Content uploaded by Agneta Gulz
Author content
All content in this area was uploaded by Agneta Gulz on Dec 30, 2020
Content may be subject to copyright.
Social enrichment by virtual characters –
differential benefits
A. Gulz
Lund University, Lund, Sweden
Abstract It is frequently held forth, within the area of virtual pedagogical characters, that such
characters are beneficial for learning as they strengthen the social dimension of electronic
learning environments. This article presents more details on this proposal together with a
survey of corresponding empirical evidence. In addition, materials from a recently conducted
empirical study are presented. Ninety school children, 12–15-year-old, were asked (i) to
grade the idea of virtual characters in electronic learning environments and (ii) to chose
between a strictly task-oriented, socially ‘shallow’ and a more socially oriented pedagogical
character. The participants were also asked to articulate the reasons behind their answers,
and to share their thoughts and opinions on the issues. The results of the study, as well as of
several of the studies reviewed, indicate that responses and attitudes towards social aspects
of virtual pedagogical characters are highly divergent. In particular, the notion that social
dimensions of virtual characters increase learners’ motivation and engagement may be less
generally applicable in a student population than is sometimes hypothesized. An ensuing
design guideline suggests interface solutions with an emphasis on flexibility regarding social
orientation and communicative style in virtual characters.
Keywords empirical, individual differences, pedagogical agents, relation oriented, secondary school
students, strictly task-oriented, virtual characters
Background
Central pedagogical theories, in various ways influ-
enced by Vygotsky, highlight the role that social as-
pects play for successful learning. They emphasize
that knowledge is socially constructed and that
learning essentially involves sharing and negotiation.
From a cognitive science perspective, the rich ability
to process various forms of social information, as well
as the motivation to do so, are essential for human
intelligence and existence (e.g. Donald 1991). No-
tably, human beings are not only able and motivated to
handle the social environments that they themselves
experience, but also represented social contexts, such
as the more or less fictitious social environments in
movies, novels, computer games, etc.
The social nature of human beings, thus described,
has instructional implications. Attention must be paid
to the social dimension of learning environments. In
the case of electronic environments, the potential of
virtual pedagogical characters in adding or develop-
ing a social dimension is frequently held forth (e.g.
Kearsley 1993; Baylor 2001; Moreno et al. 2001;
Dowling 2002; Bickmore 2003; Johnson et al. 2003).
Chou et al. (2003) state that the ‘positive impact of
research on educational agents lies in its ability to
strengthen the social learning environment’ (p. 260).
Kim and Baylor (2005) argue that whereas traditional
computer-based learning environments often failed to
provide situated social interaction, this can now be
Correspondence: Agneta Gulz, Department of Cognitive Science,
Lund University, Kungshuset, Lundagard, Lund S-222 22, Sweden.
E-mail: agneta.gulz@lucs.lu.se
Accepted: 10 October 2005
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418 405
Original article
obtained through virtual pedagogical characters,
which the authors define as ‘designed to facilitate
learning through various pedagogical strategies em-
phasizing social interaction between learners and
agents’ (p. 3).
This article sets out to present more details on how
virtual agents may develop or strengthen the social
dimension of computer-based learning environments
and thereby lead to pedagogical benefits. The first part
of the paper involves a literature survey of previous
research. Proposed pedagogical benefits and corre-
sponding empirical evidence is surveyed. The second
part is a presentation of materials from a recently
conducted empirical study. The overall goal is to add
details to our current knowledge about user responses
towards virtual pedagogical characters, specifically
their social aspects.
Proposed pedagogical benefits of social virtual
characters
The literature proposes a large number of pedagogical
benefits that virtual characters can bring about (for an
overview, see Gulz 2004). Here, the focus is on po-
tential benefits associated with the social dimension
that virtual characters may supply to electronic
learning environments.
Social capabilities in characters
The addition of a virtual character or characters with
human-like appearance and behaviour can, as such, be
said to imply a development of the social dimension of
an electronic learning environment.
1
But the social
‘nature’ of a virtual character can be emphasized,
when it is endowed with one, some or all of the fol-
lowing:
a capability for socio-emotional expressions, such
as empathy for an erring or frustrated learner; en-
couragement and praise, etc. (e.g. Lester et al. 2000;
Johnson 2001; Baylor et al. 2005);
a capacity to evoke emphatic reactions in learners
(e.g. Hall et al. 2004; Paiva et al. 2004);
an ability to apply social rules such as politeness in
dialogue and interaction (e.g. Johnson 2003; John-
son et al. 2003; Wang et al. 2005);
abilities for relation-oriented dialogue, including
elements such as small talk, conversational story
telling, joke telling/humour (e.g. Ho
¨o
¨ket al. 2000;
Bickmore 2003);
an ability to express a personality (e.g. Andre
´&
Rist 2000; Ball & Breese 2000; Churchill et. al.
2000; Johnson 2001);
a back story with personal attitudes, opinions, ex-
periences, family life, etc. and an ability to refer to
social personal issues (e.g. Ho
¨o
¨ket al. 2000; Mar-
sella et al. 2000; Oviatt & Adams 2000);
an ability for appropriate self-disclosure during a
personal interview (e.g. Gong et al. 2001) and a
capacity to use strategies of reciprocal self-dis-
closure that deepens over time (e.g. Bickmoreetal.
2005);
a capacity to remember and refer to earlier social
interactions and to social (off-task) facts, such as
what a learner has said about her favourite music, or
about her plans for the week-end (Bickmore et al.
2005);
a capacity to establish and maintain a relationship –
a social bond – with the learner (e.g. Bickmore
2003; Bickmore et al. 2005);
a capacity to evolve the relationship from stranger
to friend (e.g. Bickmore et al. 2005);
a capacity to track and evaluate the affective states
of learners (e.g. Kort et al. 2001; Conati 2002; De
Vicente & Pain 2002).
Proposed pedagogical benefits
In turn, such character capabilities are proposed to
lead to pedagogical benefits, such as:
learners relating to the character as a social and
intellectual partner for collaborating, sharing ideas
and perspectives, questioning, criticising, etc.
(Ryokai et al. 2003);
learners experiencing affiliation and identification
with a pedagogical character, and the character
functioning as a social role model (Ryokai et al.
2003; Baylor & Plant 2005);
learners experiencing a material (or an electronic
environment) as less difficult or intimidating
1
Some researchers refer to virtual characters with none of the capabilities
listed below in this section as social – in view that a human-like image
and/or voice, as such, can provide social cues and evoke social reactions
in users (e.g. Atkinson et al. 2005).
