Electrophysiological dissociation of the neural correlates of recollection and familiarity

Transcription

1 ava i l a b l e a t w w w. s c i e n c e d i r e c t. c o m w w w. e l s ev i e r. c o m / l o c a t e / b r a i n r e s Research Report Electrophysiological dissociation of the neural correlates of recollection and familiarity C. Chad Woodruff, Hiroki R. Hayama, Michael D. Rugg Center for the Neurobiology of Learning and Memory and Department of Neurobiology and Behavior, University of California, Irvine, Irvine, CA , USA A R T I C L E I N F O A B S T R A C T Article history: Accepted 2 May 2006 Available online 13 June 2006 Keywords: Confidence Dual process ERP Familiarity Recognition memory Recollection Event-related potentials (ERPs) were employed to investigate electrophysiological correlates of recognition memory in a task that allowed segregation of test items according to whether they were recollected (operationalized by introspective report) or, if recollection failed, their level of familiarity (operationalized by recognition confidence). The amplitude of a negative-going ERP deflection that onsets around 300 ms poststimulus varied inversely with familiarity strength. This effect was maximal over the left frontal scalp. It did not differ between the ERPs elicited by highly familiar versus recollected items, indicating that the recollection is not merely a consequence of strong familiarity. By contrast, a later positive deflection (onset ca. 500 ms post-stimulus) was enhanced in ERPs elicited by recollected relative to highly familiar items. This effect was maximal over the left posterior scalp and was insensitive to familiarity, as indicated by its absence in the contrast between items judged highly familiar versus highly unfamiliar. The findings constitute a double dissociation between the neural correlates of recollection and familiarity. Together with the results of a parallel functional magnetic resonance imaging study (A.P. Yonelinas et al., J. Neurosci. (2005), 25, ), they indicate that recollection and familiarity rely on qualitatively distinct neural systems and strongly support dual-process models of recognition memory Elsevier B.V. All rights reserved. 1. Introduction Several lines of evidence suggest that recognition memory is supported by two kinds of information (Yonelinas, 2002). Familiarity-based recognition reflects an acontextual sense that a test item has been previously experienced, whereas recognition based on recollection involves retrieval of contextually specific information. In one dual-process model (Yonelinas, 1994), familiarity and recollection provide independent bases for recognition. Information supporting familiarity is continuous and strength-like, whereas information supporting recollection has a threshold-like character. The validity of this and similar models is the subject of debate. Indeed, single-process models have been proposed which are claimed to give a better account of certain behavioral dissociations than dual-process formulations (Donaldson, 1996; Dunn, 2004; Slotnick and Dodson, 2005). In these models, recollection and familiarity depend on the same underlying process and reflect quantitative rather than qualitative differences in retrieved information. If recollection and familiarity are qualitatively distinct, the distinction should be evident at the neural level. In Corresponding author. address: mrugg@uci.edu (M.D. Rugg) /$ see front matter 2006 Elsevier B.V. All rights reserved. doi: /j.brainres

2 126 B R A I N R E S E A R C H ( ) support of this prediction, studies employing event-related potentials (ERPs) or functional magnetic resonance imaging (fmri) have reported that familiarity- and recollection-based judgments are associated with qualitatively different patterns of neural activity (Curran, 2000; Duarte et al., 2004; Eldridge et al., 2000; Henson et al., 2003; Rugg et al., 1998; for review, see Rugg and Yonelinas, 2003). A problem with the interpretation of these findings arises however because recollection-based judgments are usually made with high confidence, whereas the confidence of familiarity-based judgments varies widely (Yonelinas, 1994). Thus, studies employing tasks that permit only a binary segregation of recognition judgments confound the type and the confidence of the judgments. To overcome this problem, Yonelinas et al. (2005) introduced a modification to the widely employed Remember Know procedure (Tulving, 1985). The modified procedure required subjects to signal if a test item elicited recollection or, if not, to rate it on a scale ranging from confident old to confident new. Yonelinas et al. (2005) argued that neural activity associated with familiarity strength should covary with the level of confidence that an item was old. Activity associated with recollection, by contrast, should be revealed in the contrast between items endorsed as recollected and items confidently judged old in the absence of recollection. Using fmri, Yonelinas et al. (2005) reported that, when operationalized in this way, the neural correlates of recollection and familiarity were largely non-overlapping. Here, we report an ERP study employing the same procedure as Yonelinas et al. (2005). As noted, previous findings suggest that familiarity and recollection have distinct ERP correlates (Friedman and Johnson, 2000). Familiaritybased recognition is associated with attenuation of a frontal negative deflection peaking around 400 ms post-stimulus, whereas recollection is associated with enhancement of a later positive-going deflection maximal over the left parietal scalp (e.g. Curran, 2000). If this dissociation does not merely reflect differential confidence effects, it should be evident in the present study. Specifically, the frontal familiarity effect should covary with recognition confidence but should not be further enhanced for items endorsed as recollected. By contrast, the left parietal effect should be evident in the contrast between recollected items and those confidently endorsed as familiar. 