Journal of Experimental Psychology:

General
Probabilistic Learning of Emotion Categories
Rista C. Plate, Adrienne Wood, Kristina Woodard, and Seth D. Pollak
Online First Publication, December 20, 2018. http://dx.doi.org/10.1037/xge0000529

CITATION
Plate, R. C., Wood, A., Woodard, K., & Pollak, S. D. (2018, December 20). Probabilistic Learning of
Emotion Categories. Journal of Experimental Psychology: General. Advance online publication.
http://dx.doi.org/10.1037/xge0000529
 Journal of Experimental Psychology: General
© 2018 American Psychological Association 2018, Vol. 1, No. 999, 000
0096-3445/18/$12.00 http://dx.doi.org/10.1037/xge0000529

Probabilistic Learning of Emotion Categories

Rista C. Plate, Adrienne Wood, Kristina Woodard, and Seth D. Pollak
University of Wisconsin-Madison

Although the configurations of facial muscles that humans perceive vary continuously, we often represent
emotions as categories. This suggests that, as in other domains of categorical perception such as speech
and color perception, humans become attuned to features of emotion cues that map onto meaningful
thresholds for these signals given their environments. However, little is known about the learning
processes underlying the representation of these salient social signals. In Experiment 1 we test the role
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of statistical distributions of facial cues in the maintenance of an emotion category in both children (6 – 8
This document is copyrighted by the American Psychological Association or one of its allied publishers.

years old) and adults (18 –22 years old). Children and adults learned the boundary between neutral and
angry when provided with explicit feedback (supervised learning). However, after we exposed partici-
pants to different statistical distributions of facial cues, they rapidly shifted their category boundaries for
each emotion during a testing phase. In Experiments 2 and 3, we replicated this finding and also tested
the extent to which learners are able to track statistical distributions for multiple actors. Not only did
participants form actor-specific categories, but the distributions of facial cues also influenced partici-
pants’ trait judgments about the actors. Taken together, these data are consistent with the view that the
way humans construe emotion (in this case, anger) is not only flexible, but reflects complex learning
about the distributions of the myriad cues individuals experience in their social environments.

Keywords: emotion categorization, supervised learning, unsupervised learning

Faces offer rich, salient sources of information for guiding judg- flexibly update facial expression categories based upon their sensory
ments and behaviors. Human faces contain over 40 muscles that and social experiences (Barrett, 2017; Clore & Ortony, 2013; Pollak
contract and relax in patterns, producing configurations that individ- & Kistler, 2002). Yet, little data exists about how children could come
uals use to infer the mental states of others. In principle, a human face to synthesize the vast range of facial movements in a coherent way
can generate approximately 16 million muscular combinations. Be- that represents functionally meaningful patterns from their environ-
cause of this staggering amount of information, individuals must learn ments.
to attend to and ignore an extraordinary amount of information from The current experiments test whether probabilistic information
their environments in order to successfully and rapidly understand the in environments influences how individuals make judgments about
communicative signals in emotion displays. The classical view in other people’s emotions. In addition, we examine whether learning
human development holds that initial perceptual categories of facial exerts differential levels of influence earlier versus later in devel-
expression are evolutionarily preserved (Ekman, 1992; Grossmann, opment by contrasting the performance of children and adults.
2015; Shariff & Tracy, 2011). The contrasting hypothesis is that Plasticity might be expected in children, but we reasoned that if
humans can detect and track variations in the distribution of signals in such learning effects exist in mature individuals, it would suggest
environments, and use this probabilistic information to construct and that emotion categories remain fluid.

Health through grant R01MH61285 to Seth D. Pollak and by a core grant

Rista C. Plate, Adrienne Wood, Kristina Woodard, and Seth D. Pollak, to the Waisman Center from the National Institute of Child Health and
Department of Psychology, University of Wisconsin-Madison. Human Development (U54 HD090256). Rista C. Plate was supported by a
Rista C. Plate and Adrienne Wood contributed equally to the research National Science Foundation Graduate Research Fellowship (DGE-
and manuscript. All authors developed the study concept and contributed to 1256259) and the Richard L. and Jeanette A. Hoffman Wisconsin Distin-
study design. Rista C. Plate and Adrienne Wood collected the data and guished Graduate Fellowship. Adrienne Wood was supported by an Emo-
performed data analysis and interpretation. Rista C. Plate and Adrienne tion Research Training Grant (T32MH018931-24) from the National
Wood drafted the manuscript. Kristina Woodard contributed to the man- Institute of Mental Health. We thank the individuals who participated in
uscript for Experiments 2 and 3. Seth D. Pollak provided critical revisions. these experiments, and the research assistants, particularly Teresa Turco,
All authors approved the final version of the manuscript for submission. who helped with data collection. We also thank Chuck Kalish, Paula
The computerized tasks, stimuli, data, analysis scripts and output (in R) Niedenthal, and Jenny Saffran for their feedback on an earlier version of
are available online (https://osf.io/ycb3q/). Ideas and data from Study 1 this article.
were presented at Society for Personality and Social Psychology (2018, Correspondence concerning this article should be addressed to Seth D.
poster), Society for Affective Science (2018, flash talk). Studies 2 and 3 Pollak, Department of Psychology, University of Wisconsin-Madison,
were presented at the meeting of the Cognitive Science Society (2018, 1202 West Johnson Street, Madison, WI 53706. E-mail: seth.pollak@
poster). This research was supported by the National Institute of Mental wisc.edu

1
 2 PLATE, WOOD, WOODARD, AND POLLAK

From infancy, faces capture human attention, and facial config- learning, namely supervised and unsupervised learning (Love,
urations are often represented as reflecting emotion categories 2002). One way that individuals might learn about emotion cate-
(Cong et al., 2018; Diamond & Carey, 1986; Frank, Vul, & gories and category boundaries is through explicit instruction.
Johnson, 2009; Pollak, Messner, Kistler, & Cohn, 2009; Pourtois, Though not immediately obvious, some societies do explicitly
Schettino, & Vuilleumier, 2013; Russell & Bullock, 1986; Suss- teach children how to categorize emotions. For example, many
kind, Littlewort, Bartlett, Movellan, & Anderson, 2007). These North American preschools display posters depicting what differ-
categories allow observers to respond to faces quickly by ignoring ent “feelings” ought to look like. In other forms of explicit teach-
variability in facial movements to make general judgments about ing, adults may label emotions for children during nonverbal
emotion signaling (Campanella, Quinet, Bruyer, Crommelinck, & expressions (“Look, Johnny is crying, you made him feel sad”;
Guerit, 2002; Etcoff & Magee, 1992). For example, observers will Ahn, 2005; Gordon, 1991; Pollak & Thoits, 1989). We can think
not perceive a gradual transition from low to high muscle activa- of this as supervised learning, as it involves incorporating feedback
tion as a linear, continuous change, but will instead perceive a about whether the observer’s initial interpretation of a facial dis-
qualitative shift in the facial configuration at the point where a play was correct. Supervised learning likely plays a role in emotion
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

