Você está na página 1de 11

Cognitive Brain Research 20 (2004) 183 – 193

www.elsevier.com/locate/cogbrainres

Research report
Facts, rules, and strategies in single-digit multiplication:
evidence from event-related brain potentials
Kerstin Jost *, Ulrike Beinhoff, Erwin Hennighausen, Frank Rösler
Experimental and Biological Psychology, Department of Psychology, Philipps-University of Marburg, Gutenbergstrasse 18, Marburg D-35032, Germany
Accepted 27 February 2004

Available online 14 April 2004

Abstract

It has been hypothesized that zero vs. nonzero operands in single-digit multiplication problems invoke distinct solution strategies. We
studied such problems in an implicit production task with event-related brain potentials (ERPs) recorded from 61 scalp positions in 18
participants. The topography of a slow negative wave, which accompanied the implicit production of the multiplication result, varied with
problem type. In comparison to small problems, larger problems evoked a stronger negativity over fronto-central and right temporal sites, and
zero problems evoked a left anterior negativity. These topographic differences indicate not only that zero and small nonzero problems are
solved by means of distinct strategies—most likely rule application vs. fact retrieval—but also that larger, less practiced problems invoke
other processes than pure fact retrieval. Moreover, ERPs showed a positive deflection around 450 ms with a centro-parietal topography
(P300), whose amplitude reflected differences in anticipated problem difficulty.
D 2004 Elsevier B.V. All rights reserved.

Theme: Neural basis of behavior


Topic: Cognition

Keywords: Mental calculation; Multiplication facts; Problem size effect; Rule application; Slow waves; P300

1. Introduction operand, which can be solved immediately irrespective of


its size (e.g., 758  0). Further evidence for a distinction
It is widely accepted that single-digit multiplication between ‘‘rule-based’’ and ‘‘retrieval-based’’ multiplications
problems are stored as facts in long-term memory, repre- has been provided by double dissociations observed in
senting the associative structure of operands, problems, and brain-damaged patients with calculation impairments
solutions (for review, see Refs. [3,4]). LeFevre et al. [36,44]. Sokol et al. described a patient whose performance
showed that fact retrieval can account for 80% of the on fact-retrieval problems was not uniform. Error rates and
solutions of simple addition and multiplication problems, the time course of improvement varied for problems with-
while other strategies (e.g., decomposition into simpler out zero. In contrast, all multiplication problems with zero
memorized facts) are applied only if immediate retrieval were equally impaired at the beginning but improved
fails [28,29]. However, there are some other exceptions. suddenly at the same moment and to the same extent. This
Multiplications involving zero must not necessarily be suggests that an impairment of a general rule affected the
solved by fact retrieval because the corresponding result performance on all problems with zero.
can be generated by applying a rule (i.e., n  0 = n). This is Although there is little doubt that some multiplication
easily demonstrated by any multiplication with a zero problems can be solved by applying the zero rule, the
empirical evidence for how multiplications with a zero
operand are actually solved is sparse and somewhat contra-
* Corresponding author. Tel.: +49-6421-2822160; fax: +49-6421- dictory. Stazyk et al. [46] (see also Ref. [35]), for example,
2828948.
E-mail addresses: jost@staff.uni-marburg.de (K. Jost),
found that verification times of n  0 problems were un-
roesler@staff.uni-marburg.de (F. Rösler). usually long and error-prone compared to problems without
URL: http://www.staff.uni-marburg.de/~cablab/. zero, although the zero problems were uniformly rated as

0926-6410/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cogbrainres.2004.02.005
184 K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193

less difficult than nonzero problems. Thus, the authors (e.g., Ref. [7]; for review, see Ref. [3]) that attribute the
concluded that zero problems are solved by applying the problem size effect to differences in the accessibility of
appropriate rule, rather than by directly retrieving the result. results (e.g., due to differences in the frequency of
Rule knowledge might be less well established and therefore problem usage) [1,2].
its activation might take longer than fact retrieval. However, All in all, there is good reason to assume that single-
in production tasks, response latencies are usually found to digit multiplication problems are not only solved by pure
be faster for zero than for nonzero problems [22,28,32]. This fact retrieval, but that other strategies are used as well. To
discrepancy may be due to processing differences between date, however, the empirical evidence of which problem
production and verification. While production problems is solved with which strategy is inconclusive. As outlined,
require full retrieval of the result, verification problems findings with response times are inconsistent and, to some
can be solved without producing the solution [49]. Instead, degree, task-dependent. Self-reports, on the other hand,
some sort of familiarity or plausibility judgment seems often are very suggestive, but their validity is questionable.
used to decide whether the equation is valid or not (see also Kirk and Ashcraft [22], for example, demonstrated that
Ref. [9]). It could be that these processes caused the strategy reports for simple addition and multiplication
disadvantage of the n  0 problems in the Stazyk et al. problems are partially dependent on instructions. More-
study. However, even if only production tasks are consid- over, self-reports can change the behavior of some indi-
ered, response times (RTs) are inconclusive with respect to viduals (see Ref. [43]). Brain imaging studies (e.g., Refs.
the solution strategy invoked by zero problems. LeFevre et [10,25]) also addressed the question of whether arithmetic
al. [28], for example, collected introspective reports in a problems invoke distinct solution processes, but so far,
production task and found that rule application and fact single-digit multiplication problems involving zero and
retrieval both resulted in fast and correct responses to zero nonzero, or small and large operands, were not analyzed
problems. systematically with this approach. To get more insight
However, it is not only claimed that distinct solution into whether multiplications with small and large or with
strategies are possibly used for zero vs. nonzero prob- zero operands invoke different solution strategies, we
lems, or for zero problems if they are embedded in recorded event-related brain potentials (ERPs). ERPs are
different tasks. Recent studies provided evidence that on-line measures of brain activity, which reflect timing,
single-digit multiplications without zero are also not extent, and quality of processing mechanisms before an
necessarily always solved by one and the same strategy overt response is observed (e.g., see Ref. [40]). Previous
(Refs. [28,43]; for a comparable finding with addition studies demonstrated that ERPs could be successfully
problems, see Ref. [29]). In these studies, participants used to delineate the associative network structure of
solved simple arithmetic problems in a reaction time task arithmetic knowledge [20,33,34], distinct processes during
and gave immediate verbal reports on the type of strategy mental calculation [21], or strategy variations in verifying
they had used for producing the result. In contrast to complex inequalities [14].
existing theories, which state that direct memory retrieval In the present study, the production of a multiplication
is the one and only procedure to access the result of result was embedded in a chain calculation problem—a
single-digit problems (e.g., Ref. [7]; for review, see Refs. sequence of simple calculations had to be performed and a
[3,4]), participants reported substantial use of nonretrieval final result, presented at the very end of the sequence, had to
strategies, such as repeated addition, number series gen- be verified (e.g., 3  5 + 2 = 17). The advantage of such a
eration, and retrieval of derived facts. For example, the procedure is twofold. First, participants must implicitly
problem 8  7 can be solved by retrieving 7  7 and then produce the interim result of the multiplication problem
adding seven to the intermediate result, ‘‘49’’. Typically, before they can go on with the next calculation step. This
these backup strategies involve multiple steps, including prevents, as can happen in one-step verification problems
retrieval of well-known facts and applications of compu- (e.g., 3  5 = 16), that a decision is based on some sort of
tational principles, presumably the reason why most of plausibility judgment without invoking production of the
them are more time-consuming than simple, immediate result (see Refs. [9,49]). Second, the multiplication result
fact retrieval [28]. has to be produced implicitly (i.e., without any overt
LeFevre et al. found that procedures other than direct response) and this prevents movement artifacts in the
retrieval are more often used if larger operands (e.g., electroencephalogram (EEG) recording, as they are most
78), compared to smaller operands (e.g., 23), are likely in one-step production tasks with an immediate verbal
involved ([28,43]; for comparable findings with addition response.
and division problems, see Refs. [27,29]). Furthermore, Three types of multiplications—problems with small or
the authors could show that the well-established problem- large nonzero operands and problems with one zero oper-
size effect (i.e., an increase of RT with increasing size of and—were compared with a control condition not involving
the operands; e.g., see Refs. [8,32,46]) can be reduced any type of mental calculation proper. In this condition, the
dramatically if problems that invoke some nonretrieval operands of the first problem had to be stored only and one
strategy are eliminated. This finding contradicts theories of the operands had to be used for another addition or
K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193 185