2. Results 2.1. Behavioral performance The proportions of old and new items endorsed as recollected or assigned to the different confidence categories are shown in Table 1. As is evident from the table, there was a highly systematic relationship between study status and response category, such that old items predominated in the R and confident old categories whereas new items dominated the confident new category. ANOVA (factors of study status and response category) confirmed the reliability of this interaction (F 2.3,33.8 = 69.38, P < 0.001; degrees of freedom here and in all subsequent ANOVAs were Geisser Greenhouse corrected for non-sphericity). Table 1 Proportions of old and new items assigned to each response category and weighted RT associated with each category Reaction times (RTs) are also shown in Table 1. These were shorter for the extreme categories (R and confident new) than they were for the intermediate categories. Because very few or no responses were available for new items in the R and confident old categories, and for old items in the confident new category, an across-category ANOVA was performed on the weighted means of the RTs to the two classes of item. This revealed a main effect of response category (F 2.3,34.9 = 26.71, P < 0.001). Further analyses revealed (i) a significant difference between R and confident old RTs (F 1,15 = 16.03, P < 0.001); (ii) a significant difference across the four confidence categories (F 2.2,32.4 = 20.64, P < 0.001); and (iii) no significant difference between confident old and confident new RTs (P > 0.2) ERPs Recollect Conf Old Unconf Old Unconf New Conf New Old New RT The principal analyses focused on the ERPs elicited by the items assigned to each of the response categories regardless of study status (cf. Yonelinas et al., 2005) Familiarity effects Figs. 1 and 2 illustrate the grand average waveforms elicited by items accorded confident old and confident new responses, that is, items expected to differ maximally in familiarity. The waveforms were formed from a mean (range) of 33 (15 50) and 40 (18 56) trials respectively. Also shown in Fig. 2 are the waveforms elicited at a left frontal electrode by items attracting unconfident responses, which were formed from means of 35 (17 51) and 50 (30 91) trials for old and new judgments respectively. As is evident from the figures, the waveforms elicited by items confidently judged new demonstrate a frontally distributed negative-going deflection that appears to be attenuated in the waveforms for confident old items, especially over the left hemisphere. Following prior studies (Rugg et al., 1998; Curran, 2000), this putative effect of familiarity was quantified as the mean amplitude between 300 and 500 ms post-stimulus. An initial ANOVA on the electrode sites indicated in Fig. 1 (factors of response category (confident old vs. confident new), hemisphere, frontal/parietal location, and site) gave rise to a category hemisphere frontal/ parietal interaction (F 1,15 = 4.71, P < 0.05). Follow-up ANOVAs on each quadrant of electrode sites revealed a significant response category effect for the left frontal quadrant only (F 1,15 = 6.97, P < 0.025). The data in Fig. 2 suggest that the amplitude of left frontal ERPs in the ms latency range not only discriminates confident old from confident new responses but also varies monotonically with response confidence, as would be expected of a neural correlate of familiarity strength. To

3 127 Fig. 1 Grand average ERP waveforms from left and right frontal and parietal regions elicited by items accorded confident old or confident new judgments (regardless of accuracy). Insert shows an enlargement of the waveforms elicited from the F3 electrode site. assess whether the relationship between familiarity strength and frontal ERPs was statistically significant, we adopted an approach directly analogous to that employed in the prior fmri study (Yonelinas et al., 2005). For each subject, and at each left frontal electrode site, mean amplitude in the ms latency range was regressed against a dummy vector representing the 4 confidence levels. Given the null hypothesis of no relationship between amplitude and confidence, the across-subject mean of the ensuing regression coefficients has an expectation of zero. Contrary to this expectation, the mean differed significantly from zero at every left frontal site (t 15 = 1.75 to 3.05, 1-tailed 1 P < 0.05 to P < 0.005). The regional specificity of this effect was confirmed by an ANOVA contrasting the regression coefficients from each of the electrode quadrants illustrated in Fig. 1 (factors of hemisphere, frontal/parietal location, and site). This revealed a significant hemisphere frontal/parietal interaction (F 1,15 = 5.83, P < 0.05). As shown in Fig. 3, the regression coefficients were of greater magnitude over the left frontal scalp than over either the right frontal or the parietal regions. Collapsed across electrodes, the coefficient from the left frontal quadrant differed significantly from zero (t 15 = 2.53, 1-tailed P = 0.01); the coefficients from the remaining three quadrants did not significantly differ from zero. The foregoing analyses suggest that left frontal ERP amplitudes between 300 and 500 ms post-stimulus are sensitive to familiarity strength. These findings are consistent with proposals that frontal old/new effects in this latency range reflect familiarity-driven recognition memory. An alternative explanation for the present findings is possible, however. Because the ERPs were averaged without regard to the study status of the test items, the relative proportions of 1 One-tailed tests were conducted in light of the extensive prior evidence supporting an inverse relationship between the amplitude of the frontal negative deflection and familiarity (see Introduction). old and new words contributing to the ERPs from each response category systematically varied (see Table 1). Approximately 80% of the trials comprising the ERPs for items attracting confident old judgments came from old items. This proportion declined to approximately 50% for unconfident old judgments, 25% for unconfident new judgments and 14% for confident new judgments. A graded relationship between response category and left frontal ERPs would therefore be expected if ERPs were simply more positive-going for old than for new items regardless of familiarity strength. To address this issue, we contrasted ERPs according to study status of the test items while controlling for response category. For each category, all available trials were selected for the item type receiving the fewest endorsements (e.g., for confident old responses, this was invariably new items). An equal number of trials were then selected at random from those associated with the alternative item type. ERPs were formed for old and new items and collapsed across response categories, yielding waveforms that represent the activity elicited by each item class when equated for familiarity (as indexed by response category). One subject's waveforms were rejected from the analyses presented below because of a substantial baseline artifact, although inclusion of these data had no impact on the outcome. The mean numbers (and range) of trials forming the ERPs for old and new trials were 38 (21 64) and 38 (23 60) respectively. Fig. 4 illustrates the grand average waveforms from the F3 electrode site, along with those from the same site averaged according to familiarity strength (confident old vs. confident new). As is evident from the figure, when familiarity strength is equated across old and new items, the resulting old/new effect is small and far from statistical significance. A more comprehensive analysis contrasting the mean amplitudes of the ms latency regions of these ERPs failed to reveal any effect involving the factor of item type. This was the case regardless of whether the ANOVA was conducted on the data from all four electrode quadrants or from the left frontal electrodes alone (all P's > 0.3).

