INFORME
Infants Learn to Follow Gaze in Stages: Evidencia
Confirming a Robotic Prediction
Priya Silverstein1,2
, Jinzhi Feng1,3, Gert Westermann1, Eugenio Parise1,4,
and Katherine E. Twomey5
1Psychology Department, Lancaster University, Reino Unido
2Institute for Globally Distributed Open Research and Education
3School of Psychology and Neuroscience, University of St Andrews, Reino Unido
4CIMeC, Center for Mind/ Brain Sciences, University of Trento, Italia
5Division of Human Communication, Development and Hearing, University of Manchester, Reino Unido
Palabras clave: cognitive development, developmental robotics, gaze following
ABSTRACTO
Gaze following is an early-emerging skill in infancy argued to be fundamental to joint
attention and later language development. Sin embargo, how gaze following emerges is a topic of
great debate. Representational theories assume that in order to follow adults’ gaze, infantes
must have a rich sensitivity to adults’ communicative intention from birth. A diferencia de,
learning-based theories hold that infants may learn to gaze follow based on low-level social
reforzamiento, without the need to understand others’ mental states. Nagai et al. (2006)
successfully taught a robot to gaze follow through social reinforcement and found that the
robot learned in stages: first in the horizontal plane, and later in the vertical plane—a
prediction that does not follow from representational theories. En el estudio actual, probamos
this prediction in an eye-tracking paradigm. Six-month-olds did not follow gaze in either the
horizontal or vertical plane, whereas 12-month-olds and 18-month-olds only followed gaze in
the horizontal plane. These results confirm the core prediction of the robot model, sugerencia
that children may also learn to gaze follow through social reinforcement coupled with a
structured learning environment.
INTRODUCCIÓN
Gaze following, or the ability to look where a social partner is looking, is a critical milestone in
human sociocognitive development (Tomasello, 1995). Although gaze following has been
found in many nonhuman species (p.ej., Bugnyar et al., 2004; Okamoto-Barth et al., 2007;
Povinelli et al., 1996; Tomasello et al., 2007), in human infants this skill has been argued
to be a fundamental component of socialization, and infants’ early gaze following has been
linked repeatedly to their later language development (p.ej., Arroyos & Meltzoff, 2005;
Carpenter et al., 1998; Morales et al., 2000). Gaze following appears to emerge from around
five months of age (Gredebäck et al., 2018; Senju & Csibra, 2008; Szufnarowska et al., 2014)
including in non-WEIRD ( Western, educated, industrialized, rich, y democrático) popula-
ciones (Hernik & Broesch, 2018). De este modo, in human infants, gaze following appears to be an
un acceso abierto
diario
Citación: Silverstein, PAG., feng, J.,
Westermann, GRAMO., Parise, MI., & Twomey,
k. mi. (2021). Infants Learn to Follow
Gaze in Stages: Evidence Confirming
a Robotic Prediction. Mente abierta:
Descubrimientos en ciencia cognitiva, 5,
174–188. https://doi.org/10.1162/opmi
_a_00049
DOI:
https://doi.org/10.1162/opmi_a_00049
Materiales suplementarios:
https://osf.io/fqp8z/;
https://osf.io/e73fw
Recibió: 15 Puede 2020
Aceptado: 27 Agosto 2021
Conflicto de intereses:
The authors declare no conflict
de interés.
Autor correspondiente:
Priya Silverstein
priyasilverstein@gmail.com
Derechos de autor: © 2021
Instituto de Tecnología de Massachusetts
Publicado bajo Creative Commons
Atribución 4.0 Internacional
(CC POR 4.0) licencia
La prensa del MIT
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
.
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
oh
pag
metro
_
a
_
0
0
0
4
9
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
early-emerging, universal ability, which is intimately linked to our uniquely human social
capacities.
While gaze following in infancy has been repeatedly demonstrated, the origins of this abil-
ity and the cognitive representations at play remain controversial. Some influential represen-
tational theories assume that gaze following relies on infants’ capacity (not learned and present
from birth) to be sensitive to others’ communicative intentions or mental states (p.ej., Baron-
cohen, 1995; Csibra & Gergely, 2009). On these accounts, infants are born with a receptivity
to adults as intentional communicative agents, and based on this, look where adults look in
order to obtain information. In support of these accounts are studies that suggest infants may
have some rudimentary form of gaze following (gaze cueing) from birth, evidenced by the
finding that newborns detect an object on a screen faster if it appears in a location previously
cued by another’s gaze (Farroni et al., 2004). Others argue that although this is not an ability
that is present from birth, at a certain time point the ability to read others’ intentions and hence
follow gaze “switches on,” supported by the finding that 6-month-old infants need the pres-
ence of ostensive cues (direct gaze and infant directed speech) in order to follow gaze (Hernik
& Broesch, 2018; Senju & Csibra, 2008).
Sin embargo, whether infants’ ability to follow gaze is based on their existing understanding
that adults are intentional communicative agents has been subject to debate. Por ejemplo,
gaze cueing in newborns is not robust and is found only when the motion of the schematic
gaze shift is visible (Farroni et al., 2004), leaving open the possibility that apparent gaze
following reflects merely a sensitivity to motion cues. Similarmente, although 6-month-olds fol-
low gaze in the presence of ostensive cues (Senju & Csibra, 2008), this has also been found
for the non-communicative action of shivering (Gredebäck et al., 2018; Szufnarowska et al.,
2014) and in a larger study (with the power to detect smaller effects), even without any cue
en absoluto (Gredebäck et al., 2018), suggesting that infants may simply be more likely to gaze
follow even in noncommunicative situations in which their attention has been captured.
Similarmente, although the finding that 5- to 7-month-old infants in Vanuatu, a culture with fewer
face-to-face parent-infant interactions, show gaze following (Hernik & Broesch, 2018) can be
used to argue for a capacity for this ability that is present from birth, this is cautioned against by
the authors themselves, and gaze following in this paradigm (originally used by Senju & Csibra,
2008) could also be explained by motion tracking, or following of the face contour (von Hofsten
et al., 2005). These and similar results highlight the importance of designing paradigms that
assess gaze following absent of these confounds.
In contrast to these representational accounts, learning-based accounts of gaze following
assume that infants learn this skill via reinforcement learning. Específicamente, during interaction,
a caregiver may avert their gaze to look at something interesting entering the environment. Si
the infant by chance looks in the same direction, they are rewarded by the interesting in-
formación, and will be more likely to look in the same direction as the caregiver in the future
(p.ej., Triesch et al., 2006). In this scenario, social scaffolding by the caregiver, Por ejemplo,
holding objects directly in the infant’s field of view and rapid alternation of head turns from
infant to object, may increase opportunities for learning (moore & Corkum, 1994).
