DESIGNING A VIRTUAL LEARNING COACH FOR SUPPORT OF DIGITAL LITERACY OF SENIOR LEARNERS IN CONTEXT OF THE ELECTRONIC HEALTH RECORD. DESIGN CONSIDERATIONS IN THE EPA-COACH PROJECT

Abstract

This paper presents design options for a pedagogical agent as a virtual learning coach as part of the e-learning system for digital literacy of senior learners in the project ePA-Coach based on the review of current literature for designing pedagogical agents. Building on past studies and available frameworks, the paper provides an overview of different types of agents, requirements for designing pedagogical agents with focus on the preferences of senior users, design options and technologies used for designing pedagogical agents. The paper also discusses the limitations of current research. Based on this overview, the paper outlines current considerations for the design of the pedagogical agent in the ePA-Coach project and design options for four different agents with focus on the visual style, the communication and social styles, and the pedagogical roles. Finally, the paper describes next steps in research and development of the virtual learning coach in the ePA-Coach project and proposes recommendations for further research.

Full text

DESIGNING A VIRTUAL LEARNING COACH FOR SUPPORT OF
DIGITAL LITERACY OF SENIOR LEARNERS IN CONTEXT OF THE
ELECTRONIC HEALTH RECORD. DESIGN CONSIDERATIONS IN
THE EPA-COACH PROJECT
Ilona Buchem, Carolin Gellner
Beuth University of Applied Sciences, Berlin (GERMANY)
Abstract
This paper presents design options for a pedagogical agent as a virtual learning coach as part of the
e-learning system for digital literacy of senior learners in the project ePA-Coach based on the review
of current literature for designing pedagogical agents. Building on past studies and available
frameworks, the paper provides an overview of different types of agents, requirements for designing
pedagogical agents with focus on the preferences of senior users, design options and technologies
used for designing pedagogical agents. The paper also discusses the limitations of current research.
Based on this overview, the paper outlines current considerations for the design of the pedagogical
agent in the ePA-Coach project and design options for four different agents with focus on the visual
style, the communication and social styles, and the pedagogical roles. Finally, the paper describes
next steps in research and development of the virtual learning coach in the ePA-Coach project and
proposes recommendations for further research.
Keywords: virtual learning coach, pedagogical agent, senior learners, digital literacy, electronic health
records.
1 INTRODUCTION
Digital literacy and digital sovereignty in dealing with digital technologies and data are important
conditions for the participation of elderly people in modern societies. The demographic change
towards ageing societies together with the rapid developments in information technologies have
created new challenges for digital literacy. In some countries, like Germany, the increase in the
proportion of elderly people has called for new approaches to promoting digital literacy among senior
learners with focus on the demands of specific fields such as healthcare. Some of the recent
developments in the field of healthcare include the introduction of an electronic health record
(German: Elektronische Patientenakte, ePA), which enables to electronically collect, manage and
share health-related data, including demographics, medical history, laboratory test results and
medication. Statutory health insurance funds in Germany have to provide policyholders with electronic
health records from 1st January 2021 onwards. The ePA aims to become a central element of the
healthcare system, which is a highly complex and highly differentiated network and telematics
infrastructure with more than 200,000 institutions and professional groups [1]. This means that senior
citizens have to deal with a large number of different institutions and highly diverse health-related
data. In view of the electronic health record, the informational autonomy and digital sovereignty of
senior citizens becomes crucial. The ePA-Coach project addresses this challenge and aims to develop
a coaching-based e-learning system for senior learners to enhance their digital literacy against the
background of the electronic health record (ePA). In this context, digital literacy goes beyond the basic
understanding of information technologies and encompasses skills needed to exercise informational
self-determination regarding the storage, transmission, and processing of personal health data.
In this paper we describe the approach to designing a virtual learning coach (pedagogical agent) as
part of the e-learning system for digital literacy of senior learners. Pedagogical agents are characters
in multimedia learning environments used to facilitate learning. Research suggests that pedagogical
agents may have a significant impact on learning outcomes and make them more engaging and
effective. The paper starts with the review of literature and research in designing pedagogical agents
including preferences of senior users, frameworks, current limitations, and such approaches as
pedagogical agents as learning companions (PALs), Embodied Conversational Agents (ECAs) and
Animated pedagogical agents (APAs). Based on the review, specific requirements and success
factors for designing pedagogical agents in general and for senior learners including frameworks,
visual appearance, communication and interaction, and competence and role are summarized in the
Proceedings of ICERI2020 Conference
9th-10th November 2020
ISBN: 978-84-09-24232-0
7891
Buchem, I. & Gellner, C. (2020). Designing a virtual learning coach for support of digital literacy of senior learners in context of the
electronic health record. Design considerations in the ePA-Coach project. ICERI2020 Proceedings, ISBN: 978-84-09-24232-0, doi:
10.21125/iceri.2020, pp. 7891-7900. URL: library.iated.org/view/BUCHEM2020DES

third section. The fourth section describes used technologies for designing pedagogical agents. In the
fifth section of the paper there are design options and considerations for designing a virtual learning
coach as part of the e-learning system in the ePA-Coach project including visual style, communicative
and social style, and pedagogical role presented. The paper ends with conclusions including next
steps in research and development in the ePA-Coach project and recommendations for further
research.
