Information Technology Reference
In-Depth Information
with existing cultural background for cross-cultural communication. The other
is the investigations of human-agent interaction on the supposition that humans
and agents which controlled a human belong to different culture.
The rest of the paper is organized as follows. Section 2 reviews related works
and discusses achievements and non achievements of previous work. Section 3
explains the design and architecture of CEBE. Section 4 describes implemen-
tations to detect human behavior with cultural background in CEBE. Section
5 discusses concepts and limitations of CEBE to capture and analyze human
behavior with cultural background. Finally, Section 6 contains conclusions and
future works.
2 Previous Works
A number of research groups have studied the use of ECAs for intercultural
communication.
Huang et al. [6] developed a culture-adaptive virtual tour guide agent that is
implemented in a modular way with the GECA Framework [5] to minimize the
development cost. It can switch its behavior and speech language to three culture
modes: general Western, Japanese, or Croatian. This study focuses only on the
surface traits of culture, that is, languages, symbolic gestures, and probably
culture-dependent characteristics of gestures. Since they used scripts to describe
human-agent interactions, the range of possible interactions will be relatively
limited and the quality of the whole system heavily depends on the knowledge
and skill of the agent designers.
Iacobelli and Cassell [7] showed an attempt to use virtual peers to encourage
African American children to switch their language coding to increase school-
based literacy. They implemented two virtual peers by applying two models of
behavior to an existing virtual child with a racially-ambiguous appearance in
their previous study [2]. They implemented their enculturated behavior on the
basis of observations of the verbal and non-verbal behavior of children. Their
agents, therefore, had the same limitations as an agent in Huang et al [6].
Some researchers developed their systems to generate culturally adequate be-
havior based on a human's observable behavior instead of implementations of
cultural backgrounds by heuristics. For example, Rehm et al. [14] proposed a
Bayesian network model of cultural adaptation, which was then employed in
two different sample applications that illustrate the great potential of culture
adaptive systems. They, however, realized the model with empirical data from
the corpus study under standardized conditions at hand. In addition, the empir-
ical data was cross-cultural rules and habits which could be known in advance.
This is a common issue when a behavior model is developed from empirical data
at hand.
There is another method to learning cross-cultural behavior, which is through
by behavior data obtained from encultured human-agent interaction in which
an agent directly expresses behavior of a human operator. This method is our
approach. There are some systems in which an agent directly reflects human
actions.
