Post - School of Computer Science and Statistics

Do I trust you, abstract creature?: A study on personality perception of abstract
virtual faces.
Ylva Ferstl∗
Elena Kokkinara†
Graphics, Vision and Visualisation Group
Trinity College Dublin
Rachel McDonnell ‡
Figure 1: Examples of the abstract faces used as stimuli, including examples of all three different head shapes (narrow, round, wide), eyes
shapes (rectangle, almond, round) and eye sizes (small, medium, large) for each of the two characters (black, white).
Abstract
1
Introduction
Video games and animated movies often make use of abstractlooking characters that portray complex personalities or, in video
games, they may interact with the player to influence the gameplay
and experience. However, perceptual studies about the visual design of these characters have been limited and have not provided
any guidelines as to how an abstract character can be used adequately to believably portray a certain personality. Previous research on the perception of human faces has shown that there are
some rules for how certain features can influence perceptions of
specific personality traits across different faces. We sought to investigate if and how such features could be used for more abstractand less human-looking characters, in order to provide some guidelines for artists wanting to create a character intended to portray
certain personality traits.
Studies in the field of social psychology have shown evidence that
the dimensions of human facial features can directly impact the perception of personality of that human. Traits such as aggressiveness,
trustworthiness and dominance have been directly correlated with
facial features. If the same correlations were true for virtual faces,
this could be a valuable design guideline to direct the creation of
characters with intended personalities. In particular, this is relevant
for extremely abstract characters that have minimal facial features
(often seen in video games and movies), and rely heavily on these
features for portraying personality. We conducted an exploratory
study in order to retrieve insights about the way certain facial features affect the perceived personality, as well as affinity of very
abstract virtual faces. We specifically tested the effect of different
head shapes, eye shapes and eye sizes. Interestingly, our findings
show that the same rules for real human faces do not apply to the
perception of abstract faces, and in some cases are the complete reverse. These results provide us with a better understanding of the
perception of abstract virtual faces, and a starting point for the creation of guidelines for how to portray personality using minimal
facial cues.
A large number of studies have shown that perception of certain
personality traits are strongly influenced by certain human facial
features. Our brains have evolved to process these features quickly
and to draw immediate conclusions from them, such as whether the
observed person is likely a threat or is a superior that requires subordinate behavior [Sell et al. 2009]. Even more so, we can recognize
these traits not only in human counterparts, but also in non-human
primates [Kramer et al. 2011; Kramer and Ward 2012], suggesting
that this is a mechanism deeply rooted in evolution that happens
automatically on the subconscious level.
Keywords: face perception, character appeal, personality, physiognomy
We were interested in the question whether these findings could be
applied for the effective design of abstract virtual faces. This could
provide some guidelines to help artists during creation of abstract
characters, especially when aiming to create a certain personality,
or when trying to affect the player’s choices and game experience.
This can be of interest for example when presenting the player with
a moral dilemma, as frequently used in the gameplay of the Fallout
or the Witcher series, where a player’s perception of e.g. the trustworthiness and likeability of an advising character can influence the
decision process.
Concepts: •Computing methodologies → Perception;
∗ e-mail:[email protected]
† e-mail:[email protected]
‡ email:[email protected]
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are not
made or distributed for profit or commercial advantage and that copies bear
this notice and the full citation on the first page. Copyrights for components
of this work owned by others than ACM must be honored. Abstracting with
credit is permitted. To copy otherwise, or republish, to post on servers or to
redistribute to lists, requires prior specific permission and/or a fee. Request
c 2016 ACM.
permissions from [email protected]. SAP ’16, July 22-23, 2016, Anaheim, CA, USA
ISBN: 978-1-4503-4383-1/16/07$15.00
DOI: http://dx.doi.org/10.1145/2931002.2931014
We created an exploratory online survey to assess people’s perception of personality traits of abstract characters. We specifically
aimed to test the effect of the facial width-to-height ratio (FWHR)
and eye size, as previous studies on the perception of human faces
have found effects of these on personality perception. Additionally,
we manipulated the eye shape, an exploratory choice not linked to
39
previous research due to its restrictions to realistic human features.