2 BACKGROUND AND LITERATURE REVIEW
Martha and Santoso (2019) defined a pedagogical agent based on previous research as follows: „[…]
is an agent (single or multi) in the form of a virtual character equipped with artificial intelligence that
can support the students’ learning process and use various instructional strategies in an interactive
learning environment“ [2]. Research on pedagogical agents initially focused on the technical
perspective and the educational perspective has been emphasised since the late 1990s. Until 2017,
the main research focus was on pedagogical agents being able to support education in the form of
various roles such as mentors, motivators, facilitators, navigators, and collaboration assistants. Today,
research in the area focuses primarily on the agent's appearance, communication style, and how
feedback can be given to motivate learners, increase engagement, support learning activities, and
improve learning performance. Educational agents can be designed as realistic and human-like
characters. Research showed that pedagogical agents can have a significant impact on student
learning outcomes and learning motivation [2]. Lester et al. (1997) called this the persona effect [3].
The character design has a positive impact on student learning and also the role of the agent can
improve learning outcomes [2]. In this section, we describe studies on the design of pedagogical
agents, frameworks and different agent types, also addressing limitations in current research.
2.1 Design of pedagogical agents
In current research pedagogical agents are represented in the following five design forms: text, voice,
2-D character, 3-D character and human. The most common form is text and the second most popular
form is the 3D-character. In addition, the focus has been on the appearance and the role of the agents
as independent variables [2].
Link et al. (2001) examined the influence of speech parameters and facial expressions of pedagogical
agents on the perception of feedback among three groups each consisting of 30 psychology students
[4]. The feedback terms and facial expressions of the agent, called AutoTutor, were generated by the
Microsoft Agent engine. The results showed that the perception of the feedback (positive, neutral,
negative) depended on the linguistic expressions and mouth curve [4].
Baylor et al. (2003) examined the effects of pedagogical agent voice (human, machine-generated) and
animation (present, absent) on learning, motivation, and perceived pedagogical agent persona in 80
undergraduate students enrolled in an educational technology course. A positive effect of animation
on learning was found. Negative effects of animation were seen in facilitating learning, motivation,
satisfaction, and self-efficacy. In addition, there was a positive effect of the human voice on the
perceived agent persona. This seemed more engaging and more human. Learners were more
motivated when the agent was not animated and had a human voice or when the agent was animated
and had a machine-generated voice [5].
Gulz and Haake (2005) conducted a study of learner preferences regarding the visual (realism vs.
iconization) and social style (task oriented vs. relation oriented) of pedagogical agents with 43
university students and ninety students from elementary schools. Both groups of participants preferred
a relation-oriented agent with iconic style [6].
Schwind (2018) examined the preferences of users towards virtual faces based on 5 general and 32
facial parameters. First study with 431 participants from 7 to 75 years showed that smooth skin,
natural skin color, and human proportions are the most relevant factors to avoid negative feelings
regarding virtual faces. In addition, the participants preferred to create human-like faces instead of
cartoon-like faces and used very attractive features for faces they liked. Positively associated faces
had smooth skin, realistic proportions, and natural average skin color. Female faces were equipped
with full lips, snub nose, and slightly upturned eyes. Male faces got strong eyebrows, downturned
eyes, a larger throat, and thinner lips. Furthermore, if the participants were free to choose the gender
of the virtual face, three quarters of the participants chose a more female type. There seems to exist a
preference for an attractive female face. Hair color was irrelevant for perceived likability or
7892

attractiveness. Schwind named a few recommendations for designing virtual faces, such as (1)
Atypicalities for high levels of realism should be avoided because this increases negative sensations,
(2) avoid the “symptom of dead eyes” and (3) use aesthetics and appealing features [7].
Salehi and Nia (2019) examined the effect of different levels of realism (iconic, semi-iconic, and
realistic) of pedagogical agents on learning in the context of mobile-based health e-learning with 48
participants. The results showed that the group with the realistic pedagogical agent achieved better
post-test results than the group with the iconic agent [8].
