cortex 45 (2009) 11671177

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/cortex

Research report

Event-related brain potentials uncover activation dynamics in the lexicon of multiplication facts
Giovanni Galfanoa,*, Barbara Penolazzib, Ineke Vervaeckc, ` Alessandro Angrillib and Carlo Umiltab
a

Department of Developmental and Social Psychology, University of Padua, Italy Department of General Psychology, University of Padua, Italy c Department of Experimental Psychology, University of Gent, Belgium
b

article info
Article history: Received 3 June 2008 Reviewed 16 July 2008 Revised 10 September 2008 Accepted 19 September 2008 Action editor Ray Johnson Published online 1 October 2008 Keywords: Cognitive arithmetic ERPs Multiplication lexicon Spreading activation N400

abstract
The present study was designed to ascertain whether the cortical dynamics underlying multiplication fact retrieval can be modulated as a function of arithmetic relatedness. To this end, event-related brain potentials (ERPs) were recorded from 22 participants engaged in a number-matching task. The task was to decide whether a single probe number had been shown as part of a previously presented pair of cue numbers. Probes in the non-matching condition (50% of total trials) were either the product of the cue numbers (strong cue-probe association), the multiple of either cue numbers (weak cue-probe association), or an arithmetically neutral number with respect to the cue numbers (no cue-probe association). Behavioral data showed a clear interference effect (LeFevre interference), in that performance in the non-matching condition was signicantly better when the probe number was arithmetically neutral compared to when it was arithmetically related to the cue digits either strongly (i.e., through the product) or weakly (i.e., through its neighbor in the multiplication table). LeFevre interference was not statistically different in magnitude for the two arithmetically related conditions. In contrast, ERPs allowed us to dissociate cortical processing dynamics for these latter conditions: whereas initially (250350 ms) both product and neighbor probes evoked a relatively more pronounced positivity compared to neutral probes, in a later interval (350450 ms, N400-like component), neighbors critically diverged from products, showing relatively more negative values, similar to those of neutral probes. The observed dissociation in ERP measures is interpreted as evidence of activation spreading in the network of multiplication facts, with the short-lasting response elicited by neighbors likely reecting activation dissipating over time because of a weaker association to the cue digits. 2008 Elsevier Srl. All rights reserved.

1.

Introduction

The ability to solve simple mental calculation problems is critical in many contexts we face in everyday life. In the present

study, we aimed to explore the temporal dynamics underlying activation of multiplication facts stored in memory, by using event-related brain potentials (ERPs). This technique allows one to monitor online brain activity, and its use has recently

` * Corresponding author. Dipartimento di Psicologia dello Sviluppo e della Socializzazione, Universita di Padova, Via Venezia, 8, I-35131 Padova, Italy. E-mail address: giovanni.galfano@unipd.it (G. Galfano). 0010-9452/$ see front matter 2008 Elsevier Srl. All rights reserved. doi:10.1016/j.cortex.2008.09.003

1168

cortex 45 (2009) 11671177

promoted an important advance of knowledge concerning the different processes underlying arithmetic performance (e.g., Niedeggen and Rosler, 1999; Wang et al., 2000; Szucs and Csepe, 2005; Martn-Loeches et al., 2006; Nunez-Pena et al., 2006; Zhou et al., 2007). Memory retrieval seems the most-used strategy for solving simple multiplications in healthy adults (e.g., Rickard, 2005; Rusconi et al., 2006a; Campbell and Robert, 2008). However, evidence has been reported (e.g., LeFevre et al., 1996; Hecht, 1999; Lemaire and Reder, 1999) showing that nonretrieval strategies can play an important role in classic arithmetic tasks such as the arithmetic-production task (in which participants are required to generate the result for a given problem) and the arithmetic-verication task (in which participants are required to decide whether a proposed result for a given problem is correct or not). Because we aimed to address multiplication memory in isolation, in the present study, we used a different paradigm, rst proposed by LeFevre et al. (1988), and based on a number-matching task. Within this paradigm, arithmetic fact retrieval is reasonably assumed to take place implicitly, as arithmetic knowledge is neither necessary nor sufcient for performing the task (Rusconi et al., 2004). On each trial, participants are briey presented two single- or two-digit numbers (hereafter, the cue) followed, after a variable time interval, by a single one- or two-digit number (hereafter, the probe). Participants are required to decide whether the probe number was one of the cue numbers (matching trial) or not (non-matching trial). Thibodeau et al. (1996; also see Rusconi et al., 2004) have shown that performance on non-matching trials is signicantly worse when the probe is the product of the cue numbers than when it is arithmetically unrelated to the cue numbers. This interference effect (hereafter, LeFevre interference) is interpreted as evidence that participants, given the proper eliciting conditions, cannot avoid activating (i.e., involuntary retrieving) the product of the cue numbers, just as if they were treating them as operands of a multiplication problem. Activation of taskirrelevant knowledge generates a conict with task-relevant knowledge that needs to be resolved before the correct response can be executed. Critically, this holds true when the probe is the product of the cue numbers but not when the probe is arithmetically unrelated to them, which would account for the lower performance for product probes. Importantly, LeFevre interference has proved a reliable measure of the acquisition of arithmetic facts in children and adults (LeFevre and Kulak, 1994; LeFevre et al., 1994), and a useful instrument to test arithmetical abilities in neuropsychological patients (Rusconi et al., 2006b; Zamarian et al., 2006, 2007). One of the most notable advantages characterizing the number-matching paradigm resides in the fact that, as anticipated earlier, arithmetic fact mastery is tested implicitly, because participants are not asked to perform arithmetic. Therefore, LeFevre interference can only result from involuntary retrieval (i.e., activation) of arithmetic facts, because participants are not strategically involved in solving arithmetic problems. In light of this reasoning, the number-matching paradigm overcomes the problem of having participants choosing among several possible (and effortful) non-retrieval strategies that are available in both arithmetic production and verication tasks (e.g., LeFevre et al., 1996; Lemaire and Reder, 1999). Furthermore, the number-matching task, although similar to