subtraction.1 This provides an ‘‘active’’ baseline because the Distinct solution strategies—rule application for zero,
operands must be perceived and stored in each multiplica- fact retrieval for nonzero small problems, and backup
tion condition as well. Thus, computing the differences strategies for nonzero large problems, respectively—should
between multiplication conditions and the storage condition become apparent in distinct topographies of the negative
should reveal the net effects due to different solution slow waves accompanying production proper. We expected
strategies. one topographic difference between rule-based and retriev-
We analyzed two ERP phenomena: the P300 component al-based problems and another one between retrieval-based
and a subsequent slow negative wave evoked in the pro- problems involving small and large operands.
duction phase. P300 is an endogenous positive component We ran two studies: a behavioral study (Experiment I)
occurring approximately 300 – 600 ms after stimulus onset and an ERP study (Experiment II). In the behavioral study,
that reflects higher-level cognitive processes such as event the timing of the event sequences was partially controlled by
categorization [13,19]. Among others, the amplitude of the participants: they saw the first part of the problem
P300 reflects variations in anticipated task difficulty: stimuli sequence (e.g., 34) and indicated with a finger movement
which notify a subjectively more demanding task elicit when they had implicitly generated the solution to the
larger amplitudes [5,48]. Slow negative waves accompany multiplication problem and felt ready for the next operation
almost any cognitive processing with two interesting fea- (e.g., + 2). This allowed for an estimation of the implicit
tures: their topographic maximum is determined by the production times of the different problem types. We
modality and the type of task; for instance, linguistic tasks expected that responses are shorter for zero than for nonzero
evoke a left anterior maximum and spatial tasks elicit a problems, and that they show the typical problem-size effect
parietal maximum, and their amplitude reflects processing for nonzero problems (i.e., longer RTs for larger problems
effort, that is, larger negativities are evoked by more than for smaller problems).
difficult tasks) [38]. There is also strong evidence that slow The estimate of the longest implicit production time was
negative waves reflect a relative increase of excitatory then used in the ERP study to specify a constant interval
postsynaptic potentials within cortical cell assemblies [6]. between the multiplication and the next operation, which
Thus, slow negative waves indicate to what extent identical was long enough to avoid any interference between the two
or distinct neural cell assemblies are recruited during a operations.
particular processing episode.
Differences between calculation strategies should be-
come manifest in the ERPs as follows: if the three con- 2. Materials and methods
ditions—problems with zero, and small and large problems
without zero—are inherently categorized as varying in 2.1. Participants
subjective difficulty, then an amplitude modulation of the
P300 should be observed. More specifically, all multiplica- Students of the University of Marburg were recruited for
tion conditions should evoke larger amplitude than the the behavioral and the ERP study. All were right-handed,
storage condition and the amplitude should be largest for native speakers of German, had normal or corrected-to-
the nonzero problems with large operands. normal vision, and were naive with respect to the purpose of
the study. They received either course credit points or a
1 monetary compensation.
As shown by Ref. [26], arithmetic facts may be activated
automatically by a set of two digits, even if participants are not explicitly Thirteen students participated in the behavioral study.
asked to perform a calculation. In Ref. [26], subjects were shown pairs of One participant had to be excluded because of too many
numbers (e.g., 3 + 2) that were replaced, after a variable SOA, by a target errors. The final sample comprised data from 12 participants
number. Subjects were required to decide whether the target number was (nine women) with a mean age of 25 years (range 20 – 34
one of the numbers in the initial pair, referred to as number-matching task.
years).
LeFevre and Kulak found that subjects were slower to respond ‘‘no’’ when
the target was the sum of the pair (e.g., 3 + 2 = 5) than when it was unrelated
Twenty-one students took part in the ERP study, none of
to the pair (e.g., 4). Thus, our storage condition could have caused a similar whom had participated in the behavioral study. Three
automatic activation of multiplication results, although the two digits were participants had to be excluded because of too many EEG
not shown together with a multiplication sign in this condition. recording artifacts. The final sample comprised data from
However, even if such an effect was present in our control condition, it nine women and nine men. The mean age was 23 years
was of minor importance because the interference effect observed in Ref.
[26] only occurred at very short SOAs (i.e., 40 and 60 ms), indicating that (range 18– 27 years).
automatically activated results are very short-lived. Since the implicit
production of the multiplication results took much longer than 60 ms in our 2.2. Materials
study (see response times of Experiment I) and the main ERP differences
between multiplication and storage also appeared much later in time (not Operands ranging from 2 to 9 (ties were excluded; e.g.,
earlier than 300 ms after problem presentation), it seems very unlikely that
such an automatic spread of activation effect will have contaminated the
3  3) were combined to form 56 multiplication problems
contrasts of the multiplication problems with the control (storage) without zero. Two levels of problem size were defined, each
condition. containing 28 problems: Problems with both operands V 5
186 K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193