406 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
(reduced anxiety in learners) (Johnson 2003; Baylor
et al. 2004; Wang et al. 2005);
a reduced level of frustration in learners (e.g. Baylor
et al. 2005);
increased self-efficacy in learners (Baylor & Kim
2005; Ryu & Baylor 2005; Wang et al. 2005);
reduced loneliness in learners (e.g. Bickmore et al.
2005);
learners experiencing a friendship relation with the
character, with related effects such as a feeling of
responsibility towards the character (Bickmore
2003; Burleson 2004; Bickmore et al. 2005)
and increased support as well as criticism (Ryokai
et al. 2003).
A highlighted pedagogical benefit: increased
engagement and motivation
One potential benefit that is frequently pointed out is
that, as the learning environment with the addition of
pedagogical characters becomes more social, learner
motivation and engagement will increase (e.g. Andre
´
& Rist 2000; Lester et al. 2001; Moreno et al. 2001).
Moundridou and Virvou (2002) hold forth animated
pedagogical agents as a ‘promising approach to [the]
challenging aims [of] how to guarantee effective
learning, increase student’s motivation and engage-
ment [. . .] and generally enhance the learning ex-
perience’ (Moundridou and Virvou (2002, p. 253),
emphasis added). Ryu and Baylor (2005) hold forth
the ‘Engaging factor’ or ‘the positive social presence
of the agent’ as it is ‘‘there’ for the learners and mo-
tivating them’ (p. 26). Johnson (2001) as well, un-
derlines that agents can be used to improve subjective
satisfaction and ‘make experiences captivating and
exciting’ (p. 1).
It should be observed that, when motivational
benefits are brought up, there is frequently a follow up
argument. The motivating effects – gained from the
social dimensions of character interaction – are pro-
posed to affect learning outcomes in terms of im-
proved understanding, memory, problem solving, etc
(e.g. Lester et al., 2001; Moundridou & Virvou 2002;
Johnson et al. 2003). If willing to spend more time
with a system, Andre
´and Rist (2000) propose, a per-
son possibly learns more about a subject matter.
Moreno et al. (2001) suggest that a social agency
environment built around an animated pedagogical
agent will encourage learners to make a stronger effort
to understand a material.
Pedagogical benefits of social virtual characters –
empirical evidence
For most of the potential benefits listed above, the
empirical evidence is limited. Few corresponding user
studies have been carried out,
2
and in several cases
there are only single studies of pilot character ad-
dressing a particular potential benefit. But there are
exceptions. Learner experiences of friendship rela-
tions with social virtual characters, and effects of this,
have been studied by several researchers. Also, there
is a good amount of empirical evidence concerning the
notion that virtual characters can increase learner
motivation and engagement as the environment be-
comes more social.
The studies related below deal with these phenom-
ena, and regard user responses and attitudes towards
virtual pedagogical characters equipped with one or
some of the social capabilities listed above. Some
studies involve a comparison – virtual character vs.
human, character-based interface vs. non-character-
based interface – and some do not.
Learners experiencing friendship relations with social
virtual characters
Oviatt and Adams (2000) studied ten 6–10-year-olds
interacting with characters in animal shapes in a pro-
gram that teaches early elementary children about
marine biology, and observed that children ‘often
specifically mentioned that they liked ‘talking to the
animals’, whom they viewed as ‘friends’ rather than
parents or teachers’ (p. 340). The researchers also
report other evidence for friendship relations: ‘chil-
dren spoke directly to the animals using personal
pronouns, and approximately one-third of all the
content exchanged involved social questions initiated
by the child about the animated character’s name,
birthday, personal characteristics, friends, and family
life’ (p. 339). Potentially related positive effects
(feelings of responsibility towards the character, en-
gaging with the character as an intellectual partner,
etc.) were, however, not targeted in the study.
2
And to some extent it may be argued, that such studies have to await the
further development of social capabilities of virtual characters.
Social enrichment by virtual characters 407
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
Ryokai et al. (2003) studied 28 five-year-old girls,
where 14 played with props and told stories about
these props together with the virtual peer Sam,
whereas the other 14 did the same but without Sam.
The ways that the children spoke to and looked at Sam
indicate, according to the authors, that they related to
Sam as if she was a real child. Furthermore, the
children treated Sam as an intellectual and social
partner: as a collaborator and facilitator and also as a
peer that needed coaching.
Going to the other side of the age span, Bickmore
et al. (2005) let 10 participants, aged 63–85, have the
virtual coach Laura support them in the ambition to
increase their physical activity by walking, where
Laura incorporate several of the social abilities listed
in a previous section. Eleven participants in the same
age group made up a control group. This group ob-
tained a standard physical activity intervention: they
got pedometers and printed materials on the benefits of
walking for exercise. Evaluating their ‘relationship
with Laura’ on a scale ranging from stranger (1) to
close friend (7), the mean in the first group was 6.8.
3
Also their ratings of ‘friendliness of Laura’ and ‘trust
in Laura’ resulted in means over 6 with 7 as a max-
imum. In the summary Bickmore et al. (2005) state,
that Laura ‘was effective at establishing a social bond
with most users’ (p. 3). Furthermore, they report
(Bickmore et al. 2005) that several participants men-
tioned that they enjoyed the social dialog with Laura
and would have liked the opportunity to chat more
with her. These results are particularly interesting as
the interaction span was as long as 2 months, and as
many user comments about the relational dynamics
suggest that the relationship and the interaction be-
came more familiar and deepened over time. That
participants developed a feeling of responsibility to-
wards the character – which is a possible effect of a
friendship relation – was also indicated by the results.
However, contrary to Bickmore et al.’s expectations,
there were no significant differences between the
Laura group and the control group on experiences of
loneliness.
4
In an earlier study (Bickmore 2003), with a previous
version of Laura and 91 participants, mean age 25, the
responses towards a (possible) relationship with Laura
were considerably more divergent. They ranged from
subjects’ stating that they felt that they had come to
know Laura and no longer thought of her as a com-
puter character, towards statements like ‘Laura is NOT
a real person, and therefore I HAVE NO RELATION
WHATSOEVER WITH HER!’ (p. 185)
Also Ho
¨o
¨ket al. (2000) report divergency in re-
sponses among 18 participants, aged 19–41. In an
information system about films, the characters Agneta
and Frida ‘sit’ on the desktop and watch the browser,
as if watching television, and comment on what they
see. The characters were designed to express ‘per-
sonality’, ‘attitudes’, and ‘inner lives’ (Ho
¨o
¨ket al.
