Signal-detection, threshold, and dual-process models of recognition memory: ROCs and conscious recollection

In defense of the signal detection interpretation of remember/know judgments

Donaldson (1996) argued that remember/know judgments can be conceptualized within a signal detection framework by assuming that they are based on two criteria situated along a strength-of-memory decision axis. According to this model, items that exceed a high criterion receive a remember response, whereas items that only exceed a lower criterion receive a know response. Although a variety of findings have been presented in evidence against this idea, Dunn (2004) recently showed that detection theory is fully compatible with those findings. We present a variety of new results and new analyses that weigh strongly in favor of the detection interpretation. We further show that a dual-process account of recognition memory is compatible with a unidimensional detection model despite the common notion that such a model necessarily assumes a single process. The key assumption of this model is that individual recognition decisions are based on both recollection and familiarity (not on one process or the other).

fMRI Evidence for Separable and Lateralized Prefrontal Memory Monitoring Processes

Source memory research suggests that attempting to remember specific contextual aspects surrounding prior stimulus encounters results in greater left prefrontal cortex (PFC) activity than simple item-based old/new recognition judgments. Here, we tested a complementary hypothesis that predicts increases in the right PFC with tasks requiring close monitoring of item familiarity. More specifically, we compared a judgment of frequency (JOF) task to an item memory task, in which the former required estimating the number of previous picture encounters and the latter required discriminating old from new exemplars of previously seen items. In comparison to standard old/new recognition, both source memory and the JOF task examined here require more precise mnemonic judgments. However, in contrast to source memory, cognitive models suggest the JOF task relies heavily upon item familiarity, not specific contextual recollections. Event-related fMRI demonstrated greater recruitment of right, not left, dorso-lateral and frontopolar PFC regions during the JOF compared to item memory task. These data suggest a role for right PFC in the close monitoring of the familiarity of objects, which becomes critical when contextual recollection is ineffective in satisfying a memory demand.

Functional imaging and neuropsychology findings: how can they be linked?

It is argued that in poorly understood domains functional imaging and neuropsychology findings on cognitive processes can be related only through functional models of normal cognition. The psychological concept of "resource" can, however, be simply extrapolated to functional imaging. It is then argued that double dissociations can have analogous inferential power for extrapolation to models of normal cognition in functional imaging as in neuropsychology. The argument is illustrated by the example of the control processes involved in functional episodic memory imaging of experiments.

Transferring voice effects in recognition memory from remembering to knowing

In five experiments, we investigated the effects of voice congruency (same vs. different voices at study and at test) on remembering and knowing in recognition memory. With low- and medium-frequency three- or four-syllable words, a voice congruency effect occurred only in remembering. With nonwords, voice congruency effects occurred both in remembering and in knowing. With nonwords and divided attention at study, the voice congruency effect transferred almost completely from remembering to knowing. By showing a transfer of effects from remembering to knowing as encoding became more impoverished, these findings support a distinctiveness/fluency account of remembering and knowing as well as the theory that remembering and knowing indicate retrieval of events from episodic and semantic memory systems, respectively.

Recognition memory for source and occurrence: the importance of recollection

Previous recognition memory studies indicate that when both recollection and familiarity are expected to contribute to recognition performance (e.g., discriminating studied items from nonstudied items) the dual-process and the unequal-variance signal detection models provide comparable accounts of performance. When familiarity is not expected to be useful (e.g., when items from two equally familiar sources are discriminated between), the dual-process model provides a significantly better account of performance. In the present study, source recognition was tested under conditions in which familiarity could have been used to perform a list-discrimination task; participants were required to discriminate between strong studied items, weak studied items, and new items. The dual-process model provided a better account of performance than did the unequal-variance model. Moreover, the results indicated that the unequal-variance assumption in a single-process signal detection model was not a valid substitution for recollection and that recollection was used to make recognition judgments even when assessments of familiarity were useful.

The neural basis of episodic memory: evidence from functional neuroimaging

We review some of our recent research using functional neuroimaging to investigate neural activity supporting the encoding and retrieval of episodic memories, that is, memories for unique events. Findings from studies of encoding indicate that, at the cortical level, the regions responsible for the effective encoding of a stimulus event as an episodic memory include some of the regions that are also engaged to process the event 'online'. Thus, it appears that there is no single cortical site or circuit responsible for episodic encoding. The results of retrieval studies indicate that successful recollection of episodic information is associated with activation of lateral parietal cortex, along with more variable patterns of activity in dorsolateral and anterior prefrontal cortex. Whereas parietal regions may play a part in the representation of retrieved information, prefrontal areas appear to support processes that act on the products of retrieval to align behaviour with the demands of the retrieval task.

