Towards Interactive Cognitive Agents with Culturally Restricted
Luis-Felipe Rodr
, Luis A. Castro
and Omar Gonzalez-Padilla
Dept. of Computing and Design, Sonora Institute of Technology, Cd. Obreg
on, Sonora, Mexico
Catrina Labs., Guadalajara, Jalisco, Mexico
Culture, Interactive Cognitive Agent, Cognitive Architecture.
Culture is one of the most important factors that influence human interaction. Evidence shows that people’s
culture determines the type of actions they are able to perform in certain situations. Consequently, in the field
of HCI, the study of culture has become essential for the design of interactive cognitive agents embedded in
social and realistic virtual environments. However, current attempts to enculturate such agents focus on mod-
eling specific aspects of culture, leaving aside the implementation of more integrative and biologically inspired
models. In this paper, we propose an integrative framework as a novel approach to the modeling of culture
in cognitive agents for HCI applications. This integrative framework is designed to serve as the underlying
architecture of interactive cognitive agents whose behavior is influenced by specific cultural backgrounds. The
architectural design and operations of such integrative framework is based theories and models formulated in
psychology and neuroscience.
People belonging to different societies perceive and
respond to particular events in contrasting ways. The
cultural differences among societies and the increas-
ing facility of people to communicate and move
around the world makes cultural conflicts and mis-
understandings an everyday issue. Related literature
shows how diverse disciplines address many of these
types of conflicts from different perspectives, trying
to develop models and strategies to minimize the im-
pact of cultural differences on the interactions be-
tween humans or between humans and machines.
Conflicts caused by cultural differences highlight
the importance of understanding culture (Lee and
Nass, 1998). Certainly, this would enable the de-
sign of environments for human-human and human-
machine interactions that reduce the impacts of cul-
tural differences on intercultural virtual and physical
encounters. In this line, some observable causes can
be identified, including the ambiguity in understand-
ing verbal and non-verbal expressions of people from
different societies, the differences in their attitudes to-
ward same things, the variability in the intensity and
type of emotions they express when experimenting
determined events, and the diversity in the type of ac-
tions they perform as reactions to same situations.
Culture related issues can be addressed in vari-
ous ways. For example, interactive cognitive agents
whose behavior is restricted by a cultural background
may be useful in diverse multicultural scenarios.
Cultural Cognitive Agents (CCAs) are autonomous
agents whose underlying architecture includes a num-
ber of components embodying cognitive and affective
capabilities, which enable the emergence of culture-
driven behaviors. In the case of the problematic inter-
actions between students and tutors with different cul-
tural background, CCAs might play the role of a vir-
tual tutor that embodies certain culture traits accord-
ing to the culture of students, allowing thus a more
convenient and conducive environment for learning.
Research focused on the analysis and measure-
ment of how important is for an intelligent system to
recognize the user’s culture and behave accordingly
has shown that CCAs may significantly improve the
performance of HCI systems. For example, CCAs
may be able to respond to humans as they expect,
given that the agent is aware of the traditions, beliefs,
and other aspects inherent to the user’s cultural back-
ground (Payr and Trappl, 2004). Furthermore, such
research has led to the creation of a number of models
for enculturating agents (Rehm et al., 2007), which
are usually focused on the simulation of specific as-
pects of culture and are supported by the results of
Rodrà guez L., Castro L. and Gonzalez-Padilla O.
Towards Interactive Cognitive Agents with Culturally Restricted Behaviors.
DOI: 10.5220/0006583502060211
In Proceedings of the International Conference on Computer-Human Interaction Research and Applications (CHIRA 2017), pages 206-211
ISBN: 978-989-758-267-7
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
empirical observations and social-psychological the-
ories (Rehm et al., 2007). However, a general and
more integrative model of culture for cognitive agents
based on biological findings has not been addressed.
In this paper, we discuss theoretical and computa-
tional models of culture in order to identify key as-
pects to be included in the underlying architecture of
interactive cognitive agents whose behavior is cul-
turally restricted. We present an integrative cogni-
tive framework designed to include components that
model the various identified aspects of culture as well
as other cognitive and affective mechanisms associ-
ated with the culture construct.
