Finger Tracking Reveals the Covert Stages of
Mental Arithmetic
Pedro Pinheiro-Chagas1,2,3, Dror Dotan1,4, Manuela Piazza5, and Stanislas Dehaene1,3
1 Cognitive Neuroimaging Unit, CEA DRF/I2BM, INSERM, Université Paris-Sud, Université Paris-Saclay,
NeuroSpin center, Francia
2École Doctorale Cerveau-Cognition-Comportement, Université Pierre et Marie Curie, Francia
3Collège de France
4Language and Brain Lab, School of Education and the Sagol School of Neuroscience, Tel Aviv University
5Center for Mind/Brain Sciences University of Trento
un acceso abierto
diario
Palabras clave:
momentum effect
aritmética, finger
tracking,
serial processing, problem-size effect, operational
ABSTRACTO
We introduce a novel method capable of dissecting the succession of processing stages
underlying mental arithmetic, thus revealing how two numbers are transformed into a third.
We asked adults to point to the result of single-digit additions and subtractions on a number
line, while their finger trajectory was constantly monitored. We found that the two operands
are processed serially: the finger first points toward the larger operand, then slowly veers
toward the correct result. This slow deviation unfolds proportionally to the size of the smaller
operand, in both additions and subtractions. We also observed a transient operator effect: a
plus sign attracted the finger to the right and a minus sign to the left and a transient activation
of the absolute value of the subtrahend. These findings support a model whereby addition
and subtraction are computed by a stepwise displacement on the mental number line,
starting with the larger number and incrementally adding or subtracting the smaller number.
INTRODUCCIÓN
Despite decades of research in cognitive arithmetic, how the brain performs elementary arith-
metic calculations remains largely unknown. The widely replicable problem-size effect is
the finding that response times (RTs) and error rates increase as a function of the size of
the operands to be added or subtracted. By investigating the properties of the problem-size
effect across operations and during development, researchers have proposed different cogni-
tive models of arithmetic (Zbrodoff & logan, 2005).
In their seminal study, Groen and Parkman (1972) found that the best predictor of single-
digit addition RTs in first graders was the size of the smaller operand (mín.). They proposed that
children use a counting strategy to solve additions by starting from the larger operand and then
incrementing it with the min, with a slope of about 410 ms per unit. A much smaller slope,
sin embargo, was found in adults (20 ms/unit). This seemed too fast for a counting strategy, y
the authors proposed that adults directly retrieve the results from long-term memory.
Fact retrieval is thought to be a dominant strategy in adults, but some data show that
it is also supplemented by other strategies (Lefevre & Kulak, 1994; Siegler, 1987; Zbrodoff &
logan, 2005). Por ejemplo, Butterworth, Zorzi, Girelli, and Jonckheere (2001) proposed that,
given the commutativity of addition, only half of the table may be stored in long-term memory
Citación: Pinheiro-Chagas, PAG., Dotan,
D., Piazza, METRO., & Dehaene, S. (2017).
Finger tracking reveals the covert
stages of mental arithmetic. Open
Mente: Discoveries in Cognitive
Ciencia, 1(1), 30–41.
doi:10.1162/opmi_a_00003
DOI:
http://doi.org/10.1162/opmi_a_00003
Materiales suplementarios:
www.mitpressjournals.org/doi/suppl/
10.1162/opmi_a_00003
Recibió: 13 Abril 2016
Aceptado: 23 Julio 2016
Conflicto de intereses: Los autores
declare no competing interests.
Autor correspondiente:
Pedro Pinheiro-Chagas
ppinheirochagas@gmail.com
Derechos de autor: © 2017
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
oh
pag
metro
_
a
_
0
0
0
0
3
pag
d
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
(problems in which the first operand is larger than the second, which could be progressively
committed to memory as a result using the min counting strategy at a younger age). To solve
problems presented in the opposite order (p.ej., 2 + 7), participants would reorder the operands
prior to retrieval. Adults’ RTs for additions were indeed found to be higher when the first
operand was smaller, presumably due to this additional reordering stage.
Some models propose that arithmetic problems are solved by quantity manipulation,
possibly relying on a “mental number line” (Dehaene, Spelke, Pinel, Stanescu, & Tsivkin,
1999). Considerable research indicates that children and adults possess such a space-like
left-to-right numerical representation, and that arithmetic may involve internal movements on
this representation (Barrouillet & Thevenot, 2013; Dehaene & Changeux, 1993; Fayol &
Thevenot, 2012; Knops, Thirion, Hubbard, Michel, & Dehaene, 2009; Knops, Viarouge, &
Dehaene, 2009; Mathieu, Gourjon, Couderc, Thevenot, & Prado, 2016; McCrink, Dehaene,
& Dehaene-Lambertz, 2007; Restle, 1970; Uittenhove, Thevenot, & Barrouillet, 2016).
