THE INCREMENTAL DESIGN OF SCRIPTS BASED

ON MULTI-AGENT SYSTEM

Sara Boutamina

Mentouri University, B.P. 325 Route Ain El Bey, Constantine 25017, Algeria

Hassina Seridi

LabGed laboratory, Badji Mokhtar University, BP 12 Annaba 23000, Algeria

Abdelkader Gouaïch

LIRMM Laboratory, MontpellierII University Montpellier, France

Keywords: Collaboration Scripts, Tracking Learners, Traces, Design Method, Multi-Agent Systems.

Abstract: Collaborative learning is not always effective; its effects depend on the richness and intensity of interaction

between students during the collaboration (Dillenbourg, 2002). This collaboration is structured using

collaborative scripts. Hence, the design of these Scripts is not trivial; it requires information on learners and

on their interaction. We believe that when learners are the target of any design, this one needs to be

evaluated on the basis of the learners themselves. However, most of the design approaches do not use

experimental feedback on the learners’ collaboration to improve the initial design. We propose in this article

a method for the design of scripts basing on the experimental feedback. We suggest the use of multi-agent

systems to provide help and information to the scripts designers.

1 INTRODUCTION

These recent years, researchers stress more the

importance of learning design. Among these

researchers, Robe Koper and Tattersall (. Koper &

Tattersall, 2005) who state that “the key principle in

learning design is that it represents the learning

activities and the support activities that are

performed by different persons (learners, teachers)

in the context of a unit of learning”.

Koper thinks that the key of the success of the

learning environments is the activities and not the

pedagogical objects. Consequently, he proposes, to

specify the learning situations, the Educational

Modelling Language (EML) which focuses on the

pedagogical activities. This language was adopted

by the IMS Global Learning Consortium to propose

the standard IMS Learning Design (IMS LD).

The result of the design of the learning situation

is called a script which is considered as a sequence

of phases.

In our work, we affirm that this concept (script) is

linked to the concept of trace. This later can

contribute to the changing of the script, either in a

dynamic way in order to regulate the learning, or at

the end in order to evaluate and reuse this script.

In this paper, we propose an approach for the

construction of scripts taking into account the

experimental feedback on the learners’

collaboration. The idea is to track the learners when

performing the different activities prescribed by a

script which is designed at first (preliminary design)

and provide feedback on the execution of this script

in order to review the preliminary design).

2 OBJECTIVE AND

MOTIVATION

We A collaborative script (or scenario) is a set of

instructions prescribing how students should form

groups, how they should interact and collaborate and

321

Boutamina S., Seridi H. and Gouaïch A. (2009).

THE INCREMENTAL DESIGN OF SCRIPTS BASED ON MULTI-AGENT SYSTEM.

In Proceedings of the First International Conference on Computer Supported Education, pages 320-325

DOI: 10.5220/0001989303200325

 SciTePress

how they should solve a problem (Dillenbourg,

2002). It structures the collaborative process in order

to promote specific types of interactions

(Dillenbourg, 2006 (a)). A script includes multiple

activities, occurring at different various social levels

(Dillenbourg, 2006(b)): individual activities (e.g.

reading, writing…), group activities (e.g. solving a

problem with a peer…), and class wide activities

(lecturing, discussion…).

A variety of design methods of scripts have been

proposed but none of them take into account the

experimental feedback and use it in an incremental

way in the process of scripts design. In fact, these

methods rarely use the feedback to improve

incrementally the initial design and most of them

focus more on the results of collaboration rather than

the process of design itself.

Also, designers have to take in consideration the

learners and their behaviours because they are at the

end the main actors of the designed script.

Our framework of scripts design is based on the

following six ideas:

1. The process of design is incremental based

on a loop of four phases which are:

Scripting, Specification, Execution and

Evaluation (Fig 1.).

Scripting is the phase of writing, for a group of

learners, of the different rules of collaboration and

describing the different activities, the different roles,

etc. In this phase a natural language can be used.

