Observing the Uncanny Valley: Gender Differences in Perceptions of
Avatar Uncanniness
Jacqueline D. Bailey
a
, Karen L. Blackmore
b
and Robert King
College of Engineering, Science and Environment, University of Newcastle, Ring Road, Callaghan, NSW, Australia
Keywords: Uncanny Valley, Uncanniness, Avatar, Human Computer Interaction, Gender(Sex).
Abstract: The creation of avatars is a two-sided coin; on one side we see developers creating avatars with the skills,
time, and resources available to them. However, these resources (or lack thereof) may lead to avatars falling
into the uncanny valley. On the other side are the end-users who engage with the avatar, who ultimately are
the focus for these designers and developers. However, many factors can influence the perception of any
avatar created beyond the level of realism, including the physical appearance of the avatar or something more
fundamental like its gender(sex). Currently, there is a gap in understanding of the influence of gender(sex) in
avatar uncanniness perceptions, and this is mostly missing in design decisions for avatar systems. Bridging
this gap has been a source of research focus spanning the development of new technologies for avatar
development to measuring end-user perceptions of those avatars. Here we add to this discussion through an
experiment involving a set of avatars presented to participants (n = 2065) who were asked to rank them from
least to most uncanny based on their perceptions. This representative set of avatars were sourced from publicly
available methods and have different levels of realism. Our findings indicate that perceptions of avatar
uncanniness based on gender(sex) affects the overall perception of the avatar.
1 INTRODUCTION
‘Avatar’ is a common term that often refers to virtual
humans in virtual environments. Originally derived
from the Sanskrit word ‘Avatara’, referring to the
descent of a god down into the mortal world (Adams,
2014), the term also appears in novels and literature
(Stephenson, 1993). Avatars are often associated with
virtual humans in serious gaming and simulation
training scenarios.
Many factors influence the perception avatars,
including aspects of their aesthetic characteristics
such as hair colour, clothing, and gender(sex) (Fox et
al., 2013). This can also include perceived avatar
realism and uncanniness, which particularly affect to
avatar faces. The differentiation between aesthetic
and perceived characteristics may blur, for example,
where a characteristic such as realism may impact on
the look of an avatar and be the result of the tools used
to create an avatar. This can also be influenced by
how the avatar is presented (fully body or head and
shoulders). These complex interactions present
challenges for designers.
a
https://orcid.org/0000-0002-8550-902X
b
https://orcid.org/0000-0002-9111-0293
The focus of this work is an examination of
gender(sex) in end-user uncanniness perceptions of
avatar faces. Similar to the work of Stumpf et al.
(2020), we use the social construct perspective of
gender, here referred to as gender(sex), whereby
gender identification, gender expression and
performance might not necessarily align with sex.
There are numerous techniques available to create
avatars, however, issues such as available time, funds,
and other resource constraints impact the quality of
the avatar’s appearance and animation. This is true for
avatar face animation where subtle communication
cues may be required. The quality of avatar face
animation is dependent on several factors, first
determined by the fidelity of the 3Dimensional
models and textures used to represent a human face.
Creating highly realistic 3Dimensional models to
represent human faces can be achieved using
techniques such as linear transformations (Tewari et
al., 2017; Thies et al., 2015) and higher-order tensor
generalizations (Brunton et al., 2014). Data-driven
approaches can also be used to create highly realistic
textures of human faces (Saito et al., 2017). With the
Bailey, J., Blackmore, K. and King, R.
Observing the Uncanny Valley: Gender Differences in Perceptions of Avatar Uncanniness.
DOI: 10.5220/0011576000003323
In Proceedings of the 6th International Conference on Computer-Human Interaction Research and Applications (CHIRA 2022), pages 209-216
ISBN: 978-989-758-609-5; ISSN: 2184-3244
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
209
emergence of approaches, the realism of human faces
for avatars may increase, however, this is not yet the
case in many practical applications.
