![The three analyses of this paper within the framework from [Scherer 2013], visual modified from [Vinciarelli and Mohammadi 2014]](https://www.researchgate.net/profile/Tom-Giraud/publication/280076026/figure/fig1/AS:613994180145156@1523399028366/The-three-analyses-of-this-paper-within-the-framework-from-Scherer-2013-visual_Q320.jpg)
Content uploaded by Tom Giraud
Author content
All content in this area was uploaded by Tom Giraud on Oct 24, 2015
Content may be subject to copyright.
Perception of Emotion and Personality through Full-
Body Movement Qualities: a Sport Coach Case Study
TOM GIRAUD, University Paris South, LIMSI-CNRS (UPR 3251)
FLORIAN FOCONE, University Paris South, LIMSI-CNRS (UPR 3251)
VIRGINIE DEMULIER, University Paris South, LIMSI-CNRS (UPR 3251)
JEAN CLAUDE MARTIN, University Paris South, LIMSI-CNRS (UPR 3251)
BRICE ISABLEU, University Paris South, UR CIAMS (EA 4532)
Virtual sport coaches guide users through their physical activity and provide motivational support. motivation can
rapidly decay if the movements of the virtual coach are too stereotyped. Kinematics patterns generated while performing a
predefined fitness movement can elicit and help to prolong users interaction and interest in training. Human body kinematics
has been shown to convey various social attributes such as gender, identity and acted emotions. To date, no study provides
information regarding how spontaneous emotions and personality traits together are perceived from full-body movements. In
this paper, we study how people make reliable inferences regarding spontaneous emotional dimensions and personality traits of
human coaches from kinematic patterns they produced when performing a fitness sequence. Movements were presented to
participants via a virtual mannequin to isolate the influence of kinematics on perception. Kinematic patterns of biological
movement were analyzed in terms of movement qualities according to the effort-shape [Dell 1977] notation proposed by [Laban
1950]. Three studies were performed to provide an analysis of the process leading to perception
. Thirty-two participants (i.e., observers) were asked to rate the
movements of the virtual mannequin in terms of conveyed emotion dimensions, personality traits (five-factor model of
personality) and perceived movement qualities (effort-shape) from 56 fitness movement sequences. The results showed high
reliability for most of the evaluated dimensions, confirming inter-observer agreement from kinematics at zero acquaintance. A
large expressive halo merging emotional (e.g., perceived intensity) and personality aspects (e.g., extraversion) was found, driven
by perceived kinematic impulsivity and energy. Observer perceptions were partially accurate for emotion dimensions and were
not accurate for personality traits. Together, these results contribute to both the understanding of dimensions of social
perception through movement, but also to the design of expressive virtual sport coaches.
Categories and Subject Descriptors: H.5.2 [Information Interfaces and Presentation]: User Interfaces
Evaluation/methodology; J.4 [Computer applications]: Social and behavioral sciences: Psychology.
General Terms: Experimentation, Human Factors
Additional Key Words and Phrases: Full body movement, kinematics, emotion perception, personality perception, sport coach.
ACM Reference Format:
1. INTRODUCTION
Nonverbal behaviors are essential to communication, providing social meaningfulness to everyday
human-human interactions. Affects, attitudes and personalities are pervasively conveyed through
several modalities, including vocal, facial and bodily expressions [Giles and Le Poire 2006]. As the
domain of affectively aware technologies grows, more knowledge is required to guide the design of
affective interactive systems [Petta et al. 2011]. Virtual sport trainers are one these interactive
technologies. With the lack of daily physical activity being one of the main health risk factors in our
This work is supported by ANR INGREDIBLE project: ANR-12-CORD-001 (http://www.ingredible.fr).
Author's address: Tom Giraud, LIMSI-CNRS, University of Paris South, Building 508, Office 204, B.P. 133, 91403 ORSAY
cedex (France); email: tom.giraud@limsi.fr; Florian Focone, LIMSI-CNRS, University of Paris South, Building 508, Office 204,
B.P. 133, 91403 ORSAY cedex (France); email: florian.focone@limsi.fr; Virginie Demulier, LIMSI-CNRS, University of Paris
South, Building 508, Office 204, B.P. 133, 91403 ORSAY cedex (France); email: demulier@limsi.fr; Jean Claude Martin, LIMSI-
CNRS, University of Paris South, Building 508, Office 204, B.P. 133, 91403 ORSAY cedex (France); email: martin@limsi.fr;
Brice Isableu, UR CIAMS, University of Paris South, Building 335, Office 32, 91405 ORSAY (France); email: brice.isableu@u-
psud.fr.
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 permissions from Permissions@acm.org.
1
modern societies, virtual trainers are sport-oriented applications (e.g., tai-chi, biking, fitness), which
can provide physical training programs with performance feedback and motivational support [Lowe
and ÓLaighin 2012]. These physically engaging games the so-called exergames are now flourishing.
In the case of a personal anthropomorphic virtual coach which guides the user in a physical activity
(such as a fitness coach), the social aspects (dispositions, attitude and affects) conveyed during the
interaction are key determinants of the quality of the relationship between the user and his
personalized virtual trainer [Raedeke 2007, Jackson et al. 2011]. In addition, enjoyment has been
recurrently found as a key determinant of energy expenditure [Lyons et al. 2014] and physical
exercising [Dacey et al. 2008].
The context of this paper is the design of a full body interactive virtual coach that can convey social
attributes through its bodily expressive kinematics patterns (motion described through space and
time variations [Runeson and Frykholm 1983]. Of particular importance is the possibility to control
the social perception formed by users while engaging with the virtual coach. Whilst empirical studies
regarding the embodied nature of social phenomena accumulate, it appears difficult to draw general
guidelines from this fragmented body of studies for the design of our specific application [Kleinsmith
and Bianchi-Berthouze 2013]. First, only a few recent studies focus on kinematics patterns in bodily
communication and evaluate the corresponding social perceptions [Gross et al. 2010]. Second, to date,
no study has proposed to analyze conveyed social cues within the context of a coach performing a
predefined fitness movement sequence. To overcome this gap, we previously collected predefined sport
movement sequences performed by several coaches in different emotional contexts. The resulting
corpus includes variability in dispositional and affective states (blinded). Importantly, the protocol
was designed to collect spontaneous behaviors from individuals who were not familiar with acting
methods.
The aim of this paper is to study whether emotions and personality traits are accurately perceived
through kinematics patterns of bodily movements in the context of a fitness coach, who continuously
guides the non-expert user through the movements to be performed. Kinematic patterns of movement
were analyzed in terms of movement qualities according to the effort-shape notation [Dell 1977]
proposed by [Laban 1950]. Three analyses were performed to provide an analysis of the process
social perception. More specifically, thirty-two individuals were asked to rate movements in terms of
conveyed emotion dimensions, personality traits (five factor model of personality) and perceived
movement qualities (effort-shape) from fifty-six fitness movement sequences. The results from this
analysis can contribute to both the understanding of social perception through full body movement
[Thoresen et al. 2012] and the development of expressive virtual sport coaches [Tilmanne and Dutoit
2012], a virtual coach conveying motivation and positivity being more enjoyable than a neutral one.
2. RELATED WORKS
The communicative value of movement qualities has been studied with several approaches coming
from different scientific backgrounds. The first subsection below aims at reviewing these varied
origins and providing more details for the approach used in this paper: the effort-shape [Dell 1977]
analysis method of the Laban notation system. The following two subsections survey works on
associations between movement qualities and two social phenomena that are important for modeling a
virtual full body coach: emotion and personality. Both kinematic patterns of bodily expressions and
perceptions of emotion and personality are reviewed. Interactions between observed personality and
emotion (whether expressed or perceived) are discussed because the overall impression made by the
observer is often a merging of both dimensions. The last subsection considers essential methodological
aspects when studying the perception process from bodily expression to perceived social cues.
2.1 Communication through movement qualities
Nonverbal communication involves different channels such as voice, face, body and touch [App et al.
2011]. Bodily behaviors are themselves decomposed into different types of cues. For a long time,
gestures (specific isolated movements with a clear onset and offset) [Kendon 1994] and postures
(static full body geometrical configuration) [Bull 1987] have been at the heart of bodily communication
research [Dael et al. 2012]. Common to both typologies is the process of forcing a continuous behavior
into a series of discrete events [Grammer et al. 1997]. An alternative option is to embrace the analog
[Wallbott 1985]How
movement is performed ion to communicate,
but is
been evidenced with point-light displays developed by [Johansson 1973].
Movement quality is the subjective description of the objective measure of human kinematics often
related to its inner dynamics [Runeson and Frykholm 1983]. One approach for studying movement
quality has been to analyze semantically the subjective kinematic descriptions of behaviors [Wallbott
1985, Gallaher 1992, Shikanai et al. 2013]. These works revealed remarkable consistencies in their
results with a recurring four-factor structure. These movement quality dimensions are space (or
expansiveness, expansion), intensity (or expressiveness, dynamics), fluency (or coordination, stability)
and hastiness (or animation). Structural approaches propose notation systems adapted for the
detailed description of movement dynamics [Birdwhistell 1972, Sheets-Johnstone 1999]. The notation
system proposed by [Laban 1950] has been used in dance studies and more generally in studies of
bodily expressive communication. The Laban movement analysis (LMA) framework is composed of
four main categories (body, effort, shape and space) that qualify the body in movement with respect to
inner intentions. The effort-shape subdomain has received considerable interest as a systematic way
of describing body configurations and movement qualities [Dell 1977, Levy and Duke 2003]. Effort is
composed of four bipolar components including space (indirect/direct), weight (light/strong), time
(sustained/sudden) and flow (free/bound)
[Laban 1950]. Shape flow, its first component, relates to the form of the body itself (growing or
shrinking). The other two components, directional change and shaping movement, characterize the
attitude toward an external object. Interestingly, the Laban conceptualization is not so different from
the experimental results from semantic studies. They all include not only kinematic descriptions of
motion but also perceived dynamic properties. Because human kinematics inherently reveal its
underlying dynamics, the human observer tends to directly perceive the causal aspect of movement:
for example, we perceive the effort behind the movement of man lifting a heavy load [Runeson and
Frykholm 1983].
Several quantifications of these movement qualities have been proposed. In a seminal work,
[Camurri et al. 2003] proposed a quantification of the Laban effort domain based on computer vision
algorithms. Space, weight, time and flow were inferred based on automatically computed activity and
expansiveness time series extracted from video. Other solutions for the effort-shape systems have
been proposed [Samadani et al. 2013, Dyck et al. 2013] in addition to quantifications based on
[Wallbott 1985] or [Gallaher 1992] works (see [Pelachaud 2009] and [Glowinski et al. 2011]). All of
these quantifications are based on kinematic variables. An understudied research question is the
validation of the correspondence between the quantifications (e.g., computed fluency) and the
perceived movement qualities (e.g., perceived fluency). Several early works reported good
correspondence between perceived and actual movement qualities [McGowan and Gormly 1976,
Wallbott 1985]. Only one paper examined this question with modern quantification tools [Samadani et
al. 2013]. The comparison between manual annotation by a certified movement annotator and the
computed movement qualities revealed a high agreement across the effort-shape domain for acted arm
in a wider variety of movements remain to be studied. In the following sections of the paper, we make
the distinction between perceived movement qualities as proximal percepts and computed movement
qualities as distal cues [Scherer 1978, Scherer 2013].
In summary, movement qualities are subjective descriptions of the motion dynamics based on
kinematic information that convey social cues. They can be described with a few dimensions. The
Laban effort-shape system offers a way to systematically characterize them. Kinematic
quantifications have been proposed, but their correspondence with non-expert human observations
remains to be studied. The great potential of movement qualities as a communication modality relies
on this direct perception of other inner forces and constrains (dynamics) and their potential
associations with intentions (tendency to act) and emotions.
