and demographical areas. The Journal of Computers in Mathematics
and Science Teaching, 18(2), 135–143.
Cradler, J., McNabb, M., Freeman, M., & Burchett, R. (2002). How does
technology influence student learning. Learning and Leading With
Technology, 29(8), 46–56.
Cuban, L. (1993). Computers meet classroom: Classroom wins. Teachers
College Record, 95, 185–210.
Cuban, L. (2001). Oversold and underused — Computers in the classroom.
Cambridge, MA: Harvard University Press.
Dickey, E., & Kherlopian, R. (1987). A survey of teachers of mathematics,
science, and computers on the uses of computers in grades 5–9 class-
rooms. Educational Technology, 27(6), 10–14.
332 Hilton
Fatemi, E. (1999). Building the digital curriculum. Education Week – Tech-

nology Counts 1999, 19(4), 5–8.
Hedges, L. V., Konstantopoulos, S., & Thoreson, A. (2000, February 26).
Designing studies to measure the implementation and impact of
technology in American schools. Paper presented at the Conference on
the Effectiveness of Educational Technology: Research Designs for the
Next Decade, Menlo Park, CA.
Means, B., Wagner, M., Haertel, G., & Javitz, H. (2000). Investigating the
cumulative impacts of educational technology. Retrieved from May 3,
2003, www.sri.com/policy/designkt/bmeans3.html
Middleton, B. M., & Murray, R. K. (1999). The impact of instructional
technology on student academic achievement in reading and mathematics.
International Journal of Instructional Media, 26(1), 109–116.
National Center for Education Statistics. (2002). The nation’s report card:
Science highlights 2000 (No. NCES 2002452). Washington, DC:
National Center for Education Statistics.
Padilla, C., & Zalles, D. (2001). Capacity-building partnerships: Improv-
ing the evaluation of educational technology – A final report of the
Partnerships Subtask of the Evaluation of Educational Technology
Policy and Practice for the 21st Century (No. P10474). Menlo Park,
CA: SRI International.
Park, J., & Staresina, L. N. (2004). Tracking U.S. trends. Education Week
– Technology Counts 2004, 23(35), 64–67.
Schacter, J. (1999). The impact of education technology on student
achievement — What the most current research has to say. Santa
Monica, CA: Milken Family Foundation.
Skinner, R. A. (2002). Tracking tech trends. Education Week – Technology
Counts 2002, 22(35), 53–56.
Smerdon, B., Cronen, S., Lanahan, L., Anderson, J., Iannotti, N., & Angeles,
J. (2000). Teachers’ tools for the 21st century: A report on teachers’
use of technology (No. NCES 2000102). Washington, DC: National
Center for Education Statistics.
Trotter, A. (1999). Preparing teachers for the digital age. Education Week –
Technology Counts 1999, 19(4), 37–43.
U.S. Bureau of the Census. (1998). Statistical Abstract of the United
States: 1998 (118th edition). Washington, DC: U.S. Bureau of the
Census.
U.S. Department of Education. (2000). eLearning: Putting a world-class

education at the finger tips of all children. Washington, DC: U.S.
Department of Education.
Wenglinsky, H. (1998). Does it compute – The relationship between
educational technology and student achievement in mathematics.
Princeton, NJ: Educational Testing Service.
334 About the Authors
About the Authors
Eshaa M. Alkhalifa is a member of the royal family of Bahrain and the director
of the Information and Data Analysis at the Deanship of Admissions and
Registration at the University of Bahrain. She obtained her PhD in cognitive
science from the University of Edinburgh, UK, and her MSc from The George
Washington University, Washington, DC. She was awarded two science day
awards and has given numerous talks internationally. One of her main research
goals is to break the ground to allow researchers to cross the divide between
the purely theoretical findings of cognitive science and the practical applications
of computerized systems by introducing cognitively informed systems.
***
Liliana Ardissono is an associate professor at the Dipartimento di Informatica