Alternativamente, infants may learn to follow gaze by initially learning that the caregiver’s hands
are a reliable cue to the focus of the caregiver’s attention (Deák et al., 2014; Yu & Herrero,
2013). Although the mechanisms proposed in learning-based theories differ, críticamente, todo
share the assumption that, at least early in development, associative learning and reinforce-
ment are sufficient to support the emergence of gaze following given a suitably structured
learning environment—without the need for any specific, preexisting sensitivities to commu-
nicative or referential intent.
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
175
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
.
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
oh
pag
metro
_
a
_
0
0
0
4
9
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
In the current article we focus on the predictions of a developmental robotic simulation of
the emergence of gaze following based on low-level learning mechanisms. Nagai et al. (2006)
successfully taught a developmental robot with a simple neural network cognitive architecture
to gaze follow through supervised associative learning. The robot was equipped with a camera
that fed images of a human experimenter to a visual system consisting of an artificial neural
network map, which encoded these images, and a retinal smoothing layer, simulating the de-
velopment of infants’ visual acuity. Training consisted of five steps. Primero, the robot’s camera
was focused on the experimenter’s face. Próximo, the experimenter held up an object and shifted
her gaze toward it. Visual input from the camera was then processed by the neural network
after smoothing, generating a representation of the experimenter’s face. Based on this repre-
sentation, the robot generated a motor command, adjusting the joint angles in its head and
neck, resulting in a head turn and a change in its visual field. The robot was then given feed-
back based on the output error between the location of the object in the visual field and its
gaze direction: if the object was centered in the visual field, gaze following was considered
successful and no adjustments to the neural network were made. When the object was outside
the visual field or off-center, random noise was added to the connection weights in the robot’s
neural network. Across training, por lo tanto, head movements resulting in incorrect gaze follow-
ing were less likely to be produced, increasing the relative strength of connections which pro-
duced correct gaze following.1 Following training, testing with previously untrained images
demonstrated that the robot learned to follow the experimenter’s gaze. De este modo, this work raises
the possibility that low-level perceptual and proprioceptive information coupled with social
reinforcement are sufficient to support the emergence of this important ability.
Sin embargo, testing during training revealed that in the robot, the ability to follow gaze de-
veloped in stages: early in learning, the robot initially learned to follow gaze in the horizontal
plane, and only later in the vertical plane. The mechanism underlying this developmental tra-
jectory relates to characteristics of the visual input to the system. Específicamente, there was more
variability in the images that varied along the horizontal plane, providing a richer input to
learning than the less variable vertical plane. De este modo, this work makes the empirically testable
prediction that, if infants also learn to gaze follow on the basis of similar low-level mecha-
nisms, they too should initially follow gaze shifts more successfully in the horizontal than
the vertical direction. En tono rimbombante, the same prediction does not follow clearly from represen-
tational theories that assume an ability to understand others’ communicative intent, y ahí-
fore predict that infants should be able to follow gaze irrespective of gaze direction.
It is important to note that, when evaluating low-level and higher level accounts (also called
“lean” and “rich” interpretations; Haith, 1998) of infant cognition against each other, evidencia
for a low level explanation is usually also compatible with a high-level account (but not vice
versa). Por ejemplo, in the case of gaze following, if infants’ behavior can be explained on the
basis of reinforcement and learning, infants may nevertheless also understand intentions. Es
therefore not possible to directly falsify high-level accounts by showing that data are compat-
ible with low-level accounts; it is only possible to show that a certain behavior is compatible
with low-level accounts, therefore making the assumption of more complex high-level abilities
unnecessary. In the current case, we know that the robot model, which exhibited the
horizontal-then-vertical gaze following behavior, did not have any understanding of intention-
ality. To our knowledge, there is no other account making the same prediction, and specifi-
cally, it does not seem that any account assuming intention understanding in infants would
1 A formal description of the model is outside the scope of the current article; for details see Nagai et al.
(2006).
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
176
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
.
/
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
oh
pag
metro
_
a
_
0
0
0
4
9
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
make this prediction. críticamente, an important means to verify the validity of a computational
model of developmental processes is to test if its predictions are borne out in studies with chil-
niños (Asada et al., 2009; Morse & Cangelosi, 2017; Schlesinger & McMurray, 2012;
Westermann & Mareschal, 2012). In the present study we therefore provide this critical test.
Específicamente, we ask whether, like the robot, infants learn to gaze follow in stages, acquiring
horizontal gaze following before vertical gaze following. If this is the case, this would confirm
a central prediction of the robot model.
A pilot study showed that 12-month-old infants were able to follow the gaze direction in
our stimuli and were not at ceiling or floor. This age group are therefore at an intermediate
developmental stage in which gaze following is not consistently accurate, raising the possibil-
ity that at this age, we may observe the differences in horizontal/vertical tracking predicted by
the robot. Por lo tanto, in Experiment 1 participants were 12-month-old infants. In order to in-
vestigate the developmental trajectory of gaze following, we then tested 6-month-olds
(Experimento 2) and 18-month-olds (Experimento 3). This study was preregistered (https://osf.io
/e73fw) and all materials, código, and data can be found in the Supplemental Materials on the
Open Science Framework (OSF; https://osf.io/fqp8z/).
EXPERIMENT 1 (12-MONTH-OLDS)
Método
Ethics. All data were kept confidential. All three experiments were approved by the university
ethics committee and adhered to the British Psychological Society guidelines.
Participantes. Participants were recruited from a database of families at Lancaster University
Babylab, and were given a book as a gift for participation and £10 travel reimbursement.
Although an individual measure of socioeconomic status was not collected, the average in-
come in Lancaster is £24,000 annually, which is slightly less than the national average
(£29,000). Infants were not recruited for the study if they had any developmental delays or vi-
sual impairments. Parents gave informed, written consent before participation, and were free to
withdraw their consent. Sample size was determined using Bayesian sequential testing
(Schönbrodt et al., 2017). In Bayesian sequential testing, Bayes factors are calculated after every
partícipe (pending a predefined minimum number of participants) until a preregistered Bayes
factor threshold is reached for all analyses run (or a predefined maximum number of partici-
pants is reached). We determined a minimum a priori sample size of 20 infantes, with a maxi-
mum of 40 infantes, and set a threshold of a BF10 (es decir., evidence for the research hypothesis) y
BF01 (es decir., evidence for the null) de 10 or above for all preregistered analyses. We did not reach
this threshold for all tests when we reached our maximum number of infants. Por lo tanto, nuestro
final sample consisted of 40 typically developing 12-month-old infants (mean age: 364 días;
range: 349 days to 383 días; 27 femenino; all Caucasian; 34 monolingual English). Infants were
excluded if they provided less than 50% of usable trials in which they were not crying (norte = 4).
