Neuropsychologia xxx (2009) xxx xxx Contents lists available at ScienceDirect Neuropsychologia j o u r n a l h o m e p a g e : www.elsevier.com/locate/neuropsychologia Examining the neural basis of episodic memory: ERP evidence that faces are recollected differently from names Graham MacKenzie b,, David I. Donaldson a a Psychological Imaging Laboratory, Department of Psychology, University of Stirling, UK b Department of Psychology, University of Glasgow, UK a r t i c l e i n f o a b s t r a c t Article history: Received 16 October 2008 Received in revised form 26 May 2009 Accepted 31 May 2009 Available online xxx Keywords: Recognition memory Dual process theory Episodic memory Face recognition Recollection Episodic memory is supported by recollection, the conscious retrieval of contextual information associated with the encoding of a stimulus. Event-Related Potential (ERP) studies of episodic memory have identified a robust neural correlate of recollection the left parietal old/new effect that has been widely observed during recognition memory tests. This left parietal old/new effect is believed to provide an index of generic cognitive operations related to recollection; however, it has recently been suggested that the neural correlate of recollection observed when faces are used as retrieval cues has an anterior scalp distribution, raising the possibility that faces are recollected differently from other types of information. To investigate this possibility, we directly compared neural activity associated with remember responses for correctly recognized face and name retrieval cues. Compound face name stimuli were studied, and at test either a face or a name was presented alone. Participants discriminated studied from unstudied stimuli, and made a remember/familiar decision for stimuli judged old. Remembering faces was associated with anterior (500 700 ms) and late right frontal old/new effects (700 900 ms), whereas remembering names elicited mid frontal (300 500 ms) and left parietal (500 700 ms) effects. These findings demonstrate that when directly compared, with reference to common episodes, distinct cognitive operations are associated with remembering faces and names. We discuss whether faces can be remembered in the absence of recollection, or whether there may be more than one way of retrieving episodic context. 2009 Elsevier Ltd. All rights reserved. 1. Introduction The human episodic memory system allows the past to be reexperienced in the present (Tulving, 1983). The hallmark of episodic memory is recollection: the reinstatement of information about past events into conscious awareness. Recollection occurs when the processing of a retrieval cue reactivates memories associated with the cue (e.g., when the smell of baking bread brings a childhood visit to a bakery back to mind) and different types of retrieval cue are thought to elicit the same general process of recollection. In the present article we challenge the assumption that recollection is always associated with the same core cognitive operations; we present electrophysiological data that reveal dissociable neural correlates when recollection is elicited by face and name retrieval cues. Recollection is not, of course, the only basis for making episodic memory judgments. Dual process models of recognition memory Corresponding author at: Department of Psychology, Faculty of Information and Mathematical Sciences, University of Glasgow, 58 Hillhead Street, Glasgow G12 8QB, UK. Tel.: +44 0 141 330 3345; fax: +44 0 141 330 4606. E-mail address: g.mackenzie@psy.gla.ac.uk (G. MacKenzie). (e.g., Atkinson & Juola, 1974; Jacoby & Dallas, 1981; Mandler, 1980; Yonelinas, 1994) propose that a retrieval process called familiarity also supports episodic judgments. According to such models, a retrieval cue is assessed for its familiarity and if the level of familiarity is sufficiently high then the cue can be accepted as previously encountered, even if no contextual information about the original episode can be recollected. Familiarity is a fast-acting, relatively automatic process that provides a conscious feeling that an item has been experienced before. Importantly, from a theoretical perspective both familiarity and recollection are thought to be generic retrieval processes, operating across modalities and domains of information (see Yonelinas, 2002, for a review). One important feature of dual process models is that recollection and familiarity are independent retrieval processes that elicit distinct phenomenological experiences, which can be assessed using the remember/know procedure (Tulving, 1985). Remember responses are required when recognition is accompanied by the retrieval of specific details about the study episode (made on the basis of recollection), whereas know responses reflect recognition without the retrieval of specific details (made on the basis of familiarity). While the remember/know procedure has been criticized for failing to completely isolate familiarity and recollection (e.g., Yonelinas & Jacoby, 1995; Wais, Mickes, & Wixted, 0028-3932/$ see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.neuropsychologia.2009.05.025
2 G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx 2008), remember judgments are widely believed to be supported by relatively more recollection than familiarity, and as such the R/K procedure provides a useful measure of the contribution of recollection to recognition memory performance. One of the strongest bases for dual process models is evidence from event-related potentials (ERPs: an electrophysiological method that can be used to provide a record of the neural activity evoked during performance of a cognitive task). ERP studies of recognition memory reveal differences in activity between correctly identified studied (old) and unstudied (new) stimuli, referred to as ERP old/new effects (see Friedman & Johnson, 2000; Rugg & Curran, 2007, for reviews). In particular, an early (300 500 ms post-stimulus onset) modulation maximal over mid frontal scalp electrodes is associated with familiarity, while a later (500 700 ms) modulation maximal over left parietal scalp is linked with recollection. The mid frontal and left parietal old/new effects have been functionally dissociated by a number of task (e.g., Rugg et al., 1998) and stimulus (e.g., Greve, Van Rossum, & Donaldson, 2007) manipulations, providing strong evidence that the effects reflect distinct cognitive operations. While the left parietal effect appears to provide an index of recollection (Rugg & Yonelinas, 2003), the precise functional significance of the mid frontal effect remains contested, with some theorists arguing that it reflects familiarity (Rugg & Curran, 2007) and others contending that it may reflect conceptual priming processes that sometimes co-occur with episodic retrieval, particularly when words are used as stimuli (Voss & Paller, 2006; Paller, Voss, & Boehm, 2007). The left parietal and mid frontal ERP old/new effects have also been dissociated from a number of other old/new effects, including a late posterior negativity (LPN, which has been associated with heterogeneous cognitive functions such