person appears to have become “angry” (Calder, Young, Perrett, understanding across development. As one example, the extent and
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Etcoff, & Rowland, 1996; Campanella et al., 2002; Pollak & circumstances under which children experience shame versus guilt
Kistler, 2002; Wood, Lupyan, Sherrin, & Niedenthal, 2016). The are tied to overt aspects of parenting practices and socialization
location of this shift is known as the category boundary, and it has (Eisenberg, 2000).
important implications for how the observer will respond to an- However, much of human learning is thought to be unsupervised
other person’s emotion signaling. (Fisher, Pazzani, & Langley, 2014). Whereas supervised learning
Despite ample behavioral (Etcoff & Magee, 1992) and neural relies on information being directly provided to the learner, unsu-
(Campanella et al., 2002) evidence that emotions can be repre- pervised learning happens in the absence of explicit information or
sented categorically, it is not well understood how these categories feedback. During unsupervised learning, the learner extracts sta-
are acquired, whether these categories are stable, and whether tistical distributions from their environment to acquire meaningful
these categories reflect the statistical distributions of the emotion categories. Attention to statistical distributions is essential in do-
cues that individuals observe. The amount of emotion-related mains in which it would be infeasible to acquire all needed
information to be attended to and ignored is so vast that rudimen- information through explicit, supervised learning. For example,
tary categories may be innately preserved in neural architecture. unsupervised learning aids in aspects of language acquisition such
However, recent research on domain-general categorization sug- as phonemic discrimination and word segmentation (Maye,
gests that category representations emerge “ad hoc” and shift to Werker, & Gerken, 2002; Saffran, Newport, & Aslin, 1996), where
meet the demands of the environment (Casasanto & Lupyan, 2015; children could not be provided with sufficient explicit labeling of
Levari et al., 2018). Emotion categories may need to be malleable relevant distinctions between stimuli.
to allow individuals to adjust flexibly to variations in cultural and The relative contribution of supervised and unsupervised learn-
situational norms (Aviezer, Trope, & Todorov, 2012; Marchi & ing has been examined for artificial category formation of novel
Newen, 2015; Niedenthal, Rychlowska, & Wood, 2017). While all stimuli in children and adults (Kalish, Rogers, Lang, & Zhu, 2011;
emotions may require such flexibility, the focus of the present Kalish, Zhu, & Rogers, 2015). For example, in one study, 4 – 8-
investigation is on anger. Categorical representations of facial year-old children updated their category boundaries via unsuper-
anger may be particularly dependent on the social environment vised learning even when their original category boundaries were
given the potential costs of failing to detect threat signals (Pollak established through supervised learning (Kalish et al., 2015). In a
& Kistler, 2002). similar experiment, adults changed their previously formed cate-
To date, the role of learning during repeated exposure to varying gory boundaries based on exposure to statistical distributions that
intensities of facial emotions has not been examined. Additionally, differed from the original, labeled (i.e., supervised) distribution
it is unknown whether children and adults would differ in emotion (Kalish et al., 2011). However, these previous studies, by design,
category flexibility to such exposure. On the one hand, children are created completely novel stimuli, so that participants do not enter
characterized by flexibility in their learning, while adults tend to the experiment with a priori knowledge or expectations about how
rely more heavily on top-down processes (e.g., Gopnik et al., to categorize each exemplar. Scientists understand less about how
2017). This reasoning might suggest that children would be more these general learning mechanisms might operate on purportedly
likely to shift category boundaries in a brief learning episode. On privileged biological stimuli, such as human facial configurations
the other hand, adults have greater experience with interpreting that convey emotion. While there is evidence to suggest that
and responding to facial configurations across multiple contexts. unsupervised exposure to facial expressions sharpens already-
This expertise may allow them to be more successful at integrating existing category representations of emotions (Huelle, Sack, Broer,
contextual cues that they should shift their category boundaries. Komlewa, & Anders, 2014), it is unclear if unsupervised learning
Therefore, it is unknown whether statistical distributions of emo- can also shift those category representations within a feature space.
tion cues can influence category boundaries, and, if so, whether
such flexibility operates robustly in both childhood and adulthood.
Present Experiments
In Experiment 1, we tested whether probabilistic information
How Might People Update Emotion Categories?
influences an individual’s representation of emotions. We pre-
Research on categorization in children and adults—and in ma- dicted that category boundaries between emotions would be mal-
chine learning applications—identifies two overarching types of leable and would reflect the distribution of facial configurations
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 3

encountered. We expect that statistical learning plays a role in the 105 adults (75 female, 23 male, 7 unreported gender; age range ⫽
continual updating of category boundaries for all facial expres- 18 –22 years, Mage ⫽ 18.61 years, SDage ⫽ 0.95 years) participated
sions. In the present work we test this idea using anger, because in this experiment. Children ages 6 – 8-years-old were chosen
prior evidence suggests long-term social environments contribute because they have shown sensitivity to supervised and unsuper-
to individual differences in category boundaries for anger (Pollak vised learning in previous research using novel stimuli (Kalish et
& Kistler, 2002). If participants are able to use distributional al., 2015); additionally we aimed for 30 participants per condition
information of perceptual cues to alter their anger category bound- per age group (three between-subjects conditions), which exceeded
ary, even in a brief timeframe, this would suggest that the func- sample sizes of this previous research. Two additional children
tional boundaries between facial configurations are flexible, and were excluded due to a program error. Children were recruited
that this category can update according to social contexts. Testing from the community in a large Midwestern city (9% African
both children and adults, as has been done in research on super- American, 3% Asian American, 1% Hispanic, 5% multiracial,
vised and statistical learning of novel categories (Kalish et al., 79% White, 2% did not report race). Adults were undergraduates
2011, 2015), allowed us to test for developmental differences in at a large university in the same city who participated for course
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

emotion category malleability. If both children and adults update credit. Adult participants and parents of child participants gave
This document is copyrighted by the American Psychological Association or one of its allied publishers.

their categories based on extant cues, then we will have evidence informed consent; children gave verbal assent. Parents received
that emotion representations remain flexible into maturity. Exper- $20 for their time and children chose a prize for their participation.
iments 2 and 3 extend Experiment 1 by examining the robustness The Institutional Review Board approved the research.
of the learning mechanisms in adults, and begin to assess whether Face stimuli. Stimuli included images of facial emotions for
shifts in these learned categories have functional significance in one female model (Model #10) selected from the MacArthur
terms of participants’ interpretation of facial signals. Network Face Stimuli Set (Tottenham et al., 2009). The selected
model’s facial expressions were morphed in increments of 5%
Experiment 1 from a 100% neutral expression (i.e., 0% angry) to a 100% angry
expression of emotion, creating 21 equally spaced images (stimuli
For all experiments, we report how we determined our sample
from Gao & Maurer, 2009; Figure 1). Stimuli were presented with
size, data exclusions, all manipulations, and all measures. The
PsychoPy (v1.83.04).
experiment, dataset, R code, and stimuli files for all experiments
Procedure. Participants completed the task with an experi-
are available online (https://osf.io/ycb3q/).
menter in a laboratory testing room. Instructions were presented on
the screen and, for children, read by the experimenter. The exper-
Method
iment included three phases. The goal of the introductory phrase
Participants. Ninety-one children (44 female, 47 male; age was to expose participants to the model and procedure, and to
range ⫽ 6 – 8 years, Mage ⫽ 7.52 years, SDage ⫽ 0.92 years) and allow participants to practice the categorization task. The goal of

Figure 1. Experiment 1 stimuli and sampling strategies used in the supervised learning and testing phases.
Only a subset of the 21 morphed neutral-to-angry stimuli are represented in this figure. Percentages indicate the
percent angry expression present in the morphed images: In other words, the “0%” image contained 100%
neutral expression and 0% angry expression, while the “100%” image contained 0% neutral expression and
100% angry expression. The top gray rectangles illustrate the range of morphs (from 20% to 45% angry and 55%
to 80% angry) used in the supervised learning phase, and the rectangles below illustrate the percent morphs (from
0% angry to 100% angry) used in each of the three conditions (representing shifted stimuli distributions) during
the testing phase. Please note that the model depicted here is model #03 from Tottenham et al. (2009); model
#10 was used in study (not depicted due to copyright regulations). See the online article for the color version of
this figure.
 4 PLATE, WOOD, WOODARD, AND POLLAK

