=Paper=
{{Paper
|id=Vol-2141/paper3
|storemode=property
|title=An Analysis of Student Belief and Behavior in Learning by Explaining to a Digital Doppelganger
|pdfUrl=https://ceur-ws.org/Vol-2141/paper3.pdf
|volume=Vol-2141
|authors=Ning Wang,Ari Shapiro,Andrew Feng,Cindy Zhuang,David Schwartz,Stephen L. Goldberg
}}
==An Analysis of Student Belief and Behavior in Learning by Explaining to a Digital Doppelganger==
https://ceur-ws.org/Vol-2141/paper3.pdf
An Analysis of Student Belief and Behavior in
Learning by Explaining to a Digital Doppelganger
Ning Wang Ari Shapiro Andrew Feng
Institute for Creative Technologies Institute for Creative Technologies Institute for Creative Technologies
University of Southern California University of Southern California University of Southern California
Playa Vista, CA Playa Vista, CA Playa Vista, CA
nwang@ict.usc.edu shapiro@ict.usc.edu feng@ict.usc.edu
Cindy Zhuang David Schwartz Stephen L. Goldberg
Institute for Creative Technologies Department of Psychology STTC
University of Southern California University of Southern California US Army Research Laboratory
Playa Vista, CA Los Angeles, CA Orlando, FL
czhuang@ict.usc.edu davschw@usc.edu stephen.l.goldberg.civ@mail.mil
ABSTRACT adapt to learner needs in order to maximize the efficacy of the
Digital doppelgangers are virtual humans that highly resemble the personalized experience in facilitating student learning [31].
real self but behave independently. Using a low-cost and high-speed One aspect of such personalization that is less studied is the
computer graphics and character animation technology, we created agent’s appearance. Research on a pedagogical agent’s appearance
digital doppelgangers of students and placed them in a learning-by- has indicated the impact of such design decisions on learning out-
explaining task where they interacted with digital doppelgangers comes (e.g. [5], for review see [31]). However, research questions
of themselves. We investigate the research question of how does along the dimension of agent similarity with the learner are largely
increasing the similarity of the physical appearance between the left unanswered. For example, when a pedagogical agent shares
agent and the student impact learning. This paper discusses the the exact appearance and behavioral characteristics of the learner,
design and evaluation of a digital doppelganger as a virtual human will such increased resemblance improve, make no difference on,
listener in a learning-by-explaining paradigm. It presents an analy- or hamper learning? In the Teachable Agents [7] paradigm, for
sis of how students’ perceptions of the resemblance impact their example, where students cultivate their protege by teaching a ped-
learning experience and outcomes. The analysis and results offer agogical agent, would an agent that is a digital doppelganger of a
insight into the promise and limitation of the application of this student make him/her more motivated to help the protege learn?
novel technology to pedagogical agents research. To systematically study the impact of the similarity of the peda-
gogical agents’ appearance to students requires the generation of
CCS CONCEPTS such agents for a large enough population and at sufficient speed
to accommodate experiment sessions of limited duration.
• Human-centered computing → Empirical studies in HCI ; •
An emerging technology, the Rapid Avatar Capture and Sim-
Applied computing → Interactive learning environments;
ulation (RACAS) system [33], enables low-cost and high-speed
scanning of a human user and creation of a fully animatable 3D
KEYWORDS “digital doppelganger” — a virtual human that highly resembles the
pedagogical agent, learning by explaining, rapid avatar capture and real self but behaves independently [3]. This allows researchers in
simulation pedagogical agents to explore a previously unexplored research
question: how does personalizing pedagogical agent’s appearance
1 INTRODUCTION to share physical similarity with the student impact learning? A
The power of intelligent tutoring systems lies in the personalized previous study has suggested that there is limited impact of such
learning experience tailored to individual student’s needs. Much of personalization on learning outcomes in a learning-by-explaining
such personalization focuses on understanding a student’s cogni- paradigm [41]. In this paper, we discuss the design of a digital
tive, affective, and motivational states, and the adaptation of the doppelganger as a virtual human listener and a follow-up analysis
tutorial feedback to such states. The adaptation is not limited to of student perception of personalization and the impact of such
just the content of the feedback but also its delivery, such as the perception on student behavior and learning.
agent that delivers the feedback. For example, researchers in peda-
gogical agents—embodied animated virtual characters designed to 2 RELATED WORK
help students learn [21]—have experimented with many aspects of
Pedagogical Agents: There is a large body of work on pedagogical
pedagogical agent design, including animation [24], gesture [13],
agents across diverse learning domains. Most relevant to this work
voice [27], and social intelligence [40], and how such design can
is the research on agent appearance. A series of studies conducted
by Baylor and Kim examined the realism of agent appearance, its
© 2018 Copyright held by the owner/author(s).
gender, and ethnicity. Results showed that agents with a realistic
PALE’2018, June 2018, London, UK N. Wang et al.
image improved transfer of learning, while the agent’s gender and a fixed 3D sensor to acquire four key poses from a self-turning sub-
ethnicity contributed to its perceived competency, which in turn ject. Additionally, to create an animated virtual character from the
impacted student’s self-efficacy [5]. Veletsianos and colleagues ma- scanned 3D human model, the 3D model needs to be rigged with a
nipulated the agent’s appearance relevant to stereotypes (artists skeleton hierarchy and appropriate skinning weights. Traditionally,
vs. scientists) and found that such manipuations impacted the per- this process needs to be done manually and is time-consuming even
ceived knowledge level of the agent and students’ performance on for an experienced animator. An automatic skinning method was
recall tasks [38]. Domagk studied the appealingness of the agent’s proposed in [4] that can produce reasonable results while reducing
appearance and uncovered the impact of this variable on students’ the manual efforts of rigging, but requires a connected and wa-
transfer of learning [14]. In more recent work, Finkelstein and her tertight mesh to work. The method proposed by [6] complements
colleagues extended prior studies of agent appearance by looking the previous work by automatically skinning a multi-component
at not only the appearance of the agent relative to the learner (e.g., mesh. It works by detecting the boundaries between disconnected
ethnicity), but also behaviors consistent with such appearance (e.g., components to find potential joints. Such a method is suitable for
the use of dialect) [16]. Results highlighted both the benefit (e.g., rigging mechanical characters that consist of many components.
improved student learning) and potential drawback (e.g, decreased Other rigging algorithms can include manual annotation to identify
perceived agent competence) of the similarity between the agent important structures such as the wrists, knees, and neck [26]. In
and the learner. Indeed, personalizing the pedagogical agent to recent years, video-based methods have enabled the capture and
share the exact physical appearance with the student may be a reconstruction of human motion as a sequence of 3D models that
multi-faceted issue. For example, interacting with an avatar looking can preserve both the captured appearance and actor style, without
like the real self have shown to induce anxiety in a public speaking the need of a rigging step [34]. Such methods, which are capable of
task, compared to a dissimilar avatar [2]. In a learning paradigm reproducing surface and appearance details over time, have been
such as learning-by-teaching, such anxiety may hinder student used to synthesize animation by combining a set of mesh sequences
learning. [8]. However, current approaches are limited to basic locomotion
Learning by Explaining: We will study the impact of a digital such as walking, jogging, and jumping. Rigging of the video-based
doppelganger in the context of learning-by-explaining. Learning- 3D models is still needed for complex movements.
by-explaining is an effective learning technique used by human
tutors. In this technique, students are encouraged to explain a con-
3 DIGITAL DOPPELGANGERS WITH RACAS
cept either to another or themselves. Decades of research shows
that generating such explanations can lead to deep understanding The RACAS system is a virtual avatar generation system based
of the learning material [10, 28], and that these learning effects on a 3D body scan. The RACAS system has two main capabilities:
are particularly strong when the explanations are delivered in a automatic rigging transfer and interactive avatar reshaping [15, 33].
social context (i.e., explaining to a peer or tutor), as opposed to The RACAS takes scans of a user from the front, back, left, and right
explaining to oneself [12, 43]. These effects have been observed sides using an RGB-D sensor. These scans are “stitched together”
when the “other” is a computer-generated character: For example, to create a 3D model. The 3D model is then enhanced by inferring
Cassell found that kids who told stories to a virtual partner devel- a skeletal and muscular structure, as well as generating a model
oped better language skills [9]. Holmes found that students who for the deformation of the skin and clothes. To do so, RACAS first
explained to a software learning partner spent 50% more time ask- utilizes SCAPE [1] to build a morphable human model from a 3D
ing questions and making explanations and generated significantly human model database. In order to allow pose deformations via
deeper explanations compared to students who worked without linear blend skinning, RACAS researchers also manually rigged
a software learning partner [20]. Leelawong and Biswas demon- a template mesh from the database. Therefore given a 3D human
strated that students benefitted from teaching a virtual peer [22]. body scan, RACAS can fit the morphable human model produced
We will leverage this effective learning paradigm as our testbed and by SCAPE onto the input scan and establish mesh correspondences
use pedagogical agents, played by either a virtual human or a digital between them. Once RACAS establishes such correspondences,
doppelganger, to create the social context to facilitate learning by they can be used to transfer both skeleton and skin binding weights
explaining. from the template mesh onto the input scan to generate a 3D virtual
Digital Doppelgangers: Much effort in computer graphics and avatar. The users of RACAS can also interactively adjust semantic
animation has been put forth to create digital doppengangers. Cre- body attributes of the fitted model by exploring the body shape
ating a virtual character from a particular subject is not a trivial space generated from the database. Such body shape deformations
task and usually requires extensive work from a 3D artist to model, can then be transferred to the aforementioned 3D scan to further
rig, and animate the virtual character. To reconstruct a 3D model at create various virtual avatars with different body sizes and pro-
low cost, researchers have experimented with low-cost 3D cameras portions. The resulting virtual avatars can then be animated in a
(e.g., Kinect and Primesense) for 3D human shape acquisition. The simulation environment to execute various behaviors using anima-
work by [37] employs three Kinect devices and a turntable. Multiple tion retargeting. SmartBody, a character animation system, drives
shots are taken, and all frames are registered using the Embedded the animation of the 3D virtual character [32]. Using RACAS, re-
Deformation Model [35]. The work done in [45] utilizes two Kinect searchers can easily create a digital doppelganger that serves as an
sensors in front of the self-turning subject. The subject stops at ideal model for maximizing feelings of similarity, and customizing
several key poses, and the captured frame is used to update the the virtual self’s behavior to portray an optimal performance that
online model. Instead of multiple 3D sensors, the work in [42] used the physical self cannot yet achieve.
Student Belief and Behavior in Learning by Explaining to a Digital Doppelganger PALE’2018, June 2018, London, UK
4 EXPLAINING TO A DIGITAL
DOPPELGANGER
We designed a virtual human listener and incorporated the digital
doppelgangers created by RACAS to embody the virtual human
listener. A human speaker, such as a student, can interact with the
virtual human listener by talking to the it—for example, explaining
what the student has learned. The virtual human listener provides
conversational backchannel feedback as the student explains the
concepts [19]. The feedback is generated based on analysis of the
student’s nonverbal behavior, such as head nods, prosody, etc. [19].
Previous research on virtual human listeners has shown the value
of such feedback in creating rapport with the human speaker [19].
In the work discussed here, we focus on examining the impact of Figure 1: Virtual Human listeners, captured using RACAS,
personalizing the agent’s appearance on measures related to stu- from the control condition.
dent learning. We hypothesize that teaching a virtual agent who
looks just like oneself can impact a learner’s motivation and self-
regulation in learning (e.g. coping with challenges and persisting
in a learning task). As a result of the increased the motivation, stu- to match the obvious gender matching in the Digital Dop-
dents may spend more effort to explain the learning concepts to pelganger condition. Figure 1 shows the female and male
the virtual human listener, resulting in better learning outcomes virtual human listeners used in this condition.
and higher self-efficacy. Specifically, we hypothesize that a virtual
human listener that shares the appearance of the learner can im-
prove learner motivation to teach the virtual human listener in 5.2 Population
a learning-by-explaining paradigm (Hypothesis 1). Additionally,
such virtual human listeners can improve student learning of do- We recruited 41 participants from the student population of a uni-
main knowledge (Hypothesis 2) and improve student self-efficacy versity located in North America. Participants were recruited either
(Hypothesis 3). from the Psychology Department subject pool or via fliers posted
on campus. Participants recruited from the subject pool received
course credit for their participation, while participants recruited
5 EVALUATION through the fliers received $10 for their participation.
5.1 Design
In order to test our hypotheses on a pedagogical agent’s appear- 5.3 Procedure
ance’s impact on student learning, we conducted a study with the Participants first read and signed an informed consent about the
digital doppelganger serving as a virtual human listener in the task study. Then the experimenter completed two scans of the partici-
of learning-by-explaining. In this task, a student first studies a topic pants from both conditions. The first was a full-body scan using
by himself/herself, such as reading a passage on the human circula- an iPad equipped with a specialized structure sensor. A second
tory system, then interacts with the agent by verbally explaining scan was a face-only scan, captured using an Intel webcam with
the topic to the agent. Regardless of appearance, the virtual hu- depth sensors. Both scans were conducted in a lab fitted with wall-
man listener responded to the participants with the same behaviors to-ceiling diffused LED lights. Next, the participants filled out a
(e.g., backchannel behaviors) described in Section 4. We designed a Background Survey and took a Pre-Test about the domain knowl-
virtual human listener with two different appearances: edge, while the RACAS system completed the generation of the 3D
model of the participants. Then the participants read an essay on
• Digital Doppelganger condition: A virtual human listener the human circulatory system, adopted from previous studies on
constructed at the beginning of each experiment session us- expert tutoring [11], on a web browser. The participants were told
ing RACAS, thus sharing the appearance of the participants, that they would later have to teach the material to a virtual student.
was used in this condition. After that, the participants sat in front of a 30-inch computer mon-
• Virtual Human condition: A virtual human with photo- itor with the display of the virtual human listener. Two cameras
realistic appearance, not based on any resemblance to the were fitted on top of the monitor: one recorded the participants’
participant, was used in this condition. To control the real- face, and the other served as input to the virtual human listener.
ism of the virtual human listener used in both experiment The participants were told that the virtual student (i.e., the virtual
conditions, the virtual human listener in this condition was human listener) would represent him/her in a competition against
generated using captures of non-participants obtained with other virtual students in a quiz on the same subject. Participants
RACAS through the same process used in the Digital Dop- then explained what they had learned from the tutorial to the virtual
pelganger condition. During the study, the virtual human human listener. Finally, the participants filled out a Post-Interaction
listener was gender-matched to the participant, e.g., male Survey and took a Post-Test on the human circulatory system. Each
participants interacted with a male virtual human. This is study session was designed to last one hour.
PALE’2018, June 2018, London, UK N. Wang et al.
5.4 Measures and after the study on test scores and self-efficacy are statistically
The Background Survey included measures of demographic infor- significant (improved after compared to before), we did not find
mation, education, Rosenberg Self-Esteem Scale [30], Adolescent any significant differences between experiment conditions on these
Body Image Satisfaction Scale (ABISS) [23], Anxiety scale [29], and two measures. We also did not find any significant difference on the
Self-Efficacy in domain knowledge (designed by the research team). duration of explanation between the two experiment conditions.
The Self-Efficacy scale included items such as “I am confident that Details of the analysis on hypothesis testing are discussed in [41].
I know the path blood takes through the circulatory system of the This result suggests that Hypotheses 1, 2, and 3 are not supported.
human body” and “If there is a quiz on human circulatory system, I
expect to do well on the quiz.” 6.2 Alternative Hypotheses
The Post-Interaction Survey included measures of Presence (con- Because the results of the analysis suggest that there is no statisti-
structed using items from [44] and [17]), Avatar Similarity (“To cally significant difference between the two experiment conditions,
what extend do you feel that the virtual avatar resembled you?”), we conducted further analysis of the self-report and behavioral
Desired Avatar Similarity (“If you had to design your own avatar for measures to examine why that was the case.
this task, how similar to your real appearance would you make your
avatar?”), a repeated measure of Self-Efficacy in domain knowledge, 6.2.1 Number of Words in the Explanations. Upon close exam-
and Self-Efficacy in the virtual student (“I think the avatar I just ination of the audio recordings of the participants’ explanations,
taught will do well in the competition.”). we noticed that some participants spoke few words before quickly
In the Pre-Test, participants were asked to describe 10 concepts giving up teaching the virtual human listener, while others spoke
on the human circulatory system (e.g., the aorta and capillaries) at greater speed (and with greater confidence) so that they man-
and the path of blood through the body. The Post-Test included aged to cover all the material they learned in the same duration.
the same questions from the Pre-Test. Scores on these questions Thus the duration of the explanation make not paint the whole
were termed Post-Test-Repeat scores. In addition, the Post-Test picture about the participants’ motivation. To further compare the
included questions adopted from tests used in previous studies motivation to teach the virtual human listener, we analyzed the
on human tutoring [11]. Scores on these questions were termed length of the participants’ explanation in total number of words
Post-Test-NonRepeat scores. spoken, instead of the duration. We transcribed the participants’
In addition to the surveys and tests, the agent’s behaviors in re- explanations using the Temi automated online transcription service
sponse to the participants were logged. The explanations generated [36]. A member of the research team then manually inspected the
by the participants were recorded. transcription and corrected any errors in the automatically gener-
ated transcription. Independent Sample T-Tests show that there is
6 RESULTS no significant difference between the two experiment conditions
for the number of words (p = .090, MV H = 504, M DD = 360). This
Data from one participant was excluded due to extremely short
result again suggests that Hypothesis 1 regarding the similarity
participation time (in filling out surveys and tests, reading the essay,
of agent appearance and motivation to learn (and to explain and
and interacting with the agent). As a result, data from 40 partici-
teach) is not supported.
pants (25 female and 15 male, Maдe = 21.5, age range: 19.7-25.7
years) are included in the analysis. The participants came from a 6.2.2 Novelty Effect. When comparing the duration of the par-
variety of majors, ranging from psychology to fine arts, to biology, ticipants’ explanations during the hypothesis testing, we noticed
and many more. One participant had a graduate degree, while all that participants, by and large, interacted with the agent for a very
other participants had some college education. Participants were short period of time (from just under a minute to over 10 minutes).
randomly assigned to one of the conditions in the study. While RACAS is a novel technology. Participants, especially the ones in
a balanced assignment was desired, the result was that 16 partic- the Digital Doppelganger condition who had never seen themselves
ipants were assigned to the Digital Doppelganger condition and transformed into a digital character before, may have directed much
the remaining 24 participants were assigned to the Virtual Human of their attention to visually inspecting their own avatar. Such activ-
condition. ity may have distracted the participants from the learning activity,
both recalling and explaining. Behaviorally, these participants may
6.1 Hypothesis Testing experience speech disfluencies [19], such as the frequent use of
We conducted an analysis of variance test with scores on Pre-Test filler words. In our previous studies with a virtual human listener
and Post-Test-Repeat as a repeated measure and with the experi- in a story-telling task, when the virtual human listener’s behavior
ment condition (Virtual Human vs. Digital Doppelganger) as the interrupted the ”flow” of the human participant’s story-telling, such
Between-Subject factor. We also conducted an analysis of variance as non-stop staring or ill-timed back-channel feedback, the human
test with self-efficacy reported before and after the study as the participants spoke with more disfluencies [19, 39]. We annotated
repeated measure and experiment condition as the Between-Subject the explanations generated by the participants, particularly speech
factor. Finally, we conducted an Independent Sample T-Test on the disfluencies, such as “um”, “uh”. We did not find any significant
duration of participants’ explanations between experiment con- difference between the two conditions on both the number of dis-
ditions. Duration of the explanation is used in the analysis as an fluencies (p = .394, MV H = 18.6, M DD = 15.2) and frequency of
indication of the participants’ motivation to teach the virtual human disfluencies (p = .362, MV H = 4.0, M DD = 4.5, per minute). This
listener. Results show that although the within-subject effect before may be due to the fact that participants from both conditions spent
Student Belief and Behavior in Learning by Explaining to a Digital Doppelganger PALE’2018, June 2018, London, UK
an equal amount of mental effort to examine the virtual human had higher confidence that their own avatar would perform well in
listener. Because participants from both conditions were scanned, a quiz competition with other avatars.
even the participants from the non-digital doppelganger condition
may have initially devoted much mental effort to examining the
7 DISCUSSION
visual similarities between the virtual human listener and them-
selves. In this paper, we discussed the design of a pedagogical agent for
the learning-by-explaining paradigm. We applied a novel character
6.2.3 Believed Similarity with the Agent. Because the result on animation technology, RACAS, to create virtual human listeners
the main hypothesis testing indicates that the two experiment that share the physical appearance of a human learner. Evaluation
groups were very similar on the dependent measures, we won- of such virtual human listener showed that the digital doppelganger
dered how the manipulation of the independent variable was per- did not significantly impact student learning of domain knowledge,
ceived by the participants. We compared the perceived resemblance their motivation to teach the agent, or their own self-efficacy [41].
of the virtual human listener to the participants. We expect the We subsequently hypothesized that a novelty effect may be at play,
Avatar Similarity to be much lower in the Virtual Human condition, e.g., participants be distracted by the virtual human listener in dur-
compared to the Digital Doppelganger condition. Independent- ing the initial interactions. Post Hoc analysis of speech disfluency
sample T-Test shows that it is indeed the case (p = .001, MV H = of the participants’ explanations, a possible behavioral indication
1.67, M DD = 3.94). Participants from the Virtual Human condi- of such distraction, shows that there were no significant differences
tion did not perceive the virtual human listener’s appearance to in the disfluencies during the explanations. Because participants
be similar to themselves. However, participants from the Digital from all conditions were scanned by the RACAS system, they might
Doppelganger condition were also somewhat unsure whether the have all expected to see their own avatar as the virtual human lis-
virtual human listener looked like them (rated 3.94 on a 7-point tener and thus devoted equal amount of attention to examining the
Likert scale), even though the virtual human listeners were created virtual human listener during the explanation process.
using photo-realistic captures of these participants. This also led to our second alternative hypothesis that if the
We examined the participants’ free-form answers to a question participants perceive enough physical resemblance between the
regarding their reaction to seeing the virtual representation of them- virtual human listener, they may believe that they were interacting
selves on the Post-Survey. While comments from participants in with their own digital doppelganger. This is the case for several of
the Digital Doppelganger condition are consistent with the experi- our participants who interacted with a non-doppelganger virtual
ment manipulation, comments from participants from the Virtual human. Analysis of the main hypothesis based on this perceived
Human condition reveal that some of the participants thought the experiment manipulation suggests that when the participants ex-
virtual human listener looked like them and was a digital repre- plained to a virtual human listener that they believed looked like
sentation of themselves. It is plausible that those participants saw them, they performed better on the near-transfer post-test. They
enough resemblance between the gender-matched generic virtual also had higher confidence that their perceived digital doppelganger
human’s face (see Fig. 1) and their own, because participants from would perform well in a future quiz competition with other avatars.
all conditions received a face and full-body scan using the RACAS However, both effects only approach statistical significance (p = .65
system. In addition, these participants may have thought that the and .071 respectively). So the results suggest a possible trend that
body scan only captured the shape of one’s body, and not the ap- personalizing a pedagogical agent’s appearance to be similar to the
pearance (e.g., clothing), and saw enough resemblance between student’s physical appearance may play a role in the efficacy of
the gender-matched generic virtual human’s body shape and their pedagogical agents. One participant commented that “It (the digital
own. Regardless, these participants from the Virtual Human con- doppelganger) made me want to teach him the material so that he
dition believed that they saw, explained to, and taught their own could score well on the test. ” Such comments resonate with our
avatar. Based on this finding, we regrouped the participants based initial hypothesis and the results here.
on the condition they believed they were assigned to. Four partici- The RACAS system is a new technology that still requires time
pants from the Virtual Human condition are now part of the Digital to mature. The pedagogical agents created through such a process
Doppelganger condition. are less than perfect. Even very slight glitches in the virtual agent’s
Once again, we conducted the tests for the main hypothesis and appearance, such as misalignment of face and body, or animation,
found no significant main effect of the believed condition on changes such as a slight shift of the face when the eyes open/close, can
of pre- and post- test scores (i.e., scores on the questions repeated greatly distract the learner and interfere with the engagement of
from pre- and post test) and the self-efficacy. We saw marginally the learning task. Comments by the participants indicate that they
significant effects on the Post-Test-NonRepeat scores (i.e., scores certainly took notice of such imperfections. Other studies on dig-
on questions that appeared only on the post-test, p = .065, MV H = ital doppelgangers created using RACAS also found that having
29.75, M DD = 36.15) and self-efficacy in the virtual student (“I an avatar that looks like the participants improves subjective ex-
think the avatar I just taught will do well in the competition”, perience, but made no difference on performance measures (e.g.,
p = .071, MV H = 2.80, M DD = 3.9). This suggests a trend that, running in a virtual maze with mines) [25]. On the other hand,
when the virtual human listener shares enough resemblance with earlier studies on non-animated, hand-craftied, and high-fidility
the participant for them to believe that it is a virtual representation digital doppelgangers have shown great promise on attitude and
of themselves, they performed better on the near transfer Post-test, behavioral change (e.g., [18]). Further investigations on the poten-
compared to those who did not have such belief. Additionally, they tial of the personalization pf the appearance of pedagogical agents
PALE’2018, June 2018, London, UK N. Wang et al.
can be carried out once the computer graphics and animations [23] James E Leone, Elizabeth M Mullin, Suanne S Maurer-Starks, and Michael J
technologies, such as RACAS, mature. Rovito. 2014. The adolescent body image satisfaction scale for males: exploratory
factor analysis and implications for strength and conditioning professionals. The
Journal of Strength & Conditioning Research 28, 9 (2014), 2657–2668.
[24] James C Lester and Brian A Stone. 1997. Increasing believability in animated ped-
ACKNOWLEDGMENTS agogical agents. In Proceedings of the first international conference on Autonomous
This project is funded by the U.S. Army Research Laboratory. State- agents. ACM, 16–21.
[25] Gale Lucas, Evan Szablowski, Jonathan Gratch, Andrew Feng, Tiffany Huang, Jill
ments and opinions expressed do not necessarily reflect the position Boberg, and Ari Shapiro. 2016. The effect of operating a virtual doppleganger in
or the policy of the United States Government, and no official en- a 3D simulation. In Proceedings of the 9th International Conference on Motion in
dorsement should be inferred. Games. ACM Press, Burlingame, CA, 167–174.
[26] Mixamo. 2013. Mixamo auto-rigger. http://www.mixamo.com/c/auto-rigger
[27] Natalie K Person. 2003. AutoTutor improves deep learning of computer literacy:
REFERENCES Is it the dialog or the talking head? Artificial intelligence in education: Shaping
the future of learning through intelligent technologies 97 (2003), 47.
[1] Dragomir Anguelov, Praveen Srinivasan, Daphne Koller, Sebastian Thrun, Jim [28] Peter Pirolli and Margaret Recker. 1994. Learning strategies and transfer in the
Rodgers, and James Davis. 2005. SCAPE: shape completion and animation of domain of programming. Cognition and instruction 12, 3 (1994), 235–275.
people. In ACM transactions on graphics, Vol. 24. 408–416. [29] International Personality Item Pool. 2015. The Items in the Preliminary IPIP Scales
[2] Laura Aymerich-Franch, René F Kizilcec, and Jeremy N Bailenson. 2014. The Measuring Constructs Similar to Those Included in Lee and Ashton’s HEXACO
relationship between virtual self similarity and social anxiety. Frontiers in human Personality Inventory. http://ipip.ori.org/newHEXACO_PI_key.htm#Anxiety
neuroscience 8 (2014), 944. [30] Morris Rosenberg. 2015. Society and the adolescent self-image. Princeton univer-
[3] Jeremy N Bailenson. 2012. Doppelgangers-a new form of self? The Psychologist sity press.
25, 1 (2012), 36–39. [31] Noah L Schroeder, Olusola O Adesope, and Rachel Barouch Gilbert. 2013. How
[4] Ilya Baran and Jovan Popović. 2007. Automatic rigging and animation of 3d effective are pedagogical agents for learning? A meta-analytic review. Journal of
characters. ACM Transactions on Graphics (TOG) 26, 3 (2007), 72. Educational Computing Research 49, 1 (2013), 1–39.
[5] Amy L Baylor and Yanghee Kim. 2004. Pedagogical agent design: The impact of [32] Ari Shapiro. 2011. Building a character animation system. Motion in Games
agent realism, gender, ethnicity, and instructional role. In Proceedings of the 7th (2011), 98–109.
International Conference on Intelligent Tutoring Systems. Springer, 592–603. [33] Ari Shapiro, Andrew Feng, Ruizhe Wang, Hao Li, Mark Bolas, Gerard Medioni,
[6] Gaurav Bharaj, Thorsten Thormählen, Hans-Peter Seidel, and Christian Theobalt. and Evan Suma. 2014. Rapid avatar capture and simulation using commodity
2012. Automatically rigging multi-component characters. In Computer Graphics depth sensors. Computer Animation and Virtual Worlds 25, 3-4 (2014), 201–211.
Forum, Vol. 31. Wiley Online Library, 755–764. [34] Jonathan Starck and Adrian Hilton. 2007. Surface capture for performance-based
[7] Gautam Biswas, Hogyeong Jeong, John S Kinnebrew, Brian Sulcer, and ROD animation. IEEE computer graphics and applications 27, 3 (2007).
ROSCOE. 2010. Measuring self-regulated learning skills through social inter- [35] Robert W Sumner, Johannes Schmid, and Mark Pauly. 2007. Embedded deforma-
actions in a teachable agent environment. Research and Practice in Technology tion for shape manipulation. In ACM Transactions on Graphics, Vol. 26. 80.
Enhanced Learning 5, 02 (2010), 123–152. [36] Temi. [n. d.]. Audio to Text Transcription Service: Temi.
[8] Dan Casas, Marco Volino, John Collomosse, and Adrian Hilton. 2014. 4D video [37] Jing Tong, Jin Zhou, Ligang Liu, Zhigeng Pan, and Hao Yan. 2012. Scanning 3d
textures for interactive character appearance. In Computer Graphics Forum, Vol. 33. full human bodies using kinects. IEEE transactions on visualization and computer
Wiley Online Library, 371–380. graphics 18, 4 (2012), 643–650.
[9] Justine Cassell. 2004. Towards a model of technology and literacy development: [38] George Veletsianos. 2010. Contextually relevant pedagogical agents: Visual
Story listening systems. Journal of Applied Developmental Psychology 25, 1 (2004), appearance, stereotypes, and first impressions and their impact on learning.
75–105. Computers & Education 55, 2 (2010), 576–585.
[10] Michelene TH Chi and Miriam Bassok. 1989. Learning from examples via self- [39] Ning Wang and Jonathan Gratch. 2010. Don’t just stare at me!. In Proceedings of
explanations. Knowing, learning, and instruction: Essays in honor of Robert Glaser the SIGCHI Conference on Human Factors in Computing Systems. ACM, 1241–1250.
(1989), 251–282. [40] Ning Wang, W Lewis Johnson, Richard E Mayer, Paola Rizzo, Erin Shaw, and
[11] Michelene TH Chi, Stephanie A Siler, Heisawn Jeong, Takashi Yamauchi, and Heather Collins. 2008. The politeness effect: Pedagogical agents and learning
Robert G Hausmann. 2001. Learning from human tutoring. Cognitive Science 25, outcomes. International Journal of Human-Computer Studies 66, 2 (2008), 98–112.
4 (2001), 471–533. [41] Ning Wang, Ari Shapiro, Andrew Feng, Cindy Zhuang, Chirag Merchant, David
[12] National Research Council et al. 2000. How people learn: Brain, mind, experience, Schwartz, and Stephen L Goldberg. 2018. Learning by Explaining to a Digital
and school: Expanded edition. National Academies Press. Doppelganger. In Proceedings of the 14th International Conference on Intelligent
[13] Scotty D Craig, Barry Gholson, and David M Driscoll. 2002. Animated pedagogical Tutoring Systems. Springer.
agents in multimedia educational environments: Effects of agent properties, [42] Ruizhe Wang, Jongmoo Choi, and Gerard Medioni. 2012. Accurate full body
picture features and redundancy. Journal of educational psychology 94, 2 (2002), scanning from a single fixed 3d camera. In 3D Imaging, Modeling, Processing,
428. Visualization and Transmission. IEEE, 432–439.
[14] Steffi Domagk. 2010. Do pedagogical agents facilitate learner motivation and [43] Noreen M Webb. 1989. Peer interaction and learning in small groups. International
learning outcomes? Journal of media Psychology (2010). journal of Educational research 13, 1 (1989), 21–39.
[15] Andrew Feng, Dan Casas, and Ari Shapiro. 2015. Avatar reshaping and automatic [44] Bob G Witmer, Christian J Jerome, and Michael J Singer. 2005. The factor structure
rigging using a deformable model. In Motion in Games. ACM, 57–64. of the presence questionnaire. Presence: Teleoperators and Virtual Environments
[16] Samantha Finkelstein, Evelyn Yarzebinski, Callie Vaughn, Amy Ogan, and Jus- 14, 3 (2005), 298–312.
tine Cassell. 2013. The effects of culturally congruent educational technologies [45] Ming Zeng, Jiaxiang Zheng, Xuan Cheng, and Xinguo Liu. 2013. Templateless
on student achievement. In Proceedings of the 16th International Conference on quasi-rigid shape modeling with implicit loop-closure. In Computer Vision and
Artificial Intelligence in Education. Springer, 493–502. Pattern Recognition. IEEE, 145–152.
[17] Jesse Fox and Jeremy N Bailenson. 2009. Virtual self-modeling: The effects
of vicarious reinforcement and identification on exercise behaviors. Media
Psychology 12, 1 (2009), 1–25.
[18] Jesse Fox and Jeremy N Bailenson. 2010. The use of doppelgängers to promote
health behavior change. CyberTherapy & Rehabilitation 3, 2 (2010), 16–17.
[19] Jonathan Gratch, Ning Wang, Anna Okhmatovskaia, Francois Lamothe, Mathieu
Morales, Rick J van der Werf, and Louis-Philippe Morency. 2007. Can virtual
humans be more engaging than real ones?. In Proceedings of the International
Conference on Human-Computer Interaction. Springer, 286–297.
[20] Jeffrey Thomas Grant Holmes. 2003. Learning by explaining: the effects of software
agents as learning partners. Ph.D. Dissertation. University of California, Berkeley.
[21] W Lewis Johnson, Jeff W Rickel, James C Lester, et al. 2000. Animated peda-
gogical agents: Face-to-face interaction in interactive learning environments.
International Journal of Artificial intelligence in education 11, 1 (2000), 47–78.
[22] Krittaya Leelawong and Gautam Biswas. 2008. Designing learning by teaching
agents: The Betty’s Brain system. International Journal of Artificial Intelligence in
Education 18, 3 (2008), 181–208.