The nature of recollection in behavior and the brain Scott D. Slotnick

Transcription

1 Review 663 The nature of recollection in behavior and the brain Scott D. Slotnick Long-term memory is widely believed to depend on either detailed recollection or nondetailed familiarity. Although familiarity is generally thought to be a continuous/graded process, the nature of recollection is currently under debate. The present review considers evidence bearing on three separate debates on recollection. The first debate has focused on whether each medial temporal lobe subregion is differentially or similarly engaged during recollection. A meta-analysis of 5 functional MRI studies indicated that the hippocampus and the parahippocampal cortex are associated with recollection, whereas the perirhinal cortex is associated with familiarity. The second debate has focused on whether recollection is a continuous process or a threshold/all-or-none process. A meta-analysis of seven studies with 3 conditions revealed that the large majority of source/context memory receiver operating characteristics are curved, which contradicts the linear receiver operating characteristic predicted by the threshold model of recollection and supports the continuous model of recollection. The third debate has focused on whether recollection and familiarity are separate processes or recollection and familiarity represent a single process. A meta-analysis of 37 studies with 23 conditions showed that remember and know response rates closely tracked the relationship predicted by a single-process continuous model, in which recollection and familiarity reflect strong memory and weak memory, respectively. Considered together, the body of evidence indicates that recollection represents a continuous process that is not distinct from familiarity and that this single cognitive process relies on multiple brain processes. However, much less is known about the nature of recollection in the brain. NeuroReport 24: c 23 Wolters Kluwer Health Lippincott Williams & Wilkins. NeuroReport 23, 24: Keywords: continuous model, dual process, familiarity, fmri, hippocampus, recollection, review, receiver operating characteristic, single process, threshold model Department of Psychology, Boston College, Chestnut Hill, Massachusetts, USA Correspondence to Scott D. Slotnick, PhD, Department of Psychology, Boston College, Chestnut Hill, MA 2467, USA Tel: ; fax: ; sd.slotnick@bc.edu Received May 23 accepted 2 May 23 Introduction Long-term memory is widely believed to depend on either detailed recollection or nondetailed familiarity. Although familiarity is generally thought to be a continuous/graded process, the nature of recollection is currently under debate. The present review considers evidence bearing on three separate debates on recollection in an effort to consolidate our current understanding of this process and guide future research. The first debate has focused on whether the hippocampus and other cortical regions are differentially or similarly engaged during recollection. The second debate has focused on whether recollection is a continuous process or a threshold/all-or-none process. The third debate has focused on whether recollection is a separate process from familiarity or recollection and familiarity represent a single process. The nature of recollection in the hippocampus and other cortical regions The medial temporal lobe is known to be important for recollection and familiarity. However, there has been debate on whether each medial temporal lobe subregion is differentially or similarly engaged during recollection and familiarity. The subregion processing hypothesis [,2] highlights separate subregion functionality the perirhinal cortex mediates item processing, the parahippocampal cortex mediates context processing, and the hippocampus binds item and context information (Fig. a, left). This hypothesis predicts that each subregion will be differentially active during recollection and familiarity. The system processing hypothesis [3] specifies that the subregions operate together (as a system) during recollection and familiarity (Fig. a, right) and predicts that the magnitude of activity in each subregion is the same or similar during recollection and familiarity. Of importance, the magnitude of activity can vary between subregions. This debate has focused on the hippocampus, given its central role in explicit/conscious memory. A meta-analysis of 5 functional MRI (fmri) studies was conducted recently to distinguish between these hypotheses [4]. Recollection activity was isolated by contrasting accurate item memory and source/context memory versus accurate item memory and inaccurate source memory or by contrasting remember versus know /high-confidence familiarity responses (source memory, e.g. memory for an item s previous spatial location or color, is assumed to be based on recollection as it requires retrieval of detailed information). Familiarity activity was isolated by contrasting accurate item memory (and inaccurate source memory) versus forgotten items or through correlation c 23 Wolters Kluwer Health Lippincott Williams & Wilkins DOI:.97/WNR.b3e328362e47e