as retrieval fluency and action monitoring; see Herron, 2007) and a late right frontal effect (typically associated with post-retrieval monitoring; Hayama, Johnson, & Rugg, 2008). Nonetheless, whilst other ERP old/new effects are evident, the majority of ERP retrieval studies have been interpreted within a dual process framework, and a wide range of evidence suggests that the left parietal effect provides a generic index of recollection primarily because it has been observed with different stimulus materials (e.g., words Donaldson & Rugg, 1998; line drawings Curran & Cleary, 2003; landscape/object compound stimuli Tsivilis, Otten, & Rugg, 2001) and for information that is presented in different modalities (Schloerscheidt & Rugg, 2004). Thus, consistent with dual process models, the left parietal effect is typically believed to be neither material- nor modality-specific and is considered to provide an index of core cognitive operations related to recollection. Despite the foregoing evidence, in a recent study MacKenzie and Donaldson (2007) reported that recollection elicited by face retrieval cues was associated with an anterior old/new effect (400 600 ms; see Donaldson & Curran, 2007). In this experiment participants studied a series of photographs of faces, each presented with an auditory name. At test, old and new faces were presented, and participants made an initial old/new decision, and for faces judged old, were asked to report on any contextual information about the study episode that they retrieved. For recognition supported by familiarity, the ERP old/new effect was manifest over posterior scalp only (replicating the findings of Yovel & Paller, 2004), suggesting that recognition memory for faces is associated with a novel ERP signature of familiarity (and hence that there are multiple ERP signatures of familiarity). We note, however, that this finding could be interpreted as evidence against a familiarity account of the mid frontal old/new effect, providing indirect support for a priming interpretation. More importantly for present purposes, recollection-related ERPs elicited by face stimuli revealed a clear anterior old/new effect; this effect was larger when the associated names were retrieved than when other specific contextual information was retrieved. Although associated names were verbally reported and verified in the name condition, there was no verification of retrieved information in the other specifics condition; given that the two recollection-related effects were associated with the same scalp distribution the requirement to report on retrieved information did not affect the old/new effect in a qualitative manner. Critically, the anterior old/new effect was absent when contextual information could not be retrieved, suggesting a specific link with recollection. Although MacKenzie and Donaldson (2007) was the first ERP study to describe a recollection-related anterior old/new effect, a comparable neural correlate of recollection appears to be present in previous experiments using non-verbal stimuli (Duarte, Ranganath, Winward, Hayward, & Knight, 2004, Fig. 8), including faces (Paller, Gonsalves, Grabowecky, Bozic, & Yamada, 2000, Fig. 4; Yovel & Paller, 2004, Fig. 2), leaving open the possibility that recollection may be supported by different cognitive operations under certain circumstances. In particular, a recent study compared face and word retrieval using a simple old/new recognition decision (Yick & Wilding, 2008). This study reported more anteriorly distributed old/new effects for faces than words; however, a clear functional interpretation of the anterior effect was not possible because trials where recognition was supported by recollection were not isolated, and because faces and words were encoded separately. Despite evidence of anterior old/new effects, the proposal that recollection is associated with material-specific neural correlates cannot be accepted easily. Importantly, to date no ERP study has directly compared recollection elicited by word and face cues whilst holding the content of encoding episodes constant across face and name test trials (and hence limiting the possibility that differences at encoding acted as a confound). Thus, to add weight to this proposal, the present study aims to replicate the anterior recollection effect, using a similar paradigm as in MacKenzie and Donaldson (2007), but explicitly comparing recollection elicited by different types of retrieval cue. In the present experiment participants studied a series of compound visual stimuli, each consisting of a face name pair. Later, at test, a single element from each pair was presented, intermixed with unstudied faces and names. Importantly, each study episode was only ever probed once, with either the face or the name being presented as a retrieval cue. Participants were required to make an old/new discrimination for each test item, and made secondary remember/familiar decisions for each item judged old. These response options support inferences about the relative contributions of recollection and familiarity to test performance, and allow the identification of trials where recognition was supported by recollection. Given previous findings, we predicted that materialspecific neural correlates of recollection would be observed across cue types contrary to the pervasive view that recollection is supported by one homogenous set of cognitive operations. 2. Methods Twenty-six right-handed native English speakers gave informed consent and took part in the study, which was approved by the Psychology Department ethics committee at the University of Stirling. Participants reported having normal or corrected-to-normal vision, and were paid 5 per hour. Data from two participants were discarded due to contamination with ocular artifacts. Of the remaining sample, 20 participants with a mean age of 20 years (range: 18 28) contributed a sufficient number of trials for ERPs to be formed for remember responses to correctly recognized faces and names, and therefore data from these 20 participants are reported here. Faces and names were shown on a 17 in. LCD colour monitor; stimuli were presented and behavioural data were recorded with E-Prime (Psychology Software Tools; http://www.pstnet.com). Participants sat on a chair approximately one meter away from the monitor, with a button box on a desk in front of them. All faces were of young Caucasian individuals who did not wear any jewellery, glasses or facial hair. Two hundred and sixteen faces were presented during the experiment, and an additional four faces were used in a practice phase. Facial images were masked to
