You are on page 1of 14

10.

1177_1747021819844503
research-article2019
QJP0010.1177/1747021819844503The Quarterly Journal of Experimental PsychologyBuijsman and Tirado

Original Article

Quarterly Journal of Experimental

Spatial–numerical associations: Psychology


2019, Vol. 72(10) 2423­–2436
© Experimental Psychology Society 2019
Shared symbolic and non-symbolic Article reuse guidelines:
sagepub.com/journals-permissions
numerical representations DOI: 10.1177/1747021819844503
https://doi.org/10.1177/1747021819844503
qjep.sagepub.com

Stefan Buijsman1 and Carlos Tirado2

Abstract
During the last decades, there have been a large number of studies into the number-related abilities of humans. As a
result, we know that humans and non-human animals have a system known as the approximate number system that
allows them to distinguish between collections based on their number of items, separately from any counting procedures.
Dehaene and others have argued for a model on which this system uses representations for numbers that are spatial
in nature and are shared by our symbolic and non-symbolic processing of numbers. However, there is a conflicting
theoretical perspective in which there are no representations of numbers underlying the approximate number system,
but only quantity-related representations. This perspective would then suggest that there are no shared representations
between symbolic and non-symbolic processing. We review the evidence on spatial biases resulting from the activation
of numerical representations, for both non-symbolic and symbolic tests. These biases may help decide between the
theoretical differences; shared representations are expected to lead to similar biases regardless of the format, whereas
different representations more naturally explain differences in biases, and thus behaviour. The evidence is not yet
decisive, as the behavioural evidence is split: we expect bisection tasks to eventually favour shared representations,
whereas studies on the spatial–numerical association of response codes (SNARC) effect currently favour different
representations. We discuss how this impasse may be resolved, in particular, by combining these behavioural studies with
relevant neuroimaging data. If this approach is carried forward, then it may help decide which of these two theoretical
perspectives on number representations is correct.

Keywords
Numerosity; representations; triple-code model; SNARC effect; spatial–numerical associations

Received: 28 August 2018; revised: 18 March 2019; accepted: 24 March 2019

Introduction and “Analog Magnitude System” are the most common


(see Feigenson, Dehaene, & Spelke, 2004; Gebuis, Cohen
Humans, and most animals, are capable of distinguishing Kadosh, & Gevers, 2016 for reviews) and also common is
between collections of items based on the number of items. the abbreviation ANS (Gebuis & van der Smagt, 2011;
Human infants, for example, will look longer at a picture Halberda & Feigenson, 2008; Holloway & Ansari, 2009;
of four dots after having habituated to one with two dots, Lipton & Spelke, 2003; Nieder, Freedman, & Miller, 2002;
than at a picture of two different dots. Whether or not J. Wood & Spelke, 2005). While it is clear that there is
the difference in the number of items is distinguishable such a system, it is still far from clear how it works
depends on the ratio between the numbers, a fact codified (Leibovich, Katzin, Harel, & Henik, 2017). There is a
by Weber’s law. Two versus four items is feasible even for
infants, who can distinguish anything at a ratio of 1 to 2
(for the usual performance levels). They, however, cannot 1Institute
for Future Studies, Leiden, The Netherlands
distinguish between 6 and 8. This ratio improves with age, 2Gösta
Ekman Laboratory, Department of Psychology, Stockholm
up to around 7-8 in human adults (Barth, Kanwisher, & University, Stockholm, Sweden
Spelke, 2003) and sometimes even 9-10 (Guillaume, Nys,
Corresponding author:
Mussolin, & Content, 2013). Carlos Tirado, Gösta Ekman Laboratory, Department of Psychology,
The system responsible for these abilities is known under Stockholm University, Frescati hagväg 9, 114 19 Stockholm, Sweden.
a variety of names, where “Approximate Number System” Email: carlos.tirado@psychology.su.se
2424 Quarterly Journal of Experimental Psychology 72(10)

heated debate between two opposing theories of the ANS. numbers. They consider the ANS to be part of a more gen-
On one hand, theorists such as Dehaene have developed eral magnitude system, which allows us to distinguish
the triple-code model (Arsalidou & Taylor, 2011; Dehaene, between collections with different numbers of items based
1992; Dehaene, Bossini, & Giraux, 1993; Dehaene & on our ability to determine length, surface area, density,
Changeux, 1993; Dehaene, Piazza, Pinel, & Cohen, 2003; and other magnitudes. All the numerical information we
Klein et al., 2016; Knops, 2017). In this model, the repre- need is included in these magnitudes, and there is therefore
sentations of the ANS (resulting from e.g., seeing an array no separate representation of numbers. Rather, the ANS is
with nine dots) are the same as those for symbolic exact no more than the procedure of weighing the different fac-
numbers (e.g., “9”). In other words, we sense the “nine- tors to arrive at the more general judgement of one collec-
ness” of a set of nine dots with the ANS in the same way tion having more items than another. This is more general
we sense their colour and surface area with other systems. when compared to the judgement that e.g., one collection
More specifically, the triple-code model suggests that both occupies a larger surface area than another, which would
the ANS (Dehaene’s early papers refer to analogue magni- be one of the inputs for this general magnitude system.
tude representations, a terminological variant to our use of We do not yet know which of these two theories is cor-
“ANS representations”) and symbolic number representa- rect. Since more direct types of evidence are currently
tions are best seen as point activations along the same ana- inconclusive (Leibovich et al., 2017), we explore an indi-
logue number line. At the neuronal level, the model rect way of deciding between these two accounts in this
predicts that these point activations engage the bilateral review. Our main question is then which account of the
inferior parietal regions. Our numerical representations ANS is correct? One with shared representations (of num-
are, on this view, inherently spatial and shared across bers) or one with distinct representations?
formats.
The triple-code model supplements these numerical Interactions between numerical and spatial
representations, which are activated both by the ANS and
the perception of symbolic numbers, with two other types
representations
of representations (hence the “triple-code”). In the case of One way to decide which of these two perspectives is cor-
written numerals, an additional verbal number form repre- rect is to look at spatial–numerical associations: if number
sentation is activated, which is used to link the symbolic representations are the same for symbolic and non-sym-
number to the numerical representation also used by the bolic stimuli, then it is likely that the spatial–numerical
ANS. This number form representation is predicted to associations are also the same for both types of stimuli. If,
engage bilateral inferior ventral occipitotemporal regions instead, there are differences in the spatial–numerical
belonging to the ventral visual pathway. The third repre- associations, then this is support for the claim that there are
sentation is a verbal code (engaging general-purpose lan- differences in the underlying representations of numbers.
guage modules, including the left perisylvian network and The abundance of evidence on spatial associations and the
the left basal ganglia and thalamic nuclei) which affects difficulty of investigating representations directly make
the same link, but in the case of spoken number words. such an argument attractive, though not ideal. After all,
This basic framework, introduced by Dehaene (1992), behavioural evidence is no guarantee: the same behav-
has had a few recent adjustments (see Knops, 2017 and ioural signatures may be produced even though the under-
Leibovich et al., 2017 for more thorough discussions). lying processes and representations are different. Similarly,
Arsalidou and Taylor (2011) pointed out the need to different behaviour may be caused by the use of the same
include working memory in the model, in the case of cal- representations in different processes—though the tasks
culations with symbolic numbers that require multiple are similar enough, it becomes less likely that the pro-
steps (where the difficulty of the calculation determines cesses are different. While such alternative explanations
the exact role of our working memory). Klein et al. (2016) will be available, they will be less simple and, to our
suggest amending the model by extending the neural areas knowledge, have yet to be developed. Therefore, we do
associated with the numerical representation (from the think that observations of the same behaviour count as
intraparietal sulcus to an area including Broca’s area and (non-conclusive) evidence in favour of shared representa-
three others, plus their interconnectivity). They also see a tions, and observations of differing behaviour count as
more prominent role for verbal fact retrieval in the model (non-conclusive) evidence in favour of distinct representa-
when applied to arithmetical calculations. Those changes tions. This review explores, within those limits, what
are not relevant for the current review, because we have spatial–numerical associations can tell us about the degree
not included studies with calculating tasks. Such tasks do to which number representations are shared between sym-
not have clear analogues with non-symbolic stimuli. bolic and non-symbolic formats.
On the other hand, Gebuis et al. (2016), Zimmerman There are a lot of spatial–numerical associations.
(2018), and others argue that the representations of the Decades of research have established that there is a close
ANS are fundamentally different from those of symbolic link between our processing of numbers and our processing
Buijsman and Tirado 2425

