Neurophysiological markers of retrieval-induced forgetting in multiplication fact retrieval

Neurophysiological markers of retrieval-induced

forgetting in multiplication fact retrieval

GIOVANNI GALFANO,a BARBARA PENOLAZZI,b FRANCESCA FARDO,c

ELISAH DHOOGE,d ALESSANDRO ANGRILLI,c and CARLO UMILTAc

aDipartimento di Psicologia dello Sviluppo e della Socializzazione, University of Padova, Padova, ItalybDipartimento di Psicologia, University of Bologna, Bologna, ItalycDipartimento di Psicologia Generale, University of Padova, Padova, ItalydDepartment of Experimental Psychology, Ghent University, Ghent, Belgium

Abstract

Event-related potential (ERP) counterparts of practice effects in multiplication fact retrieval were examined. Par-

ticipants performed a multiplication verification task after having practiced a specific problem set. Practice was either

active (retrieval of solutions to multiplication problems) or passive (reexposure to the same operands plus the correct

result). Behavioral data showed retrieval-induced facilitation for practiced items and retrieval-induced forgetting for

related, unpracticed items, irrespective of practice type. ERPs revealed that, for the active practice group, forgetting

was reflected in a reduced N100 component time-locked to result onset. Irrespective of practice type, forgetting was

also reflected in a reduced result-locked P350 component, whereas facilitation was associated with an increased

amplitude of the same component. These results suggest that beneficial and detrimental effects of practice may be

mediated by partially distinct processes.

Descriptors: EEG/ERP, Retrieval induced forgetting, Cognitive arithmetic, Memory control

Remembering is not always beneficial to memory performance.

It is now well known that retrieval practice can give rise to a

consistent form of episodic forgetting (e.g., Bauml, Pastotter, &

Hanslmayr, 2010; Levy & Anderson, 2002). Retrieval-induced

forgetting (RIF) refers to the observation that retrievingmaterial

can hamper the subsequent retrieval of related, unpracticed ma-

terial (Anderson, Bjork, & Bjork, 1994). This phenomenon was

originally studied with the so-called retrieval practice paradigm,

in which participants first learn several category-exemplar pairs

from several categories and then perform retrieval practice for

some of the studied exemplars of some categories only. On a final

test phase, all learned exemplars are tested in a surprise recall test.

Items on this final test are divided into three critical types, prac-

ticed items from practiced categories (often referred to as

Rp1 items), unpracticed items from practiced categories (re-

ferred to as Rp� ), and unpracticed items from unpracticed cat-

egories (i.e., neutral items, referred to as Nrp). Typically, Rp1

items are recalled better than Nrp items, an effect termed re-

trieval-induced facilitation. More important, however, is the

finding that Nrp items are recalled better than Rp� items, i.e.,

the RIF effect. Hence, retrieval practice is beneficial as regards

Rp1 items, but detrimental as regards Rp� items.

RIF has generally been interpreted as reflecting two possible

mechanisms. According to the inhibitory account (e.g., Ander-

son et al., 1994; Bauml et al., 2010; Levy & Anderson, 2002),

retrieval practice of some items (Rp1) elicits active inhibition of

related memory items (Rp� ) aimed to decrease retrieval com-

petition. In contrast, according to the associative interference

account (e.g., Camp, Pecher, & Schmidt, 2007; Williams &

Zacks, 2001), retrieval practice results in strengthening associ-

ations between categories and practiced items that, in turn,

blocks or weakens access to related, unpracticed competitors

(Raaijmakers & Shiffrin, 1981; Rundus, 1973). Recent neuro-

imaging studies have provided consistent evidence that neural

mechanisms mediating RIF operate during retrieval practice and

not at test, which is inconsistent with the associative interference

account (e.g., Johansson, Aslan, Bauml, Gabel, & Mecklinger,

2007; Kuhl, Dudukovic, Kahn, &Wagner, 2007; Wimber, Ruts-

chmann, Greenlee, & Bauml, 2009). Additional evidence sup-

porting the inhibitory account comes from the observation that

retrieval-induced facilitation and RIF result from processes that

are dissociable at both the behavioral (e.g., Bauml & Ku-

hbandner, 2007; Roman, Soriano, Gomez-Ariza, & Bajo, 2009)

and the neural level (e.g., Kuhl et al., 2007; Wimber et al., 2009)

and that RIF can be observed using recognition tests (e.g., Hicks

& Starns, 2004; Spitzer & Bauml, 2007).

RIF has proved a robust effect, and has been replicated

with memory tasks other than episodic recall, including recall of

This researchwas supported by a grant from theUniversity of Padova

to Giovanni Galfano and by grants fromMIUR to Carlo Umilta and to

Alessandro Angrilli. We thank Cyma Van Petten and three anonymous

reviewers for their constructive criticisms.Address correspondence to: Giovanni Galfano, Dipartimento di Psi-

cologia dello Sviluppo e della Socializzazione, Universita di Padova, ViaVenezia, 8, I-35131, Padova, Italy. E-mail: [email protected]

Psychophysiology, 48 (2011), 1681–1691. Wiley Periodicals, Inc. Printed in the USA.Copyrightr 2011 Society for Psychophysiological ResearchDOI: 10.1111/j.1469-8986.2011.01267.x

1681

semantic knowledge (Johnson&Anderson, 2004). An important

step to show that RIF is a general phenomenon is concernedwith

the demonstration that it can take place in the context of different

memory systems. Phenix and Campbell (2004; also see Campbell

& Timm, 2000) have provided the first tentative evidence that

RIF can be observed in the semantic memory of arithmetic facts

(i.e., arithmetic operations involving single-digit operands). They

focused on operandmultiple representations and used amodified

practice-test multiplication paradigm (Manly & Spoehr, 1999).

Unlike the classic retrieval practice paradigm, there was no study

phase, as arithmetic facts constitute overlearned associations in

healthy adults. Participants underwent a practice phase in which

they were asked to solve a subset of simple multiplication prob-

lems. In the subsequent test phase, they were to complete a mul-

tiplication verification task in which they were shown both true

and false multiplication equations. Critically, these belonged to

three distinct sets of stimuli. Stimuli from the zero-operand set

consisted of multiplications in which neither of the operands had

appeared in the practice phase (Nrp). Stimuli belonging to the

one-operand set were multiplications including one of the oper-

ands that had appeared in the practice phase (Rp� ). Finally, the

two-operand set consisted of the same stimuli tested in the prac-

tice phase (Rp1). Phenix and Campbell showed that perfor-

mance on multiplications belonging to the one-operand set was

significantly worse than onmultiplications belonging to the zero-

operand set. Because correct answers to multiplications belong-

ing to the one-operand set were always multiples of practiced

operands, Phenix and Campbell interpreted this pattern as ev-

idence that actively retrieving solutions for problems presented in

the practice phase (e.g., 7 � 25 ?) impaired activation of the

multiples (i.e., one-operand problems, e.g., 6 � 25 12). Thus,

consistent with RIF, one-operand items in the test phase under-

went poorer performance compared to problems composed of

unpracticed operands (i.e., zero-operand problems, e.g., 9 �

35 27), whose results were unaffected by retrieving solutions to

problems shown in the practice phase. This result is important in

the optic of clarifying memory mechanisms operating in the lex-

icon of arithmetic facts. In this regard, abundant neuropsycho-

logical evidence is available showing that arithmetic facts

memory is dissociated from standard semantic memory (e.g.,

Cappelletti, Kopelman, Morton, & Butterworth, 2005; Zamar-

ian, Karner, Benke, Donnemiller, & Delazer, 2006). In a similar

fashion, unlike word representation, access to number represen-

tations (at least in addition tasks) has been shown to be depen-

dent on the surface format of the stimuli (Szucs & Csepe, 2004).