G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx 3 Fig. 1. Schematic depiction of study and test trial sequences. On each study trial, a grey fixation cross was replaced after 1000 ms by a face name compound stimulus, which was presented for 2000 ms. After stimulus offset, participants were prompted to rate the fit between face and name with a binary judgment, i.e., good fit or bad fit. There were two types of test trial, face trials and name trials, which were pseudo-randomly intermixed. On each test trial, a grey fixation cross was replaced after 1500 ms with either a face or a name, which was presented for 500 ms. Test items were either old (studied) or new (unstudied). A black screen followed test stimulus offset, and remained for 2000 ms during which time participants pressed either old or new. If a test item was endorsed as new then the trial terminated; if a test item was endorsed as old then a screen prompted participants to introspect upon the quality of their memory for the study episode. If recognition was accompanied by retrieval of specific information from the study episode then participants were required to press remember, which terminated the trial. If recognition was not accompanied by retrieval of specific information from the study episode then participants were required to press familiar, which terminated the trial. remove background, hair and ears, before being resized and positioned in the centre of the display. To eliminate gross differences in luminance between individual stimuli, colour was transformed 25% towards average using Psychomorph (Tiddeman, Burt, & Perrett, 2001). Faces were presented against a black background, and subtended a maximum horizontal visual angle of 2 and a maximum vertical visual angle of 5. Names were presented in 18-point bold Courier New font: during the study phase, names were presented immediately below the faces, while during the test phase, names were presented in the centre of the display, subtending a vertical angle of 0.7 and a maximum horizontal angle of 3.9. Data were acquired during 18 study test blocks. Each block contained 12 unique face name pairings presented at study, and was followed by six studied faces and six studied names intermixed with six new stimuli (three faces and three names) at test. There was a self-paced break between the study and test phases, and the test phase was initiated after the experimenter examined the EEG to ensure a clean signal was being recorded, and to prevent any contribution of short-term memory to performance. Each list contained an equal number of male and female faces and names. The test status of the stimuli was rotated across participants such that each stimulus had an equal chance of being new at test. Within each block there was random selection of stimuli to counter against order of presentation effects. The study and test phase procedures are illustrated in Fig. 1. In the study phase, each trial began with a grey fixation cross (+), which was presented in the centre of the screen against a black background for 1000 ms. The fixation cross was followed immediately by a face name presentation, which lasted for 2000 ms. After stimulus offset, the screen went blank and participants made a binary judgment as to whether the face fit the name. Participants responded with one of two buttons, and their response terminated the trial. Participants were told that the judgment was arbitrary, and that the task was designed to help them remember face name associations for the test phase. In the test phase, each trial began with a grey fixation cross, which was presented against a black background for 1500 ms. The fixation cross was replaced by a face or name presentation that lasted for 500 ms. After stimulus offset, a blank screen was displayed for 2000 ms, while participants indicated whether they thought the stimulus was old or new by pressing one of two buttons. If a new response was made, the trial terminated. If an old response was made, the blank screen was followed by a prompt indicating two response options: remember and familiar. Participants were asked to respond remember if they retrieved any information associated with the stimulus from the study episode and to respond familiar if they did not retrieve any specific information about the stimulus. The remember/familiar instructions were modeled on the other specifics/no specifics instructions described by Yovel and Paller (2004); participants were given the opportunity for questioning, to ensure that participants fully understood the distinction. The participant s response terminated the trial. The subsequent response options allowed trials where recognition was supported by recollection to be sorted for averaging EEG into ERPs. Recollection was inferred on trials where participants made a remember response to a studied stimulus. During testing, EEG was recorded from 62 silver/silver chloride electrodes embedded in an elasticized Quick-Cap (Neuromedical Supplies: http://www. neuro.com). Electrodes used were: FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8, FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8, PO7, PO5, PO3, POZ, PO4, PO6, PO8, CB1, O1, OZ, O2 and CB2. All channels were referenced to an electrode positioned between CZ and CPZ; two further electrodes were placed on the mastoid processes. Electrode positions were based on the extended International 10 20 system (Jasper, 1958). Electro-oculogram (EOG) electrodes were placed above and below the left eye, and on the outer canthi. Data were recorded and analyzed using Scan 4.3 software (http://www.neuro.com). Impedances were below 5 k before recording commenced. The data were band pass filtered between 0.1 and 40 Hz and digitized at a rate of 250 Hz (4 ms/point). Data were re-referenced off-line to recreate an average mastoid reference. EEG was segmented into 2100 ms epochs, starting 100 ms before stimulus onset. Ocular artifacts were removed using a regression procedure, and trials were excluded from the averages if drift exceeded ±75 V (measured by the difference between the first and last data points in the epoch) or where activity in any of the EEG channels anywhere in the epoch exceeded ±100 V. Waveforms were smoothed over a 5-point kernel. To ensure a good signal-to-noise ratio, a minimum of 16 artifact-free trials per condition was set as a criterion before an individual participant s data were included in grand-average ERPs. Grand-average ERPs were formed for four conditions: face remember, face correct rejection, name remember and name correct rejection. The average number of trials in these conditions was 45, 32, 51 and 43, respectively. Given that the focus of the experiment was on identifying neural correlates of recollection, during piloting an attempt was made to foster a high number of remember responses at the expense of familiar responses. Accordingly, an insufficient number of participants
