Current Biology
Dispatches
16. Wang, R., Dearing, J.A., Langdon, P.G.,
Zhang, E., Yang, X., Dakos, V., and
Scheffer, M. (2012). Flickering gives
early warning signals of a critical transition
to a eutrophic lake state. Nature 492,
419–422.
17. Costello, E.K., Stagaman, K., Dethlefsen,
L., Bohannan, B.J., and Relman, D.A.
(2012). The application of ecological
theory toward an understanding of the
Neuropsychology: How Many Emotions Are There?
Julien Dubois and Ralph Adolphs
Division of Humanities and Social Sciences, California Institute of Technology, Pasadena, CA 91125, USA
Correspondence: jcrdubois@gmail.com (J.D.), radolphs@hss.caltech.edu (R.A.)
http://dx.doi.org/10.1016/j.cub.2015.06.037
Psychological theories disagree on how we attribute emotions to people. A new neuroimaging study shows
that such attributions involve a large number of abstract features, rather than a small set of emotion
categories.
The most popular emotion theories
propose either two broad
dimensions — arousal and valence
(pleasantness) [1] — or a small number
(around six) of discrete ‘basic’ emotions
[2]. The first, dimensional, view has
the virtue of economy and is supported
by finding that many kinds of emotion
data can be mapped well into a
two-dimensional space [3,4]. The second
view derives much of its support from the
study of human facial expressions, and
also corresponds well to emotions we
would typically attribute to people and
animals (the list includes anger, fear, and
disgust). Yet a third theoretical proposal
argues that the rich emotions that people
experience unfold through a complex set
of evaluations and coping mechanisms.
Such ‘appraisal’ theories invoke a larger
vocabulary of features from which a
correspondingly larger set of emotions
can be constructed [5,6]. All three theories
have some appeal, and all three probably
reflect aspects of what people actually do
when they attribute emotions to others as
well as to themselves. Is there any way to
adjudicate further between the theories?
A new study by Skerry and Saxe [7],
reported in this issue of Current Biology,
now provides such adjudication, based
on an important and relatively new source
Current Biology 25, R654–R676, August 3, 2015 ª2015 Elsevier Ltd All rights reserved R669
human microbiome. Science 336, 1255–
1262.
18. Lenton, T.M., Held, H., Kriegler, E., Hall, J.W.,
Lucht, W., Rahmstorf, S., and Schellnhuber,
H.J. (2008). Tipping elements in the Earth’s
climate system. Proc. Natl. Acad. Sci. USA
105, 1786–1793.
19. Dearing, J.A., Wang, R., Zhang, K., Dyke,
J.G., Haberl, H., Hossain, M.S., Langdon,
of data: neuroimaging. The study had
participants attribute emotions by reading
short stories while lying in a functional
magnetic resonance imaging (fMRI)
scanner, and quantified which of the three
emotion theories best corresponded with
the patterns of evoked brain activations.
Measuring Emotion Attribution
One reason there are competing theories
of emotion is that we attribute emotions to
people and animals on the basis of a wide
range of evidence. Some data come from
people describing how they feel;
additional clues arise from interpreting
particular kinds of behaviors (facial
expressions, body posture); and yet more
information can be derived from the
circumstances in which people find
themselves. Skerry and Saxe [7] focused
on the latter type of evidence. As their
stimuli, the authors chose short, written
vignettes that explicitly described
situational causes of emotions; for
example, how do you think Dana feels
from the following actual sample stimulus:
‘‘Dana always wanted a puppy, but her
parents said it was too much of a hassle.
One summer afternoon, Dana’s parents
returned from a supposed trip to the
grocery store, and Dana heard barking
from inside her garage. She opened the
P.G., Lenton, T.M., Raworth, K., Brown, S.,
et al. (2014). Safe and just operating spaces
for regional social-ecological systems.
Global Environ. Chang. 28, 227–238.
20. Carpenter, S.R., Cole, J.J., Pace, M.L., Batt,
R., Brock, W., Cline, T., Coloso, J., Hodgson,
J.R., Kitchell, J.F., and Seekell, D.A. (2011).
Early warnings of regime shifts: a
whole-ecosystem experiment. Science 332,
1079–1082.
door to see her parents holding a golden
retriever puppy.’’
While these stimuli necessarily lack the
full complexity of real-life emotion
attribution, they have the benefit of
allowing experimental control over the
explicit information on the basis of which
subjects make the attributions. The study
used 10 different stories for each of 20
emotions (200 stimuli in total; the 20
emotions were arbitrarily chosen, with half
of them positive and half negative in
valence), and controlled for other possible
confounds, such as the complexity of the
sentences used, or the ease with which
they could be read.
Skerry and Saxe [7] first collected
behavioral ratings for their stimuli from
an independent set of people queried
over the internet, and then compared
these to the brain activations evoked in
the subjects of the neuroimaging study.
Three different feature spaces were
constructed based on the internet ratings,
corresponding to the three different
emotion theories investigated. Which of
these three spaces best matched the
neural data?
Methodological Challenges
This question was addressed using a
technique called Representational
 Current Biology