of space. Suggestions that the two are intertwined go as far in nature by other accounts, we believe that it is relevant to
back as Galton (1880), who found that some adults report include them in this review. Note also that non-symbolic
explicit visuospatial representations of numbers. It turns stimuli can still give rise to (symbolic) representations of
out that roughly 15% of the population has such representa- numbers: if we count the stimuli, we are guaranteed to end
tions (Seron, Pesenti, Noël, Deloche, & Cornet, 1992). up with representations of numbers on any account. In
There are, however, more fundamental links between our studies with non-symbolic stimuli, participants were pre-
numerical and spatial processing. Our performance on vented from counting, so we will set this slight complica-
numerical tasks is often influenced by spatial factors and tion aside; non-symbolic representations are always the
vice versa. To give two examples, we discuss further below: ones used by the ANS.
we respond faster to the question if a number is odd or even We proceed by highlighting differences and similarities
when we respond with our left hand for small numbers and between the two cases, in accordance with our central
our right hand for large numbers. We also tend to bisect question: are the representations underlying symbolic and
lines with flanking numbers on a bias towards the higher non-symbolic tasks shared? To make the argument easier
number. Recent research has focused on these numerous to follow, we have organised the review by types of spatial–
links between space and number, providing us with a sub- numerical association: for each association, we discuss,
stantial amount of behavioural (and some neuroimaging) first, the non-symbolic findings and then the symbolic
evidence on the interactions between numerical and spatial findings, with a short comparison at the end of each section.
representations (for earlier reviews, see de Hevia, Vallar, & We begin with associations as measured by bisection tasks,
Girelli, 2008; Fias & Fischer, 2005; Fias, van Dijck, & then move to spatial–numerical association of response
Gevers, 2011; Fischer & Shaki, 2014; Gevers & Lammertyn, codes (SNARC) effects, and finally discuss associations
2005; Hubbard, Piazza, Pinel, & Dehaene, 2005; McCrink measured by ordinal tasks. At the end of the review, there
& Opfer, 2014; Rossetti et al., 2011; Rugani & de Hevia, is also a discussion of the current neuroimaging studies
2017; Umilta, Priftis, & Zorzi, 2009; G. Wood, Willmes, related to spatial–numerical associations.
Nuerk, & Fischer, 2008).
The links found by these studies can be compared
Bisection tasks
between formats. Do we see the same biases associated
with non-symbolic representations as with symbolic repre- One type of interaction between numerical representations
sentations? In that case, there is reason to prefer the and spatial representations is observed in the case of line
hypothesis of shared representations (and so Dehaene’s bisection tasks. In these tasks, our ability to locate the mid-
view of the ANS). If, however, the biases are very differ- point of a line is hampered by the (automatic) activation of
ent, then there is reason to prefer the hypothesis of differ- numerical representations, leading to biases towards the
ent representations (and the general magnitude system higher number.
view of Gebuis et al. (2016)).
Such an argument is different from the one made on the
Non-symbolic tasks
basis of studies such as Izard, Sann, Spelke, and Streri
(2009). They argue for a shared representation of numbers Line bisection tasks are based on the procedure of bisect-
on the basis of tests where infants are able to match the ing a line, which is flanked by arrays of dots (or symbolic
number of dots in an array with the number of tones they numerals, which will be discussed later). de Hevia and
heard. Their argument is that these representations have to Spelke (2009) performed this test with 3- to 7-year-old
be shared, and abstract, because of their flexibility in use. children to see if they display a bias in bisecting lines when
However, experiments such as these have been discussed there are flanking arrays of dots. They found that children
by Gebuis et al. (2016), and they have argued that there is of all these ages display a bias towards the side which is
currently no decisive evidence in favour of either position, flanked by the largest number of dots. So, instead of bisect-
based on these data. However, data on spatial–numerical ing the line in the middle, as they were supposed to, they
biases have not yet been considered in this context and bisected the lines closer to one side. Furthermore, de Hevia
could also help decide between there being shared or dif- and Spelke (2009) considered several possible explana-
ferent representations. tions for this bias.
This review, therefore, separates the research on spa- First, they tested with adults whether arrays with more
tial–numerical associations based on two broad categories: dots draw preferential attention, thus resulting in a bias.
our non-symbolic and our symbolic numerical representa- Participants were supposed to locate a black triangle on the
tions. The former might not be representations of numbers; left or right side of the screen, with an array of dots or an
in fact, they do not involve numbers if the account of arabic numeral present on each side. If attention is drawn
Gebuis et al. (2016) is correct, because then they are more to the array with a larger number of dots, then it should be
general magnitude representations. Since they do underlie easier for them to find the black triangle if it is located at
number-related abilities, and are thought to be numerical the same side as that array. However, they were unable to
2426 Quarterly Journal of Experimental Psychology 72(10)

find such an effect of relative number of dots on relative used (Fischer, 2001). The bisection bias with flanking
ease of finding the black triangle. Second, they ran sepa- numbers has been tested in several other studies since, and
rate experiments with children where they controlled for there too a bias towards the larger number was found (de
contour length and other experiments where they con- Hevia, Girelli, & Vallar, 2006; Ranzini & Girelli, 2012).
trolled for the distance between the line and the flanking So, the observed bisection biases in healthy adults seem
arrays of dots. In both of these experiments, the observed related to attentional biases (Fischer & Knops, 2014).
bias persisted, suggesting that it is not caused by contour Attentional biases have been observed in other studies; for
length or flanking distance (de Hevia & Spelke, 2009). de example, when generating random numbers, participants
Hevia and Spelke (2009) also tested adults using line tend to look to the left when they are about to name a small
bisection task with flanking arrays of dots and found a bias number and to the right for a large number (Loetscher,
towards the more numerous array similar to that reported Bockisch, Nicholls, & Brugger, 2010), and brain imagery
for children. has associated cortical circuits for spatial attention with the
These findings have been complicated by a replication performance of mental arithmetic (Knops, Thirion,
study by Gebuis and Gevers (2011). They managed to rep- Hubbard, Michel, & Dehaene, 2009). These attentional
licate the bisection bias towards the higher number when biases, which result from automatic number processing,
the higher number of dots had the larger total area. When are likely to be the source of the bisection biases observed
the higher number of dots had the smallest total area, how- in symbolic tasks.
ever, the bisection bias was towards the lower number of
dots (and still highest total area). There is, at the moment,
Similarities and differences
no clear evidence for a line bisection bias based on the
number of dots. The first type of interaction between numerical and spatial
representations we looked at are biases in bisection tasks.
There are clear interactions of this type in the case of sym-
Symbolic tasks bolic stimuli, which are probably caused by attentional
There are different kinds of symbolic bisection tasks: these biases that result from automatic number processing. It is
can involve the bisection of a line or, as with the first set of therefore interesting that although the non-symbolic task
experiments we report on, the bisection of a range of num- failed to replicate properly, there are eye-tracking studies
bers. Participants in this latter study had to state the num- with infants that found a similar attentional bias for non-
ber in the middle of a range of two numbers, e.g., from 3 to symbolic stimuli.
9. Longo and Lourenco (2007, 2010) found that adults Bulf, de Hevia, and Macchi Cassia (2016) found a left-
tend to underestimate the midpoint of a number interval in ward bias in attention after perceiving a small number of
such tasks. They have, as they put it, a “leftward bias,” dots, compared to a rightward bias after perceiving a large
which seems to be correlated to the leftward bias in physi- number of dots. Infants’ attention was first focused on the
cal line bisection known as pseudoneglect. Participants centre of a screen, where subsequently an array of (two or
who displayed a larger bias in physical line bisection tasks nine) dots or a rainbow-coloured shape appeared. After
also displayed a larger bias in number interval bisection 300 ms, the cue disappeared, and a little later, a target (a
tasks (Longo & Lourenco, 2007). When they tested how flickering schematic face) appeared in a circle either to
this bias changes when physical lines and number intervals the left or to the right of the location of the cue. When the
are moved from near to far space, they found a correlated cue was an array with two dots, infants were faster at
rightward shift of the bisection bias (Longo & Lourenco, detecting the target if it was on the left, whereas with an
2010). This suggests that the two biases are related, point- array with nine dots, infants were faster if the target was
ing to an origin in attentional biases that are known to on the right. There was no difference between left and
underlie physical line bisection biases. right with any of the sizes. So, while arrays of dots pro-
If so, then we would expect to see similar results with duce attention biases, rainbow-coloured shapes do not
line bisection tasks that use flanking numbers. That is, if (Bulf et al., 2016; Figure 1).
numbers can lead to attentional biases, then it is to be The study by Bulf et al. (2016) is important for the case
expected that they also do so if they are task irrelevant. of bisection biases since, as discussed, such biases appear
Fischer (2001) tested both bisection biases with flanking to stem from attentional biases in the case of symbolic
numbers and bisection biases where participants had to number tasks. The fact that this study found similar atten-
bisect a string of numerals. For example, they were asked tional biases in the case of non-symbolic representations
to find the midpoint of 1111111. The physical line bisec- leads us to expect that further studies with non-symbolic
tion was biased towards the side with the larger number, stimuli will more clearly establish a bisection bias similar
whereas with the numeral strings, bisection was biased to the one for symbolic stimuli.
towards the left when small numbers were used and If such a similarity is found, then the only difference
towards the right when large numbers (8’s or 9’s) were between behaviour is the type of stimulus flanking the
Buijsman and Tirado 2427

