Making Intelligent Tutoring Systems culturally aware: The use of ...

Making Intelligent Tutoring Systems culturally aware:
The use of Hofstede’s cultural dimensions
Emmanuel Blanchard, Claude Frasson,

HERON Laboratory, Computer Science Department,
Université de Montréal, C.P. 6128, Succ. Centre-ville,
Montréal, Québec Canada H3C 3J7
{blanchae, frasson}@iro.umontreal.ca



Abstract:
interactive / low intrusive hints whereas girls have a
preferential tendency for the opposite.
e-Learning is the concept of teaching people through the

Gender, education and age differentiation are
Internet. By definition, an e-Learning session is not
already a norm in research fields such as management or
limited by any border, meaning that learners implicated in
psychology. Those fields have also identified others
this session could differ by many criteria which include
elements that can deeply affect behaviours of individuals in
gender, social classes, religion, nationality, occupation…
a particular context. One of the most powerful of these
It has been shown that many of those elements can impact
elements is Culture. To the best of our knowledge, it has
on learning and some are already assessed in e-Learning.
never been treated in the e-Learning research field.
However, currently, when an e-Learning system receives a
connection call from a user, the system will not make any

In this paper, we present elements able to
difference whether this user is French, Chinese, or
strengthen e-Learning systems (such as Intelligent Tutoring
American… According to cross-cultural studies, culture
Systems) by using cultural considerations. Following
has a big impact on individuals’ cognitive processes and
results of studies in both management and cross-cultural
also on how individuals understand and interact with their
psychology fields, we define what a culture is and quickly
environment and peers. In this paper, we show that e-
present ways of differentiating cultures (see Hofstede
Learning systems which want to adapt to their learners
[1980]). We show that cultures have a strong impact on
and keep them motivated will take huge benefits by
people’s learning because they affect cognitive processes
considering learners’ culture. We present some results
(such as emotion [Scollon et al, 2004]), perceptions and
from
our
study
which
support
Hofstede’s
interactions an individual can have with his/her
Individualism/Collectivism scores [Hofstede, 1980].
environment and peers. Then, we present results of a cross-
Based on those results and on cross-cultural studies in
cultural study we made. Those results seem to validate the
management and psychology, we propose different
use of Hofstede’s Individualism/Collectivism dimension in
elements of learners’ culture that an e-Learning system
future culturally aware e-Learning systems. Finally, we
should consider.
propose an architecture for Culturally AWAre Systems

(CAWAS) for e-Learning where we introduce the notion of
Keywords. e-Learning, culture, Intelligent Tutoring
Culturally Intelligent Agents (CIA).
Systems, intelligent agents, emotion.
2. Cultural differentiation
1. Introduction

Before discussing of interests of integrating cultures

e-Learning is the concept of teaching people
in e-Learning, let’s focuses on what we mean exactly by
through computer networks. By definition, an e-Learning
culture: what is a culture? This question may seem trivial at
session is not limited by any border, meaning that learners
first but, in fact, different definitions exist. As Kashima
implicated in this session could differ by many criteria
[2000] mentioned, some researchers, inspired by Vygotsky
which include age, gender, social classes, religion, level of
among others, see a culture as “a process of production
education, nationality, occupation… It has been shown that
and reproduction of meanings in particular actors’
many of those elements can impact on learning.
concrete practices (or actions or activities) in particular
contexts in time and space”. Another point of view also

Gender particularities are becoming to be assessed
enounced by Kashima [2000] is that some people think that
in e-Learning and Intelligent Tutoring Systems. For
culture is a “relatively stable system of shared meanings, a
example, Arroyo et al [2000] found relations between
repository of meaningful symbols, which provides
gender and hint interactivity: boys seem to prefer low
structure to experience”.

---

In general, when researchers are mostly interested

“Hofstede’s five dimensions” is the most famous
with the ways cultures interact with psychological
cross-cultural work and still the major reference in today’s
processes, they tend to favour the use of the first definition
cross-cultural researches. However, there is not a
but when researchers are interested in cross-cultural
consensus on Hofstede’s results (see McSweeney [2002]
comparisons, they favour the second one because the
for example).
notion of culture as a stable system is important to be able
In fact, although many attempts to define and
to establish comparisons. e-Learning could take advantage
measure culture exist, Smith and Bond [1998] concluded
of both of those views. For example:
that all of them have produced convergent results, which
• The first definition (culture as a process of production
validate the concept of cultural dimensions as enounced by
of meanings) could be used conjointly with cognitive
Hofstede.
assessments of emotional state in order to better
understand learner’s reaction in a specific emotional
context,
3. The impact of culture on learning and
• The second view (culture as a stable system) could be
cognitive processes
used to explain variations in learning results, practices
and behaviours across cultural clusters.