2.2 Movement qualities and emotion perception
An emotion is a relatively brief multi-component (cognitive, motor, physiological, and
phenomenological) episode which facilitates a response to an event of significance for the organism
(Davidson et al. 2003). Although we usually refer to emotions as states, the affective process is a
changing state and is therefore dynamic. [Sheets-Johnstone 1999] explains the relation between
emotion and movement by the congruency of their qualitative dynamics. The main theoretical
framework used to predict motion-emotion congruencies is action tendencies [Frijda 1986]. Mainly
described as discrete phenomena, emotions can alternatively be studied with dimensional approaches
[Scherer 2010]. Valence (intrinsic pleasantness of the emotion-eliciting object), arousal (degree of
-eliciting event)
are the principal axes used. These dimensions have been shown to meaningfully explain differences in
affective bodily expressions, even when the protocol is aimed at eliciting categorical emotions [Dael et
al. 2013].
Previous studies have shown that movement qualities such as proximal percepts (perceived
movement qualities) are associated with emotion recognition. At a general level, [Atkinson et al. 2004]
showed that the perceived exaggeration of acted movements increases the emotion perception
accuracy and perceived emotion intensity. Authors supposed that this general qualitative evaluation
was driven by speed and expansiveness. Using sign language, [Hietanen et al. 2004] demonstrated the
importance of movement qualities for emotion recognition. The association between high emotional
arousal (interchangeably called activation [Gross et al. 2010], intensity [Atkinson et al. 2004] or
emotion quantity [Wallbott 1998]) and a combination of high velocity, energy, force, directness and
expansiveness is a recurrent finding [Wallbott 1998, Montepare et al. 1999, Gross et al. 2010, Gross et
al. 2012, Dael et al. 2013, Crane and Gross 2013]. This has been attributed to the physical effort
mobilization or the state of readiness to act provoked by a high arousal [Frijda 1986]. Discrimination
of emotions according to valence is less consistent across studies. Fluency has been found to
differentiate anger (less fluent, jerkier) from elated joy (more fluent) [Montepare et al. 1999, Dael et al.
2013]. Other studies highlight the role of expansiveness, which is greater for elated joy than for
sadness [Gross et al. 2012, Crane and Gross 2013]. In terms of action tendencies, a negative feeling is
observed to be associated with the tendency to flee, whereas a positive feeling is related to free
activation [Frijda 1986]. This is a non-specific motor response, and it probably participates to the
difficulty in recognizing it [Fredrickson 1998].
Works based on movement qualities as distal cues (kinematic cues) demonstrated a similar pattern
of results. With movement purposely performed with different movement qualities, [Meijer 1989]
distinguished three dimensions: rejection-acceptance related to strength and directness of the
movement, withdrawal-approach associated with moving inward or outward, and preparation-
defeatedness (associated with surprise) related to sudden, fast and direct movement. Arousal is
consistently associated with velocity and acceleration [Pollick et al. 2001, Glowinski et al. 2011, Gross
et al. 2012]. [Glowinski et al. 2011] found that in cases of high arousal, jerkiness is enabled to
distinguish between negative and positive emotions. Even if the clustering algorithm defined by these
authors performed well, the pattern was less distinctive for low arousal emotions. When emotion was
induced in dancers, happiness-related expressions were observed to be expended more and were more
impulsive than expressions of sadness [Dyck et al. 2013]. Kinematic patterns of bodily expressions of
emotions have often been studied with gait analysis. Speed often came out as a main discriminative
feature [Roether et al. 2009]. Other critical features proposed are specifically based on gait
biomechanics models and, therefore, are not interpretable in terms of movement qualities [Barliya et
al. 2013].
Whether distal or proximal, movement qualities have been shown to be related to emotional
dimensions. High arousal is consistently associated with more energy and speed. Differences in
valence show a less evident influence on movement qualities and are probably more dependent on the
task. One possible issue is the interaction with the arousal dimension: valenced bodily expressions are
different under low or high arousal emotions, with high arousal being easier to discriminate than low
arousal valenced emotions. Finally, most of these studies are based on actors instructed to portray
emotions. While it ensures high recognition rates of emotional movements, it also limits the
applicability of the results to real word situations [Gross et al. 2010].
2.3 Movement qualities and personality perception
Personality traits describe invariant characteristics across both time and situations.
While several models have been proposed to capture people traits, the five-factor model of personality
has been shown to be the most robust [Costa and McCrae 1992]. This model characterizes an
its, namely, extraversion, agreeableness,
conscientiousness, neuroticism, and openness to experience [John et al. 1991].
Personality traits have been observed to be visible in nonverbal behaviors whether self-reported or
judged by others. The self-reported trait most assumed to be linked to nonverbal behaviors is
extraversion [Gifford 1991, La France et al. 2004]. Often associated with more energetic and
expressive behaviors, extraversion is consistently associated with cheerful expression [Hall et al.
2008]. This is coherent with the description of extrovert people as impulsive and uninhibited
individuals [John et al. 1991]. Other traits may have a more contextual influence on nonverbal
behavior: agreeableness will influence behaviors in a social context [Berry and Hansen 2000].
Neuroticism will modulate reactions in situations that induce negative affects [Gross and John 1995,
Riggio and Riggio 2002]. Few studies analyzed the association between self-reported personalities and
their associated bodily kinematics (distal cues). [Naumann et al. 2009] found extraversion associated
with energetic stances and with agreeable and emotionally stable people standing in a more relaxed
way. Using a music-induced movement paradigm, [Luck et al. 2010] found that extraversion and
neuroticism were the two traits most related to body movement variables. They conclude to a
concordance with the literature that extroverts are energetic and expressive, whereas neurotic people
display reduced, localized and jerkier movements.
Judged personality has also been studied in relation to perceived movement qualities (proximal
percepts). Studying political speech, [Koppensteiner and Grammer 2010] show that agreeableness is
associated with a low level of activity and variability in verticality movements, extraversion with high
activity and small movement fluctuations, and openness to experience with movement direction
changes, and finally that neuroticism was associated with smooth transitions. These findings were
confirmed in a subsequent study with a different protocol [Koppensteiner 2013]. During gait analysis,
movements were reduced to two components evaluated on the effort-shape domain of the
Labanotation system [Thoresen et al. 2012]. Agreeableness, extraversion and openness were positively
associated with perceived spatial extent. Neuroticism was negatively associated with indirect,
sustained and relaxed movement qualities. In all of these experiments, reliability was good,
confirming the consensus at zero acquaintance (when observers have no prior knowledge of the person)
from movement kinematics only. The two consistent findings emerging from these studies (perceived
extraversion associated with movement activity and perceived neuroticism associated with movement
jerkiness) are coherent with the literature on self-reported personality [Riggio and Riggio 2002].
However, having personality impressions based on good general rules (e.g., extroverts have energetic
movements) does not necessarily lead to accurate judgment. When general physical appearance is
available, only extraversion is accurately judged from the energetic stance [Naumann et al. 2009].
With only gait kinematics available, no associations between perceived and actual personality ratings
were found [Thoresen et al. 2012].
While personality judgments have been observed to be strongly influenced by emotional
expressions and evaluation of faces, only one study provides information regarding this phenomena
for movement kinematics [Thoresen et al. 2012]. Perceived positive valence was strongly associated
with adventurousness, extraversion, trustworthiness and warmth, and perceived high arousal was
moderately correlated with adventurousness, extraversion and neuroticism. This phenomenon has
been called temporal extension [Zebrowitz et al. 2007] or overgeneralization of emotional cues
[Montepare and Dobish 2003].
Overall, self-reported and judged personalities have been shown to be related to movement
qualities (distal or proximal) but without any evidence of accuracy. Observers form reliable and
coherent social perceptions regarding the target person but misinterpret these available cues, merging
personality traits and emotional dimensions via an overgeneralization process.
2.4 Methodological aspects
Studies on nonverbal behaviors often analyze the different components of the perception process
separately: people states and traits, bodily expressions, perceptions of bodily expressions and people
states and traits. To account for the whole process, some research paradigms have been proposed to
structure the domain. The Brunswikian lens model is by far the most widely used [Brunswik 1956]. It
is composed of three variables: a distal psychological variable (e.g., a personality trait) and several
available cues (e.g., nonverbal behaviors) from which a perceiver makes an imperfect evaluation (e.g.,
perceived personality) of the distal variable (evaluated criterion). The gap between the distal
psychological variable and available cues is called the ecological validity. Cue utilization is the
association between available cues and evaluated criterion. Accuracy, which is the correlation between
the distal psychological variable and evaluated criterion, is called achievement. Several of the studies
reviewed in this paper refer to this model directly to structure their analysis [Gangestad et al. 1992,
Bernieri et al. 1996, Naumann et al. 2009]. An extension of the model [Scherer 2013] proposed by
[Scherer 1978] contrasts the available cues from the Brunswikian lens into distal cues (measured and
quantified cues) and proximal percepts (evaluation of cues). This part of the analysis is called
perceptual representation. The research presented in this paper is structured following this approach.
This paper presents an analysis of co-variations of the different variables from the four lens
components. To do so, multiple Pearson correlations are presented, the multiplication of variables
making them numerous. Such an approach raises the question of multiple comparisons which
increase type 1 error risks (i.e., false positives). A common practice has been to adjust the p-value
threshold for significance according to the number of tests done, such as with the Bonferroni method.
In addition to being inconsistent (methods and scopes of adjustments vary across authors), such an
approach is highly conservative drastically increasing the risk of type II error (i.e., false negative).
Alternatively, Confidence Intervals (CI) provides information to interpret significant and non-
significant results and enable the computation of meta-analyses [Nakagawa 2004, Cumming and
Finch 2005]. As a consequence, in this paper we systematically report correlations (which are effect
sizes) and confidence intervals (with a standard confidence level of 95%). If the interval includes the
0.0 correlation, then the null hypothesis cannot be rejected. Bootstrapping procedures (n=5000) are
used to compute CI to remove the potential bias of relying on a hypothesized distribution.
Lastly, when one behavior is judged by several observers, the overall rating can be summarized in
two ways: by aggregating the observer response and then model it or by modeling it for each observer
and after average the models together. The first (mostly used) approach is a consensus or pooled
rating [Kenny 1991]. This measure is interesting when ones is interested by impression made to a
small group, like a committee [Ambady et al. 2000]. It is often confounded with the second approach
nomothetic and idiosyncratic judgment and are subject to considerable debate [Philippe et al. 1981].
Few articles undertook the approach to compute both [Bernieri et al. 1996, Naumann et al. 2009]. In
this paper, we are interesting in the impression formed while watching a fitness coach. This context
induces that the impression is rather made by a small group following the coach, thus we opted for a
consensus approach.
2.5 Points addressed in this paper
The aim of this paper is to study the extent to which spontaneous emotions and personality traits are
perceived through full body movement qualities performed by sport coaches with non-expert observers
in the context of a fitness coaching task. Knowing what drives social perceptions and whether they are
accurate will provide us guidelines and labeled data for the design of a full body virtual interactive
sport coach to provide a personalized coaching experience. For clarity purpose, in this paper
raters) in opposition to coaches who performed the movement sequences.
The originality of our approach resides in several points. First, the analysis is based on a corpus of
self-reported measures and motion capture data of non-acted behaviors: coaches (with their inter-
individual differences in terms of personality) performed the same predefined movement sequence
several times and under different contexts, eliciting different emotions. While movement variations
might be very subtle (and less perceivable than the ones collected using acted protocols), they are
spontaneous per se (i.e., not activated from an internal model of each emotion and personality trait).