4 128 B R A I N R E S E A R C H ( ) Fig. 2 Left: Grand average ERP waveforms from the F3 electrode according to the nature of the associated recognition judgment. Right: Mean amplitude across-subjects of the ms latency region of the ERPs from F3 elicited by test items attracting each class of judgment. From left to right, the data represent confident old, unconfident old, unconfident new and confident new judgments respectively. Inspection of Fig. 1 indicates that, in addition to the early frontal old/new effect discussed above, confident old items also elicited a later-onsetting positive wave. This wave is maximal over the right lateral frontal scalp and sustained until the end of the recording epoch. The effect was quantified as the mean amplitude between 800 and 1900 ms post-stimulus. ANOVA (factors of response category, hemisphere, frontal/parietal location and electrode site) revealed a main effect of response category (F 1,15 = 4.61, P < 0.05) along with significant interactions between category and site (F 2.8,42.1 = 3.14, P < 0.05) and category hemisphere frontal/parietal location (F 1,15 = 13.47, Fig. 3 Mean of the magnitudes of within-subjects parameter estimates of the relation between the amplitude of the ms latency region and familiarity strength (as indexed by response confidence). The parameter estimates have been averaged across electrodes over left frontal (LF), right frontal (RF), left parietal (LP) and right parietal (RP) scalp. Bars represent standard error of the mean. P < 0.005). A follow-up ANOVA restricted to the lateral frontal electrodes revealed a significant category hemisphere interaction (F 1,15 = 8.08, P < 0.025). The waveforms elicited at the right lateral frontal site elicited by the items from all four confidence categories are illustrated in Fig. 5, where it can be seen that the positivity is seemingly restricted to the confident old response category. This impression was confirmed by ANOVA on the ms data for all categories (F 2.7,40.9 = 7.42, P < 0.001). Pairwise contrasts revealed significant differences between the confident old category and each of the other response categories (max. P < 0.005). Finally, in an analysis relevant to the question of the specificity of the parietal old/new effect to recollection (see below), the mean amplitude of the ms region of the waveforms elicited by items assigned confident old vs. confident new judgments was subjected to ANOVA (factors of response category, hemisphere, frontal/parietal location, and electrode site). In neither the overall ANOVA, nor in a subsequent ANOVA restricted to data from the parietal electrodes, did any effect involving the factor of response category approach significance (min. P > 0.1; see Fig. 1) Recollection Grand average waveforms for items assigned R and confident old responses are illustrated in Fig. 6 [mean trial number for R ERPs was 39 (16 86)]. The waveforms show little sign of diverging until around 500 ms post-stimulus, when the ERPs for R items become more positive-going. Over the parietal scalp, this positivity is relatively phasic and left lateralized. At frontal electrodes, by contrast, the effect is sustained until the end of the recording epoch and predominates over the right hemisphere. Following prior practice, the region of the waveform encompassing the phasic parietal positivity (which we assume to be an example of the much-studied

5 129 Fig. 4 Left: Grand average ERP waveforms from the F3 electrode for items judged confident old and confident new (familiar vs. unfamiliar). Right: Waveforms for old and new items equated for familiarity (old vs. new; see text). Insert: Mean amplitude difference (and standard error) in the ms latency region for familiar minus unfamiliar (F/U) and old minus new (O/N). The familiarity effect differed significantly from zero (1-tailed P < 0.02); the old/new effect was not significant (1-tailed P > 0.1). left parietal old/new effect, Friedman and Johnson, 2000) was quantified as the mean amplitude between 500 and 800 ms post-stimulus. In keeping with the impression given by Fig. 6, ANOVA of mean amplitudes fromthe mslatency region(factorsof response category, hemisphere, frontal/parietal location, and electrode site) revealed no effects involving response category (max. P > 0.1). By contrast, ANOVA of the ms region revealed a significant response category effect (F 1,15 = 19.68, P < 0.001), which was accompanied by a category frontal/ parietal hemisphere interaction (F 1,15 = 5.44, P < 0.05). These effects reflected the greater positivity of the ERPs elicited by R items and its tendency to reverse lateralization between frontal and parietal sites. However, whereas separate ANOVAs for data from the frontal and parietal sites each revealed reliable Fig. 5 Grand average ERP waveforms from the indicated electrode (F8) according to the nature of the associated recognition judgment. category effects (F 1,15 = 10.72, P < and F 1,15 = 18.92, P < respectively), in neither case were these accompanied by significant category hemisphere interactions. The frontal positive shift distinguishing R and confident old ERPs until the end of the recording epoch was characterized by the mean amplitude between 800 and 1900 ms poststimulus. ANOVA of these data revealed a significant response category effect (F 1,15 = 6.54, P < 0.025), accompanied by category hemisphere and category electrode site interactions (respectively, F 1,15 = 6.49, P < 0.025, and F 2.3,34.5 = 3.76, P < 0.05). These interactions reflect the tendency for the response category effect to be greatest over the right hemisphere and at sites near the midline Effects of response confidence The proposal that the contrast between R and confident old items identifies the neural correlates of recollection is predicated on the assumption that the outcome of the contrast does not merely reflect greater confidence for R responses than confident old responses. This assumption can be tested by comparing ERP effects associated with recollection with those due purely to response confidence. A qualitative estimate of effects due to confidence can be obtained from the contrast between ERPs elicited by items endorsed as confident vs. unconfident collapsed across the factor of study status. In this contrast, differences in activity due to familiarity should cancel, whereas effects associated with response confidence will be preserved. Fig. 7 illustrates the relevant waveforms. It is evident from the figure that the ERPs associated with confident responding are more positivegoing than those for non-confident responses, with confidence modulating a phasic parietal positivity in manner similar to recollection. ANOVA of the ms latency region of the parietal ERPs (factors of response category, hemisphere, and electrode site) gave rise to a significant effect of category (F 1,15 = 8.25, P < 0.025), confirming the reliability of