2000), and to provide users with their ‘back story’
through comments that allude to their own everyday
lives. Participants were asked whether the Agneta and
Frida characters ‘feel like friends’ on a scale from 1 to
7, where 1 stands for ‘not at all’ and 7 for ‘very much’.
Nine out of 18 participants gave ratings between 5 and
7, and 9 gave ratings of 4 or lower.
Hall et al. (2004), finally, worked with a character-
based learning environment about bullying. Given that
the learning content as such concerns social relations
and friendships, socio-emotive experiences in the
learners, such as empathy, are essential. In a group of
127 8- to 12-year-old school children the researchers
found an overall positive response. The result was
compared with that of the 95 adult participants that
participated in the same study, where the authors
conclude that the children had a clearly more positive
attitude towards the system and characters and ‘were
more likely to have an emphatic response and found
the characters more realistic and true-to-life than
adults’ (p. 604).
Social virtual characters increasing engagement and
motivation in learners
The Ryokai et al. (2003) study supports the notion that
virtual characters can increase engagement and moti-
vation in learners. When comparing the group of
children who interacted with Sam with those who in-
teracted with a peer, those interacting with Sam both
talked more about storytelling and were more fully
engaged in storytelling-related collaboration.
Robertson et al. (2004) studied 60 twelve-year-old
pupils use a program that assists children in writing a
story. Half of the group used a version with an ani-
3
Seven answers are included in this means.
4
Neither were there any effects on exercise behaviour.
408 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
mated agent interface and the other half a version with
a more traditional graphical user interface. Children
who used the agent version indicated more strongly
that they wanted to use the program again. However,
the agent version did not, as a general effect, increase
engagement in the sense of producing more user in-
teraction with the program. Whereas this indeed
seemed to be the case for girls, boys to the contrary
tended to interact more with the non-agent version.
In the Ho
¨o
¨ket al.’s (2000) study the following
motivational issues were assessed: Did users find that
the characters made the situation nicer? Did they think
that the characters were fun? Did the characters sti-
mulate explorative behaviour? Did users want to use
the system again? Results were that two-thirds of the
18 participants seemed to appreciate the characters
and found them fun and nice. The remaining third
expressed negative views about the characters, found
them irritating or disturbing, and not fun. These par-
ticipants also, not surprisingly, expressed compara-
tively little willingness to use the system again.
In a preference study by Gulz (2002) 36 partici-
pants, aged 23–59, were asked to choose between two
interfaces to an information system about books. In
one the information was organized around a number of
virtual characters, equipped with a back story. The
other interface involved no characters but instead a
spatial structure (a park). Out of the 36 participants,
16 chose the interface with virtual characters, 18 chose
the one without characters and two answered that they
could not chose one. From the verbal motivations gi-
ven, it appeared that the social aspect of the interface
with characters was appreciated by some (e.g. ‘I chose
it because it is more personal’, ‘I like that it is about
people, individuals’, ‘I find a social context more sti-
mulating to explore’, and ‘It’s more natural and easy
for me to relate to the people than to the park’)
and disliked by others (e.g. ‘I don’t want to know
things about those people’ and ‘People only compli-
cate things’).
Bickmore (2003) reports user studies with the
character REA, a virtual realtor, in two versions. In one
version REA uses social dialogue; in the other she is
purely task-oriented and ‘sticks to the task’ of pro-
viding estate information. Bickmore observes that
REA’s engaging in small talk, which is a central social
feature in communication between humans, evokes
strong – and diverging – reactions. Some subjects
‘reported liking the social dialogue aspects of the in-
teraction: ‘It wasn’t just real estate talk, so I felt like it
made her more human’ ‘. . . It sounds like she’s on
your side when she says things are expensive’ (p. 84).
Other subjects ‘. . .clearly did not like REA’s small
talk at all. ‘. . . I come in and I shop and I get the hell
out. She seemed to want to start a basis for under-
standing each other’’ (p. 85). Also Laura, mentioned
above, came in the (Bickmore 2003) study in two
versions, a ‘non-relational version’ and a ‘relational
version’, where the latter models a number of verbal
and non-verbal relational behaviours (e.g. an increase
over time in frequency of smiles, coming closer,
making gestures; engaging in small talk, and so on).
Also in this case the relational features were very
differently received. On the one hand, some partici-
pants gave appreciative comments such as ‘I like
talking to Laura, especially those little conversations
about school, weather, interests, etc. She’s very caring.
Toward the end, I found myself looking forward to
these fresh chats that pop up every now and them.
They make Laura so much more like a real person’
(p. 184). Also, a number of participants reported that
Laura motivated them to engage in their exercise tasks
as they felt responsible towards Laura, e.g. ‘It sort of
kept me motivated, because I always do more if
I know I’m responsible to someone’ (p. 188). But on
the other hand, there were subjects who did not find
Laura motivating or engaging, stating e.g. that: ‘I’m
dealing with computers all the time. So, I really cannot
take her as a character when you talk about those
emotional kinds of things’; ‘I didn’t really like Laura
very much . . . Actually, I liked all of the software
except for the animated conversation thing’ (pp. 185–
186.). Likewise some subjects pointed out Laura’s
inability to make them feel guilty if they did not
exercise (p. 188.).
In sum
Above, potential benefits of the social dimension
created by virtual characters have been reviewed,
primarily in terms of friendship relations between
learner and character and in terms of motivational or
engagement benefits. The picture that appears is quite
disparate. In several cases, we find highly divergent
user responses towards social features in characters.
Furthermore, it is possible that some of the studies
Social enrichment by virtual characters 409
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
referred to above harbour more variability than is re-
vealed. Sometimes, what is presented is only an overall
impression or a means, for instance in cases of group
comparisons (character use vs. no character use; adults
vs. children). But this leaves us ignorant about the de-
tails making up the means. Furthermore, a number of
statements are given in terms such as ‘most’ or ‘some’
that leave the precise distributions open.