(products ranging from 6 to 20) were classified as small; al operation for 600 ms. In the storage condition, the
problems with both operands >5 (products ranging from 42 position of the third operand indicated which of the two
to 72) were classified as large (for a similar classification, stored digits should be used. Presentation on the left side of
see Refs. [21,49]). Mixed problems with one operand V5 the rectangle indicated to add/subtract on basis of the first,
and one operand >5 (e.g., 38 or 56) were defined as left-sided operand, and presentation on the right side indi-
small if they contained either a 2 or a 5. Thus, all problems cated to use the second, right-sided operand. The additional
with 2 and 5 were defined as small. This classification is operation was followed by a blank screen of 200 ms, and the
based on the observation that in several response time final result was presented until a response was given, or for
studies, problems with 2 were found to be solved faster a maximum of 1100 ms. The intertrial interval varied
than other problems and that problems with 5 show a randomly between 2 and 3 s.
latency advantage relative to other problems of the same Participants were instructed either to multiply or to store
magnitude (e.g., Refs. [8,27]). In a previous study, this the first two operands and to indicate when result production
classification had provided a reliable problem-size effect, was finished. This and the verification response had to be
in both RTs and ERPs (Ref. [20]; for a similar classification, given by a brief upward finger movement, which interrupted
see Ref. [9]).2 a light gate. Allocation of the index fingers to the two
Another 16 problems were constructed from the same response alternatives in the verification task was varied
digits by setting one operand to zero (n  0 or 0  n). All systematically across participants. Delayed and incorrect
operand combinations were also used for the storage con- responses were fed back and excluded from further analysis.
dition. Each problem was repeated seven times in the Stimulus presentation in the ERP study was identical with
multiplication and in the storage condition. In total, there one exception. Participants had a fixed interval to produce the
were 1008 trials. The additional, subsequent operation was result of the multiplication task or to store the two operands:
either an addition or a subtraction by 1, 2, or 3. Incorrect the operation sign was presented for 2800 ms (see Fig. 1),
solutions in the verification task ( p = 0.50) were realized by which is equivalent to the maximum RT observed in the
adding or subtracting 1, 2, or 3 from the correct result. behavioral study. The following additional operation and
the verification task were as described for the behavioral study.
2.3. Design and procedure In both studies, trials were presented in 28 blocks (with
36 trials each), with multiplication and storage trials appear-
Participants sat in a recording chamber in front of a ing equally often within one block. A new randomized trial
computer screen located at eye level at a distance of 70 cm.
Numeric characters (vertical size 0.9 cm) were presented in
black within a white rectangle in the center of the screen on
a light-gray background. In the behavioral study, a trial
started with a rectangle in the center of the screen in which
the first operand appeared on its left side 1000 ms later. The
next frame was presented another 500 ms later, and added
the second operand on the right side of the rectangle and the
operation sign in the center. Storage was indicated by ‘‘!’’
and multiplication by ‘‘  ’’. After another 500 ms, both
digits disappeared and the operation sign remained visible
during the ‘‘production phase.’’ The participants had to
indicate with a finger response when they had generated
the result and felt ready for the next operation. With this
procedure, solution times were measured and an appropriate
interval could be defined for the ERP study. The stimulus
sequence was continued by the presentation of the addition-

2
In other studies, multiplication problems were categorized with Fig. 1. Event sequence of a trial, grand average ERPs, and measurement
respect to the size of the product irrespective of the size of the operands, epochs. A trial started with a rectangle in which the first operand appeared
that is, all products <25 were classified as small, all >25. However, there is on its left side followed by the second operand on the right and the
some evidence that all multiples of 2 and all multiples of 5 show a response operation sign in the center. From this point on, participants had 3300 ms to
time advantage even if the second operand is >5. Therefore, we used the produce the result of the multiplication problem, or to store the two
outlined classification procedure because we think that it separates easy and operands. After the production phase, an additional operation was presented
more difficult problems better. Following the recommendation of one and followed by the final result, which had to be verified. Shading shows
reviewer, we reanalyzed the response times and the ERPs by using the more where the positivity and the slow negative wave were measured (i.e.,
traditional classification, with the cutoff set at 25. Surprisingly, this relative to the presentation of the second operand, between 300 and 600 ms
reclassification provided almost identical results as found with our and between 900 and 3100 ms, respectively). Negativity is up in this figure
classification. and in the following figure.
K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193 187