Despite these arguments suggesting a functional dissociation be-

tween the two long-term memory stores, a ‘‘lexicon’’ for arith-

metic facts is often modeled in terms of semantic memory

processes (Ashcraft, 1992; Campbell, 1995; Verguts & Fias,

2005). Hence, the presence ofRIF-like phenomena for arithmetic

facts is important also as evidence supporting the view that sim-

ilar functional mechanisms operate in both semantic and arith-

metic facts memory (e.g., Galfano, Mazza, Angrilli, & Umilta,

2004; Jost, Henninghausen, & Rosler, 2004; Niedeggen, Rosler,

& Jost, 1999; Rusconi, Galfano, Rebonato, & Umilta, 2006;

Szucs & Soltesz, 2010; Szucs, Soltesz, Czigler, & Csepe, 2007).

The purpose of the present study was to address RIF in mul-

tiplication fact retrieval at test by using event-related brain po-

tentials (ERPs) combined with the paradigm used by Phenix and

Campbell (2004). To date, only one study has used ERPs to

clarify the possible brain mechanisms underlying RIF by focus-

ing on the effects of retrieval practice on the final memory test.

Spitzer, Hanslmayr, Opitz, Mecklinger, and Bauml (2009) ex-

amined the electrophysiological correlates of RIF on recognition

memorywith the standard retrieval practice paradigmand showed

that, compared to control material from unpracticed categories

(Nrp), unpracticed items from practiced categories (Rp� ) were

characterized by poorer recognition performance (i.e., RIF)

and reduced amplitudes of the frontal P2 ERP component. In

addition, compared to Nrp items, practiced items from practiced

categories (Rp1) were associated with better recognition perfor-

mance (i.e., retrieval-induced facilitation) and a larger left parietal

positivity starting 500ms from stimulus onset. Spitzer et al. (2009)

interpreted the observation that retrieval-induced facilitation and

RIF were reflected in qualitatively distinct ERP components as

strong evidence that retrieval practice results in separable pro-

cesses: Whereas RIF would be caused by active inhibition of the

items’ representation, retrieval-induced facilitation would be in-

duced by the strengthening of associative links between items and

contextual information (Spitzer et al., 2009).

Because the present study dealt with semantic memory and

overlearned preexisting associations among items, the initial study

phase of the retrieval practice paradigm was eliminated (also see

Johnson & Anderson, 2004), and the experimental procedure

consisted of two successive phases, a practice and a test phase.

Stimuli were single-digit multiplication problems. The practice

phase comprised two different practice activities. Half of the par-

ticipants were assigned to an active practice condition in that they

were shown a problem and were explicitly asked to retrieve and

then speak the correct result aloud. The other half of the partic-

ipants were assigned to a passive practice condition in that they

were shown the same problem, followed after a blank frame by its

correct result, and were asked to relearn the problem and read the

result aloud. In accordance with Phenix and Campbell (2004),

the test phase consisted of a multiplication verification task, and

the stimuli belonged to one out of three possible categories. Stim-

uli in the two-operand set were exactly the same stimuli presented

in the practice phase (Rp1). Stimuli in the one-operand set (Rp� )

included one practiced operand (so that the result was a multiple

of a practiced problem). Finally, stimuli in the zero-operand set

included operands that were both unpracticed (Nrp).

As regards predictions at the behavioral level, for the active

practice group, we expected to replicate the RIF pattern (better

performance for multiplications belonging to the zero-operand

set compared to the one-operand set) obtained by Phenix and

Campbell (2004). On the electrophysiological side, we aimed to

ascertain whether zero-, one-, and two-operand set conditions

were able to elicit distinct ERP patterns mirroring behavioral

results. In particular, we looked at possible amplitude differences

as a function of operand set.

As for the passive practice group, it could be argued that,

following the standard inhibitory account advocated for RIF in

episodic memory studies (see Bauml, 2008, for a discussion), no

RIF should occur following the study practice phase due to the

fact that passive practice (i.e., without the explicit requirement of

active recall) does not elicit competition from related items,

which, in turn, are not being inhibited during the practice phase

(Anderson, Bjork, & Bjork, 2000; Johansson et al., 2007; Wim-

ber et al., 2009). However, other arguments may lead to the

prediction to observe a significant RIF effect also when explicit

retrieval is not required during practice. First of all, in the present

study, passive practice consisted of the presentation of the equa-

tion in two separate steps, with the onset of the problem first

(e.g., 2 � 5), followed by presentation of its correct result (i.e.,

1682 G. Galfano et al.

10). This sequential presentation of itemswas deliberately used to

trigger involuntary retrieval. In addition, strong evidence has been

provided showing that the mere presentation of two single-digit

numbers activates both their product and the neighboring mul-

tiples in the multiplication network even when arithmetic is irrel-

evant for the task at hand (Galfano, Penolazzi, Vervaeck, Angrilli,

& Umilta, 2009; Galfano, Rusconi, & Umilta, 2003; Zamarian et

al., 2007). These results have been interpreted as evidence that

memory retrieval of multiplication facts takes place without in-

tention, irrespective of task requirements. In light of these argu-

ments, the lack of behavioral and/or electrophysiological

differences between the active and the passive practice groups

might still be consistent with the inhibitory account of RIF.

As regards the viability of the accounts advocated for RIF, it is

also important to note that in the present context the sequential

presentation of the problem followed by the result in the verifi-

cation task administered in the test phase may trigger additional

recall processes. This means that, unlike studies in which standard

recognition paradigms are used at test (e.g., Spitzer et al., 2009),

inhibitory effects of retrieval practice in the present study may be

accompanied by interference-related processes like blocking.

To summarize, the purpose of the present study was to extend

RIF to the arithmetic domain, showing that RIF is a general

phenomenon. In addition, we aimed to test whether RIF in the

arithmetic domain is dependent of whether retrieval during the

practice phase is performed explicitly (as in the active practice

group) or implicitly (as in the passive practice group). Finally, we

aimed to check the plausibility, within the arithmetic memory

system, of the mechanisms assumed to cause RIF in episodic

memory and within the linguistic domain.

Methods

Participants

A total of 32 undergraduate students (20 women, age range

19–36 years) volunteered to participate in the experiment. All

participants gave their written informed consent prior to their

inclusion in the study. They 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.

Two participants were left-handed. The study was conducted in

accordance with the guidelines laid down in the Declaration of

Helsinki.

Material and Stimuli

In keeping with Phenix and Campbell (2004; also see Manly &

Spoehr, 1999), stimuli consisted of three sets of eight multipli-

cation problems each (i.e., operand sets). Each operand set was

based on two groups of digits. The first group included the digits

2, 5, 7, and 8 and the second group consisted of the digits 3, 4, 6,

and 9 (see Tables A.1 and A.2, respectively, in the Appendix).

Multiplication problems of the first operand set were created by

combining the digits in the first group (e.g., 2 � 5). Multiplica-

tion problems of the second operand set were constructed out of

the digits in the second group (e.g., 3 � 6). Five problems and

tieswere also included in order to obtain an acceptable number of

stimuli. One or the other of these operand sets was used in the

practice phase of the experiment. The operand set not used in the

practice phase served as a baseline. In line with Phenix and

Campbell (2004), these operand sets were referred to as the two-

operand set (i.e., both the operands occurred in the practice phase)

and the zero-operand set (i.e., none of the operands occurred in the

practice phase), respectively.Which operand set served as the two-

operand set was counterbalanced across participants. Multiplica-

tion problems belonging to a third operand set were obtained by

combining one digit of the first group with one digit of the second

group (e.g., 2 � 6). Thus, this operand set always had one op-

erand that was used in the practice phase andwas referred to as the

one-operand set. Hence, stimuli from both the zero- and the one-

operand sets were never shown in the practice phase.