Second, the kinematic variations of the motion sequences are analyzed both as distal cues (computed
movement qualities) and proximal percepts (perceived movement qualities) using the effort-shape
characterization proposed by [Laban 1950] and [Dell 1977]. This enables us to analyze their pairwise
similarities (e.g., computed weight with perceived weight) called perceptual representation for the first
time in non-expert observers. We expect to have positive pairwise correlations between the
kinematics-based computation of movement qualities and untrained observers perceived movement
qualities. Finally, we assessed both self-reported and judged emotions and personality traits to have a
complete understanding of how non-expert observers perceive the nonverbal behaviors performed by
the coach. For both social phenomena, ecological validity, cue utilization and achievement are assessed.
For the perception of emotions, we expect high self-reported and perceived arousals to be related to a
high value of energy (weight) and impulsiveness (time). We also hypothesize that arousal will be the
only dimension accurately evaluated by observers because several studies observed that this
dimension is the most clearly related to bodily behaviors [Glowinski et al. 2011]. Given inconsistent
results from the literature and the novelty of our task, we cannot draw expectations for the valence
dimension. For the perception of personality traits (according to the five-factor model of personality),
we expect extraversion and neuroticism to be related to computed movement qualities [Riggio and
Riggio 2002]. Coaches with high-level scores of extraversion should be higher on computed energy
(weight) and impulsiveness (time) than coaches with lower level scores of extraversion. Coaches with a
high level of neuroticism should be lower on computed smoothness (flow) and spatial extent (shape
qualities) than Coaches with a lower level of neuroticism. Given the novelty of our task and the few
available results regarding studies with kinematics only, no a priori specific hypothesis can be formed
regarding perceived personality traits. Based on previous studies, we expect observers to be
inaccurate in personality trait evaluations [Thoresen et al. 2012].
3. CORPUS OF SPORT MOVEMENTS IN AFFECTIVE CONTEXTS
The perceptual study presented in this paper is based on a corpus of predefined movement sequences
performed by coaches with different personalities under different emotional contexts. The collection of
this corpus is extensively described elsewhere (blinded). The following section details aspects that are
specifically relevant for the scope of this paper.
3.1 Coaches and Materials
Twenty French sport science students (mean age = 21.3, SD = 1.7, 11 females) participated in the
study. Coaches collected with a full body motion capture system. The marker setup
was composed of 36 markers (full body marker set provided by the Optitrack motion capture software).
The experiment room was equipped with 10 infrared cameras (S250e Optitrack system, frequency:
120 Hz, resolution: 832*832). Each Coach received one week before the experiment a video of the
motion sequence to be performed. All coaches received the same movement sequence in order to only
analyze kinematics variations. They had to learn before arrival this movement sequence involving the
whole body. The choreography was composed of two different foot movements (i.e., step, mounted
knee), repeated twice for each side, two different arm movements (i.e., waved arm, uppercut) twice for
each side, and two combinations of each previous foot and arm movement (i.e., step/wave, mounted
knee/uppercut) repeated twice for each side (Figure 2 shows key frames of the motion sequence as an
example). In each condition of the protocol (four conditions), the coach was asked to repeat this whole
choreography three times (average length of the three repetitions of 130 minute). The general
instruction was to perform the choreography in the best possible way.
3.2 Emotional contexts
Each coach performed the movement sequence under four different conditions to induce variability in
terms of emotional experience. First, the protocol began with the control condition: performing the
movement sequence without any context. The three other conditions positive, negative, and
motivated- were counterbalanced across coaches. In the positive condition, the coach received a reward
for his participation (food and electronic devi -up of
funny videos as a distraction task. In the negative condition, the experimenter explained to the coach
that the video of the performance would be streamed in real-time at a remote lecture hall in front of
hundreds of students for pedagogical purposes. Finally, in the motivated condition, the coach was
asked to imagine him/herself as a fitness trainer who had to motivate his/her audience.
3.3 Movement qualities computation
Movement data were processed using Matlab R2012b. Positions from nine segments were used (one
marker position used): hip, chest, head, arm, forearm, hand, upper-leg, leg, and foot. Five movement
qualities were computed for each segment to model the effort-shape domain proposed by Laban
[Laban 1950]: impulsiveness (time), energy (weight), directness (space), smoothness (flow) and spatial
extent (shape qualities) (synonyms given in Table 1). A standard average is used for impulsiveness
computation of spatial extent use the distance of hands and feet to the mass center. We obtained one
time series per movement sequence per movement quality. Equations to model these qualities are
described in Appendix A. Because our protocol predefines the movement trajectory, the directness
index will only vary according to speed. Therefore, we do not consider it in the analyses.
3.4 Emotion and Personality Assessment
Experienced emotions and personality traits were assessed by self-reporting. Self-reported emotion
states were assessed with the differential emotional scale [Ouss et al. 1990]. While most of the labels
were not used by coaches (i.e., all coaches reported to not feel these labeled emotions), two labels
showed good variability across coaches and conditions: enjoyment and surprise (blinded). In a
dimensional view of emotions, enjoyment represents the valence dimension (i.e., positive) and surprise
is associated with the activation dimension [Alvarado 1997]. Personality traits were considered
regarding the five-factor model [Costa and McCrae 1992],
personality across five broad traits, namely, extraversion, agreeableness, conscientiousness,
neuroticism, and openness [John et al. 1991]. The French version of the Big Five Inventory was used
[Plaisant, Courtois, Réveillère, Mendelsohn, & John, 2010].
4. PERCEPTUAL STUDY
The aim of this perceptual experiment is to collect social perceptions formed by individuals watching
expressive movement sequences performed by a sport coach. To do so, individuals evaluated
movement sequences from our corpus on perceived movement qualities, perceived emotion and
perceived personality. To isolate the influence of kinematics on perception from other sources of
information (e.g., facial expressions, clothes, etc.), we used as stimuli 3D animations that were
designed from the motion capture data described in the precedent section. Three studies provide an
analysis of the perception process for bodily signatures of emotion and personality traits (expression,
information transmission and perception) of coaches through movement qualities.
4.1 Method
4.1.1 Analysis framework. To structure the analysis of the perception process composed of the
expression, transmission and perception of social cues, we follow the paradigm proposed by [Scherer
1978, Scherer 2013]. As depicted in Figure 1, the framework describes four elements: distal
environmental criterion (e.g., personality traits), distal cues (e.g., movement speed), proximal percepts
(e.g., perceived movement speed) and evaluated criterion (e.g., perceived personality traits). These
four elements are studied via four analyses: ecological validity, cue utilization, achievement and
perceptual representation. In this paper, we organize them in three studies. First, the perceptual
representation of movement qualities is analyzed. Validating the assumed inherent link between
kinematics and the perceptual aspects of these qualities is a key point to understanding the decoding
process. Second, ecological validity, cue utilization and achievement are studied for the communication
(recognition-identification) of emotional states. Finally, ecological validity, cue utilization and
achievement are studied for the communication (recognition-identification) of personality traits.
4.1.2 Participants. Thirty-two French participants were involved in this unpaid online perceptual
study (mean age = 32.6, SD = 15.7, 15 females). The perceptual test was made available online.
Participants were contacted by email with only information on the approximate survey duration and
the conditions to undertake the survey: placed in a quiet place with good Internet access and seated at
a comfortable distance from the screen. Participants were not paid.
4.1.3 Materials. A selection of one movement sequence in each condition from fourteen coaches of
the corpus (n=20) was made to reduce the duration of the survey to ninety minutes. The fourteen
coaches were chosen randomly. This resulted in 56 movement sequences of 30 seconds each. To control
the effect of the physical appearance of coaches in the corpus (gender, size, weight, etc.) [Atkinson et
al. 2004], a video was made of a 3D virtual mannequin playing the motion capture data. The 3D
mannequin provided with the motion capture software was used. The animation being based on 3D
joint rotations, the proportions of the mannequin remain the same across coaches. Each video was
reframed to control the mannequin size. The viewpoint for the video was a front view as it is the
common setup in pupil-coach tasks. Each video was preceded by a white fixation cross and followed by
the display of one sentence asking the participant to complete the questionnaire. Figure 2 presents
the key frames of one video stimulus as an example (to read from left to right, from top to down). Self-
reported personalities (five-factor dimensions from the BFI-fr; [Plaisant et al., 2010]), self-reported
affective experiences (enjoyment and surprise scales from the DES; [Ouss et al., 1990]) and computed
movement qualities are known for each movement sequence.
Figure 1 The three analyses of this paper within the framework from [Scherer 2013], visual
modified from [Vinciarelli and Mohammadi 2014]
* Fill out the questionnaire. Click on next when you are done. Thank you.
4.1.4 Procedure. The first part of the instruction section described the general aim of the study as
The second part presented the instructions regarding the good conditions that a participant should be
in before continuing. It was made explicit that a good internet connection was required. In case of
issues with video playing, participants had to inform us. One participant data was not taken in this
study because of internet issues. It also explained the nature of the videos to be evaluated: every video
is a different movement sequence but with the same appearance (a 3D virtual manikin). Nothing
about the identities (body and face) or the conditions of the capture session was provided to the
participant. The third part asked the participant for socio-demographic information (sex, age,
practiced sport and level of expertise, contact). The following three pages provided detailed definitions
regarding the dimensions to be evaluated for each video. These dimensions were grouped under three
labels: movement qualities, emotions and personalities. For each dimension, a set of synonyms and a
left and right anchor were provided to constitute a semantic differential scale. When possible, the
labels were taken from the literature. At the end of each definition, participants were asked to report
the clarity of the scale on a five-item Likert scale. Table 1 shows the dimensions with their associated
left and right anchors (French versions are available in Appendix B). Each dimension is evaluated on
a 10-item Likert scale.
Before the evaluation part, the participants viewed and rated three training trials. These trials
were movement sequences that were not part of the set of sequences used in the perceptual study. All
of the definitions were recapped at the bottom of the page to enable the participant to read them again
if needed. For the evaluation part, the participant viewed and rated the 56 videos. The video order
was randomized across participants but not the order of questions. At each page, all dimensions had
to be rated to go on the next page.
Figure 2 Key frames of one video stimulus (as an example) presented to the participant (duration:
about 30s)
4.1.5 Item clarity and Rater reliability. All of the dimensions were rated by participants as having
very clear meanings, but the less clear dimension was Dominance (mean clarity = 4.25, on a five-item
Likert scale). Looking at the averaged pairwise inter-rater correlations averaged across all 15
dimensions, two raters were below the threshold of 0.20 (very low agreement with other raters). These
two raters were removed from the study. We performed an outlier analysis on average scores and a
standard deviation of the scores per participant on each of the 15 scales [Hoaglin et al. 1986]. Three
raters were removed based on this analysis. A second outlier analysis on standard deviation across
the 15 dimensions for each movement sequence aimed at looking for participant ratings
systematically (when not taking seriously the survey, a standard deviation of zero means rating every
scale with the same response).