In the phase of specification, the script is

specified using a specific formalism. Then, this

script will be executed and finally it will be

evaluated on the basis of the learners’ traces during

the scripts execution.

Figure 1: The different phases.

2. The Scripting must be considered as a

whole and not only through its outcome i.e.

we have to take into account:

• The starting point (the different data).

• The final point (the outcome).

• The transformation from the first point to the

second one.

3. A design method is to guide without any

constraints the designer who must take into

account the human factors involved in the

execution of the script. The execution

model of the script must be considered as a

task or activity model and not a data model.

4. Human factors play a central role in the

process of design. Designers require

information on learners and on their

collaboration in order to favour the

desirable interactions. For this reason we

suggest the tracking of the learners.

5. Formalism for the specification of scripts is

used. This formalism enables the designer

to express his choices and not only to

describe the result.

6. An integrated environment is desirable for

the scripts design in order to facilitate

continuous communication between the

various “activity spaces" of the design

process. Hence, in an incremental approach

of scripts design the designer can move

from the evaluation spaces to the

specification spaces and to the

implementation spaces. We propose that

the use of such environment provides a way

to overcome the problem and gives

designers tools to go beyond the

assumptions of standard design.

3 THE INCREMENTAL DESIGN

OF SCRIPTS

The developer of pedagogical scenario can not judge

a design choice only if he evaluates its consequences

in a real situation based on the feedback of the

learners interactions. Also, we recommend an

iterative process in which the results of the

developed scenario evaluation are analyzed and

interpreted in order to be used for the adaptation or

for the improvement of the scenario.

The execution of the script should be considered

as a task or activity model and not as a model of the

different resources offered to the learners.

• The idea is to allow the designers to

express their choices and not only describe

the script. Indeed, the script is the expected

result but some choices may be important

as they are represented. For instance they

can be used in the reuse and adaptation of

scripts.

CSEDU 2009 - International Conference on Computer Supported Education

322

Figure 2: The different activity spaces.

• To facilitate the proposed design work, an

integrated environment is essential to allow

the designer to move between different

areas of activity. Indeed, the designer is

evolving from assessment space, to

specification and implementation spaces.

• The constraints of such approach are the

automatic generation of script that can be

possible only by a formalization of the

outcome of the design and the use of a set

of artificial agents that will act in the

different activity spaces listed below.

What Activity Spaces and how to Skip from

an Area to Another?

The steps concepts must be distinguished from that

of space activity in a design. Indeed, a stage

characterizeed a specific product and design

methods are described in terms of stages. The

activity space characterizes a state of the developer’s

activity. We have identified and characterized the

different activity spaces of the design process. These

spaces and the links between them are presented in

the following figure (Fig.2.).

We can identify seven activity spaces useful for the

designer over two phases: the design and evaluation.

Each activity space represents an identifiable

viewpoint of the designer on its design task.

For the design phase, we have identified four

areas of activity spaces:

 Data and activities acquisition space in

which information is collected: pedagogical

resources, profiles, learning activities...

 The modeling and description space:

allows the designer to have key abstractions

and a clear vision and accurate information

it will use. This is similar to application

development approaches.

 Activities structure design space: the

designer main concern is developing a

model of activities or pedagogical resources

stemming from the task for which the script

is designed. This model is different from a

data model.

 The influx space: provides the designer

with the means to specify how to change

the specification of the script structure to

the instantiated structure of the target script.

For the evaluation phase, we suggest the following

three areas:

 Observation Space: The main concern is to

observe the progression of learners in

accord with their requirements and profiles.

 Analyze Space: Actions and interactions of

the learners in the groups are analyzed from

the observation delivered in the precedent

space in order to have synthesized

information about the learners’ progression

in the group and about the designed script.

 Decision Space: The main concern is to

have some decision about the script to be

presented to the script’s designer in order to

refine this later.