Stimuli and Design. Example stimuli are depicted in Figure 1. To construct our stimuli we used
five photographs of one female face looking left, bien, arriba, abajo, and directly at the camera.
The eyes from the left, bien, arriba, and down photographs were superimposed onto the face
looking directly at the camera, ensuring the face was identical apart from gaze direction.
The use of static images (as opposed to videos) has been shown to effectively elicit gaze fol-
lowing in 12-month-olds (von Hofsten et al., 2005). Target objects were selected from the
NOUN database (Horst & Hout, 2016) that had no obvious top/bottom in order to avoid
biasing infants’ attention.
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
177
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
.
/
/
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
oh
pag
metro
_
a
_
0
0
0
4
9
pag
d
.
/
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
Cifra 1.
Experimental and control conditions. Central cue (direct gaze or fixation cross) para
1,000 EM, followed by a directional cue (averted gaze or object) para 2,000 EM. Blue boxes indi-
cate areas of interest (AOIs) (visual angle of center AOI = 9.57° × 9.57°; visual angle of object
AOI = 9.57° × 8.89°).
Trials were split into control trials and experimental trials. Each trial consisted of a 3-s-long
video. Control trials consisted of a central fixation cross (1,000 EM) followed by a novel object
appearing in one of four locations at 200 pixels left, bien, arriba, or down from the center of the
pantalla (2,000 EM), and were designed to test whether infants found gaze shifts (without gaze
following) a priori easier in the horizontal than the vertical plane. Experimental trials consisted
of a human face (visual angle: 12.96° × 8.89°) looking directly at the infant, surrounded by
four images of the same object in each of the locations (1,000 EM) followed by the same face
with averted gaze to one of the four exemplars (2,000 EM). In total, 64 videos were made,
consisting of the eight objects (average visual angle = 4.11° × 4.79°) in the four locations
for both the control and experimental conditions. These videos were pseudorandomized into
four orders such that infants never saw the same object, ubicación, or trial type (control, experto-
imental) on more than two successive trials. Between trials there was a variable animated
attention-getter. All videos are available on OSF.
Infants sat on their caregiver’s lap during the experiment approximately 0.6 meters
Procedimiento.
away from a 23-in. pantalla. Parents were instructed to close their eyes during the experiment to
ensure they were blind to the stimuli. An eye-tracker (Tobii X120) captured infant looking
times and gaze locations on screen. We used Tobii Studio 3.3.1 to present stimuli and gather
eye-tracking data. We performed a 5-point calibration for all infants before beginning the
experimento. After this calibration, we instructed caregivers not to talk to or interact with their
infant, and that they could stop the experiment at any time if the infant became too fussy,
and the experiment began. Infants saw up to a maximum of 64 trials in one of the four
pseudorandomized orders (range: 28–64, m = 57; DE = 9).
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
178
yo
D
oh
w
norte
oh
a
d
mi
d
F
r
oh
metro
h
t
t
pag
:
/
/
d
i
r
mi
C
t
.
metro
i
t
.
/
mi
d
tu
oh
pag
metro
i
/
yo
a
r
t
i
C
mi
–
pag
d
F
/
d
oh
i
/
i
/
/
.
1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
oh
pag
metro
_
a
_
0
0
0
4
9
pag
d
/
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
Análisis. We performed all analyses in R 3.5.2 (R Core Team, 2017). We exported raw data
from Tobii Studio 3.3.1 and analyzed them using a combination of the eyetrackingR package
(Dink & Ferguson, 2015) and our own code (all code is available on the OSF). Areas of interest
were defined for center (400 × 400 píxeles), and left, bien, arriba, and down object locations (cada
340 × 400 píxeles). All preregistered analyses are reported either in the main article or in the
Supplemental Materials on OSF. Our outcome variables included two measures of accuracy:
proportion looking (length of looking at target AOI divided by total looking at all four object
AOIs); dichotomous target looking (whether or not the infant looked at the target AOI for any
frames after the gaze shift); and a measure of latency (time taken for the infant to look at the
target AOI). Corresponding Bayesian analyses were carried out for all frequentist analyses
(Dienes & Mclatchie, 2018; Wagenmakers et al., 2018). All Bayesian analyses used a default
Bayes factor with a wide Cauchy distribution (scale of effect = 0.707) and were calculated
using the BayesFactor R package (Morey & Rojo, 2015). For all Frequentist analyses, nosotros
used a significance threshold of p < .05. BF10 or BF01 of more than 3 are considered moderate
evidence, using the system outlined by Jeffreys (1961). Raincloud plots were created adapting
code from Allen et al. (2019).
Results
Accuracy. We submitted proportion looking to a 2 × 2 repeated measures ANOVA with main
effects of condition (gaze vs. control) and plane (horizontal vs. vertical) (Figure 2; see Figure 3
for gaze across individual conditions). Accuracy was higher for control than gaze trials—main
effect of condition: F(1, 39) = 1,222, p < .001; η2 = .85; BF10 = 2.93e63, and higher for the
horizontal than the vertical plane—main effect of plane: F(1, 39) = 9.02, p < .005; η2 = .01;
BF10 = 3.08. The ANOVA revealed a significant interaction between condition and plane,
F(1, 39) = 23.72, p < .001; η2 = .02; BF10 = 1.65. Planned, two-tailed, paired t tests revealed
that in the gaze condition, accuracy was higher for the horizontal, M = 0.34, SD = 0.15,
than the vertical plane, M = 0.18, SD = 0.14; t(39) = 4.14, p < .001; d = 1.09; BF10 = 2,774.
In contrast, in the control condition, there was no evidence for a difference in accuracy
across the two planes—horizontal: M = 0.90, SD = 0.11; vertical: M = 0.92, SD = 0.08;
t(39) = −1.77, p = .08; d = −.26; BF01 = 2.35. Finally, t tests against chance (0.5) revealed
that in the gaze condition, accuracy was above chance for the horizontal plane, t(39) = 3.72,
Figure 2. Raincloud plot showing proportion target looking across conditions (gaze vs. control)
and planes (horizontal vs. vertical) for 12-month-olds. Dashed line indicates chance performance.