Table 1 Evaluated dimensions with their left/right anchors and their raters reliability indicators
Movement qualities
Left anchora
Right anchorb
Average inter-
rater
correlations
(r)
Spearman-
Brown
Coef. (Rsb)
Cronbach
alpha
Smoothness
Abrupt
Smooth
0.18
0.89
0.86
Directness
Indirect
Direct
0.09
0.66
0.68
Energy
Limp
Energetic
0.53
0.96
0.97
Impulsivity
Restrained
Impulsive
0.46
0.95
0.96
Spatial extension
Restricted
Vast
0.41
0.95
0.95
Emotion
Left anchora
Right anchorb
Valence
Negative
emotion
Positive emotion
0.39
0.95
0.95
Intensity
Low intensity
Strong intensity
0.33
0.95
0.93
Dominance
Subdued
Dominant
0.37
0.93
0.94
Motivation
Without
motivation
Strongly motivated
0.40
0.95
0.95
Stress
Relaxed
Stressed
0.09
0.75
0.76
Personality
Left anchora
Right anchorb
Extraversion
Introverted
Extroverted
0.40
0.93
0.95
Openness
Humdrum
Adventurous
0.38
0.95
0.94
Neuroticism
Neurotic
Stable emotionally
0.17
0.87
0.84
Conscientiousness
Careless
Conscientious
0.08
0.65
0.67
Agreeableness
Disagreeable
Warm
0.31
0.93
0.92
a Score = 1
b Score = 10
= not acceptable reliability
This analysis revealed no supplementary outliers. Three different indicators of the
reliability were computed for the 30 remaining raters on each dimension (Table 1). The average inter-
rater correlations show the amount of agreement, and the Spearman-Brown coefficient shows the
effective reliability [Rosenthal 2008]. We also provide (often reported as an
alternative reliability indicator [Gross et al. 2010]). Three dimensions Directness, Stress,
Conscientiousness- showed very poor agreement (r<0.1) and reliabilities (Rsb<
removed from the analysis. For the 12 remaining scales, the (to obtain a
consensus score) for each video for the analysis.
4.1.6 Correlation bias and General impression: selecting and transforming variables. A first look at
bivariate correlations between evaluated dimensions revealed moderate to strong positive correlations
between all dimensions (r>0.6). This high common variance was confirmed by a principal component
analysis, suggesting a one-component structure accounting for 88.2% of the total variance. The
existence of sizeable pairwise correlations between rated items is a recurrent aspect of judgment
[Thorndike 1920]
emotional cohe[Morewedge and Kahneman 2010], that is, the tendency to think of a person or
an object in a generalized impression (as a whole). This exaggerated coherence is considered to be
caused by -occurrence of rated items [Podsakoff et al. 2003]. Not
considering this inherent multicolinearity in judgment studies can lead to methodological issues and
ambiguous results [Kraha et al. 2012].
Before analyzing this general impression as a social perception characteristic of our task, one
aspect should be controlled. This source of multicolinearity could result from a method bias: the
duration of the survey and the number of items to be evaluated at the same time tend to increase the
risk of correlation bias [Cooper 1981]. One way to clarify this issue is to compare the Halo effect in two
judgment conditions, the second being hypothetically less subject to associative processes [Johnson
1963]. We designed three shorter versions of the test, each one with different dimensions to be
evaluated (one with the five movement quality dimensions, one with the five emotional dimensions
and one with five personality traits dimensions). Each test had only 20 movement sequences per rater.
For these tests, the question orders were also randomized at each movement sequence. The duration
was considerably reduced (24 minutes on average compared to 96 minutes on average for the long
version). A multitrait-multimethod matrix (MTMM) [Campbell and Fiske 1959] was computed to
compare the two methods (full matrix in Appendix C). A MTMM represents the pair-wise correlations
between similar variables measured via different methods. Of particular interest for this paper are
two diagonals: the reliability diagonal (using Spearman Brown formula) and the diagonal which
indicate the inter-method correlation: they provide indication about the inter-method convergent
validity. Table 2 presents these item reliabilities and inter-method correlations. Only the neuroticism
scale showed no convergent validity across methods (the inter-method correlation is only about r=.243)
and a very poor reliability in the short form of the test. This scale was removed from the study.
Table 2 Rated dimension characteristics
Dimensions
Nb of raters
Reliability
(Rsb)
Inter-
methods
correlations
Observed /
Residuals
correlations
Long
Short
Long
Short
r
CI
Movement
qualities
Smoothness
27
10
0.89
0.81
.828
.644
.471 to .784
Energy
27
10
0.96
0.86
.949
.233
-.026 to .497
Impulsivity
27
10
0.95
0.86
.928
.242
.000 to .445
Spatial
extension
27
10
0.95
0.90
.901
.486
.220 to .686
Emotion
Valence
27
10
0.95
0.87
.805
.260
.066 to .480
Intensity
27
10
0.95
0.91
.883
.280
-.007 to .510
Dominance
27
10
0.93
0.88
.736
.267
-.015 to .543
Motivation
27
10
0.95
0.83
.869
.254
.041 to .454
Personality
Extraversion
27
12
0.93
0.92
.888
.198
-.017 to .406
Openness
27
12
0.95
0.88
.844
.113
-.190 to .382
Neuroticism
27
12
0.87
-0.16
.243
-
-
Agreeableness
27
12
0.93
0.84
.830
.363
.156 to .554
The 11 remaining scales can be considered free from method bias and subject to a large Halo effect.
The existence of this strong general impression is not surprising given our task. Raters were asked to
ery subtly different movement sequences without
any contextual information. Given this high ambiguity, they relied on prior knowledge/beliefs and
favored fast associative judgments based on the few salient elements. To disentangle the general
impression from the specific component of the evaluation, we use a procedure similar to [Landy et al.
1980]. First, we extract the general impression (G) over the remaining 11 scales with an exploratory
factorial analysis (EFA with the maximum likelihood estimation method). This method is more
appropriate than a principal component analysis because its objective is to reproduce the
intercorrelations of a set of indicators, recognizing a part of a unique variance for each [Brown 2006].
The second step is to partial out the general component from each variable by regressing them on G.
Using the residuals as unique over the shared contribution of the predictors is often performed in
cases of correlated variables [Graham 2003, Engell et al. 2007]. Extracted residuals from these
regressions form the specific part of the evaluation. Correlations between observed and specific
components for each scale are presented in Table 2 along with confidence intervals. For the rest of the
analysis, we keep the specific components that correlate significantly with their observed part. They
still contain a meaningful part of the observed variance once the general component variance is
removed. It should be noted that these five residual variables represent a very small part of the
observer ratings variance compared to the general impression dimension (88.2% of the variance) and
caution should be taken in drawing conclusions about their associations.
This procedure provided a reduced number of variables: general impression (G), perceived
smoothness (P-smoothness), perceived spatial extent (P-spatial extent), perceived valence (P-valence),
perceived motivation (P-motivation) and perceived agreeableness (P-agreeableness). In addition to a
considerably reduced multicollinearity in our data, the data provided a more representative picture of
the discriminative power of the raters. This general impression G, composed mainly of impulsivity
(r=.972), energy (r=.973), intensity (r=.962), dominance (r=.963), extraversion (r=.980) and openness
(r=.994), seems cohere, working
as an expressivity halo similar to [Bernieri et al. 1996]. Because they include movement qualities such
as emotion dimensions and personality traits, the general impression G will be considered in the three
studies.
4.2 Analysis 1: Perceptual representation of movement qualities
In this first step, we analyze the relations between movement qualities, computed or observed, within
our corpus. Our aim is twofold: to understand how movement qualities are related to each other (e.g.,
is energy independent of smoothness?) and to assess the associations be
ratings and quantified movement qualities.
4.2.1 Computed movement qualities. For each movement sequence, five movement qualities were
computed to model the effort-shape domain [Dell, 1977] proposed by Laban: energy, impulsivity,
directness, spatial extent and smoothness (Appendix A). Table 3 shows pairwise correlations between
computed movement qualities for the 56 movement sequences from our corpus.
Computed energy and impulsivity are strongly and positively correlated (r=0.89). Computed
smoothness is positively and moderately correlated to all dimensions (r>0.35). Computed spatial
extent is moderately and positively correlated to computed energy (r=0.360). These results show a
significant amount of shared variance across our movement qualities, with energy and impulsivity
being very close dimensions.
Table 3 Pearson correlations and CI between computed movement qualities (n=56)
C-energy
C-impulsivity
C-spatial
extent
C-smoothness
C-energy
1
0.89
0.36
0.50
.822 to .935
.106 to .573
.382 to .619
C-impulsivity
1
0.21
0.42
-.038 to .437
.269 to .550
C-spatial
extent
1
0.37
.159 to .592
C-smoothness
1
4.2.2 Perceived movement qualities. On the perceptual side, three movement qualities are
available: the general impression (mainly composed of energy and impulsivity), perceived smoothness
and perceived spatial extent. Table 4 shows pairwise correlations between the perceived movement
qualities for the 56 movement sequences of our corpus.
The general impression does not correlate with perceived smoothness and perceived spatial extent
per se (shared variance removed). Perceived smoothness and spatial extent are orthogonal as well
(r=0.093, p>0
independent variables.
Table 4 Pearson correlations and CI between perceived movement qualities
G
P-smoothness
P-spatial extent
G
1
0.00
0.00
/
/
P-smoothness
1
0.09
-.112 to .345
P-spatial extent
1
* for p<0.05
** for p<0.01
4.2.2 Perceptual representation. Perceived movement qualities are not necessarily highly correlated
to computed movement qualities means. For example, perceived energy could be more highly
correlated to the computed energy peaks in time (which are more salient) than the computed energy
mean. To analyze the relations between perceived movement qualities and a larger spectrum of
movement qualities computations, we computed eight values around the mean representing
intermediates between the minimum (=10%), mean (50% in figure 3) and maximum (=90%).
Considering a time series, the mean (50% in figure 3) is a value where the area under and above the
curve is equal (each represents 50% of the total area under the curve). The eight values computed are
the ordinates, where the area under the curve is increased or decreased by 10%. In figure 3, each
graph shows pairwise correlations between the three perceived movement qualities and the nine
computed variables of the treated quality.
Figure 3 Correlations between computed and perceived movement qualities
Strong positive correlations were found between general impression G and computed energy
(r=0.821 for 50%) and impulsivity (r=0.935 for 50%), regardless of the mean variations. G was
positively and moderately correlated with computed smoothness (p=0.615, for 40%). Because the
factor G is mainly composed of perceived impulsivity variance, these results show a high convergent
validity between G and the computed impulsivity mean: the highest pairwise correlation and the same
pattern of correlations with other computed qualities. There was a positive weak correlation between
the perceived spatial extent and computed energy (r=0.274, for 50%), weak positive correlations
between the perceived spatial extent and computed spatial extent (r=0.343, for 50%) and a positive
and moderate correlation between the perceived spatial extent and low values of computed
smoothness (r=0.496, for 10%). The first two results regarding the perceived spatial extent show
partial convergence between the computed and perceived spatial extents. The negative correlation
between the perceived spatial extent and low values of computed smoothness is less interpretable but
does not impair the convergent validity because the computed spatial extent (for 50%) correlates
positively with computed smoothness (r=0.323, for 10%). Perceived smoothness was negatively
correlated with high values of computed energy only (r=-0.362, for 90%). Thus, perceived smoothness
shows no convergent validity with computed smoothness. The perceptive variable was more associated
with peaks of energy within the time series.
This first analysis shows that observers were highly accurate in evaluating kinematic impulsivity
and energy. The perceived impulsivity and energy (the two dimensions are almost similar) seems to be
the kinematic basis for the general impression G. Perceived spatial extent showed partial convergent
validity and perceived smoothness with no convergence at all. Overall, untrained observers were only
heterogeneously accurate, contrasting with previous studies showing good agreement between
certified movement analysts and computed qualities. For the following two studies, we will keep both
movement qualities and evaluations (perceived and computed) in the analysis.
4.3 Analysis 2: Emotion Analysis
In this second step, we analyzed the perception process of emotions through full body movement
qualities, computed or observed, within our corpus. Our aim is twofold: to understand how movement
qualities are related to self-reported and perceived emotions and to assess the accuracy of emotional
perception.