the supervised learning phase was to train participants to a com- 100% angry (70% morph omitted to create category boundary). In
mon category boundary. The goal of the testing phase was to the neutral shifted condition (n ⫽ 30 children, 33 adults), partic-
assess whether participants shift the category boundary acquired ipants saw six repetitions of morphs ranging from 0% angry to
during the supervised learning phase based on the statistical dis- 60% angry (30% morph omitted to create category boundary).
tribution of stimuli. Participants did not receive feedback during this phase. Stimuli
Introductory phase. Participants saw an image of the model appeared in random order; however, each individual morph was
(“Jane”) and were told, “Just like everyone, sometimes Jane feels seen once before repeating.
upset and sometimes Jane feels calm. Today we need your help
figuring out if Jane is upset or calm.” Participants were then
Results
presented with an image of the model and a computerized slider
that allowed them to see all morphed facial expressions across the Analytic plan. In the results that follow, we first report anal-
entire range from neutral to angry. Of note, the emotion labels, yses of the supervised learning phase. A logistic generalized linear
“angry” or “neutral” were never provided to participants in this mixed-effect modeling approach allowed us to examine the rela-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

task. tive steepness of children’s and adults’ category boundaries, which

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Next, participants saw an image of a red-colored room contain- indicated how precise and categorical their representations of calm
ing boxing equipment and a 100% angry morphed image of Jane and upset became during the brief supervised learning phase. To
and were told, “When Jane is feeling upset she likes to go to the ensure that children and adults were able to do the task equally
red room and practice boxing.” Participants then saw an image of well, we also looked at overall accuracy. Next, we used the same
a blue-colored room containing a chair and books and a 0% angry logistic regression approach to analyze participant judgments in
(100% neutral) morphed image of Jane, with the instructions, the testing phase. In particular, we examined whether participants’
“When Jane is feeling calm she likes to go to the blue room and category boundaries shifted as a function of experimental condition
read a book. We need you to help us figure out what room Jane (unshifted, anger shifted, and neutral shifted), and whether children
should go to. If you think Jane is feeling upset, click the red button and adults differed in their sensitivity to the shifted distributions.
(left arrow) to put her in the red room. If you think Jane is feeling Results from the logistic regression are reported with odds ratios
calm, click the blue button (right arrow) to put her in the blue (OR), which indicate by how much the odds of an upset response
room.” Adults used the arrow keys to respond; children used the increase (OR ⬎ 1) or decrease (OR ⬍ 1) with a unit increase in the
arrow keys with red/blue stickers corresponding to the location of predictor variable (e.g., OR ⫽ 2 means the odds of an “upset”
the room. The side of the screen where each room appeared was response double). All analyses were conducted in the R environment
counterbalanced between participants. (R Development Core Team, 2008) with the lme4 package, Version
Finally, to ensure participants understood the task, they com- 1.1–15 (Bates, Mächler, Bolker, & Walker, 2015).
pleted six practice trials with labeled faces (e.g., “Jane is feeling Supervised learning phase.
upset. Which room should she go to?”), and the response options Individuals established a category boundary via supervised
(i.e., red room and blue room) appeared in the bottom right and left learning. First, we assessed whether children and adults were
corners. If the participant made an error, they received feedback able to learn the category boundary that we taught during the
(“Incorrect! Please try again”) and the trial was repeated until the supervised phase. Since the responses were dichotomous, we used
participant responded correctly (when the participant responded logistic generalized linear mixed-effects models, regressing re-
correctly, they received the feedback, “Correct!”). The morphs sponse (calm ⫽ 0, upset ⫽ 1) on the interaction between centered
presented during the introductory phase that were 0%, 10%, and Age Group and mean-centered Percent Anger (ranging from 20 for
20% angry were labeled as calm, and the morphs that were 80%, mostly neutral facial expression to 80 for mostly angry facial
90%, and 100% angry were labeled as upset. All participants saw expression, at 5-unit intervals), with a by-participant random slope
the same morphs, but the order was randomized between partici- for Percent Anger. The main effect of Percent Anger was signif-
pants. icant, with the odds of an image being categorized as upset
Supervised learning phase. All participants then completed increasing by a factor of 6.77 with every 10% increase in anger
the same 24 trials with “correct”/“incorrect” feedback in random- intensity, b ⫽ 1.91, ␹2(1) ⫽ 640.53, p ⬍ .001.
ized order. Stimuli consisted of two repetitions of morphs ranging Children and adults did not differ in the formation of catego-
from 20% angry to 80% angry in 5% increments. The 50% morph ries via supervised learning. The main effect of Age Group was
was omitted in order to create a category boundary at the midpoint not significant, indicating that age did not alter the overall prob-
(see Figure 1). Correct responses were defined for stimuli less than ability of someone categorizing a face as “upset”, b ⫽ .10, ␹2(1) ⫽
50% angry as “blue room” and stimuli greater than 50% angry as 0.63, p ⫽ .427, odds ratio (OR) ⫽ 1.11. The interaction between
“red room.” Note that in this phase and the testing phase, the words Age Group and Percent Anger trended toward statistical signifi-
calm and upset did not appear. cance: Shifting 10% in percent anger present in a face increased
Testing phase. The testing phase directly followed the super- the odds of it being categorized as upset by a factor of 1.25 more
vised phase and included 72 trials. Participants were randomly for adults compared to children, b ⫽ 0.22, ␹2(1) ⫽ 3.32, p ⫽ .069.
assigned to one of three between-subjects conditions (see Figure In other words, the slope of the category boundary on supervised
1). In the unshifted condition, participants (n ⫽ 31 children, 35 learning trials was marginally steeper (more categorical) for adults
adults) saw six repetitions of each morph ranging from 20% angry compared to children (see Figure 2). This marginal difference in
to 80% angry (50% morph omitted to create category boundary). category boundary steepness notwithstanding, children and adults
In the anger shifted condition (n ⫽ 30 children, 37 adults), par- seemed to learn the explicitly taught calm and upset categories and
ticipants saw six repetitions of morphs ranging from 40% angry to performed comparably.
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 5
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 2. Experiment 1 supervised learning phase: Model predictions and participant-level data. Lines are
point estimates from logistic mixed-effects models with the interaction between Age Group and Percent Anger,
and lower-order effects. Error bands represent standard errors of the point estimates. Points are individual
participants’ proportion of “upset” responses at a given facial expression morph value. See the online article for
the color version of this figure.

We next determined whether children and adults differed in way interaction between Percent AngerⴱAge GroupⴱExperimental
their overall performance in the supervised learning phase. We Condition dummy variables, plus all lower-order fixed effects and
regressed participants’ proportion of correct responses on Age a by-participant random slope for Percent Anger.
Group (coded as children ⫽ ⫺.5, adults ⫽ .5). Children (accuracy Exposure to shifted distributions altered participants’ category
M ⫽ .89, SD ⫽ .10) and adults (accuracy M ⫽ .90, SD ⫽ .08) did boundaries. Our primary question of interest was whether par-
not differ significantly in their overall proportion of correct re- ticipants would override explicitly learned emotion category
sponses, b ⫽ 0.01, F(1, 390) ⫽ 2.37, p ⫽ .125, partial ␩2 ⫽ 0.006. boundaries based upon mere passive exposure to a new statistical
Testing phase. Next, we examined the effect of shifting the distribution of facial input. Participants did shift their emotion
statistical distribution of facial images on participants’ categoriza- categories during the testing phase in response to the intensities of
tion in the testing phase. We used the same generalized linear emotions they encountered. The main effect of Experimental Con-
mixed-effect model as with the supervised phase data, with the dition was significant, ␹2(2) ⫽ 453.66, p ⬍ .001 (see Figure 3).
addition of two dummy variables to compare responses in the Dummy coded parameters indicated that the anger-response
neutral shifted and anger shifted conditions to the unshifted con- threshold was significantly earlier (i.e., occurred at a lower per-
dition. The full model regressed participant responses on a three- centage morph) in the morph continuum for participants in the
 6 PLATE, WOOD, WOODARD, AND POLLAK
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. Experiment 1 testing phase: Model predictions and participant-level data. Lines are point estimates
from logistic mixed-effects models with the three-way interaction between Age Group, Condition (dummy coded
for the sake of graphing), and Percent Anger, and all lower-order effects. Error bands represent standard errors
of the point estimates. Points are individual participants’ proportion of “upset” responses at a given morph value.
Vertical lines indicate the implied category boundary for each condition. Horizontal bars at the top indicate the
range of morphed images to which participants in each of the three conditions were exposed. See the online
article for the color version of this figure.