Dispatches
STIMULI CANDIDATE MODELS FEATURAL REPRESENTATION REPRESENTATIONAL GEOMETRY REPRESENTATIONAL
20 emotion categories (3 shown) vs. BRAIN REGION dimensionality differs representational dissimilarity matrix SIMILARITY
10 vignettes per emotion between representational spaces (all 20 emotion categories shown)
“APPRAISAL 38”
compare
expected? full RSA matrix

model A
... patterns
APPREHENSIVE

DM el C
el A
el B
PFC
safe? Euclidean
emotion #8

For months, Naomi

Ratings from M-turk subjects

knowledge

mod
mod
mod
had been struggling distance
change?
to keep up with her model A

(N = 250 total)
BASIC EMOTIONS model B
model C
afraid? DMPFC
model B

...
compare
FURIOUS surprised?
emotion #10

RDMs
After an 18-hour
disgusted?
flight, Caitlin arrived
at her vacation desti - Kendall’s τa
VALENCE/AROUSAL
model C

negative positive noise ceiling

0.12
(lower bound)
mild intense
0.08
JOYFUL OTHER EMOTION MODELS?
emotion #19

τa
Dana always wanted 0.04
a puppy, but her
fMRI subjects
neural data

... (*)
(N = 22)
parents said it was 0.00

model C
model A
model B
... (*)
... (*)
DMPFC (*) not real data Current Biology

Figure 1. Representational similarity analysis of emotion theories.

In the study [7], vignettes for 20 emotion categories (three examples are shown in the first column) were rated along dimensions posited by three different theories
(second column; model A, appraisal theory, 38 dimensions; model B, basic emotions, six dimensions; model C, valence/arousal, two dimensions). These ratings
produced representations of each of the emotions in the feature space corresponding to each model (third column). Since the representational spaces for the
different models have different dimensionalities, a transformation to ‘similarity space’ was then performed to ease comparison. Similarities between the 20
emotions were computed in the representational space of each model, yielding a Representational Dissimilarity Matrix (RDM) for each model (fourth column).
Similarly, an RDM was derived from neural data collected while subjects read the vignettes in the scanner (bottom row). Lastly, the neural RDM was
compared to each of the model RDMs using a rank correlation measure (last column). The winning model was the model with the highest correlation to the
neural RDM — and the goodness-of-fit of this model to the neural data was assessed with respect to the noise in the neural data. Possible future extensions
include the addition of further models of emotion, as well as visualization of the full RSA matrix (last column) with all brain regions included, perhaps using
techniques such as multi-dimensional scaling, to fully explore representational geometries across brain regions and emotion models.