and Leth-Steensen (2012). Yates, Nemeh, Loetscher,


Ma-Wyatt, and Nicholls (2013) also found that there is a
leftward bias for responding with “less” and a rightward
bias for responding with “more” when having to determine
if a stimulus array of dots is more or less numerous than a
reference array. For non-symbolic stimuli, the direction of
the spatial bias reverses when the answer is different
(“more” rather than “less”) and does not reverse with read-
ing direction. For symbolic stimuli (discussed below), the
direction of the bias does not reverse depending on the
answer but does reverse depending on reading direction.
Figure 1.  The experimental setup of the eye-tracking study by Such differences in reversal of directionality between sym-
Bulf, de Hevia, and Macchi Cassia (2016). bolic and non-symbolic stimuli are notable. The fact that
Children first saw a stimulus to attract their attention to the place the conditions under which the spatial bias reverses are
where the array of dots subsequently appeared; at the same time,
two circles indicated the areas of interest where a target may appear; different for symbolic and non-symbolic stimuli means
eye-tracking software was calibrated to note whether there was an that these experiments do not support the presence of a
attentional bias towards either of these areas. shared representation underlying symbolic and non-
symbolic number-related processes.
lines: either symbolic numbers or arrays of dots. The task
will be exactly the same otherwise, which does make this
Symbolic tasks
a good behavioural test to tease out any differences in the
underlying representations once more data are available. The SNARC effect was the first known interaction between
numbers and spatial organisation. Dehaene, Dupoux, and
Mehler (1990) asked participants to classify numbers as
SNARC-like effects larger or smaller than 65 by pressing a key to their left or a
More important, or at least more studied, than bisection key to their right. They found, unexpectedly, that there is a
tasks is a type of bias known as the SNARC effect in the difference in response times depending on which key par-
case of symbolic stimuli. Here, we discuss these, and the ticipants have to press. If the answer is smaller, the
closest correlates of them in the case of non-symbolic response times are lower if the corresponding key is to
stimuli, to see how closely they match. their left. If the answer is larger, the response times are
lower if the key is to their right (Dehaene et al., 1990).
A follow-up study by Dehaene et al. (1993) looked at
Non-symbolic tasks
this effect more directly. These experiments were based on
The experiments falling under this type have proceeded parity tests, where participants had to press one key if the
from the SNARC effect found for symbolic stimuli and presented number was even and another if it was odd.
looked for a similar effect with non-symbolic stimuli. After a range of experiments around this task, some with
Adults respond faster with a button to their left when numbers from 0 to 9, others with two intervals, 0-4 and
responding to small numbers and faster with a button on 5-9, and yet others with two-digit numbers, Dehaene et al.
their right when responding to large numbers (Dehaene (1993) found that in all of these cases, responses to lower
et al., 1993). Such a left-to-right bias has also been found numbers in an interval are faster with the left key and
for non-symbolic numerical stimuli (Luccio, Fumarola, responses to higher numbers in an interval are faster with
Tamburini, & Agostini, 2012; Mitchell, Bull, & Cleland, the right key. So, 3 is responded to faster on the right for
2012). Patro and Shaki (2016) used arrays of objects in the interval 0-4 but faster on the left for the interval 0-9. In
their experiment, verifying this effect for non-symbolic general, the SNARC effect is relative to the interval of
representations of numbers. While they did find a tradi- numbers chosen by the experimenters (Dehaene et al.,
tional left-to-right bias when participants were asked to 1993; Nathan, Shaki, Salti, & Algom, 2009; Ren, Nicholls,
choose the array with fewer items (from two choices), they Ma, & Chen, 2011).
observed a right-to-left bias when participants were asked A number of characteristics of the SNARC effect have
to choose the one with more items. These results were become known since the first studies by Dehaene and his
independent from reading direction, which is an important colleagues. These are organised below and briefly explained
factor for studies using numerals as stimuli, as discussed (for more extensive reviews, see Fischer & Shaki, 2014;
below (Patro & Shaki, 2016). Gevers & Lammertyn, 2005).
This difference between symbolic and non-symbolic
representations, where the directionality is not task Task irrelevance of numbers.  One of the first findings was
dependent in this way, was first noticed by Shaki, Petrusic, that for the SNARC effect to occur, there is no need for any
2428 Quarterly Journal of Experimental Psychology 72(10)

both Russian (left-to-right script) and Hebrew (right-to-


left script). These participants showed a left-to-right
SNARC effect after reading a Cyrillic text, but a right-to-
left SNARC effect after reading a Hebrew text (Shaki &
Fischer, 2008). Reading direction thus influences the
direction of the SNARC effect, possibly mediated by the
fact that counting direction is often the same as the reading
direction (see also Göbel, 2015; Shaki, Fischer, & Göbel,
2012; Shaki, Fischer, & Petrusic, 2009; Zebia, 2005).
The direction of the SNARC effect is also flexible.
However, it can be influenced in myriad ways. When west-
ern participants are asked to imagine a series of numbers
Figure 2.  A typical response time curve for the SNARC effect. arranged along a clock face, the SNARC effect reverses
There is a positive difference in response time between left and right
(i.e., RT for right minus the RT for left) for small digits and a negative
from a left-to-right direction to a right-to-left direction
difference for large digits. (Bächtold, Baumüller, & Brugger, 1998). When partici-
pants are asked to memorise a sequence of numbers (e.g.,
3, 4, 5), the SNARC effect is only shown on unrelated
explicit processing of the numbers. Even when the tasks tasks (e.g., when memorising number sequences in ascend-
had nothing to do with the numbers, participants saw there ing or random order; Lindemann, Abolafia, Pratt, &
was a measurable SNARC effect. For example, a SNARC Bekkering, 2008). Finally, Fischer, Mill, and Shaki (2010)
effect occurs when participants are asked to press buttons performed experiments where participants first read a text
for whether or not a verbal numeral contains the phoneme in which numbers were distributed either left to right (i.e.,
/e/ (Fias, Brysbaert, Geypens, & D’Ydewalle, 1996) or 1-4 on the far left side and 5-9 on the far right side) or right
when asked if a superimposed triangle is pointing upwards to left. English-speaking participants reading the left-to-
or downwards (Fias, Lauwereyns, & Lammertyn, 2001; right text exhibited a standard SNARC effect, but those
Lammertyn, Fias, & Lauwereyns, 2002). However, not all reading the right-to-left text exhibited no SNARC effect.
such tasks result in a SNARC effect. When participants Hebrew-speaking participants showed no SNARC effect
were asked to answer questions about the colour of a before reading or after reading the left-to-right text.
superimposed triangle, or whether the superimposed shape However, they did display a right-to-left SNARC effect
was a circle or square, no SNARC effect was observed after reading the text where numbers were distributed in
(Fias et al., 2001; Lammertyn et al., 2002). Interestingly, that direction (Fischer et al., 2010).
while the SNARC effect is associated with a left-to-right
orientation, similar tests with a button close to the partici- Notation dependence of the SNARC effect. The studies
pant and a button far away from the participant found that reported so far have all used Arabic numerals. Some stud-
response times are lower for the combinations of small ies have also looked at the SNARC effect in relation to
numbers and the close button and that of large numbers which numerals are used. In particular, studies have looked
and the far button (Santens & Gevers, 2008). There are into differences in SNARC effect depending on whether
also some studies that found a top/bottom SNARC effect Arabic or Chinese numerals are used. Hung, Hung, Tzeng,
(Holmes & Lourenco, 2012; Ito & Hatta, 2004; Shaki & and Wu (2008) tested the SNARC effect in Chinese speak-
Fischer, 2012). Furthermore, while it is common to test for ers with Arabic numerals and the two different numeral
the SNARC effect using buttons, it has also been observed systems used in Chinese. They found a left-to-right
where participants use pointing (Fischer, 2003), eye move- SNARC effect for Arabic numerals, but no such effect for
ments (Fischer, Warlop, Hill, & Fias, 2004) and grasping the Chinese numerals. However, they also found a top-to-
(Andres, Ostry, Florence, & Paus, 2008). Little seems to bottom SNARC effect for the simple Chinese numerals.
depend on the type of task (Figure 2). No effects were found for the complex Chinese numerals,
possibly because these are not used very frequently (Hung
Flexibility of the SNARC effect.  Another important character- et al., 2008). Kopiske et al. (2016) also looked at notation
istic is that the direction of the SNARC effect is flexible. dependence, testing Arabic numerals, simple Chinese
One factor on which the direction seems to depend is the numerals, and Chinese hand signs (as used in sign lan-
standard reading direction of the participants. Dehaene guage). They, as opposed to Hung et al. (2008), did find a
et al. (1993) found that French speakers, who read left to left-to-right SNARC effect for all three notations on a par-
right, exhibit a standard SNARC effect. Iranian speakers, ity (odd/even) test. However, in a test where participants
however, exhibit a reversed SNARC effect, probably had to decide if the presented number was larger or smaller
because they read right to left. Shaki and Fischer (2008) than 5, a clear SNARC effect was only observed for Arabic
extended this finding by studying bilingual speakers of numerals. There was weaker evidence for a SNARC effect
Buijsman and Tirado 2429