6 130 B R A I N R E S E A R C H ( ) Fig. 6 Grand average ERP waveforms from left and right frontal and parietal regions elicited by recollected items and items accorded confident old judgments. Insert shows an enlargement of the waveforms elicited from the F3 electrode site. the confidence effect at these electrodes. The question whether this effect reflects the activity of neural populations distinct from the populations responsible for the recollection effect illustrated in Fig. 6 is addressed in the next section Analysis of scalp topographies The scalp topographies of the effects due to familiarity in the ms latency region, and effects due to recollection and confidence in the ms region, are illustrated in Fig. 8. Initial ANOVAs contrasting these topographies were performed on subtraction data from the same electrode quadrants employed in the foregoing analyses. The data from each condition were range-normalized prior to analysis to eliminate confounding effects of global differences in amplitude (McCarthy and Wood, 1985) 2. To the extent that the topographies of two effects significantly differ, it can be concluded that the effects do not solely reflect differences in the timecourse or amplitude of a common neural population and are neurally (and, it is assumed, functionally) dissociable (Rugg and Coles, 1995). The ANOVA contrasting the familiarity and recollection effects revealed a significant interaction between type of memory, frontal/parietal location and hemisphere (F 1,15 = 7.28, P < 0.025). Follow-up analysis revealed a cross-over interaction for the left (but not the right) hemisphere, with the familiarity effects demonstrating a frontal maximum, and the recollection effects a parietal maximum (F 1,15 = 8.84, P < 0.01). ANOVA contrasting the topographies of the recollection and confidence effects gave rise to a significant interaction 2 The practice of amplitude normalization prior to topographic analysis has been criticized (Urbach and Kutas, 2002). The criticisms leveled at the practice by these authors that normalization results in error due to baseline artifact and the effects of residual noise apply only when data are normalized with respect to root mean square amplitude and are not relevant to data normalized with respect to range. It is the latter procedure that is employed routinely in our laboratory. between type of effect, frontal/parietal location and hemisphere (F 1,15 = 5.49, P < 0.05). In addition, the interaction of these three factors with electrode site approached significance (F 4.3,64.4 = 2.40, P < 0.06). Follow-up ANOVA of the data from the frontal electrodes revealed no effects. ANOVA of the parietal sites revealed an interaction between type of effect and hemisphere that approached significance (F 1,15 = 3.59, P < 0.08). The interaction was significant when the ANOVA was restricted to data from the three most lateral/posterior of the parietal sites (F 1,15 = 5.76, P < 0.05). In parallel with this effect, pairwise contrasts on mean amplitudes of the ms latency region at the same electrode sites revealed that ERPs to recollected items were of greater amplitude than those to items confidently recognized over both left and right hemispheres (max. P < 0.005). By contrast, the effects of response confidence were reliable over the right hemisphere only (P < 0.01). The familiarity and recollection effects illustrated in Figs. 1 and 5 both include modulation of a sustained frontal positivity. The question therefore arises whether the recollection effect is merely an exaggeration of the effect elicited by items confidently endorsed old or whether instead the two effects can be qualitatively dissociated. The scalp topographies of the effects for the ms latency region are illustrated in Fig. 9, where it can be seen that the effect for recollection has a more focal and midline distribution than does the effect for familiarity. ANOVA of the normalized data gave rise to a significant interaction between memory effect and electrode site (F 3.7,55.7 = 4.24, P < 0.01). The same effect was evident for data from the frontal sites alone (F 3.3,49.3 = 2.99, P < 0.05). 3. Discussion The findings converge with those of Yonelinas et al. (2005) to support the view that familiarity and recollection are neurally dissociable. The amplitude of a negative-going frontal

7 131 Fig. 7 Grand average ERP waveforms from left and right parietal regions elicited by items accorded confident or unconfident judgments. deflection peaking around 400 ms demonstrated an inverse relationship with familiarity strength but did not differ according to whether items were familiar or recollected. By contrast, a later-onsetting, left parietal positivity was enhanced for recollected items relative to those confidently judged old but did not differ for items accorded confident old versus confident new judgments. The double dissociation arising from this pattern of findings is difficult to reconcile with the notion that familiarity and recollection reflect variation in the strength of a common memory signal (Donaldson, 1996; Ratcliff et al., 1995; Slotnick and Dodson, 2005) Familiarity effect The negative deflection sensitive to familiarity strength corresponds to an ERP feature that has been referred to as the FN400 (Curran, 2000) or mid-frontal old/new effect (Tsivilis et al., 2001). A variety of evidence link this effect to familiarity-based recognition memory. Notably, the effect is elicited by familiarity-driven false alarms (Curran, 2000; Curran and Cleary, 2003) and, while insensitive to depth of