4 G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx Table 1 Remember hit Familiar hit Remember FA Familiar FA Face 0.48 (0.16) 0.31 (0.13) 0.06 (0.06) 0.22 (0.14) Name 0.57 (0.14) 0.23 (0.12) 0.01 (0.01) 0.04 (0.03) contributed enough trials to examine ERPs for familiar responses across faces and names. The old/new effects evoked by remember responses were first characterized for the face and name conditions separately, and then magnitude and topographic comparisons were made between the conditions. The analysis of ERP data was directed at frontal and parietal locations where old/new effects associated with recollection and familiarity have been found previously. For the initial amplitude analyses, ANOVA was performed with the factors of old/new (face remember/face correct rejection; name remember/name correct rejection), location (frontal/parietal), hemisphere (left/right) and site (superior/medial/inferior). For the magnitude analyses, subtraction waveforms were derived and the old/new factor was replaced by a factor of condition (face/name). The specific electrodes included in the analysis were: F1/3/5/2/4/6 and P1/3/5/2/4/6. Significant interactions involving the old/new factor were followed up by ANOVA performed on data from each location separately, identifying the locus of differences in the distribution of effects. The Greenhouse Geisser correction for breaches of the sphericity assumption was used as appropriate. As a final step in our analysis strategy, for each cue type and latency period, the magnitude of the old/new difference was examined numerically to determine the location of the maximal old/new difference; paired-samples t-tests were performed on the electrode where the effect was maximal to assess whether the effect was reliable at this location. Finally, data were rescaled according to the Max Min method (McCarthy & Wood, 1985) so that topographic analyses were not confounded by amplitude differences. 3. Results 3.1. Behaviour Mean probabilities for hits and false alarms are plotted in Fig. 2(panel a), which shows an equivalent hit rate across cue types and a higher false alarm rate for faces than names. ANOVA with factors of cue type (face/name) and status (hit/false alarm) revealed main effects of cue type [F(1,19) = 26.7; p < 0.001] and status [F(1,19) = 826.4; p < 0.001], along with an interaction between the factors [F(1,19) = 31.4; p < 0.001], reflecting the differential false alarm rate [t(19) = 6.5; p < 0.001]. Sensitivity (Pr) was measured by subtracting the probability of false alarms from the probability of hits separately for face (0.51) and name (0.74) cues; a pairedsamples t-test found that sensitivity is lower for faces than for names [t(19) = 16.8; p < 0.001]. Response times for the initial old/new discrimination are plotted in Fig. 2(panel b). Hits were faster for faces than for names, while all other responses for faces were slower than for names. ANOVA with factors of cue type (face/name) and response (hit/correct rejection/miss/false alarm) revealed an interaction between the factors [F(1.8,30.6) = 4.9; p < 0.05], reflecting significant differences in reaction times to faces and names for hits [t(19) = 3.7; p = 0.001] and misses [t(19) = 4.2; p < 0.001], but not for correct rejections or false alarms. 1 The raw probabilities of hits and false alarms attracting remember/familiar responses are shown in Table 1. The hit data were rescaled under the independence assumption (IRK, see Yonelinas & Jacoby, 1995) and are plotted in Fig. 2(panel c). As can be seen, names were more likely than faces to be recollected, while faces appear more likely than names to be judged familiar. ANOVA with factors of cue type (face/name) and process estimate (rec- 1 It has been noted that response times (RTs) are rather long in this study, and we feel that the reason why RTs are longer in the present study than in other item recognition studies reflects the use of a two-stage decision task. In a recent item recognition study for faces, a mean RT for 989 ms was observed (Yick & Wilding, 2008) while in a previous two-stage decision task for faces a mean RT of 1277 ms was observed (Yovel & Paller, 2004). ollection/familiarity) revealed an interaction between the factors [F(1,19) = 11.2; p < 0.01], and subsidiary paired-samples t-tests found that the only significant difference was between the estimates of recollection for names and faces [t(19) = 2.5; p < 0.05]. This pattern of results indicates that name recognition receives greater input from recollection than face recognition, while familiarity contributes equally across conditions. Response times for the correctly recognized faces and names separated by subsequent remember/familiar response are plotted in Fig. 2(panel d). As can be seen, remember responses are faster for faces than for names, while there is little difference in response time for familiar responses. ANOVA with factors of cue type (face/name) and response (remember/familiar) revealed main effects of cue type [F(1,19) = 9.4; p < 0.01] and response [F(1,19) = 18.2; p < 0.001] along with an interaction between the factors [F(1,19) = 21.2; p < 0.001]. The effect of cue type reflects faster responses for faces than for names, and the effect of response reflects faster remember responses than familiar responses. Paired-samples t-tests identified differences in response time between remember responses for faces and names [t(19) = 6.5; p < 0.001] and between remember and familiar responses for faces [t(19) = 7.8; p < 0.001], while all other comparisons were non-reliable. This pattern of results demonstrates that the interaction between cue type and response is due to remember responses for faces being faster than any other response category. In sum, the behavioural data indicate that recollection contributes more to name recognition than face recognition. It is important to note, however, that we are confident that recollection supported hit performance for faces primarily because the estimates of the contribution of familiarity to recognition performance were equivalent for faces and names. False alarm rates are also informative with respect to the contribution of familiarity to hits. Across cue types there were more false alarms receiving familiar responses than remember responses. In fact, the rates of remember false alarms were very low, suggesting that participants used the remember/familiar response options correctly, and tended to make a familiar response when recognition was primarily based on familiarity. On this basis it seems likely that the participants used the response options correctly for studied faces and names, and it therefore seems reasonable to conclude that recollection supported hits for both faces and names. 3.2. Electrophysiology Grand-average waveforms were formed for correct remember and new responses to faces and names. Waveforms were first quantified into two a priori time windows where episodic retrieval related effects for faces have been observed previously: 300 500 and 500 700 ms (Yovel & Paller, 2004; Curran & Hancock, 2007; MacKenzie & Donaldson, 2007). To assess whether apparent differences in the old/new effects for faces and names result from variations in timing across conditions, two further time windows were analyzed: 100 300 and 700 900 ms. 3.3. Face cues The top panel of Fig. 3 shows ERPs for correctly recognized old faces given remember responses along with correctly rejected new faces, at the frontal and parietal electrode sites used for analysis. As can be seen, the old waveform becomes more positive-going than the new waveform from around 400 ms. The old/new divergence persists until roughly 700 ms, and appears to be greatest at frontal electrodes. In addition to early old/new effects, from approximately 700 ms onwards there appears to be a negative-going deflection at parietal electrodes, and a sustained positive-going deflection at right frontal electrodes.