Respectivamente, mental arithmetic causes spatial biases similar to the SNARC (spatial-numerical
association of response codes) effect with single numbers (Dehaene, Bossini, & Giraux, 1993):
addition draws attention and eye movements toward the right side of space, and subtraction
toward the left (Knops, Viarouge, et al., 2009; Mathieu et al., 2016; Pinhas, Shaki, & pescador,
2014). There is also a tendency to overestimate the results of addition and to underestimate
the results of subtraction, which can be interpreted as an excessive motion on the number line
and has therefore been termed the “operational momentum” (OM) efecto (Knops, Viarouge,
et al., 2009; McCrink et al., 2007). Sin embargo, it is still unknown whether those effects betray
a genuine use of the number line during calculation, or are merely an automatic attraction to
the result after it has been calculated.
Progress in understanding mental arithmetic is impeded by the fact that RTs and error
rates provide only a single summary measure of the entire calculation process, blind to the
succession of intermediate stages. Aquí, we introduce an online measurement method that
addresses this temporal dissection problem: continuous finger tracking (Dotan & Dehaene,
2013, 2016; Song & Nakayama, 2009; see also Freeman & Ambady, 2010, for a similar
approach with mouse tracking). Participants solved single-digit additions and subtractions on
a tablet computer, and responded by pointing to the position of the result on a horizontal
number line ranging from 0 a 10, while their finger trajectory was continuously monitored.
By identifying which cognitive factors affect finger location at each time point, we aimed to
answer several questions: Are the two operands processed serially or in parallel? Is there
a stage whose duration increases linearly with the size of the numerical quantities, como
implied by models of counting (mín.) or motion on the number line? Can we visualize a re-
ordering of the two operands when solving additions, as predicted by the comparison model?
And can we determine the moment when the visuo-spatial biases underlying addition and
subtraction occur?
METHOD
Thirty right-handed French adults aged between 20 y 45 (m = 24, DE = 5) participó
en el estudio. The participants saw a series of single-digit addition and subtraction problems
on a tablet computer and were instructed to point at the position of the result on a horizontal
number line marked with 0 y 10 at its extremities (ver figura 1). On each trial, Participantes
first touched an initiation rectangle, which made a fixation cross appear above the middle of
the number line. When participants started moving their finger toward the number line, un
arithmetic operation appeared at fixation for 250 EM. Participants then continued moving their
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
31
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
oh
pag
metro
_
a
_
0
0
0
0
3
pag
d
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
Cifra 1. Task and screen layout.
finger to what they judged to be the position of the result. When the finger reached the number
line, a feedback arrow indicated the location where the finger landed.
Using an Apple iPad 2 in landscape orientation, finger position was sampled at 60 Hz
(1 ms accuracy), resampled at exactly 100 Hz using cubic spline interpolation, and smoothed
(Gaussian, σ = 20 EM). For each time point, we calculated the instantaneous direction as
the vector difference between the finger coordinates at times t−10 ms and t, and the implied
endpoint (iEP) as the position on the number line that the finger would reach if it kept moving
straight in this direction.
The experiment included two blocks, presented in random order, both mixing single-digit
additions and subtractions (individual digits were also presented, but are not reported here).
The blocks were designed to control for possible confounds in the analyses of the OM effect.
Block 1 included all single-digit addition and subtraction problems with matched operands
entre 1 y 9 (p.ej., 4 + 3 y 4 − 3), Resultando en 25 additions with larger-first operands
(denoted L + S, where L is the larger and S the smaller number) y 25 subtractions (L − S).
Each problem was repeated six times, for a total of 300 ensayos. In this block, addition results are
generally larger than subtraction results, y, de este modo, the presence of an OM effect could be due
to this bias. como control, we therefore used matched results in block 2. We started from the
54 additions and 45 subtractions with operands ranging from 0 a 9 and results ranging from 1
a 9 (thus including L + S, S + l, and L − S problems). If each problem appeared exactly once,
the distribution of addition and subtraction results would again be asymmetrical. Por lo tanto,
we overrepeated some problems to obtain exactly 20 addition and 20 subtraction trials for
each of the results 1–9 (total of 360 ensayos). Por construcción, block 2 contained all the problems
presented in block 1. Por lo tanto, the OM effect could be investigated in a most unbiased
manner by restricting the analysis to addition and subtraction problems from block 1 (con
identical operands) that were presented in block 2 (with equalized distributions of response
locations). The two blocks also allowed us to test the stability of our findings.