The level of realism of an avatar ideally meets
end-users expectations without the feeling of
uncanniness affecting their experience (MacDorman
et al., 2009). When designing avatars, some explicit
consideration of these issues is valuable, including
the possible interrelationship with gender(sex) and
uncanniness. Various perceptions can be affected by
gender(sex), such as physical attractiveness and
authority (Patel & MacDorman, 2015). Accordingly,
this research seeks to further explore the
interrelationship of observer gender(sex) on the
perception of avatar uncanniness levels. We achieve
this through a large-scale avatar ranking exercise
considering the differences in avatar rankings based
on the gender(sex) of participants. The goal is to
evaluate whether gender(sex) has an impact on the
perception of avatar uncanniness, and to generate a
robust set of ranked avatars.
Ultimately, this information is important to
inform design considerations for avatars, and
particularly for avatars that may be used in serious
games and training applications. Through this
research, we propose that avatar creators adopt a more
purposeful consideration of gender(sex) in design
considerations for avatar-based systems. This may
lead to enhanced affective outcomes for avatars in
serious purposes.
The following section provides a review of
relevant literature followed by the methodology for
our study. Next, the results of the research detail the
overall uncanniness ranks as well as potential impacts
of participant gender(sex) on these ranks. We
conclude with a discussion of the key findings,
together with recommendations for future work.
2 BACKGROUNDS
Uncanniness can apply to an avatar when a human
like avatar fails to meet the expected visual, kinetic,
or behavioural fidelity of a human. This negative
emotional response can be attributed to the Uncanny
Valley theory proposed by Mori in 1970 (Figure 1)
and is demonstrated through a sharp dip in the linear
progression of the perception of human likeness.
Mori suggests that the sense of eeriness is likely a
form of instinct that protects us from proximal, rather
than distal sources of danger, such as corpses or
members of different species. This theory has now
also been applied to human-like avatars that have
been created using computer graphic capabilities with
a
Figure 1: The Uncanny Valley (Mori et al., 2012).
a focus on boundary-pushing realism (Tinwell, 2014).
The theory states that the human-like entities, such as
avatars, that fail to meet expectations of human
behaviour fall into the Uncanny Valley, which may
stem from the consequences of realism levels seen in
avatar faces.
2.1 The Consequences of Uncanniness
Perceptions for Avatar Faces
The potential impact of uncanniness is an important
issue in the perception of avatars that should be
considered carefully. In particular, as the human
visual system excels in detecting falsehoods that may
be found in human-like facial features based on prior
knowledge of human facial features (Seyama &
Nagayama, 2007), avatars are particularly susceptible
to uncanny valley effects. These issues extend to
more than just the visual representation though.
Another key factor of avatar face animation is the
approach used to make the avatar face move and
express emotions in realistic ways. There are several
methods used to create facial animation, including
manual, motion capture, and data-driven approaches.
Traditional manual methods can include a frame-by-
frame approach or rotoscoping of facial movements.
While these manual animation methods can
effectively create facial animation, it is a highly time-
consuming process. Alternatively, techniques such as
motion capture provide a quicker means of capturing
an actor’s facial movement. These movements can be
captured as a live performance through real-time
motion capture or as a set of motions mapped post-
capture (Davison et al., 2001). Little is understood
though of the impacts of these different techniques on
perceptions of uncanniness.
Thus, based on this existing research, we
undertake a ranking exercise with a set of relatively
homogenous avatars to better understand perceptual
variability, and to create a robustly ranked set of
avatars for future research. With these avatars
CHIRA 2022 - 6th International Conference on Computer-Human Interaction Research and Applications
210
quantifiability rated in terms of uncanniness
perceptions, further analysis can be carried out from
additional data collected on this avatar set
. While the
set of avatars used in our study are of similar age and
ethnicity, they are distinctly different with varying
levels of realism as used by Seymour, Riemer
Seymour et al. (2018) and Seyama and Nagayama
(2007). In summary, we investigate the relationship
between a set of avatar faces uncanniness perceptions
and how gender(sex) may affect these perceptions.