For the active practice group, the practice phase consisted of a

production task with the stimuli of the two-operand set. Partic-

ipants were shown a multiplication problem (e.g., 2 � 55 ?) and

had to retrieve and speak the correct result aloud. For the passive

practice group, practice consisted of relearning the stimuli of the

two-operand set presented in the form of equations (e.g., 2 �

55 10) and reading the result aloud. Only correct equations were

shown. Both groups were then submitted to the same test phase,

in which they had to complete a multiplication verification task

(e.g., 2 � 55 10, TRUE or FALSE?). Half of the problems in

the test phase had a correct result, and the other half had a false

result (i.e., a lure). Although we used the same multiplication

problems as Phenix and Campbell (2004), selection of lures was

different in several ways, because we were not interested in

checking RIF in false problems. Therefore, our lures were not

divided into different types. This simplified the experimental de-

sign and allowed us to have an adequate number of trials in the

relevant conditions for the ERP analyses. In addition, Phenix

and Campbell (2004) did not match the parity status of the cor-

rect solution and the lure. This might lead to the use of different

strategies for different equations (Masse & Lemaire, 2001). The

selection of lures in the present study was done for each operand

set and problem according to the following steps. First, all mul-

tiples one and two positions above and below from the correct

solution were selected resulting in eight possible lures (e.g., 2 �

55 10 resulted in the selection of 15, 12, 5, and 8 and 20, 14, 0,

and 6 for one and two positions up and down in the table, re-

spectively). Multiples two positions above or below the correct

solutions had to be included to have lures for each equation. In a

subsequent step, all correct solutions belonging to the two-op-

erand set and the operand set in question were removed. In this

regard, it is necessary to note that which set was the two-operand

set was counterbalanced across participants, which resulted in

lures that were different depending on which operand set served

as the two-operand set.

Then, all solutions smaller than 10 and all solutions that

differed in parity from the correct solution were removed from

the list, in order to have only two-digit lures that matched the

correct solution as regards the parity status. Within these lists,

lures were selected so that, within each operand set, four of the

selected lures were one position above or below the correct so-

lution in one of the operand’s multiplication table (e.g., 30 and

20, for 5 � 55 25) and four of the selected lures were two po-

sitions above or below the correct answer in one of the operand’s

multiplication table (e.g., 63 and 35, for 7 � 75 49), thereby

creating two subsets. In addition, for each subset, lures were

selected so that two of the selected lures were smaller than

the correct solution and two of the selected lures were larger than

the correct solution. When there was more than one possibility,

the lure with the smallest distance to the correct solution was

ERPs and retrieval-induced forgetting 1683

chosen. This was done to avoid equations that were too

obviously false due to a split effect (e.g., Szucs & Csepe, 2005).

When possible, no lure was used more than once within each

operand set. Following Phenix and Campbell (2004), lures could

be reused between operand sets. For each operand set, this led to

the selection of eight different lures either one or two positions

above (i.e., larger) or one or two positions below (i.e., smaller)

the correct result of a given multiplication problem.

The restriction concerning the number of lures that were

multiples one and two positions above or below the correct so-

lution was met in all but the one-operand set (i.e., the zero-

operand set of the 3, 4, 6, and 9 numeral group), and so one extra

lure was chosen out of the set of possible lures created by the

multiples two positions above or below the correct solution in

one of the operands multiplication table. Because it was not

possible to apply this restriction to this operand set, the restric-

tion concerning the number of lures being smaller or larger than

the correct solution was not met either, so one extra lure was

chosen out of the set of possible lures created by themultiples two

positions above the correct result in one of the operands mul-

tiplication table.

In the practice phase, each participant performed 10 blocks of

32 trials each. Each block consisted of four repetitions of each

multiplication problem in the two-operand set. This resulted in

320 trials in total. In the test phase, participants performed

8 blocks of 48 trials each. In each block, each multiplication

problem for each operand set was presented with both a correct

result and a lure, which led to a total of 384 trials for the test

phase. As a result, there were 128 trials for each operand set, half

with a correct result and half with a lure.

Procedure

The experiment took place in a sound-attenuated, electrically

shielded, dimly lit room. Stimuli appeared in white against a

black background and were displayed centered on a Sony 19-in.

color monitor (640 � 480, 75 Hz) placed about 50 cm in front of

the participants. Each character measured 8 mm (0.921 of visual

angle) in height and 5 mm (0.571) in width. The distance between

each character was 9mm.Half of the participants were randomly

assigned to the passive practice condition, the other half to the

active practice condition. In both conditions, one set of multi-

plications was used (i.e., the two-operand set, see the Appendix).

Participants were informed that the experiment involved taking

part in a multiplication verification task test only after they

completed the practice phase.

Figure 1 illustrates the sequence of events in a trial for the

practice phase for the two groups (Panel a), and the sequence of

events in a trial for the test phase (Panel b). Each trial in both

phases began with a 400-ms fixation point (‘‘#’’), accompanied

by a 500-Hz warning tone lasting for 100ms. After fixation point

offset, the multiplication problem (e.g., 2 � 55 ) was shown for

350 ms. In the practice phase, after 150 ms during which a blank

frame was visible, either a question mark (‘‘?’’) or the correct

result (10) was displayed, depending on the assigned group, for

1600 ms. A 100-ms 1000-Hz tone (go signal) followed, after

which participants had 300 ms for either reading the result aloud

(passive practice group) or retrieving and speaking the correct

result aloud (active practice group). Data of the practice phase

for the active practice group were neither recorded nor analyzed,

as the occurrence of RIF in the later test phase does not depend

on whether retrieval attempts have been successful (Storm,

Bjork, Bjork, & Nestojko, 2006).

The delay between the onset of either the result or the question

mark and the acoustic go signal was used to equate the two

practice conditions as much as possible. The intertrial interval

was 400 ms. As concerns the test phase, all events proceeded in

the same way as in the practice phase until the blank frame. This

was always followed by the onset of the result for 1900 ms. The

sequential presentation of the operands and the result had the

purpose ofminimizing the possibility of eye movements (Szucs &

Csepe, 2005). The task required deciding whether the result was

correct or not by pressing one of two allowed keys of the com-

puter keyboard. Half of the participants pressed the ‘‘P’’ key for

problems with correct results and the ‘‘Q’’ key for problems with

incorrect results. The other half did the opposite. Reaction times

(RTs) were recorded from result onset. The intertrial interval in

the test phase was 800 ms. Acoustic feedback on accuracy was

provided only after wrong key presses. The instructions empha-

sized both speed and accuracy. The participants were instructed

to move as little as possible except for responding and were en-

couraged to postpone blinks after the response.

Participants were allowed short breaks between blocks of tri-

als. In line with Phenix and Campbell (2004), no training trials

were administered before either the practice or the test phase. The

order of trials was randomized separately for each participant.

The experiment, including electrode placement and execution of

the practice and the test phases, required about 120 min. An-

alyses of the test phase focused on trials in which true equations

(i.e., multiplication problems coupled with correct results) were

shown. This was motivated by the observation that false equa-

tions in verification tasks are often dealt with by processing

strategies other than retrieval (e.g., Campbell & Tarling, 1996;

Masse & Lemaire, 2001). Statistical analysis of RT data was

performed after eliminating trials with erroneous responses. RTs

for correct responses were submitted to an analysis of variance

(ANOVA) with operand set (zero-operand vs. one-operand vs.

two-operand set) as a within-participants factor and group (pas-

sive practice vs. active practice) as a between-participants factor.

A second ANOVA was performed on proportion of correct re-

sponses with the same factors as the ANOVA on RTs.