Trajectory analysis followed the method introduced in Dotan and Dehaene (2013). Primero,
for each participant, one regression was run per time point in 30-ms intervals. The dependent
variable was the iEP. Predictors were the two operands, the operator (− or +, coded as −1
y 1), the spatial-reference-points-based bias function (SRP; see Supplementary Materials),
and the result of the previous trial. The latter two predictors were added in all regressions
MENTE ABIERTA: Descubrimientos en ciencia cognitiva
32
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
oh
pag
metro
_
a
_
0
0
0
0
3
pag
d
.
i
F
b
y
gramo
tu
mi
s
t
t
oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
in order to capture maximal variance, as they were significant in previous studies (Dotan &
Dehaene 2013). As their effect was virtually identical in all conditions, they are only reported
in Supplemental Materials (Pinheiro-Chagas, Dotan, Piazza, & Dehaene, 2017).
At a second stage, we compared the b values of the different predictors (paired t test or
repeated-measures ANOVA). To examine whether a given predictor has a significant group-
level effect in each time point, we compared the participant’s b values to zero using one-sample
t test. Each b value (also called regression weight) provides a quantitative measure of the extent
to which each element of the operation influences the finger trajectory at each time point. El
reported p values are one-tailed, since we assumed that all effects of all predictors included
would be positive.
RESULTADOS
Movement Time
We first analyzed the overall movement time (MONTE) from stimulus presentation to number-line
tocar (equivalent to RT in oral calculation tasks). In both blocks, MT was longer for subtractions
compared to additions L + S (block 1: additions L + S: m = 966 EM, DE = 118 EM; subtrac-
ciones: m = 1040 EM, DE = 138 EM; t(29) = −14.72; pag < .001; d = −.58. Block 2: additions:
mean = 982 ms, SD = 127 ms; subtractions: M = 1072 ms, SD = 156 ms; t(29) = −12.55;
p < .001; d = −.64). In agreement with the COMP model of Butterworth et al.
(2001), ad-
ditions L + S were solved 14 ms faster than additions S + L (block 2: additions L + S: M =
976 ms, SD = 128 ms; additions S + L: mean = 990 ms, SD = 125 ms; t(29) = −5.23;
p < .001; d = −.11). To investigate the problem-size effect, we performed a stepwise mul-
tiple regression with MT as the dependent variable and the Min operand, Max operand, and
Result as predictors, separately for additions and subtractions in each experimental block. The
best predictor of MT was always the Min operand (insets in Figure 2). For additions, the Min
operand had a b value of 26 ms per unit (p < .001) in block 1 and of 21 ms per unit (p < .001)
in block 2. The Max operand had a small but significant negative effect in block 1 (b = 8 ms;
p < .001), but a null effect in block 2 (p = .895). Finally, the Result had a null effect in both
blocks (p > .8). For subtractions, the Min operand had a b value of 62 ms per unit (pag < .001)
in block 1 and of 44 ms per unit (p < .001) in block 2. The Max operand had a null effect
(p > 1) in both blocks. Finalmente, the Result had a small but significant effect in both blocks
(b = −4.53 ms; pag = .002 in block 1 and b = −7.07 ms; pag < .001 in block 2). Overall, the
dominant effect of the Min operand is therefore consistent across both blocks and operations.
Figure 2. Reconstructed trajectories per result and per operation, averaged across subjects.
Insets show how movement time increases as a function of the Min operand.
OPEN MIND: Discoveries in Cognitive Science
33
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
Accuracy
Next we analyzed response accuracy, that is, the location where participants landed their
finger (endpoint) in relation to the ideal location. The endpoint error is the absolute differ-
ence between the endpoint and the correct result (in numerical units). The endpoint bias is
the difference between the endpoint and the correct result, with positive values indicating
rightward bias. Subtractions produced larger endpoint errors, but this difference only reached
statistical significance in block 1 (block 1: additions L+S: M = .43, SD = .13; subtractions:
M = .47, SD = .180; t(29) = −2.47; p = .019; d = −.26; block 2: additions L+S: M = .47,
SD = .150; subtractions: M = .48, SD = .180; t(29) = −0.20; p = .841, d = −.06). Addi-
tions L+S produced a slight greater leftward endpoint bias compared to subtractions, but this
difference was only significant in block 2 (block 1: additions L + S: M = −.12, SD = .16;
subtractions: M = −.06, SD = .170; t(29) = −1.24; p = .226; d = −.37; block 2: additions
L+S: M = −.22, SD = .14; subtractions: M = .02, SD = .160; t(29) = −6.92; p < .001,
d = −1.63). Note that this effect is the opposite of the OM effect. A comprehensive analysis
of the full time course of the OM effect is presented further below (see Figure 5), but here
we simply note that in block 2 subtractions had overall larger first operands as compared to
additions (in order to yield matched results), which may have dragged responses further to
the right for subtractions. No significant differences in endpoint error or endpoint bias were
found between additions L + S and additions S + L (block 2, endpoint error: additions L + S:
M = .45, SD = .15; additions S+L: M = .47, SD = .160; t(29) = −1.95, p < .061; d = −.13;
endpoint bias: additions L+S: M = −.19, SD = .13; additions S+L: M = −.20, SD = .16;
t(29) = 0.59; p = .553; d = .07). Finally, with respect to the problem-size effect, the Min
operand, which was the best predictor of movement times, was also a significant predictor of
endpoint error in both additions and subtractions in both blocks (block 1: additions: b = .41,
p < .001, subtractions: b = .43, p < .001; block 2: additions: b = .44, p < .001, subtractions:
b = .42, p < .001).