2 664 NeuroReport 23, Vol 24 No 2 Fig. (a) Subregion processing PRC HC System processing PRC HC (b) Subregion processing System processing Item PRC HC PHC Cortical input PHC Binding Cortical input PHC Recollection index (%) PRC HC Context PHC (a) Hypotheses of medial temporal lobe subregion function. (b) Subregion and system processing predictions (top) with empirical recollection index values shown by arrows (middle) and illustrated pictorially (bottom). HC, hippocampus; PHC, parahippocampal cortex; PRC, perirhinal cortex. Reproduced with permission of Palgrave Macmillan from Slotnick [4]. with familiarity responses. Recollection activity was observed in the hippocampus in the large majority (87%) of studies, the parahippocampal cortex in the majority (57%) of studies, and was never observed in the perirhinal cortex, whereas familiarity activity was observed in the perirhinal cortex in close to half (42%) of the studies and was almost never observed in the hippocampus (8%) or the parahippocampal cortex (%; Table ). Although the same pattern of results has been reported in previous reviews [5,6], the present review includes more recent findings. For the present review, a recollection index was computed to determine the degree to which each subregion was associated with recollection and familiarity. This is the first time that such a metric has been used. The proportion of studies that reported recollection activity was divided by the sum of that value plus the proportion of studies that reported familiarity activity [e.g. 87/(87 + 8) for the hippocampus]. A recollection index of % indicates that a region is only associated with recollection, a recollection index of % indicates that a region is only associated with familiarity, and a recollection index of 5% indicates that a region is similarly associated with recollection and familiarity. The subregion processing hypothesis predicts recollection index values close to or %, whereas the system processing hypothesis predicts recollection index values close to 5% (Fig. b, top). The hippocampus had a recollection index of 92%, the parahippocampal cortex had a recollection index of %, and the perirhinal cortex had a recollection index of % (Fig. b, middle and bottom; if the system processing hypothesis was correct, all subregion ovals would have been gray). The preceding results provide strong evidence of differential subregion activity, which contradicts the system processing hypothesis and supports the subregion processing hypothesis. It has been argued that such differential activity is due to a memory strength confound in recollection versus familiarity comparisons [3,6,22]. However, in the source memory studies evaluated, an identical pattern of results was observed when source memory accuracy was greater than item memory accuracy or vice versa (Table ; each accuracy corresponded to the fmri contrast used and was scaled such that chance performance was 5%, thus reflecting an independent measure of memory strength [2]). That is, the pattern of medial temporal lobe activity did not depend on memory strength, which rules out a memory strength explanation. This is the first time that medial temporal lobe activity has been evaluated as a function of memory accuracy across studies. Future empirical studies should similarly compare item memory accuracy and source memory accuracy to further address a potential memory strength confound. Moreover, as detailed elsewhere [3], proponents of the system processing hypothesis have yet to provide any convincing evidence that subregions are similarly involved in recollection and familiarity. For instance, in a recent fmri study [22], a contrast analysis showed that both remember and know responses produced activity in the hippocampus under conditions of similar memory accuracy/strength (see Fig. 6 in Smith et al. [22]). However, analysis of the corresponding hippocampal activation timecourse magnitudes, which was carried out for this review, revealed greater than baseline activity at 6 s after stimulus onset (the time of the maximum hemodynamic response) for remember (P <.5) but not know (P >.2) responses (i.e. the

3 The nature of recollection Slotnick 665 Table Medial temporal lobe subregions associated with recollection and familiarity with relative source memory and item memory accuracy (if available) Recollection Familiarity References HC PRC PHC HC PRC PHC Accuracy Eldridge et al. [7] X X Cansino et al. [8] X X Davachi et al. [9] X X X Source memory > item memory Ranganath et al. [] X X Item memory > source memory Weis et al. [] X Source memory > item memory Woodruff et al. [2] X X Yonelinas et al. [3] X X Gold et al. [4] Kensinger and Schacter [5] X X Item memory > source memory Montaldi et al. [6] X X Kirwan et al. [7] X X Ross and Slotnick [8] X X X Source memory > item memory Staresina and Davachi [9] X Tendolkar et al. [2] X X Item memory > source memory Slotnick [2] X Item memory > source memory HC, hippocampus; PHC, parahippocampal cortex; PRC, perirhinal cortex; X, fmri activity observed in the corresponding subregion. timecourse