study processing, nonetheless discriminates hits from misses (Rugg et al., 1998; Tsivilis et al., 2001). The present findings extend these observations in two ways. First, they demonstrate that the magnitude of the effect varies linearly with familiarity strength (as indexed by rated confidence; see also Curran, 2004). Second, they indicate that, when familiarity strength is matched, the effect is insensitive to the study status of the eliciting item. Together, these findings support the proposal that the frontal old/new effect is a correlate of explicit, familiarity-based recognition memory (Rugg et al., 1998; Curran, 2000). An alternative account of the mid-frontal ERP effect argues that, rather than reflecting familiarity, the effect is a neural correlate of conceptual priming, a form of implicit memory dissociable from processes that support recognition judgments (Yovel and Paller, 2004; Voss and Paller, 2006). By this argument, the encoding conditions that enhance the familiarity of a study item also promote conceptual priming, leading to a confound between the two classes of memory and a misspecification of the functional significance of the Fig. 8 Left: Scalp topography of the difference in mean amplitude of the ms latency region for ERPs associated with confident old versus confident new judgments. The nose is at the top. Data are range-normalized, maximal positivity is indicated by dark red, maximal negativity by dark blue, with the range (in microvolts) indicated under each plot. Middle: Scalp topography of the difference in mean amplitude of the ms latency region for ERPs associated with recollected items versus confident old judgments. Right: Scalp topography of the difference in mean amplitude of the ms latency region for ERPs associated with confident versus unconfident old judgments. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

8 132 B R A I N R E S E A R C H ( ) Fig. 9 Left: Scalp topography of the difference in mean amplitude of the ms latency region for ERPs associated with recollected items versus confident old judgments. Right: Scalp topography of the difference in mean amplitude of the ms latency region for ERPs associated with confident old versus confident new judgments. See legend to Fig. 8 for further details. mid-frontal effect as it is manifest in the contrast between old (recognized) and new (correctly rejected) items. Whereas it is doubtful that the effect is a real-time manifestation of the neural computations that underlie the familiarity strength of a test item (Curran, 2004; Tsivilis et al., 2001), the present findings provide strong grounds for thinking that the effect is more closely coupled to familiarity than it is to priming. Specifically, relative to ERPs elicited by items segregated according to their rated familiarity, the effect was much diminished, and far from significant, in ERPs elicited by old and new items that had been equated for familiarity (see Fig. 4). To the extent that priming and familiarity are independent, the effect should, if anything, have been enhanced in the familiarity-equated ERPs, and certainly not eliminated. Although the generality of the relation between familiarity and the mid-frontal effect remains to be fully delineated (for example, the relation may not hold for non-verbal test items that have no pre-existing memory representation; Yovel and Paller, 2004) and its precise functional significance remains obscure (Tsivilis et al., 2001), the present findings suggest that the effect can serve as a reliable index of familiarity in recognition memory tests typical of those conducted within the verbal learning tradition. Finally, it is noteworthy the frontal effect elicited by recollected items was comparable in magnitude to that elicited by items confidently judged old. This finding is consistent with one of the major tenets of dual-process models, namely, that recollection depends upon a form of mnemonic information that differs qualitatively, rather than quantitatively, from that underlying familiarity Recollection effect Relative to ERPs elicited by items confidently judged old, recollected items elicited a phasic positivity over the left parietal scalp. We interpret the positivity as an example of the left-parietal old/new effect described in prior studies of recognition memory and held to be a correlate of recollection (Curran, 2000; Friedman and Johnson, 2000; Rugg et al., 2002). The present findings support this proposal. Notably, the effect was absent in the contrast between items attracting confident old versus new judgments, indicating that it is insensitive to differences in familiarity strength. An alternative account of the present parietal effect should be considered, however. Some parietally distributed positivities (the P300 or P3 family of components) are sensitive to such factors as the probability of the eliciting event and the confidence with which it is categorized (Polich and Kok, 1995). Given the relative rarity of R responses and the high level of confidence that is invariably associated with recollection (Yonelinas, 1994), it is possible that the present effect includes a contribution from the generators of a generic P300 (Spencer et al., 2000). This possibility gains support from the finding that response confidence modulated the amplitude of parietal ERPs in the same latency range as did recollection (Fig. 7). For two reasons, however, we think it unlikely that the parietal recollection effect can be reduced to modulation of a generic P300. First, in a study that manipulated the probability of old versus new judgments (by varying the ratio of old and new items), the amplitude of the old/new ERP effect at left parietal electrode sites remained constant (Herron et al., 2003), indicating that the effect at these sites is insensitive to generic probability effects. Second, in the present study, the effects of confidence and recollection were topographically dissociable, the latter demonstrating a more pronounced left lateralization (Fig. 8). This finding indicates that recollection and confidence effects reflect neural generators that are at least partially distinct. Together, these findings suggest that old/new effects over the left parietal