Electroencephalogram (EEG) Recording and Analysis

Electrophysiological data were recorded from 19 scalp electrodes

(Fp1, Fp2, F3, Fz, F4, F7, F8, C3, Cz, C4, P3, Pz, P4, T3, T4,

T5, T6, O1, and O2) located at standard positions according to

the International 10–20 System (Jasper, 1958), two electrodes

placed on the mastoids (M1, M2), one electrode placed on the

nasion, and four electrodes placed around the orbital regions to

measure eye movements. Cz was used as the recording reference

for all channels; then data were converted off-line to an averaged

mastoid reference ((M11M2)/2). EEG data were recorded con-

tinuously with a SynAmp system (NeuroScan Labs, Sterling,

VA), bandwidth ranged from DC to 100 Hz (6 dB/octave), am-

plitude resolution was 0.168 mV/bin, and sampling rate was set

at 500 Hz. Electrode impedance was kept below 5 kO at all scalp

locations and below 10 kO at ocular electrodes. Signal analyses

were performed by means of the Brain Vision Analyzer system,

version 1.05 (Brain Products GmbH, Germany). Vertical and

horizontal eye movements and blinks were corrected by applying

the Independent Component Analysis transformation to the

1684 G. Galfano et al.

EEG signal. Next, raw data were segmented in epochs of the

same length (� 1000 to 2000 ms), time-locked to the onset of

multiplication operands. A detrending procedure in this long

epoch was applied to minimize residual slow trends in the re-

corded signal; then epochs were further segmented in the interval

� 700 to 1000 ms. Artifact rejection was applied to all trials with

correct behavioral responses (artifacts threshold level was set to

150 mV; minimum and maximum allowed amplitude: � 75/175

mV, respectively). The remaining epochs were further visually

inspected to remove any possible residual artifact, leading to an

average trial discard of 23%. The accepted epochs were then

averaged for each of the relevant experimental conditions (i.e.,

group and operand set), corrected for baseline activity using the

mean voltage in the 200 ms preceding the onset of the operands,

and finally low-pass filtered (20 Hz, 24 dB/oct.).

Frontal (F3, Fz, F4), central (C3, Cz, C4), and parietal sites

(P3, Pz, P4) were included in the statistical analyses. We focused

on these electrode sites because in similar paradigms used pre-

viously these are the electrodes that revealed the components that

are typically elicited by arithmetic stimuli (e.g., Galfano et al.,

2004; Jost et al., 2004; Nunez-Pena, Cortinas, & Escera, 2006;

Pauli et al., 1994; Szucs & Csepe, 2004). Grandmean waveforms

elicited by the multiplication operands and correct result in the

three experimental conditions (two-operand vs. one-operand vs.

zero-operand set) are illustrated in Figure 2, showing ERPs for

the active practice (Panel a) and the passive practice (Panel b)

groups. Figure 3 illustrates grand-average waveforms in the test

phase for the active practice group and the passive practice group

as a function of operand set at the Cz recording site. On the basis

of visual inspection of waveforms, statistical analyses were fo-

cused on three main components time-locked to the onset of the

result: a negative fronto-central N100 (computed in the 80–120-

ms time interval), a positive fronto-central P250 (computed in the

200–300-ms time interval), and a centro-parietal P350 (com-

puted in the 300–400-ms time interval). In addition, a positive

component was also identified after onset of the multiplication

operands peaking at about 200 ms. This was defined as the most

positive peak between 150 and 250 ms. Because ERP compo-

nents elicited by the result seem to show early differences between

conditions that might reflect previous ongoing processes elicited

by the operands, a control analysis was performed on this com-

ponent for establishing whether differences between relevant ex-

perimental conditions may be elicited already by the

multiplication operands.

Separately for each selected timewindow (i.e., 150–250ms for

operand-locked P200, 80–120 ms for result-locked N100, 200–

300 ms for result-locked P250, and 300–400 for result-locked

P350), mean amplitude values for correct results were submitted

to anANOVAwith group (passive practice vs. active practice) as

the between-participants factor and operand set (two-operand

vs. one-operand vs. zero-operand set), caudality (anterior vs.

central vs. posterior electrodes), and laterality (left hemisphere

vs. midline vs. right hemisphere electrodes) as within-participants

factors. The Greenhouse–Geisser correction was applied when

sphericity assumptions were violated, and, in these cases, the

uncorrected degrees of freedom, e values, and the corrected

probability levels are reported. Post hoc comparisons were com-

puted using the Tukey HSD test settled with po.05. Significant

results not involving the relevant variables for evaluating RIF

and facilitation (group and operand set) are not reported.

As a final analysis, aimed at evaluating whether detrimental

and beneficial effects of retrieval practice are reflected in selective

modulations of distinct ERP components, we carried out a cor-

relation analysis between behavioral and result-locked ERP

components in order to better clarify the functional meaning of

the components in the present study and verify the possibility

that retrieval-induced facilitation and RIF reflect different pro-

cesses. In detail, we combined RTs and accuracy (proportion of

correct responses) into a single behavioral measure, named

inverse efficiency score (IE5RT/proportion correct).1 Next,

ERPs and retrieval-induced forgetting 1685

Time

Fixation (400 ms)

Operands (350 ms)

Blank (150 ms)

Fixation (400 ms)

Operands (350 ms)

Blank (150 ms)

Result (1900 ms)

Result (1600 ms)

Acoustic go signal (100 ms)

+ Response (300 ms)

Passive practice Group:

relearn the result

Active practice Group:

retrieve the result

Panel A Panel B

#

2 x 5 =

#

2 x 5 =

10

10

10

?

?

Time

Figure 1. Sequence of events in the practice and the test phases. In the practice phase (a), two operands were shown on each trial to both the passive

practice and the active practice groups. Participants in the passive practice group were then presented with the correct result and had to read it aloud and

relearn the problem. Participants in the active practice group were shown a question mark instead of the correct result and had to retrieve the result and

speak it aloud. In the test phase (b), both groups were to perform a standard multiplication verification task, with false and correct problems occurring

with the same frequency. A trial with correct result is illustrated. The stimuli are not drawn to scale.

1Inverse efficiency scores are typically used either to lower the impactof a speed–accuracy trade-off or when there is the need to refer to a singleperformance index to simplify data analysis (Townsend & Ashby, 1983).Lower scores in this measure index a better performance, in the samefashion as RTs (see, e.g., Galfano & Pavani, 2005).

indices of retrieval-induced facilitation and RIF were computed

according to the following formulas: IEtwo� operand� IEzero�operand

for retrieval-induced facilitation and IEzero�operand� IEone�operand

for RIF. More negative values in these indices reflect stronger

retrieval-induced facilitation and RIF, respectively. On the elect-

rophysiological side, waveform differences were computed by

subtractingmean amplitudes of the zero-operand condition from

mean amplitudes of the two-operand condition to index re-

trieval-induced facilitation and by subtracting mean amplitudes

of the one-operand condition frommean amplitudes of the zero-

operand condition to index RIF. Then, Spearman rank corre-

lation analyses were performed between behavioral and ERP

indexes associated with retrieval-induced facilitation and be-

tween behavioral and ERP indices associated with RIF. To avoid

1686 G. Galfano et al.

Figure 2.Electrophysiological results in the test phase for the active practice group (a) and the passive practice group (b). Grand-averagewaveformswere

recorded at F3, Fz, F4, C3, Cz, C4, P3, Pz, and P4 sites, from top to bottom. 0 ms refers to operand onset, 500 ms refers to result onset. The rectangles

superimposed on the ERP waveforms indicate the time windows included in the analyses.

inflation of significant results due to multiple comparisons, we

only considered a result as important if two or more correlations

were significant in adjacent electrodes. Correlation analyses were

performed based on the outcomes of the ANOVA. Specifically,

whenever a significant interaction involving group as factor

emerged, separate correlation analyses were performed. The a

level was adjusted for the number of comparisons according to

the false discovery rate (i.e., the expected proportion of the re-

jected null hypotheses that are erroneously rejected; Benjamini &

Hochberg, 1995).

