Memory retrieval is often challenged by other irrelevant competing memories that cause interference. This phenomenon is typically studied with the retrieval practice paradigm in which a category cue (e.g., Fruits) is presented together with an item-specific cue (e.g., Or::). Presentation of the category cue usually induces interference by reactivating competing memories (e.g., Banana, Apple, etc.), which is thought to be solved by means of inhibition, leading to retrieval-induced forgetting of these competing memories. Previous studies associated interference with an increase in medial prefrontal theta band (4–8 Hz) oscillations, but these studies could not disentangle the interference from the inhibition processes. We here used a retrieval practice procedure in which the category cue was presented before the item-specific cue to disentangle the interference from the inhibition signal. Furthermore, a competitive retrieval condition was contrasted with a noncompetitive condition. At a behavioral level, retrieval-induced forgetting was found in the competitive but not in the noncompetitive condition. At a neural level, presentation of the category cue elicited higher levels of theta power in the competitive condition, when compared with the noncompetitive retrieval condition. Importantly, this difference was localized to the ACC, which has been associated with the detection and mediation of interference. Additionally, theta power decreased upon presentation of the item-specific cue, and this difference was related to later forgetting. Our results therefore disentangle, for the first time, interference and inhibition in episodic memory retrieval and suggest that theta oscillations track the fine-grained temporal dynamics of interference during competitive memory retrieval.
Retrieval of a target memory is often challenged by other competing memories, which are irrelevant for the particular information one is trying to retrieve (Anderson & Neely, 1996; Anderson, Bjork, & Bjork, 1994). Consider, for instance, that you are trying to recall a memory from your last summer holidays. Several memories of previous holidays are prone to become activated and compete with the specific memory of the last ones, thereby eliciting retrieval interference. According to the inhibitory theory, mechanisms are needed to detect this interference and engage higher-order control mechanisms that reduce interference by suppressing the competing memories. In this study, we provide evidence that theta oscillations in the medial pFC track the time course of interference during selective memory retrieval. Employing a new experimental paradigm, we here show that theta oscillations in the medial pFC specifically reflect interference (the activation of competing information) and not its resolution, which goes beyond previous studies.
Interference and inhibition during memory retrieval are typically studied by means of the retrieval practice paradigm (Anderson et al., 1994). In this paradigm, after studying pairs of semantically related words (e.g., FRUIT-Orange; FRUIT-Banana, or ANIMAL-Horse), participants engage in a retrieval practice phase, selectively retrieving half of the exemplars from half of the categories, given a retrieval cue (e.g., FRUIT-Or::). In a final test, participants are asked to retrieve all of the exemplars from all of the previously studied categories. Results consistently show that whereas the recall of practiced items (Orange) is improved, recall of unpracticed items that belong to practiced categories (Banana) is impaired relative to control items (items from unpracticed categories; e.g., Horse). This phenomenon is termed retrieval-induced forgetting (RIF). RIF is typically explained by a reactivation of competing items during the retrieval practice phase, which are inhibited to facilitate retrieval of the target item. The results of several behavioral (e.g., Román, Soriano, Gómez-Ariza, & Bajo, 2009; Soriano, Jiménez, Román, & Bajo, 2009; Anderson, 2003; Bäuml & Hartinger, 2002; Anderson & Spellman, 1995), electrophysiological (e.g., Hellerstedt & Johansson, 2013; Waldhauser, Johansson, & Hanslmayr, 2012; Hanslmayr, Staudigl, Aslan, & Bäuml, 2010; Staudigl, Hanslmayr, & Bäuml, 2010; Johansson, Aslan, Bäuml, Gäbel, & Mecklinger, 2007), and neuroimaging studies (Wimber, Rutschmann, Greenlee, & Bäuml, 2009; Wimber et al., 2008; Kuhl, Dudukovic, Kahn, & Wagner, 2007) strongly support this inhibitory account of RIF.
On a neurocognitive level, interference has been consistently associated with oscillatory prefrontal theta band activity (4–9 Hz) in studies using response conflict tasks, such as Flanker or Stroop tasks (e.g., Cavanagh, Cohen, & Allen, 2009; Hanslmayr et al., 2008). Theta oscillations also seem to index the amount of activated information in memory (e.g., Khader & Rösler, 2011; Mecklinger, Kramer, & Strayer, 1992). Regarding studies using the retrieval practice paradigm, Hanslmayr et al. (2010) found that a competitive retrieval condition elicited higher theta power, when compared with a noncompetitive retrieval condition. This effect was most pronounced over frontal and left parietal sites and correlated with later forgetting. Similar results have been found by Waldhauser et al. (2012) and Staudigl et al. (2010).