of the Università degli Studi di Torino, Italy, where she obtained her PhD in
computer science. Her research interests include user modeling, adaptive
hypermedia, multiagent systems, and Web services. She is the author of more
than 60 papers published in international journals and conferences. Moreover,
she is co-editor of Special Issue on User Modeling and Personalization for
Television (User Modeling and User-Adapted Interaction, 2004), Personal-
ized Digital Television: Targeting Programs to Individual Users (Kluwer
Academic Publishers, 2004), and Special Issue on Human-Computer Inter-
About the Authors 335
action (Psychnology, 2004). She is program co-chair of the UM 2005

conference in Edinburgh, UK.
Meurig Beynon is a reader in computer science at the University of Warwick,

UK, where he has been employed since 1975. He received his PhD in
mathematics from King’s College London (1973). He founded the Empirical
Modelling Research Group in the 1980s and has subsequently published more
than 70 conference and journal papers relating to this innovative modeling
approach and its applications in fields as diverse as business, engineering, and
educational technology.
Teresa Chambel is an assistant professor at the University of Lisbon,

Portugal, where she received a PhD in informatics in video, hypermedia, and
learning technologies, and a BSc in computer science. Her MSc was in
electrotechnical and computer engineering at the Technical University, Lisbon,
on distributed hypermedia. She has been a member of the Human Computer
Interaction and Multimedia Group at LaSIGE/University of Lisbon, since
1998, and was previously a member of the Multimedia and Interaction
Techniques Group at INESC/Lisbon. Her research interests include multime-
dia and hypermedia, with a special emphasis on video and hypervideo,
distributed systems, and educational technology.
Matthias Finke graduated with a master’s degree from Chalmers University

in Gothenburg, Sweden. He worked at Ericsson Mobile Data in the research
department for Internet applications and then became a PhD student at the
Computer Graphics Center in Germany in 1999. Matthias Finke is involved in
the field of computer-supported collaborative learning (CSCL). He is currently
engaged in research concerning multimedia cooperative working environments
with a focus on collaborative hypervideo applications.
Cristina Gena received a degree in communications at the Università di

Torino, Italy (1998). In 2000, she obtained a master’s degree in Web
technology and security. From January to March 2002, she visited the Institute
for Software Research at the University of California, Irvine. In 2003, she
received a PhD in communications from the Department of Computer Science,
University of Turin. She is currently a temporary researcher at the Department
of Computer Science, University of Turin. Her main research interests include
user modeling, adaptive systems, ubiquitous computing, usability and empirical

evaluation, and semantic Web technologies.
Sébastien George is an associate professor in the Department of Computer

Science at INSA, Lyon (National Institute of Applied Sciences), France. He
is a member of the ICTT Research Laboratory (Collaborative Interaction, E-
Learning, E-Activities). He received his doctoral thesis from the University of
Maine in France. There he designed and developed an environment dedicated
to distant project-based learning. Afterward, he did a postdoctoral fellowship
at the TeleUniversity of Quebec in Canada. His research interests include
computer-supported collaborative learning, computer-mediated communica-
tion, and assistance to human tutoring in distance education.
June K. Hilton has taught all levels of secondary and postsecondary science
and mathematics. She has also served as mathematics and science department
chair in three secondary schools. She holds teaching credentials in Rhode
Island, New Jersey, and California, and also has National Board Certification
in Adolescent/Young Adult Science—Physics. She received her PhD in
education from Claremont Graduate University, California, in December 2003.
Her research centers on the use of technology to increase student achievement.
José Janssen holds a master’s degree in sociology from the University of

Nijmegen, The Netherlands. She worked as a researcher for the Dutch Central
Bureau of Statistics for two years before joining the Open University of The
Netherlands in 1991, where she now works as an educational technologist. Her
main area of work involves e-learning technologies, authoring tools, content
management, and workflows regarding e-learning.
Roland Kaschek is associate professor in the Department of Information