In this section, we present some results that show
the great importance of dealing with culture in order to be

Those two definitions allow us to think about many
able to understand the learner and adapt to it. Given those
different kind of cultural clusters. Hofstede’s work [1980]
different studies, we not only say that cultural awareness is
was concerning national culture but also used the concepts
useful for e-Learning but sometime necessary to avoid bad
of organisational and occupational cultures. This study is
deductions concerning learners’ behaviours and de facto to
the most recognized work on cultural differences. In 1980,
take good decisions for learners’ future evolution.
Geert Hofstede analysed results coming from more than
100 000 persons. For his study, he used a bank of IBM

Emotion has a growing importance in today’s e-
employee attitude surveys undertaken between 1967 and
Learning and Intelligent Tutoring Systems researches.
1973. Those employees were working in 66 different
Inspired by Ortony et al [1988] and by Picard [1997],
countries. In his work, he defined 4 different bipolar
many researchers are currently working on ways of
cultural dimensions (a fifth was added later). For 40 out of
assessing learners’ emotions (Conati [2002]) or inducing it
the 66 countries, he was able to give a comparative score
[Chaffar & Frasson, 2004]. Making computer systems
for each of these four dimensions. Nowadays, values for 57
emotionally intelligent should allow more proximity with
national clusters are available. Subsequent studies
the user, enhancing the quality of human-computer
concerning commercial airline pilots, students, civil service
interactions.
managers, “up-market” consumers and “elites” in a large

Concerning the link existing between emotions and
number
of
countries
have
validated
Hofstede’s
cultures, Scollon et al [2004] did a really interesting work.
assumptions.
First they showed that, depending of the culture you are

Hofstede’s dimensions are:
living in, you experience some emotions more or less
frequently. In daily life, hispanic and european Americans
• Power Distance (PDI): it “focuses on the degree of
feel positive emotions more frequently and negative
equality, or inequality, between people in the country's
emotions less frequently than Indian, Japanese and others
society.”
Asian cultures. In general, there is more cultural variability
• Individualism/Collectivism (IDV): it “focuses on the
in positives emotions than in negatives emotions.
degree the society reinforces individual or collective
achievement and interpersonal relationships.”

What is a positive or a negative emotion? This
notion also differs between cultures. For example, Kim-
• Masculinity/Feminity (MAS): it “focuses on the degree
the society reinforces, or does not reinforce, the
Prieto et al [2004] showed in a cluster analysis of 46
traditional masculine work role model of male
countries that, in general, pride is seen as a positive
achievement, control, and power.”
emotion in western societies whereas in non-western, pride
is considered as a negative affect because “it separates
• Uncertainty Avoidance (UAI): it “focuses on the level
individuals from others”. Scollon et al [2004] found similar
of tolerance for uncertainty and ambiguity within the
results for Europeans and Hispanics (western) compared to
society - i.e. unstructured situations.”
Indians (non western). Scollon and her colleagues also
• Long Term Orientation (LTO): it “focuses on the
mentioned a study of Shaver and Schwartz [1992] who
degree the society embraces, or does not embrace
found that for Chinese respondents, “love-related concepts
long-term devotion to traditional, forward thinking
clustered near sadness and other negative emotions related
values.”
to attachment and loss”. Cultural differences in test

---

anxiety [Cassady et al, 2004] and in reward allocation
Overall (N=256)
1
2
3
4
[Fischer and Smith, 2003] have also been reported.
Male < 26 (N=61)
51% 43%
0%
7%
Male >= 26 (N=74)
49% 31%
0%
20%
Female <26 (N=78)
35% 47%
1%
17%
4. Our results
Female >=26 (N=43)
37% 44%
0%
19%
France (N=105)





Following those readings, we lead a study to find
Male < 26 (N=33)
61% 30%
0%
9%
differences between groups of learners. In this section, we
Male >= 26 (N=26)
42% 38%
0%
19%
present some of our most significant results concerning
Female <26 (N=35)
43% 46%
0%
11%
gender and cultural differences.
Female >= 26 (N=11)
36% 45%
0%
18%
Non French (N=151)