The motivation for the study to be presented below,
was to go into some detail both quantitatively and
qualitatively regarding (i) learner responses towards
(the idea of) social virtual characters and (ii) learner
responses towards different degrees of sociability or
social orientation. For the latter, we were inspired by
the two kinds of communicative styles used by Bick-
more (2003) – termed ‘task condition’ vs. ‘social con-
dition’ in REA, and ‘non-relational’ vs. ‘relational’ in
Laura. But we wanted to do something that is not done
in Bickmore’s studies, namely to let learners choose
between these two communicative styles in a pedago-
gical character – strict task-orientation vs. task and
relation-orientation – and to articulate their choices.
Participants in the study were school children, aged
12–15. One reason for this choice is our belief that there
is a large potential for character-based learning en-
vironments in elementary school. Another reason is that
the literature review above indicates that virtual social
characters are generally more positively received in
children, whereas a larger proportion of adults seem
negative and sceptical towards virtual social characters.
But – as pointed out – it is possible that there exists a
hidden variability, and so, we were particularly inter-
ested in contributing details on the homogeneity/het-
erogeneity in childrens’ responses towards social
aspects of virtual pedagogical characters.
Study
Method
Participants
Ninety 12–15-year-old school children, 48 girls and
42 boys, from a Swedish secondary school, took part in
the study, which was organized in the context of their
regular arts lesson. The students came from nine dif-
ferent classes, or more precisely half classes. Nearly all
students in the groups participated. They seemed
keen to participate, and sometimes organized a queuing
system among themselves. All had some familiarity
with pedagogical programs making use of virtual
characters.
Materials and tasks
Two dummy versions of a scenario-based multimedia
program for secondary school were developed for the
study. In both versions the student takes the role of a
journalist at a magazine, being sent on various missions
to European countries in order to do article research. In
the Instructor Version the student is guided by a virtual
instructor. In the Companion Version, the student is
accompanied on the missions by a virtual companion.
5
Both dummies, created in Macromedia Director, in-
clude (i) an introduction where the program and a first
mission is presented (see Fig. 1) and (ii) a module
where the student shall choose an instructor agent or a
companion agent, respectively, from eight different
animated pedagogical agents (see Fig. 2).
The presentation of the first missions includes illus-
trations from Istanbul and traditional Turkish music. A
male speaker voice tells about the mission and presents
the students with their role as journalist. In the In-
structor Version, furthermore, the students are told that
there is a chief editor in London who will be their in-
structor. The chief editor will formulate the missions,
orient the journalist (the student) and provide necessary
information at critical stages. The journalist (the stu-
dent) is at particular times to report back to the chief
editor who will evaluate the reports and provide feed-
back on what is well done and what needs to be worked
on. In the Companion Version, the student is, instead,
told that there will be a companion journalist together
with whom she or he will conduct the missions.
Upon completion of the introduction the speaker
voice tells the students to chose which figure they
want as instructor (Instructor Version) or companion
(Companion Version). In both versions the same eight
animated characters are simultaneously placed in an
oval on the screen (see Fig. 2).
6
5
The instructor version and the companion version both, in turn, were
realized in two visual formats – realistic visual style versus iconic visual
style, as one focus of the study was user preferences as to visual style.
The screen dump in Fig. 1 comes from the realistic style version.
6
One issue in the study as a whole concerned the relation between pre-
ferred visual style – iconic or realistic, and preferred social style –
strictly task-oriented or task- and relation-oriented. This is analyzed in
Gulz and Haake (2005).
410 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
Procedure
The three experimental leaders
7
presented themselves
to the class as working with research on educational
media for the future. Students were then told that they
would be welcome to participate in a study. It was
emphasized that the purpose was to listen to student
opinions of various aspects of a program that they
would be asked to try out, that it was not a test of them
and that they were fully anonymous. The students
were asked to come, one at a time, to a small room
behind the classroom where the session took place.
After welcoming participants and asking what grade
they were in, they were asked to sit down at the
computer, press start and then wait for further in-
structions from the program.
When the student had progressed through the pro-
gram and reached the choice-of-character module
came the session part that dealt with preferences as to
the visual style of characters, and articulations of these
preferences.
8
Thereafter came the session part that is
central for this article, namely the one that regards (i)
attitudes towards virtual pedagogical characters as
such and (ii) attitudes towards a strictly task-oriented
character vs. a task- and relation-oriented character.
For assessing attitudes towards (social) virtual
pedagogical characters as such, the students were first
asked for their opinion on the idea of having virtual
learning companions (virtual instructors) in educa-
tional programs like this. They were asked to mark
their opinion on a 1–7 scale, where 1 stands for this
being a very bad idea and 7 stands for this being a very
good idea. After this they were asked the follow-up
question of why they graded the idea as they did, and
were invited to share their opinions and thoughts.
Thereafter the participants were presented with two
scenarios: (i) a scenario with an instructor/companion
that focuses strictly on the task at hand in an efficient
way (ii) a scenario with an instructor/companion that
apart from working on the tasks also attempts to de-
velop a relationship with the participant: supplying
information about him or herself, experiences, interests,
Fig. 1 A screen dump from the introduction where a first mission is presented to the participant.
7
One ran the program and helped the student out; one observed choices
and took notes; one conducted the interviews.
8
Reported in Gulz and Haake (2005).
Social enrichment by virtual characters 411
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
etc. engaging in small talk and more personal kinds of
discussions, and so on.
9
Participants were then asked
which of these two instructors/learning companions
they would prefer to have in this or a similar program
for use in school. Again there was a follow-up question
of why they answered as they did, where they were
invited to share their thoughts and opinions.
Finally, a learning style inventory was filled in. After
completion, the participant was offered refreshment,
debriefed and thanked for their valuable help. The total
time for a session was on average 18 min.
Measures
The participants’ choices were noted by one of the
experimental leaders during the session. The qualita-
tive data, i.e. how participants articulated the reasons
for their choices, were recorded and transcribed. The
data were then analysed by one of the experimental
leaders and a researcher not involved in the experi-
mental design to see whether any patterns and themes
that aligned with the focus of the study could be
identified. The first-level coding was made in-
dependently by the two coders and consisted in
grouping similar responses together, paying particular
attention to content that related to social dimensions of
virtual characters. Upon comparing the classifications
made by the two coders, a few differences occurred.
After discussing these, a joint result was arrived at.
The second-level coding involved giving each group a
summarizing label, e.g. ‘so that one doesn’t have to be
alone’ or ‘characters are tiresome’ (see Figs 1 and 2).