G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx 5 Fig. 2. Behavioural performance. Panel (a) shows the mean hit and false alarm rates for faces and names; error bars show the standard error of the mean. There is little difference in the hit rate, while there is a greater false alarm rate for faces than for names. Panel (b) shows mean response times for faces and names; error bars show the standard error of the mean. Hits were associated with shorter response times for faces than for names, while misses were associated with longer response times for faces than for names. Panel (c) shows mean probabilities of remember and familiar responses to correctly recognized items for faces and names; error bars show the standard error of the mean. A double dissociation is evident, with more remember responses than familiar responses for name recognition, and more familiar responses than remember responses for face recognition. Panel (d) shows mean response times for correctly recognized faces and names separated by subsequent response; error bars show the standard error of the mean. Remember hits are associated with faster response times for faces than for names. The analysis of data from 100 to 300 ms failed to reveal any significant differences. However, from 300 to 500 ms the analysis revealed interactions between old/new, hemisphere and site [F(1.9,35.7) = 4.0; p < 0.05] and between old/new, location, hemisphere and site [F(1.5,28.7) = 7.1; p < 0.01]. Reliable differences between the waveforms were only present at the frontal location, where an interaction between old/new, hemisphere and site [F(1.6,30.7) = 8.0; p < 0.01] was observed. This three-way interaction reflects an effect that is largest at inferior sites on the left hemisphere and becomes steadily smaller from left to right across the scalp. No differences were observed at the parietal location. Examination of the data found that the effect is only reliable at the F5 electrode [t(19) = 2.5; p < 0.05]. The analysis of data from 500 to 700 ms revealed a main effect of old/new [F(1,19) = 22.8; p < 0.001] and a marginally significant interaction between old/new, location and hemisphere Fig. 3. Grand average ERP waveforms are shown for representative frontal and parietal electrodes. ERPs are shown for correctly recognized studied items given remember responses (old) plotted along with correct rejections (new) for faces and names. The depicted epoch begins 100 ms pre-stimulus onset and ends 900 ms post-stimulus onset. Scale bars indicate the magnitude of activity (in microvolts) and the time course of activity (in milliseconds). The top of the figure shows the old/new effect for faces: the old waveform is more positive-going than the new waveform from roughly 400 700 ms post-stimulus onset, and the difference between waveforms is greatest at the frontal location. The bottom of the figure shows the old/new effect for names: the old waveform is more positive-going than the new waveform from roughly 300 700 ms post-stimulus onset, and the difference between waveforms is greatest at the left parietal electrode.
6 G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx [F(1,19) = 3.9; p = 0.06], reflecting the right-lateralization of the effect at the frontal location (subsidiary analysis of data from the frontal location revealed a main effect of old/new [F(1,19) = 18.7; p < 0.001] and a marginal interaction between old/new and hemisphere [F(1,19) = 4.2; p = 0.054], while no hemispheric differences are observed at the parietal location, where subsidiary analysis revealed a main effect of old/new only [F(1,19) = 9.8; p < 0.01]). Examination of the data found that the effect is maximal at F2 [t(19) = 4.4; p < 0.001], suggesting that remembering faces elicits an anterior effect. Finally, the analysis of data from 700 to 900 ms revealed an interaction between old/new and site [F(1.1,20.8) = 6.1; p < 0.05], reflecting a negative-going deflection for the old waveform at superior sites compared to a positive-going deflection at inferior sites. In addition, the analysis revealed an interaction between old/new, location and hemisphere [F(1,19) = 5.9; p < 0.05]: the face effect is larger on the right hemisphere than on the left hemisphere at the frontal location (subsidiary analysis revealed an interaction between old/new and hemisphere [F(1,19) = 6.3; p < 0.05]), while no hemispheric differences were observed at the parietal location. Subsidiary analysis of data from the parietal location revealed an interaction between old/new and site [F(1.1,21.6) = 7.1; p < 0.05], reflecting a negative-going deflection for the old waveform that is greater at superior sites than at inferior sites. Examination of the data found that the effect is maximal at F6 [t(19) = 2.6; p < 0.05]. Remembering faces is associated with two separate effects during this time window: a late posterior negativity (LPN) and a late right frontal effect. This pattern of results demonstrates that remember responses for face cues are associated with frontally distributed positive-going old/new effects. From 300 to 500 ms, a small old/new difference with a left-lateralized inferior frontal distribution was observed, but was only reliable at one single electrode (F5). More importantly, remembering faces elicits an anterior effect with a right superior frontal maximum from 500 to 700 ms, and from 700 to 900 ms the onset of a late right frontal effect can be observed. 3.4. Name cues The lower panel of Fig. 3 shows ERPs for correctly recognized old names given remember responses, along with corresponding correct rejections. The old waveform is more positive-going from around 300 to 700 ms. The effect appears to be most pronounced at parietal locations, where the difference between waveforms is greatest on the left hemisphere. The analysis of data from 100 to 300 ms revealed a main effect of old/new [F(1,19) = 4.5; p < 0.05], but subsidiary analysis of data from the frontal and parietal locations separately failed to identify any significant differences. Examination of the data revealed that the effect is maximal at F3 [t(19) = 2.5; p < 0.05], suggesting that remembering names elicits an early manifestation of the mid frontal effect. The analysis of data from 300 to 500 ms revealed a main effect of old/new [F(1,19) = 10.2; p = 0.05] and interactions between: old/new and hemisphere [F(1,19) = 9.7; p < 0.01]; old/new, location and site [F(1.4,25.9) = 7.5; p < 0.01]; and between old/new, location, hemisphere and site [F(1.7,32.8) = 11.3; p < 0.001]. The four-way interaction was examined with subsidiary analysis of data from the frontal and parietal locations