an arithmetic-verication task in terms of the overall number of stimuli (two cue numbers and one probe number in number matching vs two operands and one result in arithmetic verication), is much less vulnerable to the non-trivial issues related to the selection of wrong results in the latter task, which can bias participants in adopting non-retrieval strategies reected in different ERP patterns (e.g., El Yagoubi et al., 2003; Nunez Pena and Escera, 2007). Several lines of research have suggested that multiplication facts may be represented in memory as verbal associations. This hypothesis is based on abundant neuropsychological evidence (see, e.g., Dehaene et al., 2004; Delazer et al., 2006) and has recently received empirical support from studies aimed at investigating ERP correlates of multiplication fact retrieval. Niedeggen et al. (1999; also see Jost et al., 2004a) have tested this idea by contrasting ERPs elicited during verication tasks in which participants were required to evaluate the correctness of single-digit multiplication problems vs the semantic congruity of sentences. The results showed that precisely the same ERP component, known as N400 (Kutas and Hillyard, 1980), could be observed independently of whether participants were engaged in the arithmetic or the semantic task. One of the most compelling hypotheses about the functional signicance of N400 is that it may reect automatic spreading of activation in the word lexicon. This hypothesis is supported by evidence showing that N400 amplitude is modulated in semantic priming contexts under conditions of masked priming (e.g., Kiefer and Brendel, 2006) and even in attentional blink paradigms (e.g., Rolke et al., 2001). Based on this logic, Niedeggen et al. (1999) interpreted the observation that arithmetic incongruencies evoked the same N400 component as semantic incongruencies as evidence that the multiplication lexicon and the word lexicon, although functionally independent (McCloskey, 1992), may be governed by similar mechanisms, the most likely candidate being spreading activation (e.g., Collins and Loftus, 1975). This suggestion has been recently backed up by an ERP study conducted by Galfano et al. (2004), in which a numbermatching paradigm was used. Galfano and colleagues demonstrated that LeFevre interference for products resulted in an amplitude modulation of an N400-like component, with neutral probes eliciting relatively more negativity than product probes. This pattern was very similar to the so-called N400 semantic priming effect, in which N400 amplitude in a lexical decision task is larger (i.e., relatively more negative) when a probe word (e.g., lemon) is preceded by a semantically unrelated prime (e.g., house) compared to a semantically related prime (e.g., sour). Because arithmetic in the numbermatching paradigm is irrelevant to the task at hand, Galfano et al. concluded that the amplitude of this N400-like component may reect the amount of additional activation within the memory network that is needed to process a just-encountered item (also see Niedeggen and Rosler, 1999). Hence, the more a representation has been primed previously (the product probe in the number-matching paradigm; a semantically related word in a lexical-decision task), the less additional activation is necessary, and the smaller the N400 will be. In the present study, we aimed to investigate the cortical dynamics underlying activation of multiplication facts by manipulating the strength of association between the cue

cortex 45 (2009) 11671177

1169

numbers and the probe number. More specically, we attempted to determine whether multiplication fact retrieval could be modulated as a function of arithmetic relatedness. Interestingly, in their arithmetic verication studies, Niedeggen and Rosler (1999; also see Niedeggen et al., 1999, Experi ment 2) already reported a smaller N400-like amplitude for related incorrect solutions than for unrelated incorrect solutions. In the present study, however, we were interested in involuntary multiplication fact retrieval. Hence, we used the number-matching paradigm and tested possible modulations of multiplication fact activation depending on whether the degree of association among cue numbers and the probe was strong (as in the case of product probes) or weak (as in the case of table-related neighbors of the product, i.e., multiples of either cue numbers). To this end, we presented participants with a cue (e.g., 8 7) that could be followed by either a product probe (e.g., 56), a neutral probe (e.g., 12) or a multiple of either cue number (e.g., 64). Galfano et al. (2003) have already shown that neighbor probes are capable of inducing LeFevre interference in behavioral measures (14 ms, estimated by averaging the effect across experiments). Galfano et al. (2004), using similar experimental settings with product probes only, reported an 18-ms LeFevre effect. Zamarian et al. (2006, 2007) have reported two neuropsychological investigations concerning multiplication facts mastery in patients with semantic dementia and mild cognitive impairment in which LeFevre interference, among other instruments, was used to assess arithmetic facts mastery. They used a number-matching task with both product and multiple probes. Critically, however, their results showed no hint of LeFevre interference that varied in magnitude as a function of strength of association between probe and cue numbers. Although examination of the magnitude of LeFevre interference reported in the studies of Galfano et al. (2003, 2004) and Zamarian et al. (2006, 2007) seems to suggest that the two effects may not differentiate, on the basis of models of mental calculation (e.g., Ashcraft, 1987; Campbell, 1995; Verguts and Fias, 2005), we predicted that the amount of interference should vary as a function of the strength of association between cue and probe numbers. More specically, we should observe a stronger LeFevre interference for product probes than for multiple probes. In this regard, it should be noted that Zamarian et al. (2006, 2007) used a variant of the number-matching task adapted to patients, which differs in many substantial experimental aspects from that typically used with healthy adults. Subtle factors related to stimulus selection and to the fact that the task was necessarily made brief (i.e., with a reduced number of data points) for patienttesting purposes may, at least partially, account for the null differences between product and multiple probes. In light of this argument, the rst aim of the present study was to investigate LeFevre interference further with products and multiples within the same participants, using the standards of experimental testing with healthy humans. Still, nding no difference in terms of magnitude of LeFevre interference as a function of strength of association between cue and probe numbers may simply reect the low resolution of behavioral measures. Hence, the second aim of the study was to assess whether the strength of association between cue and probe numbers is mirrored in ERP differences, which might represent a more nely tuned index of automatic

spreading activation. In keeping with the parallel between the arithmetic facts lexicon and word lexicon, it could be observed that LeFevre interference evoked by neighbor probes may recall the situation known in psycholinguistic research as indirect or mediated semantic priming (e.g., Weisbrod et al., 1999; Kreher et al., 2006). In that kind of study, the prime is related to the target only indirectly, through a direct association with another word that is not explicitly presented (e.g., from lion to stripes via tiger). In the same fashion, multiples in the arithmetic facts lexicon may be activated via the product. If this were the case, then one may predict that the amplitude of the N400-like component (which reaches the maximum for neutral probes, see Galfano et al., 2004) should be less pronounced for product than for multiple probes, and in turn should be less pronounced for multiple than for neutral probes. With respect to the organization of the multiplication lexicon, this outcome may be accounted for by assuming an architecture in which activation spreads sequentially from cue numbers to product and, later in time, from product to its neighbors (i.e., multiples of either cue number; see Fig. 1, Panel A). The same pattern of amplitude modulation, however, would also be consistent with a model in which activation from cue numbers simultaneously spreads to both the product and its neighbors (see Fig. 1, Panel B). Another possible scenario related to the model illustrated in Fig. 1 (Panel A) may be expected based on the ndings reported by Galfano et al. (2003). Basically, Galfano et al. (2003) did not observe any modulation in LeFevre interference as a function of whether the probe was a multiple above or below the product (e.g., 12 vs 24 preceded by 6 and 3). They reasoned that this result was not consistent with table-like models (e.g., Ashcraft and Battaglia, 1978) in which activation rst spreads from the problem multipliers to the multiples below the product, then to the product, and then to the multiples above the product. In contrast, they interpreted the behavioral pattern as consistent with models assuming activation spreading from the product directly to closest neighbors or multiples (e.g., Ashcraft, 1987). If this model is correct, then it would be expected that multiples in the present experiment will elicit an evoked component similar to that elicited by products but, critically, in a later time window. Finally, strength of association between cue and probes may be reected in processing dynamics varying over time also for models assuming direct activation spreading from cue numbers to multiples (i.e., not mediated via the product). Specically, one may expect neighbors weaker association to cue numbers to elicit a short-lasting effect (i.e., a difference in amplitude with neutral probes that dissipates as a function of time) compared to that evoked by products. This different time course in the evoked brain activity for products and neighbors may be accounted for by activation fading away more quickly for items that are associated less strongly (see Fig. 1, Panel C).