Staudigl et al. (2010) compared two cycles of retrieval practice and observed a reduction in theta amplitude from the first to the second retrieval cycle (reflecting reduced activation of competing information), which predicted the amount of RIF. No such effect was found in a reexposure control condition. Importantly, the reduction in theta amplitude across retrieval practice cycles was localized to the ACC, confirming the results of an fMRI study showing a reduction of BOLD signal in ACC (and in the dorsolateral pFC) with an increase of retrieval attempts (Kuhl et al., 2007). These results are consistent with theories assuming that ACC plays an important role in the detection of interference and its mediation (Botvinick, Braver, Barch, Carter, & Cohen, 2001).
A problem with the prior studies, however, is that by presenting the category cue together with the item-specific retrieval cue, they cannot disentangle the effects of interference, triggered by the cue (which activates previously associated items), from the effects of inhibition, needed to reduce interference. From a theoretical perspective this is a relevant question given that competition is said to precede inhibition. In fact, without competition, there is no need for inhibition to act. So far, however, these temporal dynamics of competition and inhibition have not been addressed. To address this question, we used an adaptation of the retrieval practice paradigm previously introduced by Bajo, Gómez-Ariza, Fernandéz, and Marful (2006), where the category cue was temporally separated from the item-specific cue.
Underlying this precuing procedure is the idea that the presentation of the category cue will lead to the activation of previously associated items competing for retrieval. This has been demonstrated in previous studies showing that presenting a retrieval cue activates associated items and renders retrieval more difficult as the number of associated items increases. This effect held constant both when associations were explicitly learned during study (Anderson & Reder, 1999; Watkins, 1979) and when a given category was naturally connected to more associatively related words (Nelson, Schreiber, & Xu, 1999; Nelson, McEvoy, & Bajo, 1984).
In this study, competition (interference) is elicited by presenting the category cue. Afterwards, upon presentation of the item-specific cue, inhibitory mechanisms should come into play to reduce interference, thereby facilitating retrieval of the target item. Unlike previous experiments, this procedure allows to trace the specific temporal dynamics of competition (as prompted by the category cue), disentangling it from the inhibition-related activity, that should only be present upon presentation of the item-specific cue.
Following prior work (Hanslmayr et al., 2010; Anderson, Bjork, & Bjork, 2000), a competitive and a noncompetitive retrieval condition were contrasted. In the competitive condition, participants were presented with an occupational category that served as a category cue (e.g., Actors), followed by the face of a famous person (the item-specific cue), whose name they should retrieve (e.g., Brad Pitt). In the noncompetitive condition, participants saw the first two letters of the category cue (e.g., Ac:_), also followed by the face of a famous person, but were instructed to retrieve the category name. Note that, although both conditions demand participants to actively retrieve information (the category label or the specific name), the competitive condition requires participants to recall a specific item in the presence of competition, whereas in the noncompetitive condition participants are simply asked to retrieve that category label with no need to retrieve the specific names/faces associated to them. Therefore, the category label in the noncompetitive condition does not act as a retrieval cue for the associated items, and thus, no requirement for interference resolution mechanisms is needed.
The assumption is that, in the competitive condition, the presentation of the category cue (e.g., Actor) leads to the activation of items previously associated to it that will compete for retrieval, creating interference, which should subsequently trigger inhibitory mechanisms to reduce competition. These dynamics should be specifically reflected by medial prefrontal theta power. Accordingly, it was hypothesized (i) that presentation of the cue in the competitive retrieval condition elicits higher theta power than presentation of the cue in the noncompetitive condition, (ii) that theta power decreases upon presentation of the target face, reflecting competition reduction, and (iii) that this decrease in theta power from cue to face correlates with later forgetting.
Finally, the use of facial stimuli might help shed light on some controversial topics in the face processing and recognition literature, especially regarding the similarity of the mechanisms underlying faces and other objects' recognition (e.g., McKone, Kanwisher, & Duchaine, 2007; Haxby, Hoffman, & Gobbini, 2000; Farah, 1996) and the issue of whether face naming is subject to interference (Vitkovitch, Potton, Bakogianni, & Kinch, 2006; Darling & Valentine, 2005; Brédart & Valentine, 1992). If personal representations are inhibited (i.e., if the RIF effect is found for this type of stimuli), this would mean they are vulnerable to interference and that, at least at some instances, mechanisms underlying faces and objects' recognition are of a similar nature.
In summary, in this study, a precuing procedure was used to disentangle competition and inhibition signals during the retrieval practice paradigm. To isolate the inference signal, a competitive retrieval condition was contrasted with a noncompetitive condition. Because we were interested in the temporal dynamics of competition, we collected EEG data. Given the great temporal resolution of EEG, using this technique allowed us to trace the fine-grained temporal dynamics of competition, which would not have been possible using behavioral data only.