4.3.1 Ecological validity and Cue utilization in emotion perception. For each movement quality
(computed and perceived) and each emotion, we tested the extent to which the movement qualities
correlated with the induced emotion (ecological validity). We also tested the extent to which
movement qualities correlated with perceived emotions (cue utilization). Table 5 presents the
ecological validity and cue utilization for emotions (self-reported and perceived).
Self-reported emotions show that computed and perceived cues correlate. Coaches -reports of
enjoyment were positively correlated to the computed spatial extent and perceived general impression
and were negatively correlated to the perceived spatial extent. Self-reported surprise was positively
correlated with all cues but not with perceived smoothness. While felt surprise seems clearly related
to the general impression and impulsivity (which share some variance with all computed movement
qualities), the pattern for enjoyment is less clear: movements were found to be more expanded
kinematically but perceived as being less expanded and more impulsive.
In terms of perceived emotion, the general impression G is interpreted as emotional intensity.
most strongly correlated with computed energy and
impulsiveness and were moderately correlated computed spatial extent and computed smoothness.
O (negatively) correlated with perceived
smoothness and were moderately (positively) correlated with computed spatial extent. Perceived
motivation was only correlated with computed spatial extent. Overall, these results suggest different
patterns in cue utilization for all three perceived emotions: movements perceived as being more
positive (in affect) and motivated were recorded as more expanded whilst movement perceived more
expressive were more impulsive. In the perceptual realm, valence and motivation differ in their
association with perceived smoothness: only perceived affectively negative movements were related to
movement perceived smoother.
Both in ecological validity and cue utilization, significant correlations exist between emotions (self-
reported or perceived) and computed and perceived movement qualities. This means that sufficient
information exists for the observer to form accurate perceptions of felt emotions: experienced emotions
are kinematically encoded (i.e., embodied) and are correlated with perceived movement qualities, and
these same computed and perceived cues are correlated with perceived emotions.
Table 5 Ecological validity and cue utilization for emotions (Pearson correlations and CI)
Ecological validity
Perceptual cues
Cue utilization
S-enjoyment
S-surprise
G
P-valence
P-motivation
Distal cues
0.13
0.43
C-energy
0.82
-0.02
0.02
-.149 to .370
.138 to .644
.711 to .901
-.251 to .237
-.195 to .212
0.24
0.44
C-impulsiveness
0.94
-0.15
-0.14
.010 to .457
.123 to .667
.894 to .963
-.362 to .097
-.339 to .049
0.36
0.34
C-spatial extent
0.27
0.41
0.37
.099 to .581
.136 to .548
-.007 to .493
.181 to .603
.185 to .550
-0.02
0.40
C-smoothness
0.41
0.14
0.24
-.270 to .222
.235 to .569
.244 to .543
-.295 to .553
-.092 to .535
Proximal
percepts
0.28
0.48
G
1.00
0.00
0.00
.041 to .490
.206 to .672
/
/
/
-0.36
0.27
P-spatial extent
0.00
0.22
0.13
-.526 to -.124
-.046 to .504
/
-.011 to .401
-.137 to .342
-0.07
0.09
P-smoothness
0.00
-0.65
0.06
-.346 to .226
-.148 to .305
/
-.770 to -.483
-.236 to .357
4.3.2 Achievement between self-reported and perceived emotions. For each emotion, we tested the
extent to which the perceived emotions correlated with the coach actual emotion (achievement) to
find out if observers were accurate in their ratings. Table 6 presents the achievement for emotions
(self-reported and perceived).
Perceived intensity (general impression G) was positively correlated with both self-reported
enjoyment and surprise, more strongly with the latter. Perceived valence showed no significant
correlations. Perceived dominance was positively correlated to self-reported surprise. These results
demonstrate partial accuracy in observers Accuracy was nonexistent for valence
(represented by S-enjoyment and P-valence, respectively) and moderate for arousal (represented by S-
surprise and G, respectively). The association between general impression and felt enjoyment
demonstrates that observers had difficulties to differentiate experienced enjoyment and surprise and
that they rated them as almost being equally expressive. Perceived motivation correlation with felt
surprise confirms the partial accuracy of observers in detecting emotion activation: motivation induces
more energy expenditure (a component of arousal) even if positively connoted.
Table 6 Achievement for emotions (Pearson correlations and CI)
Evaluated criterion
G
P-valence
P-motivation
Distal criterion
S-enjoyment
0.28
0.02
-0.07
.041 to .490
-.261 to .319
-.320 to .199
S-surprise
0.48
0.11
0.20
.206 to .672
-.131 to .362
.008 to .399
4.4 Analysis 3: Personality Analysis
In the third step, we analyzed the perception process of personality through full body movement
qualities, computed or observed, within our corpus. The aim is twofold: to understand how movement
qualities are related to self-reported and perceived personality traits and to assess the accuracy of
personality perceptions.
4.3.1 Ecological validity and Cue utilization in personality perception. For each movement quality
(computed and perceived) and each personality trait, we tested the extent to which the movement
qualities correlated with the coach actual personality (ecological validity) and the extent to which
movement qualities correlated with perceived personality (cue utilization). Table 7 presents the
ecological validity and cue utilization for personality (self-reported and perceived).
Self-reported personality traits showed that computed and perceived cues correlate. Coachesself-
reports of extraversion were positively correlated to the computed spatial extent. Self-reported
openness was negatively correlated with computed energy, impulsiveness and perceived spatial extent.
Self-reported neuroticism was positively correlated with general impression only. Coaches
agreeableness scores were negatively correlated with computed spatial extent and perceived spatial
extent. Finally, self-reported consciousness was only positively correlated to perceived smoothness.
These results show that movement qualities (perceived or computed) convey some information
regarding individual personality traits: as in the literature, individual highs in extraversion make
movement more expanded, but the other traits have their associations.
Table 7 Ecological validity and cue utilization for personality traits (Pearson correlations and CI)
Ecological validity
Perceptual
cues
Cues utilization
S-extra
S-openness
S-neuro
S-agreea
S-conscie
G
P-agreeableness
Distal cues
-0.10
-0.29
0.27
-0.14
-0.22
C-energy
0.82
-0.13
-.344 to .152
-.484 to -.094
-.012 to .509
-.398 to .115
-.473 to .045
.711 to .901
-.340 to .073
-0.03
-0.26
0.26
0.00
-0.09
C-impulsiveness
0.94
-0.19
-.237 to .178
-.456 to -.072
-.011 to .498
-.261 to .203
-.346 to .183
.894 to .963
-.386 to .000
0.55
-0.01
0.16
-0.44
0.02
C-spatial extent
0.27
0.28
.330 to .714
-.288 to .345
-.130 to .387
-.674 to -.092
-.238 to .284
-.007 to .493
.075 to .492
0.25
-0.02
0.22
-0.10
0.15
C-smoothness
0.41
0.25
.027 to .441
-.234 to .221
-.076 to .440
-.328 to .171
-.134 to .376
.244 to .543
-.083 to .525
Proximal percepts
0.04
-0.15
0.40
-0.00
-0.05
G
1.00
0
-.207 to .269
-.373 to .062
.164 to .598
-.279 to .248
-.325 to .216
/
/
0.06
-0.39
-0.06
-0.29
-0.13
P-spatial extent
0.00
0.01
-.206 to .335
-.573 to -.178
-.232 to .127
-.487 to -.097
-.367 to .107
/
-.210 to .206
0.23
0.04
0.30
0.10
0.42
P-smoothness
0.00
0.57
-.008 to .434
-.191 to .266
-.031 to .576
-.111 to .350
.217 to .592
/
.343 to .736
In terms of perceived personality traits, the general impression G is mainly composed of
strongly correlated with computed energy and impulsiveness and moderately correlated with
computed spatial extent and computed smoothness. Perceived agreeableness was positively correlated
with computed spatial extent and perceived smoothness.
Both in ecological validity and cue utilization, significant correlations exist between personality
traits (self-reported or perceived) and computed and perceived movement qualities. Only self-reported
conscientiousness and neuroticism showed no computed movement quality correlations (although
some were nearly significant). This suggests that sufficient information exists for the observer to form
accurate perceptions of self-reported personality traits: personality characteristics are kinematically
encoded and are correlated with perceived movement qualities. These same computed and perceived
cues are correlated with perceived personality traits.
4.3.2 Achievement between self-reported and perceived personality. For each personality trait, we
tested the extent to which personality traits correlated with the coaches
actual personality traits (achievement). Table 8 presents the achievement for these traits (self-
reported and perceived).
Perceived extraversion and openness (within the general impression G) were positively correlated
with self-reported neuroticism. Perceived agreeableness showed positive correlations with self-
reported extraversion and openness. These results demonstrate
of personality traits. Instead, their perceptions were associated with other personality traits,
indicating wrong attributions. The wrong evaluation of self-reported neuroticism was probably caused
by the mediating variable computed energy: aroused by the protocol, more neurotic people made more
energetic movements than perceived by an extrovert and open personality person.
Table 8 Achievement for personality
Evaluated criterion
G
P-agreeableness
Distal criterion
S-extraversion
0.04
0.29
-.207 to .269
.068 to .497
S-openness
-0.15
0.27
-.373 to .062
.047 to .491
S-neuroticism
0.40
0.09
.164 to .598
-.197 to .379
S-agreeableness
-0.00
0.10
-.279 to .248
-.176 to .403
S-conscientiousness
-0.05
0.22
-.325 to .216
-.001 to .446
4.5 Discussion
The three studies presented in this paper aimed at contributing to our understanding of emotions and
personality trait perception based only on kinematic information provided by full-body movements.
The specific context of our task is a fitness coach who continuously guides non-expert users through
the movements to be performed. Thus, task movement was a predefined sequence performed in
various emotional contexts and varied according to the coach personality. Kinematic patterns of
movement were analyzed in terms of movement qualities according to the effort-shape notation [Dell
1977] proposed by [Laban 1950]. Thirty-two participants were asked to rate movements in terms of
conveyed emotion dimensions, personality traits (five-factor model of personality) and perceived
movement qualities (effort-shape) for fifty-six fitness movement sequences from our corpus.
Before conducting the three studies, the exploration of collected data from raters revealed that
high intercorrelations were present between the evaluated dimensions (movement qualities, emotions
and personality traits). Approximately 88% of the total variance could be explained by a unique
component mainly composed of perceived impulsivity, energy, intensity, dominance, extraversion and
openness. Although not often reported in the literature, this general impression is consistent with the
interpersonal halo reported by [Bernieri et al. 1996]. These authors summarize this general
impression Our methodological approach consisted of computing this
general impression to form a new variable and to extract the specific variance of other variables to
avoid multicolinearity. In the design of a virtual coach, this general impression would be the first
variable to model because it conveys most of the positive attributes. Identifying the halo does not
provide information about the sources of this perceptual generalization and future researches could
attempt to identify them. The perceptual halo can resu
more information, these associations would appear) or can be caused by the lack of relevant
information in movement kinematics (the less the individual information the more she/he relies on
assumptions which are more subject to holistic reasoning). As more ecological stimuli (i.e., with facial,
clothes and context information) have been shown to induce accurate judgments about personality and
emotions [Ambady et al. 2000], the second possibility seems more plausible. Future works could
investigate how movement kinematic complements other sources of information in the process of
impression formation. The following studies that we described decomposed the analysis in three steps:
the first analysis considered the perceptual representation of computed movement qualities; the
second analysis investigated how emotions are perceived through the kinematic modality; and the
third analyzed how personality traits are perceived through the kinematic modality.