Results

Behavioral Data

The ANOVA on RTs for correct responses revealed a significant

main effect of operand set, F(2,60)5 21.104, e5 .88, Zp25 .41,

p5 .001. Post hoc comparisons showed that participants verified

a correct result belonging to the two-operand set (M5 639 ms,

SD5 204) marginally faster (p5 .06) than a correct result be-

longing to the zero-operand set (M5 675 ms, SD5 161), in line

with a retrieval-induced facilitation effect. In addition, consistent

with RIF, participants took significantly longer to verify a cor-

rect result belonging to the one-operand set (M5 754 ms,

SD5 224) compared to the zero-operand set. Critically, no other

sources of variance were significant (lowest p5 .23). Hence, the

data showed that RIF and retrieval-induced facilitation were

unaffected by whether participants engaged in either passive or

active practice (see Table 1).

The ANOVA on proportion of correct responses yielded a

significant main effect of operand set, F(2,60)5 17.39, e5 .88,

Zp25 .49, p5 .001. Post hoc comparisons showed that partici-

pants were significantly more accurate when the correct result

belonged to the two-operand set (M5 0.95, SD5 0.03) than to

the zero-operand set (M5 0.92, SD5 0.05), consistent with a

retrieval-induced facilitation effect. In addition, participants

were less accurate in verifying a correct result belonging to the

one-operand set (M5 0.88, SD5 0.08) compared to the zero-

operand set, in line with RIF. There were no overall differences

between groups (p5 .15), but the Operand Set � Group inter-

action fell short of significance, F(2,60)5 2.59, e5 .88, Zp25 .16,

p5 .08. However, as shown in Table 1, this pattern seems to

reflect differences, if any, in retrieval-induced facilitation (and

not in RIF) as a function of group. In sum, the pattern of ac-

curacy data is fully consistent with RT data and makes the pos-

sibility of a speed-accuracy tradeoff unlikely.

ERP Data: Operand-Locked Analyses

P200. The ANOVA revealed a significant main effect of oper-

and set, F(2,60)5 3.45, e5 .99, Zp25 .195, p5 .03. Post hoc

comparisons revealed that amplitude of the P200 componentwas

significantly smaller for operands belonging to the two-operand

set compared to results belonging to the zero-operand set. Am-

plitude for operands of the one-operand set fell in between, al-

though they were not significantly different from either

conditions. Because this pattern was clearly inconsistent with

respect to RIF (see the Discussion section), no correlation an-

alyses were performed at the level of this component.

ERP Data: Result-Locked Analyses

N100. The ANOVA performed on the first result-locked com-

ponent showed a significant main effect of operand set,

F(2,60)5 4.14, e5 .84, Zp25 .121, p5 .03. Post hoc compari-

sons revealed that amplitude of the N100 component was sig-

nificantly higher for results belonging to the two-operand set

compared to results belonging to the one-operand set. Although

not significantly different from the other conditions, amplitude

for results belonging to the zero-operand set fell between those of

the one-operand and the two-operand sets. Importantly, the

Operand Set � Laterality � Group interaction was also signifi-

cant (see Figure 4), F(4,120)5 2.76, e5 .79, Zp25 .084, p5 .04.

Post hoc analyses showed that for the active practice group,

independent of laterality, N100 amplitude was significantly

smaller for results belonging to the one-operand set compared to

ERPs and retrieval-induced forgetting 1687

Figure 3. Grand-average waveforms in the test phase for the active

practice group and the passive practice group as a function of operand set

at the Cz recording site. The rectangles superimposed on the ERP

waveforms indicate the time windows included in the analyses.

Table 1. Mean Values and Standard Deviations (in Parentheses) for Reaction Times (in Milliseconds), Proportion of Correct Responses,

and Inverse Efficiency Scores (IE) as a Function of Group (Passive Practice vs. Active Practice) and Operand Set (Zero-Operand vs. One-

Operand vs. Two-Operand)

Passive practice group Active practice group

RT Proportion correct IE RT Proportion correct IE

0-operand set 712 (172) .93 (.05) 765 (203) 636 (144) .90 (.06) 708 (172)1-operand set 800 (247) .91 (.06) 889 (290) 707 (194) .86 (.09) 827 (231)2-operand set 678 (240) .95 (.05) 714 (251) 601 (159) .96 (.03) 628 (165)

results belonging to the zero-operand set, consistent with a RIF-

like pattern. In contrast, no significant amplitude difference

emerged between results belonging to the two-operand set and

the zero-operand set, so that no facilitation-like effect was visible.

As concerns the passive practice group, although a RIF-like

pattern was visible over left and midline scalp regions, the am-

plitude difference for results belonging to the one-operand and

the zero-operand sets was not significant. Moreover, unlike the

Active practice group, post hoc analyses revealed a significant

facilitation-like pattern over midline and right scalp regions, re-

sults belonging to the two-operand set being associated with a

bigger N100 than results belonging to the zero-operand set.

Spearman rank correlation analyses, conducted separately for

the two groups, did not reveal any significant effect.

P250. No sources of variance were significant in the ANOVA

performed on the second result-locked component.

P350. The ANOVA performed on the third result-locked com-

ponent revealed a significant main effect of operand set,

F(2,60)5 5.81, e5 .79, Zp25 .162, p5 .005. Post hoc compari-

sons revealed that amplitude of the P350 component was sig-

nificantly smaller for results belonging to the one-operand set

compared to results belonging to both the zero-operand and the

two-operand sets. No significant amplitude differences were ob-

served between results belonging to the zero-operand and the

two-operand set. Operand set did not interact with spatial factors

(lowest p5 .32). No significant effects involving group as a factor

were found (lowest p5 .19). The Spearman rank correlation an-

alyses showed significant negative correlations in F4 (r5 � .44,

po.05) and Fz (r5 � .37, po.05) sites between the magnitude

of RIF and the amplitude difference between the zero- and the

one-operand sets. More specifically, a larger RIF in behavioral

measures was associated with a higher decrease in amplitude for

the one-operand set compared to the zero-operand set. This in-

dicates that participants with stronger RIF in behavioral mea-

sures showed higher amplitudes in the zero-operand condition

compared with the one-operand condition. Although post hoc

analyses did not show significant amplitude differences between

results belonging to the zero-operand and the two-operand sets,

significant negative correlations were observed at F4 (r5 � .38,

po.05), Fz (r5 � .39, po.05), and Cz (r5 � .48, po.05) sites

between the magnitude of behavioral retrieval-induced facilita-

tion and amplitude differences involving zero- and two-operand

sets. This suggests that participants with stronger behavioral re-

trieval-induced facilitation showed larger amplitudes in the two-

operand condition compared with the zero-operand condition.

Discussion

The present study addressed RIF in the context of semantic

memory of multiplication facts. Our first aim was replicating the

findings reported by Phenix and Campbell (2004) to establish the

robustness of RIF in a different memory system from that ex-

plored by classic studies (e.g., Levy & Anderson, 2002) and to

test its viability as a general phenomenon associated with mem-

ory retrieval. In full accordance with Phenix and Campbell,

whose participants performed active retrieval practice, the pres-

ent behavioral results for the active practice group showed that

performance (in terms of both latency and accuracy) in the one-

operand condition was significantly worse than in the zero-

operand condition (indicating a RIF effect), and performance in

the two-operand condition was significantly better than in the

zero-operand condition (indicating a retrieval-induced facilita-

tion). This replication shows that RIF is a general phenomenon,

taking place also in semantic memory and, more specifically, in

the domain of arithmetic facts (also see Campbell & Phenix,

2009; Campbell & Timm, 2000).