activation profiles were not similar as predicted by the system processing hypothesis). Thus, the current body of evidence indicates that the hippocampus and parahippocampal cortex are associated with recollection, whereas the perirhinal cortex is associated with familiarity. Is recollection a threshold process or a continuous process? The long-standing (majority) view is that recollection is a threshold process [2,23 25], but recent evidence indicates that recollection is a continuous process [26,27]. The most convincing evidence used to distinguish between these models of recollection is based on the shape of the source memory receiver operating characteristic (ROC). In a typical source memory ROC paradigm, words are presented in a male or a female voice during the study phase. During the test phase, words are presented visually and participants make a seven-point source confidence rating from very sure female to very sure male. The continuous unequal variance (UEV) model of recollection dictates that each confidence rating response depends on the source memory strength of an item previously presented in the male or female voice, which has a Gaussian distribution, and criteria placement (C C 6 ) in decision space (Fig. 2a, left; memory strength is the distance between distribution means, and distribution variances can be unequal). For example, an item previously spoken by the male source with memory strength greater than C 6 would produce a very sure male response. To generate the ROC, hit rate is plotted against false alarm rate for each confidence rating/ criterion (Fig. 2a, right; leftmost ROC point x = the probability of a very sure male response given a female source, y = the probability of a very sure male response given a male source). The continuous model ROC is always curved (Fig. 2a, right). The two-high threshold (2HT) model of recollection dictates that there are two thresholds in decision space above and below which only one source exists (Fig. 2b, left). Recollection occurs if memory strength is beyond one of the two thresholds but otherwise fails, which is why this is referred to as the allor-none model [28 3]. The threshold model ROC is always linear (Fig. 2b, right). To date, seven source memory studies with 3 conditions have evaluated the shape of the recollection ROC [28,29,3 35]. As the familiarity ROC is curved [28,3,32], sources under all the conditions evaluated had similar levels of familiarity to eliminate a familiarity recollection confound (i.e. all the items in the analysis had been previously studied/were equally familiar such that source memory could not be based on familiarity). This ensured that the ROC only reflected source memory/recollection such that the threshold model predicted that the ROC would be linear [2,24,25,3]. For each source memory ROC, it was determined whether the addition of a quadratic (cx 2 ) component significantly improved the fit over a linear (a + bx) function. The large majority (9%) of the ROCs had significant negative curvature (i.e. the quadratic component/c-value was negative; Fig. 2c; Fig. 3). This finding contradicts the threshold model prediction of a linear ROC and supports the continuous model prediction of a negatively curved ROC. The curvature predicted by the continuous model, however, is not quadratic. As such, in five studies with 5 conditions the continuous and threshold models have also been evaluated by testing the goodness-of-fit (w 2 -value) between the predicted and the empirical ROC [26,28,3,32,35] (data from the study by Yonelinas [3] was reanalyzed by Slotnick and Dodson [28]). These source memory ROCs were usually generated to maximize recollection by restricting analysis to high-confidence old responses [28,3,32] or remember

4 666 NeuroReport 23, Vol 24 No 2 Fig. 2 (a).2 UEV model decision space UEV model source ROC Female source Male source.8 Probability C C C C C C Source memory strength (b).2 2HT model decision space 2HT model source ROC.8 Threshold Threshold 2 Probability. Female source Male source C C C C C C Source memory strength (c) Experiment source ROC Experiment 2 source ROC (a) Continuous unequal variance (UEV) model source memory decision space (left) and corresponding curved ROC (right). (b) Two-high threshold (2HT) model decision space (left) and corresponding linear ROC (right). (c) Empirical source memory ROCs (circles) with best-fit continuous (curved) and threshold (linear) ROCs. ROC, receiver operating characteristic. Reproduced with permission of Springer Science + Business Media from Slotnick and Dodson (Figs 2 and 7) [28]. responses [26] this minimized inclusion of forgotten/ nonmemorial items that can distort/flatten the ROC [28]. Although increasing source memory strength has been posited to increase familiarity [25], this is nonsensical given that source memory is assumed to rely on recollection alone [2,24,25,3]. The large majority (8%) of ROCs were adequately fit by the continuous model (either the UEV model or a modified UEV model that accounts for source misattribution [26,35]). The threshold model did not adequately fit any of the ROCs (all Ps <.).