In analysis 1, the associations between computed and perceived movement qualities (called
perceptual representation) were analyzed. The only previous work that compared manual annotation
by a certified movement annotator and the computed movement qualities revealed a high agreement
across the effort-shape domain for acted arm movements [Samadani et al. 2013]. In our work, with
non-expert observers and spontaneous full body movement in contrast to our hypothesis, we found
only partial convergence validity between movement qualities computed from kinematic variables and
perceived movement qualities. The general impression G was accurately driven by movement
impulsivity (amount of acceleration) and energy, but perceived spatial extent was only partially
related to computed spatial extent and perceived smoothness was not related to computed smoothness.
We also analyzed the relations between perceived movement qualities and a larger spectrum of
computations of the effort-shape dimensions: this analysis revealed that perceived smoothness was
correlated with higher values of energy. We can draw two conclusions from these results. First, the
general impression conveying most of the expressive attributes is accurate: in the context of a virtual
coach, varying movement impulsiveness/energy is the first parameter to explore. Second, for non-
expert observers, how the kinematics is perceived is not necessarily what kinematics is: more study is
required to evaluate what drives perceived smoothness and spatial extent. An alternative method
would be to automatically explore possible computations of perceived movement qualities [Kikhia et al.
2014].
In analysis 2, the associations between movement qualities (perceived and computed) and emotions
(perceived and self-reported) were analyzed. Based on previous works [Glowinski et al. 2011], we
hypothesized that high self-reported and perceived arousals would be related to high values of energy
(weight) and impulsiveness (time). We also hypothesized that arousal would be the only dimension
accurately evaluated by observers. Our results partially confirm our expectations: felt surprise
(considered as representing the arousal dimension) and enjoyment (considered as representing the
valence dimension) were positively correlated to the general impression, revealing moderate accuracy
and no accuracy, respectively [Gross et al. 2010]. The association between general impression and felt
enjoyment demonstrates that observers were not able to differentiate experienced enjoyment and
surprise and that they rated them as being equally expressive. While perceived valence was not
accurate, perceived motivation showed partial accuracy with a positive correlation with self-reported
surprise [Russell and Mehrabian, 1977, Dael et al., 2013]. From this analysis, we can conclude that in
our fitness task, arousal and valence are dimensions that are perceptually merged within the general
impression: a more impulsive coach will convey more positive and activated affects. Additionally, the
weakness of the associations confirm the fact that spontaneous affects are more difficult to judge than
acted ones.
In analysis 3, the associations between movement qualities (perceived and computed) and
personality traits according to the five-factor model (perceived and self-reported) were analyzed.
Based on previous works [Riggio and Riggio 2002], we hypothesized that coaches who rate high on
extraversion should be higher for computed energy (weight) and impulsiveness (time) than coaches
with lower level scores of extraversion, and coaches with a high level of neuroticism should be lower
on computed smoothness (flow) and spatial extent (shape qualities) than coaches with a lower level of
neuroticism. We also expected no accuracy in personality trait perception [Thoresen et al. 2012]. Our
results confirm our hypothesis partially: as in the vast literature on personality, self-reported
extraversion was observed to be related to computed spatial extent [La France et al. 2004]. Self-
reported neuroticism showed different associations from our expectations: more neurotic coaches had
more energetic movements and were perceived as more impulsive and smoother. We interpret this
discrepancy from the literature as a result of our protocol: the act of performing a fitness movement in
a motion capture suit in front of dozens of cameras can be considered arousing per se. Neurotic
individuals might have been more sensible to this arousing context, leading to more energy
expenditure. In terms of perceived personality traits, extraversion and openness were embedded in
the general impression G and, therefore, were associated with impulsivity. This association seems to
be consistent with a description of extrovert people as energetic individuals [Costa and McCrae 1992].
However, in our task, it was misleading: extrovert coaches performed more expanded movements, not
more impulsive movements. Perceived agreeableness was positively correlated with computed spatial
extent and perceived smoothness. Although we did not form an hypothesis regarding this dimension,
this result is interesting because agreeableness has been observed to be related to individuals
engaging in smooth interaction with others [Graziano et al. 2007]. From this last set of results, we can
draw two conclusions. First, information regarding personality traits are conveyed by coaches
kinematics computed as movement qualities, but observers are not able to accurately identify them:
more information might be needed to reduce the ambiguity of the signals. Second, two personality
traits might be advantageously expressed by a virtual coach: extraversion embedded in the general
impression driven by movement impulsivity and agreeableness, which is strongly associated with
perceived smoothness. However, additional analyses are required to identify what kinematic patterns
drive this perceived smoothness.
spatial extent, emotional valence, motivation and agreeableness) represent a very small part
(although significant) of the overall variance and should be consider with caution. Future studies
could investigate the reproducibility of these perceptual structures within similar circumstances.
Overall, these analyses revealed that information was conveyed but without accuracy, a sign of
high ambiguity. Whilst emotional activation seemed partially accurate via the accurate perception of
energy/impulsivity movement qualities, the other self-reported dimensions were not perceived directly
perhaps because of their more subtle embodiment (i.e., lower correlations with computed movement
qualities). To investigate in a more global manner the overall quantity of information conveyed,
canonical correlations could be used in future research. Such a method requires very large samples
(i.e., about 20 observations per variables in the model) and thus need more participants.
4.6 Future directions
The nature of our perceptual protocol, where observers were passively watching movement sequences,
is a limitation. As future users will actively perform the movement following the virtual coach, their
perception of emotion and personality traits might be different. An interesting question in our future
experimental setup would be to reproduce the perceptual evaluation of this study after performing an
interactive session with the virtual coach.
The presentation of movement sequences via a neutralized 3D manikin also raises several
questions. In our analyses, although accuracy is low for most dimensions, we noted that social
information was conveyed through movement qualities and social perception was related to movement
qualities. This means that observers perceived information but were not able to disambiguate it
probably as a consequence of the impoverished stimuli. Additional cues such as physical appearance
and facial expressions would probably interact with the perception of emotion and personality traits.
A possible future work would be to replicate this study by presenting a video of real coaches
performing the motion sequence. This would enable us to obtain some knowledge regarding the
complementary role of kinematics when other communicative modalities are available as performed in
[Naumann et al. 2009]. The relative contributions of the different body parts to the social perceptions
would be an additional interesting research question.
Another aspect to be considered is the asexual aspect of the 3D manikin: the virtual coach will be
sexualized to fit as much as possible a real coach condition. The additional gender information might
influence the inference made by observers. Again, replicating this study by presenting a video of real
coaches would provide some information regarding the possible influence of gender. The proportions of
the manikin might also be changed to investigate how anthropomorphism impact impression
formations. A last aspect, which impaired the ability of observers to perceive emotions and personality
traits through movement qualities, is the spontaneous nature of our corpus: not intended to be
explicitly communicated, social information leaks through kinematics. While accuracy is reduced, this
methodological choice enabled us to collect more ecological data. This will ensure the kinematics
patterns of our virtual coach are realistic and are not stereotyped.
The specific use of learned fitness movements has implications regarding the existing literature on
movement qualities. The study of everyday movements such as walking or drinking induces the collect
of more automatic motions perhaps less subject to context (i.e., emotions) and stylistic (i.e., personality)
variations. However, performing a movement sequence predefined in its form might have induced
another form of rigidity and limit the kinematic variability. An interesting future research project
would be to characterize these sources of rigidity and study how they interplay with other sources of
variabilities (e.g., emotions).
Overall, this paper provides some insights for the design of an interactive virtual fitness coach.
While varying movement impulsivity will drive most of the expressiveness attributions of the coach,
motivation and agreeableness are the other two interesting and expressive variables to play with to
adapt our coach movement qualities to the user. Using modern methods for the modeling of stylized
movements [Tilmanne and Dutoit 2012], future studies will enable us to study the influence of these
various attributes conveyed through movement qualities on energy expenditure and engagement in
the fitness task. The relevance of studying the communication of these various social attributes (i.e.,
emotions and personality) was based on the idea that a virtual coach conveying positive expressive
attributes would be more enjoyable and as a consequence foster home exercise. This assumption
should be investigated in future researches based on longitudinal experiments of virtual coach
practicing.
REFERENCES
Nancy Alvarado. 1997. Arousal and Valence in the Direct Scaling of Emotional Response to Film Clips. Motiv. Emot. 21, 4
(December 1997), 323348. DOI:http://dx.doi.org/10.1023/A:1024484306654
Nalini Ambady, Frank J. Bernieri, and Jennifer A. Richeson. 2000. Toward a histology of social behavior: Judgmental accuracy
from thin slices of the behavioral stream. In Mark P. Zanna, ed. Advances in Experimental Social Psychology.
Academic Press, 201271.
Betsy App, Daniel N. McIntosh, Catherine L. Reed, and Matthew J. Hertenstein. 2011. Nonverbal channel use in
communication of emotion: How may depend on why. Emotion 11, 3 (2011), 603617.
DOI:http://dx.doi.org/10.1037/a0023164
Anthony P. Atkinson, Winand H. Dittrich, Andrew J. Gemmell, and Andrew W. Young. 2004. Emotion perception from dynamic
and static body expressions in point-light and full-light displays. Perception 33, 6 (2004), 717 746.
DOI:http://dx.doi.org/10.1068/p5096
Avi Barliya, Lars Omlor, Martin A. Giese, Alain Berthoz, and Tamar Flash. 2013. Expression of emotion in the kinematics of
locomotion. Exp. Brain Res. 225, 2 (March 2013), 159176. DOI:http://dx.doi.org/10.1007/s00221-012-3357-4
Frank J. Bernieri, John S. Gillis, Janet M. Davis, and Jon E. Grahe. 1996. Dyad rapport and the accuracy of its judgment across
situations: A lens model analysis. J. Pers. Soc. Psychol. 71, 1 (1996), 110129. DOI:http://dx.doi.org/10.1037/0022-
3514.71.1.110
Diane S. Berry and Jane Sherman Hansen. 2000. Personality, Nonverbal Behavior, and Interaction Quality in Female Dyads.
Pers. Soc. Psychol. Bull. 26, 3 (March 2000), 278292. DOI:http://dx.doi.org/10.1177/0146167200265002
Ray L. Birdwhistell. 1972. Kinesics and Context: Essays on Body Motion Communication. Am. J. Psychol. 85, 3 (September
1972), 441. DOI:http://dx.doi.org/10.2307/1420845
Timothy A. Brown. 2006. Confirmatory Factor Analysis for Applied Research, Guilford Press.
Egon Brunswik. 1956. Perception and the Representative Design of Psychological Experiments, University of California Press.
Peter E. Bull. 1987. Posture and gesture, Elmsford, NY, US: Pergamon Press.
Donald T. Campbell and Donald W. Fiske. 1959. Convergent and discriminant validation by the multitrait-multimethod matrix.
Psychol. Bull. 56, 2 (1959), 81105. DOI:http://dx.doi.org/10.1037/h0046016
Antonio Camurri, Ingrid Lagerlöf, and Gualtiero Volpe. 2003. Recognizing emotion from dance movement: comparison of
spectator recognition and automated techniques. Int. J. Hum.-Comput. Stud. 59, 12 (July 2003), 213225.
DOI:http://dx.doi.org/10.1016/S1071-5819(03)00050-8
William H. Cooper. 1981. Ubiquitous halo. Psychol. Bull. 90, 2 (September 1981), 218244. DOI:http://dx.doi.org/10.1037/0033-
2909.90.2.218
Paul T. Costa and Robert R. McCrae. 1992. Four ways five factors are basic. Personal. Individ. Differ. 13, 6 (June 1992), 653
665. DOI:http://dx.doi.org/10.1016/0191-8869(92)90236-I
Elizabeth A. Crane and M. Melissa Gross. 2013. Effort-Shape Characteristics of Emotion-Related Body Movement. J. Nonverbal
Behav. 37, 2 (June 2013), 91105. DOI:http://dx.doi.org/10.1007/s10919-013-0144-2
Geoff Cumming and Sue Finch. 2005. Inference by Eye: Confidence Intervals and How to Read Pictures of Data. Am. Psychol.