THE INCREMENTAL DESIGN OF SCRIPTS BASED ON MULTI-AGENT SYSTEM

323

4 THE CONCEPTUAL MODEL

OF MULTI-AGENTS BASED

SYSTEM FOR THE SCRIPTS

DESIGN

Jennings and his colleagues (Jennings et al., 1998)

argue that the use of agents is attractive because they

are able to characterize naturally and easily a variety

of applications, and also to represent the different

entities of a system or a domain.

The agent paradigm is the most powerful paradigm

to provide abstractions for complex organizations

analysis and modelling. Humans and software

systems can be considered as entities which interact

and collaborate to perform their tasks in order to

achieve their goals.

In order to run a successful multi-agents

simulation in this approach, a script (scenario) must

be provided to agents. A collaborative script differs

from a program in that no explicit specification is

given in advance. It is necessary to propose a

conceptual model that models agents at an

appropriate level of abstraction by executing the

script and indexing actions and interactions of

learners in the learning environment.

The script author and an agent developer agree

upon activities as the interface between them. The

script author describes scripts using a language,

while the agent executor implements the activities to

be performed by learners and extracts interactions

from the script. The script author describes scripts

Using the learners’ activities and interactions in the

group.

The script executor conducts experiments in a

real environment, and then the experiment outcomes

are used by the script writer in order to refine the

original script.

The evaluation agent observes learners interactions

by collecting different traces of learners during their

collaboration.

5 AGENT FOR PROBLEM AREA

Learners are different and it is difficult to have an

adequate script for the entire group from the

beginning. In order to help the designer to modify

his script on the basis of learners, we suggest the use

of a set of agents having the following roles:

decision, interpretation, execution, observation and

tracking learners.

The work of the agents starts when the different

learners interact with the system.

A Graphical User Interface (GUI) is used to

facilitate the learners’ interactions with the system.

Each learner has to introduce his profile using this

interface. These profiles are stocked in the ‘Profiles

Base’ by ‘The Decision Agent’. This agent has a

direct relation with ‘The Interpretation Agent’ which

has to specify the script in a comprehensible format

for the other agents. This script is executed by ‘The

Execution Agents’.

‘The Tracking Agents: keep track of the learners

and stock the different traces in ‘the Traces Base’.

‘The Observer Agent’ controls the works of the other

agents.

The Decision Agent: Basing on the learners profiles,

the decision agent selects a script (which is adequate

to the profiles of the learners and not to the

behaviours/collaboration of learners) to be executed

in order to structure the learners’ collaboration.

Moreover, this agent is able to access directly to ‘the

Profiles Base’ and ‘the Traces Base’ in order to

provide the designer with the necessary information.

The Interpretation Agent: The script is specified

using a format which is different from that used by

agents; consequently, this script will be interpreted

by ‘the Interpretation Agents’ to make it

comprehensive.

The Execution Agent: ‘The Execution Agents’

execute the interpreted script taking into account the

different learners’ profiles.

The Observer Agent: The execution of the script will

be controlled by ‘the Observer Agents’. These

agents monitor the work of the other agents to

provide general information on the execution of the

script.

The Tracking Agent: They collect the different traces

of the learners during their collaboration.

6 A CASE STUDY ON THE

ARGUEGRAGH SCRIPT

The ArgueGragh (Dillenbourg, 2002) script is a

macro script aimed to trigger argumentation between

peers. It consists of the following five phases:

1. Each learner responds to an on-line multiple

choice questionnaire and for each answer he is

expected to argue his/her choice.

2. When all the learners answer the questionnaire

and argue, the system produces a corresponding

graph where the learners are positioned according to

their answers. Then, the teacher or the system forms

pairs of learners who provided different answers in

Phase 1.

CSEDU 2009 - International Conference on Computer Supported Education

324

3. Each pair has to respond to the same

questionnaire in Phase 1 and provide arguments.

They can see their answers and justifications

provided by each peer in Phase 1.

4. For each question the system calculates the

answers given individually and in collaboration. The

results are used in a debriefing session where

learners comment their arguments.

