Designing Computational Models of Collaborative Learning Interaction

spongereasonInternet και Εφαρμογές Web

12 Νοε 2013 (πριν από 4 χρόνια και 7 μήνες)

188 εμφανίσεις


Designing Computational
Models of Collaborative
Learning Interaction


January 7


Boulder, Colorado

CSCL 2002 Workshop: Designing Computational Models


Table of Contents




Personal Information



Presentation papers

Elements of Mediated
Collaboration in Hypermedia Co



Carlo Iacucci, Helen Pain and John Lee


Supporting the development of a working relation in a synchronous electron
collaborative writing environment



Chiel van der Puil, Jerry Andriessen

Identifying and modeling variables in complex CSCL



Jakko van der Pol


Rules of a Dialogue



Heisawn Jeong


Towards Specifyin
g Important Dimensions for Computational Models

of Collaborative Learning



Janet L. Kolodner
Paul J. Camp, Jackie Gray, Jennifer Holbrook


Towards a data model for the evaluation of participatory aspects

of collaborative learning



A. Martínez, J. A. Marcos, I. Garrachón, P. de la Fuente and Y. Dimitriadis


Towards open object
oriented models of collaborative problem solving interaction



A. Dimitracopoulou, N.M. Avouris, V. Komis, C. Fidas


Designing Computational Models of Collaborative Learning Interaction: an Activity
Theory approach



Alain Derycke


Statements of Interest

ational models to support collaborative learning in educational games



Cristina Conati


Research Interests



Ma. de los Angeles Constantino


Computational Modeling for Ill
defined Domains through AI & CSCL



Ilya M. Goldin


Workshop: Designing Computational Models of Collaborative Learning Interaction



André Koehorst


Computational Model in Agent
based Asychronous Virtual Classroom



Hiroaki Ogata, Kenji Matsuura, Yoneo Yano


Fostering computer supported collaborative learning with cooperation scripts and



Armin Weinberger, Frank Fischer, Heinz Mandl


CSCL 2002 Workshop: Designing Computational Models




Computational models are computer
based representations that help to describe, explain, and
analyze patterns of human behavior, and predict futu
re behavior. Recently, we have seen a
considerable amount of interest in using computational models to understand the processes of
collaborative learning. Because these processes are complex, and include coordination of both
task and social learning activi
ties, understanding and supporting group interaction is particularly
difficult. Modeling the processes involved in collaborative learning may help us to better analyze
and dynamically support collaborative learning activities on

Many different type
s of computational models exist. Some help to identify group members'
roles, others help scientists understand specific aspects of collaborative learning, such as
knowledge sharing or cognitive conflict. Computational models that focus specifically on soci
factors may be applied to many different domains, while those designed to facilitate task oriented
interaction may be bound to a particular domain. In this workshop, we will discuss the
requirements for modeling different aspects of interaction. For exa
mple, what data are needed
(e.g. participation statistics, coded dialog, task
based actions) to construct and maintain different
types of models, and how should this data should be represented? This workshop will aim to
address these questions, and others
along these lines that are listed below.

Issue 1: Components of computational models

What types of models exist, and how do they differ?

There are many variables that may help to characterize collaborative interaction. Are there
key indicators (e.g. part
icipation) that allow us to characterize interaction in any case?

What information, and what components are needed to develop models suited to analyze
various aspects of collaborative learning interaction?

How much and what kind of contextual, domain spe
cific information is needed? What are the
benefits gained, and perhaps the generality lost by including contextual information in a
computational model? Can we reuse AI's work and lessons learned?

Issue 2: Moving from the conceptual level to implementatio

What sort of information should be coded and logged, and at what granularity? What
compilation or abstraction methods are needed to construct a computational model from a
logfile describing the group interaction?

How do conceptual models (in terms of
, for example, roles, conflict, constructive
argumentation) translate into computational models that can be represented and manipulated
by a computer?

What is the technical cost (existing techniques, intensive computation) of making the
theoretical indica
tors we deal with operational?

Issue 3: Towards a generic description of interaction

Is a unified representation of interaction possible/desirable ? For example, would it be useful
to have a "Standard Collaborative Interaction Description Language" (SCID
L)? What about a
Standard Collaborative Interaction Log Format (SCILF) ?

How can we compare and evaluate different models of collaboration across research
disciplines? Will a unified scheme help? Would visualization tools based on a unified scheme
help sh
ape pedagogic interventions?

CSCL 2002 Workshop: Designing Computational Models


Personal Information


Jerry Andriessen

Fidas Christos

Alain Derycke

Professeur des Universités (Computer Sciences Prof.)

Laboratoire TRIGONE (laboratory)

Université des sciences et Technologies de Lille

B6 , 59655 Villeneuve d'Ascq, France

phone 33
(0)3 20 43 32 70, Fax 33
(0)3 20 43 32 79


european project DIVILAB, 5th framework IST:

DESS Multimédia & Internet pour le Commerce Electronique:

Angelique Dimitracopoulou

sistant Professor

Learning Technology and Educational Engineering Laboratory.

Department of Education

University of the Aegean

1, Av. Demokratias, 85100, Rhodes, Greece

Yannis Dimitriadis

Carlo Iacucci

Heisawn Jeo

Assistant Professor, Department of Psychology

Hallym University, Chunchon, Korea


phone: 033

Janet Kolodner

Alejandra Martinez

I am an assistan professor of programming at the University of Valladolid, Spain. My research
interests include the application of qualitative perspectives to the ev
aluation of computer
CSCL 2002 Workshop: Designing Computational Models


supported learning, with a focus in the study of social aspects of collaborative learning. I am
also interested in software engineering issues as they relate to the development of educational
software, and more precisely, with educatio
nal software components. I received a computer
science engineering at the University of Valladolid in 1997 and I am currently a doctoral

Helen Pain

Dr Pain is a Senior Lecturer at the University of Edinburgh, in the Institut
e for
Communicative and Collaborative Systems. Her main research is in the field of Artificial
Intelligence and Education. Her interests are in computer
based tools to support learners and
teachers, in particular in helping learners to improve their comm
unication and cognitive skills
though observing and analysing their difficulties and designing, developing and evaluating
learning support tools. Areas in which such research has been conducted include managing
tutorial skills, educational dialogue, studen
t modelling; second language learning (ICALL);
accessibility and special needs.

Jakko van der Pol

After studying educational sciences in Holland, I lectured at the University of Amsterdam and
m currently working at the Utrecht University as a teacher (ICT & education), programmer
(developing different versions of TC3, a computer supported collaborative text composer) and
researcher on the European SCALE
project. Also I have started my phd
amm in the field
of ICT and academic skills. My main field of interest currently is working with asynchronous
electronic discussions.

Furthermore I'm 25 years old, love to sport (bringing my snowboard to Colorado!) and am
looking forward to my first confer

Chiel van der Puil


Christina Conati

Angeles Constantino

María de los Angeles Constantino
González is an Assistant Professor at the Department of
Computer Sciences of ITESM (Monterrey Institute of Technology) Campus Laguna, México.
She received his Ph.D. i
n Artificial Intelligence in December 2000 from ITESM Campus
Monterrey. Her main advisors were Dan Suthers and José Guadalupe Escamilla. As part of her
dissertation program, Angeles worked with Dan Suthers at the Learning Research and
Development Center of

the University of Pittsburgh from summer 1997 to summer 1998.
Based on the Belvedere architecture, she initiated there the implementation of COLER, a
based environment to support students collaborate while solving database conceptual
modeling problems

in small groups. Her thesis is titled “A Computer Coach to Support
Collaboration in a Synchronous Web
based Collaborative Learning Environment”. Her main
research interest is to apply artificial intelligence techniques in the design of educational
e that supports computer
mediated collaborative learning.