The second purpose of the present study was testing whether

RIF in the domain of arithmetic facts can be elicited also when

active retrieval is not explicitly required. To this aim, we sub-

mitted a second group of participants (i.e., the passive practice

group) to a different type of practice, requiring the simple re-

learning of the same subset of items overtly recalled by the active

practice group. This practice type has consistently been shown

not to induce RIF in behavioral measures in classic episodic

memory studies, likely because passive study of items does not

give rise to competition from related material (e.g., Anderson et

al., 2000; Johansson et al., 2007). As anticipated in the Intro-

duction section, however, several arguments suggest that the

context of semanticmemory ofmultiplication factsmay be rather

different from that of semantic knowledge. As a first observation,

the long-term memory stores for general knowledge about the

world (i.e., semantic memory) and arithmetic knowledge of

numbers (i.e., semantic memory of arithmetic facts), although

often modeled according to similar processing mechanisms such

as spreading activation (see, e.g., Ashcraft, 1992), are function-

ally independent, as they can result selectively impaired in neu-

ropsychological patients (e.g., Cappelletti et al., 2005; Zamarian

et al., 2006). In addition, recent electrophysiological evidence has

shown that although both semantic and arithmetic incongruenc-

es elicit modulations at the level of the N400 component of ERPs

(e.g., Galfano et al., 2004, 2009; Jost et al., 2004; Niedeggen et

al., 1999; Niedeggen, &Rosler, 1999), thesemodulations seem to

occur earlier in time in the arithmetic domain (Jost et al., 2004;

Niedeggen et al., 1999). Finally, and most important, consistent

behavioral evidence has been reported showing that the simple

viewing of a digit pair activates the related nodes in the multi-

plication table (i.e., the product and the neighboring multiples).

More specifically, it has been shown that this knowledge is

1688 G. Galfano et al.

Figure 4.Meanvoltages (mV) for the active practice group and the passive

practice group as function of laterality (left scalp region: F3, C3, and P3;

midline scalp region: Fz, Cz, and Pz; right scalp region: F4, C4, and P4)

and operand set (zero-operand, one-operand, and two-operand) in the

analysis of the N100 (80–120 ms) component time-locked to the onset of

the result.

activated involuntarily because it interferes during execution of

tasks in which arithmetic is irrelevant (e.g., De Brauwer & Fias,

2009; Galfano et al., 2003; Rusconi, Galfano, Speriani, &

Umilta, 2004; Zamarian et al., 2006). On the basis of these ar-

guments, we have tested whether RIF would have been observed

even in participants simply asked to study simple multiplication

problems. Indeed, we have hypothesized that passive exposure in

the context of multiplication facts may entail execution of a re-

trieval process even when it is not explicitly required.

Importantly, unlike classic RIF studies, implicit retrieval was

favored in the present experiment by sequential presentation of

the to-be-studied items. Behavioral results were consistent with

this scenario in that participants in the passive practice group

showed evidence of both RIF and retrieval-induced facilitation.

ERP data, however, showed differences that make the over-

all picture much more complex than it appears from behavioral

results.

To ascertain whether differences between experimental con-

ditions emerged before the onset of the result, the first ERP an-

alyses were time-locked to the onset of the operands. A positive

deflection peaking around 200 ms after operand onset was iden-

tified. Independent of group, the amplitude of this operand-

locked P200 was significantly higher for the zero-operand

condition than for the two-operand condition, whereas ampli-

tude for the one-operand set was not significantly different from

amplitude values of either the zero- or the two-operand condi-

tions. This pattern does not appear to be consistent withRIF-like

effects in that one should have expectedmean voltage of the zero-

operand set to fall in between that of the one- and the two-

operand sets, given that, according to RIF, results belonging to

the one-operand set should be inhibited/less activated than those

belonging to the zero-operand set. In light of this reasoning and

because this component seems very early, we tend to exclude the

possibility that it reflects RIF-related processes. By contrast, we

interpret the pattern as reflecting processes related to preactiva-

tion of operands in memory, so that P200 amplitude is inversely

associated with operand activation.

The first result-locked ERP component, the N100, is much

more informative as regards the issue of establishing whether

RIF in the domain of arithmetic facts can be elicited also when

active retrieval is not explicitly required. Unlike behavioral data,

this component revealed a clear dissociation as a function of

practice type. Consistent with a RIF-like pattern, amplitude of

this component was significantly reduced, all over the scalp, in

the one-operand condition with respect to the zero-operand

condition, but only for the active practice group. In sharp con-

trast, on both right and midline electrode sites, the passive prac-

tice group showed a modulation that appears compatible with

retrieval-induced facilitation in that amplitude of the N100 was

higher for the two-operand condition than for the zero-operand

condition. This finding is relevant also with respect to the third

purpose of the present study, that is, the attempt to establish the

plausibility, in the arithmetic domain, of the mechanisms that

have been proposed as underlying RIF in the word domain for

episodic retrieval. Two main accounts have been posited as the

cause of RIF. On the one hand, following the inhibitory account

(e.g., Anderson et al., 1994; Bauml et al., 2010), retrieval practice

of a given item would elicit inhibition of related items aimed at

reducing retrieval competition. On the other hand, according to

the associative interference account (e.g., Williams & Zacks,

2001), RIF would occur as a sort of side effect due to the fact that

retrieval practice consolidates associations between practiced

items, and these strengthened associations, in turn, cause the

weakened access to related competitors due to blocking-like

phenomena (e.g., Rundus, 1973). The analyses of the result-

lockedN100, showing that amplitude wasmodulated following a

RIF-like pattern only in the active practice group, are consistent

with an inhibitory view of RIF in that this is the only account

predicting a different RIF-related response depending on prac-

tice type. This finding is particularly important in light of the fact

that, because retrieval of arithmetic knowledge is strongly auto-

matic (e.g., De Brauwer & Fias, 2009; Galfano et al., 2003), one

could have expected similar processing dynamics for the two

groups, which might have resulted in a similar ERP pattern mir-

roring behavioral results. As regards the functional meaning of

theN100, it has been suggested that amplitude of this component

is sensitive to anticipatory processes (e.g., Vogel & Luck, 2000)

that may be at work also in the current paradigm in which an-

ticipation of an equation’s result may be expected. In the present

experiment, this component may overlap with ongoing processes

elicited by operand onset and it probably reflects also memory

control processes. The presence of a facilitation-like pattern se-

lective for the passive practice group was unexpected. This find-

ing on one hand appears to confirm that practice type played an

important role, but on the other hand, it suggests that in the

present context, RIF and retrieval-induced facilitation are asso-

ciated with the modulation of the same early ERP component,

although in a different way, depending on practice type. The lack

of significant brain–behavior correlations suggests that more ex-

perimental work is needed to clarify the functional meaning of

the modulations observed at the level of N100, although the

observed dissociation as a function of practice type probably

reflects the presence of inhibitory processes for the active practice

group only.