5 The nature of recollection Slotnick 667 Fig. 3 Continuous (UEV) model Source memory ROC curvature Threshold (2HT) model Quadratic (cx 2 ) coefficient Source memory ROC curvature with continuous and threshold model predictions. 2HT, two-high threshold; ROC, receiver operating characteristic; UEV, unequal variance. The preceding ROC results are consistent with non-roc evidence such as high-confidence suprathreshold false alarms that also contradict the threshold model [3]. Recent studies that used novel tasks that did not rely on confidence ratings [36,37] or took a computational modeling approach [38] provided evidence that appeared to be more consistent with some form of threshold recollection model. However, all threshold recollection models dictate a suprathreshold hit rate that is positive in value at a false alarm rate of (i.e. a positive ROC y-axis intercept [38]), which has not been observed empirically. Future work will be needed to assess whether there are conditions under which recollection can operate in a threshold manner. That caveat aside, the extant evidence rules out the threshold model as a general description of recollection and supports the continuous model of recollection. Of relevance, there is a wide body of evidence that familiarity is well modeled by a continuous process [28,3,32]. As both recollection and familiarity appear to be continuous processes, it is next evaluated whether these processes are separate or should be characterized as a single process. Is recollection a separate process from familiarity? The belief that memory depends on recollection and familiarity is ubiquitous. This dual-process view is based, in part, on intuition/subjective experience. As illustrated by the butcher-on-the-bus phenomenon [23,27], you could have strong familiarity for someone you see on a bus (the butcher from the supermarket) without recollecting where you have seen him/her, whereas at other times you might recollect such contextual details. However, subjective memorial experience is known to be fallible, as illustrated by false memories and memory distortion [39], and thus provides weak support for the dual-process view. By comparison, process dissociations seem to provide strong empirical support for the dual-process model [23,24]. For instance, perceptual changes between study and test phases disrupt familiarity to a greater degree than recollection, and dividing attention during the test phase disrupts recollection to a greater degree than familiarity [24]. However, the evidence detailed below indicates that process dissociations are entirely compatible with the single-process model. The single-process UEV model has been evaluated in item memory studies where items are presented during the study phase and then, during the test phase, old and new items are presented and participants make remember, know, or new judgments. The continuous UEV model (Fig. 4a) dictates that each response is based on memory strength in relation to the decision criteria (C and C 2 ). According to this model, remember and know responses reflect a single process corresponding to strong memory and weak memory, respectively. A meta-analysis of 7 studies with 8 conditions showed that shifts in response criteria/bias between experiments can differentially affect measures of recollection and familiarity [4]. This provides a single-process explanation for recollection familiarity dissociations. A second metaanalysis of 72 studies with 4 conditions found that remember and know response rates reflecting all possible dissociations were well described by a singleprocess model [42]. That is, the observed remember and know response rates were very closely predicted by adjusting the UEV model memory strength for old and new items, distribution variances, and criteria placement. A third meta-analysis of 37 studies with 23 conditions assessed whether a single-process UEV model could account for empirical findings or a dual-process model was necessary [4]. The old item remember + know (adjusted RK ) response rate was plotted against the remember response rate (adjusted R ; these response rates refer to the respective probabilities of an old response and a remember response adjusted for differences in decision criteria across experiments). The empirical results (Fig. 4b, circles) closely tracked the relationship predicted by the single-process model (Fig. 4b, curve). This is problematic for the dual-process model given the infinite set of predicted response rate relationships (Fig. 4b, the upper left half of the square). If