sequence was created for each participant with the restric- centro-parietal maximum. This was also confirmed by a first
tion that no operand combination was repeated within a inspection of the data.
block. Participants received some practice trials in advance ERP effects were tested with repeated-measures
in order to familiarize themselves with the task. An oblig- ANOVAs with factors measurement epoch (10 and 11
atory break of at least 20 s separated the blocks, which could levels), electrode (6 and 17 levels), and problem type (4
be prolonged by moving both fingers out of the light gate. levels). The four levels of problem type were given by zero
Participants were advised to use the short breaks for problems, small and large problems without zero, and the
recreation. A longer break was inserted after approximately storage condition. Storage with small, large, and zero
half of the blocks. In total, the experiments took about 2.5 operands was combined because a first analysis had not
h (plus about 1.5 h of electrode montage in the ERP study). revealed any differences between them, neither for the P300
nor for the slow negative wave. A superordinate ANOVA
2.4. EEG recording, artifact handling, and signal extraction comprising all three factors revealed a significant triple
interaction, which justified more detailed analyses. To avoid
The EEG was recorded with Ag/AgCl electrodes from 61 an increase in the likelihood of type I errors, all further tests
locations and referenced to the nose tip. Additional electro- were guided by the following hierarchical constraints:
des were attached at the outer canthi of both eyes and the ‘Local’ ANOVAs for one measurement epoch at one elec-
suborbital and supraorbital ridges of the left eye for hori- trode were computed only if the particular electrode had
zontal and vertical EOG monitoring. The left mastoid served shown a significant Epoch  Problem Type interaction and
as ground. Impedances were always kept below 7 kV. the particular measurement epoch had shown a significant
Bandpass of the amplifier system (SYNAMPS; Neuro- Electrode  Problem Type interaction in superordinate
Scan) ranged from DC to 40 Hz and the signals were ANOVAs. Thus, the number of local tests was restricted
digitized with 200 Hz. A DC reset was initiated automati- in advance. Amplitude differences were considered as
cally prior to the beginning of each experimental block. reliable only if at least two consecutive time windows had
Trials containing eye blinks were detected by wavelet signaled significance ( p < 0.05) for the particular effect. F-
analysis and corrected using a linear interpolation algorithm. statistics were corrected according to the formulas of Huynh
Drift artifacts were corrected by a regression method [17]. and Feldt [18]. The uncorrected degrees of freedom, the
ERPs were extracted by averaging trials separately for corrected p value, and the respective epsilon values are
subjects, electrodes, and experimental conditions, and were reported.
based on a minimum of 90 trials each. To compare the topographies of slow negativity,
ANOVAs were also run with z-standardized scores such
2.5. Dependent variables and statistical analysis that mean and variance across electrodes were equalized for
experimental conditions and subjects. The standardization
2.5.1. Behavioral data (Experiment I) permits evaluation of pure topographic differences between
Mean response times were computed for each participant conditions that are not caused by amplitude or variance
and experimental condition. Trials with response times differences (see Refs. [30,47]).
shorter than 200 ms and larger than 10000 ms were
excluded. This procedure eliminated less than 1% of the
trials. 3. Results

2.5.2. EEG data (Experiment II) 3.1. Experiment I: behavioral data


Two intervals were defined to analyze the positive
deflection and the slow negative wave in the production Response times showed a reliable problem-size effect.
phase (see Fig. 1). P300 was measured between 300 and 600 The implicit production took longer for larger (mean 1344
ms by computing the average amplitudes of 10 consecutive ms; S.D. F708 ms) than for smaller problems (841F277
time windows of 30 ms width. The slow negative wave was ms; F(1,11)=11.55, p=0.0059). Moreover, multiplications
measured between 900 and 3100 ms by means of 11 average with zero (690F185 ms) were solved faster than both small
amplitudes, each covering 200 ms. The measures were [ F( 1,11 )= 16. 01 , p =0 .00 21 ] a nd large pro bl e m s
referenced to a baseline of 100 ms preceding the onset of [F(1,11)=14.20, p=0.0032]. The average time for the storage
the first operand (see Fig. 1). condition amounted to 744(F199) ms with no differences
To avoid an inflation of the degrees of freedom in the between zero, small, or large digits.
analysis of interactions with factor electrode, we selected 17
standard electrodes: frontal (FPz, F7, F3, Fz, F4, F8), central 3.2. Experiment II: ERPs
(C3, Cz, C4), temporal (T3, T5, T4, T6), parietal (P3, Pz,
P4), and occipital (Oz). Moreover, the analysis of P300 Grand average ERPs in the production phase (starting
effects was restricted to Pz and surrounding electrodes (P3, with the onset of the second operand and the operation sign)
P4, C3, Cz, C4) because the P300 is known to have a are presented in Fig. 2. As can be seen, both ERP phenom-
188 K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193

Fig. 2. Grand average ERPs for the four problem types in the production phase for a selected set of frontal (F7, Fz, F8), central and parietal (Cz, Pz), and
temporal electrodes (T3, T4, T5, T6). The figure shows the amplitude modulation of the centro-parietal positivity (maximum at Pz) between 300 and 600 ms
and of the subsequent slow negativity. The second operand and the operation sign were presented at time 0 s. ERPs were referenced to a 100-ms-long baseline
(between 600 and 500 ms).

ena of interest, the P300 and the subsequent slow wave, 3.2.2. Slow negative wave
were affected by the experimental conditions. The positivity was followed by a sustained negativity with
an initial phasic maximum at frontal and temporal electrodes
3.2.1. P300 and a subsequent gradual increase at central to occipital sites.
All four conditions evoked a phasic positivity with a Problem type affected the amplitude and topography of this
clear centro-parietal maximum around 450 ms (see Fig. 2). negativity significantly. The negativity was most pronounced
Latency and topography qualify this component as P300 for large problems, least pronounced for storage, and
[19]. The amplitude of the positive peak measured at Pz was intermediate for small and zero problems. This is substan-
systematically affected between 390 and 600 ms by factor tiated by a significant main effect of problem type between
problem type [minF(3,51) = 5.82, p = 0.0076, e = 0.6356; 900 and 2700 ms [minF(3,51) = 3.11, p = 0.0343, e = 1;
maxF(3,51) = 20.25, p < 0.0001, e = 0.7439]. As is obvious maxF(3,51) = 7.64, p = 0.0004, e = 0.9080]. Pairwise post-
from Fig. 2, the amplitude is smallest for storage, interme- hoc tests for the mentioned time epoch revealed that the
diate for multiplication problems with zero, and largest for average amplitudes for large problems differ significantly
multiplication problems without zero. Post-hoc tests from that of small problems (between 1100 and 2700 ms),
revealed significant amplitude differences between storage zero problems (between 1100 and 1500 ms), and storage
and multiplications with zero between 450 and 600 ms [between 900 and 2700 ms; minF(1,17) = 4.91, p = 0.0407;
[minF(1,17) = 4.71, p = 0.0444; maxF(1,17) = 8.78, maxF(1,17) = 12.77, p = 0.0023, for the mentioned con-
p = 0.0087]. The amplitude was also significantly smaller trasts]. The difference between storage and multiplications
for multiplications with zero than for multiplications with- with zero proved to be reliable between 1100 and 2300 ms
out zero between 390 and 570 ms [minF(1,17) = 8.09, [minF(1,17) = 3.21, p = 0.0911; maxF(1,17) = 8.24,
p = 0.0112; maxF(1,17) = 46.90, p < 0.0001]. Amplitude dif- p = 0.0106].
ferences are not restricted to electrode Pz. Surrounding Beside these main effects, factor problem type inter-
electrodes (i.e., P3, P4, Cz, and C4; see Cz in Fig. 2) also acted with factor electrode for the entire measurement
show reliable differences between storage and multiplication epoch [minF(48,816) = 2.87, p = 0.0006, e = 0.2851;
with zero between 450 and 570 ms [minF(1, 17) = 4.53, maxF(48,816) = 4.89, p < 0.0001, e = 0.2730], indicating
p = 0.0483; maxF(1,17) = 8.70, p = 0.0090] and between distinct scalp distributions of the negativity for the various
multiplication with and without zero between 390 and 480 problem types.
[minF(1,17) = 4.51, p = 0.0486; maxF(1,17) = 20.90, To investigate topographical differences, we first com-
p = 0.0003]. Problem size did not affect the amplitude of puted the net effect of each multiplication type by subtract-
the positivity. ing the storage (control) condition. As will be shown, the
K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193 189

three types of multiplication problems evoked the relative note that the ANOVAs were computed for 200 ms time
negativity with distinct maxima. To delineate which electro- windows). Furthermore, ANOVAs showed reliable interac-
des are relevant for the topographic and the amplitude tions of problem type and electrode, both for the contrasts of
differences, we also computed the difference scores between each multiplication type vs. storage and for the contrasts of
the multiplication conditions. Fig. 3 shows the topographies zero vs. small problems and large vs. small problems. These
of these difference waves for two consecutive time windows effects provide the basis for subsequent ‘local’ contrasts and
of 400 ms length between 1100 and 1900 ms. This interval detailed analyses of the topographic differences.
was chosen because differences between the problem types Relative to the storage condition, small multiplications
are maximal and stable throughout this interval (see below; without zero evoked the negativity with an occipital max-

Fig. 3. Topographies. Interpolated maps for two consecutive time windows of 400 ms length between 1100 and 1900 ms where the effect sizes were largest. Net
effects (multiplication minus storage) were computed for the three types of multiplication problems. Dark shading indicates a relatively more negative potential
in comparison to the storage condition. Topographies of the difference waves of zero minus small problems and large minus small problems (problem-size
effect) are presented in the second and fourth rows. Notice the different scales, which were adjusted to compensate for differences in effect size.
190 K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193

imum extending to left temporal and central sites (see Fig. 3, at the frontal (Fz) and right temporal (T6) electrodes—
middle row). The difference was significant (with p < 0.05) electrodes that already showed the maximum of the net
at C3, P3, T3, T5, Oz, and T6 for all 200 ms time epochs effect for large multiplications.
between 1100 and 1900 ms (except T6 where the effect was
significant only until 1500 ms).
Multiplications with zero (upper row) evoked a clear left 4. Discussion
anterior negativity between 1100 and 1500 ms, differing
significantly from storage at left frontal and central electro- The objective of the present study was to determine by
des (F7, F3, C3). A second maximum appeared between Oz means of event-related brain potentials whether different
and T6 (all contrasts p < 0.05). In the following time solution strategies are used to produce the results of single-
window, the relative negativity is again significant at the digit multiplication problems with small, large, or zero
mentioned electrodes, but also at left temporal (T3, T5) and operands.
parietal positions (P3). The experimental conditions affected significantly both
Amplitude differences between large multiplications ERP phenomena of interest, the P300 component peaking
without zero and storage (see lower row) existed at almost around 450 ms after problem presentation and the subse-
all electrodes (F7, F3, Fz, T3, C3, T4, T5, P3, P4, T6, Oz) quent slow negative wave. Amplitude of the P300 was
with p < 0.05 for all measurement epochs between 1100 and smallest for storage, intermediate for zero multiplications,
1900 ms, but clear maxima were present at frontal (Fz), and largest for nonzero multiplications, irrespective of their
occipital (Oz), and temporal (T5, T6) electrodes. size. The peak latency of this effect was much shorter than
To objectify the topographic differences between the the average solution times estimated in the behavioral study
three multiplication conditions, pairwise ANOVAs with z- (the fastest responses were found for zero multiplications
standardized difference scores were run for the consecutive with 690 ms). Therefore, it can be concluded that the P300
time windows of 200 ms between 1100 and 1900 ms. The reflects a processing difference that precedes the actual
topography for large multiplication problems proved to be production of a solution. This finding is compatible with
significantly distinct from those of either small multiplica- the idea that different problems are immediately categorized
tions and multiplications with zero throughout the whole according to their expected difficulty and that these differ-
epoch [comparison large vs. small: minF(16,272) = 2.07, ences are reflected by the P300 amplitude (see also Refs.
p = 0.0304, e = 0.6152; maxF(16,272) = 2.35, p = 0.0180, [5,48]).3 In this sense, the amplitude modulation can be
e = 0.5410; comparison large vs. zero: minF(16,272) = 2.58, interpreted as a measure of cognitive preparation in terms of
p = 0.0220, e = 0.3820; maxF(16,272) = 3.11, p = 0.0069, adjusting attentional and processing resources (see Ref.
e = 0.3902]. Small multiplications differed from zero prob- [48]). However, the observed amplitudes do not fully reflect
lems only marginally between 1100 and 1500 ms but signif- the task difficulty as may be inferred from the response
icantly between 1500 and 1900 ms [minF(16,272) = 1.78, times or the later negativity. RTs differed for small and large
p = 0.0948, e = 0.4464; maxF(16,272) = 2.80, p = 0.0100, nonzero multiplications (841 vs. 1344 ms) while the P300
e = 0.4370]. amplitude was the same for these two conditions.
To delineate which electrodes are responsible for the Moreover, the observation of some kind of precategori-
topographic and amplitude differences, we also computed zation also fits well with the model of arithmetic perfor-
the difference scores between multiplication conditions. The mance in children of Siegler and Shipley [42] (see also Ref.
differences between zero and small multiplications and [41]). This model contains a decision component that selects
between large and small problems are plotted as topographic a solution strategy prior to problem solution. The P300
maps in the second and fourth rows of Fig. 3. effects in the present study suggest that a similar decision
Zero problems evoked a relatively stronger negativity stage is passed in adults as well.
than small multiplications at left anterior to left central sites. The subsequent slow negative wave whose timing coin-
Towards left temporal and occipital sites, a polarity shift cides with the production phase proper was most pro-
occurred where small multiplications were more negative nounced for large multiplications, intermediate for small
than zero problems especially in the time window between and zero problems, and smallest for the storage condition.
1100 and 1500 ms (see also Fig. 2). However, only the This suggests that the different problems recruit neural
differences at left frontal and central electrodes (F3, Fz, C3) resources to a different extent.
proved to be reliable between 1500 and 1900 ms (with
p < 0.05). Thus, the topographic differences are attributable
3
to the left anterior negativity for zero problems that were not It is unlikely that event probability—in general a major factor for
observed for small problems. P300 amplitude variations—was the main determinant for the observed
The slow wave was more negative for large problems amplitude variations: zero multiplications had the smallest probability
( p = 0.111) while small and large multiplications without zero had each a
than for small problems at almost all electrodes ( p < 0.05), larger probability of 0.194, or, if both were combined to one category, of
varying between 0.5 and 2.0 AV (with nearly identical 0.389. Nevertheless, the amplitude of P300 was smallest for the zero
topography in the two time windows), but most pronounced problems with the lowest probability.
K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193 191

The most important finding, however, is that the nega- be concluded that magnitude estimation processes are trig-
tivity revealed clear topographic differences between the gered when the results of larger single-digit multiplication
problem types. These indicate an activation of distinct problems are retrieved.
generator ensembles [47] and, therefore, support the hy- The prominent frontal activation observed in this study
pothesis that single-digit multiplication problems are solved agrees with similar findings of other studies [24,45] and
by means of different solution strategies. indicates that strategies other than pure fact retrieval seem to
be involved when larger operands are multiplied. Gruber et
4.1. Problem size al. [15] reported that more complex problems involving
calculation rules and decomposition strategies showed in-
Large problems in contrast to small problems evoked a creased activity over frontal areas. They related these differ-
larger negative slow wave, suggesting greater mental load ences to linguistic and working memory functions.
for large problems (cf. Refs. [6,38]). This result agrees with Likewise, the frontal negativity in the present study could
the problem-size effect observed in the behavioral study. indicate that larger problems impose more load on working
There the RT difference between small and large problems memory and executive functions if they are decomposed
amounted to 500 ms—an effect that is substantial and larger into a chain of smaller and easily accessible facts [12].
than in other studies (cf. Refs. [8,9]).
More important is the topographic difference between 4.2. Facts vs. rules
small and large problems. Small multiplications (compared
to storage) evoked the negativity with an occipital maxi- In the pilot study, zero problems were solved much faster
mum extending to left temporal sites. In contrast, larger (690 ms) than all nonzero problems (841 and 1344 ms for
problems evoked a more substantial negativity over central- small and large problems), a finding that has also been
frontal areas (Fz) and a larger amplitude over right temporal observed in other studies [22,28,32]. Moreover, the ampli-
areas (T6). Such clearly distinct topographies are difficult to tude of the P300 also suggests that the zero problems are
reconcile with pure differences in processing difficulty. In categorized as potentially less demanding than nonzero
that case, one and the same generator configuration should multiplications.
have been affected and one and the same surface topography The negativity evoked by zero problems shows its most
should have shown an amplitude variation only. The distinct prominent amplitude over left fronto-central sites, and it
topographies are more compatible with the hypothesis that deviates in this aspect from the topography of small and
large one-digit multiplications invoke other processes than large nonzero problems. The fact that zero problems evoke a
pure fact retrieval. Of course, by means of ERPs, we cannot much larger amplitude over left anterior electrodes than
decide which strategies our participants used to solve such small nonzero problems (Fig. 3, second row) strongly
problems. ‘‘Derived facts’’ is the most reported strategy to suggests that they are solved by a strategy that is clearly
solve problems with larger operands (see Refs. [28,43]). different from fact retrieval. Most likely, the correct results
Thus, it is likely that participants in the present study are generated by applying the zero rule.
sometimes decomposed large problems into a series of It is intriguing that this negativity, which most likely
smaller, easily accessible facts. accompanies a rule-based strategy in mental calculation, has
However, P300 amplitude was the same for small and a similar topography as a conceptually related phenomenon
large problems, indicating that small and large problems observed with linguistic material, that is, the left anterior
were initially classified as equally demanding and that the negativity (LAN). LANs with a maximum at left frontal to
same amount of processing resources has been allocated to central sites have been observed in two situations: first in
the two problem types. Thus, the first categorization of the sentences with a rule violation (e.g., incorrect morphosyntax
problems seems not to fully anticipate the actual processing or incorrect usage of an auxiliary; e.g., see Refs. [16,37])
demands. From this, it can be also concluded that use of and, second, in grammatically and semantically correct
strategies other than fact retrieval might not be the default sentences with less used, and therefore less easily parsed,
option for large problems, but rather that such strategies are grammatical constructions (see Refs. [23,39]). For example,
activated only when fact retrieval fails (cf. Refs. [41,42]). Rösler et al. [39] presented German sentences in which the
Although the topography of scalp-recorded slow waves order of sentence elements (subject, indirect object, and
can give only a rough estimate of the location of underlying direct object) was systematically varied. Note that the
generators, the location of the most prominent topographic German language—due to direct case marking—allows
differences between small and large problems is neverthe- constructing grammatically correct sentences of different
less suggestive. The amplitude difference over occipital to complexity by changing the word order. In this study,
right temporal electrodes indicates stronger activation of the determiners, which function as case markers in German,
underlying areas. This agrees with previous findings evoked a left anterior negativity whenever they indicated
[20,21], which related ERP effects measured at right inferior that a noun phrase sequence would not continue in its
parietal to temporal areas to number comparison and mag- canonical (expected) word order. Rösler et al. interpreted
nitude estimation processes [11,12]. Accordingly, it might the LAN as indicating that a deviation from the canonical
192 K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193

word order triggers an additional rule-based parsing step proper in a handedness recognition task, Psychophysiology 36
(1999) 399 – 408.
(e.g., that the next noun phrase is not immediately assigned
[6] N. Birbaumer, T. Elbert, A.G. Canavan, B. Rockstroh, Slow potentials
to the syntactic role of the actor). of the cerebral cortex and behavior, Physiol. Rev. 70 (1990) 1 – 41.
The fact that a left anterior negativity is evoked with [7] J.I.D. Campbell, Mechanisms of simple addition and multiplication: a
similar topography by both linguistic and arithmetic modified network-interference theory and simulation, Math. Cogn. 1
material whenever a rule application is most likely (1995) 121 – 164.
invoked suggests some conceptual parallelism between [8] J.I.D. Campbell, D.J. Graham, Mental multiplication skill: structure,
process and acquisition, Can. J. Psychol. 39 (1985) 338 – 366.
linguistic and arithmetic processing (see Refs. [20,34] [9] J.I.D. Campbell, D.P.M. Tarling, Retrieval processes in arithmetic
for a comparison of arithmetic and semantic N400 production and verification, Mem. Cogn. 24 (1996) 156 – 172.
effects). Possibly similar mechanisms and overlapping [10] L. Cohen, S. Dehaene, F. Chochon, S. Lehéricy, L. Naccache,
neural structures are involved if linguistic and arithmetic Language and calculation within the parietal lobe: a combined
representations are modified in working memory by rule- cognitive, anatomical and fMRI study, Neuropsychologia 38
(2000) 1426 – 1440.
based operations. In other words, the common feature [11] S. Dehaene, Cerebral bases of number processing and calculation,
between the arithmetic and the linguistic LAN is seen in in: M. Gazzaniga (Ed.), The New Cognitive Neurosciences, MIT
the fact that in both cases, default strategies—as extrap- Press, Cambridge, 2000, pp. 987 – 998.
olation of overlearned grammatical structures or direct [12] S. Dehaene, L. Cohen, Towards an anatomical and functional model
access to multiplication facts—fail and have to be of number processing, Math. Cogn. 1 (1995) 83 – 120.
[13] E. Donchin, M.G. Coles, Is the P300 component a manifestation of
replaced by a distinct rule-based processing step, which context updating? Behav. Brain Sci. 11 (1988) 357 – 427.
imposes extra load on executive and/or working memory [14] R. El Yagoubi, P. Lemaire, M. Besson, Different brain mechanisms
functions. These could be reflected by the LAN. mediate two strategies in arithmetic: evidence from event-related
brain potentials, Neuropsychologia 41 (2003) 855 – 862.
[15] O. Gruber, P. Indefrey, H. Steinmetz, A. Kleinschmidt, Dissociating
neural correlates of cognitive components in metal calculation, Cereb.
5. Conclusion Cortex 11 (2001) 350 – 359.
[16] A. Hahne, A.D. Friederici, Rule-application during language compre-
By means of ERPs, we could provide evidence that hension in the adult and the child, in: A.D. Friederici, R. Menzel
strategies other than pure fact retrieval are used to solve (Eds.), Learning: Rule Extraction and Representation, Walter de
Gruyter and Co., Berlin, Germany, 1999, pp. 71 – 88.
single-digit multiplication problems. Our data sustain the
[17] E. Hennighausen, M. Heil, F. Rösler, A correction method for DC-
distinction between rule-based and fact-retrieval problems drift artifacts, Electroencephalogr. Clin. Neurophysiol. 86 (1993)
and they support the notion that problem size determines the 199 – 204.
solution process, too [28]. These functional dissociations [18] H. Huynh, L.S. Feldt, Estimation of the box correction for degrees of
observed here with ERPs are congruent with neuropsycho- freedom from sample data in randomized block and split-plot designs,
logical findings showing differential impairments of acalcu- J. Educ. Stat. 1 (1976) 69 – 82.
[19] R. Johnson, A triarchic model of P300 amplitude, Psychophysiology
lia patients, in particular, a dissociation between facts and 23 (1986) 367 – 384.
rules, on one hand [44], and problems with small and large [20] K. Jost, E. Hennighausen, F. Rösler, Comparing arithmetic and seman-
operands, on the other hand [31]. tic fact retrieval: effects of problem size and sentence constraint on
event-related brain potentials, Psychophysiology 41 (2004) 46 – 59.
[21] M. Kiefer, S. Dehaene, The time course of parietal activation in
single-digit multiplication: evidence from event-related potentials,
Acknowledgements Math. Cogn. 3 (1997) 1 – 30.
[22] E.P. Kirk, M.H. Ashcraft, Telling stories: the perils and promise of
This research was supported by grants Ro529/10 and using verbal reports to study math strategies, J. Exp. Psychol., Learn.
Ro529/17-1 from the German Research Foundation (DFG) Mem. Cogn. 27 (2001) 157 – 175.
[23] R. Kluender, M. Kutas, Bridging the gap: evidence from ERPs on
assigned to Frank Rösler.
the processing of unbounded dependencies, J. Cogn. Neurosi. 5
(1993) 196 – 214.
[24] J. Kong, Y. Wang, H. Shang, Y. Wang, X. Yang, D. Zhuang, Brain
References potentials during mental arithmetic—effects of problem difficulty on
event-related brain potentials, Neurosci. Lett. 260 (1999) 169 – 172.
[1] M.H. Ashcraft, The development of mental arithmetic: a chronometric [25] K.M. Lee, Cortical areas differentially involved in multiplication and
approach, Dev. Rev. 2 (1982) 213 – 236. subtraction: a functional magnetic resonance imaging study and cor-
[2] M.H. Ashcraft, Children’s knowledge of simple arithmetic: a devel- relation with a case of selective acalculia, Ann. Neurol. 48 (2000)
opmental model and simulation, in: C.J. Brainerd, R. Kail, J. Bisanz 657 – 661.
(Eds.), Formal Methods in Developmental Research, Springer, New [26] J. LeFevre, A.G. Kulak, Individual differences in the obligatory acti-
York, 1987, pp. 302 – 338. vation of addition facts, Mem. Cogn. 22 (1994) 188 – 200.
[3] M.H. Ashcraft, Cognitive arithmetic: a review of data and theory, [27] J. LeFevre, J. Morris, More on the relation between division and
Cognition 44 (1992) 75 – 106. multiplication in simple arithmetic: evidence for mediation of division
[4] M.H. Ashcraft, Cognitive psychology and simple arithmetic: a review solutions via multiplication, Mem. Cogn. 27 (1999) 803 – 812.
and summary of new directions, Math. Cogn. 1 (1995) 3 – 34. [28] J. LeFevre, J. Bisanz, K.E. Daley, L. Buffone, S.L. Greenham, G.S.
[5] J. Bajrič, F. Rösler, M. Heil, E. Hennighausen, On separating pro- Sadesky, Multiple routes to solution of single-digit multiplication
cesses of event categorization, task preparation, and mental rotation problems, J. Exp. Psychol. Gen. 125 (1996) 284 – 306.
K. Jost et al. / Cognitive Brain Research 20 (2004) 183–193 193