Lin et al. (2020) carried out a study on the effects of two social cues (agent vs. no agent and
conversational style vs. formal style) of pedagogical agents on learning outcomes, cognitive load, and
intrinsic motivation with 98 Chinese college students. The study showed that learning with a
pedagogical agent was more interesting for the students. In addition, learning from a lesson written in
a conversational style not only enhanced retention but also increased pressure to learn. Additionally,
having a conversational style pedagogical agent increased mental effort, while conversational style
teaching without a pedagogical agent was found to be less difficult [9].
2.2 Preferences of senior users
There is still little research into the preferences of senior citizens regarding pedagogical agents. One
of the few studies is that of Straßmann and Krämer (2017). They examined, in a qualitative study, the
preference of the appearance of virtual agents among five seniors (age 61–74) and six students, and
found age-dependent differences. Seniors preferred more realistic humanoid agents, while students
preferred more zoomorphic/machine-like agents. 2D or 3D model of the agent was irrelevant for both
groups. Seniors preferred the social effects and want to naturally interact with the agent [10].
In another study Straßmann et al. (2020) examined the effects of species, realism and embodiment
regarding age-related differences on a health-related human agent interaction with 84 students and 46
seniors (age 51-89). The seniors rated the interaction more positively than the students and showed
more bonding toward the agent regardless of the appearance. Seniors reported the highest use
intention for the cartoon-stylized humanoid agent and showed a stronger bond when interacting with
the cartoon-stylizing humanoid or the voice-only agent [11].
Esposito et al. (2019) showed in a study with 46 seniors (age 65+), that the seniors preferred female
virtual humanoid agents, regardless of their own gender. Additionally, tech-savvy seniors found the
agents less motivating, engaging, excitingly and engaging [12].
Feledichuk (2019) examined the design preferences regarding pedagogical agents for animation,
communication and voice, graphical style, agent role, competence, facial expression, gender, body
shape, ethnicity, age, and attire of 23 seniors. Results showed that there was no preference for an
animated or static agent. Regarding non-verbal communication, 16 senior citizens preferred an easy-
going demeanour. In terms of communication style, the participants preferred task-oriented and task-
and relational-oriented. Almost all seniors preferred a human-generated agent voice and many of
them wished for a realistic designed agent. The preferred agent roles were expert and teacher.
Confidence was chosen most frequently for the competence level. Regarding the facial expression, a
friendly face received best rating. There were no specific preferences for agent’s gender, body shape,
ethnicity or attire. Most seniors said they had no preference of age, or they chose the age of 41-55
years [13].
2.3 Frameworks for designing pedagogical agents
Baylor (2004) presented four dimensions of control that should be considered when designing agent-
based learning environments. This includes the instructional purpose of the environment, feedback
parameters like type, timing, amount, explicitness and learner adaptability of agent feedback, and the
relationship of the learner to agent (e.g. agent as learning companion, agent as mentor). The fourth
dimension implies that the learner must be able to develop confidence in the agent in terms of
credibility, competence and trust [14].
Ryu and Baylor (2005) developed a four-factor-model for measuring the perception of the
psychometric structure of pedagogical agents. They identified information usefulness and affective
interaction as latent variables that affect the perception of four identified factors of a pedagogical agent
persona. These include credible, facilitating, engaging and human-like [15].
7893

Veletsianos et al. (2010) presented a framework (EnALI - Enhancing Agent Lerner Interactions
Framework) for the design of pedagogical agents. This consists of three design foci – user interaction,
message, and agent characteristics, and provides useful guidelines for implementation [16].
2.4 Types of pedagogical agents
Some publications describe different types or approaches of virtual pedagogical agents. This includes
embodied (conversational) pedagogical agents (ECA, EPA), animated pedagogical agents (APA) and
pedagogical agents as learning companions (PAL). The three types are presented in more detail
below.
2.4.1 Pedagogical agents as learning companions (PALs)
Peer interaction in general has a positive impact for learning and motivation. Pedagogical agents as
learning companions (PALs) can provide an opportunity to simulate social interaction in computer-
based learning [17]. Kim et al. (2006) defined PALs „[…] as animated peer-like characters that
simulate peer interaction in computer-based learning.” [18]. Based on the theory of distributed
cognition, social Interaction and social-cognitive theory Kim and Baylor (2006) developed a social-
cognitive framework for designing PALs. This includes the following components: competency,
interaction type, gender, affect, ethnicity, multiplicity, and feedback [17]. For the competency and
interaction type Kim et al. (2006) conducted a study to examine the effects of competence (low vs.
high) and interaction type (proactive vs. responsive) of pedagogical agents as learning companions on
learning, self-efficacy and attitudes with 72 undergraduates in an introductory computer-literacy
course. Regardless of the type of interaction students who worked with the high-competency PAL
achieved higher scores in learning outcomes and showed positive attitudes towards the PAL.
However, working with a PAL with low competence resulted in significantly better self-efficacy beliefs
in the learning tasks. In addition, a proactive PAL had a significantly positive impact on the recall [18].