Systems of Massey University (Palmerston North, New Zealand). He gradu-
ated from the University of Oldenburg with a MSc (mathematics), received his
PhD in mathematics from that university, and his advanced PhD (habilitation)
in applied informatics from the University of Klagenfurt, Austria. Dr. Kaschek
taught for nearly 10 years at the University of Klagenfurt, worked for more than
2 years as a business analyst for the Swiss UBS AG in Zurich, and joined
Massey University in 2002. His major fields of interest are conceptual
modeling, including its mathematical and philosophical foundations; develop-
ment methods for information systems; and semantic models for process
modeling.
Kinshuk is associate professor of information systems and director of the

Advanced Learning Technology Research Centre at Massey University, New
Zealand. He has published more than 120 research papers in international
refereed journals, conference proceedings, and book chapters. He is the editor
of the SSCI-indexed Journal of Educational Technology & Society.
Rob Koper holds a master’s degree in educational psychology from Tilburg

University, The Netherlands, and a doctorate in educational technology from
the Open University, The Netherlands. He is a full professor in educational
technology, specifically in e-learning technologies, and is responsible for the
development of Educational Modeling Language (the predecessor of IMS
Learning Design). His research focuses on self-organized distributed learning
networks for lifelong learning, including RTD into software agents, educational
semantic web, and interoperability specifications and standards.
Karen Lee holds a PhD from the University of Surrey. She is currently a
lecturer in nursing and e-learning at the University of Dundee, UK, and an
associate lecturer with the Open University, UK, where she completed an MA
in online and distance education. She is involved in the delivery of a variety of
blended learning modules to undergraduate and postgraduate students on
campus and at a distance, and a consultant on e-learning projects with NHS
Education Scotland and the NHS University. Her particular interests are
in collaborative learning, particularly situated within communities of practice,
and she is currently researching the impact on practice of a national Web-based
education program for clinically based staff.
Chao-Lin Liu received a bachelor’s (1987) and a master’s (1989) degree in

electrical engineering from the National Tsing Hua University, Taiwan, and
earned a certificate (1996) in intelligent vehicle-highway systems and a
doctorate (1998) in computer science and engineering from the University of
Michigan in Ann Arbor. He serves as an associate professor at the National
Chengchi University in Taiwan and an executive committee member of the
Taiwanese Association for Artificial Intelligence. His research interests include
intelligent tutoring systems, probabilistic reasoning, computer-assisted deci-
sion making, natural-language processing, information retrieval, path planning,

and intelligent transportation systems. His homepage is located at http://
faculty.nccu.edu.tw/chaolin.
Jocelyn Manderveld holds a degree in educational psychology (1997,

University of Tilburg, The Netherlands). She managed a number of educational
projects for the Dutch Railways before joining the Open University of The
Netherlands (OUNL) in 1998, where she has been involved in designing and
developing flexible and rich learning environments, and in the development of
the Educational Modeling Language (EML). Jocelyn was OUNL project
manager for the standardization of EML and has participated in standardization
workgroups including IMS/LD and CEN/ISSS.
Allison J. Morgan is a doctoral candidate and a Bunton-Waller fellow in the

School of Information Sciences and Technology at Pennsylvania State Univer-
sity, University Park, USA. She earned a BA in computer-based information
systems from Howard University in Washington, DC. Her research interests
include underrepresented groups and accessibility issues with technology; the
digital divide; the social, cultural, and societal impacts of technology; Web
search engines and information retrieval; and human information behavior. She
has worked previously as an analyst for Accenture Consulting.
Panayiotis Pintelas (http://www.math.upatras.gr/~esdlab/en/members/

pintelas/index.html) is a professor of computer science in the Department of
Mathematics at the University of Patras, Greece. His recent research interests
involve computer technologies in education, concentrating in areas such as
CAL/CAI/CBT/ITS, open and distance learning (ODL), and artificial intelli-
gence techniques in education. He has numerous publications in the aforemen-
tioned fields. He is a member of the British Computer Society and the
Association for the Development of Computer-Based Instructional Systems
(ADCIS), the president of the Hellenic Association of Scientists for Informa-
tion and Communication Technologies in Education, and director of the
Educational Software Development Laboratory.
Chris Roe is a research officer in the Institute of Education at the University

of Warwick, UK. He studied for his doctorate with Meurig Beynon in the
Empirical Modelling Group and received his PhD in computer science in 2003.
He is currently attached to the Centre for New Technologies Research in