2.2 Impact of Avatar Gender(Sex)
One fundamental aspect of avatar creation is the
perceived gender(sex) of the avatar. There has been
some research in this area, with issues such as
existing gender(sex) stereotypes, the proteus effect,
and gender(sex) swapping being explored. Some
gender(sex) based stereotypes were observed by Fox
and Bailenson (2009) who suggested that female
avatars are more likely to appear either as hyper
sexualised as an ornament within games or as a
support role. The level of sexualisation of an avatar
can also affect their perceived abilities (Wang & Yeh,
2013). When considering the Proteus Effect, Yee and
Bailenson (2007) suggest that the appearance of an
avatar will lead users to conform to behavioural
expectations they have associated with a specific
gender(sex). For example, Beltrán (2013) examined
this and suggested that this conformity exists in
simulation and training contexts, finding that using a
male avatar to train professional women can
negatively affect her and her peers' achievements.
This is important to consider, as Beltrán (2013)
argues that most simulation tools show a generic male
avatar during training.
Additionally, gender(sex) swapping and
exploring an end-user's gender identity or identities is
an active area of research. Lehdonvirta et al. (2012)
suggest that male participants are more likely to seek
and receive help when disguised as a female avatar.
Also, Hussain and Griffiths (2008) suggest that there
are many social benefits to gender(sex)-swapping in
online gaming, such as male players engaging as a
female avatar to be more favourably treated by other
male players.
Another aspect of avatar gender(sex) is the ability
for end-users to explore and express their own gender
identity (Baldwin, 2018). This is important as it
allows users to express their gender identities which
may or may not reflect the gender identity they
express in their everyday lives. In video/computer
gaming settings, players may be able to remain
anonymous online while they explore their gender
identity or identities in a safe platform. To examine
the potential influence of gender(sex) in the
perception of avatar uncanniness, we conducted a
mass-scale survey, described next.
3 STUDY METHODOLOGY
Using a set of 10 homogeneous avatars we consider
the impact of participant gender(sex) in the
perception of uncanniness. This data was gathered via
an online Mturk (amazon.com, 2017) study which
took participants 15-20minutes to complete. The
participants were asked to rank the avatars from least
to most uncanny or eerie. This study was approved by
the University of Newcastle’s Human Research
Ethics Committee (H2015-0163).
In total, (n=2065) participants, with a mean age of
34.82 years completed the survey. Of these
participants, 1050 self-identified as female, 1003 as
male, three people self-identified as transgender
people, two people selected ‘other’ for their
gender(sex), and seven preferred not to say.
3.1 Avatar Set
A set of 10 avatar faces (Table 1) was used in this
study to broadly capture the varying levels of realism
obtained through different creation methods. As
shown, they are of different but homogenous faces
with assumed binary sexes (five females and five
males). These avatars are considered representative of
those found in simulation and training contexts, and
are derived from various sources (Alexander et al.,
2017; AppleInc., 2015; Battocchi, 2005; Metrics,
2018; Nao4288, 2013; von der Pahlen et al., 2014).
Table 1: Sample Images of the avatar set.
3.2 Study Procedure
To begin the ranking survey, participants first
completed a set of demographic questions on
gender(sex), age, ethnicity, English language
proficiency, current residential country, and highest
Observing the Uncanny Valley: Gender Differences in Perceptions of Avatar Uncanniness
211
education level achieved. Additional questions asked
participants to nominate whether they identified
themselves as a video/computer gamer, on which
platforms and media they interacted with avatars, and
how many hours a week they spent interacting with
avatars. Participants then ranked the avatar set from
most uncanny to least uncanny. The approach
adopted here follows Lange (2001), who used a
similar rating exercise for their study into realism
perceptions of virtual landscapes. In both cases, a
ranking rather than rating approach is adopted as
these have been shown to have higher reliability
(Mantiuk et al., 2012; Winkler, 2009).
3.3 Statistical Analysis
The avatar uncanniness ranks were first analysed
using Friedman tests. Following a significant
Friedman test, post hoc Wilcoxon signed-rank tests
using a Bonferroni adjustment for multiple tests
(significance level reduced to .0001) were conducted
to examine differences in rankings.