2.4.2 Embodied (conversational) agents (EPA and ECA)
Haake and Gulz (2009) described embodied pedagogical agents as “[…] visually represented,
computer generated characters in pedagogical roles, such as virtual instructors, mentors and learning
companions […]” [19]. Based on a framework they developed for the design of embodied pedagogical
agents; they examined the preferences of 90 school children. The framework consists of three aspects
which should be considered when designing EPAs: visual static appearance (detailed & 3D-rendered
vs. simplified & cartoonish), educational role (instructor vs. learning companion) and communication
style (strictly task-oriented vs. task- & relation-oriented). Results showed that students who chose
more stylized characters tended to be significantly more task- and relational-oriented. If the agent was
an instructor, female students tended to prefer a stylized visual character. If the agent was a learning
companion, female students showed a significant preference for task and relational agents [19].
Scholten et al. (2019) investigated the use of a pedagogical agent as an embodied conversational
agent in an eHealth self-management intervention program with 230 psychology students. The agent
was presented in four conditions: animated and speech (non-text), still and speech (non-text), still and
text (non-speech) and text-only. The students rated the following variables: autonomy, feedback,
relevance, attention, involvement, and rapport. A visible ECA led to significantly higher values for the
evaluation of the feedback and the autonomy. When comparing speech and text, only a positive effect
of speech on the evaluation of the feedback could be identified. Animation did not have a significant
effect on any of the measured variables. Furthermore, no effects could be determined between the
four different agent conditions [20].
2.4.3 Animated pedagogical agents (APA)
Lester et al. (1997) examined the impact of the communication behaviour of animated pedagogical
agents in interactive learning environments on the learning experience of 100 students. All post-test
results showed significantly higher values regardless of the agent's communication behaviour and a
positive learning effect could therefore be identified. The agents were rated differently, with significant
differences in the "fully expressive" agent in relation to the other four agents (principle-based
animated/verbal, principle-based verbal, task-specific verbal and muted) being found. This design
option also reached highest values in the evaluation [3].
Shaw et al. (2000) explored the use of an animated pedagogical agent named ADELE in a medical
learning program on 25 medical students. The students preferred the animated pedagogical agent to a
7894

text-only tutor and preferred a real voice rather than a synthesized one. The students didn't mind that
the agent's lips and voice were out of sync [21].
2.5 Limitations in current research in designing pedagogical agents
In a literature review of 28 studies, Schroeder and Gotch (2015) identified three key types of existing
limitations in research on pedagogical agents. In most studies (18) the agent was used in the role of
the information source. Ten studies used the agent for coaching scaffolding. None of the studies used
an agent for demonstration, modelling or testing. Further limitations include the methodological
conditions such as research design, measurement, and sampling. In addition, most of the studies did
not provide any information on the costs of implementing the agent or the cost efficiency [22].
3 REQUIREMENTS FOR DESIGNING PEDAGOGICAL AGENTS
Given a broad range of terms in this field of research, from virtual agents and pedagogical agents to
embodied (conversational) pedagogical agents (ECA, EPA), animated pedagogical agents (APA),
pedagogical agents as learning companions (PAL), many design options are available. However, the
question remains how to design a virtual agent in a systematic way and which criteria should be used.
Below we present relevant frameworks and design aspects which are useful for further agent design.
3.1 Frameworks
The development and design of pedagogical agents can be conducted in a systematic way, based on
the available frameworks. Some of the most popular frameworks include the four dimensions of
control by Baylor (2004) [14], the four-factor-model for measuring the perception of the psychometric
structure by Ryu & Baylor (2005) [15], the framework enALI by Veletsianos et al. (2010) [16], the
social-cognitive framework by Kim & Baylor (2006) [17], and the framework by Haake & Gulz (2009)
[19]. The frequently mentioned aspects that should be considered when designing pedagogical agents
are: visual style / appearance [15,16,17,19], communication style / interaction [16,17,19], feedback /
message [14,16,17], competence and credibility [14,15,17], and agent-role [14,19]. Some of the
components of the frameworks mentioned above were examined in previous, for example in relation to
the influence on learning and motivation as well as the preferences of the participants in the studies. It
should be noticed that most previous studies were conducted with school and university students and
that the sample sizes were rather quite small. There has been only a few studies with senior users.
Below we describe selected design aspects and research results.
3.2 Visual appearance
Regarding the visual style or appearance, the realistic pedagogical agent has achieved better post-
test results in a number of studies including [7], [8], [10] and [13]. Furthermore, a preference for
realistic humanoid agents were found among senior learners [10,13]. However, [6], [10] and [11]
identified a preference for iconic, zoomorphic/ machine-like, and cartoon-stylized human agents (last
one by seniors). The second most popular form of pedagogical agents in studies seems to be the 3D-
character [2]. A comparison of the preference between 2D and 3D characters was examined by [10].
