model can provide parameters which are more easily
controlled.
Moreover, Eysenck’ theory provided not only an
operationalized method for classifying human personalities
into different types but also a description of each type [19].
Eysenck provided eight descriptors for each of these
personality types as shown in Figure 2 [19]. The descriptors
include both traits and states, which make it easier to
generate corresponding personality types.
Figure 2. Eysenck’strait theory mode
The availability of Eysenck’s theory in constructing
personality of virtual agent has been verified by McRorie et
al [37]. In their study, four agents with distinct personalities
were built based on Eysenck’s theory and evaluated as
having high credibility. Therefore, Eysenck’s theory model
can provide an available basis for building personality types
of virtual agents.
For the above reasons, we have designed our APAs based
on Eysenck’s theoretical model. According to media
equation hypothesis, people make judgments on APAs in
the same way on human of zero-acquaintance. Therefore,
what we need to do next is to build the stereotypical
features (appearance, behavior, etc.) consistent with
descriptions of a defined personality type [9, 39]. In this
way, perceivers can make the corresponding personality
judgment from these visible physical features.
Emotional Feedback and Bodily Expression
In a virtual learning environment, APAs equipped with
emotional feedbacks play influential role. Kim et al.
indicated that the learners’ social judgments on the
instructional functionality of the APAs varied according to
APAs’ emotional states [30]. Later, Terzis et al. designed an
agent with empathetic encouragement behavior as emotional
feedback, and found a significant effect on learning intention
[51]. Therefore, it is necessary to add effective emotional
feedback to APAs.
Researchers have considered that facial expressions and
speech are important modalities of emotional feedback. By
contrast, bodily expressions were relatively neglected in
existing literatures. Actually, bodily expression conveys
important emotion information. Differing from other
emotional indicators such as speech or facial expression, it is
simple visual stimulus produced by combinations of parts of
the human body [42, 49]. Neuroscientific and psychological
studies have revealed that body movement is an important
modality of emotion expression [24, 36]. Upper-body
movements are of particular important functions.
Specifically, head and hands movements are actually most
often employed to express one’s affection, and hand
illustrator is one of the most frequent, regularly occurring
quantifiable bits of overt behavior accessible for
expressivity study [23]. Human observers focus attention
spontaneously around this body region when tempting to
infer others’ emotions [25].
In virtual reality environments, bodily expression can be an
important emotional expressive mode of virtual characters.
On the basis of hands position and trajectory patterns,
Balomenos et al. defined four classes of gesture for a
human-computer interaction system, including hands
clapping, hands over the head, “Italianate” gestures, and lift
of the hand [5]. The combinations of these gestures were
proved successful for the classification of six prototypical
emotions. Later, Glowinski et al. defined postural and
dynamic expressive gesture features, and developed a 4D
model of emotion expression [25]. The model effectively
classified emotions according to valence (positive, negative)
and arousal (high, low). This predominance of upper-body
movements in emotional communication is reinforced by
current computer interfaces and practice. Yet, these
expressive gestures have not been thoroughly applied in
virtual learning environment.
BUILDING PERSONALITY
Design of Personality Parameters
Based on Eysenck’s model, we created two sets APAs of
distinct personality types. Lee (male) and Lin (female) are
more choleric, Chen (male) and Cheng (female) are more
phlegmatic. The reason for selecting these two types is that
they have significant differences both in Extraversion and
Neuroticism. The two types can be operationalized and
defined by a combination of Eysenck’s factors (shown in
Figure 2), and measured by his test.
Personality types were constructed by manipulating physical
appearance, behavioral features and verbal features. The
features are associated with specified personality
stereotypes, which are defined as beliefs about the
characteristics of groups of individuals [14]. They are stored
in the brain as schemata, which automatically and
instantaneously lead perceivers to form expectations about
others’ personality [18]. Based on previous study results, we
associated sets of physical appearance, behavioral and
verbal features to each APA with its own personality (see
Table1).
#chi4good, CHI 2016, San Jose, CA, USA