the dual-process model was correct, why would the results so closely match the curve predicted by the single-process model? The preceding meta-analyses show that process dissociations, which have been taken to support the dual-process model, actually support the single-process model. For reasons of parsimony, a single-process model should be favored. It is notable that one of the two primary dualprocess models is mathematically equivalent to the single-process UEV model [27]. Although a meta-analysis suggested that memory strength amplitudes associated with remember know guess paradigms did not conform to the predictions of the single-process model [43], it has been argued that the small discrepancies

6 668 NeuroReport 23, Vol 24 No 2 Fig. 4 (a).2 UEV model decision space N K R (b) Probability. New items Old items Adj. RK C C 2 Item memory strength Adj. R (a) Single-process UEV model item memory decision space. (b) Old item remember + know response rate (adjusted RK ) plotted against remember response rate (adjusted R ) with the single-process model prediction (curve) and the dual-process model prediction (upper left half of the square). ROC, receiver operating characteristic; UEV, unequal variance. Adapted with permission from Dunn [4]. observed are consistent with a single-process model in which old and new item distributions have unequal variances [42,44]. A separate meta-analysis suggested that the ratios of old and new item distribution variances associated with old new versus remember know paradigms were not equivalent as predicted by the single-process model [44]; however, these ratios were largely overlapping (i.e. appeared to be equivalent), and a statistical comparison of the entire dataset was not reported. Future research along these lines will be needed to further assess whether the single-process model can adequately describe all of the experimental results. The findings reviewed in this section, along with the previous behavioral source memory ROC findings reviewed above, indicate that recollection and familiarity should be construed as strong memory and weak memory within a continuous single-process framework. Given that the terms recollection and familiarity are intimately linked to the dual-process model, it would seem prudent to replace these terms with strong memory and weak memory. The single-process behavioral findings need to be reconciled with the medial temporal lobe dissociation detailed above, which could be interpreted as supporting the dual-process model. However, a neural dissociation does not imply a behavioral/cognitive dissociation and is entirely consistent with the single-process view (for a detailed theoretical analysis, see Kalish and Dunn [45]). To illustrate, the present medial temporal lobe results indicate that the hippocampus and the perirhinal cortex are associated with strong memory and weak memory, respectively. From a single-process perspective, the hippocampus can be described as being preferentially sensitive to strong memory and the perirhinal cortex can be described as being preferentially sensitive to weak memory. This description is reminiscent of the system processing hypothesis of the medial temporal lobe, according to which each subregion preferentially but not exclusively processes a particular attribute of memory such as stimulus associations or stimulus frequency [3]. However, the present view differs from the system processing hypothesis in that the magnitude of activity in each medial temporal lobe subregion need not be of similar magnitude. Here, the activity in each medial temporal lobe subregion reflects the degree to which it processes a particular memory attribute (rather than recollection and familiarity), which is a topic of future empirical research [3]. Related to this issue, the evidence that a neural region is simply associated with recollection/strong memory does not speak to the nature of processing in that region. A recent ROC study investigated whether the hippocampus operates in a continuous manner or a threshold manner during recollection [46]. Hippocampal source memory fmri activity (Fig. 5a) was evaluated as a function of source confidence ratings. Specifically, the activation magnitudes associated with items previously presented in the left visual field or the right visual field were separated as a function of source/spatial location confidence. This event-related activation profile was used to generate the ROC (e.g. the magnitude of activity associated with a sure left response to an item previously presented in the left visual field and the right visual field, respectively, was used to compute a hit rate and false alarm rate the leftmost ROC point). The threshold model (line) provided an adequate fit to the ROC (circles), but the continuous model (curve) did not adequately fit the ROC (Fig. 5b). These results suggest