We believe that eye shape will be an important factor to consider
for abstract characters, especially when other facial cues are limited, because of the important role of eyes in our attention to faces.
ity perception; we consider our characters to have a neutral gender
and omit any investigation of gender differences, however, we never
formally tested if participants considered them to be gender neutral.
We manipulated these three factors, FWHR, eye size and eye shape,
for two different abstract virtual characters. Our results show that
previous findings about the impact of certain human facial characteristics on personality perception may not apply to abstract virtual
characters and may even be reversed in some cases. We discuss the
implications for the design of video game and movie characters.
3
2
We then created three different eye shapes for each character; we
chose to use an almond shape, inspired by the human eye shape, a
circular shape, as it is often used for non-human characters, and a
rectangular shape, which we thought would look even more abstract
and robot-like. Each of these three shapes were presented in three
different sizes. Combining all variations of eyes and head shapes,
we had a stimulus set of 54 test images (see the entire dataset in
Supplemental material).
Stimuli Creation
Two abstract virtual characters were retrieved from the game designed for the EU project POPULATE 1 . The two original models
were manipulated with Autodesk 3ds Max to create a model with a
wide-shaped, a round-shaped and a narrow-shaped head for each of
the two characters (see examples in Figure 1).
Background
Studies have shown that a number of personal traits can be quickly
detected from facial features, amongst these traits are personality
[Kramer and Ward 2010; Little and Perrett 2007], sociosexuality
[Boothroyd et al. 2008], trustworthiness [Stirrat and Perrett 2010]
and aggression [Carré et al. 2009].
Facial width-to-height ratio (FWHR), the width of the face divided
by the height of the upper face, has been hypothesised to be a cue
of threat and dominance in the human face [Carré et al. 2009]. On
a large scale meta-analysis study, Geniole et al. (2015) concluded
that individuals with larger FWHR were judged by observers as
more threatening, more dominant and less attractive, compared to
those with smaller FWHRs [Geniole et al. 2015]. This was especially the case for male faces. Another characteristic that seems to
affect personality judgments on human faces is eye size. This has
been linked to perception of dominance [Keating 1985] and honesty [Zebrowitz et al. 1996]. Specifically, people tend to perceive
faces with bigger eyes as more honest and less dominant.
4
Experiment
We created an online questionnaire where each stimulus was presented once in random order, along with a 5-item questionnaire
about the personality and affinity perception for each character. Participants were asked to rate perceived dominance, trustworthiness,
aggressiveness, appeal and eeriness of the presented character on a
7-point Likert scale (see Supplemental material).
Before starting the experiment, participants completed a set of five
training items that were modelled after the test items with some
additional explanations as to how to answer the presented questions.
A study on the effect of these features on virtual human faces suggested that these traits might not apply to virtual characters [Wang
et al. 2013]. In particular, wider virtual faces were not judged as
less trustworthy, and in contrast to results from studies on real humans, wider faces were perceived as less aggressive compared to
narrow faces. Furthermore, no link between eye size and ratings of
dominance was found in this study, while big eyes were linked to
increased honesty ratings only for female avatars.
The questionnaire was distributed via social media. We collected
data from 32 participants (11 females, ages 21-34 years, M = 26.4,
SD = 3.97), all of whom gave informed consent regarding their participation. None of the participants were familiar with the used
characters.