60, 2 (2005), 170180. DOI:http://dx.doi.org/10.1037/0003-066X.60.2.170
ward Physical
Activity. Am. J. Health Behav. 32, 6 (November 2008), 570582. DOI:http://dx.doi.org/10.5993/AJHB.32.6.2
Nele Dael, Martijn Goudbeek, and K.R. Scherer. 2013. Perceived gesture dynamics in nonverbal expression of emotion.
Perception 42, 6 (2013), 642 657. DOI:http://dx.doi.org/10.1068/p7364
Nele Dael, Marcello Mortillaro, and Klaus R. Scherer. 2012. The Body Action and Posture Coding System (BAP): Development
and Reliability. J. Nonverbal Behav. 36, 2 (June 2012), 97121. DOI:http://dx.doi.org/10.1007/s10919-012-0130-0
Richard J. Davidson, 1943-, Klaus R. (Klaus Rainer) Scherer, and H. Hill Goldsmith. 2003. Handbook of affective sciences,
Oxford ; New York : Oxford University Press.
Cecily Dell. 1977. A primer for movement description using effort-shape and supplementary concepts, Dance Notation Bureau,
Center for Movement Research and Analysis, Bureau Press.
Edith Van Dyck, Pieter-Jan Maes, Jonathan Hargreaves, Micheline Lesaffre, and Marc Leman. 2013. Expressing Induced
Emotions Through Free Dance Movement. J. Nonverbal Behav. 37, 3 (September 2013), 175190.
DOI:http://dx.doi.org/10.1007/s10919-013-0153-1
Andrew D. Engell, James V. Haxby, and Alexander Todorov. 2007. Implicit trustworthiness decisions: automatic coding of face
properties in the human amygdala. J. Cogn. Neurosci. 19, 9 (September 2007), 15081519.
DOI:http://dx.doi.org/10.1162/jocn.2007.19.9.1508
Betty H. La France, Alan D. Heisel, and Michael J. Beatty. 2004. Is there empirical evidence for a nonverbal profile of
extraversion?: a metaanalysis and critique of the literature. Commun. Monogr. 71, 1 (2004), 2848.
DOI:http://dx.doi.org/10.1080/03634520410001693148
Barbara L. Fredrickson. 1998. What good are positive emotions? Rev. Gen. Psychol. 2, 3 (1998), 300319.
DOI:http://dx.doi.org/10.1037/1089-2680.2.3.300
Nico H. Frijda. 1986. The Emotions, Cambridge University Press.
Peggy E. Gallaher. 1992. Individual differences in nonverbal behavior: Dimensions of style. J. Pers. Soc. Psychol. 63, 1 (1992),
133145. DOI:http://dx.doi.org/10.1037/0022-3514.63.1.133
Steven W. Gangestad, Jeffry A. Simpson, Kenneth DiGeronimo, and Michael Biek. 1992. Differential accuracy in person
perception across traits: Examination of a functional hypothesis. J. Pers. Soc. Psychol. 62, 4 (1992), 688698.
DOI:http://dx.doi.org/10.1037/0022-3514.62.4.688
Robert Gifford. 1991. Mapping nonverbal behavior on the interpersonal circle. J. Pers. Soc. Psychol. 61, 2 (1991), 279288.
DOI:http://dx.doi.org/10.1037/0022-3514.61.2.279
Howard Giles and Beth A. Le Poire. 2006. Introduction: The Ubiquity and Social Meaningfulness of Nonverbal Communication.
In The SAGE handbook of nonverbal communication. SAGE Publications.
Donald Glowinski, Nele Dael, Antonio Camurri, Gualtiero Volpe, Marcello Mortillaro, and Klaus Scherer. 2011. Toward a
Minimal Representation of Affective Gestures. IEEE Trans. Affect. Comput. 2, 2 (April 2011), 106118.
DOI:http://dx.doi.org/10.1109/T-AFFC.2011.7
Michael H. Graham. 2003. Confronting multicolinearity in ecological multiple regression. Ecology 84, 11 (November 2003),
28092815. DOI:http://dx.doi.org/10.1890/02-3114
Karl Grammer, Valentina Filova, and Martin Fieder. 1997. The Communication Paradox and Possible Solutions. In Alain
Schmitt, Klaus Atzwanger, Karl Grammer, & Katrin Schäfer, eds. New Aspects of Human Ethology. Springer US, 91
120.
William G. Graziano, Meara M. Habashi, Brad E. Sheese, and Renée M. Tobin. 2007. Agreeableness, empathy, and helping: A
person × situation perspective. J. Pers. Soc. Psychol. 93, 4 (2007), 583599. DOI:http://dx.doi.org/10.1037/0022-
3514.93.4.583
James J. Gross and Oliver P. John. 1995. Facets of emotional Expressivity: Three self-report factors and their correlates.
Personal. Individ. Differ. 19, 4 (October 1995), 555568. DOI:http://dx.doi.org/10.1016/0191-8869(95)00055-B
M. Melissa Gross, Elizabeth A. Crane, and Barbara L. Fredrickson. 2012. Effort-Shape and kinematic assessment of bodily
expression of emotion during gait. Hum. Mov. Sci. 31, 1 (February 2012), 202221.
DOI:http://dx.doi.org/10.1016/j.humov.2011.05.001
M. Melissa Gross, Elizabeth A. Crane, and Barbara L. Fredrickson. 2010. Methodology for Assessing Bodily Expression of
Emotion. J. Nonverbal Behav. 34, 4 (December 2010), 223248. DOI:http://dx.doi.org/10.1007/s10919-010-0094-x
Judith A. Hall, Susan A. Andrzejewski, Nora A. Murphy, Marianne Schmid Mast, and Brian A. Feinstein. 2008. Accuracy of
across tests. J. Res. Personal. 42, 6 (December 2008), 1476
1489. DOI:http://dx.doi.org/10.1016/j.jrp.2008.06.013
Jari K. Hietanen, Jukka M. Leppänen, and Ulla Lehtonen. 2004. Perception of Emotions in the Hand Movement Quality of
Finnish Sign Language. J. Nonverbal Behav. 28, 1 (March 2004), 5364.
DOI:http://dx.doi.org/10.1023/B:JONB.0000017867.70191.68
David C. Hoaglin, Boris Iglewicz, and John W. Tukey. 1986. Performance of Some Resistant Rules for Outlier Labeling. J. Am.
Stat. Assoc. 81, 396 (December 1986), 991999. DOI:http://dx.doi.org/10.1080/01621459.1986.10478363
Ben Jackson, James A. Dimmock, Daniel F. Gucciardi, and J. Robert Grove. 2011. Personality traits and relationship
perceptions in coachathlete dyads: Do opposites really attract? Psychol. Sport Exerc. 12, 3 (June 2011), 222230.
DOI:http://dx.doi.org/10.1016/j.psychsport.2010.11.005
Gunnar Johansson. 1973. Visual perception of biological motion and a model for its analysis. Percept. Psychophys. 14, 2 (June
1973), 201211. DOI:http://dx.doi.org/10.3758/BF03212378
Oliver P. John, Eileen M. Donahue, and Robert L. Kentle. 1991. The Big Five Inventory--Versions 4a and 54. (1991).
Donald M. Johnson. 1963. Reanalysis of experimental halo effects. J. Appl. Psychol. 47, 1 (1963), 4647.
DOI:http://dx.doi.org/10.1037/h0044759
Adam Kendon. 1994. Do Gestures Communicate? A Review. Res. Lang. Amp Soc. Interact. 27, 3 (1994), 175200.
DOI:http://dx.doi.org/10.1207/s15327973rlsi2703_2
David A. Kenny. 1991. A general model of consensus and accuracy in interpersonal perception. Psychol. Rev. 98, 2 (1991), 155
163. DOI:http://dx.doi.org/10.1037/0033-295X.98.2.155
Basel Kikhia, Miguel Gomez, Lara Lorna Jiménez, Josef Hallberg, Niklas Karvonen, and Kåre Synnes. 2014. Analyzing Body
Movements within the Laban Effort Framework Using a Single Accelerometer. Sensors 14, 3 (March 2014), 5725
5741. DOI:http://dx.doi.org/10.3390/s140305725
Andrea Kleinsmith and Nadia Bianchi-Berthouze. 2013. Affective Body Expression Perception and Recognition: A Survey.
IEEE Trans. Affect. Comput. 4, 1 (January 2013), 1533. DOI:http://dx.doi.org/10.1109/T-AFFC.2012.16
Markus Koppensteiner. 2013. Motion cues that make an impression. Predicting perceived personality by minimal motion
information. J. Exp. Soc. Psychol. 49, 6 (November 2013), 11371143. DOI:http://dx.doi.org/10.1016/j.jesp.2013.08.002
Markus Koppensteiner and Karl Grammer. 2010. Motion patterns in political speech and their influence on personality ratings.
J. Res. Personal. 44, 3 (June 2010), 374379. DOI:http://dx.doi.org/10.1016/j.jrp.2010.04.002
Amanda Kraha, Heather Turner, Kim Nimon, Linda Reichwein Zientek, and Robin K. Henson. 2012. Tools to support
interpreting multiple regression in the face of multicollinearity. Front. Psychol. 3 (2012), 44.
DOI:http://dx.doi.org/10.3389/fpsyg.2012.00044
Rudolf von Laban. 1950. The mastery of movement on the stage., London: MacDonald & Evans.
Frank J. Landy, Janet L. Barnes-Farrell, Robert J. Vance, and James W. Steele. 1980. Statistical Control of Halo Error in
Performance Ratings. J. Appl. Psychol. 65, 5 (October 1980), 501506.
Jacqyln A. Levy and Marshall P. Duke. 2003. The Use of Laban Movement Analysis in the Study of Personality, Emotional
Individ. Differ. Res.
1, 1 (April 2003), 3963.
Shane Lowe and Gearóid ÓLaighin. 2012. The age of the virtual trainer. Procedia Eng. 34 (2012), 242247.
DOI:http://dx.doi.org/10.1016/j.proeng.2012.04.042
Geoff Luck, Suvi Saarikallio, Birgitta Burger, Marc R. Thompson, and Petri Toiviainen. 2010. Effects of the Big Five and
musical genre on music-induced movement. J. Res. Personal. 44, 6 (December 2010), 714720.
DOI:http://dx.doi.org/10.1016/j.jrp.2010.10.001
Elizabeth J. Lyons, Deborah F. Tate, Dianne S. Ward, Kurt M. Ribisl, J. Michael Bowling, and Sriram Kalyanaraman. 2014.
Engagement, enjoyment, and energy expenditure during active video game play. Health Psychol. Off. J. Div. Health
Psychol. Am. Psychol. Assoc. 33, 2 (February 2014), 174181. DOI:http://dx.doi.org/10.1037/a0031947
John McGowan and John Gormly. 1976. Validation of personality traits: A multicriteria approach. J. Pers. Soc. Psychol. 34, 5
(1976), 791795. DOI:http://dx.doi.org/10.1037/0022-3514.34.5.791
Marco de Meijer. 1989. The contribution of general features of body movement to the attribution of emotions. J. Nonverbal
Behav. 13, 4 (December 1989), 247268. DOI:http://dx.doi.org/10.1007/BF00990296
Joann Montepare, Elissa Koff, Deborah Zaitchik, and Marilyn Albert. 1999. The Use of Body Movements and Gestures as Cues
to Emotions in Younger and Older Adults. J. Nonverbal Behav. 23, 2 (June 1999), 133152.