2.
2.1.

Methods
Participants

A total of 22 undergraduate students (14 females) took part in the experiment. All participants gave their informed consent

1170

cortex 45 (2009) 11671177

Fig. 1 Hypothetical network-based architectures for the lexicon of multiplication facts and related activation dynamics. Arrows denote activation spreading. Line thickness indicates strength of activation. A thicker line indicates stronger activation. Grey indicates activation decay as a function of time. Panel A illustrates a model in which activation rst spreads from the cue numbers to the product at time (T) 1 and, from the product to its neighbors (multiples of either cue number) at T2. According to this model, at T1 only activation of the product should be visible. In contrast, at T2, only activation of the multiples should emerge. Panel B shows a model in which activation spreads directly and simultaneously from cue numbers to both product and multiples. This model would only predict differences in strength of activation of product and multiples independent of time (as suggested by indirect semantic priming studies in the linguistic domain). Panel C illustrates a model in which activation also spreads directly from cue numbers to both product and multiples, but allows predictions concerning activation fate as a function of time. According to this model, a difference in activation between product and multiples should be expected in terms of persistence. Activation should be long-lasting for products and subjected to decay for multiples.

prior to their inclusion in the study. They received 13 euros for their participation. Their age ranged from 19 to 25. All participants had normal or corrected-to-normal vision and were naive as to the purpose of the experiment. None of them reported having any neurological or psychiatric disorder. Three participants were left-handed, 19 were right-handed. The study was conducted following the guidelines put forward by the Ethics committee at the University of Padua and in accordance with the Declaration of Helsinki.

2.2.

Stimuli

Stimuli appeared in white against a black background and were displayed centered on a Sony 19-inch color monitor

(640 480, 75 Hz) placed about 50 cm in front of participants. Each character measured 8 mm (.92 of visual angle) in height and 5 mm (.57 ) in width. The distance between the two numbers in the cue was 9 mm. Each trial consisted of a number pair cue (e.g., 3 and 6) followed by a number probe (e.g., 18). The participants task was to decide whether the probe number matched either number in the cue (i.e., number-matching task). Half of the stimuli were matching trials, half were non-matching trials. Consistent with Thibodeau et al. (1996) and Galfano et al. (2003), there were different cueprobe association types (see Appendix A). For the non-matching condition (i.e., the critical trials to assess our hypotheses), there were product trials, multiple trials, neutral trials and non-matching ller trials, occurring with the same frequency. That is, we used the same logic implemented in studies on indirect priming (e.g., Weisbrod et al., 1999; Kreher et al., 2006), in which stimuli in the three critical conditions typically occur in the same proportion. Two single-digit numbers were used as the cue for the product trials (e.g., 6 7). The probe of these trials was the product of the two digits in the cue (e.g., 42). For multiple trials, the cue was the same as for the product trials, whereas the probe was the multiple above the product related to the second number in the cue (e.g., 48 resulting from 6 8). Neutral probes had the same cues as product and multiple trials. However, in order to avoid table-related activation, neutral probes were arithmetically unrelated to either number in the cue (e.g., 38). To reiterate, the critical trials for assessing LeFevre interference all required the same motor response. The last category of non-matching trials consisted of ller trials. The cue and the probe of the ller trials included a double-digit number (e.g., 8 54, and 36), in order to have a condition in which the participants saw double-digit numbers in the cue. These trials also served the purpose of balancing the number of related and unrelated trials for the non-matching condition. However, because they served no purpose to assess our hypotheses concerning LeFevre interference, they were not included in the analysis. Unlike the non-matching stimuli, on matching trials, one of the digits in the cue matched the probe (see Appendix A). These trials, not relevant to our hypotheses, also included four categories: probe-balancing trials related to the product trials, probe-balancing trials related to the multiple trials, cue-balancing trials and matching ller trials. The probebalancing trials related to the product trials had the same probes as the product probes in the non-matching condition (e.g., 42). The probe-balancing trials related to the multiple trials had the same probes as the multiple probes in the nonmatching condition (e.g., 48). The cue-balancing trials had the same cue as the product, multiple and neutral trials in the non-matching condition (e.g., 6 7). The matching ller trials had the same probe as non-matching neutral trials (e.g., 38). For the crucial conditions (i.e., non-matching product, multiple, and neutral trials), several criteria for stimulus selection were adopted in order to minimize possible confounds. First, combinations of cues and probes that might have elicited activation on the basis of arithmetic relations other than multiplication (e.g., subtraction, 9 3, and 6) were eliminated from the stimulus set. Second, ties (e.g., 4 4) were also excluded, because they are assumed to have

cortex 45 (2009) 11671177

1171

an easier access to the memory store than other problems (e.g., Campbell, 1995). Cues and probes containing 0, 1, or 5 were also discarded from the stimulus set, as problems with these numbers as operands seem to elicit a set of rules instead of results (e.g., McCloskey et al., 1991; Jost et al., 2004b). Third, all probes were even, in order to avoid possible strategies based on the so called oddeven rule (Krueger, 1986). Fourth, each probe in the critical conditions appeared the same number of times. Finally, the average magnitude of the probes in the non-matching related condition (product and multiple trials) was the same as in the non-matching neutral condition (36.4). This constraint was built in to rule out the possibility that participants responded more slowly on product trials because the distance between cue and probe was larger than the distance in neutral trials (e.g., Dehaene and Akhavein, 1995). Note that, because the criteria illustrated here are very strict, in order to have a reasonable number of stimuli, we also included stimuli in which the single-digit numbers in the cues matched either number in the doubledigit probes (see Appendix A). Hence, one stimulus had a partial match between one of the numbers in the cue and the probe. This, however, was true for every probe condition, and previous experiments in which partial matches were included in the stimulus set have shown that this factor does not inuence behavioral data (Galfano et al., 2003). With all the constraints taken into account, we selected a list of ve stimuli for each category. This list was repeated 14 times,1 so that we had 70 data points for each cell of the design, and a total of 560 trials for each participant. The sequence of trials was randomized separately for each participant.

Fig. 2 Sequence of events in the number-matching task. A non-matching product trial (on the left), a non-matching multiple trial (middle), and a non-matching neutral trial (on the right) are illustrated. The three types of trial occurred with same frequency.