Education, and his research interests are primarily concerned with program-
ming in education, to which he has contributed over 10 conference papers.
Hermina Tabachneck-Schijf received her undergraduate degree in psychol-

ogy from Wayne State University in Detroit, Michigan, and an MSc and PhD
in cognitive psychology from Carnegie Mellon University, Pittsburgh, Pennsyl-
vania. After postdoctorates in cognition and in robotics/computer science at
Carnegie Mellon, she moved to The Netherlands and is continuing her
academic career as an assistant professor at the Information Sciences Depart-
ment at Utrecht University, The Netherlands. Her interests lie in representa-
tional aspects, such as the effects of multiple representations on learning, how
to visually represent other-modal information and elicit such information, and
how representing (externalizing) problem aspects affects learning and planning.
Colin Tattersall studied computational science before working on his PhD at

the Computer Based Learning Unit, Leeds University, UK. He subsequently
moved to The Netherlands to work for the research and development arm of
one of the major Dutch telecommunications operators. In the mid-1990s, he
moved into the software industry, working as product manager for a company
specializing in support systems for knowledge-intensive processes. In mid-
2002, he joined the Open University of The Netherlands as an educational
technologist, where his responsibilities cover work-related to innovation in e-
learning and learning technology standardization.
Eileen M. Trauth is a professor of information sciences and technology and

director of the Center for the Information Society at Pennsylvania State
University, University Park, USA. Her research is concerned with societal,
cultural, and organizational influences on information technology and the
information technology professions. Dr. Trauth’s investigation of sociocultural
influences is published in her book, The Culture of an Information Economy:
Influences and Impacts in the Republic of Ireland. She is currently engaged
in a multicountry study of women in the information technology professions in
Australia, New Zealand, Ireland, England, and the U.S. She is associate editor
of Information and Organization and serves on the editorial boards of several
international journals.
Alexei Tretiakov graduated from Moscow State University (Russia) and

received a PhD in information sciences from Tohoku University (Sendai,
Japan). He taught for a number of years at the Graduate School of Information
Sciences of Tohoku University and at the Department of Information Systems
of Massey University (Palmerston North, New Zealand). Dr. Tretiakov’s
research interests range from stochastic cellular automata and object-oriented
database management systems to Web-based systems and e-learning.
Tasos Triantis (http://www.math.upatras.gr/~esdlab/en/members/triantis/

index.html) received his Diploma in Mathematics from the Department of
Mathematics, University of Patras, Greece. He is now a postgraduate student
pursuing a PhD in the field of artificial intelligence and intelligent agents. His
research interests include intelligent agents and multiagent systems, expert
systems, intelligent tutoring systems, intelligent user interface, fuzzy systems,
and virtual reality applications. He joined the ESDLab in 1997.
Bert van den Berg studied theology (1984) at University of Tilburg in The
Netherlands and has been working as a teacher in the vocational education
sector for 5 years. In 1989, he began working for the Dutch Open University
at the Education Technology Expertise Centre. From 1993 to 1999, he worked
for the European Association for Distance Education as project manager in the
field of technology, and coordinated several projects for the European Com-
mission. Back at the Open University his work is currently related to the
application of the educational modeling language and in general with technolo-
gies for e-learning.
René van Es holds a degree in educational science and technology from the
University of Twente, The Netherlands, where he specialized in the application
of media in education. Prior to attending university, he studied electronics at a
vocational level and thereafter technical teacher training. He worked for one of
the major manufacturers of telecommunication equipment as a performance
technologist in the department of customer training and documentation. In
2002, he joined the Open University of The Netherlands to work for the RTD
program as an educational technologist focusing on learning technology stan-
dards.
Christof van Nimwegen started a career in interaction design at the Arts