We review some of the most influential theories in the
development of computational models of culture.
2.1 Theoretical Models
The development of computational models of culture
has been mainly inspired by theoretical findings origi-
nated in areas such as social sciences and psychology.
One of the most influential theories is the dimensional
model of Hofstede (Hofstede, 1991). This is a so-
cial model developed on the basis of data from a sur-
vey conducted among employees at IBM and students
from 23 countries. The main hypothesis of Hofstede’s
theory is that people’s culture can be represented as a
point in a five-dimensional space:
1. Power distance: level of people’s acceptance in
the distribution of power among members of or-
ganizations, whether equitable or inequitable.
2. Individualism-collectivism: degree to which peo-
ple act as members of a group or as individuals
seeking personal achievement.
3. Masculinity-femininity: emphasizes on gender eq-
uity, and women’s and men’s values.
4. Uncertainty avoidance index: level of tolerance of
people to face uncertainty and thus change or not
their way of living.
5. Long-term Orientation versus short-term Orien-
tation: degree to which people are interested in
the future, in the present and the past.
Other computational models of culture have also
been developed on the basis of the cultural schema
theory (Nishida, 1999). This is a cognitive-based ap-
proach that tries to explain how culture is learned and
represented in the human’s brain. This theory charac-
terizes culture as an organized and structured network
of knowledge or schemas, which are supposed to be
retrieved in cultural situations in order to provide a
person with the appropriate knowledge to understand
the current scene, create expectations, and respond ac-
cordingly (Nishida, 1999).
2.2 Computational Models
A variety of computational models have been pro-
posed to develop Autonomous Agents (AAs) whose
behavior reflects certain cultural traits. For example,
Jan et al. (Jan et al., 2007) propose a model for devel-
oping conversational AAs that incorporate a specific
cultural background. This proposal focuses on the
modeling of cultural differences in proxemics, gaze,
and overlap in turn taking in North American, Mexi-
can, and Arabic cultures. Jan et al. use data derived
from empirical observations reported in related litera-
ture to assign proper values to each these three aspects
in conversational agents embodying a determined cul-
ture. In the underlying architecture of these agents,
proxemics is represented by the beliefs of the agent’s
relationship with other agents, gaze is modeled using
a probabilistic schema that allows moving between
five predefined gaze states, and overlaps in turn tak-
ing is implemented by using a gaussian distribution in
which mean and variation reflect cultural traits.
Rehm et al. (Rehm et al., 2007) propose a compu-
tational model of culture based on Hofstede dimen-
sions. In this model the recognition of user’s cultural
background is used to adjust the behavior of AAs.
Agents interact with users by embodying their same
cultural background, which is matched to one of the
following prototypes: Arab, Chinese, Germanic, Is-
raeli, Japanese, Swedish, Thai, and North American.
Five steps are implemented to achieve such proce-
dure. Behavior observation: a Wii remote controller
is used to collect user’s non-verbal information by
sensing their movements in a 3D space. Appraisal:
the data collected from users (spatial extent, speed
and power factors) are analyzed to match one of the
eight culture prototypes. Mode: the user’s culture is
linked to the agent’s culture; in this case, the agent
adopts the same culture of the user. Simulation: the
agent’s non-verbal behaviors such as sound, spatial
extension, speed, and power are calculated according
to the designated culture. Behavior display: the agent
displays accurate culture-driven behaviors. In a more
recent publication, Rehm et al. (Rehm et al., 2009)
describe a similar model of culture in which the be-
haviors developed by AAs imitate the actions of Ger-
man and Japanese individuals identified in video clips
recorded by the authors. Three scenarios were estab-
lished for such videos: meeting someone for the first
time, negotiations, and interactions with people with
a higher social status.
A computational model of culture based on the
schema theory and the appraisal theory of emotions
(which explains the elicitation and differentiation of
emotions on the basis of the relationship between
individuals and their environment) is proposed by
Taylor and Sims (Taylor and Sims, 2009), which
aims to develop interactive 3D cultural characters for
cross-cultural training. The authors propose a gen-
eral framework called the Cultural Cogntive Archi-
tecture, which makes use of particular schemas of
knowledge to elicit culture-driven behaviors. In this
model, cultural schemas contain data that is used for
the establishment of adequate sequences of behaviors
that cultural characteres must develop or expect in de-
termined situations. Moreover, appraisal processes
provide mechanisms for assessing cultural events un-
der certain appraisal dimensions, such as goal con-
duciveness and compatibility with internal and exter-
nal standards. This process is restricted by cultural
schemas leading to emotional data useful for generat-
ing coherent agent’s behaviors.