The variable ‘observer gender(sex)’ was
investigated to determine whether the gender(sex) of
a participant was related to the realism and
uncanniness rankings of the avatar faces. To measure
the impact of gender(sex), we used the approach
suggested by (Conover & Iman, 1981) who suggest
that rank transformation procedures allow the use of
parametric methods. This would normally be done
with observed scores being converted to ranks first.
However, in this study, the ranks were provided by
the participants, so the transformation step was not
needed.
The impact of gender(sex)was tested using the
General Linear Model (GLM) as the parametric
procedure. For the effect of observer gender(sex), a
full factorial two-way ANOVA model was fit with
avatar and gender(sex)being the two factors. Follow
up testing with pairwise comparisons were used to
determine significant differences with a Bonferroni
adjustment to control the familywise error rate due to
multiple testing.
4 RESULTS
We first present results of the Friedman tests to
compare the ranks for our avatar set’s uncanniness
perceptions, followed by the impact of participant
gender(sex) on these rankings.
4.1 Uncanniness Ranking Scores
Based on the Freidman test, there was a statistically
significant difference in the perception of the avatars’
uncanniness
χ
2
(9) = 156.254, p < .001. Interestingly,
we see that the means scores for the uncanniness
ranks are all clustered between 4.9-5.8. This suggests
that there is some variation in the uncanniness
perceptions, but it may be very subtle.
The rankings here show that the Mid1 realism
avatars frame the two extremes of these ranks, with
Victor (Mid1 realism male) being rated as the
uncanniest (M= 4.99, SD= 2.505), whereas his female
counterpart Ilana (M= 5.87, SD =2.324) is considered
the least uncanny of the avatars. However, Ira (high
realism male) (M= 5.14, SD= 2.805) was considered
uncannier than his counterpart Emily (M= 5.37, SD=
3.006). Leo (Mid2-Low realism male) was ranked
fourth most uncanny by participants (M= 5.47, SD =
2.739), while Rycroft (real human male) is ranked as
the fifth most uncanny avatar (M = 5.51, SD = 3.114).
Interestingly, Rose (real human female) (M= 5.65,
SD= 3.360) was considered less uncanny than her
counterpart. Despite being considered the least
realistic, the Bailie (M= 5.66, SD= 3.214) avatar was
not considered to be the uncanniest.
A Wilcoxon signed ranked test using a Bonferroni
correction, determined where the differences
occurred for the uncanniness ranks. Although there
were no large differences between the scores for the
avatars, there were some significant differences in the
ranks of the following avatars. When examining the
scores for the high realism male and the real humans,
both comparisons were statistically significant; Ira
and Rycroft (Z = -4.527, p. < 0.001) and Ira and Rose
the real human female (Z = -6.229, p. < 0.001).
However, the scores for Emily (high realism female)
were not statistically significantly different when
compared to real humans, which may suggest that the
gender(sex)of the avatar may influence uncanniness
when compared to real humans. Another pattern
occurs in the comparison of the uncanniness rank for
Victor (Mid1 realism male); it is statistically
significantly different to other avatars, including the
female avatar from the same source. This again may
suggest some gender(sex)-based perceptual
differences.
4.2 The Impact of End-user
Gender(Sex) on the Rank Scores
Following the analysis of differences in the overall
uncanniness rankings, we consider gender(sex)-based
variability in uncanniness ranks. Using a GLM with a
CHIRA 2022 - 6th International Conference on Computer-Human Interaction Research and Applications
212
Bonferroni correction, we note a statistically
significant interaction of our avatar set uncanniness
ranks and the participants’ gender(sex) F(9, 19910) =
10.453, p < .001).