CSCL 2002 Workshop: Designing Computational Models


Ilya Goldin

Ilya Goldin is a PhD student in the Intelligent Systems Program and a Graduate Student
Researcher at the Learning Research and Development Center. His resea
rch focuses on
Supported Collaborative Learning. He works on PETE (the Professional Ethics
Tutoring Environment) together with his advisor Kevin Ashley, an LRDC Senior Scientist,
and Rosa Pinkus, director of the Consortium Ethics Program. PETE is
a Web site that aims to
help bioengineering ethics students learn to analyze ethics cases.

Andre Koehorst

Drs. André M. Koehorst

Educational Technologist

Learning Lab

ersiteit Maastricht

André Koehorst has a masters in Psychology and has been working in the field of educational
technology since 1989. He started as a researcher at the Research Institute for Knowledge
Systems, where he worked on

several experimental authoring tools for educational systems.
The educational scope of these systems was wide and ranged from intelligent tutoring to
collaborative knowledge building. He was involved in the development of an assessment
system for project
based learning at the Faculty of General Sciences of Universiteit Maastricht
and held a position as assistant professor in the Faculty of Medicine of Universiteit
Maastricht, where he did the daily coordination and the educational work (design, evaluation)

of a small task force on computer assisted learning and developed and tested a WWW
for patient management problems.

He joined the Learning Lab in 1999 where he works on the development of learning
environments for problem based learning. In current

projects he is involved in the design and
evaluation POLARIS, a system for computer supported collaborative learning.

His research interests include modeling approaches to the specification and evaluation of tools
for learning, interface design for comput
er supported collaborative learning, and
methodological issues in learning tools research.

Hiroaki Ogata

Name: Hiroaki OGATA, Ph.D.

Position: Associate Professor


Dept. of Information Science and Intelligent Systems,

Faculty of Engineering,

The University of Tokushima, Japan

mail: ogata@is.tokushima

URL: http://www

Research interests:

Computer Supported Social Networking, and Virtual Communities for Lifelong Learning.

Armin Weinberger

Armin Weinberg
er, M. A.

Institute for Educational Psychology

CSCL 2002 Workshop: Designing Computational Models


Prof. Dr. H. Mandl

Leopoldstr. 13

80802 Munich / Germany

Mail: weinberg@edupsy.uni


Tel: ++49 89 2180

Fax: ++49 89 2180


Patrick Jermann

Patrick Jermann is currently a research associate at the Learning Research and Developement
Center at the University of Pittsburgh. His work includes design and development of online
learning activities for the Netlearn

project. He is also a doctoral student at TECFA (University
of Geneva, Switzerland) under the direction of Pierre Dillenbourg. His dissertation aims at
studying metacognitive mechanisms in collaborative problem solving. He recently developed
and tested in
teraction meters, i.e. visual representations that are updated dynamically and
inform problem solvers on various aspects of their interaction. In Geneva he contributed to the
design and development of the TECFA virtual campus.

Martin Muehlenbrock

Research Scientist, Xerox Research Centre Europe, Grenoble, France


earch Interests

based collaboration analysis for group learning

supported collaborative learning, shared workspaces environments

Artificial intelligence (in education), machine learning

Contextual computing and devices & intermediation

Selected Publications

Muehlenbrock, M. (2001). Action
based Collaboration Analysis for Group Learning. IOS
Press, Amsterdam.

Muehlenbrock, M., editor (2000). Proceedings
of the workshop Analysis and Modelling
of Collaborative Learning Interactions at the European Conference on Artificial
Intelligence in Education, ECAI
2000, Berlin, Germany, August.

Muehlenbrock, M., & Hoppe, U. (1999). Computer supported interaction analysis of
group problem solving. Proceedings of the Conference on Computer Supported
Collaborative Learning CSCL
99 (pages 398
405). Palo Alto, CA, December.

Amy Soller

Amy Soller is a doctoral candidate in the Intelligent Systems Program at the University of
Pittsburgh. Her research involves developing artificial intelligence techniques for analy
and supporting on
line collaborative learning activities. As part of her dissertation work, she is
training probabilistic models to characterize instances of effective student knowledge sharing.
CSCL 2002 Workshop: Designing Computational Models


Amy began her work in advanced collaborative learning te
chnology while working at the
MITRE Corporation from 1995 to 1998. Since hen, she has continued this research with her
advisor, Alan Lesgold, at the Learning Research and Development Center.

Pierre Dillenbourg

Ulrich Hoppe



Alan Lesgold

CSCL 2002 Workshop: Designing Computational Models


Elements of Mediated Collaboration in

Hypermedia Co

The case of reconciling tutorial dialogue

videos with written literacy

Carlo Iacucci, Helen Pain and John Lee


Institute for Communicating and Collaborative Syste

Division of Informatics, University of Edinburgh


This paper aims to contribute to the discussion of what elements should be included in
models of collaborative learning interactions. We consider the context of educational
hypermedia design f
or supporting the social evolution of multimedia documents. We
address issues of collaborative learning to better understand the nature and the role of
collaboration in hypermedia authoring.

When learners co
author multimedia presentations by selecting an
d combining
representations of different modalities, their practices are mediated by the
representations they share in the authoring process. The means they are provided with,
for creating intertextual links and building discursive structures, play a relev
ant role in
this. Their practices of interpretation and negotiation occur as visible actions on a
distinct 'textual surface'.

Our inquiry draws on an empirical study in which learners were engaged in creating
multimedia presentations out of pre
existing d
ialogue videos and printed notes. We were
concerned with devising particular constructive and collaborative learning practices.
The results of the study suggest that some outcomes of collaboration can be tracked by
looking at the strategies and other resou
rces that learners use in collaboratively
authoring the media
texts. To better account for such strategies, issues from the
traditional literature on CSCL are considered.


When learners collaborate in constructing media texts, their interpre
tations and negotiations
occur as visible actions on a distinct 'textual surface'. Such educational collaborative practices
need to be studied and supported in particular ways, because they are mediated. We have been
looking at how learners collaboratively

build media
texts out of lecture notes and tutorial
dialogue videos.

These learners rely on traces and a breaches that engage co
authors and readers in speculative
activities. Modelling collaboration in these contexts requires an understanding of how part
negotiate and share an understanding of the practices themselves. Relevant aspects upon which
participants rely for making sense of their work are displayed and locally observable in the details
of their activities, and we may study them by posing

ethnomethodological questions. Since our
research is exploratory, and we aim to study learning practices while devising them, we are
setting our research agenda by starting with a reconsideration of relevant concepts we find in the
CSCL literature.

t research has investigated the effectiveness of tutorial dialogue videos as educational
resources (Cox et al. 1998) and the possibility to foster the active participation of `overhearers´ to
such videos (Lee et al. 1998). More generally, numerous systems
have been designed that provide
CSCL 2002 Workshop: Designing Computational Models


ways of supporting collaboration and combining together video and other media
texts (Liestøl
1994, Plowman et al. 1998, Cadiz et al. 2000, LeeTiernan and Grudin 2001). Several problems
affecting learners' experiences have be
en documented in such contexts. They are not all easily
generalizable to our context. However, we note that accounts of those applications point to
similar problems: learners being 'overwhelmed' by distracting functions of the technology, being
by the 'syntax' of the task or being 'disoriented' or 'overloaded' with information in their
'exploratory' activities (e.g., Liestøl and Morrison 1998, Laurillard et al. 2000).