Twenty (10 women) students from the University of Granada participated in this study (mean age = 24 years, SD = 3.1 years). All participants were Spanish or had been living in Spain for at least 15 years. All of them reported normal or corrected-to-normal vision. Participants gave written informed consent before the study and received either course credits or a monetary reward for their participation.
Forty-eight pictures of famous people in Spain from eight occupational categories (male actors, politicians, football players, writers, and TV hosts, and female singers, royalty members, and tabloid stars) were chosen from a set of pictures collected before the experiment to assess the familiarity of each item. The most familiar pictures were selected from this set with the constraint that none of the names of the famous people shared the first two letters. Two additional categories (radio personalities and bull fighters) with three exemplars each were used as fillers to prevent for primacy and recency effects and were not taken into consideration in any of the analyses described. The pictures (5.19 cm × 6.99 cm) were presented in color against a white background. The faces displayed a neutral to mildly positive expression. An oval template was applied around each picture to standardize silhouettes and background (see Young, Ellis, Flude, McWeeny, & Hay, 1986).
The experiment consisted of two blocks (within-participant design). Each block comprised a study phase, a retrieval practice phase, and a final test. The two blocks differed in the list of items provided as well as in the type of selective retrieval performed—competitive or noncompetitive (Figure 1A). From the total of pictures, half (24) were shown in the competitive condition (C), and the other half were presented in the noncompetitive condition (NC). A similar procedure has been used in different RIF studies, such as Gómez-Ariza, Fernandéz, and Bajo (2012), Hanslmayr et al. (2010), or Anderson et al. (2000).
In the C condition, participants were shown the category cue, which consisted of the occupational category of the target face (e.g., Actor), followed by the item-specific cue (a previously studied face belonging to that same category; e.g., Brad Pitt). Participants were instructed to retrieve the name of each famous person. In the NC condition, participants saw the first two letters of the to-be-retrieved category (e.g., Ac:_) and subsequently a face with the corresponding name written below. In this condition, the participants' task consisted in retrieving the category name (“Actor,” for instance).
In both conditions, retrieval was performed for half of the items from three of the four studied categories. This created three different types of items: practiced items, unpracticed items (from practiced categories), and control items from nonpracticed categories, which served as baseline items.
Across participants, all items served equally often as practiced, unpracticed, and control items, in both C and NC conditions. Assignment of list to block and presentation order of the blocks was counterbalanced across participants.
Each block started with a study list, during which the 24 items were presented successively on a 15-in. computer screen. After a fixation cross (1 sec), participants were presented a face together with its respective name and category displayed below (see Figure 1A) for 5 sec. Presentation order of the items was randomized with the restriction that the first and the last four faces were filler items, and two faces from the same category did not appear sequentially. Participants were instructed to memorize all faces as well as their name and category, as they would be asked to recall them later.
Retrieval Practice Phase
After studying the 24 items, participants engaged in nine retrieval practice trials, where half of the exemplars from three of the studied categories were presented. In the C condition, a trial started with a fixation cross, with a variable duration (1–1.5 sec) followed by a category cue for 2 sec (e.g., Actor) and a blank screen (1 sec). Thereafter, the item-specific cue (i.e., the face) appeared on the screen for 2.5 sec, followed by a blank screen (4 sec), during which participants were asked to overtly retrieve the corresponding name (e.g., Brad Pitt). In the NC condition, a trial started with a jittered fixation cross (1–1.5 sec) followed by the first letters of the category cue (e.g., Ac::). Then, a blank screen was presented (1 sec) and subsequently a face belonging to that category, along with the corresponding name displayed below (2.5 sec). During the following 4 sec, participants were asked to overtly retrieve the corresponding category name (e.g., Actors). To prevent possible speech artifacts in the EEG, participants were instructed to only respond during the 4-sec blank interval.
After a 5-min distractor task (the vocabulary test from the Wechsler Adult Intelligence Scale), each block ended with a memory test, during which each of the studied faces were presented individually, and participants were asked to produce the corresponding name. After a fixation cross (1 sec), each face appeared on the screen for 4 sec, and participants were instructed to name the person as quickly as possible. The order for testing was pseudorandomized, such that unpracticed items and half of the control items were shown first, followed by practiced and the other half of the controls. This was done to prevent blocking effects, that is, to prevent retrieval of the practiced items to block access to the unpracticed ones, which would confound the effects of memory inhibition (Anderson et al., 1994). After a short break, the second block of the experiment was conducted.
The EEG was recorded from 64 scalp electrodes mounted on an elastic cap according to the standard 10–20 system. The continuous electrical activity was recorded with Neuroscan Synamps2 amplifiers (El Paso, TX). The EEG was initially recorded against an electrode placed in the midline of the cap (between Cz and CPz) and re-referenced off-line against a common average reference. Each EEG channel was amplified with a band pass of 0.01–100 Hz and digitized at a sampling rate of 500 Hz. Impedances were kept below 5 kΩ. To control for vertical and horizontal eye movements, two additional electrodes were located above and below the left eye and another two at the outer side of each eye. Before data analysis, a high-pass filter (at 1 Hz) was applied, and data were corrected for artifacts such as blinks, horizontal eye movement, and EKG, by performing an independent component analysis that allows an identification of components corresponding to eye blinks, horizontal movements, or EKG artifacts. Remaining artifacts, because of muscle activity or poor artifact correction, were excluded by careful visual inspection.