The Posthoc analysis shows some statistically
significant differences between the responses of
female and male participants. Specifically, the female
participants rated Rycroft (real human male) as less
uncanny than the male participants (Rycroft p = .001,
(Female (M= 5.68, SD= .093), Male (M= 5.26, SD=
.088))).This pattern continues with scores for Rose
(real human female) where the female participant
scores are higher than the male participants (Rose p <
.001, (Female (M=5.85, SD= .093), Male (M= 5.33,
SD= .088))). The high realism female is less uncanny
by female participants when compared to the male
participants (Emily p = .001, (Female (M= 5.54, SD=
.093), Male (M= 5.12, SD= .088))).
In contrast, the high realism male avatar (Ira) was
ranked uncannier by male participants (Ira p = .005,
(Female (M= 5.27, SD= .093), Male (M= 4.90, SD=
.088))). The independently sourced avatar (Baillie)
was deemed less uncanny by male participants than
the female participants (Bailie p < .001, (Female (M=
5.47, SD= .093), Male (M= 5.98, SD= .088))).
Finally, Macaw (Mid2-Low male avatar) was
regarded as less uncanny by male participants when
compared to scores from female participants (Macaw
p=.001, (Female (M= 5.30, SD= .093), Male (M=
5.75, SD= .088))). Generally, male participants are
more likely to rank high realism avatars as uncanny.
Further, we note limited variation in the responses
as the scores only vary in a range between 4.5 and 6.5
when compared by participant gender(sex). However,
when examining the distribution of uncanniness
scores for each avatar and grouped by participant
gender(sex), several patterns emerge as shown in
Figure 2.
The initial comparison considers differences in
the distributions of the uncanniness scores for each
avatar for female and male participant responses.
From these histograms (Figure 2), we observe the
Mid1 avatars (Ilana and Victor) and the Mid2-Low
male avatar (Macaw) scores resemble a normal
distribution. In contrast, Rose and Rycroft (real-world
humans) and the Mid2-Low female avatar (Bailie)
scores have a distinctly bi-modal distribution
indicating participants have a divided opinion of these
avatars. The remaining avatars present a somewhat
flat, uniform like distribution of scores.
Using a 2 sample Kolmogorov-Smirnov Test to
compare the shapes of the distributions, and with a
Bonferroni correction for testing multiple pairs, we
compared the female and male participant
Figure 2: Rank scores comparison by avatar and participant
gender(sex).
uncanniness score distributions. We see statistically
significant differences in the distribution of the scores
for some avatars when comparing the female and
male distributions, including Macaw D(1993) =
1.796, p.= .003, Rose D(1993) = 1.785, p.< .001,
Bailie D(1993) = 1.785, p.= .003 and Rycroft
D(1993) = 1.819, p.= 003.
The second key comparison is an examination of
the differences in the distribution of scores when
comparing avatars between female and male
participants. Using a series of Kolmogorov-Smirnov
Tests, with a Bonferroni correction to examine these
differences, reveals some statistically significant
differences. When comparing Leo – Liliwen, the
distribution of the female scores were statistically
different while the male participants scores were not
(Female(D(2090) = 1.947, p.< .001., Male (D(1896)
= 1.355, p.= .051.)). We also note that the female
participant scores when comparing Rose (real human
female) and Emily (high realism female) are
statistically different when to the male participant
scores (Female (D(2090) =3.412, p.< .001.), Male
(D(1896) = 1.447, p.=.030.)) This pattern continues
with Rycroft (real human male) and Rose (real human
female) with female participants’ scores statistically
significant while the male scores are not statistically
different (Female (D(2090) =1.312, p.= .064), Male
(D(1896) = 3.834, p.< .001)). When comparing the
both the Mid2-Low female avatars (Liliwen Bailie),
the female participant scores are statistically
significant, and the male participants’ are not (Female
Observing the Uncanny Valley: Gender Differences in Perceptions of Avatar Uncanniness
213
(D(2090) = 2.166, p.< .001., Male(D(1896) = 1.699,
p.= .051.)).