separately. At the frontal location, the analysis revealed a main effect of old/new [F(1,19) = 9.4; p < 0.01], along with an interaction between old/new and site [F(1.4,26.2) = 7.6; p < 0.01], reflecting a bigger effect at superior sites. The analysis of parietal sites revealed a main effect of old/new [F(1,19) = 5.8; p < 0.05], and interactions between old/new and hemisphere [F(1,19) = 26.3; p < 0.001] and between old/new, hemisphere and site [F(1.2,22.2) = 7.1; p < 0.05]. The effect is only present on the left hemisphere, where it is bigger at inferior sites. Examination of the data found that the effect is maximal at F1 [t(19) = 3.9; p = 0.001], suggesting that remembering names elicits a mid frontal effect. The analysis of data from 500 to 700 ms revealed a main effect of old/new [F(1,19) = 22.6; p < 0.001] along with interactions between old/new, location and hemisphere [F(1,19) = 8.3; p < 0.01] and between old/new, location, hemisphere and site [F(1.6,29.7) = 8.2; p < 0.01]. Subsidiary analysis revealed a main effect of old/new [F(1,19) = 6.5; p < 0.05] at the frontal location compared to a main effect of old/new [F(1,19) = 27.0; p < 0.001] and an interaction between old/new and hemisphere [F(1,19) = 9.9; p = 0.005] at the parietal location, reflecting a bigger effect on the left hemisphere. Examination of the data found that the effect is maximal at P3 [t(19) = 6.6; p < 0.001], suggesting that remembering names also elicits a left parietal old/new effect. The analysis of data from 700 to 900 ms revealed a main effect of old/new [F(1,19) = 5.6; p < 0.05] along with interactions between old/new and site [F(1.1,21.4) = 6.1; p < 0.05] and between old/new, location and site [F(1.1,21.8) = 6.4; p < 0.05]. These interactions reflect a negative-going deflection for the old waveform that is reliable at the parietal location [F(1,19) = 13.1; p < 0.01] but not at the frontal location; this negativity is bigger at superior sites than at inferior sites (interaction between old/new and site at the parietal location [F(1.1,21.5) = 12.2; p < 0.01]). Examination of the data revealed that the effect is maximal at P2 [t(19) = 4.3; p < 0.001], suggesting that remembering names elicits an LPN. In sum, this pattern of results suggests that remembering names elicits a mid frontal effect (maximal from 300 to 500 ms) and a left parietal effect (maximal from 500 to 700 ms). From 700 to 900 ms, the positive-going deflection of the old waveform can no longer be observed, while an LPN is present over parietal electrodes: given that the LPN is not the focus of this paper, no further consideration will be given to this effect. 3.5. Magnitude comparisons Difference waveforms were computed by subtracting the correct rejection waveform from the remember waveform for both faces and names. These difference waveforms reflect the magnitudes of the old/new differences and allow inferences to be drawn concerning the relative activity of underlying cognitive operations across conditions. Data were quantified into the same four latency periods used for the amplitude analyses, and were submitted to ANOVA with factors of condition (face/name), location (frontal/parietal), hemisphere (left/right) and site (superior/medial/inferior). During the 100 300 ms latency period, the analysis revealed a four-way interaction [F(1.7,32.5) = 4.3; p > 0.05]. Subsidiary analyses were performed on data from the frontal and parietal locations separately; while no differences were observed at the frontal location, an interaction between condition, hemisphere and site was observed at the parietal location [F(1.3,24.5) = 4.4; p < 0.05]. This interaction appears to reflect differences in distribution between the two effects: the name effect is larger than the face effect on the left hemisphere, where it is maximal at inferior electrodes; by contrast, on the right hemisphere the name effect is larger than the face effect at the superior electrode only, while the face effect is larger than the name effect at the medial electrode and little difference was observed across cue types at the inferior electrode. The name effect is larger than the face effect at all frontal and parietal electrodes except for P4. From 300 to 500 ms, the analysis revealed an interaction between condition and hemisphere [F(1,19) = 6.2; p < 0.05], which reflects the right-lateralization of the face effect compared to the left-lateralization of the name effect. In addition, the analysis revealed an interaction between condition, location, hemisphere
G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx 7 and site [F(1.7,32.0) = 19.6; p < 0.001]. Subsidiary analyses revealed an interaction between condition, hemisphere and site at the frontal location [F(1.4,26.6) = 6.8; p < 0.01], reflecting differences on the left hemisphere, where the face effect is maximal at the inferior electrode and the name effect is maximal at superior electrode (interaction between condition and site [F(1.2, 22.4) = 5.9; p < 0.05]). At the parietal location, subsidiary analyses revealed an interaction between condition and hemisphere [F(1.19) = 13.3; p < 0.01], reflecting the right-lateralization of the face effect compared to the left-lateralization of the name effect. In addition, at the parietal location an interaction between condition, hemisphere and site was observed [F(1.3,25.6) = 7.6; p < 0.01], reflecting differences on the right hemisphere, where the face effect is maximal at the medial electrode while the name effect is maximal at the superior electrode (interaction between condition and site [F(1.3,23.8) = 4.8; p < 0.05]. As in the 100 300 ms latency period, the name effect is larger than the face effect at all frontal and parietal electrodes except for P4. During the 500 700 ms latency period, the analysis revealed a four-way interaction [F(1.8,33.4) = 7.0; p < 0.01]. Subsidiary analyses were performed on data from the frontal and parietal locations separately. The face effect is larger than the name effect at the frontal location, while at the parietal location the name effect is larger; furthermore, at the parietal location, an interaction between condition and hemisphere was observed [F(1,19) = 6.0; p < 0.05], reflecting a face effect with a bilateral distribution compared to a left-lateralized name effect. The overall four-way interaction appears to reflect differences in distribution that are restricted to the right hemisphere, where the face effect is maximal at the superior electrode at the frontal location (relative to a name effect that is roughly equivalent across sites); while at the parietal location the face effect is maximal at the medial electrode whereas the name effect is maximal at the superior electrode. In contrast to the two earlier latency periods, the name effect is only larger than the face effect at left parietal electrodes, while the face effect is larger at the frontal location and at right parietal electrodes. Finally, from 700 to 900 ms, the analysis revealed an interaction between condition and hemisphere [F(1,19) = 6.1; p < 0.05], reflecting the right-lateralization of the face effect relative to the leftlateralization of the name effect. In addition, the analysis revealed an interaction between condition, location and hemisphere [F(1,19) = 6.3; p < 0.05]. Subsidiary analyses identified the aforementioned laterality differences at the frontal location (interaction between condition and hemisphere [F(1,19) = 7.8; p < 0.05]) while no differences were observed at the parietal location. The face effect is larger than the name effect at all frontal and parietal electrodes. 