The rationales for an open-ended questioning in
contrast to a more structured procedure were (i) the
explorative nature of the study and (ii) the ambition
to create an informal and causal atmosphere where
students would talk and associate freely.
Results
Virtual pedagogical characters – pros and cons
A majority (80%, n572) of the participants, 37 girls and
35 boys, declared a positive attitude towards the idea of
Fig. 2 A screen dump of the module where the participant shall choose an instructor character or a companion character,
respectively, from eight different animated pedagogical characters.
9
The exact wordings differed in details depending upon whether the
student had been using the companion version or instructor version.
412 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
virtual characters in pedagogical programs (score 5, 6,
and 7 on the Likert-scale where 7 stood for ‘this is a very
good idea’ and 1 stood for ‘this is a very bad idea’).
A minority (20%, n518) of the participants, 11
girls and seven boys, gave lower scores (3 or 4 on the
Likert-scale), where some expressed doubtfulness or
negative attitudes towards virtual characters in peda-
gogical programs. No participants, however, used the
lowest part of the scale (1 or 2 on the Likert-scale).
Overall, a positive attitude was manifest in a majority
(approx. 80%) of the participants, a result that is in
concordance with the studies of Robertson et al.
(2004) and Hall et al. (2004), related above.
The most common pro argument for a character
(21 answers) was that a character makes it more fun
and/or more interesting. Another common line of rea-
soning (11 answers) was that a character would be
advantageous in a learning context, because one might
get help from the character and might learn more. A
third relatively common argument (seven answers) was
that with a character one does not have to be alone.
Three types of arguments against virtual pedago-
gical characters came forth: a character is superfluous
and unnecessary (four answers), a character makes
things more complicated (four answers) and a char-
acter is tiresome (three answers).
The most common kinds of pro and con arguments,
respectively, are presented and exemplified in Table 1.
It can be observed that some of the comments relate to
social aspects of virtual characters.
Preferred social style of character – strictly task-
oriented vs. task- and relation-oriented
Fifty-three participants (31 girls and 22 boys) chose a
task- and relation-oriented character, and 37 (17 girls
and 20 boys) chose a strictly task-oriented character.
Seventeen participants motivated their choice of a
task- and relation-oriented character in terms of it
being more fun, nice or interesting, without develop-
ing the argument any further. Another 18 participants
developed the argument: it is nicer and more fun with
a socially rich character because you get to know the
character better. Some explicitly spoke about the im-
portance of personal relations and personal contact/
connection. Eight participants motivated their pre-
ference for a task- and relation-oriented character in
terms of this being more playful and easy-going. The
task-oriented character, it was held, would make the
task too serious and hard. Four participants, finally,
motivated their preference for a task- and relation-
oriented character in terms of what is normal/
common. Two of those felt, that a task- and relation-
oriented character is more interesting because it is ‘not
so common’, whereas two found a task- and relation-
oriented character ‘more normal and common’.
All arguments in favour of a strictly task-oriented
character, notably, were negatively formulated, as
arguments against a task- and relation-oriented char-
acter. The three most common categories of argu-
ments, each occurring in seven instances, are related to
one another. Seven participants held that a task- and
relation-oriented character would be trying, tiresome
and a nuisance. Another seven pointed at the risk of
getting distracted by a task- and relation-oriented
character, and a third group of seven participants
spoke of a task- and relation-oriented character as one
that does unnecessary or meaningless things instead of
focusing on what is important. Finally, five partici-
pants held forth that the character is a computer
character and not a human being and that they there-
fore do not want to be personal with it.
Two participants explicitly responded that the ideal
would be to have both versions available to choose
from: ‘Sometimes one would feel like talking more and
chatting, but sometimes one would prefer a companion
that is quiet and sticks to the task. The best I think
would be if one could chose between companions that
have different personalities’, ‘It depends, sometimes I
would like one that is talkative and social, but some-
times I can’t stand that and want to be spared from it’.
Notable in both pro and con arguments are refer-
ences to various social aspects; such as socio-emo-
tional feelings of not being alone, of collaborating, of
having a personal relation, of getting in touch per-
sonally – vs. a desire to be spared ‘listening to chatter’
and these ‘unnecessary and meaningless social things’.
Table 2 presents common arguments pro and con
the two kinds of communicative style.
In summary, participants’ views on strictly task-oriented
vs. task- and relation-oriented characters clearly diverged.
Limitations
In interpreting the results, a number of methodological
limits should be observed.
Social enrichment by virtual characters 413
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
The study session as such gives the participant only
limited experience with a character-based learning
environment, in the form of a relatively short exposure
to a prototype program including an introduction
about the program concept. As concerns the partici-
pant’s response towards communication style in
characters, the study session only provides scenario
descriptions of the two kinds of communicative style,
but no actual encounters with implementations of
them. Nevertheless, bearing these conditions in mind
the results – in particular the nature and divergence of
positive and negative responses towards social or-
ientation in characters – are informative.
A more problematic limitation, in my opinion, is the
contrast between the study situation in terms of ‘a
break from the usual’ and an authentic learning ac-
tivity, ‘for real’ in a school situation. I speculate that
participants perhaps would be more likely to prefer a
strictly task-oriented character before a task- and re-
lation-oriented character in the context of a ‘serious’
learning task or other school activity that they want to
complete and, possibly, ‘get done’ with. In order to
test this, further investigations are required.
Furthermore, the study, as the majority of user
studies regarding character-based learning environ-
ments, does not inform us about repeated use during a
longer period of time. It is possible that the actual
results involve a novelty effect. A positive attitude
towards characters in general, as well as positive re-
sponses towards socially and relationally oriented
characters, may be amplified due to such an effect.
10
One of the central tasks for future research is to
Table 1. Pros and cons for virtual pedagogical characters
Arguments Frequency
13
Pro virtual characters
It is more fun/more interesting
‘It’s much more fun than just the computer saying things’. ‘It’s more interesting, a character is more alive’. ‘It can be
interesting to see how it reacts and behaves towards you’. ‘It adds something, it gets more fun’. ‘To have someone along
makes it more interesting’. ‘It’s a really great idea, it’ll be more fun then’. ‘A character is more fun than text-boxes’. ‘Text-
boxes are not that fun’. ‘It will engage more, because it becomes more alive’. ‘It’s good as well as entertaining’. ‘It is a
good idea, it can be quite fun – but it depends on how the idea is realised’.
21
You learn better that way/you get help
‘I think it’s good, I think you learn pretty much that way’. ‘It’s good to think that there is someone to help you out’. ‘Good
to know who to turn to’. ‘It would be good to work together, and you could learn from it’. ‘If it’s a game where you are
supposed to learn things, it is absolutely a great idea’. ‘Good, because then you also get help’. ‘Always when you play a
game where you shall learn things, it’s good to have a figure by your side’.
11
So that one doesn’t have to be alone
‘It’s good because then you don’t have to be alone as much’. ‘It’s a good idea because sometimes it’s boring to play alone’.
‘It is good, because then you feel that you have someone with you, it looks as a human being or what to say, ‘Pretty good;
sometimes it is boring to play alone’. ‘Always when you play a game where you shall learn things, it’s good to have a figure
by your side’.
7
A character is better than only a voice/only text-boxes
‘It is better with a figure that talks to you than only a voice’. ‘A figure that talks instead of only a voice will create more of a
whole’. ‘It’s better than only text-boxes’.
5
Con virtual characters
Characters are superfluous and unnecessary
‘It is unnecessary’. ‘They really aren’t needed’.
4
Characters complicate things
‘Characters only complicate things’. ‘I don’t think they add anything, only make things more complicated’.
4
Characters are tiresome
‘Such characters are tiresome’. ‘It seems tiresome, it’s easier if you’re allowed to do things yourself, it’ll probably only
complain all the time’. ‘I usually get weary of such characters’.
3
13
Frequency indicates the number of arguments within each category. Only kinds of arguments that occur more than once are included.
10
Bickmore et al. (2005), however, relate an opposite effect in their long-
term study (p. 28).
414 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
examine repeated use during extended periods of time
(cf. Gulz 2004).
11
However, there is reason to believe that the main
result of the study, namely the demonstration of di-
verging attitudes towards strict task orientation vs.
task and relation orientation, where we are not talking
of minority groups of participants, will have some
bearing also in ‘real’ school situations and in activities
over time. The relative distribution of preferences will
certainly vary, among other things because of the
quality and design of particular implementations, but
it seems unlikely that none of the identified divergence
will reappear.
From a somewhat different angle, Batliner et al.
(2005) carried out a study that corroborates the ex-
istence of diverging approaches in children towards to
the social aspects of computational artefacts. In this
study 81 children, aged 4–14, were allowed to freely
interact with a Sony AIBO robot dog, and notably two
distinctly different approaches towards the robot dog
appeared. One group of children spontaneously
Table 2. Pros and cons for strictly task-oriented versus task- and relation-oriented characters
Arguments Frequency
14
Pro task- and relation-oriented character
More fun, nicer ‘It may be a little bit more fun to have something like that’. ‘It’s more fun to work in that way’. ‘It will
be a bit more fun and a bit nicer’. ‘More fun and more interesting’. ‘More interesting. You won’t get as tired’.
17
You get to learn the character/you get a personal connection ‘Because one gets to know the person more’. ‘It will get
more interesting if one gets to know the person somewhat’. ‘Unconsciously, I think, one will listen more to a person where
one knows her better’. ‘It’s better if he wants to tell more about himself, so that you know who he is’. ‘It’s more fun to get
to know more about what the person is like’. ‘It’s much more fun if you can hear fun details about the one you are
working for’. ‘Then, even if it is a computer game, it gets like more personal and like more fun’. ‘A teacher should be
personal, it should be possible to have fun with a teacher’. ‘It’s good if you can get a personal connection to it’. ‘It’s more
fun to get in touch personally with it, that will make you feel more like really participating’. ‘It’ll be more like a person’.
18
More playful and easygoing ‘It’s nicer to be able to get away from work and have a break’. ‘It will be more fun to use
such a program then, it’ll be some play also and not just as serious’. ‘Otherwise it would be too strict’. ‘It will be easier, not
as strained’. ‘It becomes like . . . a bit easier that way’.
8
Something different/not plain common ‘It’s much more fun if it works like that, I have never seen a game like that’.
‘That sounds much more fun, the other version is just as usual’.
2
Common and normal ‘Because that would feel more as usual, more normal. The other one [agent] that you can’t talk
to about different things . . . that would feel more extreme’.
2
Pro strictly task-oriented character
Trying, tiresome and a nuisance ‘I think I would find it trying if it would talk about such other things’. ‘Such a game
would be totally tiresome’. ‘I would get tired of it, at least in the long run’. ‘I like when it is focused on the mission, I want
to be spared listening to other chatter’.
7
Risk to get distracted ‘[With the task and relation oriented character] you can easily lose track’. ‘I think it would be easy
to lose one self, forget what one is there to do’. ‘One may lose one’s balance’. ‘[The strictly task-oriented character] would
be better, because then you can concentrate better’. ‘[With the task and relation oriented character] I would get
disturbed’.
7
The character should do what it is there for, not do unnecessary things ‘It’s so unnecessary to hear about their interests
and such things’. ‘The first is better because you save time, it’s unnecessary and meaningless with these social things’.
‘Then you get done with it’. ‘It’s somewhat unnecessary these other things, one wants to get on with the mission’. ‘The
second one is completely meaningless. Listen to bullshit’. ‘You don’t really want to know things about this person, you
want to complete the mission’. ‘The program is about learning and not talking about that person’s favourite music. That is
not part of the game, and then it’s not as fun to talk about his family and friends and so on’.
7
It is a computer character, not a human being ‘It can become plastic like’. ‘It’ll be strange if a computer figure tells
about it’s interests, because you still know that it is a computer figure, you cannot change that’. ‘I think it would be
strange to talk about such things, if it is a computer game . . .’. ‘It feels a bit ‘suspicious’ being personal with a computer
character’. ‘In a computer program it’s not so interesting to hear about its family. It could fall very flat somehow’.
5
14
Frequency indicates the number of arguments within each category. Only kinds of arguments that occur more than once are included.
11
It seems more likely that fully accomplished educational programs for
commercial use will be intended for use over time.
Social enrichment by virtual characters 415
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
approached it as an interaction partner and social
partner, encouraging it, reprimanding it, mothering it,
getting angry with it, and so on. Another group of
children approached the dog as a sort of remote con-
trol tool, attempting to figure out how it could be
manipulated. The central observation, however, is that
both groups of children seemed to be engaged and
stimulated by the activities that they undertook (A.
Batliner, personal communication, July 2005).
Discussion
Sometimes it has been asked whether virtual char-
acters are a good idea: do positive or negative learning
effects dominate? Do learners find virtual characters
engaging and invest socially and emotionally in them
– or do they find them distracting as they imply an
overload on working memory? (Moreno et al. 2001).
The evidence presented in this article suggests that
it is more fruitful to ask what kinds of characters have
positive/negative effects for which and how large
groups of learners. Specifically the use of socially
oriented characters, capable of social dialogue, of
building relationships, of referring to personal issues,
and so on, may constitute less of an overall pedago-
gical benefit than is sometimes proposed.
Something that stands out in the results of the pre-
sented study, namely, is the contrast between how
socially oriented characters for some participants, in-
deed, seem engaging (‘It’s more fun to get in touch
personally with it’, ‘It would be a stimulus, like having
someone that might cheer you up when its dull’, ‘It
would be interesting, I am always curious about hu-
mans inner lives’, ‘It gets more interesting if one gets
to know the person’) whereas for others they seem all
but engaging (‘It would be totally tiresome’, ‘I want to
focus on the task, I want to be spared listening to other
chatter’, ‘It’s so unnecessary and meaningless with
these social things’).
The question is: even if increasingly high quality
systems and characters of this kind are developed, will
they be beneficial for all learners? My tentative answer
is that, at least, the proposed benefits by social di-
mensions of virtual characters of increasing motiva-
tion and engagement
12
is less general in a student
population than is sometimes hypothesized. Therefore,
one should be more considerate when it comes to the
communicative style of pedagogical characters. Out
from the results presented in this article, I would
propose the following tentative design guidelines:
(i) Supply an adequate variety and flexibility in ped-
agogical characters with respect to communicative
style, where users may choose for themselves.
(ii) Likewise, in products that use multiple characters
(teams of characters, multiple agent systems, etc.)
ensure an adequate diversity with respect to commu-
nicative style (cf. Hietala & Niemirepo 1998).
Because while is true that human beings are social
creatures and that it is crucial to develop the social
dimensions of electronic learning environments, there
is a simultaneous need to consider individual differ-
ences in learning with social interfaces. Opting for
design solutions with an emphasis on flexibility in this
aspect may increase the positive effects from char-
acter-based learning environments.
Acknowledgements
The author wishes to acknowledge the support of the
Swedish Science Council (Vetenskapsra
˚det). Appre-
ciation is also extended to the students who partici-
pated in the study reported on, and to the collaborators
Magnus Haake, the Department of Design Sciences,
Lund Institute of Technology and Martin Jonasson,
Lund University Cognitive Science. Finally, the au-
thor wants to thank two anonymous reviewers for
highly valuable comments on a previous version of
this manuscript.
References
Andre
´E. & Rist T. (2000) Presenting through performing:
on the use of multiple lifelike characters in knowledge-
based presentation systems. IUI 2000, pp. 1–8, ACM
Press, New Orleans, LA, USA.
Atkinson R.K., Mayer R.E. & Merrill M.M. (2005) Fostering
social agency in multimedia learning: examining the
impact of an animated agent’s voice. Contemporary
Educational Psychology 30, 117–139.
Ball G. & Breese J. (2000) Emotion and personality in a
conversational agent. In Embodied Conversational
Agents (eds J. Cassell, J. Sullivan, S. Prevost & E.
Churchill), pp. 189–219. MIT Press, Cambridge, MA.
12
And possible learning gains ensuing from this.
416 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
Batliner A., Steidl S., Hacker C. & Noth E. (2005) Private
emotions vs social interaction. Towards new dimensions
in Research on emotions. 10th International Conference
on User Modeling, UM05. Retrieved 15/09/2005 from
http: //www.di.uniba.it/intint/UM05/list-ws-um05.html
Baylor A. (2001) Permutations of control: cognitive guide-
lines for agent-based learning environments. Journal of
Interactive Learning Research 12, 403–425.
Baylor A.L. & Kim Y. (2005) Simulating instructional roles
through pedagogical agents. International Journal of
Artificial Intelligence in Education 15, 1–18.
Baylor A.L. & Plant E.A. (2005) Pedagogical agents as
social models for engineering: the influence of appear-
ance on female choice. Proceedings of AI-ED (Artificial
Intelligence in Education), Amsterdam. Retrieved 15/09/
2005 from http://ritl.fsu.edu/publications_proceedings.
html
Baylor A., Shen E. & Warren D. (2004) Supporting learners
with math anxiety: the impact of pedagogical agent
emotional and motivational support. International Con-
ference on Intelligent User Interfaces, San Diego, CA.
Retrieved 14/05/2005 from http://ritl.fsu.edu/pub
lications_proceedings.html
Baylor A.L., Warren D., Park S., Shen E. & Perez R. (2005)
The impact of frustration-mitigating messages delivered
by an interface agent. Proceedings of AI-ED (Artificial
Intelligence in Education), Amsterdam. Retrieved 15/09/
2005 from http://ritl.fsu.edu/publications_proceedings.
html
Bickmore T. (2003) Relational agents: effecting change
through human-computer relationships. PhD Thesis,
Media Arts & Sciences, Massachusetts Institute of
Technology. Retrieved 19/05/2004 from http//web.me-
dia.mit.edu/ bickmore/bickmore-thesis.pdf
Bickmore T., Caruso L., Clough-Gorr K. & Heeren T. (in
press)‘It’s just like you talk to a friend’ relational agents
for older adults: Interacting with computers. Available
on-line at www.sciencedirect.com
Burleson W. (2004) Affective Learning Companions, 7th
International Conference on Intelligent Tutoring Systems,
Maceio – Alagoas, Brasil. Retrieved 14/02/2005 from
http://affect.media.mit.edu/publications.php
Chou C.-Y., Chan T.-W. & Lin C.-J. (2003) Redefining the
learning companion: the past, present, and future of edu-
cational agents. Computers & Education 40, 255–269.
Churchill E.F., Cook L., Hodgson P., Prevost S. & Sullivan
J.W. (2000) ‘May I Help You?’: designing embodied
conversational agent allies. In Embodied Conversational
Agents (eds J. Cassell, J. Sullivan, S. Prevost & E.
Churchill), pp. 64–94. MIT Press, Cambridge, MA.
Conati C. (2002) Probabilistic assessment of user’s emotions
in educational games. Journal of Applied Artificial In-
telligence 16, 555–575.