OPEN MIND: Discoveries in Cognitive Science
179
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
Figure 3. Raincloud plot showing proportion target looking across conditions (gaze vs. control)
and positions (down, left, right, up) for 12-month-olds.
p < .001; d = .83; BF10 = 70, and below chance for the vertical plane, t(39) = −3.14, p < .01;
d = −.70; BF10 = 15. Dichotomous target looking results mirrored the proportion looking results
and can be found in the Supplemental Materials on OSF (Figure S1).
Latency. For trials where infants looked at the target AOI at any point after the gaze shift, we
submitted latency to a 2 × 2 repeated measures ANOVA with main effects of condition (gaze
vs. control) and plane (horizontal vs. vertical) (Figure S4). The ANOVA revealed moderate
evidence for no interaction between condition and plane for latency, F(1, 36) = 0.02, p =
−5, BF01 = 5.66. Latency was shorter for control than gaze trials—main effect
.90; η2 = 8.71e
−1, BF10 = 283—and there was moderate
of condition: F(1, 36) = 20.49, p < .001; η2 = 1.03e
evidence for no difference in latency in the horizontal and vertical planes—main effect of
plane: F(1, 36) = 0.001, p = .98; η2 = 6.00e
−6, BF01 = 5.71.
Discussion
In this experiment we tested the predictions of Nagai et al.’s (2006) robotic implementation of
the development of gaze following, specifically that infants in the early stages of gaze follow-
ing should first learn in the horizontal plane, and then the vertical plane. Our results capture
this prediction: they suggest that 12-month-old infants are able to follow horizontal gaze shifts
from static images of a face, whereas they are not yet able to do this for vertical gaze shifts. We
find no such difference in our control condition, where objects appear in the same locations
but are not cued with gaze. We find that latency is faster in the control than the gaze condi-
tion, which could be due to it being easier to disengage from the fixation cross than the face,
and the lack of competing stimuli to look at. We find no difference in latency across planes,
which suggests that infants are not slower to make gaze shifts in either plane. Importantly, the
results of our control condition indicate that the difference found in the experimental condi-
tion is not because infants have difficulty shifting attention to salient vertical targets. Our re-
sults are compatible with learning-based theories, specifically Nagai and colleagues’ robotic
implementation, in which a combination of associative learning mechanisms and social rein-
forcement were sufficient to support the emergence of gaze following. In contrast, represen-
tational accounts offer no mechanism by which a difference between horizontal and vertical
gaze following might arise. Importantly, however, to strengthen the learning-based account of
OPEN MIND: Discoveries in Cognitive Science
180
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
.
/
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
gaze following, we should be able to demonstrate developmental changes in horizontal and
vertical gaze following. Specifically, if gaze following is learned in stages (rather than there
being a general bias for horizontal gaze in infants), we should be able to find a stage at which
younger infants do not follow either horizontal or vertical gaze. Thus, in Experiment 2, we
tested 6-month-old infants in the same paradigm.
Although in Experiment 1, when using Bayesian sequential testing we reached our maxi-
mum number of infants, this was due to not meeting our threshold for all analyses. This is due
to the fact it takes a larger amount of evidence to reach any given Bayes factor in support of the
null hypothesis than against it ( Johnson & Rossell, 2010). However, our frequentist results
were the same at 20 participants as at 40 participants, that is, all results significant at the
p < .05 threshold at 40 participants also were significant at 20 participants. This is likely due
to the high number of trials (64) compared to many infancy studies. Hence, in each further
experiment reported, we collected a final sample of 20 infants.
EXPERIMENT 2 (6-MONTH-OLDS)
Method
All stimuli and procedures were identical to Experiment 1. Infants viewed up to 64 trials
(range: 28–64, M = 57; SD = 10). As we did not observe a difference between horizontal
and vertical gaze following latency in Experiment 1 (12-month-olds), we did not analyze
latency in Experiments 2 and 3 (6- and 18-month-olds, respectively).
Participants. Twenty typically developing 6-month-old infants took part (mean age: 182 days;
range: 166 days to 196 days; 7 female; all Caucasian; 18 monolingual English). Participants
were from the same demographic background as participants in Experiment 1.
Results
We submitted proportion looking to a 2 × 2 repeated measures ANOVA with main effects of
condition (gaze vs. control) and plane (horizontal vs. vertical) (Figure 4). Accuracy was signif-
icantly higher for control than gaze trials—main effect of condition: F(1, 19) = 175, p < .001;
Figure 4. Raincloud plot showing proportion target looking across conditions (gaze vs. control)
and planes (horizontal vs. vertical) for 6-month-olds. Dashed line indicates chance performance.
OPEN MIND: Discoveries in Cognitive Science
181
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
.
/
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
η2 = .75; BF10 = 5.09e21, and we found moderate evidence for no difference between hori-
zontal and vertical accuracy—main effect of plane: F(1, 19) = 0.08, p = .78; η2 = .0002; BF01 =
4.28. The ANOVA revealed moderate evidence for no interaction between condition and
plane for proportion looking—horizontal gaze: M = 0.33, SD = 0.19; vertical gaze: M =
0.34, SD = 0.23; horizontal control: M = 0.91, SD = 0.13; vertical control: M = 0.87, SD =
0.22; F(1, 19) = 1.34, p = .26; η2 = .004; BF10 = 3.81. Finally, planned t tests found no ev-
idence for accuracy at levels greater than expected by chance for either plane in the gaze
condition—horizontal: t(19) = 0.36, p = .73; d = .11; BF01 = 3.12; vertical: t(19) = −1.13, p =
.27; d = −.36; BF01 = 1.96. Dichotomous target looking results mirrored the proportion look-
ing results and can be found in the Supplemental Materials on OSF (Figure S2).
We note that it is possible that for 6-month-olds true effect sizes are smaller, and so we
cannot guarantee that a larger sample of infants would not have found effects where we
found null results. However, all Bayes factors but one were moderate (BF01 of above 3),
making this interpretation less likely. The only Bayes factor that was inconclusive is for ver-
tical gaze against chance, which is qualitatively below chance level (as is horizontal gaze
against chance).
In order to assess whether 6-month-olds were indeed perform-
Comparison With Experiment 1.
ing differently to 12-month-olds, we submitted proportion looking in gaze trials to a 2 × 2
mixed ANOVA with main effects of age (12 months vs. 6 months) and plane (horizontal vs.
vertical). Although according to our frequentist analysis the ANOVA did not reveal a signifi-
cant interaction between age and plane for proportion looking, Bayes factor analysis revealed
strong evidence for an interaction, F(1, 58) = 2.73, p = .10; η2 = 0.03; BF10 = 14.56.