scalp reflect neural activity associated primarily with recollection. This is not to say, however, that the P300 makes no contribution to parietal ERP effects elicited by recollected items, especially at electrode sites near the midline or overlying the right hemisphere Late frontal effects In addition to the effects discussed above, recollection and familiarity were further dissociated by later-onsetting frontal positivities (Fig. 9). Relative to items confidently judged new, confident old items elicited a positive-going shift maximal over the lateral right frontal scalp. Recollected items, by contrast, elicited both this effect and an additional overlapping positivity with a more midline and anterior maximum. A right lateralized ERP retrieval effect (the right frontal old/ new effect ) was first highlighted by Wilding and Rugg (1996), who proposed that it reflected the engagement of postretrieval monitoring. According to one recent account (Rugg et al., 2002), monitoring is engaged (and the right frontal effect elicited) whenever the outcome of a retrieval attempt is ambiguous. This could occur because the retrieved information requires evaluation before a response can be selected (as in source memory tasks), or because the information is impoverished, causing uncertainty about the item's study status (as in the case of an unrecollected item eliciting a weak familiarity signal). This account has difficulty accounting for the present findings. The experience of recollection should

9 133 have been sufficient to permit an R response to be selected without further evaluation. Moreover, items accorded unconfident responses should have engaged monitoring to a greater extent than those attracting confident responses (cf. Henson et al., 2000). It is unclear how these disparities can be accommodated by the above account. In any case, the present findings demonstrate that the right frontal old/new ERP effect is neither neurally nor functionally homogeneous Neural generators of ERP effects A key question concerns the loci of the generators of the present ERP effects. Robust general methods for localizing the sources of scalp electrical fields have yet to be developed, but the question can to some extent be addressed informally. One source of information about the generators of the ERP effects comes from their scalp distributions. A second important line of evidence comes from functional neuroimaging studies (for review, see Rugg and Henson, 2002), notably the study that motivated the present experiment (Yonelinas et al., 2005). This study identified regions where neural activity varied linearly according to recognition confidence (putative familiarity effects), as well as regions where activity differed according to whether items were recollected or confidently judged old (putative recollection effects). Below, we ask whether any of these regions is a plausible generator of the functionally analogous ERP effect. Importantly, if an ERP and an fmri effect can be linked, useful information is gained not only about the intra-cerebral origin of the ERP effect, but also about the timecourse of the fmri effect. The scalp distribution of the ERP correlate of familiarity suggests an origin in left frontal cortex. This fits well with the finding of Yonelinas et al. (2005) that left inferior and anterior prefrontal cortex were among the most prominent regions where activity covaried with familiarity strength. Thus, the two sets of findings converge to suggest that the prefrontal cortex plays a role in familiarity-driven recognition (see also Duarte et al., 2005). The early onset of the ERP effects (ca. 300 ms) suggests that this role extends beyond the postretrieval monitoring operations that are held to onset relatively late in the course of a retrieval attempt (Schacter et al., 1997; Wilding and Rugg, 1996). As with the familiarity effect, the distribution of the putative ERP correlate of recollection, which was maximal over left parietal scalp, is suggestive of a generator in subjacent cortex. On the basis of numerous reports that recognized test items elicit enhanced lateral parietal activity (for a review, see Rugg and Henson, 2002; Wagner et al., 2005), it has been proposed previously that the parietal ERP old/new effect is generated in this region (Rugg et al., 2002; Wagner et al., 2005). The findings of Yonelinas et al. (2005), who reported that activity in left inferior lateral parietal cortex was enhanced for recollected items but was insensitive to familiarity strength, offer further support for this proposal. The functional role of the lateral parietal cortex in recollection is uncertain (Wagner et al., 2005), although the two most discussed possibilities that the region supports attentional orienting to recollected information or alternatively that it supports the representation of that information both predict that recollection-related lateral parietal activity should onset as soon as recollected information is available. The onset latency of the ERP old/new effect, at around 400 ms, appears broadly consistent with this prediction (Hintzman and Curran, 1994). The final ERP effect with a parallel in the findings of the preceding fmri experiment is the sustained frontal positivity elicited specifically by recollected items (Fig. 6). Yonelinas et al. (2005) identified an extensive region of left medial anterior prefrontal cortex where activity was enhanced exclusively for recollected items. Since neural activity localized to the medial surface of a hemisphere generates a scalp field with a contralateral maximum (Barrett et al., 1976; Brunia and Vingerhoets, 1981), it seems plausible to suppose that these ERP and fmri effects are related. This parallel, if valid, has implications for understanding the functional role of the anterior medial prefrontal cortex in recollection. Yonelinas et al. (2005) discussed two possible roles for the recollectionrelated neural activity that they observed in this region. They conjectured that the activity might act as a signal that recollection had occurred or that it might support postrecollective monitoring triggered by such a signal. The late onset and sustained time-course of the ERP effect suggest that the latter is more likely Concluding comments The present findings lend strong support to dual-process models of recognition memory and add weight to prior proposals that frontal and left parietal ERP old/new effects are respectively correlates of familiarity and recollection. Furthermore, the findings converge with those from a parallel fmri study (Yonelinas et al., 2005) to shed light on the timecourse and possible functional significance of the cortical regions selectively engaged during familiarity- and recollection-based recognition. 