Trajectory Dynamics
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Next, we analyzed the full trajectories (Figure 3). The first question we considered was whether
the operands are processed in parallel or serially. In additions with larger operand first (L + S),
regressions indicated that the finger first moved according to the first operand, and only then a
significant effect of the second operand emerged. The first operand has a significantly higher
effect compared to the second starting at 420 ms (p = .017) in block 1 and at 450 ms in
block 2 (p < .001) and this difference remained significant until the end of the trajectory.
Remarkably, for additions with smaller operand first (S + L), the order was reversed: the
second operand (larger number) had a higher weight compared to the first operand (smaller
number) from 390 ms on (p < .047), compatible with the assumption that the two operands
are reordered prior to effecting the finger movement. This pattern remained stable until
the end of the trajectory. In fact, the second operand (larger) deviated from zero at 390 ms
(b[signed_op2] > .057, R2 > 0.036, t(29) > 2.998, pag < .034), 120 ms before the effect of the
first operand at 510 ms (b[op1] > .064, R2 > .020, t(29) > 1.943, pag < .030).
In subtractions, a serial effect was again observed. The first operand had a higher
effect than the second during almost the entire trajectory (from 390 ms on in block 1 and
from 360 ms on in block 2). Both operations therefore indicate that the operands are pro-
cessed serially: participants start processing the larger operand followed by the smaller, regard-
less of the order in which they appeared. Additional analyses (see Supplemental Materials,
OPEN MIND: Discoveries in Cognitive Science
34
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Figure 3. Time course of the regression effects in block 1 (a, b) and in block 2 (c, d, e) per
condition. The b values were averaged over participants and plotted as a function of time. The
b values were compared to zero (t test), black dots denote p < .05. Error bars represent 95%
confidence intervals.
Pinheiro-Chagas et al., 2017) revealed that the finger first moved according to the larger
operand L at the same time in all arithmetic operations, and then a correction was introduced
for the smaller operand S at different delays (L + S < S + L < L − S).
To directly visualize this serial processing pattern, we returned to the individual trajecto-
ries for specific problems. Figure 4a shows the example of subtraction problems “9 - S” (where
S ranges from 0 to 8), in which we could investigate the full spectrum of results 1–9. The plot
shows that the finger first deviates toward the right (i.e., in the direction of the larger operand
9) and then to the correct result. Additionally, the latter correction seems to be progressive, as
if the finger goes through intermediate stages. This is most clearly seen for the problem 9 - 8:
the trajectory first coincides with that for 9 - 0, then 9 - 1, 9 - 2, and so on.
OPEN MIND: Discoveries in Cognitive Science
35
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Figure 4. Reconstructed trajectories per operation averaged across subjects (a, c, e) and subtracted
consecutive trajectories (b, d, f). Insets show the divergence time as a function of the number being
subtracted, for different ΔX coordinate threshold.
To characterize these effects, we subtracted consecutive trajectories (9 - 7) - (9 - 8), (9 - 6) -
(9 - 7), and so on, and plotted the resulting difference trajectories, thus revealing the time course
of their divergence (Figure 4b). We then examined whether the divergence times increased
with the Min Operand. Divergence time was measured as the moment when the difference in
finger location (Δx) achieved a threshold value. Regardless of the particular choice of thresh-
old, regressions indicated that the divergence time increased with the number being subtracted
(p < .001) with a stable slope of about 60 ms per additional unit. Similar results were found
for subtractions 8-S, 7-S, 6-S, and so on (slopes = 75, 66, 68 ms, respectively). In order to test
whether the same pattern was also present in additions, we selected the problems 5 + S and
5 − S, since they have the larger range of S that can be matched between operations. Both
subtractions 5 − S (mean slope = 55 ms) and additions 5 + S (mean slopes = 52 ms) showed
a progressive deviation from the larger operand, proportional to the size of the Min Operand.