2.3.

Procedure

The experiment and the recordings took place in a soundattenuated, electrically shielded, dimly lit room. The participants were seated in a comfortable chair, about 50 cm from the monitor. Before the experiment started, participants performed some practice trials until they became familiar with the task. The sequence of events in a trial for the three critical conditions is illustrated in Fig. 2. Each trial began with a 500-Hz tone as a warning signal lasting for 100 ms. Along with this signal, the xation point (#) was shown for 400 ms at the center of the screen (see Fig. 2). The two digits of the cue were presented synchronously to xation point offset. The cue frame lasted 60 ms and was immediately followed by a mask frame of 40 ms consisting of seven # symbols. After mask frame offset, there was an interstimulus interval (ISI) of 20 ms.
1 It could be argued that using only ve stimuli for each category may represent a very narrow picture of multiplication lexicon. However, this stimulus set was chosen in order to rule out alternative interpretations of differences (if any) among product, multiple and neutral probes based on numeric properties other than those related to multiplicative association between cue and probes. In addition, the fact that each critical stimulus was repeated so many times, if anything, should have worked against our hypotheses of nding differences in processing product and multiple probes. That is because high repetition may strengthen pre-existing associations between cue numbers and multiples. On the grounds of this reasoning, we argue that our experimental scheme was conservative.

This resulted in a xed 120-ms stimulus onset asynchrony (SOA). Then, the probe appeared and lasted until participants responded or 1900 ms had elapsed, whichever came rst. The SOA incorporated in the present experiment was chosen because previous studies with similar experimental settings have proved that a 120-ms SOA is indeed able to reveal LeFevre interference for both products (e.g., Galfano et al., 2004) and multiples (Galfano et al., 2003). The intertrial interval was set at 1100 ms. The participants were required to press the P button on the keyboard when the probe matched either number in the cue, and the Q button when there was no match. Response key assignment was counterbalanced across participants. When the participants responded correctly, the message correct response appeared on the screen. When they either responded incorrectly or no response was given within 1900 ms, the message error appeared on the screen along with a 500-Hz tone that lasted for 500 ms. Participants were explicitly encouraged to make eye-blinks during the intertrial interval only, and to avoid making any movements between xation point onset and response. They were instructed to respond as quickly and accurate as possible. Each participant performed a single experimental session consisting of 14 blocks of 40 trials each. Participants were allowed short breaks between blocks of trials. The entire session, including electrode placement, practice and experimental trials required about 120 min.

2.4.

EEG recording and analysis

EEG was sampled continuously, using SynAmp ampliers (Neuroscan system), in DC mode, with a .0150-Hz bandpass lter, and sampling rate of 500 Hz. Impedance was kept below 5 kU. Brain electrical signals were recorded by means of 25 tin electrodes. Nineteen electrodes were placed on an elastic cap (Electrocap) according to the extended International 1020 System (Oostenveld and Praamstra, 2001). The remaining seven electrodes were positioned below each eye (Io1, Io2), on the external canthii (F9, F10), on the mastoids (M1, M2) and on the nasion (Nz). Cz was used as recording reference for all channels. Data were converted ofine to a calculated linked mastoids reference ((M1 M2)/2). EEGblink artifacts were corrected by using vertical electrooculogram channels as reference. Next, data were epoched in the

1172

cortex 45 (2009) 11671177

interval from 400 to 2000 ms with respect to probe onset, and then aligned to the 220120-ms pre-stimulus baseline. Trials with residual artifacts (exceeding 100 mV) were visually inspected and rejected, leading to an average trial discard of 21.37%. The accepted epochs were then averaged for each of the crucial experimental conditions (neutral, product and multiple trials), low-pass ltered (20 Hz, 24 dB/oct) and nally cut in the 200/1500-ms interval. Frontal (F3, Fz, F4), central (C3, Cz, C4), and parietal sites (P3, Pz, P4) were included in the statistical analyses. Fig. 3 shows grand mean waveforms of

the three experimental conditions, averaged over all participants, for the midline electrodes. In order to detect the exact time course of the activity elicited by the three probe types, we computed statistical analyses on mean amplitude values within four contiguous 50-ms time windows, starting from 250 ms. This ne-grained analysis was aimed to capture variations in brain activity across time among the three probe types expected on the basis of the models of spreading activation highlighted in Fig. 1. Critically, on the grounds of the results reported by Galfano et al. (2004), we expected to replicate the difference between neutral and product probes in both the 350400- and 400450-ms time windows (i.e., the so called arithmetic N400-like effect). A four-way ANOVA was carried out including the following factors: probe type (three levels: neutral vs product vs multiple), time window (four levels: 250300 vs 300350 vs 350400 vs 400 450 ms), caudality (three levels: anterior vs central vs posterior electrodes), and laterality (three levels: left vs midline vs right electrodes). In addition, to further investigate the late component detected in grand average curves, we performed, within the 500900-ms time window, an additional three-way ANOVA, including the within-group factors probe type, caudality and laterality. The HuynhFeldt correction was applied when sphericity assumptions were violated (Huynh and Feldt, 1970). In these cases, the uncorrected degrees of freedom, epsilon values and the corrected probability levels were reported. Posthoc comparisons were computed using the NewmanKeuls test.

3.
3.1.

Results
Behavioral data

Participants behavioral performance was assessed by means of correct response times (RTs) and proportion of correct responses. Data for matching stimuli and non-matching ller stimuli are shown in Table 1, but they were not further analyzed, because they served no purpose for assessing LeFevre interference.2 This effect was measured by comparing performance on product and multiple trials, and neutral trials (see also Thibodeau et al., 1996; Galfano et al., 2003, 2004). A one-way ANOVA with probe type as factor (neutral vs product vs multiple) was conducted on RTs for correct responses. It revealed a signicant main effect, F(2,42) 5.09, MSe 681.48, 3 .98, p < .01. Posthoc comparisons revealed that participants rejected a neutral probe (RT 709 ms, SD 157) faster than both multiple
One may notice that in the set of stimuli used in the present study (see Appendix A), cues consisting of two one-digit numbers required a non-matching response on 75% of total trials and may wonder whether this unbalance may have affected results. We discard this possibility on the basis of the following arguments. First, the cue was always short-lasting and backward-masked, which should make the possibility for participants to use strategies highly unlikely. Second, it is well known that increasing the relative frequency of a given response results in a decrease of RTs for that response (e.g., Sternberg, 1969). In accordance, had our participants noted the unbalance, they should have been faster on non-matching trials than on matching trials with the same cue (cue balancing trials). Inspection of Table 1 shows clearly that this was not the case.
2

Fig. 3 Electrophysiological results. Grand-average waveforms recorded at Fz, Cz, and Pz sites, from top to bottom. Grey vertical bars identify the different time windows included in the analyses.