4. Experimental procedures 4.1. Subjects Twenty four subjects (9 males), ranging in age from 18 to 27 (mean 21), participated in return for payment at the rate of $15/h. All were right-handed native English speakers recruited from the UCI undergraduate community and had normal or corrected-to-normal vision. Informed consent was obtained in accordance with UCI Institutional Review Board guidelines. Of these 24 subjects, 8 were rejected from analysis. Seven subjects were rejected due to insufficient trials (<15) in one or more of the critical experimental conditions. The remaining subject was rejected due to excessive eye movement artifact Stimuli The critical experimental stimuli were drawn from a pool of 300 words (3 13 letters long with a mean length 5.6 letters, and a mean written frequency of 21.2 counts per million; Kucera and Francis, 1967). Four study lists of 150 words were constructed by randomly selecting words from the pool. Each subject receiving one of these four lists (re-randomized

10 134 B R A I N R E S E A R C H ( ) for each subject). The remaining 150 items that were not selected for the study list were designated as new items and were pseudo-randomly intermixed with study list items to form a test list (item ordering was constrained so that no more than three items of the same type occurred consecutively). Study and test lists were both buffered at the beginning with two items drawn from a pool of additional items with the same parameters as the critical pool. The study list had an additional two buffers at its end. A separate pool of 100 items (same parameters as above) was used to construct a study list (50 items) and a test list (100 items) for the practice phase. Subjects were seated 1 m from the computer monitor on which stimuli were displayed. The stimuli were presented within a constantly present uniform gray frame, subtending a visual angle of , centered on a black background. Words (maximum height 0.7, maximum length 4.1 ) and fixation crosses ( ) were superimposed on this frame and were presented in black 30-point Helvetica font with the fixation cross turning red 500 ms prior to the onset of each test word. Timing of stimulus events was identical for the practice and the experimental phases. While the practice phase only involved one study block (50 items) followed by one test block (100 items), the experimental phase involved one study block (154 items) followed by two test blocks (152 items each). The break between study and test was approximately 5 min as was the break between test blocks one and two. A trial began with a red fixation cross appearing for 500 ms followed by the word displayed for another 500 ms. The word was replaced by a black fixation cross for 3200 ms which then turned red to signify the beginning of the next trial Procedure Following informed consent, subjects were seated in a soundattenuated room in front of a computer monitor upon which experimental stimuli were displayed. They performed a practice version of the experiment in which they were given instructions for the study phase followed by practice followed by instructions for the test phase followed by practice. The EEG cap and electrodes were then applied, following which study and test instructions were repeated. For the study phase, instructions were to press one of two buttons with the index or middle finger depending on whether the presented word represented an animate or inanimate item, respectively. For the test phase, instructions were to indicate the old/new status of each item by pressing one of five buttons. If the word was judged old (i.e. presented during the study phase) and, in addition, something specific about the study episode could be recollected, a thumb-press was required using the designated hand (a Remember or R response). In the absence of recollection, one of four alternative responses was required using the fingers of the other hand. Subjects were instructed to make a thumb-press if they were confident that the item was old despite the absence of recollective detail, to depress their index finger if they thought the item was old but were unconfident, to depress their middle finger if they were unconfident that the item was new (unstudied), and to use their ring finger when confident that the item was new. Response hands were counterbalanced across subjects so that half used the thumb on their left hand to indicate an R response, while the other half used the thumb on their right hand. During the first quarter of trials in the practice test phase, subjects were requested to justify their responses on a trialby-trial basis. In particular, following an R response, subjects were requested to report the contextual information that was retrieved along with the studied item. The experimenter took care to ensure the subject fully understood the basis of each response type ERP recording and analysis EEG was recorded continuously from 60 active silver/silver chloride electrodes embedded in an elastic cap and from two electrodes placed one each to the left and right mastoids. Electrode arrangement was based on the International 10/10 system (American Electroencephalographic Society; see montage 11 at excluding electrodes Oz, TP9, TP10, and FT10). Vertical and horizontal eye movement was monitored by recording EOG from electrode pairs placed above and below the left eye and one on each outer canthus. Data acquisition was carried out using the BioSemi (Amsterdam, Netherlands) Active Two system at a 256-Hz sampling rate and an amplifier bandwidth of 0 67 Hz ( 3 db). EEG data were digitally highpass filtered ( 3 db at 0.1 Hz, zero-phase) and epoched (2048 ms duration; 102 ms pre-stimulus baseline) offline. The