OPEN MIND: Discoveries in Cognitive Science
36
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
Figure 5. Time course of the regression effects in block 1 (a) and in block 2 with the selected trials
of block 1 (b) for additions and subtractions together.
To investigate the OM effect, we pooled additions and subtractions (Figure 5). In block 1,
the operator had a significant transient effect from 480 ms to 810 ms (b[operator] > .048,
R2 > .005, t(29) > 2.335, pag < .013), reaching its peak value at 630 ms (b[operator] = .170,
R2 = .014, t(29) = 8.673, p < .001). The effect was positive, indicating that an addition sign
transiently distorts the trajectory toward the right side of the number line, and a subtraction
sign toward the left side, as expected from the OM effect. The results remained essentially
unchanged when we analyzed the same arithmetic problems as in block 1, but now using
the data from block 2, that is, with an unbiased distribution of result size for addition and
subtraction. There was a significant positive transient effect of the operator (210 ms) from
(b[operator] > .069, R2 > .009, t(29) > 1.913, pag = .033), reaching its
510 ms to 720 EM.
peak b value at 600 EM (b[operator] = .096, R2 = .008, t(29) = 3.973, pag < .001) and then
progressively dropping until losing statistical significance at 750 ms.
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
An interesting visualization of the OM effect is provided by addition and subtraction
problems involving zero (x + 0 and x − 0 problems), which can be directly compared with
single-digit controls where X alone is presented. As shown in Figure S3 (Pinheiro-Chagas
et al., 2017), when plotting the implied endpoints per time, subtractions show a systematic
transient leftward bias, additions a rightward bias, and single digits tend to fall in between.
This variant of the OM effect was statistically confirmed in a regression analysis comparing
additions x + 0 and subtractions x − 0 (insets in Figure S3 (Pinheiro-Chagas et al., 2017)).
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Interestingly, at about the same time as the effect of the operator, we observed a signifi-
cant transient effect of the absolute value of the second operand in both blocks (block 1: from
330 ms on b[abs_2op] = .017, R2 = .003, t(29) = 1.994, p < .028, peak b at 510 ms; block 2
from 510 ms to 960 ms, b[abs_2op] > .078, R2 > .012, t(29) > 3.233, pag < .001; peak b at
690 ms). The effects of b[op1] and b[signed_op2] were highly similar to the ones found in
previous regression analyses, with the effect of b[signed_op2] delayed as compared to b[op1].
In brief, the temporal dynamics of the OM effect revealed a spatial bias induced by the
operator coinciding with the time that participants are processing the second operand. Further-
more, participants also seemed to transiently represent the absolute value of the subtrahend.
OPEN MIND: Discoveries in Cognitive Science
37
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
DISCUSSION
By continuously measuring finger position in a novel calculate-and-point task, we obtained a
Importantly,
detailed picture of the processing stages underlying addition and subtraction.
overall movement time and accuracy replicated previous findings in cognitive arithmetic,
indicating that our task did not depart radically from previous oral calculation tasks. Sub-
tractions were solved slower and less accurately than additions, and within additions, those
with a smaller first operand were solved 14 ms slower than those with a larger first operand
(Butterworth et al., 2001, reported a 13 ms difference). This effect is not trivial, since it
runs opposite to what could have been predicted in the light of spatial-numerical congruency
effects, such as the SNARC (Dehaene, Bossini, & Giraux, 1993)—that is, that additions S + L,
where digits are in the proper left-to-right order, would be solved faster than L + S.
In line with Barrouillet and Thevenot (2013) and Uittenhove et al.
(2016), we also
detected a robust problem-size effect in single-digit additions, even with operands in the range
0–4. The best predictor of movement time was the smaller operand (Min) in both additions
and subtractions. The slopes that we found for additions (21 ms per unit in block 1 and
26 ms in block 2) match the slope of 26 ms reported by Barrouillet and Thevenot (2013).
Finally, subtractions showed a higher problem size effect than additions, also consistent with
previous studies (Seyler, Kirk, & Ashcraft, 2003).
The main contribution of our study is to investigate the covert processing stages under-
lying arithmetic calculations. Regression analyses on instantaneous finger direction allowed
us to uncover two new effects. First, in both additions and subtractions, the operands were
processed in a serial way: participants started processing the larger operand, then the smaller
operand, regardless of the order in which they were presented. In particular, we could directly
observe a reordering in S + L additions, as predicted by the COMP model (Butterworth et al.,
2001). The presence of an additional reordering stage may also explain why S + L additions
were slower than L + S additions.