5
It is still unknown whether these factors have a similar effect on
perceptions of personality of extremely abstract-looking characters
that have minimal facial features. So far, studies have shown that
for virtual characters, different render styles affect judgments about
personality; for example, a cartoon style was rated as having a more
agreeable personality than an ill-looking realistic rendering style
[Zibrek and McDonnell 2014], while perception of likeability and
intelligence were also judged differently in realistic-looking versus cartoon characters [Hyde et al. 2013]. Additionally, stylization
parameters such as shape and material of virtual faces seem to affect perceptions of virtual characters [Zell et al. 2015]. However,
it is still unknown whether facial features that have been found to
influence perception of human faces could similarly influence the
perception of abstract characters. Conceivably, the perception of
a virtual character that has only a pair of round eyes on a simple
spherical head might rely heavily on the configuration of these limited cues.
Results
We examined the effect of all three factors (head shape, eye shape
and eye size) on the ratings given for each of the 5 questions
(dominance, trustworthiness, aggressiveness, appeal, eeriness). We
treated the questionnaire scores as ordinal data and analyzed the
data by fitting a cumulative link model for the three factors, using clm from the R ordinal package [Christensen 2015]. Diverging
stacked bar charts of questionnaire results for each factor can be
seen in Figure 2.
Perception of aggressiveness was significantly influenced by head
shape (χ2 (2) = 62.81, p < .0001) as well as eye shape (χ2 (2) =
62.38, p < .0001), but not eye size. The narrow head shape was
perceived as more aggressive than both the round and the wide head
(both p < .0001), with no significant difference between the round
and wide head. For the eye shape, the almond shape got the highest aggressiveness rating (p < .0001 compared to both other eye
shapes), and rectangular eyes were perceived more aggressive than
round eyes (p < .05).
In our study, we explore how facial features of abstract characters
affect perception of aggressiveness, dominance, trustworthiness,
appeal and eeriness. In addition to the two factors discussed above,
FWHR and eye size, we chose to manipulate eye shape; as previous
studies have focused on human avatars, with the normal human eye
shape, the potential of different eye shapes to influence personality
perception has not been researched yet. Previous studies for human
faces have included gender as an influencing factor for personal-
For the personality trait of dominance, there was a significant effect
of head shape (χ2 (2) = 84.77, p < .0001) and eye shape (χ2 (2) =
60.48, p < .0001), but no effect of eye size. Our results show that
the narrow head shape was perceived as more dominant than both
1 http://www.populateproject.eu/
40
χ2 (2) = 25.15, p < .0001). The round head shape was rated as
most appealing (p < .05 compared to narrow shape and p < .001
compared to wide shape). For the eyes, the round shape was rated as
most appealing (p < .0.01 compared to both rectangle and almond
eyes), as well as the medium eye size (p < .05 compared to small
and p < .0001 compared to large), with small eyes rated as more
appealing than large eyes (p < .05).
Eeriness was also significantly influenced by all three factors (head
shape: χ2 (2) = 37.95, p < .0001; eye shape: χ2 (2) = 10.12, p <
.01; eye size: χ2 (2) = 22.93, p < .0001). The narrow head
shape was perceived as most eerie (p < .0001 compared to both
other shapes). Almond shaped eyes were perceived as significantly
more eerie than both round and rectangle eyes (both p < 0.05).
Large eyes were rated as significantly more eerie than medium
(p < .0001) and small eyes (p < .05).
We used the likelihood ratio test to assess whether the interaction
between the three factors are supported by the data for each question. The likelihood ratio statistic was small in all cases, meaning
that interactions are not supported by the data.
Furthermore, we tested whether the type of character was a significant factor influencing the personality and affinity ratings by
adding it as a factor on the cumulative link model. We found
that the character had a significant main effect on three of the ratings (aggressiveness (χ2 (1) = 30.31, p < .0001), dominance
(χ2 (1) = 93.33, p < .0001) and trustworthiness (χ2 (1) =
16.16, p < .0001)). Main effects of the other factors remained
the same when character was an additional factor. As previously,
the likelihood ratio for the interaction term was small in all cases.