neutral shifted condition, b ⫽ 2.07, z ⫽ 10.32, p ⬍ .001, OR ⫽ boundaries were no less steep if they were in a shifted condition,
7.93 and significantly later (i.e., occurred at a higher percentage ␹2(2) ⫽ 1.09, p ⫽ .580. In other words, to the extent that the
morph) in the morph continuum for the anger shifted participants, sharpness of a category boundary indicates the precision of the
b ⫽ ⫺2.44, z ⫽ ⫺12.36, p ⬍ .001, OR ⫽ 0.09, in comparison with participant’s category representation, participants’ representations
the participants in the unshifted condition. These estimates indicate were equally precise even when their category boundaries moved
that both children and adults adapted their categories about which substantially between the supervised learning and testing phases.
faces constituted anger in accord with the facial configurations Adults more readily adapted to the range of emotion expres-
encountered in the shifted experimental conditions. Furthermore, sion intensity. Children and adults did not differ in their overall
the Experimental Condition ⫻ Age Group interaction was not use of calm and upset responses in the unshifted condition (con-
significant, ␹2(2) ⫽ 3.09, p ⫽ .214. In other words, children and trolling for all other variables, b ⫽ ⫺0.01, ␹2(1) ⫽ .001, p ⫽ .970).
adults shifted their boundaries to the same extent. However, the interaction between Age Group and Percent Anger
Finally, the Experimental Condition ⫻ Percent Anger interac- was significant, with the effect of 10% increases in morphed anger
tion was not significant, indicating that participants’ category being greater by a factor of 2.816 for adults compared to children,
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 7

b ⫽ 1.04, ␹2(1) ⫽ 27.52, p ⬍ .001. This effect can be observed by use of these differences in forming judgments of the expressers.
the slightly steeper curves for adults compared to children in The influence of distributional information on perceivers’ explicit
Figure 3. Thus, adults have more precise categories across exper- judgments about expressers has implications for unpacking the
imental conditions compared with children. The three-way inter- relation between perceptual cues and social behavior. Participants
action between Age Group, Percent Anger, and Experimental were constrained to adults for Experiments 2 and 3 for two reasons:
Condition was not significant, ␹2(2) ⫽ 4.21, p ⫽ .122. (a) there were no developmental differences in the extent to which
children and adults shifted boundaries in Experiment 1 and (b) mov-
Discussion ing to a within-subjects design required a greater number of trials,
which we thought would be demanding for children.
After just a brief exposure to a new statistical distribution of
facial emotions, children and adults changed their thresholds for
what they considered to be someone feeling upset. Yet despite Method
updating their category boundaries, participants in the neutral and Participants. Participants were 55 adults (34 female, 18 male,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

anger shifted conditions showed comparable categorical precision

3 unreported gender; age range ⫽ 18 –21 years, Mage ⫽ 18.77
This document is copyrighted by the American Psychological Association or one of its allied publishers.

to the perceptual judgments of participants in the unshifted con-

years, SDage ⫽ 0.70 years). We aimed for 40 participants because
dition. In other words, the new categorical representations were as
30 participants per cell was sufficient in Experiment 1. Because of
distinct as those for which participants could continue to rely on
overscheduling, we collected data on 55 participants. One partic-
the explicitly taught category boundary. While these data do not
ipant was excluded for inattention (as defined by only using one
necessarily address how children initially acquire emotion catego-
response option throughout the task). Adults were undergraduates
ries, the data provide evidence that these perceptual categories are
at the same university as Experiment 1 and participated for course
flexible. This evidence supports the view that statistical distribu-
credit (7% African American, 16% Asian American, 4% Hispanic,
tions of facial configurations influence people’s representations of
4% multiracial, 67% White, 4% unreported). Participants gave
emotion categories. And, in turn, Experiment 1 suggests that
informed consent. The Institutional Review Board approved the
categories for, and interpretations of, anger are malleable and can
research.
be adjusted according to the patterns in the environment.
Face stimuli. Stimuli included images of facial expressions of
Adults established (during the supervised and testing phases)
emotion for two male models (Model #24 and Model #42) and one
somewhat steeper category boundaries compared to children,
likely reflecting that adults have more precise category represen- female model (Model #10; same as in Experiment 1) selected from
tations of facial anger overall. It may be that categorization be- the MacArthur Network Face Stimuli Set (Tottenham et al., 2009).
comes more robust and efficient with experience, particularly as As in Experiment 1, all models’ facial expressions were morphed
adults prioritize facial cues, whereas children are learning how to in increments of 5% from a 100% neutral expression (i.e., 0%
interpret faces and divide their attention equally between faces and angry) to a 100% angry expression of emotion, creating 21 equally
other contextual information (Leitzke & Pollak, 2016). However, spaced images (Gao & Maurer, 2009). Stimuli were presented with
children’s performance was nevertheless characterized by the high PsychoPy (v1.83.04).
degree of flexibility in learning that has been observed across Procedure. The procedure was identical to Experiment 1 ex-
cognitive domains (e.g., Gopnik et al., 2017). cept for the following minor changes to the instructions, the
number of trials in the introductory, supervised, and testing phases,
and the addition of actor trait judgments at the end of the task. In
Experiment 2 the introductory phase, participants were introduced to three mod-
In Experiment 1, participants were able to rapidly update their els (Jane, Tom, and Brian). All other instructions were the same,
category boundary between neutral and angry for a single model. with Jane, Tom, and Brian replacing Jane. Eighteen practice trials
We suggest such emotion perception flexibility allows perceivers were completed (six per actor). In the supervised phase, the num-
to adjust to the expressive tendencies of the people in their current ber of trials was adjusted to account for the new actors. Stimuli
environmental context. But some social environments contain peo- consisted of three repetitions per actor, using the same morphing
ple with diverging expressive styles—for instance, one friend may increments as before, for a total of 108 trials. In the testing phase,
be highly expressive, with unambiguous anger displays, and an- the unshifted, neutral shifted, and anger shifted conditions all
other friend may have expressions of anger that are much more occurred within-participants. Participants saw 72 trials of each
subtle. When a perceiver encounters multiple people with distinct model for a total of 216 trials. As the number of trials per model
ranges of expression, does their category boundary reflect an was the same as in Experiment 1, we used the same distributions
average of all the individuals’ expressive ranges? Or do perceivers of morphs. The actor assigned to each distribution (unshifted,
track the probabilistic distributions of individual expressers and anger shifted, or neutral shifted) was randomized across partici-
establish a unique category boundary for each social actor? To pants.
address this question, we trained adult participants on a common At the end of the task, participants completed an assessment
category boundary for three different actors (within-subject), and of trait judgments about the actors whose faces they viewed
then exposed participants to a testing phase, where each actor’s during the experiment. For each actor, they were shown the
expressive distribution was shifted to a different distribution of actor’s 100% happy expression and were asked to, “Rate your
intensity of facial displays. In addition to assessing whether per- impression of Jane/Brian/Tom: How likeable/approachable/irrita-
ceivers are sensitive to, and can track, intraindividual variation in ble/angry/friendly is this person?” Each actor was judged on each
emotion expression, we asked whether perceivers make functional trait, for a total of 15 questions; the order of the actors was
 8 PLATE, WOOD, WOODARD, AND POLLAK

randomized between participants. Because of a programming er- terpretations of Jane’s expressions. To test this possibility, we
ror, trait judgment data was lost on five participants. analyzed the adult data from Experiment 1 along with trials
from Experiment 2 in which Jane was the target (since Jane was
the only target in Experiment 1) across all possible shifts
Results
(neutral shifted, unshifted, anger shifted). We regressed partic-
Analytic plan. Here we report results from the testing phase. ipant responses on the Actor Condition ⫻ Percent Anger inter-
Results from the supervised phase are available online. We fol- action (as above), plus each of those variables’ interactions with
lowed the same data processing and analysis procedure as Exper- Experiment (Experiment 1 ⫽ ⫺.5, Experiment 2 ⫽ .5). Next,
iment 1, making changes only as demanded by the modified design we included by-participant random slopes for the Actor Condi-
of Experiment 2. Specifically, there was no Age variable (because tion ⫻ Percent Anger interaction. The interaction between
we only included adults in this study) and the random effects Experiment and Actor Condition was significant, ␹2(2) ⫽
structure for the testing phase model was adjusted to reflect the 45.98, p ⬍ .001, ORs ⫽ 0.31 and 9.41, demonstrating that
within-subject nature of the Actor Condition variable (dummy participants did not shift their boundaries for Jane as far when
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

variable that coded for the neutral shifted, unshifted, or anger they were also seeing Tom and Brian’s competing expression
This document is copyrighted by the American Psychological Association or one of its allied publishers.