3.6. Topographic analysis Fig. 4 shows the distributions of the face and name remember effects. For faces, the most pronounced activity can be observed over right superior frontal electrodes from 500 to 700 ms, while for names classic mid frontal (300 500 ms) and left parietal (500 700 ms) effects can be observed. The most important topographic feature concerns the differential activity from 500 to 700 ms when faces elicit an anterior effect and, by contrast, a left parietal effect can be observed for names. Finally, the topography of the face effect appears to shift from a superior frontal distribution (500 700 ms) to an inferior, right-sided frontal distribution (700 900 ms), reflecting the onset of the late right frontal effect. Topographic analyses of rescaled difference waveforms were performed to assess whether faces and names are associated with qualitatively distinct old/new effects. The first analysis compared the face and name conditions to see if there is any evidence that the effects present from 500 to 700 ms have different distributions. Data were submitted to ANOVA with the factors of cue type (face/name), location (frontal/parietal), hemisphere (left/right) and Fig. 4. Topographic distribution of recollection old/new effects for faces and names. Each cartoon shows the distribution of the difference between correctly recognized old items given remember responses and correctly rejected new items, averaged over a 200 ms time period. The front of the head is at the top of each map, and the left hemisphere is on the left-hand side. Each dot represents a recording electrode. The scale bar indicates the range of activity (in microvolts). The distribution of the face recollection effect has an anterior maximum, whereas the distribution of the name recollection effect has topographically dissociable mid frontal (300 500 ms) and left parietal (500 700 ms) maxima. site (superior/medial/inferior). The analysis revealed a four-way interaction [F(1.7,33.1) = 7.5; p < 0.01], reflecting the fact that the face effect is bigger on the right hemisphere at superior, frontal sites, whereas the name effect is bigger on the left hemisphere at medial, parietal sites. This topographic dissociation indicates that the anterior and left parietal effects had different scalp distributions, providing evidence that remembering can be associated with a different pattern of underlying cognitive operations for faces and names. A second analysis was performed to assess whether the distribution of the face recollection effect (500 700 ms) differs from the distribution of the name recollection effect present in the early time window (300 500 ms). Although these two effects were present in different time windows, both were maximal at anterior sites and it is conceivable that the face recollection effect represents nothing more than a delayed manifestation of the early mid frontal effect seen for names. ANOVA was performed with factors of condition (face 500 700 ms/name 300 500 ms), location (frontal/parietal), hemisphere (left/right) and site (superior/medial/inferior). The analysis revealed an interaction between condition and hemisphere [F(1,19) = 10.0; p = 0.005], reflecting the fact that the face effect is right-lateralized whereas the name effect is left-lateralized. The analysis also revealed an interaction between condition, location, hemisphere and site [F(2.0,37.2) = 10.1; p < 0.001]. The name effect is left-lateralized at both locations whereas the face effect is right-lateralized at the frontal location and bilateral at the parietal location. In addition, at the parietal location, the name effect is smaller at inferior sites than at superior sites, whereas the face effect is larger at inferior sites than at superior sites. This pattern of topographic differences between the mid frontal effect for names (300 500 ms) and the anterior effect for faces (500 700 ms) is critical because it suggests that the anterior effect observed for faces cannot simply be explained as a delayed manifestation of the mid frontal effect observed for names. 2 2 The 300 500 ms latency period allows comparison of the face old/new effect with the classic mid frontal effect, but it does not represent the best fit for the face effect, which begins to diverge around 400 ms. We examined the possibility that
8 G. MacKenzie, D.I. Donaldson / Neuropsychologia xxx (2009) xxx xxx A third analysis was performed to assess whether the anterior old/new effect seen for faces (500 700 ms) differs from the late right frontal effect that emerges during the next time window (700 900 ms). ANOVA was performed with the factors of latency (500 700 ms/700 900 ms), location (frontal/parietal), hemisphere (left/right) and site (superior/medial/inferior). The analysis revealed an interaction between latency and site [F(1.1,20.3) = 13.8; p = 0.001], reflecting a superior distribution from 500 to 700 ms which differs from the inferior distribution of the effect observed from 700 to 900 ms. The analysis also revealed an interaction between latency, location, hemisphere and site [F(1.4,26.1) = 5.0; p < 0.05]. Interpretation of this interaction is confounded by the presence of an LPN in the later time window, and therefore subsidiary ANOVA were performed on data from the frontal location to assess whether the anterior and right frontal old/new effects can be dissociated. The analysis revealed a marginally significant interaction between latency, hemisphere and site [F(1.2,23.4) = 3.2; p = 0.08] that reflects a combination of no significant differences on the left hemisphere (F < 2.7) and highly significant differences on the right hemisphere (interaction between latency and