5. Each student writes a summary of all the

arguments collected for a specific question. The

summary should be structured according to the

framework used in the debriefing session.

Application of the Proposed Model to the

ArgueGraph Script

When the learners are present, the decision agent

informs the interpretation agents in order to rewrite

the script in a comprehensible format for the other

agents. Then, this script is executed by a set of

execution agents. These agents provide the learners

with the questionnaire and each learner responds to

it and argues his/her choice.

According to the learners answers the decision

agent produces the corresponding graph. Then, the

teacher or this agent forms pairs of learners who

have conflictual answers.

Each pair has to respond to the same questionnaire

in Phase 1 and provide arguments. The decision

agent allows the learners to see their answers and

justifications in Phase 1.

This agent calculates for each question the

answers given individually and in collaboration.

The results are used in a debriefing session where

the learners comment their arguments.

The different interactions of the learners with the

system are collected by the tracking agents and

during all these phases, the observer agents monitor

the other agents in order to provide a general idea on

the execution of the script.

In this way the designer can modify his script

and adapt it on the basis of the learners assisted by a

set of artificial agents which gave him the necessary

information about the learners’ interactions and

actions.

6 THE AGENTS

IMPLEMENTATION

To allow learners to access the learning system, a

distributed learning environment is proposed for

learners located anywhere and connected to learn at

any times. It’s a multi-agent based distributed

learning environment which provides a multitude of

learning object for learners of the group which are

referenced by the script author.

The learning system consists of the client side and

the server side. On the client side it has a JSP (Java

Server Page) user interface. On the server side, the

servlets and a multi-agent platform implemented

using JADE (jade: http://jade.tilab.com).

JADE (Java Agent Development Framework) is

a software framework for the development of multi-

agent systems and conforms to the FIPA

specifications (fipa : http://www.fipa.org/).

When learners log on the system through Web

based applications, a learner agent upload the profile

and requirements and the learner is affected to the

assigned group. The script is uploaded and the

execution of the script will be performed.

7 CONCLUSIONS

Collaboration has certain advantages for learning.

To profit from these advantages, the learners’

collaboration should be structured and organized.

Hence, scripts are used to structure the desired

interactions among learners.

The design of these scripts in not easy, for this

reason we suggest the use of an incremental script to

help the designer to take into account the behaviours

of learners and their interactions.

In this paper we presented a multi-agent based

system for the incremental design of collaborative

scripts. The main agents of this system are, namely,

‘The Decision Agent’, ‘The Interpreter Agents’, ‘The

Execution Agents’, ‘The Tracking Agents’ and ‘‘The

Observer Agents. These agents have the following

roles: decision, interpretation, execution, observation

and tracking learners.

REFERENCES

Dillenbourg, P, 2002. Over-scripting CSCL: the risks of

blending collaborative learning with instructional

design. In P. A. Kirschner(Ed) Three worlds of CSCL.

Can we support CSC, Heerlen, Open UniversiteitNed.

pp: 61-91. Netherland. 2002.

Dillenbourg, P, 2006(a). The solo/duo gap. Computers in

Human Behavior. Elsevier Ltd. 22, pp. 155- 159.

Dillenbourg, P, 2006(b). Orchestrating integrated learning

scenarios.Proceedings of the 23

annual ascilite

conference: who’s learning? whose technology? 2006.

Foundation of Intelligent Physical Agents (FIPA),

http://www.fipa.org/

THE INCREMENTAL DESIGN OF SCRIPTS BASED ON MULTI-AGENT SYSTEM

325

Java Agent Development Environment,

http://jade.tilab.com/

Jennings, N. R, Sycara K. P, and Wooldridge, M, 1998. A

roadmap of agent research and development. Journal

of Autonomous Agents and Multi-Agent Systems, 1(1):

pp. 7–36.

Koper, R, and Tattersall, C, 2005. Learning design: a

handbook on modelling and delivering networked

education and training .Berlin; New York: Springer.

CSEDU 2009 - International Conference on Computer Supported Education

326