Comparing each predictor across additions and subtractions further revealed that the
larger operand is processed at about the same time in all conditions, and then a correc-
tion is introduced for the smaller operand, at a variable delay (additions L + S < additions
S + L < subtractions). A possible explanation for the higher delay in subtraction, revealed
by regression, is that participants transiently represent the absolute value of the subtrahend.
One additional processing stage may therefore be involved in subtractions: the conversion of
the absolute value of the subtrahend into a negative number. Additionally, within addition
problems, a possible explanation for the difference between L + S and S + L additions is that
the selection of the smaller operand is faster when the numbers are in the appropriate left-to-
right order.
Most importantly, we discovered a possible origin of the Min effect, reflected by a serial
influence of the size of the second operand on finger trajectories. During an operation such
as 9 - 8, the finger did not instantaneously point to the result 1, but slowly veered toward it,
successively pointing to each intermediate location 9 - 0, 9 - 1, 9 - 2, and so on. Analysis
confirmed that, when the second operand increased, the trajectories diverged at increasingly
later times: 9 - 8 diverged from 9 - 7 at a later time than 9 - 7 diverged from 9 - 6, and so on.
Thus, calculation always starts with the larger operand and then the correct result is attained
in a slow and incremental manner, proportional to the Min operand, with a slope of about
50–60 ms per unit.
OPEN MIND: Discoveries in Cognitive Science
38
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
The two serial stages that we identified—first point to the larger operand, then serially
deviate toward the target number—strongly constrain the models of mental arithmetic that we
outlined in the introduction. Our results are most compatible with models where arithmetic
operations are solved by serial quantity manipulation, possibly relying on a mental number
line (Dehaene et al., 1999). They fit precisely with the original Groen and Parkman (1972) Min
Model of addition, in which subjects start with the larger number and count up the smaller
quantity. The slope of that serial process was thought to be too fast for counting, but the
present results do suggest a slow and serial incremental process.
Its fast speed suggests a
series of “jumps” on the mental number line, where quantities are chained by successor and
predecessor operations, rather than an explicit verbal counting process.
Our finger-tracking task could have biased subjects to use approximation or other quantity-
based manipulations. This possibility is unlikely, however, given that our movement-time
results converge with classical studies of the problem-size effect (Groen & Parkman, 1972) and
two of its most recent reexaminations. Using a classical calculation task with oral responses,
(2016) presented robust evidence that
Barrouillet and Thevenot (2013) and Uittenhove et al.
additions are solved using “fast procedures that scroll an ordered representation such as a num-
ber line.” The convergence of results suggest that the proposed model may not be restricted
to the current experimental design, but may be generalized to other methods of measuring
mental calculation. While previous studies could infer this underlying mechanism only in-
directly, from the patterns of RT, our method provides a more direct look at the underlying
processing stages.
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
the results, namely,
The hypothesis of a mental displacement in quantity space also accounts for another as-
pect of
the operational momentum (OM). As previously reported
(McCrink et al., 2007), additions produced a transient bias toward larger numbers and subtrac-
tions toward smaller numbers. Our results reject the hypothesis that OM originates solely from
post-calculation processes, when the retrieved result attracts attention left or right. Rather, the
present work shows that the OM effect coincides with the processing of the second operand,
suggesting that movement on the number line occurs during the single-digit calculation pro-
cess. This result concurs with the only study that actually investigated the timing of the OM
effect during calculation, using mouse tracking rather than finger tracking (Marghetis, Núñez, &
Bergen, 2014). It also fits with other studies suggesting that the visual-spatial attention system
is actively involved in arithmetic calculations (Knops, Thirion, et al., 2009; Knops, Viarouge,
et al., 2009; Mathieu et al., 2016; McCrink et al., 2007).
Our results impose strong restrictions on retrieval-based models of single-digit additions
(Ashcraft, 1992; Campbell, 1987; Siegler, 1987; Zbrodoff & Logan, 2005). These models
postulate a direct access to a memory for arithmetic facts and therefore have no reason to
postulate either a faster influence of the larger number, a linear effect of the smaller number,
or an OM effect.
Indeed, existing models of arithmetical retrieval typically assume that (1)
the ease of retrieval depends on the frequency with which a problem has been encountered
(Hamann & Ashcraft, 1986); (2) after a competition process, only a single result is ultimately
selected (Ashcraft, 1992; Campbell, 1987; Siegler, 1987). Both properties fail to explain
why the finger first points to the larger number, and then slowly veers toward the correct result,
in proportion to the min.