for simple Chinese numerals, but no evidence for a This raises the question why the non-symbolic effects,
SNARC effect for Chinese hand signs (Kopiske et al., which would attach to the ANS representation, are not
2016). In the absence of more studies, it is hard to con- observed in symbolic tasks (recall that there are a number
clude whether the SNARC effect is notation dependent, of situations in which no spatial bias was found and that
but there are some suggestions that notation may matter. the SNARC effect is not standardly in a left-to-right direc-
To sum up, the important characteristics of the SNARC tion). If the non-symbolic ANS representation and pro-
effect are, first, the relativity of the effect: which stimuli cesses have an important role in processing symbolic
are responded to faster with the left hand depends on the numbers, as the triple-code model suggests, we would also
range of stimuli that is presented. Second, a SNARC effect expect the supposedly spatial nature of the ANS represen-
can be observed whether numbers are relevant to the task tation to filter through.
or not. Third, the SNARC effect is flexible in the sense that This seems to happen in a related case slightly out of
it depends on reading direction and can disappear or even the scope of our review (because the non-symbolic biases
reverse when participants read incongruent texts or have to are not explicitly spatial): the operational momentum
memorise conflicting sequences of numbers. Fourth, the effect observed with mental calculation. McCrink,
SNARC effect may even be dependent on the notation sys- Dehaene, and Dehaene-Lambertz (2007) observed that
tem that is used. there is an upwards bias for mental addition with non-
symbolic stimuli and a downwards bias for mental sub-
traction with such stimuli. Adults, and 9-month-old
Similarities and differences
children (McCrink & Wynn, 2009), more readily accept
The SNARC effect, as we know it from symbolic stimuli, non-symbolic solutions to addition problems that are too
has a number of important features that are not mirrored in large and vice versa for subtraction problems. The idea is
the effects observed with non-symbolic stimuli on other- that this is caused by the mental number line: addition car-
wise identical tasks. The flexibility with respect to varia- ries the representations rightward, towards higher num-
bles such as reading direction of the SNARC effect is not bers, and thus may overshoot. Subtraction moves leftward,
found in the case of non-symbolic stimuli. Oddly enough, towards smaller numbers, and therefore can move too far
the non-symbolic effect is flexible with respect to the task to the left. Pinhas and Fischer (2008) observed this hypoth-
formulation (formulated in terms of “more” or “less”), esised spatial bias for symbolic stimuli when adults had to
whereas the symbolic SNARC effect is not flexible in this locate the mentally computed result of an addition or sub-
way. It should be noted, though, that these tasks were traction on a number line. They explicitly link this to the
slightly different: for non-symbolic stimuli, both stimuli non-symbolic effect, which as we point out should also be
were visible at the same time, whereas in symbolic SNARC the case for the SNARC effects.
tasks, the numbers are usually presented one at a time. The SNARC data are easier to explain on the compet-
Furthermore, the more/less task common to non-symbolic ing account of the ANS by e.g., Gebuis et al. (2016). In
formats is not standard for the symbolic SNARC tasks that case, there is no direct link between the representa-
(parity tests are more usual), so these differences in task tions of the ANS, which are responsible for the inflexible
setup do make the comparison imperfect. Still, the cur- SNARC-like effects discussed here, and our representa-
rently observed differences in behaviour do suggest a dif- tions of symbolic numbers, which are responsible for the
ference in representations. Or is it possible to explain the (flexible) SNARC effect. This makes it harder to explain
behavioural differences on the hypothesis that the repre- the operational momentum effect, but perhaps these theo-
sentations are the same? rists can locate similarities in the way addition and sub-
One theory of the SNARC effect holds that these effects traction are implemented, which underlie the similar
are the result of reading habits (Dehaene et al., 1993; biases. So, while the SNARC data are far from a decisive
Nuerk et al., 2015; Shaki & Fischer, 2008). One may think point in favour of accounts with distinct representations, it
that such an account can explain the difference, as this dif- is easier to explain the absence of the inflexible effect with
ferentiates the SNARC effect from any innate biases that symbolic stimuli if the representations differ.
might be responsible for the effects observed with non-
symbolic stimuli. The SNARC effect would attach to the
Ordinal tasks
representation of the visual number form (the symbol-spe-
cific part of the triple-code model). The non-symbolic The behavioural tests discussed so far have focussed on
effect would be specific to the numerical representations cardinal numbers, as the ANS is best known for its func-
of the ANS. tion with amounts (e.g., seeing the difference between four
However, such a simple solution is not readily availa- apples and eight apples). There are, however, also interac-
ble. Both the visual number form and the ANS representa- tions with spatial representations in the case of tasks with
tion are activated in the symbolic tasks in question, ordinal numbers. It is important to keep in mind, though,
according to the triple-code model (Klein et al., 2016). that these tasks may not tell us much about the ANS. The
2430 Quarterly Journal of Experimental Psychology 72(10)

current evidence suggests that ordinality is processed dif- de Hevia, & Girelli, 2010; Van Opstal, Fias, Peigneux, &
ferently from cardinality. Instead of the ANS, it probably Verguts, 2009; Zhang et al., 2016).
relies on short-term memory mechanisms (Lyons, Vogel,
& Ansari, 2016 for a review). We nevertheless chose to
Symbolic tasks
include these studies, since the effects are very similar to
the SNARC effect, and the tasks do involve numbers. As a On the symbolic side, similar results have been found for
result, we think that the studies should be considered even sequences of numbers which participants have to memo-
if it is only for the sake of demarcating the limits to study- rise. van Dijck and Fias (2011) found a SNARC effect
ing the nature of the ANS on the basis of spatial–numerical related to the position of a number in the sequence that par-
associations. ticipants memorised in a short period of time, in a test
where they had to answer if the number shown was part of
the sequence. Aside from short-term memory, verbal work-
Non-symbolic tasks ing memory also influences spatial attention: retrieving
Another source of spatial biases is not related to cardinal items occurring later in a memorised sequence causes
numbers (4, 5, etc.) but to ordinal numbers (4th, 5th, etc.). attentional shifts towards the right (van Dijck, Abrahamse,
For these numbers, too, there are a number of spatial– Acar, Ketels, & Fias, 2014; van Dijck, Abrahamse, Majerus,
numerical associations. One study looked at infants’ & Fias, 2013).
responses to increasing and decreasing series consisting of Not all experiments with ordinal sequences found a
arrays of dots, when presented left to right and right to left SNARC effect. Cheung and Lourenco (2016) tested
(de Hevia, Girelli, Addabbo, & Macchi Cassia, 2014). whether there is a SNARC effect in tasks where partici-
They found that increasing, but not decreasing, series of pants have to decide whether two numbers/letters are in
three arrays of rectangles presented in a left-to-right for- ascending or descending order. They found that in the case
mat were detected by infants. Only for increasing series of numbers, there is a SNARC effect: participants respond
did they look longer at test displays with the opposite (i.e., faster with the left button for ascending pairs and faster
decreasing) series presented in the same format (de Hevia with the right for descending pairs. They found no such
et al., 2014). effects for letters, where ascending order meant alphabeti-
A later study by de Hevia, Veggiotti, Streri, and Bonn cal order (Cheung & Lourenco, 2016). Another difference
(2017) used increasing and decreasing sequences of tones between numerical and non-numerical sequences has been
to see if 0- to 3-day-old infants associate increasing reported by Schroeder, Nuerk, and Plewnia (2017). They
sequences with the right side of space and decreasing tested SNARC effects for both the numbers 1-5 and the
sequences with the left side. Infants only seem to make this weekdays Monday-Friday. First, they found left-to-right
association when numerical information is available. In SNARC effects for both sequences. Second, when they
control studies where the number of tones in the sequence used transcranial direct current stimulation (tDCS) to test
was kept constant, but duration and the length of associ- whether the same prefrontal circuit was responsible for
ated lines was varied, children failed to associate the both, they found a difference between the two. During
increasing sequence with the right side of space (and vice anodal tDCS, participants still had a left-to-right SNARC
versa for the decreasing sequence). They do make the effect with the number sequence, but displayed a right-to-
association, however, when the number of tones varies but left SNARC effect with the weekdays (Schroeder et al.,
duration or line length is kept constant. Even very young 2017). This gives us some reasons to think that the SNARC
infants associate “few” with the left side of space and effect with other ordinally ordered sequences is not caused
“many” with the right side (de Hevia et al., 2017). in exactly the same way as the SNARC effect for the num-
Finally, there is a suggestion that a SNARC-like effect bers. Furthermore, this prefrontal flexibility is consistent
can be observed even for notations that are not numerals with the flexibility of the SNARC effect that was previ-
but do have an ordinal ordering (i.e., a fixed first element, ously shown in studies with different linguistic groups.
second element, etc.). Because no numbers are involved,
these stimuli are non-symbolic even though they are
Similarities and differences
linguistic.
Gevers, Reynvoet, and Fias (2003) first tested this by Spatial biases observed with ordinal number tasks are very
looking for a SNARC effect with letters and names of the similar to those observed in the case of cardinal numbers.
months. A SNARC effect was observed both when the ordi- Furthermore, it seems that the SNARC-like effects found
nal information was task relevant and when it was task for non-symbolic stimuli (non-numerical sequences) are
irrelevant (Gevers et al., 2003; Gevers, Reynvoet, & Fias, very similar to those found for symbolic stimuli. However,
2004). In fact, SNARC effects have even been observed the evidential value of these experiments for the current
for weakly (ordinally) ordered sequences such as the question (whether the representations of the ANS are shared
Chinese colour words and for arbitrary sequences (Previtali, with the representations of symbolic numbers) is limited.
Buijsman and Tirado 2431