7 The nature of recollection Slotnick 669 Fig. 5 (a) Source memory activity (b) Hippocampal source memory ROC (a) Source memory functional MRI activity in the hippocampus (circled). (b) Hippocampal source memory ROC (circles) with best-fit threshold (line) and continuous (curve) models. ROC, receiver operating characteristic. Reprinted with permission of Wolters Kluwer Health from Slotnick and Thakral [46]. that the hippocampus operates in a threshold manner during recollection. How can a threshold process in the hippocampus give rise to continuous recollection in behavior? Many brain regions mediate recollection including the hippocampus, the parahippocampal cortex, the prefrontal cortex, and the parietal cortex [47]. The hippocampus is involved in binding item and source information, but other regions, such as the parietal cortex, that reflect subjective memorial experience [2] would be expected to mediate continuous behavioral recollection. Additional studies along the lines of the hippocampal ROC study above will be required to evaluate the nature of processing in other cortical regions during recollection. Conclusion A wide body of behavioral evidence indicates that recollection and familiarity should be characterized as a single continuous process. This is at odds with the current majority view that recollection is a threshold process that is distinct from familiarity. Recent results suggest that the hippocampus operates in a threshold manner during recollection/strong memory. This highlights the fact that the nature of recollection need not be unitary in behavior and in the brain. Critically, although the nature of recollection in behavior is relatively well delineated, much less is known about the nature of recollection in the brain. Future work will be needed to address this gap in understanding. Acknowledgements Conflicts of interest There are no conflicts of interest. References Ranganath C. A unified framework for the functional organization of the medial temporal lobes and the phenomenology of episodic memory. Hippocampus 2; 2: Yonelinas AP, Aly M, Wang WC, Koen JD. Recollection and familiarity: examining controversial assumptions and new directions. Hippocampus 2; 2: Squire LR, Wixted JT, Clark RE. Recognition memory and the medial temporal lobe: a new perspective. Nat Rev Neurosci 27; 8: Slotnick SD. Controversies in cognitive neuroscience. Basingstoke, UK: Palgrave Macmillan; Eichenbaum H, Yonelinas AP, Ranganath C. The medial temporal lobe and recognition memory. Annu Rev Neurosci 27; 3: Wais PE. fmri signals associated with memory strength in the medial temporal lobes: a meta-analysis. Neuropsychologia 28; 46: Eldridge LL, Knowlton BJ, Furmanski CS, Bookheimer SY, Engel SA. Remembering episodes: a selective role for the hippocampus during retrieval. Nat Neurosci 2; 3: Cansino S, Maquet P, Dolan RJ, Rugg MD. Brain activity underlying encoding and retrieval of source memory. Cereb Cortex 22; 2: Davachi L, Mitchell JP, Wagner AD. Multiple routes to memory: distinct medial temporal lobe processes build item and source memories. Proc Natl Acad Sci USA 23; : Ranganath C, Yonelinas AP, Cohen MX, Dy CJ, Tom SM, D Esposito M. Dissociable correlates of recollection and familiarity within the medial temporal lobes. Neuropsychologia 24; 42:2 3. Weis S, Specht K, Klaver P, Tendolkar I, Willmes K, Ruhlmann J, et al. Process dissociation between contextual retrieval and item recognition. Neuroreport 24; 5: Woodruff CC, Johnson JD, Uncapher MR, Rugg MD. Content-specificity of the neural correlates of recollection. Neuropsychologia 25; 43: Yonelinas AP, Otten LJ, Shaw KN, Rugg MD. Separating the brain regions involved in recollection and familiarity in recognition memory. J Neurosci 25; 25: Gold JJ, Smith CN, Bayley PJ, Shrager Y, Brewer JB, Stark CE, et al. Item memory, source memory, and the medial temporal lobe: concordant findings from fmri and memory-impaired patients. Proc Natl Acad Sci USA 26; 3: Kensinger EA, Schacter DL. Amygdala activity is associated with the successful encoding of item, but not source, information for positive and negative stimuli. J Neurosci 26; 26: Montaldi D, Spencer TJ, Roberts N, Mayes AR. The neural system that mediates familiarity memory. Hippocampus 26; 6:54 52.