De Vicente A. & Pain H. (2002) Informing the detection of
the students’ motivational state: an empirical study. In
Intelligent Tutoring Systems (eds S.A. Cerri, G. Gouar-
de
´res & F. Paraguacu), pp. 933–943. Springer, Berlin.
Donald M. (1991) Origins of the Modern Mind: Three
Stages in the Evolution of Culture and Cognition. Har-
vard University Press, Cambridge, MA.
Dowling C. (2002) The socially interactive pedagogical age-
nt within online learning communities. ICCE, Auckland,
New Zealand, pp. 20–34
Gong L., Nass C., Simard C. & Takhteyev Y. (2001) When
non-human is better than semi-human: consistency in
speech interfaces. In Usability Evaluation and Interface
Design: Cognitive Engineering, Intelligent Agents, and
Virtual Reality (eds M.J. Smith, G. Salvendy, D. Harris &
R. Koubek), pp. 1558–1562. Lawrence Erlbaum As-
sociates, Mahwah, NJ.
Gulz A. (2002) Spatially oriented and person oriented
thinking – implications for user interface design. Edu-
cation and Information Technologies 7, 67–80.
Gulz A. (2004) Benefits of virtual characters in computer
based learning environments: claims and evidence. In-
ternational Journal of Artificial Intelligence in Education
14, 313–334.
Gulz A. & Haake M. Social and visual style in virtual
pedagogical agents. 10th International Conference on
User Modeling, UM05. Retrieved 15/09/2005 from http:
//www.di.uniba.it/intint/UM05/list-ws-um05.html
Hall L.E., Woods S., Dautenhahn K., Sobral D., Paiva A.,
Wolke D. & Newall. L. (2004) Designing emphatic
agents: adults versus kids. In Intelligent Tutoring System
2004 (eds J.C. Lester, R.M. Vicari & F. Paraguac¸u),
pp. 604–613. Springer Verlag, Berlin, Heidelberg.
Hietala P. & Niemirepo T. (1998) The competence of
learning companion agents. International Journal of Ar-
tificial Intelligence in Education 9, 178–192.
Ho
¨o
¨k K., Persson P. & Sjo
¨linder M. (2000) Evaluating users’
experience of a character-enhanced information space. AI
Communications: The European Journal on Artificial
Intelligence 13, 195–212.
Johnson W.L. (2001) Pedagogical agent research at CARTE
– Articles. AI Magazine, 2001. Retrieved 14/04/05 from
http://articles.findarticles.com/p/articles/mi_m2483/
is_4_22/ai_82129229/print
Johnson W.L. (2003) Interaction tactics for socially in-
telligent pedagogical agents. IUI 2003. pp. 251–253,
ACM Press, New York.
Social enrichment by virtual characters 417
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
Johnson W.L., Kole S., Shaw E. & Pain H. (2003) Socially
intelligent learner–agent interaction tactics. AI-ED
2003. Retrieved 19/05/04 from www.cs.usyd.edu.
au/ aied/papers_short.html
Kearsley G. (1993) Intelligent agents and instructional sys-
tems: implications of a new paradigm. Journal of Artifi-
cial Intelligence in Education 4, 295–304.
Kim Y. & Baylor A.L. (accepted) A social cognitive fra-
mework for designing pedagogical agents as learning
companions. Educational Technology Research and De-
velopment. Retrieved 15/09/05 from http://ritl.fsu.edu/
publications_journals.html
Kort B., Reilly R. & Picard R. (2001) An affective model of
interplay between emotions and learning: reengineering
educational pedagogy – building a learning companion.
Proceedings of the IEE International Conference on
Advanced Learning Technologies, pp. 43–46. Madison,
WI, USA.
Lester J., Towns S., Callaway C., Voerman J. & Fitzgerald
P. (2000) Deictic and emotive communication in
animated pedagogical agents. In Embodied Conversa-
tional Agents (eds J. Cassell, J. Sullivan, S. Prevost & E.
Churchill), pp. 123–154. MIT Press, Cambridge, MA.
Lester J., Callaway C., Gre
´goire J., Stelling G. Towns, S., &
Zettlemoyer, L. (2001) Animated pedagogical agents in
knowledge-based learning environments.. In Smart Ma-
chines in Education (eds K. Forbus & P. Feltovich), pp.
269–298. AAAI/MIT Press, Menlo Park.
Marsella S.C., Johnson L.W. & LaBore C. (2000) Interactive
pedagogical drama. Agents 2000 Barcelona Spain, pp.
301–308. ACM Press, New York.
Moreno R., Mayer R., Spires H. & Lester J. (2001) The case for
social agency in computer-based teaching: do students learn
more deeply when they interact with animated pedago-
gical agents? Cognition and Instruction 19, 177–213.
Moundridou M. & Virvou M. (2002) Evaluating the persona
effect of an interface agent in a tutoring system. Journal
of Computer Assisted Learning 18, 253–261.
Oviatt S. & Adams B. (2000) Designing and evaluating
conversational interfaces with animated characters. In
Embodied Conversational Agents (eds J. Cassell, J. Sul-
livan, S. Prevost & E. Churchill), pp. 319–345. MIT
Press, Cambridge, MA.
Paiva A., Dias J., Sobral D., Aylett R., Sobreperez P., Woods
S., Zoll C. & Hall L. (2004) Caring for agents and agents
that care: building emphatic relations with synthetic agents.
AAMAS 2004, pp. 194–201, ACM Press, New York.
Robertson J., Cross B., Macleod H. & Wiemer-Hastings P.
(2004) Children’s interactions with animated agents in an
intelligent tutoring system. International Journal of Ar-
tificial Intelligence in Education 14, 335–357.
Ryu J. & Baylor A.L. (2005) The psychometric structure of
pedagogical agent persona. Technology, Instruction,
Cognition & Learning 2, 291–319.
Ryokai K., Vaucelle C. & Cassell J. (2003) Virtual peers as
partners in storytelling and literacy learning. Journal of
Computer Assisted Learning 19, 195–208.
Wang N., Johnson W.L., Rizzo P., Shaw E. & Mayer R.E.
(2005) Experimental evaluation of polite interaction
tactics for pedagogical agents. International Conference
on Intelligent User Interfaces, San Diego, CA, pp. 12–19,
ACM Press, New York.
418 A. Gulz
&Blackwell Publishing Ltd 2005 Journal of Computer Assisted Learning 21, pp405–418