Discussion
In Experiment 2 we tested whether 6-month-old infants could follow gaze in either the hori-
zontal or vertical plane. In contrast to 12-month-olds, this younger group showed no differ-
ence in their horizontal and vertical gaze following and did not show above-chance accuracy
in either plane. As in Experiment 1, these results capture the predictions of the robotic model
presented by Nagai et al. (2006), suggesting that younger infants have not yet learned to follow
gaze in either plane. Thus, the difference in gaze following ability across the horizontal and
vertical planes that we find in 12-month-olds is not an artifact of our stimuli (as the stimuli used
were identical). Further, infants do not exhibit a general bias to follow only horizontal gaze,
strengthening our interpretation that infants may learn to follow gaze in stages.
In Experiment 3, we investigated whether, as we predicted according to the robot model,
older infants can follow gaze equally successfully in the horizontal and vertical plane.
EXPERIMENT 3 (18-MONTH-OLDS)
Method
All stimuli, procedures, and analyses were identical to Experiments 1 and 2. Infants viewed up
to 64 trials (range: 42–64, M = 60; SD = 7).
Participants. Twenty typically developing 18-month-old infants took part (mean age: 546
days; range: 530 days to 567 days; 10 females; all Caucasian; all monolingual English).
Participants were from the same demographic background as participants in Experiment 1.
Two additional infants were excluded due to fussiness.
OPEN MIND: Discoveries in Cognitive Science
182
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
Results
We submitted proportion looking to a 2 × 2 repeated measures ANOVA with main effects of
condition (gaze vs. control) and plane (horizontal vs. vertical) (Figure 5; see Figure 6 for gaze
across individual conditions). Accuracy was greater for control than gaze trials—main effect of
condition: F(1, 19) = 356, p < .001; η2 = .82; BF10 = 7.38e26—and greater for the horizontal
than the vertical plane—main effect of plane: F(1, 19) = 9.37, p < .006; η2 = .02; BF10 = 2.40.
The ANOVA revealed a significant interaction between condition and plane for proportion
looking, F(1, 19) = 9.92, p < .01; η2 = .01; BF10 = 2.83. Planned, two-tailed, paired t tests
revealed that in the gaze condition, accuracy was higher for the horizontal, M = 0.34, SD =
0.13, than the vertical plane, M = 0.17, SD = 0.13; t(19) = 3.52, p < .005; d = 1.24;
BF10 = 72.68, with moderate evidence for no difference in the control condition—horizontal:
M = 0.87, SD = 0.14; vertical: M = 0.86, SD = 0.14; t(19) = 0.53, p = .60; d = .10; BF01 = 3.13.
Finally, t tests against chance (0.5) revealed that in the gaze condition, accuracy was signifi-
cantly above chance for the horizontal plane, t(19) = 2.93, p < .01; d = .93; BF10 = 7.66, and
significantly below chance for the vertical plane, t(19) = −2.61, p < 0.05; d = −.82; BF10 = 4.07.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
.
/
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 5. Raincloud plot showing proportion target looking across conditions (gaze vs. control)
and planes (horizontal vs. vertical) for 18-month-olds. Dashed line indicates chance performance.
Figure 6. Raincloud plot showing proportion target looking across conditions (gaze vs. control)
and positions (down, left, right, up) for 18-month-olds.
OPEN MIND: Discoveries in Cognitive Science
183
Gaze Following in Infants
Silverstein et al.
Dichotomous target looking results mirrored the proportion looking results and can be found on
OSF (Figure S3).
Discussion
In this experiment, we tested whether 18-month-infants were yet able to follow gaze in both
the horizontal and vertical plane. We found that they showed the same pattern as the 12-month-
olds and were only able to follow gaze in the horizontal plane. Thus, in Experiment 3 we did not
identify the age at which infants have acquired both horizontal and vertical gaze following
abilities. Critically, however, we replicated the 12-month-old results with 18-month-olds;
the results of Experiment 3 are therefore consistent with an account of gaze following in which
infants learn in stages, suggesting that infants remain at the horizontal stage in this age group.
Note that in all three experiments we investigate gaze following of just the eye gaze shift,
which is a much more subtle cue than the more commonly used head turn and eye gaze shift
(Moore et al., 1997). Interestingly, we found that rather than being at chance, infants were
significantly below chance in their vertical gaze following. Tentatively, we suggest that this
could be due to infants persevering with horizontal saccades after successfully following gaze
in the horizontal condition, but future research would be needed to determine whether this
unexpected effect could even be replicated, and if so, what the mechanism may be.
GENERAL DISCUSSION
In the current study we tested the predictions of Nagai et al.’s (2006) robotic implementation of
the development of gaze following. In Nagai et al. (2006), the robot learned to gaze follow
based on associative learning and visuomotor input and did so in stages: first in the horizontal
plane, followed by the vertical plane. Our results capture this prediction: while 6-month-old
infants are unable to gaze follow in either plane in our paradigm, by 12 months they are able
to follow gaze in the horizontal plane, but critically, not the vertical plane. This pattern is,
unexpectedly, replicated in 18-month-olds, suggesting that the ability to follow vertical gaze
shifts without a head turn may develop after this age. We find no such difference in our control
conditions, where objects are not cued with gaze; thus, the difference we observe in the ex-
perimental conditions is specific to gaze following and is not driven by motoric differences in
the ease of horizontal and vertical eye movements. Our results are therefore compatible with
the robotic prediction and learning-based theories of gaze following, suggesting that a com-
bination of associative learning mechanisms and social reinforcement may be sufficient for the
emergence of gaze following. Importantly, Nagai and colleagues’ model offers an explicit
mechanism by which the horizontal-first gaze following behavior we observed may emerge.
In the robotic model, visual input in the horizontal plane was more variable than visual input
in the vertical plane. The model therefore used this additional information for learning in the
first instance, leading to better gaze following in the horizontal plane early in learning. Thus,
taken together with our replication of the model’s behavioral prediction, this work makes the
further, empirically testable prediction that gaze input to infants, in particular gaze direction
information, should be more variable in the horizontal than the vertical plane.
