1
Scientific RepoRtS | (2019) 9:15292 | https://doi.org/10.1038/s41598-019-51783-9
www.nature.com/scientificreports
Reexive gaze following in
common marmoset monkeys
Silvia Spadacenta
*
, Peter W. Dicke & Peter Thier
*
The ability to extract the direction of the other’s gaze allows us to shift our attention to an object of
interest to the other and to establish joint attention. By mapping one’s own intentions on the object of
joint attention, humans develop a Theory of (the other’s) Mind (TOM), a functional sequence possibly
disrupted in autism. Gaze following of both humans and old world monkeys is orchestrated by very
similar cortical architectures, strongly suggesting homology. Also new world monkeys, a primate
suborder that split from the old world monkey line about 35 million years ago, have complex social
structures and one member of this group, the common marmosets (Callithrix jacchus) are known to
follow human head-gaze. However, the question is if they use gaze following to establish joint attention
with conspecics. Here we show that this is indeed the case. In a free choice task, head-restrained
marmosets prefer objects gazed at by a conspecic and, moreover, they exhibit considerably shorter
choice reaction times for the same objects. These ndings support the assumption of an evolutionarily
old domain specic faculty shared within the primate order and they underline the potential value of
marmosets in studies of normal and disturbed joint attention.
Gaze following is a an important social skill, which enables humans, great apes and old world primates to establish
joint attention on an object of interest to the other. Although old world monkeys probably do not possess a TOM,
namely they are not able to map one’s own intentions, desires and expectations on the object of interest, they fol-
low the other’s gaze and they establish joint attention
1
. Like humans, gaze following of old world monkeys fullls
Fodor’s criteria of a domain specic function. Common marmosets are also well known for having a peculiar
interest in faces
2,3
. Unlike macaques, the species of old world primates studied best, and other non-human pri-
mate species, they oen engage in mutual gaze, for example in the context of joint action tasks
4
. Many individuals
even seek eye contact with their human caretakers (personal observations). Common marmosets also care about
the orientation of a human face as demonstrated by the fact that human head-gaze biases choices in an object
selection task
5
. While this latter behavior may indicate an inherent capacity for gaze following, it remains to be
shown that it can also be triggered by conspecics. By the same token the lack of high resolution behavioral data
has as yet precluded well-founded inferences about the relationship of marmoset gaze following to gaze following
exhibited by humans and rhesus monkeys, the two species of old world primates for which detailed behavioral
and neuronal data are available
1,6
. Gaze following of macaques and humans is reex-like in the sense that it is fast
and hard to suppress, two features that have contributed to the assumption of a domain specic faculty
7–12
based
on a dedicated neural system
13
. Do marmosets follow the gaze of conspecics in the same reex-like manner? An
armative response would support the notion that gaze following in extant primate lines is homologous, i.e. a
reection of shared ancestry.
In order to address these questions, we trained 3 common marmosets (2 females, 1 male) to execute a free
choice task in a well-controlled experimental setup that allowed us to head-restrain the animals to precisely track
eye movements. A conspecic’s face, oriented either to the le or to the right, was presented on a monitor for a
variable time ranging between 100 and 600 ms in steps of 100 ms (Fig.1a) and the observing animal was allowed
to scrutinize the face with eye movements conned by the boundaries of the portrait. e facial portrait was fol-
lowed by the appearance of two targets placed at −5° and + 5° from the center on the horizontal axis. e animals
had to freely choose one of the two targets, a human face (2° × 3°extension), by making an indicative saccade into
a window of 2° centered on the target within 500 ms. Independent of the orientation of the conspecic’s face, both
possible choices were rewarded, provided that the eyes had met the xation requirements.