In order to preserve the memory retrieval model, one would have to propose a table-
search model (Ashcraft & Battaglia, 1978) according to which calculation time reflects a search
for the proper entry in a stored table of arithmetic facts. For instance, for a problem like 6 + 3,
OPEN MIND: Discoveries in Cognitive Science
39
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
subjects would first identify the larger number (6), then list all 6 + x problems, and finally
search serially among them (6 + 1 = 7, 6 + 2 = 8, 6 + 3 = 9). Such a model is functionally
equivalent to the above proposal, the only difference being that the movement occurs on a
memorized table rather than a number line.
Overall, our findings highlight how a precisely timed series of operations underlies sim-
ple arithmetic. They also demonstrate that even complex mental operations can be contin-
uously reflected in finger-pointing movements, as previously demonstrated in simpler cases
(Song & Nakayama, 2009). Within the existing methods for investigating covert serial pro-
cesses (King & Dehaene, 2014; Resulaj, Kiani, Wolpert, & Shadlen, 2009; Sternberg, 1969;
Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995), finger tracking may play a special
role as a simple and powerful behavioral method.
ACKNOWLEDGMENTS
This research was sponsored by INSERM, CEA, and the Bettencourt-Schueller Foundation. DD
is grateful to the Azrieli Foundation for the award of an Azrieli Fellowship. PPC gratefully
acknowledges a Science Without Borders Fellowship from CNPq (no. 246750/2012-0).
AUTHOR CONTRIBUTIONS
All authors conceived and designed the study. PPC collect the data. PPC analyzed the data
with the support of DD. All authors helped interpret the data. PPC, SD, and DD wrote the
manuscript.
REFERENCES
Ashcraft, M. H. (1992). Cognitive arithmetic: A review of data and
theory. Cognition, 44(1-2), 75–106.
Ashcraft, M. H., & Battaglia,
(1978). Cognitive arithmetic:
J.
Evidence for retrieval and decision processes in mental addi-
tion. Journal of Experimental Psychology: Human Learning &
Memory, 4(5), 527–538. doi.org/10.1037/0278-7393.4.5.527
Barrouillet, P., & Thevenot, C. (2013). On the problem-size effect in
small additions: Can we really discard any counting-based ac-
count?. Cognition, 128(1), 35–44. doi.org/10.1016/j.cognition.
2013.02.018
Butterworth, B., Zorzi, M., Girelli, L., & Jonckheere, A. R. (2001).
Storage and retrieval of addition facts: The role of number com-
parison. The Quarterly Journal of Experimental Psychology A,
Human Experimental Psychology, 54(4), 1005–1029. doi.org/10.
1080/713756007
Campbell, J. I. (1987). Network interference and mental multipli-
cation. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 13(1), 109–123. doi.org/10.1037/0278-7393.13.
1.109
Dehaene, S., Bossini, S., & Giraux, P. (1993). The mental represen-
tation of parity and number magnitude. Journal of Experimental
Psychology: General, 122(3), 371.
Dehaene, S., & Changeux, J. P. (1993). Development of elementary
numerical abilities: A neuronal model. Journal of Cognitive Neu-
roscience, 5(4), 390–407. doi.org/10.1162/jocn.1993.5.4.390
Dehaene, S., Spelke, E., Pinel, P., Stanescu, R., & Tsivkin, S. (1999).
Sources of mathematical thinking: Behavioral and brain-imaging
evidence. Science, 284(5416), 970–974. Retrieved from http://
www.ncbi.nlm.nih.gov/pubmed/10320379
Dotan, D., & Dehaene, S. (2013). How do we convert a number
into a finger trajectory? Cognition, 129(3), 512–529. doi.org/10.
1016/j.cognition.2013.07.007
Dotan, D., & Dehaene, S. (2016). On the origins of logarithmic
number-to-position mapping. Psychological Review, 123(6),
637–666. doi.org/10.1037/rev0000038
Fayol, M., & Thevenot, C. (2012). The use of procedural knowledge
in simple addition and subtraction problems. Cognition, 123(3),
392–403. doi.org/10.1016/j.cognition.2012.02.008
Freeman, J. B., & Ambady, N. (2010). MouseTracker: Software for
studying real-time mental processing using a computer mouse-
tracking method. Behavior Research Methods, 42(1), 226–241.
doi.org/10.3758/BRM.42.1.226
Groen, G. J., & Parkman, J. M. (1972). A chronometric analysis of
simple addition. Psychological Review, 79(4), 329–343. doi.org/
10.1037/h0032950
Hamann, M. S., & Ashcraft, M. H. (1986). Textbook presentations
the basic addition facts. Cognition and Instruction, 3(3),
of
173–192.
King, J.-R., & Dehaene, S. (2014). Characterizing the dynamics of
mental representations: The temporal generalization method.