In summary, most of the current attempts to model
culture traits in AAs focus on the generation of certain
cultural behaviors, leaving aside the creation of more
general models which allow AAs to implement any
behavior on the basis of their cultural background.
For example, the model proposed by Jan et al. (Jan
et al., 2007) focuses on modeling proxemics, gaze,
and overlap in turn taking, neglecting aspects such as
the utility of goals an emotions. Furthermore, such
proposals are not concerned with developing models
of culture that allow the integration of new cultural
behaviors or cultural traits in AAs.
The word culture may refer to very different fea-
tures depending on the context, ranging from super-
ficial manifestations such as clothes and language,
to deeper mental models influencing human aspects
such as socialization and needs (Hofstede, 1991).
However, what do all these features have in common?,
which are the main features of culture? This section
discusses the common features of culture.
Culture is acquired. People learn cultural values
by interacting with other individuals and the environ-
ment. Tylor (Tylor, 1874) states that culture is a com-
plex whole including capabilities and habits acquired
by an individual as a member of a society. Keesing
(Keesing, 1974), Vatrapu and Suthers (Vatrapu and
Suthers, 2007) and ScHall (Schall, 2010) prioritize
the influence of the context in culture. Hofstede (Hof-
stede, 1991) defines culture as a collective program-
ming of the mind that distinguishes the members of
one group from another.
Culture is collective. Individuals belonging to the
same groups share similar cultural values. Hofst-
ede (Hofstede, 1991) affirms that each culture is con-
stituted by groups such as those formed in family,
school, religion, and work, which are considered as
subcultures that influence the behavior of individu-
als in different ways. Vatrapu and Suthers (Vatrapu
and Suthers, 2007) state that culture is shared knowl-
edge, created by a group in order to rule the way in
which people perceive, interpret, and express the re-
ality around them. Straub et al. (Straub et al., 2002)
state that culture is shared by individuals which share
the same space and engage in similar activities.
Culture is subjective. The perception of an indi-
vidual about the context depends on his own cultural
values. Researchers studying culture issues assert that
the ideas and conceptions of people are true only so
far as their civilization goes (Kitayama and Cohen,
2010). Sumner (Sumner, 2013) developed the term
ethnocentrism, which states that the individual’s own
culture is particularly important, and that other cul-
tures are measured in relation to the one’s own. Her-
skovits (Herskovits, 1972) affirms that judgments are
based on experience, which is strongly influenced by
the culture of individuals.
Culture is dynamic. Culture evolves over time in
order to adapt to external influences. According to
Keesing (Keesing, 1974), the technological develop-
ment and economic, politic, and social changes play
a main role in culture evolution. Hofstede (Hofst-
ede, 1991) and ScHall (Schall, 2010) affirm that when
members from different cultures meet and interact,
they are mutually influenced, producing changes in
their respective cultures. In addition, Scherer and
Brosch (Scherer and Brosch, 2009) consider that own
beliefs and thoughts change over time, producing an
evolution in culture. Hofstede (Hofstede, 1991) states
that culture is changeable as members of groups are
exposed to external influences.
Culture influences behaviors, emotions, and cog-
nition. Many studies about culture are based on ob-
servable behaviors of individuals. Hofstede (Hof-
stede, 1991) describes culture in terms of ve cul-
tural dimensions, each of them explained in terms of
attachment to stereotypical behaviors. Trompenaars
(Trompenaars and Hampden-Turner, 1998) studies
culture in terms of the way in which people solve
problems. Hall (Hall, 1966) investigates the influ-
ence of culture in physical distances and gazes while
people communicate. ScHall (Schall, 2010) identifies
cultural differences as the main influence in the way
in which people obtain, interpret, and share informa-
tion. Regarding emotions, several studies state that
culture represents a high influence in the way in which
people manage and express emotions; Trompenaars
(Trompenaars and Hampden-Turner, 1998) identifies
certain patterns of behavior defining whether express-
ing emotions is acceptable or not in a given culture.