Both students and seniors found it irrelevant whether the character was designed in 2D or 3D [10].
Regarding gender, the senior learners seem to prefer female virtual humanoid agents [12] or have no
preference for gender [13]. Senior participants tend to chose a more female type, if they are free to
choose [7]. Regarding the facial expressions, seniors learners seem to prefer friendly faces [13] with
smooth skin, realistic proportions, and natural average skin color [7]. Female faces are associated with
full lips, snub nose and slightly upturned eyes, while male faces with got strong eyebrows, downturned
eyes, a larger throat, and thin lips [7]. The hair color seems to be irrelevant for perceived likability [7].
Another aspect of the appearance is the animation. [5] found a positive effect of animation on learning,
while students in the study by [21] preferred the animated agent to a text-only tutor. In addition, in the
study by [3], students preferred the animated "fully expressive" agent. On the contrary, in the study by
[5], negative effects of animation were seen in facilitating learning, motivation, satisfaction and self-
efficacy. The study by [20] found no significant effect of animation and seniors and the study by [13]
showed no preference for an animated or static agent.
In summary, the results for the visual style of pedagogical agents are contradictory. The preferences
for a more realistic / human-like vs. a more unrealistic designed agent are not consistent across the
studies, but there is a tendency for a preference for human-like agents, especially among senior
7895

users. Regarding the gender of the agent, there is a tendency to like female agents and there are no
preferences in relation to the form (2D or 3D). Although 3D-agents were used more often in previous
studies, there was no comparison to 2D-agents. In addition, it seems that there tends to be a positive
attitude towards animated pedagogical agents, but there are no clear results in this regard. Past study
results however indicate that the facial expressions should be friendly and typical for a given gender.
3.3 Communication and interaction
In terms of communication style and interaction, the study by [6] revealed a preference for relational-
oriented. The study by [13] showed that the seniors preferred task-oriented and task- & relational-
oriented interaction. [6] found a preference for task- & relational-oriented (compared to strictly task-
oriented) agents when a more stylized character. In addition, female students showed a significant
tendency for a preference for task and relational agents if the agent was a learning companion [19].
[20] identified a positive effect of speech (compared to text) on the evaluation of the feedback, but no
effect for autonomy, relevance, attention, involvement, and rapport. [4] found that the perception of
feedback (positive, neutral, negative) depended on the linguistic expressions and mouth curve. [3]
found that students preferred the verbal (and animated) "fully expressive" agent. Regarding non-verbal
communication, a large part of the senior citizens in the study by [13] preferred an easy-going
demeanour. [9] found that learning from a lesson written in a conversational style or having a
conversational style pedagogical agent enhanced retention, but also increased pressure and mental
effort. Regarding the voice of pedagogical agents, the students in the study by [21] preferred a real
voice rather than a synthesized one. Seniors learners in the study by [13] also liked a more human-
generated agent voice. [5] found on the one hand a positive effect of the human voice on the
perceived agent persona and a greater motivation when the agent was not animated and had a
human voice. On the other hand, the students were more motivated when the agent combined
animation and machine-generated voice [5].
In summary, there seems to be a preference for more relational-oriented agents or a combination of
task-oriented and relational-oriented interaction styles. Regarding the use of speech and
conversational style a slight positive trend can be seen and there seems to be a preference and more
positive effects for a real, human-generated voice. In general, the results related to the preferences of
learners for specific communication style are still very ambiguous.
3.4 Competence and role
In terms of competence and role, [18] used an pedagogical agent as a learning companion (PAL) and
found that the high-competency PAL achieved higher scores in learning outcomes and a low-
competence PAL resulted in better self-efficacy beliefs in learning tasks. In addition, a proactive PAL
had a positive impact on the recall [18]. Seniors seem to chose the most "confident" competence level
and prefer experts and teachers as agent roles [13]. As observed by [22], most studies the agent was
used in the role of an information source and only a few studies used the agent for coaching and
scaffolding [22]. In the studies described in this paper, the agent was used mostly for giving help and
feedback [3, 4, 5, 6, 11, 20] as well as an expert also giving instructions [5, 8, 9, 15, 19, 20]. In
summary, there seems to be a tendency to prefer agents with high competence in their role as
information provider and supporter.
4 TECHNOLOGIES FOR DESIGNING PEDAGOGICAL AGENTS
In seven of the studies described above, which included an implemented agent, no details were given
on the technical implementation of the agents [8, 9, 11, 13, 15, 18, 20]. [3] developed an agent with
the help of an interdisciplinary team of computer scientists, graphic designers, and animators, but did
not provide specific information on the tools and techniques applied in the design.