Interesting effects were also found at the level of the P350

component time-locked to the onset of the result. This posterior

positive deflection has already been reported in several studies

using the arithmetic verification task (see, e.g., Niedeggen et al.,

1999). In this kind of paradigm, this component is typically ob-

served in response to correct results in a time window similar to

that examined here (Niedeggen & Rosler, 1999). In the present

study, examination of such a component revealed that, indepen-

dent of group, amplitude was significantly smaller for results

belonging to the one-operand set than for results belonging to the

zero-operand set, consistent with a RIF-like pattern. In addition,

although no overall retrieval-induced facilitation pattern

emerged in the ANOVA, significant correlations over fronto-

central sites were found between P350 amplitude differences and

both retrieval-induced facilitation and RIF. In detail, partici-

pants with stronger retrieval-induced facilitation in behavioral

data showed larger amplitudes in the two-operand condition

compared with the zero-operand condition, at least at some

electrodes. At the same time, participants with stronger behav-

ioral RIF showed higher amplitudes in the zero-operand condi-

tion comparedwith the one-operand condition. One possibility is

that this pattern of results may reflect variations in recollective

processes, so that better recollection could be expected for prac-

ticed items, and, if inhibition occurs, subsequent impaired recol-

lection for unpracticed items may arise as well. In other words,

the P350 in the present context may represent an index of the

level of activation of cue–item associations, a proposal that will

be strengthened if the association between late positive amplitude

and facilitation is confirmed by replication. Specifically, this

component may reflect variations in cue–item association

ERPs and retrieval-induced forgetting 1689

strength, according to which increased activation of practiced

associations (accounting for retrieval-induced facilitation) would

cause the subsequent blocking or reduced access to unpracticed,

related competitors (accounting for RIF; e.g., Williams & Zacks,

2001). To note, in line with this interpretation, recent functional

magnetic resonance imaging evidence has been reported docu-

menting both an inhibition and a blocking component of RIF at

test (Wimber et al., 2008). In this regard, it is important to em-

phasize that, in the present study, the test phase consisted of the

presentation of the equation in two separate steps, with the onset

of the problem first (e.g., 2 � 5), followed by presentation of the

result to be verified (i.e., 10). This sequential presentation of

items may have elicited recall processes with the possibility that

blocking-related processes may have played a role to some ex-

tent. Thus, it is possible that the specific experimental task de-

vised to investigate RIF in the present experiment (which is

typically used to address arithmetic knowledge in EEG studies)

played a role in eliciting blocking-like phenomena along with

inhibition. The reason why Spitzer et al. (2009) found ERP ev-

idence for inhibition in isolation (i.e., without blocking-like pro-

cesses) is likely due to the adoption of a widely interference-free

test (i.e., item recognition), whereas in the present study we used

a recall situation, in which RIF is likely to reflect both inhibition

and some degree of interference (also see Wimber et al., 2008).

Overall, our ERP findings appear to be consistent, at least in

part, with the inhibitory account of RIF. In particular, we in-

terpret the fact that RIF was reflected in the amplitude modu-

lation of an ERP component (i.e., the N100) only for the group

that was engaged in active, deliberate, retrieval at practice as

evidence of inhibitory processes operating during the practice

phase. The specific amplitude gradient involving RIF and re-

trieval-induced facilitation at the level of P350 and the related

correlation analyses suggest that RIF in the domain of semantic

retrieval of arithmetic facts may, at least in part, be linked to

retrieval interference processes that cause retrieval-induced fa-

cilitation and that are likely to reflect quantitative differences in

spreading activation within the lexicon of arithmetic facts

(Censabella, & Noel, 2004). Alternatively, as discussed earlier,

the P350 pattern that emerged in the present experiment may

simply reflect the fact that our task in the test phase favored the

occurrence of blocking-like phenomena, including some degree

of interference (see also Wimber et al., 2008).

REFERENCES

Anderson, M. C., Bjork, E. L., & Bjork, R. A. (2000). Retrieval-inducedforgetting: Evidence for a recall-specific mechanism. PsychonomicBulletin and Review, 7, 522–530.

Anderson, M. C., Bjork, R. A., & Bjork, E. L. (1994). Remembering cancause forgetting: Retrieval dynamics in long-termmemory. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 20,1063–1087.

Ashcraft, M. H. (1992). Cognitive arithmetic: A review of data and the-ory. Cognition, 44, 75–106.

Bauml, K.-H. (2008). Inhibitory processes. In H. L. Roediger III (Ed.),Cognitive psychology of memory. Vol. 2 of Learning and memory: Acomprehensive reference (pp. 195–220). Oxford: Elsevier.

Bauml, K.-H., & Kuhbandner, C. (2007). Remembering can cause for-gettingFbut not in negative moods. Psychological Science, 18, 111–115.

Bauml, K.-H., Pastotter, B., & Hanslmayr, S. (2010). Binding and in-hibition in episodic memory: Cognitive, emotional, and neural pro-cesses. Neuroscience and Biobehavioral Reviews, 34, 1047–1054.

Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discoveryrate: A practical and powerful approach tomultiple testing. Journal ofthe Royal Statistical Society Series B, 57, 289–300.

Camp, G., Pecher, D., & Schmidt, H. G. (2007). No retrieval-inducedforgetting using item-specific independent cues: Evidence against ageneral inhibitory account. Journal of Experimental Psychology:Learning, Memory, and Cognition, 33, 950–958.

Campbell, J. I. D. (1995). Mechanisms of simple addition and multipli-cation: A modified network interference theory and simulation.Mathematical Cognition, 1, 121–164.

Campbell, J. I. D., & Phenix, T. L. (2009). Target strength and retrieval-induced forgetting in semantic recall.Memory and Cognition, 37, 65–72.

Campbell, J. I. D., & Tarling, D. P. M. (1996). Retrieval processes inarithmetic production and verification. Memory and Cognition, 24,156–172.

Campbell, J. I. D., & Timm, J. C. (2000). Adults’ strategy choices forsimple addition: Effects of retrieval interference.Psychonomic Bulletinand Review, 7, 692–699.

Cappelletti, M., Kopelman, M. D., Morton, J., & Butterworth, B.(2005). Dissociations in numerical abilities revealed by progressivecognitive decline in a patient with semantic dementia. Cognitive Neu-ropsychology, 22, 771–793.

Censabella, S., &Noel,M.-P. (2004). Interference in arithmetic facts: Areactive suppression processes involved when performing simple mentalarithmetic? Cahiers de Psychologie Cognitive/Current Psychology ofCognition, 22, 635–671.

De Brauwer, J., & Fias, W. (2009). A longitudinal study of children’ssimple multiplication and division problems. Developmental Psychol-ogy, 45, 1480–1496.

Galfano, G., Mazza, V., Angrilli, A., & Umilta, C. (2004). Electrophys-iological correlates of stimulus-driven multiplication facts retrieval.Neuropsychologia, 42, 1370–1382.

Galfano, G., & Pavani, F. (2005). Long-lasting capture of tactile atten-tion by body shadows. Experimental Brain Research, 166, 518–527.

Galfano, G., Penolazzi, B., Vervaeck, I., Angrilli, A., & Umilta, C.(2009). Event-related brain potentials uncover activation dynamics inthe lexicon of multiplication facts. Cortex, 45, 1167–1177.

Galfano, G., Rusconi, E., & Umilta, C. (2003). Automatic activation ofmultiplication facts: Evidence from the nodes adjacent to the product.Quarterly Journal of Experimental Psychology, 56A, 31–61.

Hicks, J. L., & Starns, J. J. (2004). Retrieval-induced forgetting occurs intests of item recognition. Psychonomic Bulletin and Review, 11, 125–130.

Jasper, H. H. (1958). The ten-twenty electrode system of the Interna-tional Federation. Electroencephalography and Clinical Neurophysi-ology, 10, 371–375.

Johansson, M., Aslan, A., Bauml, K.-H., Gabel, A., & Mecklinger, A.(2007). When remembering induces forgetting: Electrophysiologicalcorrelates of retrieval-induced forgetting. Cerebral Cortex, 17, 1335–1341.

Johnson, S. K., & Anderson,M. C. (2004). The role of inhibitory controlin forgetting semantic knowledge.Psychological Science, 15, 448–453.