4.1. Methodology
Male < 26 (N=28)
39% 57%
0%
4%

Our study was available on the Internet. Participants
Male >= 26 (N=48)
52% 27%
0%
21%
were first asked to give information such as gender, age,
Female <26 (N=43)
28% 49%
2%
21%
mother tongue, nationality, childhood countries, level of
Female >= 26 (N=32)
38% 44%
0%
19%
education, current occupation… Then they had to answer a
Figure 1: preference of style of explanation relatively to age,
set of questions concerning learning preferences,
gender and nationality
personality (using some questions from the Myers-Brigg
questionnaire), relations with computers, and trust in

For the two next results presented, we asked people
eServices. The questionnaire was transmitted via web
to tell us how much they endorsed a given affirmation.
groups and sometimes mailing lists. A total of 256 persons
They used a 7 degrees scale to answer (going from 1:
answered. Most significant national groups were French
“corresponds not at all” to 7: “corresponds exactly”; 4 was
(number N=105), Canadians (N=51), Brazilians (N=38)
the mean value). Results were obtained by finding the
and Iranians (N=26). Our questionnaire was available in
mean score for each cluster. In general, Standard Deviation
English, French, Farsi and German but most of the answers
was between 1.5 and 2, which make us think that those
were given in English or French.
results must be seen as a cluster tendency, not a ground
truth. We used the same sub group than in the precedent
question and figure 2 shows the answers to the following
4.2. Results concerning gender differences
affirmation (taken in the Myers-Brigg questionnaire):

We first decided to consider 3 parameters: gender,
“I consider the scientific approach to be the best.”
age: under 26 or equal/over 26, and nationality: overall

sample (N=256), French (N=105) and non-French
Non French

Overall
French
(N=151). One of our principal questions was:
“When you have to learn a new concept or to understand a
6
new idea, you prefer it to be explained:
5,5
A - With diagrams/drawings.
5
Male
B - With oral explanations when you can interact with the
4,5
Female
speaker.
4
C - With oral explanations when you can just listen to the
3,5
speaker.
3
1
2
3
4
5
6
D - With written explanations.”


Figure 1 shows the results for this question.
Figure 2: endorsement of the scientific approach relatively to age,
gender and country.

Answers A and B have been privileged but we
Columns 1, 3 and 5 represent answers of people under 26
found that, except one group (non-French men under 26),
whereas columns 2, 4 and 6 concerns people over 26. It
men
preferred
to
be
explained
with
graphical
appears that while men tend to give more importance to the
representations whereas women preferred human-to-human
scientific approach, women may consider alternative
interactions.
approaches.

4.3. Results concerning cultural differences


Our main result concerns of course cross-cultural

differences. Using the same scale than in the last question,

we gave the following affirmation:

---

corresponded exactly to Hofstede’s order between those
“I prefer to work alone than in a group.”
nations.

In figure 3, we present results depending on

Those results encourage us to use Hofstede’s data
nationality, restricted to college/undergraduate students.
in future Culturally AWAre systems for e-Learning that

Canadians report greater interest for individualistic
take into account influence of culture. Although we found
work than French. However Brazilians disagree with this
some relations between learners’ preferences and the
affirmation. When we compare those results with
Hofstede’s IDV dimension, we suppose that others
Hofstede’s IDV national scores, it appears that Canada is
dimensions could also be useful in many e-Learning related
seen as highly individualistic (Canada IDV=80), France
topics (emotional assessment [Conati, 2002] and induction
also is supposed to be individualistic (France IDV=71)
[Chaffar & Frasson, 2004], politeness [Johnson & Rizzo,
whereas Brazil is seen as relatively collectivist (Brazil
2004], Socially Intelligent Agents [Johnson et al, 2003]
IVD=38). Analysis using age or childhood countries as a
discriminator provided similar results. Our results match
Hofstede’s IDV scores.
5. Presentation of CAWAS: Culturally
AWAre Systems for e-Learning
College and undergraduate students
5

A common critic to Hofstede’s dimensions and
other’s “cultural scores” is that it could reflect mostly
4,5
cultural stereotypes instead of true specificities of cultures.
Canada - col ege/undergraduate
(N=34)
For example, a classical argument in this sense is to remark
4
France - college/undergraduate
that, whether you live in a collectivist country, you can
(N=34)
have an individualist attitude and the opposite is also true.
Brazil - col ege/undergraduate
3,5
(N=28)
To our opinion, the idea when dealing with “cultural
scores” is to take them as reflecting a cultural group
3
tendency. But how to use group tendencies when we want
to adapt to a learner who is, by definition, an individual?
2,5
Our answer is based on the use of a “Culturally Intelligent