Components of episodic memory: the contribution of recollection and familiarity

The examination of recognition memory confidence judgements indicates that there are two separate components or processes underlying episodic memory. A model that accounts for these results is described in which a recollection process and a familiarity process are assumed to contribute to recognition memory performance. Recollection is assumed to reflect a threshold process whereby qualitative information about the study event is retrieved, whereas familiarity reflects a classical signal-detection process whereby items exceeding a familiarity response criterion are accepted as having been studied. Evidence from cognitive, neuropsychological and neuroimaging studies indicate that the model is in agreement with the existing recognition results, and indicate that recollection and familiarity are behaviourally, neurally and phenomenologically distinct memory retrieval processes.

The systematic discrepancy between A' for overall recognition and remembering: a dual-process account

Signal detection accounts of recognition assume that all item endorsements arise from the assessment of a single continuous indication of memory strength, even when subjects claim to categorically separate items accompanied by contextual recollection from those that are not (viz., remembering vs. knowing). Dissociations of these response types are held to occur because the former require a higher response criterion for item strength than does the latter. Meta-analytic and individual subject data suggest that when the A' metric is used, accuracy for remembering can systematically deviate from that of overall responding for individual subjects. This occurs because, unlike the symmetric and rigid receiver-operating characteristic (ROC) implied under A', empirical ROCs are asymmetric and plastic. A dual-process model predicted that the magnitude of the deviation would vary as a systematic function of the proportion of overall recognition accompanied by subjective remember reports for individual subjects. The predictions were confirmed using multiple regression on Monte Carlo and experimental data sets and were also shown to generalize to the double equal-threshold, single high-threshold [i.e., H - FA; (H - FA)/(1 - FA)], and the equal variance signal detection d' corrections. The unequal variance signal detection model was also shown to mirror the data, but only under the post hoc assumption that every subject adopts a very similar remember criterion placement rule. The results demonstrate that the systematic failure of tightly constrained models of recognition constitutes valuable regression data for more complex models and simultaneously highlights why single-point measures of accuracy are unsuitable as summaries across conditions or groups. Furthermore, the results show that remember rates carry unique information regarding the underlying processes governing individual subject performance that cannot be gleaned from the overall hit and false alarm rates in isolation.

Predicting individual false alarm rates and signal detection theory: a role for remembering

The relationships between hit, remember, and false alarm rates were examined across individual subjects in three remember-know experiments in order to determine whether signal detection theory would be consistent with the observed data. The experimental data differed from signal detection predictions in two critical ways. First, remember reports were unrelated, or slightly negatively related, to the commission of false alarms. Second, both response types (remembers and false alarms) were uniquely related to hit rates, which demonstrated that the hit rate cannot be viewed as the result of a single underlying strength process. These results are consistent with the dual-process signal detection model of Yonelinas (1994), in which performance is determined by two independent processes--retrieval of categorical context information (remembering) and discriminations based on continuous item strength. Remember and false alarm rates selectively tap these processes, whereas the hit rate is jointly determined. Monte Carlo simulations in which the dual-process model was used successfully reproduced the pattern in the experimental data, whereas simulations in which a signal detection model, with separate "old" and "remember" criteria, was used, did not. The results demonstrate the utility of examining individual differences in response types when one is evaluating memory models.

Confidence in recognition memory for words: dissociating right prefrontal roles in episodic retrieval

We used event-related functional magnetic resonance imaging (efMRI) to investigate brain regions showing differential responses as a function of confidence in an episodic word recognition task. Twelve healthy volunteers indicated whether their old-new judgments were made with high or low confidence. Hemodynamic responses associated with each judgment were modeled with an "early" and a "late" response function. As predicted by the monitoring hypothesis generated from a previous recognition study [Henson, R. N. A., Rugg, M. D., Shallice, T., Josephs, O., & Dolan, R. J. (1999a). Recollection and familiarity in recognition memory: An event-related fMRI study. Journal of Neuroscience, 19, 3962-3972], a right dorsolateral prefrontal region showed a greater response to correct low- versus correct high-confidence judgements. Several regions, including the precuneus, posterior cingulate, and left lateral parietal cortex, showed greater responses to correct old than correct new judgements. The anterior left and right prefrontal regions also showed an old-new difference, but for these regions the difference emerged relatively later in time. These results further support the proposal that different subregions of the prefrontal cortex subserve different functions during episodic retrieval. These functions are discussed in relation to a monitoring process, which operates when familiarity levels are close to response criterion and is associated with nonconfident judgements, and a recollective process, which is associated with the confident recognition of old words.