The SNARC-like effects observed with ordinal tasks do also drew clock faces to test for lateralised spatial attention
not seem to stem from the same source as the SNARC bias. Their research shows that interhemispheric competi-
effect with cardinal numbers. The spatial biases in the tion can help to explain the directionality of numerical
ordinal case seem related to working memory, as is the magnitude and distinguish these spatial biases from gen-
case with ordinal number processing in general (Lyons eral behavioural attention bias. Dynamic interhemispheric
et al., 2016). The study by Schroeder et al. (2017) further competition seems to underlie the spatial biases for sym-
confirms this hypothesis of a relevant difference between bolic numbers. The dynamic nature of this competition
ordinal and cardinal numbers. As a result, it is unlikely that also helps explain the flexibility of the SNARC effect
ordinal tasks can tell us anything about the representations somewhat, though no experiments have been done yet to
of the ANS, even though the spatial biases are very similar confirm that the situations in which the SNARC effect
and the tasks can involve symbolic representations of car- reverses fit the hypothesis that reversals are due to dynamic
dinal numbers (as in the case of experiments with the interhemispheric competition.
sequence 1-5). The similarity between the symbolic and Another finding that supports interhemispheric compe-
non-symbolic case is, as a result, more likely to be due to tition is the one found by Cohen Kadosh, Kadosh, Kaas,
the reliance on working memory with both tasks than due Henik, and Goebel (2007). They showed that representa-
to some feature of the numerical representations. tions of numbers are located in the parietal lobes in a func-
tional magnetic resonance adaptation (fMRA) study
looking at brain activation when perceiving natural lan-
Neuroimaging studies
guage and symbolic numerals (“two” and “2”). Participants
Most of the research on the spatial representation of num- were presented with different stimuli of all symbolic
bers has been behavioural, which means that there are numerals, all natural language numerals, or a mix. Brain
currently no well-established hypotheses regarding the activation patterns were then compared for the different
underlying brain mechanisms (and is the reason that most stimuli. These suggest that both types of numerals are pro-
of the evidence reviewed here is behavioural). However, cessed (in the same way) in the parietal lobes. They also
the evidence indicates that arithmetic and other areas found evidence of a left–right asymmetry in the parietal
linked to numerosity tasks are mostly located in the lobe function regarding number representations. Despite
left occipitotemporal and bilateral intraparietal cortex that their findings seem to support a shared representation
(see Dehaene & Cohen, 2007 for a review). Therefore, a between written and spoken numerals, Cohen Kadosh
neuro-recycling hypothesis seems suitable to explain et al. (2007) do not discard that other neuronal populations
how these mechanisms build number representations. are involved in numeric processing in the left parietal lobe,
According to Gould and Vrba (1982), the neuronal recy- but are being masked. It may be that the resolution limita-
cling hypothesis proposes that neuronal mechanisms that tions of non-invasive imaging techniques have an effect on
evolved for different purposes are now being reassigned the mapping of these neuron populations. Still, in combi-
to processes, such as reading or arithmetic, without nation with the previous study, such an interpretation is
changes in the genetic characteristics of the species (for less likely (Figure 3).
evidence on this hypothesis, see Formisano et al., 2003; In addition, Harvey, Klein, Petridou, and Dumoulin
Lee, 2000; Pica, Lemer, Izard, & Dehaene, 2004). This (2013) used high-field magnetic resonance imaging to
means that representations of numbers need some degree map the neural populations involved with numerosity
of modal input and that the mechanisms have been mod- tasks in the brain. They found that the neural population
erated by our evolutionary process. are organised topographically in the parietal cortex and
To continue along the lines of Schroeder et al. (2017), that numerical representations require a certain degree of
there have been studies into the SNARC effect, number sensorimotor activation. Piazza and Eger (2016) review
representation, and numerosity tasks that use neuroimag- findings which show that intraparietal activation occurs
ing to locate which brain areas are involved with number in humans and macaque monkeys’ brains when encoding
processing. Arshad et al. (2016) were interested in finding for numerosity tasks. Following this line, other studies
the neural mechanism that modulated numerical cognition have found a similar link between numerosity and num-
and, more specifically, numerical magnitude. To achieve ber representation and its link to the parietal cortex (in
this, they combined visual (two elongated rectangles) and line with the triple-code model). To be more precise, the
vestibular (cold or warm water flowing past one ear) stim- inferior parietal lobes are needed for processing numeri-
ulation to induce interhemispheric conflict. Follow-up cal and visual spatial information (see Chochon, Cohen,
experiments using tDCS confirmed the occurrence of such Moortele, & Dehaene, 1999; Dehaene et al., 2003; Fias,
conflict and subsequent unihemispheric inhibition. Lammertyn, Reynvoet, Dupont, & Orban, 2003; Pesenti,
Participants performed a number pair bisection task, hav- Thioux, Seron, & De Volder, 2000). The parietal cortex
ing to name the number in the middle of the interval given seems to be the key area regarding numerosity and spatial
by a pair of numbers in both types of experiment. They coding, with additional areas activated for more complex
2432 Quarterly Journal of Experimental Psychology 72(10)

Figure 3.  The stimuli used by Cohen Kadosh, Kadosh, Kaas, Henik, and Goebel (2007) to test for the locations of numerical
processing in the brain.
They compared imaging data for observations of digits, words and a mix of the two.

numerical tasks as mentioned in the introduction. The can be an important test if done more thoroughly. While
overlap between the numerical and spatial areas is why it they are currently inconclusive due to replication failure,
is possible to accept that visual spatial coding is vital for in the spatial–numerical bias on the non-symbolic side, we
numerosity tasks (Knops et al., 2009). So, the data are in expect that there may be such effects since arrays of dots
accordance with the neuro-recycling hypothesis men- induce the same type of attentional bias as symbolic num-
tioned above. bers. Those attentional biases are in turn responsible for
line bisection biases, so it would be surprising if there are
no bisection biases for non-symbolic stimuli even though
Discussion there are relevant attentional biases.
We set out to compare spatial–numerical associations for For now, though, the little behavioural evidence we
symbolic and non-symbolic stimuli, to see to what extent have on bisection tasks suggests that there are different
such a comparison can help decide whether the represen- underlying representations—since we cannot be sure that
tations of the ANS are shared with our symbolic number there are relevant spatial biases for non-symbolic stimuli.
representations. This question is one way in which the This is consistent with the findings from our comparison
competing views of Dehaene et al. (2003) and Gebuis of SNARC effects with symbolic and non-symbolic stim-
et al. (2016) can be distinguished. For the triple-code uli. Even on otherwise identical tasks (numerical compari-
model to be correct, there has to be a representation of sons with numbers vs numerical comparisons with arrays
numbers that is shared between symbolic and non-sym- of dots), the behavioural signature is different. The bias
bolic formats. For the sensory integration view of the ANS with symbolic representations is flexible, whereas the only
to be correct, there have to be different representations for kind of flexibility found with non-symbolic stimuli (the
the symbolic and non-symbolic case, since the non-sym- phrasing of the question) is one of the few points in which
bolic representations are not numerical but magnitude the symbolic bias seems to be unchanged.
based. Evidence that tries to decide between these two These differences in behaviour are more easily
views more directly is currently inconclusive (Gebuis explained if the underlying representations are different.
et al., 2016; Leibovich et al., 2017). As this review has The main reason is that, on the triple-code model, the rep-
shown, evidence based on spatial–numeric associations is resentations and processes of the ANS are also used in
also not yet conclusive, but has the potential to help decide tasks involving symbolic numbers. Since these representa-
the question. tions are activated on the triple-code model, and are sup-
The behavioural evidence is mixed, but we think it may posed to be inherently spatial, one would expect to see
swing in favour of shared representations between the some effects of the spatial biases observed with non-sym-
ANS and the symbolic number system. Line bisection bolic stimuli in the case of symbolic stimuli. Currently, the
tasks, which are good measures, since the only alteration is triple-code model does not seem to have an explanation as
to exchange flanking numbers for flanking arrays of dots, to why this is not observed with SNARC-like effects.
Buijsman and Tirado 2433