cortex 45 (2009) 11671177

1173

Table 1 Mean values and standard deviations for reaction times (in ms) and proportion of correct responses in all the matching and non-matching conditions. RTs M
Non-matching Product probes Multiple probes Neutral probes Fillers Matching Product probe-balancing Multiple probe-balancing Neutral probe-balancing Cue-balancing 732 729 709 711 654

Proportion correct M
.85 .85 .87 .92 .86 .85 .93 .85

SD
161 170 157 162 130

SD
.11 .11 .11 .11 .11 .09 .10 .11

663 124 627 138 663 155

(RT 729 ms, SD 170) and product probes (RT 732 ms, SD 161), p .006 and p .001, respectively. In contrast, no signicant difference in mean RTs was found between product and multiple trials, p .71. Hence, a robust LeFevre interference effect emerged in RTs for both product and multiple probes.3 This pattern replicates the results of Galfano et al. (2003, 2004), who, in different studies with similar experimental settings, reported a signicant LeFevre interference effect for multiple probes, and product probes, respectively. A second ANOVA was performed on proportion of correct responses with the same factor as the ANOVA on RTs. No signicant main effect was found, F(2,42) 1.24, MSe .001, 3 1, p .3, indicating that the proportion of correct responses for product probes (M .85, SD .11) was not signicantly different from multiple (M .85, SD .11) and neutral trials (M .87, SD .11). This pattern, although not showing evidence of LeFevre interference, makes the possibility of a speed-accuracy tradeoff unlikely, since the proportion of correct responses for the three probe types was in the same direction as RT data.

3.2.

ERP data

less negative waves. In contrast, neutral probes always evoke the most negative waves, whereas multiple probes, throughout this interval, show shifting voltage values from those of product condition towards those of neutral condition. The size of the difference between relatively less negative curves evoked by products, on the one hand, and relatively more negative curves evoked by neutral and multiple probes, on the other hand, plainly becomes very broad in the following, long-lasting interval from around 500 to 900 ms. The rst ANOVA, performed on contiguous time windows, revealed a signicant main effect of time window, F(3,63) 10.137, MSe 31.78, 3 .58, p < .001, showing that from the rst to the third time window (250400 ms) there was a general increase in negativity, which was followed, in the last window (400450 ms), by a return towards amplitudes slightly more positive than in the previous interval. More important to the purpose of the study, a signicant main effect of probe type emerged, F(2,42) 3.365, MSe 17.09, 3 1.00, p .044). Posthoc comparisons revealed that the potential elicited by product probes (M 4.28 mV) was signicantly relatively less negative (p .036) than that elicited by neutral probes (M 3.74 mV). The potential evoked by multiple probes (M 3.95 mV), although showing values in between those of the other two probe types, was not statistically different from the product or neutral probes. The ANOVA also revealed a signicant probe type time interval interaction, F(6,126) 2.483, MSe 2.13, 3 .83, p .037 (see Fig. 4). Post-hoc comparisons conrmed the differences observed in grand average waveforms (see Fig. 3). Namely, in the rst time window (250300 ms), product (M 5.12 mV) and multiple (M 5.11 mV) probes elicited a smaller relative negativity ( p .017 and p .007, respectively) than neutral probes (M 4.72 mV). In the following time window (300350 ms), even though the multiple condition shifted towards more negative values, the same pattern of evoked activity was observed as in the 250300-ms window. That is, neutral probes (M 3.76 mV) always showed a larger relative negativity than both product (M 4.34 mV) and multiple (M 4.22 mV) probes (p < .001, and p < .005, respectively). However, starting from the subsequent time

Electrophysiological data were measured in terms of the components (mean amplitude values) evoked by the probe only for the neutral, product, and multiple conditions. Differences in the mean amplitude of the components elicited by the three conditions are clearly visible starting around 250 ms after probe onset and up to 450 ms. In particular, product probes were consistently associated to the relatively
3 In three out of ve of our stimuli (see Appendix A), the split between neutral and product probes was greater than the split between multiple and product probes. Critics may argue that the fact that neutral probes were responded to more efciently than multiple probes reected a split effect (Ashcraft and Battaglia, 1978), typically observed in arithmetic-verication tasks, instead of task-irrelevant activation of arithmetic knowledge. However, in a previous number-matching study (Galfano et al., 2003, Experiments 5 and 6), it was shown that making the split between product and neutral probes smaller than the split between products and multiples had no impact on LeFevre interference. In light of this evidence, the alternative account based on the split effect can be safely discarded.

Fig. 4 Mean voltages of the electrodes entered in the analyses for each probe type as a function of time window (interval) in the analysis of the arithmetic N400-like component. At the rst two time windows, product and multiple probes elicited less relatively negative activity than neutral probes. At the later two time windows, this pattern held true only for product probes. See text for details.

1174

cortex 45 (2009) 11671177

window (350400 ms) up to the end of the considered time course, multiples evoked potentials the extent of which were close to those elicited by neutral probes. In more detail, at the 350400-ms time window, product probes (M 3.73 mV) were associated with a signicantly smaller relative negativity than that elicited by both neutral (M 3.16 mV) and multiple (M 3.17 mV) probes (p < .001 and p < .001, respectively). The same pattern was conrmed in the 400450-ms interval, with product probes (M 3.92 mV) showing a signicantly less pronounced relative negativity compared to both neutral (M 3.33 mV) and multiple (M 3.32 mV) probes (p < .001 and p < .001, respectively). This pattern was present in all scalp regions, because no interactions between probe type and spatial factors (i.e., caudality and laterality) were signicant. The ANOVA computed on the last time window (500900 ms) showed a marginally signicant main effect of probe type, F(2,42) 3.107, MSe 6.97, 3 1.00, p .055. This was due to a similar pattern as that found in the previous analysis, that product probes (M 3.34 mV) were characterized by relatively larger potentials than neutral (M 2.86 mV) and multiple (M 2.72 mV) probes. The ANOVA also yielded a marginally signicant probe type caudality interaction, F(4,84) 2.81, MSe 1.41, 3 .67, p .052). As shown in Fig. 5, whereas mean voltages at anterior locations were very similar for the three probe types, in center-posterior sites the pattern of evoked cortical activity resembled that reported for the previous time windows. In more detail, a relatively smaller relative negativity for product probes with respect to both neutral and multiple probes can be observed in both central and posterior scalp regions (see Fig. 5). No other interactions involving probe type as factor were even close to signicance.

4.