Culture is personal and unique. Despite of con-
sidering culture as a collective phenomenon, there are
personal traits that produce variations in the cultural
background of individuals in a same culture. Hofstede
(Hofstede, 1991) recognizes the mutual influence of
personality and culture, which produces unique cul-
tural configurations in each individual. According to
Hall (Hall, 1966), each individual develops his main
culture, maintaining a continuous process to adapt it
according to the contexts in which he is involved in
the course of his life.
In the previous section we identified some key char-
acteristics of culture. The identified set of features
shows that a synthetic model of culture may become
as complex as needed, and that there will always exist
a trade-off between complexity and realism. We pro-
pose interactive cognitive agents as a good approxi-
mation for facing such trade-off. In this section we
briefly discuss some cognitive capabilities needed in
CCAs in order to illustrate the complexity of model-
ing culture. We then introduce a cognitive integrative
framework that aims to serve as the underlying archi-
tecture of CCAs, which provides a favorable environ-
ment for modeling the different aspects of culture.
4.1 Cognitive Capabilities for CCAs
Cognitive architectures of AAs are regarded as in-
tegrative frameworks aimed at unifying a number
of heterogeneous components whose interaction pro-
duce cognitive behavior. Often, these components
represent computational models of cognition and
emotion and implement mechanisms to reproduce the
behavior of processes such as perception, reasoning,
and decision making. The main assumption is that
from the joint operation of this type of process will
stem believable and intelligent behavior in cognitive
agents, which are software components that proac-
tively act in order to reach some objectives, based on
knowledge, skills, and the information retrieved from
the context. In order to show such kind of behavior,
cognitive agents have the following capabilities:
Autonomous execution and proactive behavior.
Agents have knowledge about their goals and au-
tomatically perform actions for achieving them.
Representation and storage of knowledge. Agents
can maintain a repository of structured knowledge
(e.g., using an ontology, trees and rules).
Reasoning capabilities. Agents implement infer-
ence mechanisms to produce new knowledge by
reasoning over their knowledge base.
Perception of the environment. Agents perceive
information of the context in order to update their
knowledge and therefore adapt their behavior.
Learning from experience. Agents evaluate the re-
sults of performing certain actions under certain
situations. They are able to adapt their actions in
order to improve their performance.
Communication and cooperation with other
agents. Agents are able to communicate in order
to coordinate their activities and achieve global
The main reason of proposing cognitive agents as
the underlying technology for developing a synthetic
model of culture is that they present many capabili-
ties which simulate to some extent the cognitive ca-
pabilities of human beings. As discussed in Section
3, culture involves a series of aspects that CCAs must
model in order to capture the complexity of cultural
behaviors. The acquisition of a cultural configura-
tion by the agent requires the capacity for represen-
tation and storage of Knowledge. Knowledge bases
are a convenient approach for such requirement given
that they are attached to structures like ontologies and
desicion trees which allow computations and infer-
ences. For example, ontologies are suitable for rep-
resenting models such as those proposed by Hofstede
(Hofstede, 1991), Hall (Hall, 1966), and Trompenaars
(Trompenaars and Hampden-Turner, 1998), which
consist of a set of independent variables that describe
cultural preferences in people. In order to model such
acquisition process, CCAs require to take information
from the environment and construct a cultural config-
uration from such information. Two main capabilities
are involved: communication and perception. While
the former allows the agent to capture information
about cultural configurations of other agents and com-
pute an influence over its own cultural configurations,
the latter enables the agent to take information from
symbols in the environment.
CCAs must be assigned a specific cultural config-
uration, which is stored in their knowledge bases. The
structure of such the knowledge bases depends on the
needs of the application. For example, if the applica-
Figure 1: Integrative approach for modeling culture in cog-
nitive agents.
tion is based on the model of Hofstede, such configu-
ration can be represented as a set of values represent-
ing the cultural dimension of the model. Furthermore,
CCAs must be able to tune their culture by perceiv-
ing their environment and learning from the actions of
humans and other agents. For example, a CCA whose
objective is to model the culture of a user could obtain
an initial cultural configuration based on the demo-
graphical information of the user (e.g., its nationality,
age, and sex) and personalize such configuration as
the agent learns from the actions of the user.