EEG analysis was performed using FieldTrip Matlab toolbox software (Oostenveld, Fries, Maris, & Schoffelen, 2011) and in-house Matlab (The MathWorks, Munich, Germany) codes. The EEG data were segmented into time windows ranging from −2000 msec before and 4000 msec after the onset of the retrieval cue (cues in the C and NC conditions) and from −3000 msec to 4000 msec around face presentation (both C and NC). These broad time windows were chosen to prevent filter artifacts at the edges of the epochs. Analyses were restricted to a 2500-msec time window, ranging from −500 to 2000 msec.
Analysis of the Oscillatory Power
Time–frequency analyses were conducted applying Morlet wavelet transformation (7 cycles) to derive the time–frequency representation. Data were filtered in a frequency range of 1–30 Hz and exported in bins of 50 msec and 1 Hz. To analyze event-related changes, power changes were calculated in relation to a prestimulus baseline (set to 500–0 msec before stimulus onset). Analyses were restricted to theta band (6–8 Hz), as explained in the Results section.
To estimate the sources of activity that contributed to the effects at the sensor level, the dynamic imaging of coherent sources (DICS) Beamformer approach was used (Gross et al., 2001). The reliability of Beamforming methods in localizing the source of EEG activity was demonstrated by several combined EEG-fMRI, MEG-fMRI, and MEG-intracranial EEG studies (see, e.g., Singh, 2012; Hanslmayr et al., 2011; Dalal et al., 2009).
For source reconstruction, a standardized boundary element model was used to calculate the leadfield. The standard boundary element model was derived from an averaged T1-weighted MRI data set (MNI, www.mni.mcgill.ca). A previous study demonstrated that similar results are obtained from such a standard head model and individual head models (Fuchs, Kastner, Wagner, Hawes, & Ebersole, 2002).
The DICS algorithm allows performing source reconstruction in a frequency domain, given a time latency and frequency range defined by the user. The Beamformer computes the changes in power from a prestimulus baseline to a poststimulus interval, transforming data into standard MNI space (Montreal Neurological Institute, Montreal, Quebec, Canada). The prestimulus baseline was set from 500 to 0 msec before stimulus onset. The time window for the poststimulus interval and the frequency band were chosen according to the effects on the sensor level.
In an attempt to replicate previous findings, ERPs were also analyzed (Hanslmayr et al., 2010). The ERPs were computed for each participant and condition, in a time window from −500 to 2000 msec. The waveforms were low-pass filtered at 15 Hz and high-pass filtered at 1 Hz. For the interaction between the C and NC conditions (cue–face), a time window ranging from 400 to 500 msec was chosen for further analysis, upon visual inspection of the grand-averaged waveform. Analyzing the grand-averaged waveforms, when comparing C and NC conditions upon presentation of the category cue, two time windows were chosen for subsequent analysis: 350–400 and 410–450 msec, according to Hanslmayr et al. (2010). For the comparison between the conditions upon presentation of the item-specific cue (the face), one time window was chosen, ranging from 150 to 180 msec, comprising the well-known N170 component.
Two 2 × 2 repeated-measures ANOVAs were conducted to analyze the behavioral data. To assess the forgetting effect, we calculated an ANOVA with the factors Item Type (unpracticed vs. control) and Retrieval Condition (C vs. NC). Likewise, a second ANOVA was conducted for the facilitation effect, taking as factors Item Type (practiced vs. control) and the Retrieval Condition. The significant effects from these analyses were then followed up by performing planned comparisons with two-tailed paired-samples t tests.
In a first step, an interaction analysis (cue minus face × condition) of the power differences averaged across all electrode sites was conducted to define time–frequency windows for subsequent analyses. To this end, the difference in oscillatory theta power between the face and the cue was computed for each participant for the C and NC conditions, respectively. This difference was then subjected to a dependent samples t test. To account for multiple testing, a Monte Carlo randomization procedure was employed, following Maris and Oostenveld (2007).
This method randomizes the observations and recalculates the statistic for the randomized data after each run. The frequency bands and time windows exhibiting significant interaction effects are then subjected to planned comparisons using cluster-based, dependent sample t tests with Monte Carlo randomization (Maris & Oostenveld, 2007).