Discussion

The present study was devoted to examining whether, in a number-matching task, interference induced by involuntary activation of arithmetic facts (i.e., LeFevre interference)

Fig. 5 Mean voltages for each probe type as a function of caudality (Anterior scalp region: F3, Fz, and F4. Central scalp region: C3, Cz, C4. Posterior scalp region: P3, Pz, P4) in the analysis of the late (500900 ms) component. The difference between product probes on the one hand, and multiple and neutral probes on the other side, is visible only over central and posterior sites.

was modulated by the strength of association in the multiplication lexicon. Associative strength between cue and probe numbers was manipulated by presenting product probes (assumed to be strongly associated to the cue numbers) and multiple probes (assumed to be weakly associated to cue numbers; Ashcraft, 1987). Although it is known that LeFevre interference can be generated by both products (e.g., Galfano et al., 2004; Rusconi et al., 2004) and their closest neighbors in multiplication tables (e.g., Galfano et al., 2003), no such comparison was performed within the same participants and in the same experiment with healthy adults. The present study was aimed at addressing this issue by assessing possible differences between strong and weak multiplicative associations with a combination of traditional behavioral measures and ERPs. This ERP technique is particularly suited to investigating such issues, because it provides temporal information about the time-course of the cognitive processes used in the task, which is essential to exploring the activation dynamics in lexical networks (e.g., Barber and Kutas, 2007). On the behavioral side, RTs showed signicant LeFevre interference for both product and multiple probes. This pattern of results conrms the results reported by Galfano et al. (2003) for multiples and Rusconi et al. (2004) for products (also see Zamarian et al., 2006, 2007). No reliable difference, however, emerged in LeFevre interference as a function of activation strength between cue and probe numbers. Indeed, although multiples induced less interference than products (20 vs 23 ms), the two effects were not statistically different. This outcome does not seem consistent with predictions stemming from inuential models of mental calculation (e.g., Ashcraft, 1987). According to these models, in which the lexicon of arithmetic facts is described as a network mainly governed by spreading activation (Collins and Loftus, 1975), the product should be activated more strongly than multiples, in that it would be directly linked to the cue numbers. That should result in a better lexical access. Therefore, there should have been a larger LeFevre interference compared to that evoked by multiples, which clearly was not the case. Still, as anticipated earlier, one cannot rule out the possibility that behavioral measures may be less than ideal to reveal these differences, since they represent the nal outcome of several processing stages. Unlike behavioral data, ERPs proved to be particularly sensitive in uncovering these differences. Strength of association between cue and probe was reected in a different time course of the scalp activity elicited by the two related conditions with respect to the neutral condition. Specically, in an early time window starting from 250 ms, brain activity evoked by product and multiple, although signicantly different from that elicited by neutral probes, was highly overlapping, thus in agreement with behavioral measures. We interpret the results observed at the earliest time window as reecting implicit access to the lexicon of multiplication facts and the associated activation spreading from cue numbers to product and multiples, which does not take place for neutral probes unrelated to cue numbers. Interestingly, this effect, which is perfectly matched by behavioral data, occurs very early. This nding is consistent with recent psycholinguistic studies documenting that lexical access can be reected in ERP modulations even within

cortex 45 (2009) 11671177

1175

200 ms from word onset (e.g., Penolazzi et al., 2007; Spironelli and Angrilli, 2007). In line with the literature on the arithmetic N400-like component (e.g., Niedeggen et al., 1999; Galfano et al., 2004), starting from 350 ms, the amplitude difference between product and neutral probes was still reliable, again reecting a difference in activation between these two probe types. However, multiples probes, whose evoked activity was similar to that elicited by products in the previous time window, generated a pattern overlapping with the one evoked by neutral probes. Hence, multiples, accessed automatically as testied by the early component we detected, showed an amplitude uctuation as a function of time that suggests a fast decay of activation, which was still present for products instead. This latter pattern of evoked activity (i.e., multiples and neutral probes showing overlapping potentials, both consistently different from those of product probes) was long-lasting and spread all over the scalp. At the latest time window (500900 ms), this pattern was somehow unexpected. Although participants responses were delivered within this time window, we are inclined to think that this slow positive complex reects activity independent of motor or responserelated processes as well (cf. Galfano et al., 2004). Looking at behavioral measures, multiple and product probes are characterized by longer RTs than neutral probes (i.e., both produce LeFevre interference). On the ERP side, by contrast, brain activity evoked by multiple probes is similar to that evoked by neutral than product probes. In light of this discrepancy between behavioral and electrophysiological data, we tend to interpret the data in the 500900-ms time window as indicative of residual activation of arithmetic networks, which may be independent of task-relevant knowledge (also see Galfano et al., 2004). More specically, it might be possible that the relative greater positivity characterizing product probes compared to multiples reects a stronger persistence in activation of task-irrelevant knowledge that is more strongly related to cue numbers. This interpretation may also be supported by the observation that the topography of this slow complex seems at least partially consistent with that reported in neuroimaging studies that identify posterior areas as the neural correlates of the arithmetic fact lexicon (e.g., Dehaene et al., 2004). The ERP effects found here replicate and extend those reported by Galfano et al. (2004), in which a number-matching paradigm was used with products as the only probes related to the cue numbers. However, it can be observed in the present study that an arithmetic N400-like effect seemed to arise earlier than the one observed by Galfano et al. (2004). In our view, among the factors that may account for this anticipated effect, stimulus sample may have played a role. Specically, compared to Galfano et al.s setting, the proportion of related trials in the present experiment was higher (1/4 vs 1/6 of total trials), and two related conditions were included instead of one (product and multiple vs product only). However, even if we are inclined to interpret these modications as the main reason responsible for the earlier onset of the N400-like effect, we are also reluctant to posit that such change might have inuenced the strategies participants used to perform the number-matching task. Indeed, critics may argue that the higher proportion of related trials may have favored a strategic