Representing and modelling collaboration can contribute to supporting and o
pening to
inquiry the value of educational hypermedia. Challenges and issues to be addressed to this end,
that have been pointed to in the past (e.g., Joyce 1995) are still relevant:


making knowledge construction overt;


maintaining attention to learning go


providing process
relevant feedback;


giving learners responsibility for contributing to each others' learning.

challenging goal for addressing such problems is to devise collaborative practices in which
learners themselves interpret and frame the r
epresentations in discursive structures and then open
the artefacts produced to inquiry. By engaging learners in such activities, one can try and reify
the collaborative process and trace learning in the artefacts produced. In this endeavour, we make
assumption that the dynamics of collaborative learning often rely upon misunderstanding. In
the perspective we intend to take, the ultimate goal is not to avoid misunderstandings, but to
enhance the process, to facilitate it, and to create opportunities fo
r it. We intend that learners
make mistakes, misunderstand, rephrase, help each other, and thereby they learn. A starting
consideration will be that these problems are context
specific. In the case of hypermedia co
authoring, mediated collaboration require
s specific solutions

Figure 1. An example of the collaborative practices we have been devising is the “documentary
exercise”, which is a

to make sense of the dialogue episodes. The dialogue episodes are
treated as
open texts
, and documentaries ai
m to place the episodes in a linear perspective..

practices are of interest for how learners are engaged in collaborating in different ways
for interpreting and framing mutlimedia objects. Educational objectives lie in the task of
reconciling oral f
orms and written literacy. Such learners' activities are mediated by the
representations they share in the authoring process, and by the means they are provided with for
CSCL 2002 Workshop: Designing Computational Models


creating intertextual references between the multimedia objects. There is a need for b
accounts of such mediations in collaborative learning. We intend to contribute to an
understanding of what the possibilities for active participation by the media
texts' readers are, and
how negotiation and forms of

shared understanding are achieved
between participants.

Overview of the empirical study

We engaged learners in the collaborative construction of hypermedia presentations out of pre
existing tutorial dialogues video
clips and lecture notes (Figure 1 outlines an example). This
paper will no
t go into details of the empirical study, but mention the most relevant aspects. A
requirement that was explicitly set in the task is that the dialogue episodes should be provided
with what they lack in order to be part of a linear presentation. Hence, lea
rners are engaged in
understanding the episodes content and in working as media designers. The tutorial dialogues
contained in the videos addressed the same issues that the authored `documentary´ presentations
were asked to address. Hence, the learners who

authored the hypermedia presentations performed
a reflexive task. In fact, by producing collaboratively multimedia presentations out of these
objects, they give a form to their own learning or to the learning of learners in the videos on the
same issues t
hat they are learning themselves, with the purpose of learning them better.

The empirical study has been carried out by proceeding in an explorative way. We assigned
an open
ended task to learners, and observed ways of structuring the presentations that e
from such activities: argument, narrative and montage. Our research agenda has some
consequences we shall try to make clear. First, the same representations we might want to use to
track or model collaboration

around the discursive structures that

we have identified

impose some structuring on the authoring process, on the shared representations and media that
are used. Such structuring can be achieved in many ways, direct or indirect, intentional or
unintentional. Secondly, by imposing such
structure, inevitably the collaborative practices
become structured in non
neutral ways. These concerns introduce the research questions.

Research questions

The problem of accounting for collaboration in these kinds of contexts is complex because of
the o
penness of the educational task and of the varied nature of the intertextual references
learners can create. Previous reviewers of our research have expressed concerns about the
possibility of determining whether and how the practices are successful on the

educational side.
In particular, do students learn because of the task being constructive? Do they learn because of
some aspect of collaboration? Do they learn thanks to the features of the multimedia presentations
construction as opposed to 'traditional'

essay writing? Or just because they spend more time on
the task? These concerns can be regarded as not merely a problem of determining objective
measures. The very opportunity of concluding anything about the benefits of the collaborative
practices can be

put into question.

Addressing these questions involves making claims for the educational value of the practices.
But this would require a much more ambitious research agenda that we have set so far. It would
require to take the participants' background i
nto account. It would involve, for example, seeing
the proposed co
authoring activity as an intervention that occurs in relation to some pre
classroom activity, and will have rich and somewhat unpredictable effects that depend quite a bit
on this
relation. An improvement to our understanding of these issues could be reached by finding
better terminology. In the following, we reconsider traditional concepts of CSCL. In particular,
we will look for a better account for how negotiation and a shared un
derstanding are achieved
between learners in our context. We will also question what happens to the collaborative practices
if we impose structure on the interface to the shared presentations. Again, reconsidering
traditional accounts from the literature o
n CSCL will provide us with some guidance.

The goal of this paper is to look at collaborative practices that emerged in the course of an
empirical study. We will point out what elements can be tracked in new forms of discourse
CSCL 2002 Workshop: Designing Computational Models


between participants and in t
heir mediated activities through distinctive media. Hence, our hope
is not to use technology to enhance traditional practices. Rather, by such inquiry we intend to
inform the way we address the questions above, and this can involve significantly rethinking

existing models of collaboration. Such phenomena are not new to applications that have been
studied in the CSCL literature. In particular, we reconsider the case of Belvedere as an example
of a system for supporting the construction of a joint presentatio
n (which in the case of Belvedere
is an
) in a structured way (which in the case of Belvedere it aims for students'
). The problem is not only how to define a model that best accounts for students'
collaborative learning. We als
o need to uncover the effects of the interaction of students with the
tools and the models themselves. This ultimately poses ethnomethodological questions. We can
find some answers by observing how learners achieve some goals and some ordered behaviour
ough their competency in dealing with ordinary daily activities. These sequences of conduct
are embodied, situated and locally observable, just as they are in other contexts of mediated
collaboration in building a shared representation. As we argue below,
Belvedere provides
examples from which we can draw our hypotheses.

Background concepts and related research

By researching computational models of collaboration, one can address two questions that
approach the problem of understanding readers' participati
on: (1) How can sharing a
representation lead to a shared understanding? (2) How can a shared space for producing media
texts benefit participants' learning?

Accounts of CSCL have been traditionally founded upon the description of appropriate
patterns. In order to address the role of the media and the tools in achieving a shared
understanding, we need to look at results of CSCL. In particular, we shall consider aspects related
to 'mutuality' and 'grounding', which are critical in mediated commu

“Grounding”, “intersubjectivity”, “common understanding”

Does sharing a representation lead to a shared understanding? It depends on what we mean by
'shared understanding'. If by 'shared understanding' we mean what participants in a conversation

suppose has been assumed 'so far' in the conversation, we might be pointing to the achievement of
common ground. This is a process that has been defined for studying dialogue. According to
Clark and Schaefer (1989), common ground is achieved by sharing as
sumptions about what has
been said (along with other aspects of the dialogue context). In this definition, 'shared' does not
mean completely shared. It may mean just 'shared' for the purpose of accomplishing the current

We might also mean the 'mental

models' that participants are building about the collaborative
task, and about each other. In this case, we are perhaps referring to a different picture, containing
the mental overlaps between these mental models. This property has been called intersubjec
(e.g., Sawyer 2000), and can be used to set some requirements for working together, such as
having a mental model of the partner's mental model. We could also extend this to aspects not
considered in the previous two definitions, such as different
representations of the same shared

CSCL 2002 Workshop: Designing Computational Models


How does structuring shared representations mediate collaboration?