Nevertheless, some authors provide insights into the technologies used for designing pedagogical
agents. For example, the agent in the study by [21] was developed as a simple Java applet that runs
on the client side. In six studies the agent was developed using already existing tools and software [4,
5, 6, 7, 12, 19]. For the implementation of the voice in [4, 5], the Microsoft Agent speech engine was
used and recorded human voice [5]. [12] used the website Natural Reader (www.naturalreaders.com)
in combination with the free software Audacity for recording.
For the implementation of the appearance, the Microsoft Agent software was used in [4], the 3D
computer graphics program Poser in [5], Macromedia Director and 3D Studio Max 5 (with plug-in
7896

module FacialStudio) in [6] and [19], and the website BOTLIBRE (www.botlibre.com) in [12]. [7]
developed a 3D avatar generator called facemaker (http://facemaker.uvrg.org/) and used the Face
Database of the Park Aging Mind Laboratory (PAL), PsychoMorph, the FaceGen4 software and
PhotoFit feature, Autodesk Maya 2014 and Mudbox 2014 for the development. This generator can be
used by others to design the face of a 3D agent. In the study by [6], the Lo-Fi paper sheet prototypes
of an imaginary pedagogical multimedia program were used as an analog method. In addition, the
free generator Avatar Maker (https://avatarmaker.com/) for designing 2D faces can be used in design.
In summary, past studies provide only limited information about the technical tools and the
implementation of agents. Only a few of the tools used in previous studies are freely available. In
addition some tools such as the Microsoft Agent tool, are out of date and no longer available [23].
5 DESIGN OPTIONS AND CONSIDERATIONS FOR EPA-COACH PROJECT
Based on the results of the literature review described above, we defined design options for four
virtual agents (Lisa, Maria, Max, and Peter) to be tested in the ePA-Coach project with the group of
senior learners. The design options include the visual style, communication and social style, and the
competence and role. The agent "Lisa" incorporates all the preferred properties from the studies
described above to represents the "perfect" virtual agent. The other three agents have some similar
properties to Lisa and additionally have opposing properties for which the research results are still
inconsistent as compared to Lisa's properties. We plan to design and test the mockups of the four
agents and choose the most likeable option based on the preferences of senior learners.
5.1 Visual style
As previously described, a slight preference for female agents and faces was identified in the past
studies. The ePA-Coach agent mockups will be designed use female and male agents to check
possible gender preferences among senior learners. Sine some studies showed preferences for the
animation of the agent, wo of the four ePA-Coach agents will be animated to test possible
preferences. Past studies showed no clear preferences for age, and no one chose an age below 26.
Accordingly, we would first assign different age classes to the four agents. The shape of the agent (2D
or 3D) seemed irrelevant, with most of the studies using 3D characters. To have another comparison
in this regard, we will use both 2D and 3D variants. Regarding the realism of the agent, there seems to
be a tendency for agents who are realistic and human-like. Therefore, the four ePA-Coach agents will
be designed in a realistic and human-like style. Facial expression and face style of ePA-Coach agents
will be based on the animation and gender-specific preferences from the study by [7]. The design
decisions for ePA-Coach agents are summarized in Table 1 below.
Table 1: Visual Style for virtual agent in ePA-Coach project
Lisa (1)
Maria (2)
Max (3)
Peter (4)
Gender
female
female
male
male
Animation
yes
no
yes
no
Age
35
60
35
60
Form
3D
2D
3D
2D
Realism
human-like
human-like
human-like
human-like
Facial expr.
mouth: smiling
(default), neutral, sad,
open, closed
mouth: slightly smiling
(fixed)
mouth: slightly smiling
(fixed)
mouth: smiling
(default), neutral, sad,
open, closed
Face style
smooth skin, realistic proportions,
natural skin color, full lips, snub nose, slightly
upturned eyes
smooth skin, realistic proportions, natural skin
color, strong eyebrows, downturned eyes,
larger throat, and thin lips
5.2 Communication and social style
A preference for relational-oriented agents or a combination of task-oriented and relational-oriented
could be found in some studies described before. Therefore, two of the ePA-Coach agents will be
designed according to the relational-oriented social style and the other two agents will have the
7897

combination of both styles. In addition, we will test the preference for speech vs. text communication.
The results for voice are rather ambiguous, with a slight preference for human-generated voice. We
will design one agent with a human-generated voice and one with a machine- generated voice. The
two agents with the text speech style will have no voice. These design decisions are summarized in
Table 2. In addition, we will follow the guidelines of the EnALI-Framework by [16].