6
Discussion
Previous studies for human faces have shown that some facial features predict the perception of personality traits for the observed
face. We wanted to investigate whether the rules found for these associations also hold true for the perception of abstract virtual characters. Our results stand in contrast to some of the findings of personality perception for human faces. While a larger FWHR has
been consistently linked with increased ratings of dominance and
aggressiveness [Geniole et al. 2015], and may be linked with decreased ratings of trustworthiness [Stirrat and Perrett 2010; Wilson
and Rule 2015], our results indicate that this is different for abstract
virtual characters. Contrary to human face perception, we found
narrow faces to be perceived as most aggressive as well as most
dominant. Narrow faces were also perceived as most eerie, which
could have negatively affected ratings of aggressiveness and dominance. Villains in animated movies also tend to have thin, long
facial features (e.g in many Disney movies, such as Snow White,
101 Dalmatians, Peter Pan, The Lion King, etc.), which could affect FWHR associations for computer-generated characters.
Figure 2: Stacked bar charts for the questionnaire scores. Plotted
is the frequency of responses for each of the 7 rating scores, for all
head shapes, eye shapes and eye sizes. (The y-axis represents the
frequency of responses.)
the round and the wide head (both p < .0001). For the eye shape,
almond eyes were rated most dominant (p < .0001 compared to
both other eye shapes), and rectangle eyes were perceived more
dominant than round eyes (p < .01).
Another facial feature previous research has linked to personality
perception for human faces is eye size, with larger eyes being associated with increased ratings of trustworthiness [Zebrowitz et al.
1996], as well as decreased ratings of dominance [Keating 1985].
However, our results show that medium-sized eyes, not large eyes,
were linked to the highest ratings of trustworthiness, while there
was no effect on dominance ratings. Medium eyes also showed the
highest ratings for appeal and the lowest for eeriness; large eyes
received the lowest ratings for trustworthiness and appeal, and the
highest ratings for eeriness. Hence a medium size was generally
rated most positively. One reason for this difference to findings
for human faces could be that natural human eye sizes have much
lower variance than is possible for the creation of virtual characters. Therefore what is considered large for human eyes is still
Perceived trustworthiness was significantly influenced by all three
factors, head shape (χ2 (2) = 19.09, p < .0001), eye shape
(χ2 (2) = 46.7, p < .0001) and eye size (χ2 (2) = 20.6, p <
.0001). The narrow head shape was perceived as significantly
less trustworthy than both the round (p < .0001) and the wide
head (p < .05). Almond shaped eyes were perceived as significantly less trustworthy than both rectangle and round eyes (both
p < .0001). Round eyes were rated more trustworthy than rectangle eyes (p < .05). The medium eye size was rated as most trustworthy (p < .01 compared to small eyes and p < .0001 compared
to large eyes).
Appeal was dependent on all three factors (head shape: χ2 (2) =
14.33, p < .001; eye shape: χ2 (2) = 9.22, p < .01; eye size:
41
Acknowledgements
somewhat close to an average eye size; if eyes were enlarged artificially on photo-realistic stimuli, this might then lead to similar
pattern as we observed here. Furthermore, the decreased ratings
of dominance for human faces with larger eyes could be caused
by these being a youthful feature [Keating 1985]; previous studies
have shown that baby-faced adults were judged as less dominant,
but more honest, warm and kind [Berry and McArthur 1985; Zebrowitz 1997; Todorov et al. 2008]. In an abstract character, large
eyes may not elicit as strong an impression of immaturity due to
lack of familarity. This might be one reason why we did not find an
effect of eye size on perceived dominance.
This research was funded by the Horizon 2020 EU POPULATE
(644655) project, funded under the research and innovation programme, and sponsored by Science Foundation Ireland as part of
the Cartoon Motion project.
References
B ERRY, D. S., AND M C A RTHUR , L. Z. 1985. Some components
and consequences of a babyface. Journal of personality and social psychology 48, 2, 312.