If we want to understand the difference between these terms, we need to take into account the
nature of the collaborative activities. In the
case of mediated collaboration, we can get further
clarifications by looking at studies on 'levels of mutuality of knowledge' with different media.
The work by Schober and Clark (1989) considers various tasks in which communication is
mediated by different

media, and where the different information is shared. They point to relevant
distinctions in the levels of mutuality of the knowledge achieved. In particular, they consider the
difference between the cases in which information (either facts or inferences)

is shared just by
talking, and the cases in which such information is shared through a whiteboard, where facts or
inferences are written. The result they obtained from experimental data is shown in figure 2.
Hence, the effort and the tasks required to gro
und a conversation change with different media.
Similarly, in our study, language gets structured through the co
authoring of the presentations.




(no effort)


Sharing facts in talk

Sharing inferences in talk

Sharing facts on a whiteboard

Sharing inferences on a whiteboard

Figure 2 Redrawn from (Schob
er and Clark 1989). The effort and tasks required to ground a
conversation change with the support of different media. This picture suggests that we need to be
clear about how the media involved will change the collaborative tasks, and not only about their

properties as representations.

Hypertexts constitute a medium for negotiation within communities of learners, in which
members struggle to reach a shared interpretation (Brown and Duguid 1996). Recent research on
document reviewing and publishing points
to various discourse forms such negotiations can take.
For instance, Sumner and Buckingham Shum (1998) explored different socially
based publishing
models in which multimedia documents serve as a medium for negotiation. The results of their
studies have b
een implemented in shared interfaces for managing and reviewing multimedia
documents. The way these interfaces structure the collaborative activities of authors can be better
understood by applying the above considerations to other cases in the CSCL litera
ture, in which
shared representations are collaboratively constructed by learners through an interface. We
consider the case of the Belvedere system, and develop this point further in discussing how
sharing a representation influences the task of achievin
g a shared interpretation.

CSCL 2002 Workshop: Designing Computational Models


The case of Belvedere

Argumentation as a model for collaboration has long been considered in the CSCL literature.
In the Belvedere system

(Suthers et al. 1997), this is done explicitly by gearing collaboration
around the cons
truction of argument as a shared artefact. Belvedere supports collaborative
argumentation, by providing a structured representation of the construction of knowledge in the
form of scientific inquiry. It provides a concept map type representation to graphic
ally lay out the
argument and structuring a scientific inquiry. One of the ways the system can be used is by a
couple of learners who collaboratively reify the structure of the argument through building a
representation on the shared
. Here we in
tend to point to some features of Belvedere that
allow the participants to get a shared understanding of certain aspects of their argument. This is
not only achieved by collaboratively creating a representation, but also by collaboratively
creating ways of

communicating and authoring through the system, as we have witnessed in
recent trials with Belvedere

These benefits are not concerned with the originally intended goal of Belvedere, and with its
way of teaching an epistemology and shaping inquiries, no
r with its proposed ontology of
rhetorical relationships. Rather, they are side
effects of its structured view of argument. By
applying the Belvedere's structured view of argument, learners are forced to accommodate a
wider range of aspects of their inquir
y and ways of communicating that are not foreseen nor
allowed in it. Hence, learners need to spend more effort and accommodate the tasks to ground the
mediated conversation by acting upon the textual surface of the representation they are sharing.
Some 'im
proper' use of the Belvedere interface is not rare among truly engaged learners, who
must force Belvedere's ontology in order to accommodate their specific (and local) difficulties
while arguing. For example, before attempting to identify what categories t
o choose among those
allowed in the system in order to add a justification to the argument, we have witnessed
participants negotiating to determine how to represent the very propositions.

In conclusion, the structured interface has the potential to provid
e opportunities and
motivations to learn more about the process of arguing, than what is embedded in its ontology. In
fact, by looking at how learners argued through the shared interface we could observe that often
there was no real 'reified' structure whi
ch could be properly called an 'argument'. Rather, during
the process of arguing
the system, they were facilitated in monitoring the differences
between their ways of arguing and several problems in communication became apparent and
could be explic
itly addressed through the interface. These benefits were not in the scope of the
original Belvedere's pedagogical aims. But they were often achieved thanks to a creative use that
some learners made of the system



Empirical trials at the 5

Intermedia Resea
rch School, University of Oslo, May 2001.


Such use is a drift from its intended use. In other words, Belvedere allowed for a form of bricolage, as the
phenomenon indicated by Levi
Strauss and that can be found also in studies on information systems and
roupware (Ciborra 1996).

CSCL 2002 Workshop: Designing Computational Models



Our empirical investigatio
ns on open
ended tasks for hypermedia co
authoring suggest that
not only participants collaboratively work out what media
texts are saying. As an additional
feature of the activities, this working out can occur as visible actions on a distinct ‘textual
face’, that is the multimedia presentations. In other words, it doesn’t occur simply as a matter
of thoughtful interpretation, but it is achieved through the practice of actions that change shared

Argument, montage and narrative seem to be bas
ic categories to describe the possible
structures of the authored presentations. Here we will not be concerned with the pedagogical
functions of such structures themselves. Rather, we address the mechanisms of mediated
collaboration, which requires some an
alytical delimitation of the object of study. In fact,
collaborative learning can take place in various forms. Different kinds of shared understanding
can be achieved, through different kinds of interaction patterns, and technology
mediated co
What seems specific to our context in this respect is that learners as authors act as
interpreters and translators. They articulate intertextual relationships in discursive structures in
which they aim to fix meanings. These features call for different met
hods and resources
according to argument, montage or narrative. In particular, referring to their sub
different forms of access, visibility, understanding and agreement are involved.

Reconsidering interaction patterns and other phenomena from
the CSCL literature

How can a shared space for producing such media
texts favour collaborative learning
between participants? In order to address this question, we can draw from our previous
considerations and look at patterns of behaviour that have tradi
tionally been considered in
accounts of CSCL (e.g. Dillenbourg 1999). In particular, let us consider the aspects concerned
with negotiation and argumentation:


relates to how learners jointly decide about facts, methods, goals, or
inferences, at

the level of the task at hand, or through

. These matters raise questions about the nature of the collaborative
setting, such as the ‘degree of negotiability’, the issue of authority, and
possibilities for negotiating how to in
teract (meta
communication), or how to
space for negotiation

(Dillenbourg and Baker 1996). Negotiation can be
inhibited by several aspects of the collaborative situation. For example, if partners
play too definite roles, or too trivial tasks, there i
s less to disagree upon. Also, in
tasks with too semantically rich and ambiguous representations there is nothing to
disagree upon. The boundary between misunderstanding and disagreement is
narrow. If we do not understand each other, one cannot say that we

(Dillenbourg and Traum 1996).