Academy in Utrecht, The Netherlands. After that and a brief excursion to
artificial intelligence, he studied cognitive psychology at Utrecht University, The
Netherlands from which he received his master’s degree in cognitive ergonom-
ics. Following graduation, he worked several years as usability engineer and
interaction designer in Internet-related businesses in the Netherlands and
abroad. After that, he became a junior teacher at Utrecht University for several
years, after which he enrolled in a PhD project in 2003 concerning represen-
tations in interfaces and human computer interaction in general.
Herre van Oostendorp received his master’s degree in experimental cogni-

tive psychology and his PhD from the University of Amsterdam, The Nether-
lands. He is currently an associate professor at the Center for Content and
Knowledge Engineering at the Institute of Information and Computing Sci-
ences, Utrecht University, The Netherlands. His research interests lie in the
areas of text processing, updating of mental models, cognitive models for
navigating the Web, and learning from screen. He has authored many journal
articles in the area of text comprehension, cognitive ergonomics, and informa-
tion science, and (co)edited several books, including Cognition in a Digital
World (Erlbaum, 2003) and Creation, Use and Deployment of Digital
Information (Erlbaum, 2005).
Michael Verhaart is a principal academic staff member at the Eastern Institute

of Technology, Hawke’s Bay, New Zealand. He has over 20 years of
experience in tertiary education, and is currently enrolled in a PhD program in
information systems. His areas of specialty include the Internet and Web
design, multimedia, and database.
Carmen Zahn (http://www.iwm-kmrc.de/cza.html) is a postdoctoral re-

searcher at the Knowledge Media Research Center in Tuebingen, Germany.
She received her PhD in cognitive psychology from the University of Tuebingen;
her research project, “Knowledge Communication by Hypervideo Technol-
ogy,” was funded by the German Science Foundation (DFG). Her research
interests include learning by collaboratively designing multimedia and
hypermedia, with a special emphasis on advanced digital video and school
education. Dr. Zahn is a member of the German Cognitive Science Society
(GK), the European Association for Research on Learning and Instruction
(EARLI), and the German Communication Association (DGPuK).
342 Index
Index
A B
access 270 Bayesian networks 283
accessibility 238 behavior 76
ACQUA Project 147 behaviourism 187
active learning 215 “berry picking” 267
activity theory 105 Blackboard 193
adaptive concept mapping 128 branching movies 28
adaptive system 144
adaptive Web site 142 C
addressee 105 California 312
addresser 104 CAT (computer-assisted training) 240
agent 221 CAT (computerized adaptive testing) 294
annotated information 51 certification 240
anomalous states of knowledge (ASK) 268 Citizen Kane 28
arc directions 293 code 105
arithmetic 7 cognition 12, 26, 103
association rules 154 cognitive apprenticeship 192
attendance 312 cognitive capacity 7
attention 6 cognitive engagement 1
attention span 14 cognitive learner 11
attitude 87 cognitive load 35
audiovisual information 34 cognitive modes 29
auditory modality 7 cognitive needs 74
aural learning style 30 cognitive recall 50
aural media 16 cognitive science 2, 75
avatars 122 cognitive tools 4
Index 343
cognitively informed system 1 disorientation 35