The pattern identified in the preceding paragraph
is reversed for some of the avatars. For male avatars
(Macaw Ira), the high realism male comparison
shows the male scores being statistically significant
and the females do not (Male(D(1896) = 3.834, p.<
.001), Female (D(2090) = 1.312, p.= .064)). This
continues with one of the Mid2-Low female avatars
(Bailie) to the real human male (Ira) comparison
where the distribution of the male scores is
statistically different and the female participant
scores are not (Male (D(1896) = 2.825, p.< .001),
Female (D(2090) = 1.444, p.= .031)). Again, the
comparison of the Mid2-Low male avatar (Leo) to the
high realism male (Ira) distributions are statistically
significant for male participants but not the female
participants (Male (D(1896) = 3.789, p.< .001),
Female (D(2090) = 1.553, p.= .016)). Lastly, we see
a statistically significant difference between the
distribution of scores for the Mid2-Low realism male
avatar (Leo) to the high realism female avatar (Emily)
for male participants but not for the females (Male
(D(1896) = 3.330, p.< .001), Female (D(2090)
=1.837,p.=.016)). The remaining comparisons were
not statistically significant for either gender(sex).
5 DISCUSSION AND
CONCLUSION
Our investigation into the impact of gender(sex)
considered if a participants’ gender(sex) affects the
perception of an avatar’s uncanniness ratings. Male
participants rated the higher realism avatars as
uncannier than all other avatars except for the Mid1
realism male avatar. This aligns to previous findings
where male participants were more sensitive to
uncanniness in human-like agents (Tinwell, 2014).
Of interest, the top four ranks for female
participants were populated by male avatar faces,
suggesting that these avatars may have triggered
negative responses for participants who identified as
female. Existing research on avatar faces features and
uncanniness suggest that avatars failing to display
empathy or reacting appropriately may lead to
assumptions of psychopathy in an avatar (Tinwell,
2014). As such, the critical dimensions of
uncanniness, such as threat avoidance and alignment
to terror management theory (MacDorman et al.,
2009) may be more enhanced in female participants.
It is also interesting that while there was some
consistency in the avatars ranked as most uncanny,
the same consensus was not present in those ranked
least uncanny, with female and male participants
responding differently. For both genders(sexes), the
avatars ranked as most uncanny were female,
although they were different avatars. These
differences led to further investigation of the
distribution of uncanniness ranks by participant
gender(sex) to reveal further insights.
From examination of the distribution of scores, we
see some interesting differences in the female and
male participant scores for each avatar. The real
humans and one of the Mid2-Low realism female
avatars scores exhibit a bi-modal distribution,
suggesting that the general opinion of these avatars is
divided. Additionally, despite both real humans
achieving high realism ranks, these avatars fall
around the middle of the uncanniness rankings. These
distributions were found to be statistically
significantly different when comparing the scores
when grouped by the participants’ gender(sex). These
differences may suggest that some participants were
convinced that these avatars were computer-
generated images as opposed to real humans, or
perhaps factors other than avatar realism contribute to
perceptions of uncanniness. Further, we see that the
one of the Mid2-Low realism female avatars (Bailie)
was ranked as the least realistic but ranked only 8th
on the uncanny scale, indicating that the avatar was
considered unrealistic but not uncanny. This supports
existing literature suggesting lower realism levels can
lead to avatars being perceived as less uncanny
(Tinwell, 2014).
Our findings highlight the importance of
gender(sex) in avatar design decisions due to the
impact of this variable on the perceptions of avatar
uncanniness. As previously identified, a detailed
consideration of the influence of gender(sex) in avatar
uncanniness perceptions is mostly missing in the
design decisions for avatar systems. Further, it is
evident from the literature that the unavoidable design
choice of gender(sex) for an avatar may have
underlying cues and expectations placed on them
based solely on their perceived gender(sex).
Together, the findings of the research provide key
insights into gender(sex) based perception of avatars.
Despite the significant contributions of this work,
it is not without some limitations. Firstly, there is a
lack of ethnic diversity in our surveys’ sample. The
primary ethnicities that completed this survey were of
White/Caucasian and Asian background, which may
lead to some bias in the interpretation of the data.