shifted actor). We then made a cross-study analysis of the current distributions (see Figure 4). The interaction between Experi-
data and the adults’ data from Experiment 1 to compare participant ment and Percent Anger was also significant, such that partic-
efficiency in learning the shifted category boundaries when par- ipants had steeper (more categorical) category boundaries in
ticipants needed to track the distributions of multiple people’s Experiment 1 compared to Experiment 2, ␹2(1) ⫽ 63.53, p ⬍
facial expressions. Last, we report analyses of the actor trait .001, OR ⫽ 0.41.
judgments. The effect of shifted distributions on trait judgments of
Testing phase. We regressed participants’ responses (0 ⫽ actors. Participants’ ratings of each of the three actors’ anger,
calm, 1 ⫽ upset) on the interaction between Actor Condition irritability, likability, friendliness, and approachability were
and Percent Anger. The maximal model with by-participant strongly correlated (rs ⬎ .64), so we computed an average score of
random slopes for the interaction term and lower order terms Overall Evaluation (higher scores indicate more positive evalua-
failed to converge, so we removed the lower order random tions of the target; models analyzing each of the five judgments
slopes (Brauer & Curtin, 2018). The main effect of Actor separately are online). Participants were expected to rate the actors
Condition was significant, ␹2(2) ⫽ 251.53, p ⬍ .001, indicating in the neutral shifted condition most positively, followed by the
that participants established distinct category boundaries for unshifted actors, then the anger shifted actors, since the latter had
each of the three actors. The dummy coded parameters indi- the most “intense” negative facial movements. This prediction
cated that the anger-response threshold was significantly earlier translated into a linear contrast (neutral shifted ⫽ ⫺.5, unshifted ⫽
(i.e., at a lower anger percentage) in the morph continuum when 0, anger shifted ⫽ .5), and created an orthogonal quadratic contrast
the actors were neutral shifted, b ⫽ 0.82, z ⫽ 9.62, p ⬍ .001,
variable (neutral and anger shifted ⫽ ⫺.33, unshifted ⫽ .66). We
OR ⫽ 2.27, and significantly later (i.e., at a higher anger
then regressed the Overall Evaluation scores on the two contrast
percentage) in the morph continuum when the actors were anger
variables. After initially fitting a linear mixed-effect model with
shifted, b ⫽ ⫺0.62, z ⫽ ⫺7.45, p ⬍ .001, OR ⫽ 0.54, compared
random slopes for both contrast variables, we removed the orthog-
to when the actors were unshifted. The intercept was not sig-
onal quadratic contrast’s random slope when the model failed to
nificantly different from 0, indicating that participants’ cate-
converge (Brauer & Curtin, 2018). The linear variable trended
gory boundaries for the unshifted actor remained at the 50%
toward significance, b ⫽ ⫺0.26, t(71.47) ⫽ ⫺1.85, p ⫽ .06, OR ⫽
anger point (the boundary trained during the supervised phase),
0.77 (the orthogonal contrast was not significant). In other words,
b ⫽ ⫺0.06, ␹2(1) ⫽ 0.25, p ⫽ .619, OR ⫽ 0.94.
we found suggestive— but not statistically significant— evidence
The interaction between Percent Anger and Actor Condition
that participants made more negative trait attributions about actors
was also significant, ␹2(2) ⫽ 7.40, p ⫽ .025, indicating that the
with more extreme anger expressions compared to the actors with
steepness of the category boundary slopes was not uniform
more subtle expressive distributions during the task (even though
across the three shifted Conditions. The dummy coded interac-
the actors displayed the same, 100% happy expression, when trait
tion terms suggest that the slope of the category boundary for
ratings were obtained). Thus, participants not only updated their
the neutral shifted actors was steeper than the slope for the
unshifted actors, b ⫽ 0.35, z ⫽ 2.67, p ⫽ .008, OR ⫽ 1.42, category boundaries in response to the shifted distributions of
while the slope for the anger shifted actors only trended toward anger expressions, but may have also formed impressions of the
being steeper than the unshifted actors slope, b ⫽ 0.18, z ⫽ actors’ emotional traits based on those distributions.
1.65, p ⫽ .10, OR ⫽ 1.20.
Adapting emotion representations for multiple expressers. Discussion
The above analyses indicate that participants did not maintain a
single emotion category for each of the different actors in the Experiment 2 demonstrates that, when faced with multiple peo-
testing phase. Rather, participants were able to encode each actor’s ple differing in anger expressivity, perceivers formed individual-
particular distribution of facial displays and formed distinct cate- specific category boundaries. This effect indicates robustness of
gory boundaries for each actor. statistical learning, allowing perceivers to both track and integrate
It is possible that the multiple actor context caused partici- multiple distributions simultaneously. Additionally, a trend in the
pants to draw comparisons between the actors. For instance, data suggests that perceivers formed trait judgments that varied
perhaps viewing Tom’s distributions colored participants’ in- based on the statistical distribution of each individual’s expressiv-
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 9
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 4. Comparison of the testing phase responses of Experiment 1 adult participants and Experiment 2
participants on all conditions (neutral shifted, unshifted, anger shifted) of “Jane” trials. While participants
updated their category boundaries for Jane based on her facial expression distribution, the subsequent category
shift was also influenced by the facial distributions of the other actors encountered in the environment of
Experiment 2. See the online article for the color version of this figure.

ity. While the effect on actor trait judgments was small, if valid, it to replicate Experiment 2, but removed the supervised learning
would indicate that perceptual categories have a functional impli- phase. Therefore, the testing phase of Experiment 3 fully con-
cation in how individuals engage with social partners. stitutes unsupervised learning in that participants are exposed to
One notable limitation for Experiment 2 (and Experiment 1) the stimuli without any prior feedback regarding categorization.
is the presence of the supervised phase that occurred before
participants engaged in the testing phase. In the supervised Experiment 3
phase, participants were trained to a midpoint category bound-
ary. Perhaps, as a result of this training, participants acquired an
Method
assumption that they should respond “red room” for 50% of the
trials and respond “blue room” for 50% of the trials for each Participants. Participants were 40 adults (21 female, 19 male;
actor. Participants acting in accordance with this assumption age range ⫽ 18 –21 years, Mage ⫽ 18.88 years, SDage ⫽ 0.88
would demonstrate the same pattern of results observed in years). We aimed for 40 participants as in Experiment 2. Adults
Experiment 2. To address this potential confound, we attempted were undergraduates at the same university as in Experiments 1
 10 PLATE, WOOD, WOODARD, AND POLLAK