cognitivism 187 display-based problem solving 78
coherence 35 distance learning 102, 107, 167, 236, 244
collaboration 195, 239 domain expert agent 245
collaborative knowledge building 26 “drill and kill” software 212
collaborative learning 193 Dublin Core 51
color 15 dynamic information spaces (DIS) 40
common ground 31
communication 9, 27, 102, 239 E
communication view 43 e-learning 105, 130, 214
community of practice 31, 191 e-Muse project 213
comprehension 9 education 17
computer programming 216 education level 265
computer support 210 educational scenario 108
computer tools 103 EM (empirical modelling) 211
computer use 312 emotion 12
computer-assisted learning 284 empirical modelling (EM) 211, 219
computer-assisted training (CAT) 240 environment 145, 185
computerized adaptive testing (CAT) 294 expertise 189
conceptual helper 9 eXtensible Markup Language (XML) 51
conceptualization 30 external cognition 77
CONFOR (CONtextual FORum) 104 external memory 77
consistency 35 externalization of information 76
construction 30
constructionism 210 F
constructivism 188
contact 105 familiarity 35
contemporary learner 239 Finland 52
content management systems 67 Freud 12
context 35, 52, 105
contextual information 54
G
contextual reuse 50 gender 11
continuity 35 goals 274
copyright 57 graphical representation 8
curriculum 315 group knowledge 31
group learning 30
D guesses 286
data analysis, 145
deduction 10
H
derived metadata 64 HERM 123
dialogue 30, 267 heterogeneous hypervideo 28
digital asset 51 homogeneous hypervideo 28
digital divide 261, 269 HTIMEL 36
digital media 50 HTML (hypertext markup language) 58
digital video 26 human cognition 29
DIS (dynamic information space) 40 human communication 102
344 Index
human tutor 237 learning phases 29

human-computer interaction 126 learning styles 29
human-computer interface 41 learning track 173
hyperfilms 28 link awareness 35
Hypertext Markup Language (HTML) 58 logo programming 210
Hypertext Markup Language with Time
Extensions 36 M
hypervideo 26, 27 mathematics 316
I Media Vocabulary Markup Language
(MVML) 57
Imagine logo 214 memory 6, 76
individual differences 261 memory encoding 7
individual learning 26, 103 mental models theory 11
information retrieval (IR) 265 mental representation 8
information system 143 message 104
information view 42 metadata 53
instruction 239 metaphor 117
intelligence 2 metaphor enabler 130
“intelligent” tutor 237 metaphor language 119
intelligent tutoring systems (ITS) 236, 283 MI-ADAPT 296
interaction 213 MMOs (multimedia objects) 59
interactivity 15 mobility 239
interference 8 modality 7
internalization of information 76 motivation 31
Internet 51, 128, 238, 262 multiagent educational applications 236
Italian Public Administration 142 multiagent system 237
item response theory 299 multimedia 41
item selection 295 multimedia objects (MMOs) 50, 59
ITS (intelligent tutoring system) 236 multimedia tutoring system 12
Multimedia Vocabulary Markup Language
J (MVML) 59
Jasper Woodbury Series 32 MVML (Media Vocabulary Markup
JAVA 41 Language) 57
MVML (Multimedia Vocabulary Markup
K Language) 59
kinesthetic learning style 30 N

knowledge 2, 87, 109, 120, 145, 190, 239
knowledge management 239 National Telecommunications and Informa-
tion Administration 270
L natural-language comprehension 9
navigation history 38
language 118, 217, 324 navigation view 43
learner 173, 239 network-based tutoring system 240
learner-centred 185 network-like hypervideo 28
learning 26, 103, 202, 284, 294, 312 No Child Left Behind Act 313
learning environment 36
Index 345
O science 2
screenshot 222
online 120 search agents 246
open learning 167 search engine 262
OWL (Web ontology language) 118 security 244
self-directed learner 168
P self-organisation 170
pattern recognition 6 shareable digital assets 54
PDF (portable document file) 55 shared knowledge 31
perception 5 short-term memory 7
performance 239 situated action theory 105
physics 4 situated cognition 191
Picasso 55 situational annotation 55
plan-based problem solving 78 social constructivism 188
PNG (portable network graphic) 55 social constructivist 185
portable document format (PDF) 55 social negotiation 195
portable network graphic (PNG) 55 socio-cognitive 31
Power Principle 213 software agents 122
problem solving 74, 78 spatio-temporal link 26
programming 215 speech comprehension 9
programming language command 217 stimuli 6
property rights 240 student achievement 312
student assessment 296
Q student modeling 283
student population 317
QoS (quality of service) 238 students 193
qualitative parameter 152 swarm-based wayfinding 167
quality of service (QoS) 238 symbol system 33
quantitative parameter 152 synchronization 35
R system architecture 16
RDF (resource definition framework) 51 T