Hertie Institute for Clinical Brain Research, Department of Cognitive Neurology, Otfried-Müller-Str. 27, 72076,
Tübingen, Germany. *email: silvia.spadacenta@uni-tuebingen.de; thier@uni-tuebingen.de
open
2
Scientific RepoRtS | (2019) 9:15292 | https://doi.org/10.1038/s41598-019-51783-9
www.nature.com/scientificreports
www.nature.com/scientificreports/
Results and Discussion
Common marmosets follow the gaze of a conspecic in a quasi-reexive manner. Figure1b, le
panel, plots the percentage of target choices in the direction of the face orientation (“congruent choices”) as func-
tion of the duration of the availability of the portrait. e graph depicts data pooled over all three animals and the
two possible face orientations: congruent choices exceeded chance level signicantly (binomial probability test,
p < 0.05), indicating that the observing monkey tended to follow the gaze of the portrayed monkey. is prefer-
ence was already apparent aer a presentation duration of the portrait of only 100 ms and got stronger for longer
presentation durations peaking at 300 ms exposure time (see also Supplementary Fig.S1). is dependence on
exposure duration is similar to the one exhibited by human observers when exposed to symbolic central cues
such as pointing arrows. ey typically demonstrate a gradual buildup of their spatial target preferences cued by
central stimuli, reaching an optimum at 300 ms
14,15
. As shown in the right panel of Fig.1b, the dependence of the
choice bias on presentation duration was the same in all three animals for up to 300 ms. Only later, the individual
plots start to diverge: interestingly, two of our animals (M2 and M3) showed a clear second peak, overall convey-
ing the impression of an oscillatory pattern with a period of about 250–300 ms. Periodic uctuations of attention
between two locations with a period of 4 HZ have also been described for human and macaque spatial vision
16,17
.
Figure 1. Oriented faces bias the animal’s choices to targets congruent with gaze direction. (a) Behavioral
paradigm. e trial started with the presentation of a central xation dot. Once xation was established, the
oriented face of a conspecic (replaced by other stimuli in control experiments) appeared for a variable time
(100–600 ms in steps of 100 ms). e disappearance of the conspecic’s face or the control stimuli and the
simultaneous appearance of two peripheral targets was the go signal for the animals to freely choose one of two
peripheral targets presented on the horizontal axis at 5° right and le of the center respectively, by means of a
saccade. e animal received a reward if the xation requirements were met. e xation window for the central
dot had a size of 2° × 2°, a size of 2° × 3° for the peripheral targets portraits of a human and for the central
portraits/control stimuli it corresponded to the extent the centrally presented image (7° × 6°). e subject
depicted in gure has agreed for her portrait to be published. (b) Le panel: plot of the percentage of target
choices congruent with portrait orientation as function of the duration of presentation. Pooled data (monkeys
M1, M2, M3). Binomial probability: ***p < 0.001. Right panel: plot of the percentage of target choices
congruent with portrait orientation for the individual animals M1, M2 and M3. Binomial probability: *p < 0.05,
***p < 0.001. (c) Monkey specic plots of the percentage of target choices congruent with portrait orientation,
separating portraits oriented to the le and right respectively. In each panel (le, M1; center, M2; right, M3) the
solid line stands for le oriented portraits and the dashed one for right oriented portraits.
3
Scientific RepoRtS | (2019) 9:15292 | https://doi.org/10.1038/s41598-019-51783-9
www.nature.com/scientificreports
www.nature.com/scientificreports/
Yet, given the fact that the third animal exhibited a dierent pattern, characterized by an absence of a second gaze
following peak and a constant choice at chance level aer 300 ms, further studies will be needed to critically assess
the possibility of periodicity.
All three animals exhibited individually dierent directional biases for the le and the right respectively, mod-
ifying their choice behavior on top of the inuence of directional information provided by facial orientation.
Directional biases became apparent when plotting the dependence of choice preference on head gaze direction
for the three individual animals independently for head gaze to the le and to the right (Fig.1c). For example,
a bias to the le side boosted the correct responses for the le oriented head gaze portraits (M1 and M2, le
and central panel respectively), and for the right oriented when the bias fell on the other side (M3, right panel).