use of related trials to perform the task. However, had this been the case, then we should have found the use of such strategies reected in the behavioral results, with possible reverse interference effects, due to participants intentional use of arithmetic information to provide a fast non-matching response (i.e., a strategy that may sound like the probe is related to the cue numbers, hence it does not match them). Because behavioral data did not reveal such pattern, we discard this possibility. As regards ERP topography, another difference that becomes evident when comparing the present results with those reported by Galfano et al. (2004) is in the observation that the arithmetic N400-like effect was spread all over the scalp in the present study, whereas Galfano et al. reported an effect mainly visible over the right hemisphere. Such divergence, along with differences characterizing event-related activity at later time windows, may be related to the different proportion of related items in the two studies and (with reference to these later time windows) to methodological modications such as the fact that, unlike Galfano et al. (2004), response keys were perfectly counterbalanced in the present study. The important point, however, is that, besides these differences, the basic pattern of results was consistent across the two studies, which supports the robustness of the paradigm, and its suitability to address involuntary access and activation of the multiplication lexicon. As regards the possible models of multiplication lexicon illustrated in Fig. 1, the present ERP data allow one to get a more ne-grained picture than with behavioral results alone. More specically, whereas behavioral measures are consistent with all three different architectures discussed in the introduction, the ERP data unambiguously speak against two of them. It is worth noting that these models can be distinguished in two broad categories, based on the mechanism governing activation spreading in the lexicon. In more detail, architectures similar to the one depicted in Fig. 1 (Panel A) assume sequential activation spreading from cue numbers (or operands) to multiples indirectly, mediated through activation of the product. This mechanism is very similar to the one likely operating in the word lexicon, as shown by indirect semantic priming studies in the linguistic domain (e.g., Weisbrod et al., 1999; Kreher et al., 2006). By contrast, other possible architectures assume activation spreading from cue numbers (or operands) simultaneously to both products and multiples. Still, differences in strength of association between cue and related probes (products and multiples) can be reected in modulation of their activation in terms of either intensity (see Fig. 1, Panel B) or duration (see Fig. 1, Panel C). The observation of an early complete overlap of evoked activity for products and multiples seem to indicate that activation of these latter probes is not indirectly mediated via activation of the product, as suggested by the network retrieval model (Ashcraft, 1987), and hypothesized by Galfano et al. (2003). The model illustrated in Fig. 1, Panel A, explicitly assumes activation spreading rst from cue numbers to closest multiples via the product. This should have resulted in multiples eliciting an evoked activity similar to that evoked by products but, critically, either different in amplitude or shifted in time, which clearly was not the case. As a consequence, this model is not able to account for the present

1176

cortex 45 (2009) 11671177

electrophysiological data.4 The observation that the same early ERP component elicited by products and multiples did not differ in amplitude allows us to rule out also the model depicted in Fig. 1, Panel B. According to this model, the weaker strength of association for multiples compared to products should determine a difference in intensity of activation, which in turn should result in a different amplitude of the evoked activity of products and multiples. Because this prediction was not conrmed by the data, this kind of model seems neurophysiologically implausible. Our nding that products and multiples evoked an overlapping brain activity in the earliest time window indicates that both kinds of related probes receive activation from cue numbers directly and simultaneously. In addition, the observation that activity evoked by multiples, starting at 350 ms, becomes overlapping with that elicited by neutral probes suggests that activation of multiples, unlike that of products, is subjected to rapid decay. These results are overall consistent with the model in which strength of association between cue and probes is reected in processing dynamics that vary over time (Fig. 1, Panel C). More specically, multiples weaker association to cue numbers elicits a short-lasting effect (i.e., a difference in amplitude with neutral probes that dissipates as a function of time) compared to that evoked by products. This different time course of brain activity evoked by products and multiples can only be accounted for by a model assuming activation fading away more quickly for items that are associated less strongly (see Fig. 1, Panel C).

` MIUR to Carlo Umilta and to Alessandro Angrilli. We are indebted to Francesca Fardo for assistance in data collection, Veronica Mazza for helpful discussions, and Massimiliano Pastore for advice in data analysis. We also thank Ray Johnson Jr. and two anonymous reviewers for useful suggestions and comments on a previous draft.

Appendix A
Stimuli in the non-matching and matching conditions used in the experiment are shown in Table A.1.

Table A.1
Non-matching stimuli Products Multiples Neutrals Fillers 46 24 46 28 46 26 67 8 58 24 6 24 28 4 28 7 26 26 46 4 87 56 87 64 87 12 34 9 68 56 7 56 64 3 64 12 6 12 87 8 28 16 28 18 28 54 23 7 46 16 9 16 18 7 18 54 3 54 28 2 48 32 48 36 48 52 3 72 82 8 32 32 9 36 36 52 6 52 48 4 67 42 67 48 67 38 8 54 36 3 42 42 6 48 48 9 38 38 67 6

Matching stimuli

Product probebalancing Multiple probebalancing Neutral probebalancing

5.

Conclusion

Cue-balancing

The present study is the rst attempt to dissociate effects induced by products and multiples in a number-matching task in the same participants, using both behavioral and ERP measures. Unlike behavioral data, which are consistent with most models of multiplication lexicon (e.g., Ashcraft, 1987; Campbell, 1995; Verguts and Fias, 2005), the ERP results revealed activation dynamics that allowed us to conrm the neurophysiological plausibility of only one of these models. Specically, the data are consistent with a model that postulates a simultaneous spreading of activation from cue numbers to both products and multiples. Activation of multiples seems to possess the same intensity as that of products. However, activation of multiples is subjected to rapid decay, whereas is long-lasting in the case of products.

references

Acknowledgements
This research was supported in part by a grant from the University of Padova to Giovanni Galfano and by grants from
It could be objected that one cannot rule out the possibility for sequential activation of product and multiples to take place within 200 ms between cue and probe onset such that product and neighbors would have been primed already by the time that related probes deviate from unrelated probes. Although we cannot rule out this possibility, we discard the model depicted in Fig. 1 (Panel A) on the grounds of the observation that the difference in evoked activity between product and neutral probes was very long-lasting. This clearly violates the prediction that activation of products should be vanished at T2.
4

Ashcraft MH. Childrens knowledge of simple arithmetic: a developmental model and simulation. In Brainerd CJ, Kail R, and Bisanz J (Eds), Formal Methods in Developmental Psychology: Progress in Cognitive Development Research. New York: SpringerVerlag, 1987: 302338. Ashcraft MH and Battaglia J. Cognitive arithmetic: evidence for retrieval and decision processes in mental addition. Journal of Experimental Psychology: Human Learning and Memory, 4: 527 538, 1978. Barber HA and Kutas M. Interplay between computational models and cognitive electrophysiology in visual word recognition. Brain Research Reviews, 53: 98123, 2007. Campbell JID. Mechanisms of simple addition and multiplication: a modied network interference theory and simulation. Mathematical Cognition, 1: 121164, 1995. Campbell JID and Robert ND. Bidirectional associations in multiplication memory: conditions of negative and positive transfer. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34: 546555, 2008. Collins AM and Loftus EF. A spreading activation theory of semantic processing. Psychological Review, 82: 407428, 1975. Dehaene S and Akhavein R. Attention, automaticity, and levels of representation in number processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21: 314326, 1995. Dehaene S, Molko N, Cohen L, and Wilson AJ. Arithmetic and the brain. Current Opinion in Neurobiology, 14: 218224, 2004.