This is relevant to how different modes of production of hypermedia presentation allow for
different degrees of mutuality of knowledge (the definitions of the concepts mentioned in figure
2). The resources learners us
e to fix meaning in such tasks include quotations



structures mentioned above.


relates to how learners can be dealing with quite different tasks.
For example, identifying the form of arguments and the criteria of truth; the
of representing propositions; how to set up alternatives for the unfolding of the
arguments, such as the different possibilities of finding evidence or justification.

CSCL 2002 Workshop: Designing Computational Models


A critical issue is how to represent collaboration in these cases. A first considera
tion can be
drawn from the case of Belvedere mentioned above, which suggests at least two points in our


Modes of collaboration can happen in opposition to the structure of the interface. We should
be modest when defining an epistemology that learner
s should follow. Also, it is worth
seeking insights from the details of situated and locally observable participants’ practices to
discover the modes of collaboration. In fact, it’s very easy to obtain some non
structuring of the language and parti
cipation through a structured medium, but the most
valuable effect for educational purposes could well be different from those intended in the
first place, and they can be inspected during the unfolding of the practice.


Rather than reifying the thoughts
and the products of the collaboration, and regarding them as
‘the’ valuable results of the collaborative learning task, it could be more proficient to find
ways of reifying the collaborative process in the shared representations. This will involve
in detail at how the collaborative practices are mediated in situated interactions.

The practice of intertextuality in mediated collaboration

How can we build upon the above considerations? Some relevant aspects of the authoring
practices must be outlin
ed in order to understand how this can be supported. For instance, they
are built upon a set of ‘autonomous’ items, which are chosen among others. However, they can
only be selected among a given set of multimedia items and it is difficult to improve upon
simplify their aesthetic


Several activities in the task are based upon the practice of
. The raw matter
includes video streams of dialogues that are often meandering and inconclusive discussions
characterised by digression. Often
, quotations are operated in order to reconcile them with
lecture notes. Here are some relevant aspects of the practice of quotations that aim to overcome
this aspect and affect the collaborative practices:


Often quotations focus the readers' attention on
textual functioning rather than on text


The quotations generate a tension between the original integrity of the video episode and
the possibility of reintegration.


Quotations constitute both a trace and a breach, and engage readers in spe
activities to reconstruct the original context and the intended interpretation.

The following considerations address the ‘conditions’ under which learners’ authoring
process takes place. These are direct consequences of how the collaborative prac
tices is set in the
first place:

Seeking unity

The main activities in which learners are engaged aim to create unity
and coherence among a given set of multimedia items. For instance, more low
level activities
aim to isolate those items for which causal

or associative links of some type can be found or
established with the others. However, this is not the only paradigm through which learners
. It depends on the genre of presentation, and in order to share an understanding of
the activity, the g
enre itself


be negotiated.

Inaccessibility of the “a priori” conditions

By the nature of this practice, learners are
prompted to choose an interpretation and display a point of view. But the elementary objects
are chosen, not produced, by the same
author 'a priori'. This is a considerable limitation in
their authoring activity. As a consequence of it, they will be engaged in framing and
annotating such items in order to achieve a point of view and maybe favour a 'desired
reading': this can be a poin
t of misinterpretation between participants.

CSCL 2002 Workshop: Designing Computational Models


Interpretation as production

If we can regard such products as being authored, and
provided with aesthetic qualities by which we can judge their authors, it is because they are
the end of an interpretation. S
uch interpretation is at the same time a production because it
has realised itself through selection and composition, and this is the very nature of such an
artful task. By framing the videos in hypertexts the dialogue utterances get ‘relocated’ (Linell
98), learners de
contextualise and re
contextualise them.


mediated collaboration in hypermedia co
authoring, one can rely on the
detailed and contextualised observation of learners' practices as they happen as mediated
ivities through distinctive media. Such observations point to what elements of collaboration
can be tracked, and to new forms of discourse between participants that can be identified. The
benefits of such observations seem to be limited by the possibility
of using them in addressing
more general questions and in applying collaboration models. For example, how to make claims
for the educational value of the effects? How to be informed by similar collaborative practices
that are institutionalised and accounte
d for in other contexts? To these ends, we have advocated
abandoning traditional learning
outcome measures. Instead, we aimed to describe how through
hypermedia co
authoring learners act as interpreters and translators, by relying for example on
how they c
reate intertextual references. The hope is to provide a basis for addressing higher level
issues. For instance, such detailed observations are necessary if we are to understand how much
of the collaborative process can be reified in the media, or what are
the roles of the different
discursive structures that are created in the produced media texts. Motivations for this
methodological interest can be found in the CSCL literature. In particular, we have drawn from
the observation that accounting for how learn
ing occurs through the use of Belvedere, in some
cases, can need to be informed by how participants make sense out of the collaborative practice
through detailed and situated activities. And these are not carried out by applying the same
categories provide
d by the system.


Baker, M.J. and Lund, K. (1996). Flexibly structuring the interaction in a CSCL environment. In P. Brna,
A. Paiva and J. Self (eds),
Proceedings of the European Conference on Artificial Intelligence in
sbon, Portugal, Sept 20

Oct 2, 401

Brown, J.S. and Duguid, P. (1996). The Social life of Documents.
First Monday
, 1, 1996.

Cadiz, J.J., Balachandran, A., Sanocki, E., Gupta, A. and Grudin, J. (200
Distance Learning Through
Distributed Collaborative Video Viewing.

CSCW 2000
, 135

Ciborra, C.U. (1994).
From Thinking to Tinkering, In C.U. Ciborra and T. Jelassi (eds.)
Information Systems
, J. Wiley.

Clark, H.H. and Schaefer, E.F. (1989). Contributing to Discourse. In
Cognitive Science

), pp. 259

Cox, R., McKendree, J., Tobin, R., Lee, J. and Mayes, T. (1998). Vicarious learning from dialogue and
discourse: A controlled comparison,
Instructional Science

, Kluwer Academic Publishers, 431

Dillenbourg, P. and Baker, M.J. (1996)
. Negotiation spaces in Human
Computer Collaborative Learning.
Proceedings of the International Conference on Cooperative Systems (COOP’96)
, Juan
France, June 12

Dillenbourg, P. and Traum, D. (1996). Grounding in multi
modal task

collaboration. In P. Brna,
A. Paiva and J. Self (eds),
Proceedings of the European Conference on Artificial Intelligence in
Lisbon, Portugal, Sept 20

Oct 2, pp 401

Dillenbourg, P. (1999). What do you mean by ‘collaborative learning’? In
P. Dillenbourg (ed)
learning: Cognitive and Computational Approaches
, Oxford: Elsevier, 1

Dourish, P. and Button, G. (1998). On Technomethodology: Foundational Relationships between
Ethnomethodology and System Design,
Human Computer Inter
(4), 395

CSCL 2002 Workshop: Designing Computational Models


Joyce, M. (1999). Siren Shapes: Exploratory and Constructive Hypertexts,
Academic Computing
, no. 4,
14, 37

, D., Stratford, M., Luckin, R., Plowman, L. and Taylor, J. (2000). Affordances for Learning in a
inear Narrative Medium.
Journal of Interactive Media in Education
, 2000 (2).