This puts the behavioural evidence in a tricky position. neuroimaging studies that are available do point at fruitful
We expect the bisection tasks to show similar behavioural avenues of research, though. A combination of methods
signatures, but the SNARC tasks are unlikely to turn out has lead to the hypothesis that spatial–numerical biases for
more similar across formats with continued testing. We see symbolic numbers are based on dynamic interhemispheric
two ways out of this position. On one hand, the observations competition. If these studies are supplemented with exper-
with bisection tasks may be due to more general attentional iments that look at the SNARC-like effects observed with
biases. In other words, the effects may be the result of differ- non-symbolic stimuli, then the source of the difference
ent representations (in line with the SNARC findings) if between the two should become a lot clearer. There are,
magnitude-related variables such as total surface area can more importantly, suggestions that there is a shared repre-
explain the attentional bias with non-symbolic stimuli. The sentation for the different formats of symbolic numerical
one eye-tracking study with non-symbolic stimuli suggests stimuli. Since no studies have been done with non-symbolic
otherwise, but since it is a single study which is in conflict stimuli, it is too early to say anything based on brain imag-
with the replication study on bisection biases, there is good ing studies, but it seems that future studies that include
reason to attempt a replication similar to what Gebuis and non-symbolic stimuli would be helpful for deciding the
Gevers (2011) did for bisection tasks in tandem to more question raised in this review.
studies with bisection tasks using non-symbolic stimuli. In general, we have found that there is no clear support
The alternative is that the bisection biases are specific to for either of the two positions; the triple-code model with
numerical stimuli (whether symbolic or non-symbolic) and shared representations across formats or the general mag-
stem from the spatial nature of our numerical representa- nitude system with different representations for symbolic
tions. In this case, the discrepancy in SNARC-like effects and non-symbolic stimuli. The situation may become
requires explanation. The brain studies discussed may be of clearer if there is further research on a number of points.
some help here. The interhemispheric competition found for First, the data on line bisection tasks are unclear, which
symbolic representations can, with further studies, be com- calls for further research on bisection tasks with non-sym-
pared to the source of the SNARC-like effect with non-sym- bolic stimuli together with research on attentional biases
bolic stimuli. A clearer picture of the neural differences may due to non-symbolic stimuli. Since line bisection with
help pinpoint why we do not observe the non-symbolic symbolic stimuli is the result of attentional biases, the cur-
biases when testing with symbolic stimuli. It may be possible rently supported combination of attentional bias without
for the triple-code model to be refined in a way that explains bisection bias seems unlikely. This will also clarify whether
why the spatial nature of the ANS representations (suppos- there is a specific numerical bisection/attentional bias,
edly demonstrated by the non-symbolic SNARC-like effect) which would point in favour of the triple-code model.
does not influence our performance with tasks that demon- Second, the reason for the marked difference between
strate a symbolic SNARC effect. Neuroimaging studies the non-symbolic SNARC-like effect and the symbolic
focussing on this difference may help the theoretical work SNARC effect requires further study. We expect that the
necessary to fit this discrepancy into the triple-code model. behavioural effects are sufficiently well-studied for there
The review also discussed findings that tried to establish to be few surprises from further behavioural testing.
what neural mechanisms underlie these spatial–numeric However, since there is a clear hypothesis on the neural
associations. These findings can also help to decide the mechanism behind the symbolic SNARC effect, it will be
central question. Should brain studies find that the spatial– helpful if similar studies are carried out for the non-symbolic
numeric associations for non-symbolic and symbolic effect.
numerical stimuli are based on the same neural mecha- Third, there is a total absence of brain studies that look
nisms, then this would support the hypothesis of shared at numerical processes across symbolic and non-symbolic
representations. Vice versa, an observation of different neu- formats. Such studies seem extremely relevant given the
ral mechanisms for the two cases supports the hypothesis heated debate between the triple-code model and the gen-
that there are different representations. Furthermore, eral magnitude system. Furthermore, since a similar study
they are an important addition to the behavioural studies has been carried out successfully with different symbolic
discussed earlier. It is difficult to say that the observed formats, there is a template readily available for such
behaviour fits, or does not fit, the existence of shared repre- cross-format studies.
sentations when we are unsure which processes produced Fourth, we want to encourage behavioural studies that
the behaviour. By looking for overlap in the processes make the comparison between spatial biases with sym-
across formats (e.g., the symbolic SNARC effect and the bolic and non-symbolic stimuli with the same subjects.
non-symbolic SNARC-like effect), we get a better sense of Such an intraindividual comparison can be helpful particu-
the relation between the observed behaviour and the under- larly for the differences observed with the SNARC effect.
lying representations and processes. We have assumed that the results for non-symbolic stimuli
So far, however, there are not enough data available to are culture independent, but to the best of our knowledge,
further clarify the behavioural studies discussed here. The this has not been investigated thoroughly. There may be
2434 Quarterly Journal of Experimental Psychology 72(10)