Table 2: Communication and social style for virtual agent in ePA-Coach project
Lisa (1)
Maria (2)
Max (3)
Peter (4)
Social style
relational-oriented
task- & relational-
oriented
task- & relational-
oriented
task-oriented
Speech style
speech
text
speech
text
Voice
human
-
machine
-
5.3 Pedagogical role
Regarding the pedagogical role and competence of the agent, there seems to be a preference for
agents with high competence as information source and supporter. In the ePA-Coach project the
agent should act as a coach for senior learners but is open about who exactly this coach represents
and how the agent should coach the seniors. Table 3 shows four possibilities that we have
considered.
Table 3: Pedagogical role for virtual agent in ePA-Coach project
Lisa (1)
Maria (2)
Max (3)
Peter (4)
role
pedagogic-expert
eLearning-expert
ePA-expert
health-expert
job /
qualification
geriatric educator
professor for
educational technology
Gematik GmbH
employee
doctor for geriatrics
competence
pedagogic
geriatrics
ePA
technology
high
middle
low
low
middle
low
low
high
low
low
high
high
low
high
middle
low
6 CONCLUSIONS
In this paper we described the current literature of designing pedagogical agents and derived design
options for a pedagogical agent as a virtual learning coach as part of the e-learning system for digital
literacy of senior learners in the project ePA-Coach. As part of the literature review, we described
studies and frameworks for the design of pedagogical agents. In addition, we presented different types
of agents (PAL, EPA, ECA and APA) and showed limitations in current research. In the next step, we
summarized requirements for designing virtual agents including the main aspects of existing
frameworks and state of research in relation to visual appearance, communication and interaction, and
competence and role. We showed that preferences for pedagogical agents are not consistent or even
contradictory and past studies could identify only slight tendencies with small samples. Some of the
preferences includes human-like designed agents, especially among seniors, or a slight preference for
high competence agents and female agents. In particular, there is a lack of studies on the preferences
among senior learners. We also showed which technologies were used for designing pedagogical
agents and identified that many studies give no details on the technical implementation. Finally, we
described the considerations for the ePA-Coach project and the design options for four different
agents including the visual style, the communication and social style, and the pedagogical role.
The next steps in research and development for the realization of a virtual learning coach
(pedagogical agent) in the ePA-Coach will be the design and testing of the mock-ups of the four
agents with the goal to identify preferred options for designing a final learning coach version in the
7898

ePA-Coach project. We will examine the possibilities for the implementation of the agent intelligence
and behaviour and will design and implement at least one prototype for a virtual learning coach.
Further research is necessary for designing virtual agents, especially research with larger and diverse
samples, including senior learners. Research results from past research are ambiguous, and no
generalisable findings could be derived from the literature review conducted in this paper. In particular,
there is a lack of research on the agent preferences of senior users. This research is necessary
because studies have shown that there can be relevant differences in preferences among different
target groups. In addition, future studies should not only show effects and preferences for pedagogical
agents, but also provide detailed information on the technical tools and the implementation. Finally,
virtual agents should be designed and examined in diverse roles, for example in the role of a
(learning) mentor or a coach. Most past studies, as described in this paper, focused only on the role of
agents as instructors and sources of information.
ACKNOWLEDGEMENTS
This publication was produced as part of the project ePA-Coach: Digital sovereignty in context of the
electronic patient file, founded by the Federal Ministry of Education and Research under the program
Human-technology interaction for digital sovereignty. For more information please visit: https://technik-
zum-menschen-bringen.de/projekte/epa-coach
REFERENCES
[1] A. Pott, “EHR and PHR : digital records in the German healthcare system,” Healthcare industry,
2019. https://www.gesundheitsindustrie-bw.de/en/article/news/
ehr-and-phr-digital-records-in-the-german-healthcare-system (accessed Aug. 13, 2020).
[2] A. S. D. Martha and H. B. Santoso, “The design and impact of the pedagogical agent: A
systematic literature review,” J. Educ. Online, vol. 16, no. 1, 2019, doi: 10.9743/jeo.2019.16.1.8.
[3] J. C. Lester, S. T. Barlow, S. A. Converse, B. A. Stone, S. E. Kahler, and R. S. Bhogal, “Persona
effect: Affective impact of animated pedagogical agents,” Conf. Hum. Factors Comput. Syst. -
Proc., pp. 359–366, 1997.
[4] K. E. Link, R. J. Kreuz, and A. C. Graesser, “Factors that influence the perception of feedback
delivered by a pedagogical agent,” Int. J. Speech Technol., vol. 4, no. 2, pp. 145–153, 2001, doi:
10.1023/A:1017383528041.
[5] A. L. Baylor, J. Ryu, and E. Shen, “The effects of pedagogical agent voice and animation on
learning, motivation and perceived persona,” ED-MEDIA 2003 Proc. World Conf. Educ.
Multimedia, Hypermedia Telecommun., pp. 452–458, 2003.
[6] A. Gulz and M. Haake, “Social and Visual Style in Virtual Pedagogical Agents,” Proc. Work.
Adapt. Interact. Style to Affect. Factors, 10th Int. Conf. User Model., 2005.
[7] V. Schwind, “Implications of the Uncanny Valley of Avatars and Virtual Characters for Human-
Computer Interaction,” Universität Stuttgart, 2018.
[8] V. Salehi and F. T. Nia, “Effect of levels of realism in mobile-based pedagogical agents on health
e-learning,” Futur. Med. Educ. J., vol. 9, no. 2, pp. 40–45, 2019, doi:
10.22038/fmej.2019.39824.1261.
[9] L. Lin, P. Ginns, T. Wang, and P. Zhang, “Using a pedagogical agent to deliver conversational
style instruction: What benefits can you obtain?,” Comput. Educ., vol. 143, 2020, doi:
10.1016/j.compedu.2019.103658.
[10] C. Straßmann and N. C. Krämer, “A categorization of virtual agent appearances and a qualitative
study on age-related user preferences Categorization of appearance variables,” in Intelligent
Virtual Agents. IVA 2017. Lecture Notes in Computer Science, 10498th ed., J. Beskow, C. Peters,
G. Castellano, C. O’Sullivan, I. Leite, and S. Kopp, Eds. Springer, Cham, 2017, pp. 413–422.
[11] C. Straßmann, N. C. Krämer, H. Buschmeier, and S. Kopp, “Age-Related Differences in the
Evaluation of a Virtual Health Agent ’ s Appearance and Embodiment in a Health-Related
Interaction : Experimental Lab Study Corresponding Author :,” J. Med. INTERNET Res., vol. 22,
no. 4, pp. 1–15, 2020, doi: 10.2196/13726.
7899

[12] A. Esposito et al., “Seniors’ acceptance of virtual humanoid agents,” in Ambient Assisted Living,
Lecture Notes in Electrical Engineering, 544th ed., vol. 544, A. Leone, C. A., G. Rescio, D. G., and
P. Siciliano, Eds. Springer Nature Switzerland AG 2019, 2019, pp. 429–443.
[13] D. G. Feledichuk, “The preferred design of pedagogical agents by older adults for self-
management health programs for chronic conditions,” University of Alberta, 2019.
[14] A. L. Baylor, “Permutations of Control: Cognitive Considerations for Agent-Based Learning
Environments.,” J. Interact. Learn. Res., vol. 15, no. 4, pp. 403–425, 2004.
[15] J. Ryu and A. L. Baylor, “The Psychometric Structure of Pedagogical Agent Persona,” Technol.
Instr. Cogn. Learn., vol. 2, no. 4, pp. 291–315, 2005.
[16] G. Veletsianos, C. Miller, and A. Doering, “EnALI: A Research and Design Framework for Virtual
Characters and Pedagogical Agents,” J. Educ. Comput. Res., vol. 41, no. 2, pp. 171–194, 2010.
[17] Y. Kim and A. L. Baylor, “A social-cognitive framework for pedagogical agents as learning
companions,” Educ. Technol. Res. Dev., vol. 54, no. 6, pp. 569–596, 2006, doi: 10.1007/s11423-
006-0637-3.
[18] Y. Kim, A. Baylor, and P. Group, “Pedagogical Agents as Learning Companions : The Role of
Agent Competency and Type of Interaction,” ITLS Fac. Publ., 2006, doi: 10.1007/s11423-006-
8805-z.
[19] M. Haake and A. Gulz, “A Look at the Roles of Look & Roles in Embodied Pedagogical Agents –
A User Preference Perspective,” Artif. Intell. Educ., vol. 19, pp. 39–71, 2009.
[20] M. R. Scholten, S. M. Kelders, and J. E. W. . C. Van Gemert-Pijnen, “An Empirical Study of a
Pedagogical Agent as an Adjunct to an eHealth Self-Management Intervention : What Modalities
Does It Need to Successfully Support and Motivate Users ?,” Front. Psychol., vol. 10, no. 1063,
2019, doi: 10.3389/fpsyg.2019.01063.
[21] E. Shaw, R. Ganeshan, W. L. Johnson, and D. Millar, “Building a case for agent-assisted learning
as a catalyst for curriculum reform in medical education,” Med. Educ. AI-ED’ 99 Work. Animat.
Pers. Pedagog. Agents, Le Mans, pp. 509–516, 2000.
[22] N. L. Schroeder and C. M. Gotch, “Persisting Issues in Pedagogical Agent,” J. Educ. Comput.
Res., vol. 53, no. 2, pp. 183–204, 2015, doi: 10.1177/0735633115597625.
[23] Microsoft, “Microsoft Agent,” 2018. https://docs.microsoft.com/en-us/windows/win32/lwef/
microsoft-agent (accessed Sep. 22, 2020).
7900