Recognition memory for faces: when familiarity supports associative recognition judgments

Recognition memory for single items can be dissociated from recognition memory for the associations between items. For example, recognition tests for single words produce curvilinear receiver operating characteristics (ROCs), but associative recognition tests for word pairs produce linear ROCs. These dissociations are consistent with dual-process theories of recognition and suggest that associative recognition relies on recollection but that item recognition relies on a combination of recollection and assessments of familiarity. In the present study, we examined associative recognition ROCs for facial stimuli by manipulating the central and external features, in order to determine whether linear ROCs would be observed for stimuli other than arbitrary word pairs. When the faces were presented upright, familiarity estimates were significantly above zero, and the associative ROCs were curvilinear, suggesting that familiarity contributed to associative judgments. However, presenting the faces upside down effectively eliminated the contribution of familiarity to associative recognition, and the ROCs were linear. The results suggest that familiarity can support associative recognition judgments, if the associated components are encoded as a coherent gestalt, as in upright faces.

Recollection and familiarity in recognition memory: an event-related functional magnetic resonance imaging study

The question of whether recognition memory judgments with and without recollection reflect dissociable patterns of brain activity is unresolved. We used event-related, functional magnetic resonance imaging (fMRI) of 12 healthy volunteers to measure hemodynamic responses associated with both studying and recognizing words. Volunteers made one of three judgments to each word during recognition: whether they recollected seeing it during study (R judgments), whether they experienced a feeling of familiarity in the absence of recollection (K judgments), or whether they did not remember seeing it during study (N judgments). Both R and K judgments for studied words were associated with enhanced responses in left prefrontal and left parietal cortices relative to N judgments for unstudied words. The opposite pattern was observed in bilateral temporoccipital regions and amygdalae. R judgments for studied words were associated with enhanced responses in anterior left prefrontal, left parietal, and posterior cingulate regions relative to K judgments. At study, a posterior left prefrontal region exhibited an enhanced response to words subsequently given R versus K judgments, but the response of this region during recognition did not differentiate R and K judgments. K judgments for studied words were associated with enhanced responses in right lateral and medial prefrontal cortex relative to both R judgments for studied words and N judgments for unstudied words, a difference we attribute to greater monitoring demands when memory judgments are less certain. These results suggest that the responses of different brain regions do dissociate according to the phenomenology associated with memory retrieval.

Experiences of remembering, knowing, and guessing

This article presents and discusses transcripts of some 270 explanations subjects provided subsequently for recognition memory decisions that had been associated with remember, know, or guess responses at the time the recognition decisions were made. Only transcripts for remember responses included reports of recollective experiences, which seemed mostly to reflect either effortful elaborative encoding or involuntary reminding at study, especially in relation to the self. Transcripts for know responses included claims of just knowing, and of feelings of familiarity. These transcripts indicated that subjects were often quite confident of the accuracy of their decisions, compared with those for guess responses. Transcripts for decisions associated with guess responses also expressed feelings of familiarity but additionally revealed various strategies and inferences that did not directly reflect memory for studied items. The article concludes with a historical and theoretical overview of some interpretations of the states of awareness measured by these responses.

Recognition memory ROCs for item and associative information: the contribution of recollection and familiarity

Receiver-operating characteristics (ROCs) were examined in three recognition memory experiments. ROCs for item information (i.e., was this word presented?) were found to be curvilinear. However, ROCs for associative information (i.e., were these two words presented together?) were found to be linear. The results are in agreement with the predictions of a dual-process model that assumes that recognition judgments are based on familiarity and recollection. Familiarity reflects the assessment of a continuous strength dimension and is well described as a signal detection process, whereas recollection reflects the retrieval of qualitative information about the study episode and behaves like a discrete threshold process. The results showed that memory judgments about items relied on a combination of recollection and familiarity, but that judgments about associations relied primarily on recollection. Further examination of the associative ROCs suggested that subjects were able to recollect that old pairs of items were in the study list, and, under some conditions, that new pairs were not in the study list.