A CCA must embody cognitive processes to be
able to combine and make operations over different
cultural configurations in order to represent the mix-
ing of cultures of different groups. Given that cultural
configurations are stored in knowledge bases, mixing
operations are performed by reasoning and making in-
ferences over such knowledge. For example, consider
an application aimed at supporting collaboration of
users with the same nationality. It is agreed that cul-
ture of individuals with the same nationality may be
quite diverse depending on their age, sex, social po-
sition, and interests. In such case, cultural configura-
tions of different CCAs, would be based on the same
initial cultural configuration (i.e. the national culture)
and each of them will be adapted by computing the
impact of different groups to which the user belongs.
In applications involving several CCAs, each of them
must maintain a knowledge base of his own cultural
configuration, but also an own perception of the cul-
tural configuration of other CCAs.
4.2 Cognitive Architecture for CCAs
The analysis in the previous section reveals that cul-
ture is a human phenomenon that results from the op-
eration of several cognitive processes. In this line,
a computational model of culture must implement
several mechanisms associated to various cognitive
and affective processes as well as psychological con-
structs. We present an integrative cognitive frame-
work that aims to provide a favorable environment for
the unification of several computational models im-
plementing diverse cognitive and affective processes
associated to culture, such as decision making, learn-
ing, and personality (see Figure 1).
As Figure 1 shows, this integrative framework rep-
resents the underlying architecture of the CCA (level
1 in figure), which enables the emergence of cultur-
ally restricted behaviors. The architectural design
and operational assumptions of such agent architec-
ture are based on concepts, theories, and models for-
mulated in disciplines devoted to investigate human
behavior and its relation to the information process-
ing in the brain, such as psychology and neuroscience.
On the one hand, psychology offers high-level ex-
planations of the brain processes underlying human
behavior (including cultural behavior), enabling the
understanding of their main functions, their working
assumptions, and their interactions with other brain
processes. On the other hand, neuroscience pro-
vides theories and models that investigate the cog-
nitive and affective processes involved in culture in
terms of brain functions, brain structures, and neural
pathways, which is useful to understand in more detail
the internal procedures necessary for the modeling of
cultural behavior.
As explained above, culture is a complex phe-
nomenon that is subject of study in various disci-
plines. This psychological construct is being ad-
dressed at different levels of abstraction and from dif-
ferent perspectives. As a result, several theories and
models explaining culture in individuals and groups
have been formulated. Hence, in order to keep up with
such advances and develop a computational model of
culture able to capture new evidence, the cognitive ar-
chitecture of a CCA should meet the following two
1. Integrative and scalable architecture: CCAs have
to be designed so that they incorporate frame-
works that consistently unify theories and mod-
els that explain diverse aspects of culture, and
other cognitive and affective aspects as well. Ad-
ditionally, such architectures should provide suit-
able environments for the steadily incorporation
of new findings about this human phenomenon.
2. Consistent model for the interaction of various
cognitive and affective processes: since cultural
behaviors are the result of the operation of a se-
ries of cognitive processes, computational models
of culture must embody proper interfaces for the
interaction of the affective and cognitive functions
underlying culture.
The framework introduced above is based on evi-
dence from psychology and neuroscience, which sim-
plifies the achievement of these two requirements. As
shown in Figure 1, our proposed framework includes
a series of computational models of cognitive and
affective processes that represent psychological con-
structs or brain functions (level 2 in Figure), such as
personality, learning, and culture. The interaction be-
tween these components enables the dynamics of cul-
tural behaviors in CCAs. Furthermore, they represent
abstract models whose behavior emerges from the op-
eration of a number of architectural components (level
3 in Figure), which simulate the functionality and ar-
chitecture of brain structures.
In this manner, such proposed cognitive frame-
work allows us to implement an integrative and scal-
able computational model of culture, since any avail-
able or new neuroscientific theory addressing some
brain function or psychological construct can be im-
plemented by using the structural and operational
basis within the agent architecture. Similarly, this
framework meets the second requirement since com-
putational models of cognitive and affective functions
are implemented using the same or compatible struc-
tural and operational machinery used to implement
the model of culture.