Lee, J., McKendree, J., Dineen, F. and Cox, R. (1998). Vicarious Learning: dialogue and multimodality,
Proceedings of the Second International Conference in Computer Multimodal Co

LeeTiernan, S. and Grudin, J. (2001). Fostering Engagement in Asynchronous Learning Through
Collaborative Multimedia Annotation,

Liestøl, G. (1994). Aesthetic and Rhetorical Aspects of Linking Video in Hypermedia, In
s of

Liestøl, G. and Morrison, A. (1998). Making Sense of Multimedia Training, In
Proceedings of ICCE’98

Linell, P. (1998).
Approaching Dialogue: Talk, Interaction and Contexts in Dialogical Perspectives
, John
Benjamins: Amsterdam.

Plowman, L., L
uckin, R., Laurillard, D., Stratfold, M. and Taylor, J. (1999). Designing Multimedia for
Learning: Narrative Guidance and Narrative Construction.
Proceedings of CHI’99: ACM Conference
on Human Factors in Computing Systems
, 310

Sawyer, R.K. (2000). Em
ergence in collective behavior: A case study of improvisational theater
performance. Paper presented at the
Annual meeting of the American Sociological Association

Schober, M.F. and Clark, H.H. (1989). Understanding by addressees and overhearers.
21, 211

Sumner, T. and Buckingham Shum, S. (1998). From Documents to Discourse: Shifting Conceptions of
Scholarly Publishing. In

Proceedings of CHI 98: Human Factors in Computing Systems
, 18
23 April,
1998, Los Angeles, CA. ACM Press: New


Suthers, D., Erdosne Toth, E. and Weiner, A. (1997). An Integrated Approach to Implementing
Collaborative Inquiry in the Classroom.
December 10
, Toronto

CSCL 2002 Workshop: Designing Computational Models


Supporting the development of a working relation in a
synchronous electronic co
llaborative writing

Chiel van der Puil, Jerry Andriessen

Dept. of Educational Sciences,
Utrecht University;


In this

contribution, we report on the design of an experiment in which we examine the
development of interpersonal communication during electronic collaborative writing. The work is
based on earlier research findings (Veerman, 2000; Andriessen, Erkens, Peters, v
an de Laak &
Coirier, in press) with university students. In that research we could distinguish pairs of students
that focused on the meaning of conceptual information, trying to engage in epistemic discussions,
and pairs of students whose main goal was to

deliver a result, such as the solution to the problem,
or a common text. While the first type of collaborators more closely behaved according to the
goal of the task set by the experimenters, they also appeared more sensitive to characteristics of
the CSC
L environment they were using. As a consequence, the focus of their discussions also
involved many problems that hindered meaningful discussion, such as technical matters, planning
the task, and issues concerning the synchronization of the discussion. The
purpose of this
experiment is to investigate the effect of working with interfaces that aim at supporting (1)
students in developing a better working relationship, in terms of regulation of social, task
oriented, and conceptual common ground; and (2) stude
nts in becoming more engaged in
epistemic activities. The experiment is part of a project in which the regulation and coordination
of interaction during several collaborative learning tasks is studied.


Developing a shared understanding of concep
ts is an important process in collaborative
learning, but it seems that realizing this type of interaction (involving negotiation about the
meaning of concepts) is especially difficult to in computer
mediated communication (Veerman,
2000). Veermans’ analys
is of focus and focus
shifts indicates that some characteristics of the
synchronous chat
interface may hinder the coordination and regulation needed to develop a
shared understanding.

In trying to make sense of how the interface exerts, or is able to exert
, its influence on
mediated interaction, we investigate how the working
relationship between
collaborative pairs evolves during a collaborative writing task. We hypothesize that better support
of the social and task oriented aspects of collaborati
on, by offering appropriate sentence
at the interface, will foster the development of better communication, which in turn will foster
negotiation and co
construction of knowledge.

The hypothesis is derived from Veermans (2000) comparison of collabo
rators who pursued
making activities versus the majority of collaborators who focused on the use of
concepts. The latter were better able to reach their goals. Two problems are observable, a) most
collaborators do not pursue meaning making activiti
es and b) when pursuing this strategy
collaborators run into problems. The tasks involved using CMC for open
problem solving, and
defending decisions in the context of designing an educational computer program.

An interaction thus seems to exist between th
e way students perceive a task (focus on task
completion and the use of concepts versus focus on concept meaning) and their ability to handle
the task in a CMC
setting. A possible explanation for this phenomenon is offered by grounding
theory. According to

the grounding criterion of Clark and Schaefer (1987)
“speaker and
addressees mutually believe that the addressees have understood what the speaker meant to a
CSCL 2002 Workshop: Designing Computational Models


criterion sufficient for current purposes”
. This would mean that the meaning
making dyads in
mans’ research set their goals higher regarding mutual understanding compared to dyads
focusing on the use of concepts. It is from here a next step to suppose that the higher goal
(according to the grounding criterion) of meaning
making asks for a better w
orking relation
between participants, a high
level working relationship that might be difficult to develop and
maintain through computer mediated interaction, leading to not pursuing this type of interaction,
and difficulties in maintaining such interactio

Negotiation in collaborative writing

Composing and discussing information during composing of an argumentative text requires
collaborators to form an opinion to defend, to evaluate arguments pro and contra, and to assess
information to its’ relevance in

the local and global text
context. In text composition, several
subtasks, like idea
generation and revision, combine to result in a coherent and deep
understanding of the subject. The role of a partner with whom one works synchronously on the
same text is

not unambiguous, he or she can be facilitating but also inhibiting. Research into
brainstorming, for example, shows that groups do not always perform better than individuals do,
showing the importance of the framework in which such a task is performed.

esearch by Andriessen et al. (in press) discusses the process of
knowledge negotiation

collaborative writing, which they describe as discussion for agreement about concepts and their
interpretation in the context of a learning task. Negotiation may be a
bout problem solutions,
meanings of concepts, and other things. Argumentation is one of the forms that dialogues in
negotiation may take (Baker, 1994; Dillenbourg and Baker, 1996). The extent to which
negotiation fosters individual learning depends on spec
ific meaningful exchanges between
individual participants. The analysis showed that university students that collaboratively produce
an argumentative text through CMC do not tend to negotiate each item of content extensively.
When they do, as in elaborate
argumentation, this immediately increases the variety of arguments
in the discussion, in terms of their orientation, pro

and counter the main position of the text.
Also, information discussed as elaborate argumentation is immediately entered into the text

construction. However, in most of the discussion, participants are quite willing to accept most
content put forward by the other.

Of course, from a learning perspective, such an attitude is undesirable, and has to be
overcome by redesigning the edu
cational context of such tasks (Andriessen, Baker & Suthers, in
press). Our aim is more modest here, as we try to investigate whether, within the constraints of a
school environment, redesigning the interface could affect this minimal behavior. For example
Jermann & Dillenbourg (in press) show that when the computer tool (by its design) orients the
students towards expressing differences in opinion, students abandon their usual ‘standing pat’
attitude (Baker, 1994). Certain tool features served to perceive

the task in a different way, which
worked out in the context of the educational scenario designed by Jermann & Dillenbourg.

Establishing a working relationship through chat

Working together on an argumentative text requires more than just interaction abou
t content,
probably as important is the interaction that communicates, direct or indirect, how the ‘together’
is shaped. This is not only of interest to the researcher (with a birds eye view), also the participant
relies on this shaping of a social collabo
rative context for how to proceed. As proposed before,
chat/CMC as an interaction medium could possibly be interfering this process, resulting in below
standard conceptual interaction. Compared to ‘natural’ face
face interaction, chat lacks all non
ual cues. Generally perceived positive effects of chat are:

An equalizing effect (concerning interpersonal interaction): participants are more free to
express themselves, feel more anonymous, possibilities of exercising power are limited.