With cluster statistics, dependent-sample t tests are first calculated for every sample (channel–frequency–time). Then, samples with t values higher than the specified threshold (α = .05) are selected and clustered. Cluster statistics are calculated by taking the maximum sum of t values within every cluster. This result is the test statistic by means of which the effect of the experimental conditions is evaluated. Observations in the data are then randomized, and statistics are recalculated for this randomized data (again, taking the maximum sum of the cluster t values). This procedure is repeated several times, and the proportion of observations that resulted in a larger test statistic than the observed one is calculated. This proportion is the Monte Carlo significance probability (p value). Importantly, for two-tailed tests, such as in this study, alpha value is corrected (that is, considered divided by 2), so that each tail is actually tested with α = .025. From this procedure, clusters of electrodes that significantly differed for each condition (pcorr < .05) are obtained.
Following our hypotheses, planned comparisons were made between power changes from cue to face presentation (both in C and NC conditions) and upon presentation of the cue (C vs. NC).
Source level statistics were conducted for both retrieval conditions (similarly to oscillatory power: cue to face (in C and NC conditions) and cue (C vs. NC). For the statistical analysis at the source level, we used a parametric test that results in analytic probabilities for the null hypothesis and performed dependent-sample t tests on these probabilities. Only clusters bigger than 100 voxels were taken into account. As reported in the results section, the only comparison that met this criterion was between presentation of the category cue in the C and NC conditions (i.e., all other comparisons resulted in clusters < 100 voxels).
The mean amplitudes in the time windows previously chosen by analyzing the grand-averaged waveforms (400–500 msec for the interaction, 350–400 and 410–450 msec for cue C vs. NC, and 150–180 msec for face C vs. NC) were subjected to a dependent samples t test, using the Monte Carlo randomization procedure mentioned above.
Mean recall performance in the C condition was 66% (SD = 19%) for unpracticed items and 81% (SD = 19%) for control ones. In the NC condition, mean recall performance was 73% (SD = 21%) for unpracticed items and 67% (SD = 32%) for control items, as depicted in Figure 1B.
A two-way ANOVA (Type of Item × Condition), revealed a significant Item Type × Condition interaction, F(1, 19) = 7.37, p < .01. No main effects of Item Type or Retrieval Condition were found (p > .05). Post hoc t tests showed impaired memory performance for unpracticed items compared with control items, t(19) = 3.80, p < .001, in the C condition, whereas no difference was found for the NC condition, t(19) = .79, p > .05. These results reveal the typical pattern of RIF in the competitive condition, which was not evident in the noncompetitive condition.
Mean recall performance for practiced items was 95% (SD = 12%) for the C condition and 81% (SD = 19%) for the NC one, as shown in Figure 1C.
Regarding the facilitation effect, the two-way ANOVA did not yield a significant interaction between Item Type and Retrieval Condition (F < 1). There was a main effect of Item Type, F(1, 19) = 12.39, p < .01, as well as a main effect of Competition, F(1, 19) = 10.26, p < .01. Post hoc analyses showed that practiced items were recalled significantly better than control items [t(19) = 3.16, p < .01 for the C condition and t(19) = 2.19, p < .05 for the NC]. Also, mean recall performance was higher in the C (M = .88, SD = .18) than in the NC condition (M = .74, SD = .27), t(19) = 3.22, p < .01.
Oscillatory Power Results
For the interaction analysis, the difference in oscillatory theta power between the face and the cue was computed for each participant for the C and NC conditions, respectively. The interaction (cue minus face × condition) yielded a significant effect in the theta frequency range (6–8 Hz, 500–1000 msec), as depicted in Figure 2A. For this time window and frequency range, the difference in theta power between cue and face was significantly bigger in C than in NC condition, with the C condition showing a larger decrease in theta power, from cue to face, than the NC condition. As the topography illustrates, the difference was most pronounced over frontal and parietal electrode sites (pcorr < .01; Figure 2B). All of the following analyses were based on this time window and frequency range. Theta power results (6–8 Hz, 500–1000 msec) for both retrieval conditions (C and NC), averaged over significant clusters of electrodes, upon presentation of the cue and of the face are depicted in Figure 2C. Planned comparisons on these results are described below.
Cue to Face
Theta power in the C condition decreased upon presentation of the face, over central and left parietal sites (Figure 3A). Nonetheless, this difference was only marginally significant (pcorr = .07). Regarding the NC condition, significant differences were also found when comparing theta activity upon presentation of cue and face, with face presentation inducing a significantly larger increase of theta power than the presentation of the cue (pcorr < .005), over left parietal and frontal regions (Figure 3B).
Cue-C versus Cue-NC
Cluster-based permutation tests revealed that theta power (6–8 Hz) significantly increased after presentation of the cue in the C as compared with NC condition (pcorr < .01), from 500 to 1000 msec (Figure 4A). This increase was evident over fronto-central and left parietal sites, although only the fronto-central electrodes exceeded the statistical threshold (Figure 4B). Source localization analysis (DICS Beamformer; Gross et al., 2001) of this theta power effect indicated that the biggest cluster of significant voxels (366 voxels) was located in the dorsal part of the right ACC (MNI coordinates: x = 4, y = 23, z = 40; ∼BA 32), as depicted in Figure 4C.
Theta Power and Behavioral RIF Effect
To test the hypothesis that high theta power is related to stronger RIF effects, we median-split participants according to the difference in theta power from cue to face presentation and compared the between group differences (larger vs. smaller theta power difference) in the forgetting score (recall of control items minus recall of unpracticed items). An independent samples t test showed significant differences between the two groups, t(18) = −2.1, p < .05. The large theta difference group had higher forgetting scores (M = .22, SD = .18) than the small theta difference group (M = .07, SD = .14). No such pattern emerged either for the facilitation scores, t(18) = .35, p > .05, or for the NC condition [for the forgetting score: t(18) = −.19, p > .05; for the facilitation score: t(18) = .00, p > .05].
The competitive retrieval condition (cue–face) elicited a more positive ERP in comparison with the noncompetitive one, with a midfrontal topography in a time window raging from 400 to 500 msec (pcorr < .05) and led to a more negative ERP than the NC condition over midparietal sites (pcorr < .05) in the same time window (Figure 5).
Cue-C versus Cue-NC
A similar pattern was found for the category cue, where the competitive retrieval cue elicited a more positive ERP in comparison with the cue in the NC condition, both in a time window from 350 to 400 msec and from 410 to 450 msec (pcorr < .05) over frontal–central sites. The opposite pattern emerged over midparietal sites, with the C cue showing an ERP wave more negative than the NC (pcorr < .05) in the first time window (Figure 6).
Face in the C condition elicited a more negative N170 (150–180 msec) component when compared with face presentation in the NC condition (pcorr < .05) over parietal sites (Figure 7).
The behavioral results replicate those from previous studies, showing that selectively retrieving a subset of relevant items impairs the later recall of related but irrelevant items (Anderson et al., 2000). Importantly, and in contrast to the competitive retrieval condition, the noncompetitive retrieval practice did not induce significant forgetting. Whereas retrieving a subset of items impaired later recall of competing associates, mere exposure to this subset of items did not. These results speak in favor of an inhibitory account of RIF, assuming that related but competing items interfered during the retrieval of target items and were suppressed to reduce this interference in the competitive condition (Anderson, 2003). No such effect was found in the noncompetitive condition, where no interference occurred, and thus, no inhibition was needed. Moreover, our results confirm that the noncompetitive condition can be used as a neural baseline for the competitive condition to isolate effects of interference and interference resolution (Hanslmayr et al., 2010).
Although the inhibitory view of the RIF effect has been questioned by alternative cognitive theories (Anderson, 1983; Raaijmakers & Shiffrin, 1981; see Verde, 2012; Anderson, Bjork, & Bjork, 1994, for a review) there is strong evidence from neurophysiological investigations of RIF, showing that RIF relies on the reactivation of competing items and their active suppression (e.g., Waldhauser et al., 2012; Kuhl et al., 2011). Waldhauser and colleagues (2012), for instance, showed that, during selective retrieval of the target items, alpha/beta power increased exactly at those brain regions storing the competitor's memory trace. As increased alpha/beta power has been closely linked to neural inhibition (Jensen & Mazaheri, 2010; Klimesch, Sauseng, & Hanslmayr, 2007) and because this increase in alpha/beta power predicted later forgetting, these results clearly speak against noninhibitory accounts of RIF.
We expected that the competitive condition elicited higher levels of theta power than the noncompetitive condition, reflecting generally higher levels of interference during competitive memory retrieval (Hanslmayr et al., 2010; Staudigl et al., 2010). Going beyond the prior studies, the design of the current study allowed us to disentangle the effects of competition from the effects of inhibition, as the presentation of the category cue was temporally separated from the presentation of the item-specific cue (face). If theta oscillations track the reactivation of memories and thus memory competition, theta power should already increase upon category cue presentation, in the absence of the item-specific memory cue. Our assumption was that the presentation of the category cue during retrieval practice activated competitor memories. This assumption comes from different lines of research. For example, classic studies on blocking effects showed impairment in recalling semantic information after having retrieved associated items (Brown, Cattoi, & Bradley, 1985; Blaxton & Neely, 1983; Brown, 1981). Evidence for this also comes from a recent study by Hellerstedt and Johansson (2013) who, by varying the associative strength of cue and competitors, found a competition-sensitive ERP modulation after presentation of a category cue, reflecting the retrieval of the semantically associated competitors. Finally, many studies have shown that retrieval is more difficult as the number of associated items increases (Anderson & Reder, 1999; Nelson et al., 1984, 1999).