[29] J. LeFevre, G.S. Sadesky, J. Bisanz, Selection of procedures in mental [40] M.D. Rugg, M.G.H. Coles (Eds.), Electrophysiology of Mind: Event-
addition: reassessing the problem size effect in adults, J. Exp. Psy- Related Brain Potentials and Cognition, Oxford Univ. Press, London,
chol., Learn. Mem. Cogn. 22 (1996) 216 – 230. 1995.
[30] G. McCarthy, C.C. Wood, Scalp distributions of event-related poten- [41] J. Shrager, R.S. Siegler, SCADS: a model of children’s strategy
tials: an ambiguity associated with analysis of variance models, Elec- choices and strategy discoveries, Psychol. Sci. 9 (1998) 405 – 410.
troencephalogr. Clin. Neurophysiol. 62 (1985) 203 – 208. [42] R.S. Siegler, C. Shipley, Variation, selection, and cognitive change,
[31] M. McCloskey, W. Harley, S.M. Sokol, Models of arithmetic fact in: T. Simon, G. Halford (Eds.), Developing Cognitive Competence:
retrieval: an evaluation in light of findings from normal and brain- New Approaches to Process Modeling, Erlbaum, Hillsdale, NJ, 1995,
damaged subjects, J. Exp. Psychol., Learn. Mem. Cogn. 17 (1991) pp. 31 – 76.
377 – 397. [43] B.L. Smith-Chant, J. LeFevre, Doing as they are told and telling it
[32] K. Miller, M. Perlmutter, D. Keating, Cognitive arithmetic: compar- like it is: self-reports in mental arithmetic, Mem. Cogn. 31 (2003)
ison of operations, J. Exp. Psychol., Learn. Mem. Cogn. 10 (1984) 516 – 528.
46 – 60. [44] S.M. Sokol, M. McCloskey, N.J. Cohen, D. Aliminosa, Cognitive
[33] M. Niedeggen, F. Rösler, N400 effects reflect activation spread during representations and processes in arithmetic: evidence from the perfor-
retrieval of arithmetic facts, Psychol. Sci. 10 (1999) 271 – 276. mance of brain-damaged patients, J. Exp. Psychol., Learn. Mem.
[34] M. Niedeggen, F. Rösler, K. Jost, Processing of incongruous mental Cogn. 17 (1991) 355 – 376.
calculation problems: evidence for an arithmetic N400-effect, Psycho- [45] R. Stanescu-Cosson, P. Pinel, P.F. van de Moortele, D. Le Bihan, L.
physiology 36 (1999) 307 – 324. Cohen, S. Dehaene, Understanding dissociations in dyscalculia: a
[35] J.M. Parkman, Temporal aspects of simple multiplication and com- brain imaging study of the impact of number size on the cerebral
parison, J. Exp. Psychol. 95 (1972) 437 – 444. networks for exact and approximate calculation, Brain 123 (2000)
[36] M. Pesenti, N. Depoorter, X. Seron, Noncommutability of the N + 0 2240 – 2255.
arithmetical rule: a case study of dissociated impairment, Cortex 36 [46] E.H. Stazyk, M.H. Ashcraft, M.S. Hamann, A network approach to
(2000) 445 – 454. mental multiplication, J. Exp. Psychol., Learn. Mem. Cogn. 8 (1982)
[37] F. Rösler, A. Friederici, P. Pütz, A. Hahne, Event-related brain poten- 320 – 335.
tials while encountering semantic and syntactic constraint violations, [47] T.P. Urbach, M. Kutas, The intractability of scaling scalp
J. Cogn. Neurosci. 5 (1993) 345 – 362. distributions to infer neuroelectric sources, Psychophysiology 39
[38] F. Rösler, M. Heil, B. Röder, Slow negative brain potentials as reflec- (2002) 791 – 808.
tions of specific modular resources of cognition, Biol. Psychol. 45 [48] G.F. Wilson, C.R. Swain, P. Ullsperger, ERP components elicited in
(1997) 109 – 141. response to warning stimuli: the influence of task difficulty, Biol.
[39] F. Rösler, T. Pechmann, J. Streb, B. Röder, E. Hennighausen, Parsing Psychol. 47 (1998) 137 – 158.
of sentences in a language with varying word order: word-by-word [49] N.J. Zbrodoff, G.D. Logan, On the relation between production and
variations of processing demands are revealed by event-related brain verification tasks in the psychology of simple arithmetic, J. Exp.
potentials, J. Mem. Lang. 38 (1998) 150 – 176. Psychol., Learn. Mem. Cogn. 16 (1990) 83 – 97.

Você também pode gostar