While the current data offer support for learning-based accounts of gaze following, we can-
not rule out alternative accounts based on cognitive processes present from birth. In particular,
the shape of the human eye is such that horizontal eye movements are easier to see, due to the
amount of visible white sclera being larger and therefore there being a clearer contrast be-
tween iris and sclera. For this reason, one could argue that the perceptual system is simply
tuned to spot more perceptually contrasting movements (and for evidence of gaze cueing in
OPEN MIND: Discoveries in Cognitive Science
184
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
newborns, see Farroni et al., 2004). However, if this were driving our results, in the current
study we might expect to find better gaze following for upward vertical gaze (where there is
more visible sclera) than the downward vertical gaze (where no sclera is visible), which is not
the case. In fact, we find worse accuracy for upward looks than all three other positions
(Figures 3 and 6), in contrast to an upward bias found in adults (Bock et al., 2008).
Intuitively, this is consistent with learning theories, as infants are typically situated below
adults’ line of gaze, which would elicit more frequent downwards than upwards looks from
the adult. In the same vein, it is also possible that as well as being more perceptually variable,
horizontal gaze is just more common in input to infants. It is also important to note that infants
can only see an adult’s gaze cues when the adult is facing the infant. In this context, adults
may be more likely to exhibit lateral gaze behavior (to the right or left of the infant) as opposed
to looking above (i.e., behind) the infant as they engage with them. Gaze behavior while en-
gaging with the infant is also likely to be followed by the adult labeling an object or the infant
turning to look in the direction of the adult’s gaze. As a result, this adult gaze bias along the
horizontal plane, coupled with feedback from the environment about the lateral object to
which the adult is looking, could serve as an associative mechanism for the horizontal gaze
bias. Again, empirical evidence from naturalistic input to infants can assess these possibilities.
It is also possible that other morphological features of human eyes or face (e.g. the fact that
eyes are arranged in the horizontal axis themselves) are responsible for this effect in human
infants, however this would likely affect all eye movements, and so induce a bias in our con-
trol condition (which it does not).
Nonetheless, while elements of the learning environment (such as increased variability in
the horizontal plane) may shape infants’ behavioral responses, they may still possess an ability
to read others’ mental states. However, theories proposing that infants learn to follow gaze by
understanding the communicative intentions of others do not offer a mechanism that could
explain our observation that infants learn to follow gaze earlier in the horizontal than the ver-
tical plane. For example, as our stimuli begin with an ostensive cue (direct gaze), Natural
Pedagogy theory (Csibra & Gergely, 2009, 2011) would predict instead that even the 6-
month-olds in Experiment 2 would follow gaze, due to the referential expectation induced
by the ostensive cue. We cannot rule out the possibility that some other, representation-based
mechanism biases gaze following to horizontal planes. However, we are not aware of any
representational theory making this prediction.
Importantly, we know that a learning theory without intention understanding is sufficient
to produce this behavior (Nagai et al., 2006), and we interpret our results within a theory
that is as simple as possible, and as complex as necessary. We may, however, consider what
evidence would be needed to be able to provide definitive support for a capacity present
from birth to understand others’ intentions. The strongest evidence would come from evi-
dence of newborn gaze following, which to date has not been found. Newborn gaze cueing
has been observed (Farroni et al., 2004), but this is not the same as gaze following, and can
be explained by a much simpler mechanism. In these studies, a gaze shift is followed by a
target appearing on a cued or uncued side. Infants are found to be faster to detect targets
that appear on the cued side, which is taken by some to be evidence of rudimentary gaze
following ability. However, in these studies infants could be tracking the motion of the gaze
shift, and are simply quicker to fixate on a cued target because it is a shorter distance from
the cued target than the uncued target to the location of the termination of the movement.
Beyond this, it is very difficult (if not impossible) to measure intention understanding itself
without a verbal response.
OPEN MIND: Discoveries in Cognitive Science
185
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
.
/
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
Our paradigm could be altered (by accessing our open materials on the OSF) in many ways
to gain further theoretical understanding of what this horizontal bias in infant gaze following
means. Although our chosen control condition controls well for any horizontal bias in infant
eye movements, other theoretically interesting comparisons could be made to our experimen-
tal condition. Using a control that was non-social but still directed infant attention to one of
four possible objects would reveal whether infants find it more difficult to disengage from the
face vertically than horizontally, independent of social cues. One way to test this would be to
have a central face with direct gaze remaining on screen, while one of the four targets flashed.
In addition, our stimuli (by design) were artificial, and so have many features that make
them unlike naturalistic social interactions. The face was artificially static and presented alone
without a body, the gaze shift was engineered from two pictures as opposed to being a video,
the scene was absent of natural clutter, and trials were repetitive. However, we do not believe
that the repetitive nature of the paradigm affected our results, as we do not see behavior
changing in a systematic way across trials (see Figures S5–S7). Fine-grained control of the stim-
uli enabled us to effectively investigate the difference between horizontal and vertical gaze
following while ensuring horizontal and vertical stimuli were as visually similar as possible.
However, future research should investigate whether our findings can be replicated in more
naturalistic scenarios.
It is interesting to note that infants in our study are ‘bad’ at gaze following; that is, even the
18-month-olds were only above chance for horizontal gaze following. This is to be expected,
as it has been shown repeatedly that infants respond best to head and eye movements in com-
bination (Michel et al., 2021; Moore et al., 1997). The movement of just the eyes is a very
subtle cue and so makes it understandably more difficult for infants to disengage from the face,
where infants are known to fixate for the majority of the time even when there is a combined
head and gaze shift (e.g., von Hofsten et al., 2005). We chose to manipulate only eyes in this
experiment in order to create stimuli that were as controlled as possible, and to address a pos-
sible confound of combined head and gaze cue stimuli. Studies where head and gaze cues are
confounded leave open the possibility that infants may just be attracted to the face contour
itself, and so look in the direction that the head is turned, leading to the object in their periph-
eral vision being closer and more likely to catch their attention (von Hofsten et al., 2005).
Nevertheless, future research should explore whether our findings can be replicated for com-
bined head movement and eye cues, since, if gaze following is learned, these additional cues
should also be integrated by the learning mechanism.
Finally, it is possible that factors affecting gaze following also affect infant perception of
other object related actions, for example, object tracking. Indeed, it has been found that 5-
to 9-month-old infants are better at object tracking in the horizontal than vertical plane
(Grönqvist et al., 2006). Further research could determine whether these biases are due to a
common underlying factor (for example, visual input of a wide variety of scenes being more
variable in the horizontal plane, not just gaze events), or whether these are unrelated
phenomena.