Similarity Analysis (RSA) [8,9], the flow of
which we summarize in Figure 1. Take a
given stimulus, say a vignette evoking an
attribution of apprehensiveness (blue
stimuli in Figure 1, first column). Each of
the three theories (second column) posits
a representation of this emotion in a
specific space (38 dimensions for the
appraisal model, six for the basic
emotions model, and two for the
valence/arousal model). The 20 emotions
can be represented in each feature space
using average behavioral ratings (the third
column shows the representations of
apprehensive, furious, and joyful; see also
Figure 2A–C in the paper [7], which shows
the data for all the emotions used). Next
one compares the similarity between all
possible pairs of emotions — how far
apart the 20 emotions lie in each of the
three different representational spaces.
For example, in the valence/arousal
space, apprehensive and furious appear
more similar to one another than to joyful
(Figure 1, third column). This intuition can
be formalized into a Representational
Dissimilarity Matrix (RDM), computed
using a Euclidean distance measure for
each of the competing theories (Figure 1,
fourth column; see also Figure 3, top, in rightmost column). The study [7] found
the paper [7]). This analysis essentially compelling evidence that the neural
maps the similarity structure of different RDMs in all brain regions correlate best
stimuli, when rated according to the three with the model feature space of appraisal
different emotion theories: which theory. With the particular stimuli used
emotions cluster together as similar, and in the study, the six-emotion and
which seem very different from one valence/arousal models failed to capture
another? the representational geometry measured
An analogous approach can be taken in any of the investigated brain regions.
for the neuroimaging data, generating a Yet appraisal advocates should not
neural RDM (for dorsomedial prefrontal rejoice too hastily. The methods used
cortex as one example: Figure 1, bottom here lack the sensitivity to conclude that a
row; see also Figure 3, center, in the 38-dimensional appraisal theory is the
original paper [7]). In this case, it is not the ‘‘right’’ model: the noise ceiling
profiles of how people rate the stimuli on (the expected RDM correlation achieved
various features that determine the by the unknown true model) has a lower
similarity between the stimuli, but rather bound less than 0.1 for all brain regions
the different patterns of activation evoked (see dotted lines in Figure 4 and Figure S3
across voxels in a given brain region when in the paper [7]). As a correlation-based
participants read the stimuli in the measure, the theoretically best value in
scanner. the absence of noise is 1.0; Nili et al. [8]
The question now is: which feature report a value of about 0.25 for a study of
space (from the behavioral ratings) looks human IT fMRI. One possible reason is
most like the neural RDM (from the that the stimuli used in the Skerry and
neuroimaging) — the one for valence/ Saxe study [7] did not elicit sufficiently
arousal, the one for the six basic strong attributions of emotion in the
emotions, or the one built on appraisal subjects’ brains. Another may be the low
theory? This comparison can be done signal-to-noise ratio of the fMRI data.
with a rank correlation measure (Figure 1, Though there are complex trade-offs in