Additionally, we assume that the competitive and noncompetitive conditions differ in the mental operations performed by the participants. When engaging in the competitive condition, participants are aware they will need to retrieve only one item in each trial and this specific retrieval mode should orient their act of retrieval accordingly (Tulving, 1983). Given that participants know that each presentation of a category cue is followed by the retrieval of one specific item, all previously studied items are likely to get preactivated by the category cue and kept in working memory until the target retrieval. The fact that associated items are activated and held in memory until the item-specific cue is presented is what we termed interference in the context of this study.
In contrast, in the noncompetitive condition, where participants only need to retrieve the category name, interference should not be present to the same extent. Thus, even if some items do come into mind upon presentation of the category cue, there is no need for participants to hold them in memory, and therefore, no competition should occur.
Indeed, the present results demonstrate higher levels of category cue-related theta power in the competitive condition than in the noncompetitive condition. The presentation of the category cue in the competitive condition elicited higher theta power than in the noncompetitive one in the time window ranging from 500 to 1000 msec. Hence, shortly after the category cue is presented, items associated to that cue are reactivated in memory leading to interference as reflected by theta oscillations (Staudigl et al., 2010). In line with prior studies (Staudigl et al., 2010; Hanslmayr et al., 2008), the source of this theta interference signal was source localized to the dorsal part of the ACC, suggesting that medial prefrontal theta oscillations track interference, but not its resolution, given that no retrieval (and thus no interference resolution) is actually needed until the presentation of the item-specific cue.
When the item-specific cue is presented, interference should be solved, which should be reflected by a decrease in theta power in the competitive condition from the presentation of the category cue to the presentation of the item-specific cue (i.e., the face). Furthermore, this decrease should correlate with later forgetting, if we assume that interference is resolved by inhibition of the competing items. The obtained results corroborate these hypotheses by showing that the theta power decrease from the category cue to the item-specific cue was related to later forgetting.
Using the precuing procedure, that is, temporally separating the presentation of the category cue and the item-specific cue, it can be assumed that the results obtained in theta track the time course of the rise and fall of interference during competitive memory retrieval.
As mentioned, the difference in theta power between the category cue in the competitive and the noncompetitive condition was localized to ACC (see Staudigl et al., 2010, for similar localization), which has been consistently associated to the detection of interference (Wimber et al., 2009; Kuhl et al., 2007; Botvinick et al., 2001). Importantly, ACC seems to be involved not only in response conflict (e.g., Aron & Poldrack, 2006; Aron, Robins, & Poldrack, 2004; Menon, Adleman, White, Glover, & Reiss, 2001), but in cognitive conflict in general, as seen in studies using Flankers or Stroop tasks (Cavanagh et al., 2009; Hanslmayr et al., 2008). We argue that, in the current study, there is a conflict between different mnemonic representations that compete for retrieval. According to Botvinick et al. (2001), the cingulate cortex is the responsible for mediating and detecting interference, and theta oscillations seem to underlie such activity (Staudigl et al., 2010).
An alternate explanation would be that theta oscillations reflect task effort or difficulty rather than interference. Namely, the competitive condition seems to be harder than the noncompetitive one, and thus, theta oscillations could be reflecting increased task effort instead of competition. However, previous studies have shown that this is unlikely to be the case because theta oscillations at retrieval actually decrease for items that are difficult to retrieve (e.g., Spitzer et al., 2009; Klimesch et al., 2006; see Hanslmayr et al., 2010, for a similar explanation), which indicates that theta oscillations do not relate positively to retrieval effort.
Regarding interference resolution, it has been shown to depend on other brain regions (namely the dorsolateral pFC; Kuhl et al., 2007; Botvinick et al., 2001). According to the cognitive theory, interference during retrieval practice should be resolved by means of inhibition (e.g., Hellerstedt & Johansson, 2013; Waldhauser et al., 2012; Hanslmayr et al., 2010; Staudigl et al., 2010; Román et al., 2009; Soriano et al., 2009; Wimber et al., 2008, 2009; Kuhl et al., 2007; Anderson, 2003; Bäuml & Hartinger, 2002; Anderson & Spellman, 1995).
As mentioned earlier, Waldhauser et al. (2012) were able to disentangle the neural representation of the competitor and the target memory via lateralized presentation and found increased alpha/beta power to reflect inhibition of competing visual memories. In the current study, we were not able to investigate such material specific inhibitory markers as the material was presented foveally and therefore encoded in highly overlapping neural assemblies. Presenting stimuli in a lateralized fashion requires that stimuli are shown on the screen for a very short period of time, which would not be optimal when the stimuli being presented are such complex ones as faces.