epochs were downsampled to a 125-Hz sampling rate and algebraically referenced to linked mastoids. Movement artifact, horizontal EOG artifact, vertical EOG artifact (excluding blinks), and excessive baseline drift served as a basis for rejecting trials. The averaged ERPs were digitally smoothed ( 3 db at 19.4 Hz, zero-phase). Blink artifacts were removed by subtracting the weighted vertical EOG trace for each condition from the corresponding ERP waveforms, estimating the weights for each electrode site by regressing the average blink artifact in the EOG channel against the averaged artifact in each channel of EEG. Acknowledgments HRH was supported by NIMH National Research Service Award MH A1. This research was supported by NIMH grant 5R01MH CCW is now at the Department of Psychology, Pomona College, Claremont, CA R E F E R E N C E S Barrett, G., Blumhardt, L., Halliday, A.M., Halliday, E., Kriss, A., Paradox in lateralization of visual evoked-response. Nature 261, Brunia, C.H., Vingerhoets, A.J., Opposite hemisphere differences in movement related potentials preceding foot and finger flexions. Biol. Psychol. 13, Curran, T., Brain potentials of recollection and familiarity. Mem. Cogn. 28, Curran, T., Effects of attention and confidence on the hypothesized ERP correlates of recollection and familiarity. Neuropsychologia 42,

11 135 Curran, T., Cleary, A.M., Using ERPs to dissociate recollection from familiarity in picture recognition. Brain Res. Cogn. Brain Res. 15, Donaldson, W., The role of decision processes in remembering and knowing. Mem. Cogn. 24, Duarte, A., Ranganath, C., Winward, L., Hayward, D., Knight, R.T., Dissociable neural correlates for familiarity and recollection during the encoding and retrieval of pictures. Brain Res. Cogn. Brain Res. 18, Duarte, A., Ranganath, C., Knight, R.T., Effects of unilateral prefrontal lesions on familiarity, recollection, and source memory. J. Neurosci. 25, Dunn, J.C., Remember know: a matter of confidence. Psychol. Rev. 111, Eldridge, L.L., Knowlton, B.J., Furmanski, C.S., Bookheimer, S.Y., Engel, S.A., Remembering episodes: a selective role for the hippocampus during retrieval. Nat. Neurosci. 3, Friedman, D., Johnson, J.R., Event-related potential (ERP) studies of memory encoding and retrieval: a selective review. Microsc. Res. Tech. 51, Henson, R.N., Rugg, M.D., Shallice, T., Dolan, R.J., Confidence in recognition memory for words: dissociating right prefrontal roles in episodic retrieval. J. Cogn. Neurosci. 12, Henson, R.N., Goshen-Gottstein, Y., Ganel, T., Otten, L.J., Quayle, A., Rugg, M.D., Electrophysiological and haemodynamic correlates of face perception, recognition and priming. Cereb. Cortex 13, Herron, J.E., Quayle, A.H., Rugg, M.D., Probability effects on event-related potential correlates of recognition memory. Brain Res. Cogn. Brain Res. 16, Hintzman, D.L., Curran, T., Retrieval dynamics of recognition and frequency judgments Evidence for separate processes of familiarity and recall. J. Mem. Lang. 33, Kucera, H., Francis, W.N., Computational Analysis of Present-day American English. Brown Univ. Press, Providence, RI. McCarthy, G., Wood, C.C., Scalp distributions of eventrelated potentials: an ambiguity associated with analysis of variance models. Electroencephalogr. Clin. Neurophysiol. 62, Polich, J., Kok, A., Cognitive and biological determinants of P300: an integrative review. Biol. Psychol. 41, Ratcliff, R., Vanzandt, T., McKoon, G., Process dissociation, single-process theories, and recognition memory. J. Exp. Psychol. Gen. 124, Rugg, M.D., Coles, M.G.H., The ERP and cognitive psychology: conceptual issues. In: Rugg, M.D., Coles, M.G.H. (Eds.), Electrophysiology of Mind: Event-Related Brain Potentials and Cognition. Oxford UP, New York, pp Rugg, M.D., Henson, R.N., Episodic memory retrieval: an (event-related) functional neuroimaging perspective. In: Parker, A.E., Wilding, E.L., Bussey, T. (Eds.), The Cognitive Neuroscience of Memory Encoding and Retrieval. Psychology Press, Hove, UK, pp Rugg, M.D., Yonelinas, A.P., Human recognition memory: a cognitive neuroscience perspective. Trends Cogn. Sci. 7, Rugg, M.D., Mark, R.E., Walla, P., Schloerscheidt, A.M., Birch, C.S., Allan, K., Dissociation of the neural correlates of implicit and explicit memory. Nature 392, Rugg, M.D., Otten, L.J., Henson, R.N, The neural basis of episodic memory: evidence from functional neuroimaging. Philos. Trans. R. Soc. Lond., Ser. B Biol. Sci. 357, Schacter, D.L., Buckner, R.L., Koutstaal, W., Dale, A.M., Rosen, B.R., Late onset of anterior prefrontal activity during true and false recognition: an event-related fmri study. NeuroImage 6, Slotnick, S.D., Dodson, C.S., Support for a continuous (single-process) model of recognition memory and source memory. Mem. Cogn. 33, Spencer, K.M., Vila Abad, E., Donchin, E., On the search for the neurophysiological manifestation of recollective experience. Psychophysiology 37, Tsivilis, D., Otten, L.J., Rugg, M.D., Context effects on the neural correlates of recognition memory: an electrophysiological study. Neuron 31, Tulving, E., Memory and consciousness. Can. Psychol. 26, Urbach, T.P., Kutas, M., The intractability of scaling scalp distributions to infer neuroelectric sources. Psychophysiology 39, Voss, J.L., Paller, K.A., Fluent conceptual processing and explicit memory for faces are electrophysiologically distinct. J. Neurosci. 26, Wagner, A.D., Shannon, B.J., Kahn, I., Buckner, R.L., Parietal lobe contributions to episodic memory retrieval. Trends Cogn. Sci. 9, Wilding, E.L., Rugg, M.D., An event-related potential study of recognition memory with and without retrieval of source. Brain 119, Yonelinas, A.P., Receiver-operating characteristics in recognition memory: evidence for a dual-process model. J. Exper. Psychol., Learn., Mem., Cogn. 20, Yonelinas, A.P., The nature of recollection and familiarity: a review of 30 years of research. J. Mem. Lang. 33, Yonelinas, A.P., Otten, L.J., Shaw, K.N., Rugg, M.D., Separating the brain regions involved in recollection and familiarity in recognition memory. J. Neurosci. 25, Yovel, G., Paller, K.A., The neural basis of the butcher-on-the-bus phenomenon: when a face seems familiar but is not remembered. NeuroImage 21,