Nonetheless, the bias never altered the overall response curve shape with a peak for congruent choices at around
300 ms. A signicant dominance of congruent choices peaking at 300 ms could be seen in M1 and M2 even for
congruent choices prompted by portraits oriented towards the animal’s non preferred side (binomial probability,
p < 0.001). A comparable tendency in M3 did not reach signicance (binomial probability, p = 0.1). e basis of
the directional bias remains unclear. e fact that it diers between individuals indicates that hidden imbalances
in the setup that might bind attention can hardly matter.
When the animals were confronted with direct gaze of a conspecic with the face turned straight or alterna-
tively, with black or grey disks of a similar size, likewise lacking directional information, target choices of all three
animals did not dier signicantly from chance level at most exposure times with the exception of the shortest
one (Fig.2a, pooled data). In particular the choice peaks for 300 and 600 ms could no longer been seen. For
100 ms exposure, overall pooled choices to the le were signicantly more frequent than to the right. Two of the
Figure 2. Direct gaze and disk or triangular stimuli attract the animals’ attention towards the center. (a) Plot
of le target choice as function of the duration of a central stimulus, either a conspecic’s face looking straight
at the observer (direct gaze) or a circular grey or black disk of similar size. Data pooled over the three animals
(monkeys M1, M2, M3). A choice bias is evident at the shortest duration time, whereas for longer exposures
the animals chose the targets on the right and le at random. Binomial probability: ***p < 0.001. (b) Bar plot
of percentage of choices congruent with the darker half of a bipartite disk. Data pooled over the three animals
(monkeys M1, M2, M3). For 100 ms presentation duration, the animals exhibited a signicant preference for the
target on the brighter side, i.e. opposite to the side preference to be expected based on a mechanism exploiting
the luminance asymmetry associated with face orientation. At 300 ms the choices did not indicate a preference.
Binomial probability: **p < 0.01; ***p < 0.001. (c) Bar plot of percentage of choices congruent with the side
cued by the oriented triangles. Data pooled over the three animals (monkeys M1, M2, M3). No signicant bias
was detected.
4
Scientific RepoRtS | (2019) 9:15292 | https://doi.org/10.1038/s41598-019-51783-9
www.nature.com/scientificreports
www.nature.com/scientificreports/
individual monkeys (M1, M2) exhibited this preference for the le, whereas the third one (monkey M3) a prefer-
ence for the right. e individual directional preferences for the le and the right corresponded to the direction
of the biases seen in the responses to oriented gaze (Fig.1c), yet, now conned to the shortest exposure only. We
think that the disappearance of the directional bias for longer exposure times might be a consequence of increas-
ing attraction towards the central object, overriding the bias, no matter if the central object is the neutral disk or
the portrait of a conspecic looking straight. is interpretation has interesting implications for the experiments
with oriented faces, which showed a persistence of the directional biases independent of exposure time. Here the
directional gaze seems to suppress the buildup of attraction to the central object, facilitating the readiness to look
elsewhere as determined by the resultant of the other’s gaze direction and an internal directional bias.
e white ear tus on the le and right of the darker central face of a straight ahead marmoset oer a symmet-
ric luminance prole. Once the animal turns the head to the side, symmetry is lost as the visible area of the ear
tu on the side of the head turn will decrease, whereas the area of the other one will increase (see Supplementary
Fig.S3a,b). Hence, gauging the extent of the luminance asymmetry may be a simple way to determine the other’s
head gaze direction without the need to process other aspects of the face. To test whether le-right dierences in
the luminance of an object prompt an orienting response of the observer, we exposed all 3 animals to bipartite
disks replacing the marmoset portraits. e disks were black on the le and light grey on the right or vice versa.
ese two versions of the bipartite disks were presented randomly interleaved for 100 ms or 300 ms, two portrait
exposure times that had prompted clear gaze following in the main experiment. Against the backdrop of the
preceding considerations, we had hypothesized that the animals might prefer the target on the side of the darker
half of the bipartite disk for both exposure times. However, contrary to our expectation, the animal preferred the
target on the side of the brighter half of the disk, independently if positioned on the right or le side and, more-
over, only for 100 ms exposure time. For 300 ms choices did not exhibit any preference (Fig.2b). is result does
not support the hypothesis that marmoset gaze following is determined by a simple mechanism, conned to the
comparison of the two ear tu areas.