read-write learning style 30 task knowledge 145
reading 9 teaching 312
reasoning 10 technology 312
recognition 5 text comprehension 9
redundancy 14 theory of interpretation 11
representations 8 thinking 11
resource definition framework (RDF) 51 traffic 244
robots 122 tutor 237, 283
route 173
Russia 52 U
S unequal access 270
universal resource locator (URL) 54
scaffolding 4 University of Lisbon 36
scaffolds 196 URL (universal resource locator) 54
346 Index
user 125, 145, 267

user annotation 55
user behavior 267
user interface 43
user-centered approach 142
V
Vanderbilt University 32
VARK perceptual learning styles 30
vCard 51
verbal presentation 16
verbal reasoning 7
video 32
video course material 39
video projects 33
video view 42
video-based hypermedia 26
video-centered page 37
virtual learning 185
virtual learning environment (VLE) 186
visual learning style 30
visual modality 7
visual system 5
VLE (virtual learning environment) 186
W
wayfinding 167
wayfinding support 167
Web information systems (WIS) 117
Web Ontology Language (OWL) 118
Web searching 261
Web site 142, 265
Web-based user interface 43
Welles, O. 28
WIS (Web information system) 117
working memory 7
World Wide Web (WWW) 236
written representation 8
X
XML (eXtensible Markup Language) 51
Single Journal Articles and Case Studies Are
Now Right at Your Fingertips!
Purchase any single journal article or

teaching case for only $18.00!
Idea Group Publishing offers an extensive collection of research articles and
teaching cases that are available for electronic purchase by visiting
www.idea-group.com/articles. You will find over 980 journal articles and
over 275 case studies from over 20 journals available for only $18.00. The
website also offers a new capability of searching journal articles and case
studies by category. To take advantage of this new feature, please use the
link above to search within these available categories:
Business Process Reengineering Data and Database Management

Distance Learning E-commerce
Emerging and Innovative Technologies End User Computing
Healthcare Human Side of IT
Information Resource Management Internet-Based Technologies
IS/IT Planning IT Education
IT Management Knowledge Management
Organization Politics and Culture Software Engineering Tools
Systems Planning Decision Support Systems
Telecommunication and Networking Virtual Offices
Client Server Technology Strategic Information Systems
Design, Implementation
You can now view the table of contents for each journal so it is easier to
locate and purchase one specific article from the journal of your choice.
Case studies are also available through XanEdu, to start building your
perfect coursepack, please visit www.xanedu.com.
For more information, contact [email protected] or 717-533-8845

ext. 10.
www.idea-group.com
Experience the latest full-text research in the fields
of Information Science, Technology & Management
InfoSci-Online
InfoSci-Online is available to libraries to help keep students,
faculty and researchers up-to-date with the latest research in
the ever-growing field of information science, technology, and
management.
The InfoSci-Online collection includes:

Scholarly and scientific book chapters
Peer-reviewed journal articles
Comprehensive teaching cases
Conference proceeding papers
All entries have abstracts and citation information
The full text of every entry is downloadable in .pdf format
Some topics covered: “…The theoretical bent

Business Management of many of the titles
Computer Science covered, and the ease
Education Technologies of adding chapters to
Electronic Commerce reading lists, makes it
InfoSci-Online Environmental IS particularly good for
features: Healthcare Information Systems institutions with strong
Easy-to-use Information Systems information science
6,000+ full-text Library Science curricula.”
entries Multimedia Information Systems — Issues in Science and
Aggregated Public Information Systems Technology Librarianship
Multi-user access Social Science and Technologies
To receive your free 30-day trial access subscription contact:

Andrew Bundy
Email: [email protected] • Phone: 717/533-8845 x29
Web Address: www.infosci-online.com
A PRODUCT OF
Publishers of Idea Group Publishing, Information Science Publishing, CyberTech Publishing, and IRM Press infosci-online.com

Eshaa M. Alkhalifa - Eshaa Alkhalifa - Cognitively Informed Systems - Utilizing Practical Approaches To Enrich Information Presentation and Transfer (2006, Idea Group Publishing) - Libgen - Li

Uploaded by

Copyright:

Available Formats

Eshaa M. Alkhalifa - Eshaa Alkhalifa - Cognitively Informed Systems - Utilizing Practical Approaches To Enrich Information Presentation and Transfer (2006, Idea Group Publishing) - Libgen - Li

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Eshaa M. Alkhalifa - Eshaa Alkhalifa - Cognitively Informed Systems - Utilizing Practical Approaches To Enrich Information Presentation and Transfer (2006, Idea Group Publishing) - Libgen - Li

Uploaded by

Copyright:

Available Formats

Cognitively

IDEA GROUP PUBLISHING

Published in the United States of America by

and in the United Kingdom by

Library of Congress Cataloging-in-Publication Data

Cognitively informed systems : utilizing practical approaches to enrich information presentation

British Cataloguing in Publication Data

Preface ................................................................................................. vii

About the Authors .............................................................................. 334

Index ................................................................................................ 342

It is the laborious life of scientific research that requires us to continually pre-

If the words that are said,

Eshaa M. Alkhalifa, University of Bahrain, Bahrain

Cognitively informed systems as introduced by Alkhalifa (2005b) is a

Paul Thagard (2004) describes cognitive science as the interdisciplinary study

found themselves a place in modern-day systems in online help, decision

Cognitively informed systems is a term that represents all computerized systems

Main Areas of Focus in

1. Perception and Recognition: The aim is to learn from various findings

a. Theories of attention focus on limitations in the cognitive capacity to

a. The first competition was between the two basic approaches to

Albacete and VanLehn (2000a) attempted to utilize the findings on the

a. Comprehension of text or speech involves the analysis of the gram-

The primary aim of utilizing these findings is to either generate language or

a. Johnson-Laird and Byrne (1993) indicate that deductive reasoning is

competing theories and to solve problems.” It is these domains that

a. Jonassen and Grabowski (1993) give a detailed account of basic

to imagine concepts versus the verbalizing learner who likes to learn

Al Balooshi and Alkhalifa (2002) showed that cognitive differences of

A concrete example is represented in multimedia educational systems (Alkhalifa,

A Cognitively Informed Design of a Multimedia System

Amount or Complexity of the Media Offered

• Perception and recognition

Delivering Information in Parallel through the Multiple Windows

• Redundancy could cause interference

• Affects attention focus

Most of the animations were as simple as they could be by including only

Cognitive-load reduction is supposed to occur by actually showing the students

This decision affects the use of aural cues or reading of materials.

• Speech perception issues such as accent and clarity

No aural media was allowed to avoid any external influences.

Verbal Presentation of Material

Evaluation of Multimedia Systems

The consideration of the presented cognitive areas of relevance will necessitate

1st Dimension: System Architecture

• Full description of system modules and complete check of interaction

2nd Dimension: Educational Impact

• Students grouped according to their mean grade in a quiz.

3rd Dimension: Affective Measures

isolation whenever possible to identify the strengths and weaknesses of that

Albacete, P. L., & VanLehn, K. (2000a). The conceptual helper: An intelligent

Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Lawrence Erlbaum.

Thagard, P. (2004). Cognitive science. Stanford Encyclopedia of Philoso-

Carmen Zahn, Knowledge Media Research Center, Germany

Matthias Finke, Computer Graphics Center, Germany

This chapter discusses how advanced digital video technologies, such as

New technologies do not only meet existing needs in terms of communication

tional information elements, such as text, photos, graphics, audio, or other

• Homogeneous hypervideo, where video is the only medium involved,