and 2, who participated for course credit (5% African American, Results
10% Asian American, 8% Hispanic, 3% Multiracial, 75% White).
Participants gave informed consent. The Institutional Review Analytic Plan. We followed the same data processing and
Board approved the research. analysis procedure as Experiment 2.
Face stimuli. Stimuli used were identical to those in Experi- Testing phase. The model was identical to Experiment 2. The
ment 2. intercept term was significantly different from 0, suggesting that in
Procedure. The procedure was identical to Experiment 2 ex- the absence of the supervised phase, participants’ category bound-
cept that the introductory phase was shortened and the supervised ary for the unshifted actors was not at 50% anger, b ⫽ 1.88,
learning phase was omitted. The introductory phase consisted of ␹2(1) ⫽ 68.85, p ⬍ .001, OR ⫽ 6.55. Participants who have never
six trials in which each actor’s 0% angry and 100% angry morphs received feedback on their responses tended to categorize an
were presented. Participants also completed the survey of actor expression as upset at a lower anger percentage (see Figure 5). As
judgments at the end of task, though because of program errors, in the previous experiments, the effect of Actor Condition was
only 37 of the participants had complete survey data. significant, ␹2(2) ⫽ 32.10, p ⬍ .001. The dummy coded parame-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 5. Comparing Experiments 2 and 3 testing phase responses. Participants in Experiment 3 did not
complete initial supervised trials. Therefore, they tended to start from a default category boundary that was much
lower in anger as compared to the 50% boundary taught to participants in Experiment 2 in the supervised
learning phase. Despite this main effect of removing the supervised trials (illustrated by the dashed Experiment
3 lines being further to the left), Experiment 3 participants nonetheless learned distinct category boundaries for
each of the three actors. See the online article for the color version of this figure.
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 11

ters indicated that the category boundaries for the neutral shifted General Discussion
actors were at a lower anger percentage than for the unshifted
actors, b ⫽ 0.52, z ⫽ 3.92, p ⬍ .001, OR ⫽ 1.68. Participants’ People represent facial configurations as members of categories
boundaries for anger shifted actors trended toward being at a (Cong et al., 2018; de Gelder, Teunisse, & Benson, 1997). But it
higher anger percentage than for the unshifted actors, b ⫽ ⫺0.21, was unknown whether and how people flexibly change them to
z ⫽ ⫺1.69, p ⫽ .09, OR ⫽ 0.81. Thus we replicated the overall reflect their social contexts, and whether this tendency changes
effect from Experiment 2, even after removing the potential con- with age. The present data provide support for the view that the
found of the supervised phase. statistical distribution of observed expressions operates on peo-
The effect of initial supervised learning: Comparing Exper- ple’s representations of emotions, highlighting the malleability of
iments 2 and 3. We combined the testing phase data from emotion categories across ages, and (though not directly tested
Experiments 2 and 3 (coded as ⫺.5 and .5, respectively) to here) suggesting a potential mechanism through which emotion
examine whether the initial supervised phase (which was only categories might be formed. Further, the specificity and efficiency
present in Experiment 2, and trained participants to an initial with which the boundaries shifted during exposure to the distri-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

boundary location of 50% for all actors) changed participants’ bution of faces— even with multiple actors—suggests that these
This document is copyrighted by the American Psychological Association or one of its allied publishers.

response patterns in the subsequent testing phase. The model learning processes are both robust and flexible.
structure was identical to that of the model comparing Experiments
1 and 2.
The main effect of Experiment was significant, such that par- Exploring a Mechanism for Adjusting to the
ticipants in Experiment 3 made significantly more upset categori- Expressive Styles of Others
zations than calm, b ⫽ 1.95, ␹2(1) ⫽ 63.42, p ⬍ .001, OR ⫽ 7.03.
This is illustrated in Figure 5 by the overall leftward shift of the The current work suggests that unsupervised statistical learning
Experiment 3 lines. The interaction between Actor Condition and is one way in which perceivers quickly adjust their interpretation
Experiment was also significant, ␹2(2) ⫽ 16.14, p ⬍ .001. The of facial emotions according to individual differences in expres-
dummy coded interaction terms indicate that the Experiment 2 sivity. People often encounter shifts in how emotions are ex-
participants shifted their responses on anger shifted trials signifi- pressed, both in the short term (e.g., with a particularly expressive
cantly further from their unshifted trial responses, when compared social partner) and in the long term (e.g., when visiting a new
to Experiment 3 participants, b ⫽ 0.38, z ⫽ 2.56, p ⫽ .01, OR ⫽ culture). Substantial variability in expressivity exists (Friedman,
1.46. The moderating effect of Experiment on the distance be- Prince, Riggio, & DiMatteo, 1980), some resulting from person-
tween the neutral shifted boundary and the unshifted boundary was ality (Friedman, DiMatteo, & Taranta, 1980) and gender (Kring &
not significant, b ⫽ ⫺.23, z ⫽ ⫺1.55, p ⫽ .12, OR ⫽ 0.79. Gordon, 1998), such that some individuals are more expressive
The effect of shifted distributions on actor judgments. We facially than others. To successfully engage with others, observers
took an identical approach to analyzing participants’ ratings of the must be able to track and adjust to such individual differences. At
three actors’ anger, irritability, likability, friendliness, and ap- the same time, and consistent with the data reported here, patterns
proachability, which we averaged into an Overall Evaluation of individual variation are not independent, as perceivers also
score. The linear contrast for Actor Condition was significant, such integrate exemplars with reference to each other.
that participants’ evaluations of the three actors became increas- On a larger scale, entire cultures vary in production of facial
ingly negative as their expressive distributions shifted toward the expressions (Niedenthal et al., 2017). Cross-cultural differences in
more angry end of the continuum, b ⫽ ⫺0.39, t(48.63) ⫽ ⫺2.15, expressivity are thought to be cultural adaptations to social and
p ⫽ .04, OR ⫽ 0.68. ecological pressures, and interactions with people from other cul-
tures partly rely on an observer’s ability to quickly adjust to the
Discussion expressive style of their new social partner (Girard & McDuff,
2017; Rychlowska et al., 2015; Wood, Rychlowska, & Niedenthal,
Experiment 3 was identical to Experiment 2 but lacked the
2016). The current work elucidates one way in which perceivers
initial supervised phase. We did this to determine whether partic-
quickly adjust their interpretation of facial emotions, that is, de-
ipants’ seeming sensitivity to each actor’s distribution merely
pending upon statistical properties of their social partner’s expres-
reflected a tendency to categorize half of each actor’s expressions
sivity. To successfully engage with others, observers must be able
as upset, a task demand that the supervised phase may have
to learn from and adjust to these variations. Future research can
inadvertently enforced. In comparing Experiment 2 and Experi-
ment 3, it appears that the supervised phase in Experiment 2 served explore the possibility that this process of tracking the distribu-
to: (a) bring all participants away from their default anger detec- tional properties of emotional displays contributes to children’s
tion threshold to a common 50% anger boundary, and (b) make initial acquisition of emotion representations and expressivity
participants somewhat more sensitive to the distinct expression norms as they navigate the social world.
distributions for each actor. The latter effect may be due to Finally, the current work suggests that emotion perception re-
Experiment 2’s task demand to respond “calm” and “upset” searchers should proceed with caution whenever using emotion
equally for each actor. However the effect of shifting each actor’s categorization or detection tasks to measure individual differences
distribution in Experiment 3 did not go away when the supervised they assume to be stable (e.g., Baron-Cohen, Wheelwright, Hill,
phase was removed. This suggests people can and do encode the Raste, & Plumb, 2001). Here we found that people’s emotion
unique expressive ranges of individuals and categorize their ex- category representations are flexible and responsive to the distri-
pressions accordingly, regardless of task demands. bution of cues in the social context.
 12 PLATE, WOOD, WOODARD, AND POLLAK