Turning the head to the side will not only change the balance between paler and darker parts of the face but
also disrupt the symmetry of the facial contour. A face that appears symmetric when seen in a straight ahead
orientation will become asymmetric when turned to the side, an asymmetry that oers directional information.
If marmoset gaze following relied on an ability to exploit contour asymmetries, one might expect that even asym-
metric non-face shapes might redirect attention. In order to test this possibility we exposed our experimental
animals to the vision of lled out black triangles, covering an area similar to the one covered by the portraits. e
triangles were presented with one of their three corners pointing either to the le or to the right. e orientation
was chosen randomly interleaved and the triangles were presented for 100 ms or 300 ms in separate blocks of
trials, i.e. for the presentation durations that had provided the strongest eects of oriented marmoset faces. e
results clearly showed that both for the le and the right oriented triangle the target choice remained at chance
level independent of the presentation duration (see Fig.2c). In sum, the luminance and shape control stimuli,
mimicking isolated low level features of oriented faces, turned out to be unable to redirect spatial attention. In
other words, marmoset gaze following seems to require a more holistic perceptual interpretation of faces, based
on the integrative extraction of more than an isolated parameter.
Congruent choices are accompanied by faster reaction times already at short exposure
times. In the main experiment, the latencies of saccades indicating congruent choices were shorter than the
ones for incongruent choices for exposure times up to 400 ms duration (Fig.3). Actually, this facilitation eect
was strongest for the shortest exposure time, gradually decreased with exposure time and no longer reached sig-
nicance for the longest durations tested (500 and 600 ms; see Fig.3 legend for statistics), consistent with a time
course of reexive rather than volitional orienting. A similar facilitation eect for comparably short exposure
durations has been seen in studies of macaque monkeys
11
and humans
8
. However, these studies did not report a
gradual increase of reaction times with the time of exposure seen in our experiments on marmosets. is dier-
ence might be a consequence of the specic paradigm we used. In our experiments, the animals had to choose
between two targets of equal appearance, rather than to follow the other’s gaze to a specic target as in the work
on macaque monkeys and humans. Hence, our animals may have tended to extract additional information from
the other’s face beyond gaze direction in an attempt to facilitate their choices, provided that this portrait was avail-
able long enough. is increased interest in the other’s face, gated by longer exposure times, can be expected to
compromise the ability to quickly disengage attention at the time of the go-signal. is interpretation is supported
by the experiments with control stimuli and the eye movements prompted by the appearance of the portraits we
discuss below.
Saccades associated with the straight ahead face (“direct gaze”) exhibited latencies that were not dierent from
the ones associated with incongruent choices to oriented faces. Interestingly, latencies of saccades associated with
neutral disks showed an inuence of exposure time that was qualitatively opposite to the inuence on saccades for
congruent choices: while being similar to saccades for straight faces for short presentation durations, they became
shorter with increasing exposure time (see Fig.3). e same held for the bicolor disk control stimuli and for the
oriented triangles (Supplementary Fig.S2). ese results indicate that for marmosets, the attraction of the other’s
face but not the attraction of non-biological objects increases with time exposure and correspondingly attentional
disengagement is delayed. Neutral objects were associated with relatively long saccadic reaction times when pre-
sented briey, probably because of the need to scrutinize the object in order to assess its signicance. Once its
irrelevance is established aer some 200 ms of presentation, the observer disengages her/his attention in order to
prepare a fast saccadic choice. A short exposure to the oriented face can cause a profound shortening of saccadic
reaction time, because the drive to follow gaze direction is already fully expressed whereas facial attraction is still
building up. e idea that the development of facial attraction and in general the perception of faces may need a
much longer time is also supported by a consideration of the pattern of saccadic exploration of the portraits (see
5
Scientific RepoRtS | (2019) 9:15292 | https://doi.org/10.1038/s41598-019-51783-9
www.nature.com/scientificreports
www.nature.com/scientificreports/
Supplementary Fig.S3a,b for details), whose complexity keeps growing with exposure time. Hence, the question
is why the drive to follow gaze is fully expressed in saccadic reaction times for short exposure times, arguably
too short to allow a detailed scrutiny of the face whereas the choice bias increases further with exposure time for
up to 300 ms. We think that this dissociation between reaction times and choice probabilities might reect the
concerted action of two systems controlling gaze following. e rst is fast, probably subcortical, controlling gaze
following based on a rough and potentially error prone analysis of the other’s face, too limited to provide informa-
tion on other aspects of the face like the identity or mood of the agent. With longer exposure and concomitantly
processing time, this information becomes available, on the one hand binding attention but, on the other hand,
also improving the directional precision of decisions.