Secondly, the results cannot be extended to nonbinary
genders(sexes) due to small sample sizes. However,
this does present an avenue for future work. Lastly,
CHIRA 2022 - 6th International Conference on Computer-Human Interaction Research and Applications
214
while the participants in our study come from a wide
range of ages (18-87 years old (M= 34.82, SD=
11.52)), a restriction put in place by Mturk is that
participants must be at least 18 years old. Therefore,
the results presented in our study may not be
applicable to those under 18.
Another area of limitation relates to the avatars
themselves. There is a lack of diversity in the ten
avatars used in this study. The sample of avatars is
primarily of an Anglo-Saxon appearance with little
subjective difference in the facial features. Lastly, the
current study asks participants to consider the avatars
without the context of how they are used. We note
that perceptions might differ with context as
previously identified (Rosen, 2008). The complexity
of the ranking task necessitated the use of a minimal
avatar set. However, given that avatars, as virtual
representations of humans, could potentially reflect
the full diversity of the human form, it is arguable
how large a set would be required to be
representative. Thus, future work may seek to expand
the avatars evaluated to examine the differences in a
more diverse set.
Despite these limitations, the work presented here
has produced interesting insights into gender(sex)
differences in the perception of avatars and generates
several avenues for future research. First, the work
presented here has focused on perceptual effects of
the gender(sex) of both participants and avatars.
Future analysis may extend this to explore the
differences in the rankings associated with
perceptions of avatar-participant self-similarity and
avatar sex. An area for further analysis considers the
individual attributes of each of the ten avatars through
a gender(sex)-swapped lens to further explore
gender(sex) as a contributor to perceptions of
uncanniness. In summary, the work presented here
provides the basis for extending current
understanding of gender(sex) differences in the
perceptions of avatars.
ACKNOWLEDGEMENTS
This research was supported by an Australian
Government Research Training Program (RTP)
Scholarship. The authors would like to thank Mr. Kim
Colyvas from the Statistical Support Services,
University of Newcastle, for his assistance with the
analysis of data for this manuscript.
REFERENCES
Adams, E. (2014). Fundamentals of game design. Pearson
Education.
Alexander, O., Rogers, M., Lambeth, W., Chiang, J., Ma,
W., Wang, C., & Debevec, P. (2017). The Digital Emily
Project: Achieving a Photoreal Digital Actor. USC
Institute for Creative Technologies.
amazon.com. (2017). amazon mechanical turk.
AppleInc. (2015). FaceShift. In (Version 2014.2.01)
FaceShift AG.
Baldwin, K. (2018). Virtual Avatars: Trans Experiences of
Ideal Selves Through Gaming. Markets, Globalization
& Development Review (MGDR), 3(3).
Battocchi, A., Pianesi, F., Goren-Bar, D. (2005). A first
evaluation study of a database of kinetic facial
expressions (dafex) Proceedings of the 7th international
conference on Multimodal interfaces, Trento, Italy.
Beltrán, M. (2013). The importance of the avatar gender in
training simulators based on virtual reality. Proceedings
of the 18th ACM conference on Innovation and
technology in computer science education,
Brunton, A., Bolkart, T., & Wuhrer, S. (2014). Multilinear
wavelets: A statistical shape space for human faces.
European Conference on Computer Vision,
Conover, W. J., & Iman, R. L. (1981). Rank
transformations as a bridge between parametric and
nonparametric statistics. The American Statistician,
35(3), 124-129.
Davison, A. J., Deutscher, J., & Reid, I. D. (2001).
Markerless motion capture of complex full-body
movement for character animation. In Computer
Animation and Simulation 2001 (pp. 3-14). Springer.
Fox, & Bailenson. (2009). Virtual virgins and vamps: The
effects of exposure to female characters’ sexualized
appearance and gaze in an immersive virtual
environment. Sex Roles, 61(3-4), 147-157.
Fox, J., Bailenson, J. N., & Tricase, L. (2013). The
embodiment of sexualized virtual selves: The Proteus
effect and experiences of self-objectification via
avatars. Comput Human Behav, 29(3), 930-938.