Although this direct marker of inhibition could not be obtained within this study, its effect, namely interference resolution, could be revealed as indicated by a reduction in theta power from cue to face presentation in the competitive condition. Once participants see an item-specific cue (i.e., a particular face), theta power decreases, presumably indicating interference resolution in the competitive condition (but not in the noncompetitive one). Furthermore, this difference selectively relates to later forgetting, but not with facilitation. Thus, participants with a greater difference in theta power from cue to face presentation (or, in other words, those able to reduce interference), had higher scores of later forgetting. This supports the assumption that interference was solved by the suppression of competing items.
Concerning the ERP effects, the present results showed that the competitive condition elicited midfrontal positivity and right parieto-occipital negativity, when compared with the noncompetitive condition. Inspecting the waveform, this pattern seems to be reversed after 500 msec, which is exactly what one would expect if these components were originated by a theta oscillation. These results should however be interpreted with caution, given that the comparison includes stimuli of different natures, namely words (for the category cue) and faces (for the item-specific cue), which are known to elicit very different ERP components.
ERP results upon presentation of the category cue are considerably more important, and again, the present results showed that the competitive category cue elicited both fronto-central positivity and right parieto-occipital negativity in contrast to the noncompetitive retrieval cue (Hanslmayr et al., 2010). Importantly, these results go beyond prior studies as they demonstrate that these ERP components reflect solely interference and not inhibition in episodic memory retrieval.
For the ERPs obtained upon face presentation, as expected, both conditions elicited a strong N170 component, typically shown to be modulated by faces (e.g., Rossion & Jacques, 2008). However, the competitive face evoked a more negative N170 than the noncompetitive one. This is possibly because of the fact that in the first condition participants were presented with the face alone, whereas in the later, the face was presented along with the corresponding written name below it, which might explain the less negativity found on the N170 for this noncompetitive condition.
An additional goal of this study was to explore the role of memory inhibition of personal representations. According to previous studies, face processing does not engage the same mechanisms as other objects (e.g., McKone et al., 2007; Haxby et al., 2000; Farah, 1996). This raises the question if RIF, which has been shown for a variety of materials, can also be found using faces or other personal representations. Studies investigating object and face naming, point to material specific effects of interference. For objects, naming a target can be impaired by the presentation of a semantically related distractor (Glaser & Düngelhoff, 1984; Lupker, 1979; Rosinski, Golinkoff, & Kukish, 1975); however, using facial stimuli, Vitkovitch et al. (2006) did not find any effects of interference in naming (see Darling & Valentine, 2005; Brédart & Valentine, 1992, for inconsistent findings). Thus, these studies imply that faces may not be vulnerable to interference, and as a consequence, no inhibitory mechanism would be needed to resolve it.
Models of face processing, on the other hand, do assume that interference may arise between competing personal representations, at different levels of face recognition (e.g., Burton, Bruce, & Hancock, 1999; Brédart, Valentine, Calder, & Gassi, 1995; Burton, Bruce, & Johnston, 1990; Burton et al., 1990; Bruce & Young, 1986). These models, however, either fail to give a solution for how we deal with interference (e.g., Bruce & Young) or propose very quick, automatic mechanisms that inhibit activated competing representations whenever we try to retrieve a particular face or name. In any of these models, inhibitory processes of a more controlled nature (such as the one found in RIF studies) are taken into account.
The results found in this study, both at a behavioral and neural level, could indicate that personal representations are prone to suffer from interference and that resolution of interference between these type of stimuli could depend on inhibitory processes of a controlled nature. Our design does not allow us to examine where this interference is actually taking place (if at a perceptual or at a semantic level), but both levels should be vulnerable to interference. Thus, it seems that, at least in some instances, mechanisms underlying interference for personal representations and other objects could be similar. These conclusions, however, cannot be fully drawn from the current set of data; thus, more research is clearly needed on this topic.
Finally, our results seem to indicate that the memory interference traced by ACC theta effects generalizes to materials other than words, which could suggest a domain general marker of interference.
Taken together, this study shows that (i) theta oscillations track the temporal dynamics of interference during competitive memory retrieval, disentangling interference from inhibition; (ii) the sources of medial prefrontal theta oscillations are located in the ACC; and (iii) the RIF effect generalizes to personal representations.
This research was supported by Doctoral Research grant AP2009-2215 to C. S. F. from the Spanish Ministry of Education and by grants EDU2008-01111 and CSD2008-00048 by the Spanish Ministry of Science and Innovation and grants P07-HUM-2510 and P08-HUM-3600 by the Andalusian Government to T. B. and A. M., and P07-HUM-2510 by the Andalusian Government to T. B. The research has also been supported by a grant from the Deutsche Forschungsgemeinschaft (Project HA 5622/1-1) awarded to S. H.
Reprint requests should be sent to Catarina S. Ferreira, Facultad de Psicologia, Universidad de Granada, Campus de Cartuja s/n, 18071 Granada, Spain, or via e-mail: email@example.com.