Jost, K., Henninghausen, E., & Rosler, F. (2004). Comparing arithmeticand semantic fact retrieval: Effects of problem size and sentence con-straint on event-related brain potentials.Psychophysiology, 41, 46–59.

Kuhl, B. A., Dudukovic, N. M., Kahn, I., & Wagner, A. D. (2007).Decreased demands on cognitive control reveal the neural processingbenefits of forgetting. Nature Neuroscience, 10, 908–914.

Levy, B. J., & Anderson, M. C. (2002). Inhibitory processes and thecontrol of memory retrieval. Trends in Cognitive Sciences, 6, 299–305.

Manly, C. F., & Spoehr, K. T. (1999). Mental multiplication: Nothingbut the facts? Memory and Cognition, 27, 1087–1096.

Masse, C., & Lemaire, P. (2001). Do people combine the parity- and five-rule checking strategies in product verification? Psychological Re-search, 65, 28–33.

Niedeggen, M., & Rosler, F. (1999). N400 effects reflect activationspread during retrieval of arithmetic facts. Psychological Science, 10,271–276.

Niedeggen, M., Rosler, F., & Jost, K. (1999). Processing of incongruousmental calculation problems: Evidence for an arithmetic N400 effect.Psychophysiology, 36, 1–18.

1690 G. Galfano et al.

Nunez-Pena, M. I., Cortinas, M., & Escera, C. (2006). Problem sizeeffect and processing strategies inmental arithmetic.NeuroReport, 17,357–360.

Pauli, P., Lutzenberger, W., Rau, H., Birbaumer, N., Rickard, T. C.,Yaroush, R. A., & Bourne, L. E. Jr (1994). Brain potentials duringmental arithmetic: Effects of extensive practice and problem diffi-culty. Cognitive Brain Research, 2, 21–29.

Phenix, T. L., & Campbell, J. I. D. (2004). Effects of multiplicationpractice on product verification: Integrated structures model or re-trieval-induced forgetting? Memory and Cognition, 32, 324–335.

Raaijmakers, J. W., & Shiffrin, R. M. (1981). Search of associativememory. Psychological Review, 88, 93–134.

Roman, P., Soriano, M. F., Gomez-Ariza, C. J., & Bajo, M. T. (2009).Retrieval-induced forgetting and executive control. PsychologicalScience, 20, 1053–1058.

Rundus, D. (1973). Negative effects of using list items as retrieval cues.Journal of Verbal Learning and Verbal Behavior, 12, 43–50.

Rusconi, E., Galfano, G., Rebonato, E., & Umilta, C. (2006). Bidirec-tional links in the network of multiplication facts. Psychological Re-search, 70, 32–42.

Rusconi, E., Galfano, G., Speriani, V., & Umilta, C. (2004). Capacityand contextual constraints on product activation: Evidence fromtask-irrelevant fact retrieval. Quarterly Journal of Experimental Psy-chology, 57A, 1485–1511.

Spitzer, B., & Bauml, K.-H. (2007). Retrieval-induced forgetting in itemrecognition: Evidence for a reduction in general memory strength.Journal of Experimental Psychology: Learning, Memory, and Cogni-tion, 33, 863–875.

Spitzer, B., Hanslmayr, S., Opitz, B., Mecklinger, A., & Bauml, K.-H.(2009). Oscillatory correlates of retrieval-induced forgetting in rec-ognition memory. Journal of Cognitive Neuroscience, 21, 976–990.

Storm, B. C., Bjork, E. L., Bjork, R. A., & Nestojko, J. F. (2006). Isretrieval success a necessary condition for retrieval-induced forget-ting? Psychonomic Bulletin and Review, 13, 1023–1027.

Szucs, D., &Csepe, V. (2004). Access to numerical information is dependenton themodality of stimulus presentation inmental addition: A combinedERP and behavioral study. Cognitive Brain Research, 19, 10–27.

Szucs, D., & Csepe, V. (2005). The effect of numerical distance and stim-ulus probability on ERP components elicited by numerical incongru-encies in mental addition. Cognitive Brain Research, 22, 282–300.

Szucs, D., & Soltesz, F. (2010). Event-related brain potentials to viola-tions of arithmetic syntax represented by place value structure. Bi-ological Psychology, 84, 354–367.

Szucs, D., Soltesz, F., Czigler, I., & Csepe, V. (2007). Electroencephalo-graphy effects to semantic and non-semantic mismatch in propertiesof visually presented single-characters: The N2b and the N400.Neuroscience Letters, 412, 18–23.

Townsend, J. T., & Ashby, F. G. (1983). Stochastic modeling of elemen-tary psychological processes. New York: Cambridge University Press.

Verguts, T., & Fias, W. (2005). Interacting neighbors: A connectionistmodel of retrieval in single-digit multiplication. Memory and Cogni-tion, 33, 1–16.

Vogel, E. K., & Luck, S. J. (2000). The visual N1 component as an indexof a discrimination process. Psychophysiology, 37, 190–203.

Williams, C. C., & Zacks, R. T. (2001). Is retrieval-induced forgetting aninhibitory process? American Journal of Psychology, 114, 329–354.

Wimber, M., Bauml, K.-H., Bergstrom, Z., Markopoulos, G., Heinze,H.-J., & Richardson-Klavehn, A. (2008). Neural markers of inhibitionin human memory retrieval. Journal of Neuroscience, 28, 13419–13427.

Wimber, M., Rutschmann, R. M., Greenlee, M. W., & Bauml, K.-H.(2009). Retrieval from episodic memory: Neural mechanisms of in-terference resolution. Journal of Cognitive Neuroscience, 21, 538–549.

Zamarian, L., Karner, E., Benke, T., Donnemiller, E., & Delazer, M.(2006). Knowing 7 � 8, but not the meaning of ‘Elephant’: Evidencefor the dissociation between numerical and non-numerical semanticknowledge. Neuropsychologia, 44, 1708–1723.

Zamarian, L., Stadelmann, E., Nurk, H.-C., Gamboz, N., Marksteiner,J., &Delazer,M. (2007). Effects of age andmild cognitive impairmenton direct and indirect access to arithmetic knowledge. Ne-uropsychologia, 45, 1511–1521.

(Received May 31, 2010; Accepted May 24, 2011)

Appendix

ERPs and retrieval-induced forgetting 1691

Table A.1. Multiplication Problems Including Correct Results and

Lures When the Two-Operand Set Consisted of 2, 5, 7, and 8

Problems Correct results Lures

Two-operand set2 � 55 10 207 � 25 14 128 � 25 16 185 � 55 25 155 � 75 35 458 � 55 40 327 � 75 49 637 � 85 56 64

One-operand set3 � 55 15 216 � 25 12 188 � 35 24 325 � 45 20 284 � 75 28 249 � 55 45 276 � 75 42 367 � 95 63 45

Zero-operand set3 � 45 12 186 � 35 18 124 � 45 16 204 � 65 24 329 � 35 27 219 � 45 36 286 � 65 36 426 � 95 54 48

Table A.2. Multiplication Problems Including Correct Results and

Lures When the Two-Operand Set Consisted of 3, 4, 6, and 9

Problems Correct results Lures

Two-operand set3 � 45 12 206 � 35 18 304 � 45 16 204 � 65 24 289 � 35 27 219 � 45 36 326 � 65 36 486 � 95 54 48

One-operand set3 � 55 15 216 � 25 12 108 � 35 24 165 � 45 20 284 � 75 28 329 � 55 45 356 � 75 42 487 � 95 63 81

Zero-operand set2 � 55 10 207 � 25 14 108 � 25 16 145 � 55 25 355 � 75 35 218 � 55 40 327 � 75 49 637 � 85 56 64