Lehdonvirta, M., Nagashima, Y., Lehdonvirta, V., & Baba,
A. (2012). The stoic male: How avatar gender affects
help-seeking behavior in an online game. Games and
Culture, 7(1), 29-47.
MacDorman, K. F., Green, R. D., Ho, C.-C., & Koch, C. T.
(2009). Too real for comfort? Uncanny responses to
computer generated faces. Comput Human Behav,
25(3), 695-710.
Mantiuk, R. K., Tomaszewska, A., & Mantiuk, R. (2012).
Comparison of four subjective methods for image
quality assessment. Computer Graphics Forum, 31(8),
2478-2491.
Metrics, I. (2018).
Faceware. In (Version 3.1) Faceware
Tech.
Mori, M., MacDorman, K. F., & Kageki, N. (2012). The
uncanny valley [from the field]. IEEE Robotics &
Automation Magazine, 19(2), 98-100.
Observing the Uncanny Valley: Gender Differences in Perceptions of Avatar Uncanniness
215
Nao4288, n. (2013). Female Facial Animation ver 1.0 (No
Sound) [YouTube Video]. https://tinyurl.com/yc6k6
e3k
Patel, H., & MacDorman, K. F. (2015). Sending an avatar
to do a human's job: Compliance with authority persists
despite the uncanny valley. Presence, 24(1), 1-23.
Rosen, K. R. (2008). The history of medical simulation.
Journal of critical care, 23(2), 157-166.
Saito, S., Wei, L., Hu, L., Nagano, K., & Li, H. (2017).
Photorealistic facial texture inference using deep neural
networks. Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition, Honolulu,
Hawaii, United States.
Seyama, J. i., & Nagayama, R. S. (2007). The uncanny
valley: Effect of realism on the impression of artificial
human faces. Presence: Teleoperators and Virtual
Environments, 16(4), 337-351.
Seymour, M., Riemer, K., & Kay, J. (2018). Actors, avatars
and agents: potentials and implications of natural face
technology for the creation of realistic visual presence.
Journal of the Association for Information Systems,
19(10), 4.
Stephenson, N. (1993). Snow Crash. Bantam Books.
Stumpf, S., Peters, A., Bardzell, S., Burnett, M., Busse, D.,
Cauchard, J., & Churchill, E. (2020). Gender-inclusive
HCI research and design: A conceptual review.
Foundations and Trends in Human–Computer
Interaction, 13(1), 1-69.
Tewari, A., Zollhofer, M., Kim, H., Garrido, P., Bernard,
F., Perez, P., & Theobalt, C. (2017). Mofa: Model-
based deep convolutional face autoencoder for
unsupervised monocular reconstruction. Proceedings of
the IEEE International Conference on Computer
Vision, Italy.
Thies, J., Zollhöfer, M., Nießner, M., Valgaerts, L.,
Stamminger, M., & Theobalt, C. (2015). Real-time
expression transfer for facial reenactment. ACM Trans.
Graph., 34(6), 183:181-183:114.
Tinwell, A. (2014). The uncanny valley in games and
animation. AK Peters/CRC Press.
von der Pahlen, J., Jimenez, J., Danvoye, E., Debevec, P.,
Fyffe, G., & Alexander, O. (2014). Digital ira and
beyond: creating real-time photoreal digital actors.
ACM SIGGRAPH 2014 Courses,
Wang, C.-C., & Yeh, W.-J. (2013). Avatars with sex appeal
as pedagogical agents: Attractiveness, trustworthiness,
expertise, and gender differences. Journal of
Educational Computing Research, 48(4), 403-429.
Winkler, S. (2009). On the properties of subjective ratings
in video quality experiments. Quality of Multimedia
Experience, 2009. QoMEx 2009. International
Workshop on Quality of Multimedia Experience, USA.
Yee, N., & Bailenson, J. (2007). The Proteus effect: The
effect of transformed self-representation on behavior.
Human communication research, 33
(3), 271-290.
CHIRA 2022 - 6th International Conference on Computer-Human Interaction Research and Applications
216