Trends in Cognitive Sciences, 18(4), 203–210. doi.org/10.1016/
j.tics.2014.01.002
Knops, A., Thirion, B., Hubbard, E. M., Michel, V., & Dehaene, S.
(2009). Recruitment of an area involved in eye movements during
OPEN MIND: Discoveries in Cognitive Science
40
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
Covert Stages of Mental Arithmetic
Pinheiro-Chagas, Dotan, Piazza, Dehaene
mental arithmetic. Science, 324(5934), 1583–1585. doi.org/10.
1126/science.1171599
Knops, A., Viarouge, A., & Dehaene, S. (2009). Dynamic repre-
sentations underlying symbolic and nonsymbolic calculation:
Evidence from the operational momentum effect. Attention
Perception Psychophysics, 71(4), 803–821. Retrieved from http://
app.psychonomic-journals.org/content/71/4/803.full.pdf
Lefevre, J. A., & Kulak, A. G. (1994). Individual differences in the
obligatory activation of addition facts. Memory & Cognition,
22(2), 188–200.
Marghetis, T., Núñez, R., & Bergen, B. K. (2014). Doing arithmetic
by hand: Hand movements during exact arithmetic reveal sys-
tematic, dynamic spatial processing. The Quarterly Journal of
Experimental Psychology, 67(8), 1579–1596. doi.org/10.1080/
17470218.2014.897359
Mathieu, R., Gourjon, A., Couderc, A., Thevenot, C., & Prado,
J. (2016). Running the number line: Rapid shifts of attention
in single-digit arithmetic. Cognition, 146, 229–239. doi.org/10.
1016/j.cognition.2015.10.002
McCrink, K., Dehaene, S., & Dehaene-Lambertz, G. (2007). Moving
along the number line: Operational momentum in nonsymbolic
arithmetic. Perception & Psychophysics, 69(8), 1324–1333. Re-
trieved from http://www.ncbi.nlm.nih.gov/pubmed/18078224
Pinhas, M., Shaki, S., & Fischer, M. H. (2014). Heed the signs:
Operation signs have spatial associations. Quarterly Journal of
Experimental Psychology, 67(8), 1527–1540. doi.org/10.1080/
17470218.2014.892516
Pinheiro-Chagas, P., Dotan, D., Piazza, M., & Dehaene, S. (2017).
Supplementary material for “Finger tracking reveals the covert
stages of mental arithmetic.” Open Mind: Discoveries in Cogni-
tive Science 1(1). doi.org:10.1162/opmi_a_00003
Restle, F. (1970). Speed of adding and comparing numbers. Journal
of Experimental Psychology, 83(2, Pt. 1), 274–278.
Resulaj, A., Kiani, R., Wolpert, D. M., & Shadlen, M. N. (2009).
Changes of mind in decision-making. Nature, 461(7261),
263–266. doi.org/10.1038/nature08275
Seyler, D. J., Kirk, E. P., & Ashcraft, M. H. (2003). Elementary sub-
traction. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 29(6), 1339–1352. doi.org/10.1037/0278-7393.
29.6.1339
Siegler, R. S. (1987). The perils of averaging data over strategies:
An example from children’s addition. Journal of Experimental
Psychology: General, 116(3), 250.
Song, J.-H., & Nakayama, K. (2009). Hidden cognitive states re-
vealed in choice reaching tasks. Trends in Cognitive Sciences,
13(8), 360–366. doi.org/10.1016/j.tics.2009.04.009
Sternberg, S. (1969). The discovery of processing stages: Exten-
sions of Donders’ method. Acta Psychologica, 30, 276–315.
Retrieved from http://www.sciencedirect.com/science/article/
pii/0001691869900559
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., &
Sedivy,
Integration of visual and linguistic
information in spoken language comprehension. Science,
268(5217), 1632–1634.
(1995).
J. C.
Uittenhove, K., Thevenot, C., & Barrouillet, P. (2016). Fast auto-
mated counting procedures in addition problem solving: When
are they used and why are they mistaken for retrieval? Cognition,
146, 289–303. doi.org/10.1016/j.cognition.2015.10.008
Zbrodoff, J. N., & Logan, G. D. (2005). What everyone finds: The
problem size effect. In J. I. D. Campbell (Ed.), The handbook
of mathematical cognition (pp. 331–345). New York, NY:
Psychology Press.
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
/
/
/
/
/
1
1
3
0
1
8
6
8
2
2
5
o
p
m
_
a
_
0
0
0
0
3
p
d
.
i
f
b
y
g
u
e
s
t
t
o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3
OPEN MIND: Discoveries in Cognitive Science
41