In this paper we identified key characteristics to
be modeled in interactive cognitive agents aimed at
addressing the complex requirements of contempo-
rary applications in HCI. We presented an integra-
tive framework as the underlying architecture of such
interactive cognitive agents that attempts to integrate
the various components of culture. Considering that
the current state of knowledge of most aspects related
to culture in individuals is still limited, but in de-
velopment, this integrative approach becomes conve-
nient for the computational modeling of culture. Ad-
ditionally, neuroscience offers theoretical models of
the processes that underlie human behavior and which
are common to all individuals, allowing the model-
ing of the process of culture in autonomous agents
through the implementation of the basic mechanisms
from which all kind of cultural behaviors emerge.
Thus, instead of dealing with the question of why and
how culture is different among people and societies,
we focus on the synthesis of the brain mechanisms
that underlie the development of cultural aspects in
Hall, E. T. (1966). The hidden dimension.
Herskovits, M. J. (1972). Cultural relativism; perspectives
in cultural pluralism.
Hofstede, G. (1991). Cultures and organizations: software
of the mind. Mcgraw-Hill.
Jan, D., Herrera, D., Martinovski, B., Novick, D., and
Traum, D. (2007). A computational model of culture-
specific conversational behavior. In International
Workshop on Intelligent Virtual Agents, pages 45–56.
Keesing, R. M. (1974). Theories of culture. Annual review
of anthropology, 3(1):73–97.
Kitayama, S. and Cohen, D. (2010). Handbook of cultural
psychology. Guilford Press.
Lee, E. and Nass, C. (1998). Does the ethnicity of a com-
puter agent matter? an experimental comparison of
human-computer interaction and computer-mediated
communication. In Proceedings of the 1998 Workshop
on Embodied Conversational Characters.
Nishida, H. (1999). A cognitive approach to intercultural
communication based on schema theory. Interna-
tional Journal of Intercultural Relations, 23(5):753–
Payr, S. and Trappl, R. (2004). Agent culture: human-agent
interaction in a multicultural world. CRC Press.
Rehm, M., Bee, N., Endrass, B., Wissner, M., and Andr
E. (2007). Too close for comfort?: adapting to the
user’s cultural background. In Proceedings of the
international workshop on Human-centered multime-
dia, pages 85–94. ACM.
Rehm, M., Nakano, Y., Andr
e, E., Nishida, T., Bee, N., En-
drass, B., Wissner, M., Lipi, A. A., and Huang, H.-H.
(2009). From observation to simulation: generating
culture-specific behavior for interactive systems. AI
& society, 24(3):267–280.
Schall, J. M. (2010). Cultural exploration through mapping.
The Social Studies, 101(4):166–173.
Scherer, K. R. and Brosch, T. (2009). Culture-specific ap-
praisal biases contribute to emotion dispositions. Eu-
ropean Journal of Personality, 23(3):265–288.
Straub, D., Loch, K., Evaristo, R., Karahanna, E., and
Srite, M. (2002). Toward a theory-based measure-
ment of culture. Human factors in information sys-
tems, 10(1):61–65.
Sumner, W. G. (2013). Folkways-A Study Of The Sociologi-
cal Importance Of Usages, Manners, Customs, Mores
And Morals. Read Books Ltd.
Taylor, G. and Sims, E. (2009). Developing believable inter-
active cultural characters for cross-cultural training. In
International Conference on Online Communities and
Social Computing, pages 282–291. Springer.
Trompenaars, F. and Hampden-Turner, C. (1998). Rid-
ing the waves of culture: Understanding diversity in
global business. Nueva York: Mc Graw Hill.[Links].
Tylor, E. B. (1874). Primitive culture: Researches into the
development of mythology, philosophy, religion, lan-
guage, art and customs, volume 1. H. Holt.
Vatrapu, R. and Suthers, D. (2007). Culture and computers:
A review of the concept of culture and implications
for intercultural collaborative online learning. In In-
tercultural Collaboration, pages 260–275. Springer.