Limitations and Future Directions category representation. Inspired by recent interest in supervised
and unsupervised learning mechanisms, we investigated these
In one sense, the laboratory paradigm is a very simplified ideas in the domain of emotion. The robustness of the mechanism
version of emotion categorization. Yet, it still provides insights was probed through replication across three experiments, in envi-
into real-world behavior. Rather than having participants simply ronments with a single versus multiple expressers, and in the
label a face with emotion words (something people arguably rarely relation between statistical learning and explicit trait ratings. We
do in everyday life, but often are asked do in the laboratory), are continuing this line of research to further the understanding of
participants predicted the actor’s likeliest next behavior. No emo- the influence of individual differences and contextual factors in
tion labels were encountered in the supervised and testing phases emotion representations and the flexibility of emotion categories.
of the experiment. We suggest the current task bears closer resem-
blance to what people do with the information they extract from
References
the faces of people around them (Martin, Rychlowska, Wood, &
Niedenthal, 2017). However, while we avoided labels that imply Ahn, H. J. (2005). Teachers’ discussions of emotion in child care centers.
an internal state, such as anger, and did not use labels as the Early Childhood Education Journal, 32, 237–242. http://dx.doi.org/10
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

forced-choice options, our use of the terms calm and upset during .1007/s10643-004-1424-6
This document is copyrighted by the American Psychological Association or one of its allied publishers.

the instructions and practice trials may have influenced partici- Aviezer, H., Trope, Y., & Todorov, A. (2012). Holistic person processing:
pants’ representations of the stimuli (see Doyle & Lindquist, 2018; Faces with bodies tell the whole story. Journal of Personality and Social
Psychology, 103, 20 –37. http://dx.doi.org/10.1037/a0027411
Lupyan, Rakison, & McClelland, 2007).
Baron-Cohen, S., Wheelwright, S., Hill, J., Raste, Y., & Plumb, I. (2001).
Future research can also consider how this learning and updat- The “Reading the Mind in the Eyes” test revised version: A study with
ing process generalizes across emotions. The current study ex- normal adults, and adults with Asperger syndrome or high-functioning
plored the effects of supervised and unsupervised learning on autism. Journal of Child Psychology and Psychiatry, 42, 241–251.
anger; future research can confirm that similar processes operate http://dx.doi.org/10.1111/1469-7610.00715
across other emotion categories. Anger indicates a potential threat, Barrett, L. F. (2017). Functionalism cannot save the classical view of
and is an especially salient and attention-grabbing stimulus. It emotion. Social Cognitive and Affective Neuroscience, 12, 34 –36.
could be that adults and children are more attuned to the patterns Barrett, L. F., Mesquita, B., & Gendron, M. (2011). Context in emotion
of anger expressions than other expressions. Additionally, an in- perception. Current Directions in Psychological Science, 20, 286 –290.
teresting next step will be to examine how these learning processes http://dx.doi.org/10.1177/0963721411422522
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear
operate when embedded in richer contexts that mimic the social
mixed-effects models using lme4. Journal of Statistical Software, 67,
world. For example, antecedent events and behavioral conse- 1– 48. http://dx.doi.org/10.18637/jss.v067.i01
quences that precede and follow a facial expression can shape or Brauer, M., & Curtin, J. J. (2018). Linear mixed-effects models and the
reinforce a perceiver’s current and future interpretations of the analysis of nonindependent data: A unified framework to analyze cate-
expression (Barrett, Mesquita, & Gendron, 2011). gorical and continuous independent variables that vary within-subjects
and/or within-items. Psychological Methods, 23, 389 – 411. http://dx.doi
.org/10.1037/met0000159
Conclusion Calder, A. J., Young, A. W., Perrett, D. I., Etcoff, N. L., & Rowland, D.
(1996). Categorical perception of morphed facial expressions. Visual
The current experiment is the first known demonstration that
Cognition, 3, 81–118. http://dx.doi.org/10.1080/713756735
exposure to a particular statistical distribution of facial emotion Campanella, S., Quinet, P., Bruyer, R., Crommelinck, M., & Guerit, J.-M.
changed people’s emotion category boundaries. That emotion cat- (2002). Categorical perception of happiness and fear facial expressions:
egories are malleable and responsive to environmental statistics An ERP study. Journal of Cognitive Neuroscience, 14, 210 –227. http://
raises new possibilities for understanding human emotion percep- dx.doi.org/10.1162/089892902317236858
tion. Such flexibility also allows individuals to adjust emotion Casasanto, D., & Lupyan, G. (2015). All concepts are ad hoc concepts. In
concepts across contexts and organize appropriate behavioral re- E. Margolis & S. Laurence (Eds.), The conceptual mind: New Directions
sponses based on the available sociocultural cues. in the study of concepts (pp. 543–566). Cambridge, UK: MIT Press.
Clore, G. L., & Ortony, A. (2013). Psychological construction in the OCC
model of emotion. Emotion Review, 5, 335–343. http://dx.doi.org/10
Research in Context .1177/1754073913489751
Cong, Y.-Q., Junge, C., Aktar, E., Raijmakers, M., Franklin, A., & Sauter,
Evidence suggests that people perceive facial configurations as D. (2018). Pre-verbal infants perceive emotional facial expressions
members of emotion categories, even though the facial muscles categorically. Cognition and Emotion. Advance online publication.
making up expressions can vary continuously across many feature http://dx.doi.org/10.1080/02699931.2018.1455640
dimensions. As with other perceptual categories, perceivers must de Gelder, B., Teunisse, J.-P., & Benson, P. J. (1997). Categorical percep-
learn the diagnostic features of emotion categories and become tion of facial expressions: Categories and their internal structure. Cog-
sensitive to the boundaries between those categories (e.g., detect- nition and Emotion, 11, 1–23. http://dx.doi.org/10.1080/0269
ing when another person has become angry). Adding to this 99397380005
Diamond, R., & Carey, S. (1986). Why faces are and are not special: An
perceptual challenge, perceivers must also update category repre-
effect of expertise. Journal of Experimental Psychology: General, 115,
sentations based on differences in expressivity across individuals 107–117. http://dx.doi.org/10.1037/0096-3445.115.2.107
and cultures. Given the authors’ combined background in emo- Doyle, C. M., & Lindquist, K. A. (2018). When a word is worth a thousand
tional development, cognitive science, and the influence of culture pictures: Language shapes perceptual memory for emotion. Journal of
on emotion representations, we were interested in identifying Experimental Psychology: General, 147, 62–73. http://dx.doi.org/10
potential learning processes that support this flexible emotion .1037/xge0000361
 PROBABILISTIC LEARNING OF EMOTION CATEGORIES 13

Eisenberg, N. (2000). Emotion, regulation, and moral development. Annual Levari, D. E., Gilbert, D. T., Wilson, T. D., Sievers, B., Amodio, D. M., &
Review of Psychology, 51, 665– 697. http://dx.doi.org/10.1146/annurev Wheatley, T. (2018). Prevalence-induced concept change in human
.psych.51.1.665 judgment. Science, 360, 1465–1467. http://dx.doi.org/10.1126/science
Ekman, P. (1992). An argument for basic emotions. Cognition & Emotion, .aap8731
6, 169 –200. http://dx.doi.org/10.1080/02699939208411068 Love, B. C. (2002). Comparing supervised and unsupervised category
Etcoff, N. L., & Magee, J. J. (1992). Categorical perception of facial learning. Psychonomic Bulletin & Review, 9, 829 – 835. http://dx.doi.org/
expressions. Cognition, 44, 227–240. http://dx.doi.org/10.1016/0010-02 10.3758/BF03196342
77(92)90002-Y Lupyan, G., Rakison, D. H., & McClelland, J. L. (2007). Language is not
Fisher, D. H., Pazzani, M. J., & Langley, P. (2014). Concept formation: just for talking: Redundant labels facilitate learning of novel categories.
Knowledge and experience in unsupervised learning. San Mateo, CA: Psychological Science, 18, 1077–1083. http://dx.doi.org/10.1111/j
Morgan Kaufmann Publishers. .1467-9280.2007.02028.x
Frank, M. C., Vul, E., & Johnson, S. P. (2009). Development of infants’ Marchi, F., & Newen, A. (2015). Cognitive penetrability and emotion
attention to faces during the first year. Cognition, 110, 160 –170. http:// recognition in human facial expressions. Frontiers in Psychology, 6,
dx.doi.org/10.1016/j.cognition.2008.11.010 828. http://dx.doi.org/10.3389/fpsyg.2015.00828
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Friedman, H. S., DiMatteo, M. R., & Taranta, A. (1980). A study of the Martin, J., Rychlowska, M., Wood, A., & Niedenthal, P. (2017). Smiles as
This document is copyrighted by the American Psychological Association or one of its allied publishers.