An effect of hav
ing to be more explicit in communicating meaning an intentions

CSCL 2002 Workshop: Designing Computational Models


As a negative counterbalance, it is more difficult to build an interpersonal relationship via
chat. Trust, responsibility for the other and for the collaborative work are built much easier in a

face situation. Chat, for several possible reasons, causes difficulties for the coordination
and regulation of learning interaction:

On a linguistic level, collaborators might have to adapt their way of interaction to the
medium, the written commu
nication medium might for example invite focus changes
from topic to topic easier, because of lack of cues available in f2f

The cost of subsequent repair, a form of grounding (the correct and sufficient
understanding of each other), is high co
mpared to face
face interaction. This and other
mechanisms might not be utilized as much as is necessary.

This could then disturb the working relationship on an interpersonal level.

The above illustrates how on different levels (linguistics,

interaction and interpersonal),
reasons can be found for the failure to establish a positive working relationship. An option to
overcome these kind of problems is to support chat
communication. We have chosen to use
sentence openers for they can help main
tain focus and they can lower the social cost of making a
repair (or the initiating costs). In the experiment two sets of sentence openers are tested; one set
focuses on the ‘we’ aspect, the other on the ‘relevance to the task’. Simple comments like ‘good
idea’ or ‘good plan’ are an easy way to stimulate the partner and maintain a positive atmosphere,
and a ‘grmbl’ is an easy way to show annoyance, that otherwise might indirectly show via cynical
remarks, that are difficult to ‘read’ via chat. Sentence open
ers can lower borders (socially and
initiating) for showing uncertainty, for example “what do you mean by…” or “I do not
understand…” prompts for sharing information and thoughts, while at the same time focusing on
a specific topic.

The experiment

The foc
us of this research is on how the environment enables interaction between
collaborators and how it enables handling and presentation of information/knowledge. Here we
use a synchronous writing tool for collaborative dyads, using chat for communication and
shared text editor for writing/information presentation/handling. Subjects are 152 high
students (15
17 years old), being immersed in a studio scenario, characterized by frequent
collaborative and project
based assignments. Topics are negotiated w
ith the teacher and the task
is part of regular classroom assignments, to be used and rated by the teacher. In the design, (1)

CSCL 2002 Workshop: Designing Computational Models


individual background knowledge on a topic is offered as a starting point for the (2) individual
construction of diagrammatic rep
resentations. Then, (3) pairs of students are formed on the basis
of the similar degree of elaboration of their individual diagrams. Subsequently, (4) these pairs of
students are confronted with each others’ representations and are asked to collaborate on
production of an argumentative text, with the comparison of the diagrams as a starting point.
After the collaborative session, (5) students are asked to revise their diagrams according to the
changes in their personal viewpoints on the basis of the dis
cussion. Students are informed the
goal of the task is to further develop and refine their knowledge of the topic.

Collaboration is further supported at the level of the interface through sentence openers
focusing on social and task oriented aspects of han
dling information. We aim to focus attention to
these aspects of collaboration, ultimately aiming at elaborate negotiation about the meanings of
concepts. The analysis focuses on the relationship between social, task

and content
interaction by use

of sentence openers (see below).


In order to reach insight in how the experimental manipulation
the use of sentence openers

influences collaborative interaction, the goal of the analysis is to link the developing social
relationship and task re
lationship to the quality and success of content development as a
measurement of collaborative interaction. Initially episodes will be distinguished (by topic
change) that are social, task
related or epistemic in nature. Social and task relationship (toget
the working relationship) are viewed as characterizations of episodes. An analysis scheme has to
be developed to qualify episodes on how they contribute to a positive social environment and a
clear grasp and organization of the task. ‘Sharing personal

information’, ‘gossip’ and
‘commenting on each other’, are categories perceivable in the development of social relationship,
while ‘discussing/agreeing on goals and task perception’ and ‘making deals about division of
work/roles’, can point toward the dev
elopment of a task

Success of content development, the effectiveness in relation to delivered products, is
measured individually, using the changes in the initial diagrams the students made after the
collaborative writing task. The changes in
diagrams thus provide a measurement of perceived
important changes in their knowledge representation. By comparing these changes between
collaborators, we expect to be able to classify the outcomes of the collaborative process as:

Mutually convergent: both

diagrams change equally and become more similar, indicating
an equal collaborative process of co

Single sided convergent: one diagram changes more than the other, becoming more
similar, indicating a non
equal process of explaining, rather tha
n co

Divergent: diagrams grow dissimilar, indicating problems in the collaborative process
like unresolved disagreement or problems in developing a shared understanding.

On the group level, the developed content of the collaboratively produce
d text is another
measure of success in collaborative content development.

The quality of content development in interaction (through chat and the use of the text editor)
will be assessed in relation to the collaborativeness of that development. For conten
interaction episodes the (collaborative) nature of developing knowledge co
construction will be
scored, categorized as follows:


Adding information (without interaction)


Explaining (transmission of information)


Building shared knowledge (collabo
rative knowledge construction)


Triggering (non
content related social stimulation)


Grounding (for example, rephrasing or repeating)


Building on each other (content related elaboration/argumentation)

This categorization then (where 3c represents the collabo
rative end of the scale) can be
related to the changes in diagrams and the appearance of content in the text, showing if the
CSCL 2002 Workshop: Designing Computational Models


collaborativeness of epistemic interaction leaves an impression on individual knowledge and the
‘group’ knowledge respectively.

ter this global analysis of the session, a more detailed analysis will take place, zooming in
on the epistemic episodes. Within these episodes, for each contribution social and task aspects as
well as epistemic aspects of each contribution are analyzed to
see how on a local level, the
buildup of a working relationship shapes successful and qualitatively good knowledge

Effects of the interface manipulation will be tested against differences between student
changes in diagrammatic repres
entation, on the classification of epistemic interaction episodes,
and on scores of (local and global) social and task related interaction, every step coming closer to
what we directly try to manipulate. The effect will be tested for a) availability of sen
tence openers
(experimental conditions) and b) counts of the actual use of sentence openers.

What is still neglected in this analysis is an explicit account of the writing process, and how
this influences collaboration. This factor will be part of future a
nalysis, making clear if and how
collaborative learning interaction is affected by the specific context of collaborative writing.

A modeling proposal: from transmission to studio

Establishment of a better working relationship between participants in a col
laborative task is
indirectly affected by the educational context, through the participants’ perception of the task.
Also, from the modeling perspective, desired user support is derived from the didactical scenario
that a user is working in. The propositio
n that we briefly like to develop here is that the didactical
question is primary for deciding about modeling issues. Desired user support defines the
knowledge a system needs in order to model a user and user progress. In our case, users can be
zed as being involved in a gradual shift from learning as knowledge transmission to a
studio scenario (Andriessen & Sandberg, 1999; Andriessen, Baker & Suthers, in press).