Agent” (CIA) that is inspired by Earley and Mosakowski
Figure 3: preference of an individualistic manner of working
[2004]’s notion of Cultural Intelligence, i.e. “an outsider’s
depending on countries for undergraduate and college students
seemingly natural ability to interpret someone’s unfamiliar

We repeated this measure but this time, we used the
and ambiguous gestures the way that person’s compatriots
overall national population in order to add the Iranian
would”. More precisely this CIA is formed by three
cluster as a fourth national cluster. Results are shown in
specific agents: a “Cultural Action Agent”, and a
figure 4.
“Cultural Transcriptor Agent” that interact with additional
components into a general architecture (Figure 4) of an e-
Overall population
Learning system entitled CAWAS (a Culturally AWAre
System).
5
4,5
Canada (N=51)
4
France (N=105)
Brazil (N=38)
3,5
Iran (N=26)
3
2,5

Figure 4: preference of an individualistic manner of working
depending on countries for the overall population

Results stayed coherent and Iran which was also a
more or less collectivist country (Iran IDV=41) reported a
mean score around 3.6 (quite neutral). We have to notice
that the order we found between our four national clusters

---

The SMA provides a complete status of the learner
Cultural Knowledge Base
profile to the Cultural Action Agent (CAA). This profile
includes the level of knowledge of the learner, information


on his personality traits, on his cognitive state (emotion
Dynamic
Static

and motivation) and on his membership to specific cultural
Culture
Culture
System
Database
groups obtained from the Matching Culture Agent. The
CAA asks the Cultural Knowledge Base to obtain the rules
associated to the cultural profile of the learner. Given all
the information provided by the SMA, the CAA is then in
CMA
MCA
charge of planning the learning session, determining

Learning
learning strategies to use and selecting information in the
Strategy
curriculum and in the system database in order to present
Base
Student Modeller Agent
the course. The CAA has a Decision-Making Engine that
determines the next action to do, using non monotonic
C.I.A.
logic in order to be able to process eventual opposite rules.

CAA
CTA
Curriculum
6. Conclusion and future works
Observers


In this paper, we tried to stress the importance of
Figure 4: architecture of CAWAS (Culturally AWAre System)
learners’ cultures for an e-Learning system. It has been
for e-Learning
shown in many cross-cultural studies that cultures affect

The learner is first observed according to a variety
observable behaviours and some cognitive processes (such
of parameters able to distinguish specific behaviours.
as emotions). Underestimating the culture’s role in e-
Learning may lead to misunderstanding learners’ reactions

These data are culturally interpreted by the Cultural
to different kinds of stimuli, which could lead to an error of
Transcriptor Agent (CTA) using cultural knowledge
adaptation to learners’ needs.
obtained from the Culture Knowledge Base (for instance,
“pride can be considered as a positive emotion for a

To support the idea of making e-Learning systems
learner of such country”). As we have seen before, cultural
aware of culture, we have presented results of a study.
data express more a tendency than the exact attitude for
Although our sample of individuals was modest, we found
every member of a cultural group.
clear relations with Hofstede’s cross-cultural study [1980].
We think that using results of such large scale cross-

The “Student Modeller Agent” (SMA) receives both
cultural studies in e-Learning is not difficult and could be
data from the observer and the CTA. The SMA asks the
very profitable. To this extent, we have proposed an
“Matching Culture Agent” to know the learner’s cultural
architecture for Culturally AWAre Systems (CAWAS).
types. It also transmits the learner profile to the “Culture
This architecture contains different agents which take into
Modeler Agent” (CMA). This one generates new cultural
account learners’ cultural particularities. Because we think
clusters which are stored in the “Dynamic Culture”
results of cross-cultural studies are cultural tendencies, we
module. Those clusters are composed of a set of empirical
think that it is useful to have two levels of cultural
rules deducted from the use of the system. Dominant
information: a theoretical one (containing the results of
meanings techniques [Abdelrazek et al, 2003] could also
cross-cultural studies) and an empirical one (related to
be adapted to detect the cultural core of a group of
cultural behaviours inside our system).
learners.