site [F(1.1,21.4) = 9.6; p < 0.01]). Taken together, the topographic analyses reveal that the superior distribution of the anterior effect (500 700 ms) differs from the inferior distribution of the late right frontal effect (700 900 ms). This pattern of results suggests that different cognitive operations are engaged when the superior frontal effect is manifest compared to when the late right frontal and late posterior negativities are observed. Finally, an analysis was performed to assess whether the mid frontal (300 500 ms) and left parietal (500 700 ms) effects can be dissociated on topographic grounds. ANOVA was performed with factors of latency period (300 500 ms/500 700 ms), location (frontal/parietal), hemisphere (left/right) and site (superior/medial/inferior). The analysis revealed interactions between condition and location [F(1,19) = 7.8; p < 0.05], condition and hemisphere [F(1.19) = 4.4; p < 0.05], and between condition, location and site [F(1.1,20.8) = 5.4; p < 0.05]. These results imply that the mid frontal and left parietal effects are associated with different scalp topographies. 3.7. Summary Topographic analysis identified qualitative differences between the old/new effects associated with remembering faces and names. Face cues elicited superior frontal (500 700 ms) and late right frontal (700 900 ms) effects whereas name cues elicited mid frontal (300 500 ms) and left parietal effects (500 700 ms). The topographic analysis supports the view that different cognitive operations underlie the anterior face effect when compared to the mid frontal and left parietal effects observed for names, and when compared to the late right frontal effect seen for faces. Overall, this pattern of results implies that remembering faces recruits different retrieval processes from remembering names. 4. Discussion In this recognition memory experiment qualitatively distinct neural correlates of retrieval were observed for remember responses to face and name cues. During the critical 500 700 ms the dissociation between the anterior face effect and the mid frontal name effect is an artifact of inappropriate time window selection by segmenting waveforms into consecutive 50 ms latency periods from 300 to 700 ms. Amplitude analysis revealed that the face effect achieves significance at 450 ms with a right superior frontal distribution. The distribution remains stable from 450 700 ms, and therefore it is unlikely that the dissociation between the anterior face and mid frontal name effects rests upon inappropriate time window selection. latency period, faces were associated with an anterior ERP old/new effect, whereas names were associated with the classic left parietal effect linked with recollection. The anterior effect was topographically dissociable from both the mid frontal effect observed for names and the late right frontal effect observed for faces, suggesting that the anterior effect cannot be accounted for straightforwardly in terms of familiarity or post-retrieval monitoring. Most importantly, the anterior and left parietal effects had different scalp distributions, providing evidence that remembering can be associated with different underlying cognitive operations for faces and names. Below, we discuss the potential role of familiarity and postretrieval processing during remembering, and we then consider whether the anterior effect observed for faces represents a blend of mid frontal and late right frontal effects, or whether in fact recollection processing differs for face and name cues. 4.1. The role of familiarity and post-retrieval processing in remembering Remember judgments should in theory be supported by recollection, but most researchers would agree that familiarity also contributes towards remembering. The relationship between familiarity and recollection is disputed, however, with at least three models having some currency: exclusivity, independence and redundancy (Jones, 1987). Clearly, if either independence or redundancy are assumed, and familiarity can co-occur with recollection, then familiarity is likely to contribute to remembering. Here, the behavioural evidence suggests that this is the case; in fact, the estimates of the contribution of familiarity to recognition performance were equivalent across cue types. Recently, the remember/know procedure has been criticized for not providing a pure estimate of familiarity (Wais et al., 2008); however, the consensus remains that remember responses receive contributions from both recollection and familiarity. In comparing old/new effects associated with remember responses, therefore, it is to be expected that both recollection and familiarity signals should be present. For names, the remember old/new effect featured both mid frontal (300 500 ms) and left parietal (500 700 ms) effects. As described in the introduction, the mid frontal effect has been associated with both familiarity (Rugg et al., 1998; Curran, 2000; Curran & Cleary, 2003; Azimian-Faridani & Wilding, 2006) and conceptual priming (Voss & Paller, 2006; Paller et al., 2007), while the left parietal effect is associated with recollection (Wilding & Rugg, 1996; Allan, Wilding, & Rugg, 1998; Donaldson & Rugg, 1998; Johansson, Mecklinger, & Treese, 2004; Johnson, Minton, & Rugg, 2008). For faces, however, the remember old/new effect had an anterior maximum associated with recollection (MacKenzie & Donaldson, 2007) but no distinct component related to familiarity was observed. A posterior old/new effect has been linked to familiarity by studies examining recognition of faces in the absence of contextual retrieval (see Yovel & Paller, 2004; MacKenzie & Donaldson, 2007), and therefore it may be the case that a posterior familiarity signal is present in the current data for faces but that it is obscured by the anterior recollection effect. Whilst the behavioural data suggest equivalent rates of familiarity across cue types, there were fewer remember responses for faces than for names. Thus, although the topographic analyses reveal distinct ERP old/new effects for faces and names, it is possible that this reflects nothing more than the fact that recollection was not sufficiently engaged to elicit a reliable left parietal effect when faces were used as retrieval cues. By this account, in the absence of recollection, one might assume that the anterior effect seen for faces may therefore reflect a delayed mid frontal effect or the early onset of the late right frontal effect; that is, performance is supported by a combination of familiarity and post-retrieval processing. As mentioned above, however, topographic differences are present between the