In transmission, the goal of education is the development of personal understanding

of expert
knowledge. In transmission, collaboration is mainly considered as a reasoning or problem solving
process, in which learners try to articulate strong and relevant arguments to arrive at approved
conclusions or solutions. The situational constrain
t that drives the argumentation in the first place
is the teacher, or in a more abstract sense, the domain expert, not the interests or personal goals of
the participants themselves. Hence, what is being confronted during collaboration between two

concerns not only their personal representations, but also those of experts, represented in
the teacher’s words or the textbook contents. In other words, for the participants the issue at stake
during argumentation in knowledge transmission is: is this in
? Ideally, this
would require on
line expert feedback and prompting, originally one of the main goals of
Intelligent Tutoring Systems (Andriessen & Sandberg, 1999). Collaboration in transmission does
not work effectively because transmissi
on focuses on individual knowledge and the correctness of
that knowledge.

The second educational scenario distinguished by Andriessen and Sandberg (1999) was
. Instead of focusing on personal understanding of normative information as in
ission, the studio scenario concerns the acquisition of self
regulated learning skills. The
learning environment is seen as a collection of tools and tasks to be used by learners to adapt their
learning to their needs and goals. Collaborative learning is o
ne of the skills that learners are
supposed to master. Instead of collaboration being acquired as a by
product of knowledge
acquisition, now it is addressed in a number of different tasks in which different functions of
collaboration and roles of participa
nts are the focus of attention. While studio scenarios
presuppose that learners have the ability to understand given information, the new focus is now
on their individual roles as learners and collaborators that have to apply and extend their

to different tasks. Typically, the tasks are open, with no fixed solutions or solution
paths, and allow adaptation to specific individual knowledge and skills. In studio, learners are
learning to arrive at
shared understanding
, by dialogue and communicat
ion of information and
CSCL 2002 Workshop: Designing Computational Models


knowledge through the use of various tools available in the environment. This involves, among
other things, discussing different viewpoints, and integrating personal beliefs, other peoples’
ideas and information from different source
s in a process of argumentative learning (Stahl, 2000).
In studio, the context is ripe for the development and use of CSCL environments that (1)

(as opposed to enable) collaborative (and argumentative) learning, (2) allow
generic t

that are not task

and domain
specific as in transmission, and (3) do not treat
collaboration as a single type of activity, but instead allow attention for

of users,
differentiation of user

and distinction of different
in task
based discussions. ). The
focus of support in studio is on the

of collaborative argumentation (when, why, and
how to argue, and for what purpose?) with the aim of (1) reducing the space of misunderstanding
during collaborative learning (Dillenb
ourg, 2000), (2) increasing metacognitive awareness of
learning, collaboration and the role of argumentation (e.g. Kuhn, Shaw & Felton, 1997), and (3)
preparing learners for the practice of group work and knowledge building discourse (e.g. Bereiter,
in pre

We suppose that our users are in some transition phase between knowledge transmission and
studio. This requires explicit support with respect to the

of collaborative learning. A
system designed to support such learning serves to gradually c
hanging users’ representation of
collaboration, by focusing on the process of communication, monitoring user roles in terms of
ontologies that represent features of dialogue episodes that characterize types of collaboration in a
task context or task phase.

Monitoring growing awareness of collaboration may be seen as
collecting evidence (in user profiles) of richer dialogues (argumentative, epistemic, social, task
related) during task execution. Figure 1 and table 1 may somewhat clarify how we view such a
deling process. Table 1 (below) shows expected/observed behavior of students in the current
experiment in the context of the pedagogical
ideological transformation (further explained in
figure 1). The current research should provide information about the i
nterrelatedness of working
relationship and content development and the effectiveness of ‘static feedback’ using sentence
openers. The table shows we expect no effect of sentence openers when students engage in a
transmission scenario. More dynamic feedbac
k, based on computational modeling could,
assessing the used model, give more situation dependent feedback, prompting students to engage
in or (even better) to develop studio
scenario behavior.

In the perspective of modeling collaboration, we do not think
a general model for
collaboration is feasible, as collaboration stands for different forms of interaction and
communication in many different specific task contexts. A general model would seem to
characterize collaboration as a uniform concept, somewhat si
milar to writing, communicating, or
acquisition of metacognitive skills. In contrast, what we propose (for discussion in the workshop)
at least for the studio scenario, is the development of task and user
specific collaboration

that allow a syst
em to evaluate user
roles and user development in terms of progress in
the context of a collaborative learning scenario. This only works in appropriate educational
contexts that help to tune user task perception towards desired goals.

Conclusion & Discussi
on for the workshop (Issue 1)

In this paper we argued that, next to a content related product, collaborators have to build a
working relationship too. This working relationship forms the basis for collaborative learning
activities and involves general soci
al interaction and a more situation
specific shared task
perception. Hypothesized is that if these components of the working relationship are better
developed between collaborative pairs, they will perform better in co
constructive activities and
more spec
ific in meaning oriented discussion.

By manipulating the communication interface, by exposing student
pairs with sentence
openers that focus either on the social aspect of interaction or on the task aspect of interaction, we
hope to stimulate a better work
ing relationship compared to subjects not receiving such stimulus.
Dependent on results concerning the effectiveness of the interface manipulation, a next
CSCL 2002 Workshop: Designing Computational Models


experiment could focus on awareness of different roles in the collaborative process (like

social cohesion, or task management) or in the writing process (idea generation, idea
development (argumentation) integration, revision). This would extend the current experiment,
where just one aspect of collaboration is stimulated in a more interactive
stimulus, where choice
of roles and their complementarities should help maintaining focus, both on epistemic aspects of
the task and on social and task
related coordination and regulation.

We have four questions for discussion in the workshop. First, the d
efinition of collaborative
learning determines how to construct a model about the interaction that constitutes collaborative
learning. Here the ‘dichotomy’ collaboration (synchronous)
cooperation (division of tasks) is an
issue, as is role taking or any o
ther division of work.
What kind of interaction do we expect to
characterize the nature of the collaborative task and setting?

Second, the medium used for
interaction and its affordances should be taken into account.
What type of interaction can one
ly expect through a specific medium?

Third, the question is
what information can be
reliably measured and used for model construction
; participation is easily to measure, but what
does it say about the contribution to the collaborative learning process? Wh
at type of model or
theory is needed? Finally, but most importantly, the constraints of the setting point at the
(didactical) issue of giving feedback:
what aspects of the interaction should we be supporting, and
what do we leave for students to develop th

When, for example, the computer gives
cues about every step in the learning process, will students learn how to plan their work?

CSCL 2002 Workshop: Designing Computational Models


. Identifiers and characteristics for transmission and studio scenarios. The charac
teristics of
'this experiment' are based on the task design and on observations.



Transition phase

(this experiment)


(+ learning to


No truly collaborative

Skills for efficient


collaborative situation
(fight or flight)

Utilization of non
standard skills
(attempts to conform to
collaborative situation)

Commitment to
collaborative situation

Development &
collaborative skills

Working relation

ial cohesion

Authority (knowledge)

Emotional relationship
development (within
and outside
collaborative situation)

Social grounding:
humor, accept (social)
confrontation, taking
shared responsibility,
dealing with


Product orientation

Goal interpretation

Professionalism: role
taking strategies, own
goal setting, negotiation

Content development

Explaining, check
reliability to source

referencing, factseeking,
confirmation of facts


of relevant
source material

Questioning &
comparison of source

Identification of
interrelated key

True dialogue,
prompting, argument
development, content