CONCLUSION
Taken together, our results and those of Nagai et al. provide evidence that a low-level learning
mechanism provided with structured input may be sufficient to support the development of
gaze following in human infants. Importantly, however, current prominent theories in devel-
opmental psychology (e.g., Baron-Cohen, 1995; Csibra & Gergely, 2009) do not predict that
there would be differences in accuracy of gaze following across the horizontal and vertical
OPEN MIND: Discoveries in Cognitive Science
186
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Gaze Following in Infants
Silverstein et al.
planes, as an understanding of shared attention or communicative intentions (the basis for
these theories) is not dependent on direction of gaze. As such, the current results point to
new opportunities for the development of mechanistic and mentalistic theories that can account
for these behavioral data.
ACKNOWLEDGMENTS
We thank Anna Barnett for appearing in the stimuli, Malcolm Wong for help with data collec-
tion, Katharina Kaduk for her role as lab manager at Lancaster Babylab, and the infants and
caregivers who took part in the study.
FUNDING INFORMATION
PS, Leverhulme Trust (https://dx.doi.org/10.13039/501100000275), Award ID: DS-2014-014.
(https://dx.doi.org/10.13039
GW, EP, Economic and Social Research Council
/501100000269), Award ID: ES/L008955/1. GW and KET, Award ID: ES/S007113/1.
AUTHOR CONTRIBUTIONS
PS: Conceptualization: Equal; Data curation: Lead; Formal analysis: Lead; Investigation:
Lead; Methodology: Lead; Project administration: Lead; Visualization: Lead; Writing – original
draft: Lead; Writing – review & editing: Lead. JF: Investigation: Supporting; Project admin-
istration: Supporting; Writing – original draft: Supporting. GW: Funding acquisition: Lead;
Methodology: Equal; Supervision: Equal; Writing – review & editing: Equal. EP: Funding
acquisition: Equal; Methodology: Equal; Supervision: Equal; Writing – review & editing:
Equal. KET: Conceptualization: Equal; Methodology: Equal; Supervision: Lead; Writing – review
& editing: Equal.
REFERENCES
Allen, M., Poggiali, D., Whitaker, K., Marshall, T. R., & Kievit, R. A.
(2019). Raincloud plots: A multi-platform tool for robust data vi-
sualization [version 1; peer review: 2 approved]. Wellcome
Open Research, 4, Article 63. https://doi.org/10.12688
/wellcomeopenres.15191.1, PubMed: 31069261
Asada, M., Hosoda, K., Kuniyoshi, Y., Ishiguro, H., Inui, T.,
Yoshikawa, Y., Ogino, M., & Yoshida, C. (2009). Cognitive de-
velopmental robotics: A survey. IEEE Transactions on
Autonomous Mental Development, 1(1), 12–34. https://doi.org
/10.1109/TAMD.2009.2021702
Baron-Cohen, S. (1995). The eye direction detector (EDD) and the
shared attention mechanism (SAM): Two cases for evolutionary
psychology. In C. Moore & P. J. Dunham (Eds.), Joint attention:
Its origins and role in development (pp. 41–59). Lawrence
Erlbaum Associates.
Bock, S. W., Dicke, P., & Thier, P. (2008). How precise is gaze fol-
lowing in humans? Vision Research, 48(7), 946–957. https://doi
.org/10.1016/j.visres.2008.01.011, PubMed: 18294671
Brooks, R., & Meltzoff, A. N. (2005). The development of gaze fol-
lowing and its relation to language. Developmental Science, 8(6),
535–543. https://doi.org/10.1111/j.1467-7687.2005.00445.x,
PubMed: 16246245
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Bugnyar, T., Stöwe, M., & Heinrich, B. (2004). Ravens, Corvus corax,
follow gaze direction of humans around obstacles. Proceedings
of the Royal Society of London. Series B: Biological Sciences,
271(1546), 1331–1336. https://doi.org/10.1098/rspb.2004.2738,
PubMed: 15306330
Carpenter, M., Nagell, K., Tomasello, M., Butterworth, G., &
Moore, C. (1998). Social cognition, joint attention, and commu-
nicative competence from 9 to 15 months of age. Monographs of
the Society for Research in Child Development, 63(4), i. https://
doi.org/10.2307/1166214, PubMed: 9835078
Csibra, G., & Gergely, G. (2009). Natural pedagogy. Trends in
Cognitive Sciences, 13(4), 148–153. https://doi.org/10.1016/j
.tics.2009.01.005, PubMed: 19285912
Csibra, G., & Gergely, G. (2011). Natural pedagogy as evolutionary
adaptation. Philosophical Transactions of the Royal Society B:
Biological Sciences, 366(1567), 1149–1157. https://doi.org/10
.1098/rstb.2010.0319, PubMed: 21357237
Deák, G. O., Krasno, A. M., Triesch, J., Lewis, J., & Sepeta, L.
(2014). Watch the hands: Infants can learn to follow gaze by
seeing adults manipulate objects. Developmental Science, 17(2),
270–281. https://doi.org/10.1111/desc.12122, PubMed:
24387193
OPEN MIND: Discoveries in Cognitive Science
187
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
/
.
i
Gaze Following in Infants
Silverstein et al.
Dienes, Z., & Mclatchie, N. (2018). Four reasons to prefer Bayesian
analyses over significance testing. Psychonomic Bulletin &
Review, 25, 207–218. https://doi.org/10.3758/s13423-017-1266-z,
PubMed: 28353065
Dink, J. W., & Ferguson, B. (2015). EyetrackingR: An R Library for
Eye-tracking Data Analysis. https://www.eyetrackingr.com
Farroni, T., Massaccesi, S., Pividori, D., & Johnson, M. H. (2004).
Gaze following in newborns. Infancy, 5(1), 39–60. https://doi.org
/10.1207/s15327078in0501_2
Gredebäck, G., Astor, K., & Fawcett, C. (2018). Gaze following is
not dependent on ostensive cues: A critical test of natural peda-
gogy. Child Development, 89(6), 2091–2098. https://doi.org/10
.1111/cdev.13026, PubMed: 29315501
Grönqvist, H., Gredebäck, G., & Hofsten, C. von. (2006).
Developmental asymmetries between horizontal and vertical
tracking. Vision Research, 46(11), 1754–1761. https://doi.org/10
.1016/j.visres.2005.11.007, PubMed: 16376401
Haith, M. M. (1998). Who put the cog in infant cognition? Is rich
interpretation too costly? Infant Behavior and Development, 21(2),
167–179. https://doi.org/10.1016/S0163-6383(98)90001-7
Hernik, M., & Broesch, T. (2018). Infant gaze following depends on
communicative signals: An eye-tracking study of 5- to 7-month-
olds in Vanuatu. Developmental Science, 22(4), Article e12779.
https://doi.org/10.1111/desc.12779, PubMed: 30506550
Horst, J. S., & Hout, M. C. (2016). The Novel Object and Unusual
Name (NOUN) database: A collection of novel images for use in
experimental research. Behavior Research Methods, 48(4),
1393–1409. https://doi.org/10.3758/s13428-015-0647-3,
PubMed: 26424438
Jeffreys, H. (1961). Theory of probability (3rd ed.). Oxford
University Press.