Finally, an important part of CAWAS depends of a

The Static Culture module, on the other side,
new class of cognitive agents: Culturally Intelligent Agents
contains theoretical rules and assumptions on the cultural
(CIAs) which are based on the notion of Cultural
behaviours (for instance assumptions deducted from
Intelligence [Earley & Mosakowski, 2004]. We think that
Hofstede’s values and cross-cultural studies like “if
CIAs could also be applied in many large-scale deployment
Hofstede’s IDV is high, people will have a tendency to
systems, including games, e-Commerce, e-Government and
work individually”). A factor of certainty is also allowed to
e-Services in general.
each of the cultural rules. These factors will evolve given
cultural outcomes produced by the learner while he uses
CAWAS. We think using Active Learner Modelling
techniques [McCalla et al, 2000] could also be interesting

in a CAWAS architecture.

---

Acknowledgements
failure of analysis. Journal of Human Relations, 55(1), pp. 89-
118.
We acknowledge the support for this work from
Ortony A., Clore G. L. & Collins A. (1988). The cognitive
Valorisation Recherche Québec (VRQ). This research is
structure of emotions. Cambridge University Press, MA.
part of the DIVA project: 2200-106.
Picard R. (1997). Affective Computing. The MIT Press,
Cambridge, MA.
References
Shaver P. R., Schwartz J. C. (1992). Cross-cultural similarities
and differences in emotion and its representation. In C.Z.
Abdelrazek M, Frasson C., Kaltenbach M. (2003). Dominant
Malatesta & C. Izard (eds), Emotions in adult development. pp.
Meanings Classification Model for Web Information. Third
319- 338. Beverly Hills, CA: Sage.
International Conference on Hybrid Intelligent Systems.,
Melbourne, Australia
Scollon C. N., Diener E., Oishi S., Biswas-Diener R. (2004).
Emotions across cultures and methods. Journal of cross-
Arroyo I., Beck J. E., Woolf B. P., Beal C. R., Shultz K. (2000).
cultural psychology, 35(3), pp. 304-326.
Macroadapting animalwatch to gender and cognitive
difference with respect to hint Interactivity and symbolism.
Smith P. B., Bond M. H. (1998). Social psychology across
International Conference on Intelligent Tutoring Systems,
cultures. Prentice-Hall.
Montreal, Canada, pp. 574-583.
Zhang Y. (2005). Age, gender and Internet attitudes among
Cassady J. C., Mohammed A., Mathieu L. (2004). Cross-
employees in the business world. Journal of computers in
cultural differences in test perceptions: Women in Kuwait
Human Behavior, 21(1), pp. 1-10.
and the United States. Journal of cross-cultural psychology,
35(6), pp. 713-718.
Chaffar S., Frasson C. (2004). Inducing optimal emotional state
for learning in Intelligent Tutoring Systems. International
Conference on Intelligent Tutoring Systems, Maceïo, Brazil, pp.
45-54.
Conati C. (2002). Probabilistic assessment of user’s emotions
in educational games. Journal of Applied Artificial Intelligence,
pp 555-575.
Earley C. and Mosakowski E. (2004). Cultural Intelligence.
Harvard Business Review, October 2004, pp. 139-146.
Fischer R., Smith P. (2003). Reward allocation and culture: a
meta-analysis. Journal of cross-cultural psychology, 34(3), pp.
251-268.
Hofstede G. (1980). Culture’s consequences: international
differences in work-related values, Beverly Hills, CA: Sage.
(see also www.geert-hofstede.com).
Johnson W. L., Kole S., Shaw E., Pain H. (2003). Socially
intelligent learner-agent interaction tactics. International
Conference on Artificial Intelligence in Education. Sydney,
Australia.
Johnson W. L., Rizzo P. (2004). Politeness in tutoring dialogs:
“run the factory, that’s what I‘d do”. International Conference
on Intelligent Tutoring Systems, Maceïo, Brazil, pp. 67-76.
Kashima Y. (2000). Conceptions of culture and person for
psychology. Journal of cross-cultural psychology, 31(1).
Kim-Prieto C., Fujita F., Diener E. (2004). Culture and
structure of emotional experience. Unpublished Manuscript,
University of Illinois, Urbana-Champaign.
McCalla G., Vassileva J., Greer J., Bull S. (2000). Active
Learner Modelling. International Conference on Intelligent
Tutoring Systems, Montreal, Canada, pp. 53-62.
McSweeney B. (2002). Hofstede’s model of national cultural
differences and their consequences: a triumph of faith – a

---