cortex 45 (2009) 11671177

1177

Delazer M, Karner E, Zamarian L, Donnemiller E, and Benke T. Number processing in posterior cortical atrophy a neuropsychological case study. Neuropsychologia, 44: 3651, 2006. El Yagoubi R, Lemaire P, and Besson M. Different brain mechanisms mediate two strategies in arithmetic: evidence from event-related brain potentials. Neuropsychologia, 41: 855 862, 2003. ` Galfano G, Rusconi E, and Umilta C. Automatic activation of multiplication facts: evidence from the nodes adjacent to the product. Quarterly Journal of Experimental Psychology, 56A: 3161, 2003. ` Galfano G, Mazza V, Angrilli A, and Umilta C. Electrophysiological correlates of stimulus-driven multiplication facts retrieval. Neuropsychologia, 42: 13701382, 2004. Hecht SA. Individual solution processes while solving addition and multiplication math facts in adults. Memory and Cognition, 27: 10971107, 1999. Huynh H and Feldt LS. Conditions under which mean square ratios in repeated measurements designs have exact Fdistribution. Journal of the American Statistical Association, 65: 15821589, 1970. Jost K, Henninghausen E, and Rosler F. Comparing arithmetic and semantic fact retrieval: effects of problem size and sentence constraint on event-related brain potentials. Psychophysiology, 41: 4659, 2004a. Jost K, Beinhoff U, Henninghausen E, and Rosler F. Facts, rules, and strategies in single-digit multiplication: evidence from event-related brain potentials. Cognitive Brain Research, 20: 183 193, 2004b. Kiefer M and Brendel D. Attentional modulation of unconscious automatic processes: evidence from event-related potentials in a masked priming paradigm. Journal of Cognitive Neuroscience, 18: 184198, 2006. Kreher DA, Holcomb PJ, and Kuperberg GR. An electrophysiological investigation of indirect semantic priming. Psychophysiology, 43: 550563, 2006. Krueger LE. Why 2 2 5 looks so wrong: on the odd-even rule in product verication. Memory and Cognition, 14: 141149, 1986. Kutas M and Hillyard SA. Reading senseless sentences: brain potentials reect semantic incongruity. Science, 207: 203205, 1980. Lemaire P and Reder L. What affects strategy selection in mental arithmetic? An example of parity and ve effects on product verication. Memory and Cognition, 27: 364382, 1999. LeFevre J-A and Kulak AG. Individual differences in the obligatory activation of addition facts. Memory and Cognition, 22: 188200, 1994. LeFevre J-A, Bisanz J, and Mrkonjic L. Cognitive arithmetic: evidence for obligatory activation of arithmetic facts. Memory and Cognition, 16: 4553, 1988. LeFevre J-A, Kulak AG, and Bisanz J. Individual differences and developmental change in the associative relations among numbers. Journal of Experimental Child Psychology, 52: 256274, 1994. LeFevre J-A, Bisanz J, Daley KE, Buffone L, Greenham SL, and Sadesky GS. Multiple routes to solution of single-digit multiplication problems. Journal of Experimental Psychology: General, 125: 216230, 1996. Martn-Loeches M, Casado P, Gonzalo R, De Heras L, and Fernandez-Fras C. Brain-potentials to mathematical syntax problems. Psychophysiology, 43: 579591, 2006. McCloskey M. Cognitive mechanisms in numerical processing: evidence from acquired dyscalculia. Cognition, 44: 107157, 1992. McCloskey M, Aliminosa D, and Sokol S. Facts, rules, ad procedures in normal calculation: evidence from multiple single-patient studies of impaired arithmetic fact retrieval. Brain and Cognition, 17: 154203, 1991.

Niedeggen M and Rosler F. N400 effects reect activation spread during retrieval of arithmetic facts. Psychological Science, 10: 271 276, 1999. Niedeggen M, Rosler F, and Jost K. Processing of incongruous mental calculation problems: evidence for an arithmetic N400 effect. Psychophysiology, 36: 307324, 1999. Nunez-Pena MI and Escera C. An event-related brain potential study of the arithmetic split effect. International Journal of Psychophysiology, 64: 164173, 2007. Nunez-Pena MI, Cortinas M, and Escera C. Problem size effect and processing strategies in mental arithmetic. NeuroReport, 17: 357360, 2006. Oostenveld R and Praamstra P. The ve percent electrode system for high-resolution EEG and ERP measurements. Clinical Neurophysiology, 112: 713719, 2001. Penolazzi B, Hauk O, and Pulvermuller F. Early semantic context integration and lexical access as revealed by event-related brain potentials. Biological Psychology, 74: 374388, 2007. Rickard TC. A revised identical elements model of arithmetic facts representation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31: 250257, 2005. Rolke B, Heil M, Streb J, and Henninghausen E. Missed prime words within the attentional blink evoke an N400 semantic priming effect. Psychophysiology, 38: 165174, 2001. ` Rusconi E, Galfano G, Speriani V, and Umilta C. Capacity and contextual constraints on product activation: evidence form task-irrelevant fact retrieval. Quarterly Journal of Experimental Psychology, 57A: 14851511, 2004. ` Rusconi E, Galfano G, Rebonato E, and Umilta C. Bidirectional links in the network of multiplication facts. Psychological Research, 70: 3242, 2006a. ` Rusconi E, Priftis K, Rusconi ML, and Umilta C. Arithmetic priming from neglected numbers. Cognitive Neuropsychology, 23: 227 239, 2006b. Spironelli C and Angrilli A. Inuence of phonological, semantic and orthographic tasks on the early linguistic components N150 and N350. International Journal of Psychophysiology, 64: 190198, 2007. Sternberg S. Memory-scanning: mental processes revealed by reaction-time experiments. American Scientist, 57: 421457, 1969. Szucs D and Csepe V. The effect of numerical distance and stimulus probability on ERP components elicited by numerical incongruencies in mental addition. Cognitive Brain Research, 22: 289300, 2005. Thibodeau MH, LeFevre J-A, and Bisanz J. The extension of the interference effect to multiplication. Canadian Journal of Experimental Psychology, 50: 393396, 1996. Verguts T and Fias W. Interacting neighbors: a connectionist model of retrieval in single-digit multiplication. Memory and Cognition, 33: 116, 2005. Wang Y, Kong J, Tang X, Zhuang D, and Li S. Event-related potential N270 is elicited by mental conict processing in human brain. Neuroscience Letters, 293: 1720, 2000. Weisbrod M, Kiefer M, Winkler S, Maier S, Hill H, Roesch-Ely D, et al. Electrophysiological correlates of direct versus indirect semantic priming in normal volunteers. Cognitive Brain Research, 8: 289298, 1999. Zamarian L, Karner E, Benke T, Donnemiller E, and Delazer M. Knowing 7 8, but not the meaning of elephant: evidence for the dissociation between numerical and non-numerical semantic knowledge. Neuropsychologia, 44: 17081723, 2006. Zamarian L, Stadelmann E, Nurk H-C, Gamboz N, Marksteiner J, and Delazer M. Effects of age and mild cognitive impairment on direct and indirect access to arithmetic knowledge. Neuropsychologia, 45: 15111521, 2007. Zhou X, Chen C, Zhang H, Chen C, Zhou R, and Dong Q. The operand-order effect in single-digit multiplication. Neuroscience Letters, 414: 4144, 2007.