R670 Current Biology 25, R654–R676, August 3, 2015 ª2015 Elsevier Ltd All rights reserved
Current Biology
Dispatches
neuroimaging, future studies will likely be
able to substantially improve sensitivity
(raise the noise ceiling) by imaging with
smaller voxels, faster acquisition times,
and by using higher field strengths. These
hardware advances will go hand in hand
with processing refinements, such as
surface-based analyses and less data
smoothing [10]. Finally, it is of course well
known that BOLD-fMRI, the
neuroimaging method used in the paper
[7], is both indirect and macroscopic;
fMRI pattern analysis as used here may
not retrieve the relevant information
present in underlying populations of
single neurons [11], adding uncertainty to
the interpretation of the current findings.
An even more challenging question is
what models of emotion to use in the first
place. The present study compared three
emotion models, because these
correspond to well entrenched theories.
But of course an unbounded number of
other models are possible: how many
features should there be, and which
features should one pick (why did Skerry
and Saxe [7] choose the particular
38 features for their Appraisal Space?).
The authors are appropriately cautious in
what can be said about the ‘‘best’’
emotion space at this stage, but that still
leaves us with questions about what to
conclude.
Conceptual Interpretation
Prior studies have suggested that specific
basic emotions might be associated with
specific brain regions. For instance, the
amygdala has often been linked to fear
[12], and medial prefrontal cortex to
valence [13]. Skerry and Saxe [7] looked
everywhere in the brain using a
searchlight approach. The set of brain
regions where the searchlight approach
revealed information about the 20
emotions overlapped largely with an
independently localized set of regions
involved in attributing beliefs to people
(the so-called ‘theory of mind’ network
[14]). As with the present finding [7] that a
larger set of abstract features best
describes how the brain attributes
emotions, so too might such an expanded
feature set best describe ‘theory of mind’
more generally [15]. That is, we always
represent what other people are thinking
and feeling from a large and diverse set of
features that can flexibly describe many
different situations. The apparent
Current Biology 25, R654–R676, August 3, 2015 ª2015 Elsevier Ltd All rights reserved R671
simplicity of some findings (only six basic
emotions; only attributing beliefs) may
have more to do with the simplicity of the
stimuli or the task used in a particular
study, rather than reflecting how the
brain actually represents other people’s
minds in general. It thus appears that
there is a neural system for thinking
about other people, but exactly what it is
that the regions comprising this system
represent is typically complex and
abstract. Indeed, the situation is probably
more complicated yet, since there may be
no single neural system, but rather a
collection of systems that only partly
overlap [16].
One possible next step may be to
extend the investigation by combining the
competing theories. One could imagine
constructing a more complex framework
consisting of an underlying dimensionality
of valence and arousal, a more
fine-grained classification into six or so
‘basic’ emotion categories, and a very
fine-grained and more flexible attribution
based on appraisal features [17]. Perhaps
we use a mixture of these when we
attribute emotions to others, or when we
experience them ourselves (another
recent study suggests as much [18]).
Perhaps the extent to which one of the
three schemes dominates depends on the
details of the stimuli and the task. Perhaps
this also shows interesting individual
differences. Maybe children begin with
the simpler types of attributions, whereas
adults engage in more appraisal.
Acknowledging this flexibility and
variability in how we attribute emotions
would open up a large set of new studies
that would also link to psychopathology.
It is important to remember that all such
studies are still studies of how laypeople
attribute emotions to others in everyday
life. The participants in the studies are not
emotion researchers themselves. What
can be said about how scientists should
attribute emotions to people, let alone
how they should attribute them to other
animals? What is the best science of
emotion? That project might not use the
words we normally use for emotions at all
[19], and might attempt a broader survey
of features that all emotions share across
species [20]. Could a future science of
emotion look completely alien to the
layperson, as do current scientific
theories in physics, for example? Skerry
and Saxe [7] suggest a more optimistic
view: plausibly, one reason for the relative
success of psychological emotion
theories is that they haven’t got it
completely wrong, even if they haven’t got
it completely right. By telling us that a
higher-dimensional abstract feature
space best explains how our brains
ordinarily represent emotions, the study
may be telling the emotion scientist that
this would be a fruitful place to start as
well.
REFERENCES
1. Russell, J.A. (2003). Core affect and the
psychological construction of emotion. Psych.
Rev. 110, 145–172.
2. Ekman, P. (1994). Strong evidence for
universals in facial expressions: a reply to
Russell’s mistaken critique. Psych. Bull. 115,
268–287.
3. Russell, J.A., and Bullock, M. (1985).
Multidimensional scaling of emotional facial
expressions: similarity from preschoolers to
adults. J. Pers. Soc. Psych. 48, 1290–1298.
4. Rolls, E.T. (1999). The Brain and Emotion (New
York: Oxford University Press).
5. Lazarus, R.S. (1991). Emotion and Adaptation
(New York: Oxford University Press).
6. Scherer, K. (1984). On the nature and function
of emotion: A component process approach.
In Approaches to Emotion, K.R. Scherer, and
P. Ekman, eds. (New York: Lawrence Erlbaum
Associates), pp. 293–317.
7. Skerry, A.E., and Saxe, R. (2015). Neural
representations of emotion are organized
around abstract event features. Curr. Biol. 25,
1945–1954.
8. Nili, H., Wingfield, C., Walther, A., Su, L.,
Marslen-Wilson, W., and Kriegeskorte, N.
(2014). A toolbox for representational similarity
analysis. PLoS Comput. Biol. 10, e1003553.
9. Kriegeskorte, N., and Kievit, R.A. (2013).
Representational geometry: integrating
cognition, computation, and the brain. Trends
Cogn. Sci. 17, 401–412.
10. Glasser, M.F., Sotiropoulos, S.N., Wilson, J.A.,
Coalson, T.S., Fischl, B., Andersson, J.L., Xu,
J., Jbabdi, S., Webster, M., Polimeni, J.R.,
et al. (2013). The minimal preprocessing
pipelines for the Human Connectome Project.
Neuroimage 80, 105–124.
11. Dubois, J., deBerker, A.O., and Tsao, D.Y.
(2015). Single-unit recordings in the macaque
face patch system reveal limitations of fMRI
MVPA. J. Neurosci. 35, 2791–2802.
12. Calder, A.J., Lawrence, A.D., and Young, A.W.
(2001). Neuropsychology of fear and loathing.
Nat. Rev. Neurosci. 2, 352–363.
13. Kringelbach, M.L. (2005). The human
orbitofrontal cortex: linking reward to hedonic
experience. Nat. Rev. Neurosci. 6, 691–702.
14. Saxe, R. (2006). Why and how to study Theory
of Mind with fMRI. Brain Res. 1079, 57–65.