Next to FWHR and eye size, the third factor we manipulated was
the characters’ eye shape. Our results indicate that the most humanlike shape, the almond shape, was perceived as most aggressive,
most dominant, least trustworthy and most eerie. The reason for
this negative rating for almond eyes might be that they were perceived as looking like the character was squinting its eyes due to a
lack of other facial cues that would indicate this in a human face,
such as a wrinkling of the skin around the eyes. Squinted eyes may
make a face look more angry or mistrusting. The presentation of
this eye shape in-between stimuli with other eye shapes might have
increased this effect. The classic round eye shape was perceived as
the least aggressive and the least dominant. Though this might intuitively be know by artists, to our knowledge this has not been tested
before for this context. Possibilities to draw parallels to previous
studies is limited due to the lack of variety in human eye shapes;
e.g. findings such as the “just below the eyes” rule [Peterson and
Eckstein 2012] may not apply for low-reaching rectangular eyes.
B OOTHROYD , L. G., J ONES , B. C., B URT, D. M., D E B RUINE ,
L. M., AND P ERRETT, D. I. 2008. Facial correlates of sociosexuality. Evolution and Human Behavior 29, 3, 211–218.
C ARR É , J. M., M C C ORMICK , C. M., AND M ONDLOCH , C. J.
2009. Facial structure is a reliable cue of aggressive behavior.
Psychological Science 20, 10, 1194–1198.
C HRISTENSEN , R. H. B., 2015. ordinal—regression models for
ordinal data. R package version 2015.6-28. http://www.cran.rproject.org/package=ordinal/.
G ENIOLE , S. N., D ENSON , T. F., D IXSON , B. J., C ARR É , J. M.,
AND M C C ORMICK , C. M. 2015. Evidence from meta-analyses
of the facial width-to-height ratio as an evolved cue of threat.
PloS one 10, 7, e0132726.
Other effects were less clear and seem to be dependent on the specific character in use. For future studies, a larger pool of different
characters should be considered in order to rule out effects being
dependent on a specific character. It may also be preferential to
use more plain-looking characters; part of the differences between
the perceptions of our characters may be due to their different accessories or their color. Furthermore, while the characters in our
study only had eyes as facial features, previous studies have also explored how the nose and ears may influence personality judgments
[Wang et al. 2013; McArthur and Apatow 1984]. In future studies,
more facial features, such as ears, a nose and mouth, could be used
for different abstract characters. In addition, in this study, we only
used explicit, subjective measures of personality perceptions; future studies could for instance employ decisions in moral dilemmas
or trust games [Stirrat and Perrett 2010] as an objective measure
of participants’ perceptions of characters. Also, we used static images of the characters, but future research might consider the effect
of presenting animated characters instead; animations have been
shown to influence subjects’s perception of virtual characters [Hyde
et al. 2016; Hyde et al. 2015]. Lastly, we did not collect any information about participants’ ethnic background; future studies could
look at possible cultural effects on facial feature perception.
H YDE , J., C ARTER , E. J., K IESLER , S., AND H ODGINS , J. K.
2013. Perceptual effects of damped and exaggerated facial motion in animated characters. In Automatic Face and Gesture
Recognition (FG), 2013 10th IEEE International Conference
and Workshops on, IEEE, 1–6.
H YDE , J., C ARTER , E. J., K IESLER , S., AND H ODGINS , J. K.
2015. Using an interactive avatar’s facial expressiveness to increase persuasiveness and socialness. In Proceedings of the 33rd
Annual ACM Conference on Human Factors in Computing Systems, ACM, 1719–1728.
H YDE , J., C ARTER , E. J., K IESLER , S., AND H ODGINS , J. K.
2016. Evaluating animated characters: Facial motion magnitude
influences personality perceptions. ACM Transactions on Applied Perception (TAP) 13, 2, 8.
K EATING , C. F. 1985. Gender and the physiognomy of dominance
and attractiveness. Social psychology quarterly, 61–70.