some variability in the non-symbolic effect in line with the Chochon, F., Cohen, L., Moortele, P. V. D., & Dehaene, S.
variability in the dominant direction of the SNARC effect. (1999). Differential contributions of the left and right
In other words, the difference in flexibility may be more a inferior parietal lobules to number processing. Journal of
difference of degree than a difference in kind. Cognitive Neuroscience, 11, 617–630.
Cohen Kadosh, R., Kadosh, K., Kaas, A., Henik, A., & Goebel,
Altogether, we think these are promising directions
R. (2007). Notation-dependent and -independent representa-
of research that will help us decide between accounts
tions of numbers in the parietal lobes. Neuron, 53, 307–314.
such as the triple-code model on which symbolic and Dehaene, S. (1992). Varieties of numerical abilities. Cognition,
non-symbolic representations are shared and accounts 44, 1–42.
that prefer a general magnitude system with distinct Dehaene, S., Bossini, S., & Giraux, P. (1993). The mental rep-
representations. Currently, the behavioural data are resentation of parity and number magnitude. Journal of
unclear, even taken aside from the difficult to say that Experimental Psychology: General, 122, 371–396.
behaviour on similar tasks is not the result of different Dehaene, S., & Changeux, J. (1993). Developments of elemen-
processes. However, with the directions outlined here, tary numerical abilities: A neuronal model. Journal of
the behavioural evidence should become clearer. By Cognitive Neuroscience, 5, 390–407.
linking it to neuroimaging studies, it should, moreover, Dehaene, S., & Cohen, L. (2007). Cultural recycling of cortical
maps. Neuron, 56, 384–398.
be possible to say with more certainty that the observed
Dehaene, S., Dupoux, E., & Mehler, J. (1990). Is numerical com-
behaviours are relevant to understanding our numerical
parison digital? Analogical and symbolic effects in two-digit
representations. So, even though this review has not number comparison. Journal of Experimental Psychology:
managed to add strong support for either position, we Human Perception and Performance, 16, 626–641.
do think that it suggests a way forward that may help Dehaene, S., Piazza, M., Pinel, P., & Cohen, L. (2003).
resolve the debate. Three parietal circuits of number processing. Cognitive
Neuropsychology, 20, 487–506.
Declaration of conflicting interests de Hevia, M., Girelli, L., Addabbo, M., & Macchi Cassia, V.
(2014). Human infants’ preference for left-to-right oriented
The author(s) declared no potential conflicts of interest with
increasing numerical sequences. PLoS ONE, 9(5), e96412.
respect to the research, authorship, and/or publication of this
de Hevia, M., Girelli, L., & Vallar, G. (2006). Numbers and
article.
space: A cognitive illusion? Experimental Brain Research,
168, 256–264.
Funding de Hevia, M., & Spelke, E. (2009). Spontaneous mapping of
The author(s) received no financial support for the research, number and space in adults and young children. Cognition,
authorship, and/or publication of this article. 110, 198–207.
de Hevia, M., Vallar, G., & Girelli, L. (2008). Visualizing num-
ORCID iD bers in the mind’s eye: The role of visuo-spatial processes
in numerical abilities. Neuroscience and Biobehavioral
Carlos Tirado https://orcid.org/0000-0003-3937-9372
Reviews, 32, 1361–1372.
de Hevia, M., Veggiotti, L., Streri, A., & Bonn, C. (2017). At
References
birth, humans associate “few” with left and “many” with
Andres, M., Ostry, D., Florence, N., & Paus, T. (2008). Time right. Current Biology, 27, 3879–3884.
course of number magnitude interference during grasping. Feigenson, L., Dehaene, S., & Spelke, E. (2004). Core systems of
Cortex, 44, 414–419. number. Trends in Cognitive Sciences, 8, 307–314.
Arsalidou, M., & Taylor, M. (2011). Is 2+2=4? Meta-analyses Fias, W., Brysbaert, M., Geypens, F., & D’Ydewalle, G. (1996).
of brain areas needed for numbers and calculations. The importance of magnitude information in numerical pro-
NeuroImage, 54, 2382–2393. cessing: Evidence from the SNARC effect. Mathematical
Arshad, Q., Nigmatullina, Y., Nigmatullin, R., Asavarut, P., Goga, Cognition, 2, 95–110.
U., Khan, S., . . . Bronstein, A. M. (2016). Bidirectional Fias, W., & Fischer, M. (2005). Spatial representation of num-
modulation of numerical magnitude. Cerebral Cortex, 26, ber. In J. I. D. Campbell (Ed.), Handbook of mathematical
2311–2324. cognition (pp. 43–54). New York, NY: Psychology Press.
Bächtold, D., Baumüller, M., & Brugger, P. (1998). Stimulus- Fias, W., Lammertyn, J., Reynvoet, B., Dupont, P., & Orban, G. A.
response compatibility in representational space. (2003). Parietal representation of symbolic and nonsymbolic
Neuropsychologia, 36, 731–735. magnitude. Journal of Cognitive Neuroscience, 15, 47–56.
Barth, H., Kanwisher, N., & Spelke, E. (2003). The construc- Fias, W., Lauwereyns, J., & Lammertyn, J. (2001). Irrelevant dig-
tion of large number representations in adults. Cognition, its affect feature based attention depending on the overlap of
86, 201–221. neural circuits. Cognitive Brain Research, 12, 415–423.
Bulf, H., de Hevia, M., & Macchi Cassia, V. (2016). Small on the left, Fias, W., van Dijck, J., & Gevers, W. (2011). How number is
large on the right: Numbers orient visual attention onto space in associated with space? The role of working memory. In S.
preverbal infants. Developmental Science, 19, 394–401. Dehaene & E. M. Brannon (Eds.), Space, time and number
Cheung, C., & Lourenco, S. (2016). The associations between in the brain: Searching for the foundations of mathemati-
space and order in numerical and non-numerical sequences. cal thought (pp. 133–148). Amsterdam, The Netherlands:
Consciousness and Cognition, 45, 124–134. Elsevier Science.
Buijsman and Tirado 2435

Fischer, M. (2001). Number processing induces spatial perfor- Quarterly Journal of Experimental Psychology, 65, 1044–
mance biases. Neurology, 57, 822–826. 1051.
Fischer, M. (2003). Spatial representations in number process- Hubbard, E., Piazza, M., Pinel, P., & Dehaene, S. (2005).
ing—Evidence from a pointing task. Visual Cognition, 10, Interactions between number and space in parietal cortex.
493–508. Nature Reviews Neuroscience, 6, 435–448.
Fischer, M., & Knops, A. (2014). Attentional cueing in numeri- Hung, Y., Hung, D., Tzeng, O., & Wu, D. (2008). Flexible spatial
cal cognition. Frontiers in Psychology, 5, Article 1381. mapping of different notations of numbers in Chinese read-
Fischer, M., Mills, R., & Shaki, S. (2010). How to cook a ers. Cognition, 106, 1441–1450.
SNARC: Number placement in text rapidly changes spatial– Ito, Y., & Hatta, T. (2004). Spatial structure of quantitative rep-
numerical associations. Brain and Cognition, 72, 333–336. resentation of numbers: Evidence from the SNARC effect.
Fischer, M., & Shaki, S. (2014). Spatial associations in numerical Memory & Cognition, 32, 662–673.
cognition—From single digits to arithmetic. The Quarterly Izard, V., Sann, C., Spelke, E. S., & Streri, A. (2009). Newborn
Journal of Experimental Psychology, 67, 1461–1483. infants perceive abstract numbers. Proceedings of the
Fischer, M., Warlop, N., Hill, R., & Fias, W. (2004). Oculomotor National Academy of Sciences, 106, 10382–10385.
bias induced by number perception. Experimental Psycho- Klein, E., Suchan, J., Moeller, K., Karnath, H., Knops, A., Wood,
logy, 51, 1–7. G., . . . Willmes, K. (2016). Considering structural connectiv-
Formisano, E., Kim, D. S., Di Salle, F., van de Moortele, P. F., ity in the triple code model of numerical cognition: Differential
Ugurbil, K., & Goebel, R. (2003). Mirror-symmetric tono- connectivity for magnitude processing and arithmetic facts.
topic maps in human primary auditory cortex. Neuron, 40, Brain Structure and Function, 221, 979–995.
859–869. Knops, A. (2017). Probing the neural correlates of number pro-
Galton, F. (1880). Visualised numerals. Nature, 21, 252–256. cessing. The Neuroscientist, 23, 264–274.
Gebuis, T., Cohen Kadosh, R., & Gevers, W. (2016). Sensory- Knops, A., Thirion, B., Hubbard, E., Michel, V., & Dehaene, S.
integration system rather than approximate number system (2009). Recruitment of an area involved in eye movements
underlies numerosity processing: A critical review. Acta during mental arithmetic. Science, 324, 1583–1585.
Psychologica, 171, 17–35. Kopiske, K., Löwenkap, C., Eloka, O., Schiller, F., Kao, C., Wu,
Gebuis, T., & Gevers, W. (2011). Numerosities and space; C., . . . Franz, V. (2016). The SNARC effect in Chinese
Indeed a cognitive illusion! A reply to de Hevia and Spelke numerals: Do visual properties of characters and hand signs
(2009). Cognition, 121, 253–255. influence number processing? PLoS ONE, 11(9), e0163897.
Gebuis, T., & van der Smagt, M. (2011). False approximations of Lammertyn, J., Fias, W., & Lauwereyns, J. (2002). Semantic
the approximate number system? PLoS ONE, 6(10), e25405. influences on feature-based attention due to overlap of neu-
Gevers, W., & Lammertyn, J. (2005). The hunt for SNARC. ral circuits. Cortex, 38, 878–882.
Psychology Science, 47, 10–21. Lee, K. (2000). Cortical areas differentially involved in multi-
Gevers, W., Reynvoet, B., & Fias, W. (2003). The mental rep- plication and subtraction: A functional magnetic resonance
resentation of ordinal sequences is spatially organized. imaging study and correlation with a case of selective acal-
Cognition, 87(3), B87–95. culia. Annals of Neurology, 48, 657–661.
Gevers, W., Reynvoet, B., & Fias, W. (2004). The mental rep- Leibovich, T., Katzin, N., Harel, M., & Henik, A. (2017). From
resentation of ordinal sequences is spatially organised: “sense of number” to “sense of magnitude”: The role of con-
Evidence from days of the week. Cortex, 40, 171–172. tinuous magnitudes in numerical cognition. Behavioral and
Göbel, S. (2015). Up or down? Reading direction influences ver- Brain Sciences, 40, e164.
tical counting direction in the horizontal plane—A cross- Lindemann, O., Abolafia, J., Pratt, J., & Bekkering, H. (2008).
cultural comparison. Frontiers in Psychology, 6, Article Coding strategies in number space: Memory requirements
228. influence spatial–numerical associations. The Quarterly
Gould, S. J., & Vrba, E. S. (1982). Exaptation –a missing term in Journal of Experimental Psychology, 61, 515–524.
the science of form. Paleobiology, 8, 4–15. Lipton, J., & Spelke, E. (2003). Origins of number sense. Large-
Guillaume, M., Nys, J., Mussolin, C., & Content, A. (2013). number discrimination in human infants. Psychological
Differences in the acuity of the Approximate Number Science, 14, 396–401.
System in adults: The effect of mathematical ability. Acta Loetscher, T., Bockisch, C., Nicholls, M., & Brugger, P. (2010).
Psychologica, 144, 506–512. Eye position predicts what number you have in mind.
Halberda, J., & Feigenson, L. (2008). Developmental change Current Biology, 20, R264–R265.
in the acuity of the “Number Sense”: The Approximate Longo, M., & Lourenco, S. (2007). Spatial attention and the men-
Number System in 3-, 4-, 5-, and 6-year-olds and adults. tal number line: Evidence for characteristic biases and com-
Developmental Psychology, 44, 1457–1465. pression. Neuropsychologia, 45, 1400–1407.
Harvey, B. M., Klein, B. P., Petridou, N., & Dumoulin, S. O. Longo, M., & Lourenco, S. (2010). Bisecting the mental num-
(2013). Topographic representation of numerosity in the ber line in near and far space. Brain and Cognition, 72,
human parietal cortex. Science, 341, 1123–1126. 362–367.
Holloway, I. D., & Ansari, D. (2009). Mapping numerical magni- Luccio, R., Fumarola, A., Tamburini, G., & Agostini, T. (2012).
tudes onto symbols: The numerical distance effect and indi- The spatial representation of non-symbolic numerical quan-
vidual differences in children’s mathematics achievement. tities. Proceedings of Fechner Day, 28, 321–327.
Journal of Experimental Child Psychology, 103, 17–29. Lyons, I., Vogel, S., & Ansari, D. (2016). On the ordinality
Holmes, K., & Lourenco, S. (2012). Orienting numbers in men- of numbers: A review of neural and behavioural studies.
tal space: Horizontal organization trumps vertical. The Progress in Brain Research, 227, 187–221.
2436 Quarterly Journal of Experimental Psychology 72(10)