Conclusions
Gaze following is prevalent among numerous species but its strength and exibility varies substantially between
them
18
. As shown here, gaze following is also well developed in common marmosets, a new world monkey spe-
cies. Marmoset gaze following is characterized by strong similarities with the gaze following behavior of the two
old world primate species studied extensively, macaques and humans. e strongest argument for correspond-
ence is the similar dependence on the time of exposure to the other’s gaze direction. In all three species the other’s
gaze biases decisions on potential targets already aer only 100 ms of exposure to the other’s gaze, too short to
accommodate a more detailed scrutiny of the other’s face. However, given more time to explore the other’s face,
the bias gets stronger in all three, in line with the assumption that primate gaze following is a faculty, consisting of
an early reex-line component that is complemented by a later, more exible component, arguably also respon-
sible for the more sophisticated emotional and cognitive control known to modulate gaze following
19–21
. e
behavioral similarities between the gaze following behavior of marmosets, macaque monkeys and humans are
in principle in line with the assumption of a homologous faculty, already available before the split of the new and
old world monkey primate lines. is conclusion may strengthen the view that the marmoset may indeed become
a useful model system for research into the underpinnings of disturbed human social interactions like autism,
related to decient gaze following and joint attention
22
. However, although compelling, the behavioral similarities
established in our study may as well reect behavioral convergence. Hence, comparative physiological and genetic
studies of the underlying neural systems will be needed to strengthen the case for homology.
Materials and Methods
Experimental model and subject details. Common marmosets. We trained 3 adult common marmoset
monkeys (Callithrix jacchus; two females and one male, aged 7 years) to voluntarily enter a custom made monkey
chair by means of positive reinforcement training and to accept the restriction of head movements through a head
holder. Animals were all born in captivity and kept in a marmoset husbandry at approximately 26 °C, 40%–60%
relative humidity and a 12 h:12 h light-dark cycle. Access to water was always ad libitum, while food intake was
controlled according to body weight (weight loss never less than 10% of the ad libitum weight) and amount of
reward received in the experiment. Food consisted in fresh fruits and vegetables and standard commercial chow.
Figure 3. Oriented faces speed up reaction times for congruent choices. Plots of saccadic reaction times as
function of the duration of presentation of the central stimuli. Data pooled over the three animals (monkeys
M1, M2, M3) and choice direction. Saccadic reaction times indicating congruent choices were signicantly
shorter compared to the incongruent ones up to 400 ms of presentation duration (Wilcoxon rank-sum test,
100 ms: zval = −4.8221, p < 0.001; 200 ms: zval = −3.8449, p < 0.001; 300 ms: zval = −2.0341, p = 0.04; 400 ms:
zval = −2.4745, p = 0.01). e individual plots are tted with 2nd degree polynomial functions in an attempt to
improve the visibility. e two ts that showed a signicant dependence of saccade latency on the presentation
duration of the central stimulus were the one for congruent choices prompted by oriented faces and the one for
neutral disk stimuli. e former exhibited a gradual increase with duration from a substantially shorted reaction
time for a duration of 100 ms (adjusted r
2
= 0.86). e latter, on the other hand, showed a gradual decrease with
duration (adjusted r
2
= 0.89).