15. Schaafsma, S., Pfaff, D.W., Spunt, R.P., and
Adolphs, R. (2015). Deconstructing and
reconstructing theory of mind. Trends Cogn.
Sci. 19, 65–72.
16. Schurz, M., Radua, J., Aichhorn, M., Richlan,
F., and Perner, J. (2014). Fractionating theory
of mind: a meta-analysis of functional brain
Handedness: What Kangaroos Tell Us about Our
Lopsided Brains
Giorgio Vallortigara
Centre for Mind/Brain Sciences, University of Trento, Piazza della Manifattura 1, I-38068 Rovereto, Italy
Correspondence: giorgio.vallortigara@unitn.it
http://dx.doi.org/10.1016/j.cub.2015.06.036
Brain asymmetry is widespread, but the presence of handedness in non-human animals is debated. A new
study now provides evidence for handedness in bipedal — but not quadrupedal — marsupials.
A persistent myth in neuroscience has
been the idea that brain asymmetry — the
different functions of the left and right
sides of the nervous system — is a
uniquely human trait. Of course, there
could be uniquely human biological traits
that also show asymmetry (e.g. language),
but brain asymmetry in itself is so
widespread in the animal kingdom that it
can be very plausibly considered a
fundamental principle of organization of
their nervous systems. In recent years,
besides having been shown in vertebrates
[1], laterality has been found to occur also
in invertebrates, such as Caenorhabditis
elegans [2] and the honeybee [3], thus
revealing that brain size or number of
neurons is certainly not a key factor in the
emergence of cerebral asymmetry.
However, in spite of the huge amount of
evidence for brain asymmetry in different
animal groups, there is still debate on one
particular manifestation of cerebral and
functional asymmetry in behaviour in the
form of handedness. A recent study in
Current Biology by Giljov et al. [4] now
provides for the first time evidence for true
handedness in some species of
marsupials.
The term ‘handedness’ describes
manual asymmetries at the population
level, whereas ‘hand preference’ refers
usually to the individual level. Though
R672 Current Biology 25, R654–R676, August 3, 2015 ª2015 Elsevier Ltd All rights reserved
imaging studies. Neurosci. Biobehav. Rev. 42,
9–34.
17. Barrett, L.F., Mesquita, B., Ochsner, K.N., and
Gross, J.J. (2007). The experience of emotion.
Annu. Rev. Psychol. 58, 373–403.
18. Kragel, P.A., and LaBar, K.S. (2015).
Multivariate neural biomarkers of emotional
there might be some geographical
variation [5], around 90% of humans are
right-handed. In non-human primates,
whereas there is little doubt that
individuals may show hand preferences,
differences in the methods used to study
manual asymmetries (tasks, sample size,
etc.) have sometimes produced
discrepant findings [6]. Nonetheless,
some striking evidence for handedness
has emerged, particularly in great apes.
Bill Hopkins and his collaborators [7] have
collected data for more than 700
chimpanzees from four different
populations who have been tested for
hand use on a task requiring coordinated
bimanual actions, revealing a significant
right-hand bias in each sampled
population. The captive chimp colonies
that Hopkins studied are 60–70 percent
right-handed, regardless of the
proportion of individuals in each colony
that were human-reared. Thus, it has
been suggested that, whereas there
might be a genetic basis for handedness
in chimpanzees, it must be expressed
less strongly than in humans. This may be
due to the fact that right-handedness in
humans is associated with the left
hemisphere’s specialization for language
and speech production. Yet, the
association is far from straightforward,
because handedness in humans is only
Current Biology
Dispatches
states are categorically distinct. Soc. Cogn.
Affect. Neurosci. pii, nsv032.
19. LeDoux, J. (2012). Rethinking the emotional
brain. Neuron 73, 653–676.
20. Anderson, D.J., and Adolphs, R. (2014). A
framework for investigating emotion across
species. Cell 157, 187–200.
weakly correlated with cerebral
dominance for either praxis or language
[8]. Besides, the argument based on the
strength of handedness between human
and non-human animals is problematic.
First, because the measures of
handedness in humans (largely based on
questionnaires) rarely compare with those
carried out in non-human animals (for
exceptions, see [9]). Second, if we look at
other animals, the idea that handedness is
present with maximal strength only in our
species is untenable. Parrots, for
instance, show preferred use of a foot, the
left foot in most species, to hold food
objects with percentages of bias at the
population level that parallel or exceed
those of human handedness [10].
Even among non-human primates the
presence or absence of handedness
should be not taken as a cue of the
presence of other biases that indicate the
presence of functional asymmetry in the
brain. Common marmosets (Callithrix
jacchus), for instance, have a preferred
hand in simple reaching tasks which
develops by the time they are
8–12 months old and each individual uses
the same preferred hand across its
lifetime [11], but no population bias is
apparent. However, the same group of
marmosets has been shown to display a
striking group bias in a completely