McCrink, K., Dehaene, S., & Dehaene-Lambertz, G. (2007). Moving Schroeder, P., Nuerk, H., & Plewnia, C. (2017). Prefrontal neu-
along the number line: Operational momentum in nonsymbolic romodulation reverses spatial associations of non-numerical
arithmetic. Perception & Psychophysics, 69, 1324–1333. sequences, but not numbers. Biological Psychology, 128,
McCrink, K., & Opfer, J. (2014). Development of spatial-numerical 39–49.
associations. Current Directions in Psychological Science, Seron, X., Pesenti, M., Noël, M., Deloche, G., & Cornet, J.
23, 439–445. (1992). Images of numbers, or when 98 is upper left and 6
McCrink, K., & Wynn, K. (2009). Operational momentum in sky blue. Cognition, 44, 159–196.
large-number addition and subtraction by 9-month olds. Shaki, S., & Fischer, M. (2008). Reading space into numbers—
Journal of Experimental Child Psychology, 103, 400–408. A cross-linguistic comparison of the SNARC effect.
Mitchell, T., Bull, R., & Cleland, A. (2012). Implicit response- Cognition, 108, 590–599.
irrelevant number information triggers the SNARC effect: Shaki, S., & Fischer, M. (2012). Multiple spatial mappings in
Evidence using a neural overlap paradigm. The Quarterly numerical cognition. Journal of Experimental Psychology:
Journal of Experimental Psychology, 65, 1945–1961. Human Perception and Performance, 38, 804–809.
Nathan, M., Shaki, S., Salti, M., & Algom, D. (2009). Numbers Shaki, S., Fischer, M., & Göbel, S. (2012). Direction counts: A
and space: Associations and dissociations. Psychonomic comparative study of spatially directional counting biases
Bulletin & Review, 16, 578–582. in cultures with different reading directions. Journal of
Nieder, A., Freedman, D., & Miller, E. (2002). Representation of Experimental Child Psychology, 112, 275–281.
the quantity of visual items in the primate prefrontal cortex. Shaki, S., Fischer, M., & Petrusic, W. (2009). Reading habits for
Science, 297, 1708–1711. both words and numbers contribute to the SNARC effect.
Nuerk, H., Patro, K., Cress, U., Schild, U., Friedrich, C., & Göbel, Psychonomic Bulletin & Review, 16, 328–331.
S. (2015). How space-number associations may be created Shaki, S., Petrusic, W., & Leth-Steensen, C. (2012). SNARC
in preliterate children: Six distinct mechanisms. Frontiers in effects with numerical and non-numerical symbolic com-
Psychology, 6, Article 215. parative judgments: Instructional and cultural dependencies.
Patro, K., & Shaki, S. (2016). SNARC for numerosities is modu- Journal of Experimental Psychology Human Perception &
lated by comparative instruction (and resembles some non- Performance, 38, 515–530.
numerical effects). Cognitive Processing, 17, 127–137. Umilta, C., Priftis, K., & Zorzi, M. (2009). The spatial repre-
Pesenti, M., Thioux, M., Seron, X., & Volder, A. D. (2000). sentation of numbers: Evidence from neglect and pseudon-
Neuroanatomical substrates of Arabic number processing, eglect. Experimental Brain Research, 192, 561–569.
numerical comparison, and simple addition: A PET study. van Dijck, J., Abrahamse, E., Acar, F., Ketels, B., & Fias, W.
Journal of Cognitive Neuroscience, 12, 461–479. (2014). A working memory account of the interaction
Piazza, M., & Eger, E. (2016). Neural foundations and functional between numbers and spatial attention. Quarterly Journal
specificity of number representations. Neuropsychologia, of Experimental Psychology, 67, 1500–1513.
83, 257–273. van Dijck, J., Abrahamse, E., Majerus, S., & Fias, W. (2013). Spatial
Pica, P., Lemer, C., Izard, V., & Dehaene, S. (2004). Exact and attention interacts with serial-order retrieval from verbal work-
approximate arithmetic in an Amazonian indigene group. ing memory. Psychological Science, 24, 1854–1859.
Science, 306, 499–503. van Dijck, J. P., & Fias, W. (2011). A working memory account
Pinhas, M., & Fischer, M. (2008). Mental movements without for spatial -numerical associations. Cognition, 119, 114–119.
magnitude? A study of spatial biases in symbolic arithmetic. Van Opstal, F., Fias, W., Peigneux, P., & Verguts, T. (2009).
Cognition, 109, 408–415. The neural representation of extensively trained ordered
Previtali, P., de Hevia, M., & Girelli, L. (2010). Placing order in sequences. NeuroImage, 47, 367–375.
space: The SNARC effect in serial learning. Experimental Wood, G., Willmes, K., Nuerk, H., & Fischer, M. (2008). On the
Brain Research, 201, 599–605. cognitive link between space and number: A meta-analysis
Ranzini, M., & Girelli, L. (2012). Exploiting illusory effects to of the SNARC effect. Psychology Science Quarterly, 50,
disclose similarities in numerical and luminance processing. 489–525.
Attention, Perception, & Psychophysics, 75, 1001–1008. Wood, J., & Spelke, E. (2005). Infants’ enumeration of
Ren, P., Nicholls, M., Ma, Y., & Chen, L. (2011). Size matters: actions: Numerical discrimination and its signature limits.
Non-numerical magnitude affects the spatial coding of Developmental Science, 8, 173–181.
response. PLoS ONE, 6, e23553. Yates, M., Nemeh, F., Loetscher, T., Ma-Wyatt, A., & Nicholls,
Rossetti, Y., Jacquin-Courtois, S., Aiello, M., Ishihara, M., M. (2013). Numerosity is represented spatially: Evidence
Brozzolli, C., & Doricchi, F. (2011). Neglect “around the from a “SNARC” task. Perception, 42, 132–133.
clock”: Dissociating number and spatial neglect in right Zebia, S. (2005). Linkages between number concepts, spatial
brain damage. In S. Dehaene & E. M. Brannon (Eds.), Space, thinking, and directionality of writing: The SNARC effect
time and number in the brain: Searching for the foundations and the REVERSE SNARC effect in English and Arabic
of mathematical thought (pp. 149–171). Amsterdam, The monoliterates, biliterates, and illiterate Arabic speakers.
Netherlands: Elsevier Science. Journal of Cognition and Culture, 5, 165–190.
Rugani, R., & de Hevia, M. (2017). Number-space associations Zhang, M., Gao, X., Li, B., Yu, S., Gong, T., Jiang, T., & Chen,
without language: Evidence from preverbal human infants Y. (2016). Spatial representation of ordinal information.
and non-human animal species. Psychonomic Bulletin and Frontiers in Psychology, 7, Article 505.
Review, 24, 352–369. Zimmerman, E. (2018). Small numbers are sensed directly, high
Santens, W., & Gevers, W. (2008). The SNARC effect does not numbers constructed from size and density. Cognition, 173,
imply a mental number line. Cognition, 108, 263–270. 1–7.

You might also like