DOI:http://dx.doi.org/10.1023/A:1021435526134
Joann M. Montepare and Heidi Dobish. 2003. The Contribution of Emotion Perceptions and Their Overgeneralizations to Trait
Impressions. J. Nonverbal Behav. 27, 4 (December 2003), 237254. DOI:http://dx.doi.org/10.1023/A:1027332800296
Carey K. Morewedge and Daniel Kahneman. 2010. Associative processes in intuitive judgment. Trends Cogn. Sci. 14, 10
(October 2010), 435440. DOI:http://dx.doi.org/10.1016/j.tics.2010.07.004
Shinichi Nakagawa. 2004. A farewell to Bonferroni: the problems of low statistical power and publication bias. Behav. Ecol. 15,
6 (November 2004), 10441045. DOI:http://dx.doi.org/10.1093/beheco/arh107
Laura P. Naumann, Simine Vazire, Peter J. Rentfrow, and Samuel D. Gosling. 2009. Personality Judgments Based on Physical
Appearance. Pers. Soc. Psychol. Bull. (September 2009). DOI:http://dx.doi.org/10.1177/0146167209346309
L. Ouss, S. Carton, Roland Jouvent, and D. Widloêcher. émotions différentielle
: exploration de la qualification verbale des émotions. Encéphale 16, 6 (1990), 453458.
Catherine Pelachaud. 2009. Studies on gesture expressivity for a virtual agent. Speech Commun. 51, 7 (July 2009), 630639.
DOI:http://dx.doi.org/10.1016/j.specom.2008.04.009
Paolo Petta, Catherine Pelachaud, and Roddy Cowie. 2011. Emotion-Oriented Systems - The Humaine Handbook, Springer.
Philippe J. Rushton, Douglas N. Jackson, and Sampo V. Paunonen. 1981. Personality: Nomothetic or idiographic? A response to
Kenrick and Stringfield. Psychol. Rev. 88, 6 (1981), 582589. DOI:http://dx.doi.org/10.1037/0033-295X.88.6.582
Odile Plaisant, Robert Courtois, Christian Réveillère, Gerald Mendelson and Olivier John. 2010. Validation par analyse
factorielle du Big Five Inventory français (BFI-Fr). Analyse convergente avec le NEO-PI-R. Annales Medico-
Psychologiques, 168, 2, (2010), 97-106.
Philip M. Podsakoff, Scott B. MacKenzie, Jeong-Yeon Lee, and Nathan P. Podsakoff. 2003. Common method biases in
behavioral research: a critical review of the literature and recommended remedies. J. Appl. Psychol. 88, 5 (October
2003), 879903. DOI:http://dx.doi.org/10.1037/0021-9010.88.5.879
Frank E. Pollick, Helena M. Paterson, Armin Bruderlin, and Anthony J. Sanford. 2001. Perceiving affect from arm movement.
Cognition 82, 2 (December 2001), B51B61. DOI:http://dx.doi.org/10.1016/S0010-0277(01)00147-0
Thomas D. Raedeke. 2007. The Relationship Between Enjoyment and Affective Responses to Exercise. J. Appl. Sport Psychol.
19, 1 (2007), 105115. DOI:http://dx.doi.org/10.1080/10413200601113638
Heidi R. Riggio and Ronald E. Riggio. 2002. Emotional Expressiveness, Extraversion, and Neuroticism: A Meta-Analysis. J.
Nonverbal Behav. 26, 4 (December 2002), 195218. DOI:http://dx.doi.org/10.1023/A:1022117500440
Claire L. Roether, Lars Omlor, Andrea Christensen, and Martin A. Giese. 2009. Critical features for the perception of emotion
from gait. J. Vis. 9, 6 (June 2009), 15. DOI:http://dx.doi.org/10.1167/9.6.15
Robert Rosenthal. 2008. Conducting Judgment Studies: In Jinni Harrigan, Robert Rosenthal, & Klaus Scherer, eds. The New
Handbook of Methods in Nonverbal Behavior Research. Oxford University Press, 199234.
Sverker Runeson and Gunilla Frykholm. 1983. Kinematic specification of dynamics as an informational basis for person-and-
action perception: Expectation, gender recognition, and deceptive intention. J. Exp. Psychol. Gen. 112, 4 (1983), 585
615. DOI:http://dx.doi.org/10.1037/0096-3445.112.4.585
James A. Russell and Albert Mehrabian. 1977. Evidence for a three-factor theory of emotions. J. Res. Personal. 11, 3
(September 1977), 273294. DOI:http://dx.doi.org/10.1016/0092-6566(77)90037-X
Ali .A. Samadani, Sarahjane Burton, Rob Gorbet, and Dana Kulic. 2013. Laban Effort and Shape Analysis of Affective Hand
and Arm Movements. In 2013 Humaine Association Conference on Affective Computing and Intelligent Interaction
(ACII). 343348. DOI:http://dx.doi.org/10.1109/ACII.2013.63
Klaus R. Scherer. 2010. Emotion and emotional competence: conceptual and theoretical issues for modelling agents. In
Blueprint for Affective Computing, A Sourcebook. Oxford University Press, 320.
Klaus R. Scherer. 1978. Personality inference from voice quality: The loud voice of extroversion. Eur. J. Soc. Psychol. 8, 4
(October 1978), 467487. DOI:http://dx.doi.org/10.1002/ejsp.2420080405
Klaus R. Scherer. 2013. Vocal markers of emotion: Comparing induction and acting elicitation. Comput. Speech Lang. 27, 1
(January 2013), 4058. DOI:http://dx.doi.org/10.1016/j.csl.2011.11.003
Maxine Sheets-Johnstone. 1999. The Primacy of Movement, John Benjamins Publishing.
Nao Shikanai, Misako Sawada, and Motonobu Ishii. 2013. Development of the Movements Impressions Emotions Model:
Evaluation of Movements and Impressions Related to the Perception of Emotions in Dance. J. Nonverbal Behav. 37, 2
(June 2013), 107121. DOI:http://dx.doi.org/10.1007/s10919-013-0148-y
John C. Thoresen, Quoc C. Vuong, and Anthony P. Atkinson. 2012. First impressions: Gait cues drive reliable trait judgements.
Cognition 124, 3 (September 2012), 261271. DOI:http://dx.doi.org/10.1016/j.cognition.2012.05.018
Edward L. Thorndike. 1920. A constant error in psychological ratings. J. Appl. Psychol. 4, 1 (1920), 2529.
DOI:http://dx.doi.org/10.1037/h0071663
Joëlle Tilmanne and Thierry Dutoit. 2012. Continuous Control of Style and Style Transitions through Linear Interpolation in
Hidden Markov Model Based Walk Synthesis. In Marina L. Gavrilova & C. J. Kenneth Tan, eds. Transactions on
Computational Science XVI. Lecture Notes in Computer Science. Springer Berlin Heidelberg, 3454.
Allessandro Vinciarelli and Gelareh Mohammadi. 2014. A Survey of Personality Computing. IEEE Trans. Affect. Comput. 5, 3
(July 2014), 273291. DOI:http://dx.doi.org/10.1109/TAFFC.2014.2330816
Harald G. Wallbott. 1998. Bodily expression of emotion. Eur. J. Soc. Psychol. 28, 6 (November 1998), 879896.
DOI:http://dx.doi.org/10.1002/(SICI)1099-0992(1998110)28:6<879::AID-EJSP901>3.0.CO;2-W
Harald G. Wallbott. 1985. Hand movement quality: A neglected aspect of nonverbal behavior in clinical judgment and person
perception. J. Clin. Psychol. 41, 3 (May 1985), 345359. DOI:http://dx.doi.org/10.1002/1097-
4679(198505)41:3<345::AID-JCLP2270410307>3.0.CO;2-9
Leslie A. Zebrowitz, Masako Kikuchi, and Jean-Marc Fellous. 2007. Are Effects of Emotion Expression on Trait Impressions
Mediated by Babyfaceness? Evidence From Connectionist Modeling. Pers. Soc. Psychol. Bull. (April 2007).
DOI:http://dx.doi.org/10.1177/0146167206297399
Online Appendix to:
Emotion and personality perception through full-body
movement qualities: a sport coach case study
XX, XX
A. MOVEMENT QUALITIES COMPUTATION
Times series - Descriptive
Equation
Impulsiveness (Equation 1): Time effort can be
determined as the net acceleration at the body parts over
time. Large values of net acceleration indicate sudden
movements characterized by a high impulsiveness.
, velocity of the member.
Energy (Equation 2): Weight effort can be determined as
energy at each instant (t) over time. is the
approximation of the mass of each member according to
the Winter table (Winter 2004). Large values of energy
indicate strong movements.
=
, velocity of the member.
Directness (Equation 3): Space effort is computed as the
inner product of chest and member displacement (i.e.,
right hand, left hand, right foot, left foot) trajectories.
Direct movements are thus usually characterized by a
small number of peaks.
, tangents to chest and member positions.
Smoothness (Equation 4): Flow effort is determined as
the 3D curvature for each segment for each time. The
curvature is a rapport between velocity and acceleration.
Computation of curvature gives larger values for smooth
movement and high values for jerky movements. A
negative sign is added to correspond to the label
movements.
indicate the first and second derivatives of the member
position at frame i, respectively.
Spatial Extent (Equation 5): Shape Qualities describe
the way the body is changing toward space. Spatial extent
is associated to the position of each segment according to
the center of mass at each time. Values of Expansiveness
are close to zero for dense movements and are high for
expanded movements.
D
DIx, DIy, DIz are the sum of distances between the mass center
coordinate ( ) and the n-th member coordinate ( )
at frame i.
Adapted from [Piana et al 2013, Kapadia et al 2013, Samadani et al 2013, Chen et al 2011].
@ 2010 ACM 1544-3558/2010/05-ART1 $15.00
DOI 10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000
1
B. FRENCH VERSION OF THE EVALUATED DIMENSIONS
Movement qualities
Left anchora
Right anchorb
English
French
English
French
English
French
Smoothness
Fluidité
Abrupt
Abrupt
Smooth
Fluide
Directness
Directivité
Indirect
Indirect
Direct
Direct
Energy
Energie
Limp
Mou
Energetic
Energique
Impulsivity
Impulsivité
Restrained
Retenu
Impulsive
Impulsif
Spatial extension
Extension
spatiale
Restricted
Restreint
Vast
Etendu
Emotion
Left anchora
Right anchorb
English
French
English
French
English
French
Valence
Valence
Negative
emotion
Emotion
negative
Positive emotion
Emotion
positive
Intensity
Intensité
Low intensity
Faible
intensité
Strong intensity
Forte intensité
Dominance
Dominance
Subdued
Soumise
Dominant
Dominante
Motivation
Motivation
Without
motivation
Sans
motivation
Strongly
motivated
Très motivé
Stress
Stress
Relaxed
Pas stressé
Stressed
Très stressé
Personality
Left anchora
Right anchorb
English
French
English
French
English
French
Extraversion
Extraverti
Introverted
Introverti
Extroverted
Extraverti
Openness
Aventureux
Humdrum
Routinier
Adventurous
Aventureux
Neuroticism
Stable
émotionellement
Neurotic
Névrosé
Stable emotionally
Stable
Conscientiousness
Consciencieux
Careless
Négligent
Conscientious
Consciencieux
Agreeableness
Chaleureux
Disagreeable
Désagréable
Warm
Chaleureux
a Score = 1
b Score = 10
Italic scales inverted from French to English
@ 2010 ACM 1544-3558/2010/05-ART1 $15.00
DOI 10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000C.