K RAMER , R. S., AND WARD , R. 2010. Internal facial features
are signals of personality and health. The Quarterly Journal of
Experimental Psychology 63, 11, 2273–2287.
Together, our results indicate that different rules should be considered when creating abstract virtual characters versus human characters. While wide faces may portray an aggressive and dominant
personality for human faces, this may better be achieved with narrow faces for abstract characters. Furthermore, the human almond
eye shape, when used for abstract characters, may be perceived as
aggressive, dominant, untrustworthy and eerie. In our case, this
could however be due to a lack of other facial cues around the eyes
that could indicate whether the character is squinting, such as wrinkled skin around the eyes. Additionally, while “bigger is better”
may have applied to human eyes so far, we find that the medium
eye size was rated most positively, whereas big eyes can be used to
induce perceptions of eeriness for the characters. These results can
be used as guidelines for the creation of abstract characters that are
intended to portray certain personality features.
K RAMER , R. S., AND WARD , R. 2012. Cues to personality and
health in the facial appearance of chimpanzees (pan troglodytes).
Evolutionary Psychology 10, 2, 147470491201000210.
K RAMER , R. S., K ING , J. E., AND WARD , R. 2011. Identifying
personality from the static, nonexpressive face in humans and
chimpanzees: evidence of a shared system for signaling personality. Evolution and Human Behavior 32, 3, 179–185.
L ITTLE , A. C., AND P ERRETT, D. I. 2007. Using composite images to assess accuracy in personality attribution to faces. British
Journal of Psychology 98, 1, 111–126.
M C A RTHUR , L. Z., AND A PATOW, K. 1984. Impressions of babyfaced adults. Social Cognition 2, 4, 315.
42
P ETERSON , M. F., AND E CKSTEIN , M. P. 2012. Looking just
below the eyes is optimal across face recognition tasks. Proceedings of the National Academy of Sciences 109, 48, E3314–
E3323.
S ELL , A., C OSMIDES , L., T OOBY, J., S ZNYCER , D., VON RUE DEN , C., AND G URVEN , M. 2009. Human adaptations for the
visual assessment of strength and fighting ability from the body
and face. Proceedings of the Royal Society of London B: Biological Sciences 276, 1656, 575–584.
S TIRRAT, M., AND P ERRETT, D. I. 2010. Valid facial cues to cooperation and trust male facial width and trustworthiness. Psychological science 21, 3, 349–354.
T ODOROV, A., S AID , C. P., E NGELL , A. D., AND O OSTERHOF,
N. N. 2008. Understanding evaluation of faces on social dimensions. Trends in cognitive sciences 12, 12, 455–460.
WANG , Y., G EIGEL , J., AND H ERBERT, A. 2013. Reading personality: Avatar vs. human faces. In Affective Computing and
Intelligent Interaction (ACII), 2013 Humaine Association Conference on, IEEE, 479–484.
W ILSON , J. P., AND RULE , N. O. 2015. Facial trustworthiness
predicts extreme criminal-sentencing outcomes. Psychological
science 26, 8, 1325–1331.
Z EBROWITZ , L. A., VOINESCU , L., AND C OLLINS , M. A. 1996.
”wide-eyed” and ”crooked-faced”: Determinants of perceived
and real honesty across the life span. Personality and social psychology bulletin 22, 12, 1258–1269.
Z EBROWITZ , L. A. 1997. Reading faces: Window to the soul?
Westview Press.
Z ELL , E., A LIAGA , C., JARABO , A., Z IBREK , K., G UTIERREZ ,
D., M C D ONNELL , R., AND B OTSCH , M. 2015. To stylize or
not to stylize?: the effect of shape and material stylization on the
perception of computer-generated faces. ACM Transactions on
Graphics (TOG) 34, 6, 184.
Z IBREK , K., AND M C D ONNELL , R. 2014. Does render style affect perception of personality in virtual humans? In Proceedings
of the ACM Symposium on Applied Perception, ACM, 111–115.
43