relationship between individual differences in nonverbal expressiveness multipurpose social signals. Trends in Cognitive Sciences, 21, 864 – 877.
and factors of personality and social interaction. Journal of Research in http://dx.doi.org/10.1016/j.tics.2017.08.007
Personality, 14, 351–364. http://dx.doi.org/10.1016/0092-6566(80) Maye, J., Werker, J. F., & Gerken, L. (2002). Infant sensitivity to distri-
90018-5 butional information can affect phonetic discrimination. Cognition, 82,
Friedman, H. S., Prince, L. M., Riggio, R. E., & DiMatteo, M. R. (1980). B101–B111. http://dx.doi.org/10.1016/S0010-0277(01)00157-3
Understanding and assessing nonverbal expressiveness: The affective Niedenthal, P. M., Rychlowska, M., & Wood, A. (2017). Feelings and
communication test. Journal of Personality and Social Psychology, 39, contexts: Socioecological influences on the nonverbal expression of
333–351. http://dx.doi.org/10.1037/0022-3514.39.2.333 emotion. Current Opinion in Psychology, 17, 170 –175. http://dx.doi
Gao, X., & Maurer, D. (2009). Influence of intensity on children’s sensi- .org/10.1016/j.copsyc.2017.07.025
Pollak, L. H., & Thoits, P. A. (1989). Processes in emotional socialization.
tivity to happy, sad, and fearful facial expressions. Journal of Experi-
Social Psychology Quarterly, 52, 22–34. http://dx.doi.org/10.2307/278
mental Child Psychology, 102, 503–521. http://dx.doi.org/10.1016/j.jecp
6901
.2008.11.002
Pollak, S. D., & Kistler, D. J. (2002). Early experience is associated with
Girard, J. M., & McDuff, D. (2017). Historical heterogeneity predicts
the development of categorical representations for facial expressions of
smiling: Evidence from large-scale observational analyses. In Proceed-
emotion. Proceedings of the National Academy of Sciences of the United
ings of the 12th IEEE International Conference on Automatic Face &
States of America, 99, 9072–9076. http://dx.doi.org/10.1073/pnas
Gesture Recognition (pp. 719 –726). http://dx.doi.org/10.1109/FG.2017
.142165999
.135
Pollak, S. D., Messner, M., Kistler, D. J., & Cohn, J. F. (2009). Develop-
Gopnik, A., O’Grady, S., Lucas, C. G., Griffiths, T. L., Wente, A.,
ment of perceptual expertise in emotion recognition. Cognition, 110,
Bridgers, S., . . . Dahl, R. E. (2017). Changes in cognitive flexibility and
242–247. http://dx.doi.org/10.1016/j.cognition.2008.10.010
hypothesis search across human life history from childhood to adoles-
Pourtois, G., Schettino, A., & Vuilleumier, P. (2013). Brain mechanisms
cence to adulthood. Proceedings of the National Academy of Sciences of
for emotional influences on perception and attention: What is magic and
the United States of America, 114, 7892–7899. http://dx.doi.org/10
what is not. Biological Psychology, 92, 492–512. http://dx.doi.org/10
.1073/pnas.1700811114
.1016/j.biopsycho.2012.02.007
Gordon, S. L. (1991). 12 The socialization of children’s emotions: Emo- R Development Core Team. (2008). R: A language and environment for
tional culture, competence, and exposure. In C. Saarni & P. L. Harris statistical computing [Computer software]. Vienna, Austria : R Foun-
(Eds.), Children’s understanding of emotion (pp. 319 –349). New York, dation for Statistical Computing. Retrieved from http://www.R-project.
NY: Cambridge University Press. org
Grossmann, T. (2015). The development of social brain functions in Russell, J. A., & Bullock, M. (1986). Fuzzy concepts and the perception of
infancy. Psychological Bulletin, 141, 1266 –1287. http://dx.doi.org/10 emotion in facial expressions. Social Cognition, 4, 309 –341. http://dx
.1037/bul0000002 .doi.org/10.1521/soco.1986.4.3.309
Huelle, J. O., Sack, B., Broer, K., Komlewa, I., & Anders, S. (2014). Rychlowska, M., Miyamoto, Y., Matsumoto, D., Hess, U., Gilboa-
Unsupervised learning of facial emotion decoding skills. Frontiers in Schechtman, E., Kamble, S., . . . Niedenthal, P. M. (2015). Hetero-
Human Neuroscience, 8, 77. http://dx.doi.org/10.3389/fnhum.2014 geneity of long-history migration explains cultural differences in
.00077 reports of emotional expressivity and the functions of smiles. Pro-
Kalish, C. W., Rogers, T. T., Lang, J., & Zhu, X. (2011). Can semi- ceedings of the National Academy of Sciences of the United States of
supervised learning explain incorrect beliefs about categories? Cogni- America, 112, E2429 –E2436. http://dx.doi.org/10.1073/pnas
tion, 120, 106 –118. http://dx.doi.org/10.1016/j.cognition.2011.03.002 .1413661112
Kalish, C. W., Zhu, X., & Rogers, T. T. (2015). Drift in children’s Saffran, J. R., Newport, E. L., & Aslin, R. N. (1996). Word segmentation:
categories: When experienced distributions conflict with prior learning. The role of distributional cues. Journal of Memory and Language, 35,
Developmental Science, 18, 940 –956. http://dx.doi.org/10.1111/desc 606 – 621. http://dx.doi.org/10.1006/jmla.1996.0032
.12280 Shariff, A. F., & Tracy, J. L. (2011). Emotion expressions: On signals,
Kring, A. M., & Gordon, A. H. (1998). Sex differences in emotion: symbols, and spandrels—A response to Barrett (2011). Current Direc-
Expression, experience, and physiology. Journal of Personality and tions in Psychological Science, 20, 407– 408. http://dx.doi.org/10.1177/
Social Psychology, 74, 686 –703. http://dx.doi.org/10.1037/0022-3514 0963721411429126
.74.3.686 Susskind, J. M., Littlewort, G., Bartlett, M. S., Movellan, J., & Anderson,
Leitzke, B. T., & Pollak, S. D. (2016). Developmental changes in the A. K. (2007). Human and computer recognition of facial expressions of
primacy of facial cues for emotion recognition. Developmental Psychol- emotion. Neuropsychologia, 45, 152–162. http://dx.doi.org/10.1016/j
ogy, 52, 572–581. http://dx.doi.org/10.1037/a0040067 .neuropsychologia.2006.05.001
 14 PLATE, WOOD, WOODARD, AND POLLAK

Tottenham, N., Tanaka, J. W., Leon, A. C., McCarry, T., Nurse, M., Wood, A., Rychlowska, M., & Niedenthal, P. M. (2016). Heterogeneity of
Hare, T. A., . . . Nelson, C. (2009). The NimStim set of facial long-history migration predicts emotion recognition accuracy. Emotion,
expressions: Judgments from untrained research participants. Psychi- 16, 413– 420. http://dx.doi.org/10.1037/emo0000137
atry Research, 168, 242–249. http://dx.doi.org/10.1016/j.psychres
.2008.05.006
Wood, A., Lupyan, G., Sherrin, S., & Niedenthal, P. (2016). Altering
sensorimotor feedback disrupts visual discrimination of facial expres- Received May 17, 2018
sions. Psychonomic Bulletin & Review, 23, 1150 –1156. http://dx.doi Revision received August 7, 2018
.org/10.3758/s13423-015-0974-5 Accepted September 25, 2018 䡲
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.