Johnson, V. E., & Rossell, D. (2010). On the use of non-local prior
densities in Bayesian hypothesis tests. Journal of the Royal
Statistical Society: Series B (Statistical Methodology), 72(2),
143–170. https://doi.org/10.1111/j.1467-9868.2009.00730.x
Michel, C., Kayhan, E., Pauen, S., & Hoehl, S. (2021). Effects of re-
inforcement learning on gaze following of gaze and head direc-
tion in early infancy: An interactive eye-tracking study. Child
Development. https://doi.org/10.1111/cdev.13497, PubMed:
33427319
Moore, C., Angelopoulos, M., & Bennett, P. (1997). The role of
movement in the development of joint visual attention. Infant
Behavior and Development, 20(1), 83–92. https://doi.org/10
.1016/S0163-6383(97)90063-1
Moore, C., & Corkum, V. (1994). Social understanding at the end
of the first year. Developmental Review, 14, 349–372. https://doi
.org/10.1006/drev.1994.1014
Morales, M., Mundy, P., Delgado, C. E. F., Yale, M., Neal, R., &
Schwartz, H. K. (2000). Gaze following, temperament, and
language development in 6-month-olds: A replication and exten-
sion. Infant Behavior and Development, 23(2), 231–236. https://
doi.org/10.1016/S0163-6383(01)00038-8
Morey, R. D., & Rouder, J. N. (2015). Bayes factor: Computation of
Bayes factors for common designs. R package version 0.9.12-2.
https://CRAN.R-project.org/package=BayesFactor
Morse, A. F., & Cangelosi, A. (2017). Why are there developmental
stages in language learning? A developmental robotics model of
language development. Cognitive Science, 41, 32–51. https://doi
.org/10.1111/cogs.12390, PubMed: 27680660
Nagai, Y., Asada, M., & Hosoda, K. (2006). Learning for joint attention
helped by functional development. Advanced Robotics, 20(10),
1165–1181. https://doi.org/10.1163/156855306778522497
Okamoto-Barth, S., Call, J., & Tomasello, M. (2007). Great apes’
understanding of other individuals’ line of sight. Psychological
Science, 18(5), 462–468. https://doi.org/10.1111/j.1467-9280
.2007.01922.x, PubMed: 17576288
Povinelli, D. J., Eddy, T. J., Hobson, R. P., & Tomasello, M. (1996).
What young chimpanzees know about seeing. Monographs of
the Society for Research in Child Development, 61(3), i. https://
doi.org/10.2307/1166159, PubMed: 8795292
R Core Team. (2017). R (3.5.2) [Computer software]. R Foundation
for Statistical Computing. https://www.R-project.org/
Schlesinger, M., & McMurray, B. (2012). The past, present, and
future of computational models of cognitive development.
Cognitive Development, 27(4), 326–348. https://doi.org/10
.1016/j.cogdev.2012.07.002
Schönbrodt, F. D., Zehetleitner, M., Wagenmakers, E.-J., &
Perugini, M. (2017). Sequential hypothesis testing with Bayes fac-
tors: Efficiently testing mean differences. Psychological Methods,
22(2), 322–339. https://doi.org/10.1037/met0000061, PubMed:
26651986
Senju, A., & Csibra, G. (2008). Gaze following in human infants
depends on communicative signals. Current Biology, 18(9),
668–671. https://doi.org/10.1016/j.cub.2008.03.059, PubMed:
18439827
Szufnarowska, J., Rohlfing, K. J., Fawcett, C., & Gredebäck, G.
(2014). Is ostension any more than attention? Scientific Reports,
4(1), Article 5304. https://doi.org/10.1038/srep05304, PubMed:
24931735
Tomasello, M. (1995). Joint attention as social cognition. In Joint
attention: Its origins and role in development (pp. 103–130).
Lawrence Erlbaum Associates.
Tomasello, M., Hare, B., Lehmann, H., & Call, J. (2007). Reliance
on head versus eyes in the gaze following of great apes and
human infants: The cooperative eye hypothesis. Journal of
Human Evolution, 52(3), 314–320. https://doi.org/10.1016/j
.jhevol.2006.10.001, PubMed: 17140637
Triesch, J., Teuscher, C., Deak, G. O., & Carlson, E. (2006). Gaze
following: Why (not) learn it? Developmental Science, 9(2),
125–147. https://doi.org/10.1111/j.1467-7687.2006.00470.x,
PubMed: 16472311
von Hofsten, C., Dahlström, E., & Fredriksson, Y. (2005). 12-month-old
infants’ perception of attention direction in static video images.
Infancy, 8(3), 217–231. https://doi.org/10.1207/s15327078in0803_2
Wagenmakers, E. J., Marsman, M., Jamil, T., Ly, A., Verhagen, J.,
Love, J., Selker, R., Gronau, Q. F., Šmíra, M., Epskamp, S., Matzke,
D., Rouder, J. N., & Morey, R. D. (2018). Bayesian inference for
psychology. Part I: Theoretical advantages and practical ramifica-
tions. Psychonomic Bulletin & Review, 25, 35–57. https://doi.org
/10.3758/s13423-017-1343-3, PubMed: 28779455
Westermann, G., & Mareschal, D. (2012). Mechanisms of develop-
mental change in infant categorization. Cognitive Development,
27(4), 367–382. https://doi.org/10.1016/j.cogdev.2012.08.004
Yu, C., & Smith, L. B. (2013). Joint attention without gaze following:
Human infants and their parents coordinate visual attention to
objects through eye-hand coordination. PLoS ONE, 8(11),
Article e79659. https://doi.org/10.1371/journal.pone.0079659,
PubMed: 24236151
OPEN MIND: Discoveries in Cognitive Science
188
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
o
p
m
i
/
l
a
r
t
i
c
e
-
p
d
f
/
d
o
i
/
i
/
/
.
1
0
1
1
6
2
o
p
m
_
a
_
0
0
0
4
9
1
9
7
4
9
2
4
o
p
m
_
a
_
0
0
0
4
9
p
d
.
/
i
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