8 67 NeuroReport 23, Vol 24 No 2 7 Kirwan CB, Wixted JT, Squire LR. Activity in the medial temporal lobe predicts memory strength, whereas activity in the prefrontal cortex predicts recollection. J Neurosci 28; 28: Ross RS, Slotnick SD. The hippocampus is preferentially associated with memory for spatial context. J Cogn Neurosci 28; 2: Staresina BP, Davachi L. Selective and shared contributions of the hippocampus and perirhinal cortex to episodic item and associative encoding. J Cogn Neurosci 28; 2: Tendolkar I, Arnold J, Petersson KM, Weis S, Brockhaus-Dumke A, van Eijndhoven P, et al. Contributions of the medial temporal lobe to declarative memory retrieval: manipulating the amount of contextual retrieval. Learn Mem 28; 5: Slotnick SD. Does the hippocampus mediate objective binding or subjective remembering? Neuroimage 2; 49: Smith CN, Wixted JT, Squire LR. The hippocampus supports both recollection and familiarity when memories are strong. J Neurosci 2; 3: Mandler G. Recognizing: the judgment of previous occurrence. Psychol Rev 98; 87: Yonelinas AP. The nature of recollection and familiarity: a review of 3 years of research. J Mem Lang 22; 46: Yonelinas AP, Parks CM. Receiver operating characteristics (ROCs) in recognition memory: a review. Psychol Bull 27; 33: Slotnick SD. Remember source memory ROCs indicate recollection is a continuous process. Memory 2; 8: Wixted JT, Mickes L. A continuous dual-process model of remember/know judgments. Psychol Rev 2; 7: Slotnick SD, Dodson CS. Support for a continuous (single-process) model of recognition memory and source memory. Mem Cognit 25; 33: Hilford A, Glanzer M, Kim K, DeCarlo LT. Regularities of source recognition: ROC analysis. J Exp Psychol Gen 22; 3: Wixted JT, Stretch V. In defense of the signal detection interpretation of remember/know judgments. Psychon Bull Rev 24; : Yonelinas AP. The contribution of recollection and familiarity to recognition and source-memory judgments: a formal dual-process model and an analysis of receiver operating characteristics. J Exp Psychol Learn Mem Cogn 999; 25: Slotnick SD, Klein SA, Dodson CS, Shimamura AP. An analysis of signal detection and threshold models of source memory. J Exp Psychol Learn Mem Cogn 2; 26: Qin J, Raye CL, Johnson MK, Mitchell KJ. Source ROCs are (typically) curvilinear: comment on Yonelinas (999). J Exp Psychol Learn Mem Cogn 2; 27: Glanzer M, Hilford A, Kim K. 6 regularities of source recognition. JExp Psychol Learn Mem Cogn 24; 3: Dodson CS, Bawa S, SD Slotnick. Aging, source memory, and misrecollections. J Exp Psychol Learn Mem Cogn 27; 33: Parks CM, Yonelinas AP. Evidence for a memory threshold in second-choice recognition memory responses. Proc Natl Acad Sci USA 29; 6: Harlow IM, Donaldson DI. Source accuracy data reveal the thresholded nature of human episodic memory. Psychon Bull Rev 23; 2: Norman KA. How hippocampus and cortex contribute to recognition memory: revisiting the complementary learning systems model. Hippocampus 2; 2: Schacter DL. The seven sins of memory: how the mind forgets and remembers. Boston: Houghton Mifflin; 2. 4 Dunn JC. The dimensionality of the remember-know task: a state-trace analysis. Psychol Rev 28; 5: Donaldson W. The role of decision processes in remembering and knowing. Mem Cognit 996; 24: Dunn JC. Remember-know: a matter of confidence. Psychol Rev 24; : Gardiner JM, Ramponi C, Richardson-Klavehn A. Recognition memory and decision processes: a meta-analysis of remember, know, and guess responses. Memory 22; : Rotello CM, Macmillan NA, Reeder JA. Sum-difference theory of remembering and knowing: a two-dimensional signal-detection model. Psychol Rev 24; : Kalish ML, Dunn JC. What could cognitive neuroscience tell us about recognition memory? Aust J Psychol 22; 64: Slotnick SD, Thakral PP. The hippocampus operates in a threshold manner during spatial source memory. Neuroreport 23; 24: Mitchell KJ, Johnson MK. Source monitoring 5 years later: what have we learned from fmri about the neural mechanisms of source memory? Psychol Bull 29; 35: