Dual-Process Signal Detection (Remember/Know Paradigm) – John M. Gardiner & Andrew P. Yonelinas

Human recognition memory—the capacity to judge whether an event, stimulus, or individual has been previously encountered—lies at the intersection of cognitive psychology, psychophysics, and systems neuroscience. For decades, memory research was dominated by single-process formulations positing that recognition judgments are made along a unidimensional continuum of memory trace strength. In this classical view, an item is recognized as old simply if its associated trace strength exceeds a particular response criterion. However, this parsimonious conception routinely struggled to capture the rich, multidimensional phenomenology of human retrieval: specifically, the qualitative divergence between suddenly re-experiencing the precise episodic context of an initial encounter versus experiencing an intuitive, context-free feeling that an item is familiar.

The transformation of recognition memory from a simplistic, scalar psychophysical problem into a sophisticated dual-process framework represents one of the most intellectually vibrant chapters in cognitive science. This shift was catalyzed by two complementary research traditions: the phenomenological systematization initiated by John M. Gardiner, who operationalized Endel Tulving’s theoretical concepts of autonoetic and noetic consciousness via the Remember/Know procedure, and the psychophysical and quantitative architecture formalized by Andrew P. Yonelinas, who synthesized qualitative retrieval thresholds with classical signal detection mechanics to produce the Dual-Process Signal Detection (DPSD) model. Together, these paradigms bridged subjective introspective experiences and rigorous mathematical formalizations, fundamentally reshaping our understanding of human mnemonic architecture.

This comprehensive treatise examines the historical, theoretical, mathematical, and neurobiological dimensions of the Remember/Know paradigm and the Dual-Process Signal Detection model. Across twelve detailed sections, we trace how epistemological debates regarding human consciousness evolved into formal receiver operating characteristic (ROC) curves; how empirical double dissociations dismantled monolithic strength accounts; how neural systems within the medial temporal lobe, prefrontal cortex, and posterior parietal regions divide the computational labor between threshold recollection and continuous familiarity; and how these foundational methodologies continue to illuminate human cognition across the lifespan, in clinical pathology, and within contemporary computational neuroscience.

1. Historical and Theoretical Foundations of Recognition Memory

1.1 Early Epistemological Views on Human Memory Retrieval

The philosophical and scientific interrogation of recognition memory possesses a lineage that extends back to classical antiquity. In Aristotle’s treatise De Memoria et Reminiscentia, a pivotal distinction was forged between mneme (the passive preservation of an experience) and anamnesis (the active, effortful recollection of an event through associative chains). This distinction foreshadowed the fundamental tension that modern cognitive psychology would confront centuries later: the contrast between an immediate, non-deliberate recognition and a deliberate, reconstructive retrieval process. Associationist philosophers of the eighteenth and nineteenth centuries, including David Hartley, David Hume, and James Mill, expanded upon this foundation by proposing that recognition is governed by the strength and frequency of mental associations formed between sensory impressions.

When William James established the foundations of modern American psychology in 1890, he introduced an exceptionally prescient taxonomy of retrieval experiences. James posited a functional separation between primary memory—the fleeting contents of immediate consciousness—and secondary memory, which involves the retrieval of knowledge that has slipped from awareness and must be brought back into the conscious sphere. Crucially, James observed that the act of remembering an event is not simply the revival of an isolated psychic image; it requires that the image be experienced with a sense of pastness and accompanied by contextual associates. James wrote that an event must be recognized as having occurred within the personal continuum of one’s past, surrounded by idiosyncratic peripheral details, anticipations, and affective tones, an observation that directly prefigured the contemporary concept of episodic recollection.

Despite James’s nuanced phenomenological insights, the experimental psychology of the early and mid-twentieth century largely abandoned qualitative descriptions in favor of behaviorally tractable, unidimensional metrics. Dominated by the verbal learning tradition established by Hermann Ebbinghaus and later refined by associationist paradigms, researchers conceptualized recognition memory through single-factor models. Memory traces were operationalized as scalar quantities, often termed “trace strength” or “habit strength.” Recognition was viewed as an elementary discrimination task: if the perceptual or conceptual trace strength of a test probe exceeded a threshold value, the participant responded affirmatively; otherwise, the probe was rejected. This single-strength paradigm treated recall and recognition not as manifestations of qualitatively distinct retrieval operations, but merely as tasks differing in their sensitivity to a singular underlying memory trace. Recall was viewed as requiring a high threshold of strength, whereas recognition was assumed to demand a considerably lower threshold, a conceptualization formalized in early “generate-recognize” models.

By the late 1960s and 1970s, the limits of these single-factor formulations became increasingly apparent within the cognitive revolution. Researchers observed that verbal memory performance was systematically influenced by manipulations that could not be readily integrated into a unidimensional strength axis. Investigators such as George Mandler began drawing attention to the reality that a stimulus could be recognized via two demonstrably distinct mechanisms. Mandler famously illustrated this dichotomy through the “butcher-on-the-bus” phenomenon: an individual encounters their local butcher seated on a public bus and immediately recognizes the face as intensely familiar, yet remains utterly incapable of retrieving the person’s name, profession, or the circumstances under which they are known. Only after prolonged, effortful search does the contextual information flood into consciousness: the white apron, the meat counter, and the grocery store. This intuitive yet theoretically disruptive demonstration underscored that the subjective certainty of an encounter can dissociate completely from the retrieval of episodic associative context, exposing the conceptual inadequacy of monolithic trace-strength models.

1.2 Tulving’s Distinction Between Episodic Consciousness Modes

The modern era of dual-process memory research was fundamentally inaugurated by Endel Tulving in the early 1980s. Tulving had previously altered cognitive theory by proposing the structural separation between episodic memory (the memory system dedicated to personally experienced events situated in specific spatial and temporal matrices) and semantic memory (the repository of decontextualized factual knowledge about the world). In his seminal 1985 paper, “Memory and Consciousness,” Tulving moved beyond purely structural classifications of memory stores to address the qualitative, phenomenological textures of conscious awareness that accompany the act of retrieval.

Tulving introduced a tripartite taxonomy of conscious states, linking each to a distinct memory system:

• Anoetic consciousness: Characterized as temporally non-reflective and tied to the immediate present; associated with procedural memory and non-declarative learning systems.
• Noetic consciousness: Defined as an awareness of the world and its facts without any subjective sense of personal re-experiencing; the hallmark of semantic memory retrieval.
• Autonoetic consciousness: Defined as “self-knowing” consciousness; the unique phenomenological signature of episodic memory retrieval, enabling an individual to engage in “mental time travel.”

Through autonoetic consciousness, an individual does not merely possess information that an event transpired; they mentally project themselves back across subjective time to re-experience the event within its original perceptual, affective, and spatiotemporal environment.

To capture these elusive subjective states in an empirical laboratory setting, Tulving introduced the Remember/Know (R/K) procedure. In this testing paradigm, participants were presented with a study list of items (typically words) and subsequently given a recognition memory test. However, rather than simply rendering a binary “Old” or “New” decision, participants who judged an item to be “Old” were required to provide a meta-cognitive, phenomenological report: they were instructed to designate the item as “Remember” if the recognition was accompanied by conscious recollection of specific details, context, or thoughts from the encoding episode (autonoetic awareness), or as “Know” if the item felt familiar and recognized as having appeared on the list, yet lacked any conscious retrieval of its original encoding context (noetic awareness).

Tulving’s conceptualization carried profound evolutionary implications. He posited that autonoetic consciousness and episodic memory were phylogenetically recent evolutionary developments, likely unique to hominids or shared only with a narrow range of advanced mammals. The capacity to engage in mental time travel was not an incidental byproduct of cognitive processing, but an evolutionary adaptation allowing hominids to mentally simulate future scenarios, evaluate potential consequences of planned actions, and build complex, temporally extended social coalitions based on detailed autobiographical records. By establishing that episodic retrieval is fundamentally characterized by autonoetic awareness, Tulving provided the theoretical foundation for transforming the study of recognition memory from a mechanistic exercise in perceptual discrimination into an exploration of human conscious experience.

1.3 The Evolution from Phenomenological Reports to Quantitative Modeling

While Tulving’s Remember/Know paradigm offered an elegant tool for examining states of conscious awareness, its early reception within experimental psychology was marked by skepticism. Traditional experimental psychophysics and mathematical cognitive psychology had long maintained a profound wariness toward introspective reporting methods. Critics argued that the R/K distinction might represent nothing more than arbitrary linguistic labels that human subjects map onto varying levels of subjective confidence. According to this skeptical perspective, a “Remember” response was simply an old judgment executed with high confidence, whereas a “Know” response was an old judgment executed with moderate or low confidence, both arising from a unidimensional continuum of memory strength.

This epistemological tension created a clear methodological imperative: if the distinction between autonoetic recollection and noetic familiarity represented genuine, functionally independent cognitive sub-processes rather than semantic artifacts or confidence gradations, the phenomenon had to be formalized within a quantitative mathematical architecture. The field required an analytical framework capable of disentangling genuine memory sensitivity from participant-specific response criteria, response biases, and subjective confidence thresholds. The primary quantitative model of the era, classical Signal Detection Theory (SDT), had established itself as the gold standard for measuring perceptual discrimination, yet SDT had historically evolved under the assumption of a single, continuous, normally distributed sensory evidence variable.

The challenge was to construct a mathematical synthesis that could simultaneously respect the qualitative richness of phenomenological reports while preserving the psychophysical rigor of signal detection metrics. Researchers realized that if recollection was truly an all-or-none, threshold-based episodic retrieval event, and familiarity was a graded, continuous signal-detection variable, their mathematical interaction would generate distinct empirical signatures—specifically within the shapes of Receiver Operating Characteristic (ROC) curves—that would be impossible to reconcile with classical single-process models. Thus began the transition from descriptive phenomenological psychology to computational dual-process modeling, a development that would revolutionize the empirical study of the human mind.

2. John M. Gardiner and the Systematization of the Remember/Know Paradigm

2.1 Gardiner’s Methodological Standardization of Remember and Know Instructions

Although Endel Tulving devised the initial concept of the Remember/Know procedure, it was the British cognitive psychologist John M. Gardiner who methodologically refined, standardized, and systematically defended the paradigm against early criticisms. When Gardiner began his systematic investigations in the late 1980s, he recognized that the greatest threat to the validity of the R/K paradigm was instructional ambiguity. If experimental participants misunderstood the distinction between “Remembering” and “Knowing,” or if they systematically conflated “Knowing” with guessing, the experimental results would be hopelessly confounded by fluctuating subjective response criteria.

Gardiner established standardized behavioral instructional protocols designed to rigorously isolate autonoetic recollection from noetic familiarity. In his protocols, participants were explicitly trained before the test phase using comprehensive real-world analogies and standardized scripts. A “Remember” response was defined strictly as the conscious retrieval of specific episodic associations that occurred at the moment the word was presented: this could include conscious recollection of a visual image formed in response to the word, an association with a personal event, awareness of the physical appearance of the font, or the retrieval of a word presented immediately prior or subsequent to the probe. In sharp contrast, a “Know” response was operationalized as a definitive, confident recognition that the word had been on the studied list, but where the individual was entirely unable to retrieve any specific conscious thought, image, or context from the study episode.

To prevent statistical and conceptual contamination, Gardiner took the crucial step of introducing an explicit “Guess” response category in tripartite testing protocols (Remember, Know, or Guess). Prior to this innovation, participants who were uncertain of an item’s status frequently defaulted to a “Know” response, creating an artificial inflation of the noetic category with low-confidence noise. By providing an independent outlet for pure guesses, Gardiner ensured that “Know” responses reflected genuine, high-confidence familiarity rather than uncertainty. Gardiner’s instructional scripts became the gold standard across international cognitive laboratories, yielding exceptional cross-laboratory reliability and establishing the empirical foundation necessary to test the reality of dual cognitive mechanisms.

2.2 Empirical Functional Dissociations Identified by Gardiner

Armed with a standardized methodology, Gardiner embarked on an extensive program of experimental research throughout the late 1980s and the 1990s, aimed at demonstrating that Remember and Know responses could be functionally dissociated. If Remember and Know were merely different points along a single continuum of memory trace strength, then any independent variable that increased or decreased memory strength should cause Remember and Know judgments to move in parallel, or at least in a predictable, monotonic fashion. If, however, they reflected functionally distinct memory systems or processing modes, experimental manipulations should be capable of producing double dissociations—selectively influencing one retrieval experience while leaving the other completely unchanged, or affecting them in diametrically opposing directions.

Gardiner and his colleagues successfully identified a series of dramatic empirical dissociations:

• Levels-of-Processing Manipulations: In seminal studies (e.g., Gardiner, 1988), participants encoded words under either shallow perceptual conditions (e.g., counting syllables, judging font characteristics) or deep semantic conditions (e.g., generating semantic associations, judging pleasantness). The results were unambiguous: deep semantic processing dramatically increased the proportion of “Remember” responses relative to shallow processing, but left the proportion of “Know” responses virtually unaffected or slightly reduced. This finding demonstrated that autonoetic recollection is selectively sensitive to semantic elaboration during encoding.
• Perceptual and Modality Shifts: Conversely, when Gardiner altered the physical, surface characteristics of stimuli between encoding and retrieval (such as shifting presentation modality from auditory at study to visual at test, or changing the visual typography), he observed the inverse pattern. Perceptual shifts had minimal effect on the rate of “Remember” responses, but significantly suppressed “Know” responses, proving that noetic familiarity is heavily contingent upon perceptual fluency and data-driven overlap between study and test representations.
• Divided Attention at Encoding: Gardiner and Parkin (1990) manipulated cognitive load during the encoding phase by requiring participants to perform a secondary auditory continuous reaction-time task. Dividing attention during study decimated subsequent “Remember” responses, demonstrating that the binding of episodic features into recollective traces requires deliberate, central executive resources. Remarkable, however, was the finding that divided attention left “Know” responses completely invariant: familiarity-based recognition was encoded automatically, without demanding conscious, focal attention.
• Differential Retention Intervals: Temporal decay functions revealed striking asymmetries. Over extended retention intervals spanning hours, days, and weeks, the proportion of “Remember” judgments exhibited rapid, steep decay curves, whereas “Know” judgments exhibited remarkable stability over time, sometimes even demonstrating relative increases as recollected details faded and left behind lingering feelings of uncontextualized familiarity.

2.3 Theoretical Assertions: Separate Memory Systems versus Processing Modes

The accumulation of these robust functional dissociations led Gardiner to engage deeply with the theoretical architecture of human memory. Initially, Gardiner aligned his work closely with Tulving’s structural view, arguing that the Remember/Know paradigm provided empirical validation for the physical existence of distinct, dissociable memory systems—namely, that Remember responses were direct readouts of the Episodic Memory System, while Know responses reflected the operation of the Semantic Memory System.

However, as the empirical database expanded, Gardiner refined his theoretical assertions, adopting a more flexible stance that incorporated processing-oriented frameworks. Drawing upon the transfer-appropriate processing framework advanced by Roediger, Weldon, and Challis, Gardiner proposed that Remember and Know judgments might reflect the engagement of qualitatively distinct cognitive processing modes rather than rigidly separated anatomical compartments. Specifically, he conceptualized Remember judgments as emerging from conceptually driven, relational cognitive processing that integrates items into elaborate associative networks. In contrast, Know judgments were conceptualized as the product of data-driven, item-specific perceptual processing that relies heavily on sensory integration and processing fluency.

Central to Gardiner’s theoretical writing was the resolute rejection of the unidimensional strength view. He argued that the existence of crossover and double dissociations could not be reconciled with any model positing that Remember and Know merely represented arbitrary, tiered response criteria placed along a single sensory continuum. Gardiner synthesized these empirical outcomes with Tulving’s concept of *synergetic ecphory*—the process by which retrieval cues interact with stored engrams to generate conscious retrieval. Gardiner maintained that the nature of the ecphoric experience was fundamentally dichotomous: an ecphoric event either succeeds in reinstating the qualitative, autonoetic context of the personal past, or it yields an isolated noetic awareness of past occurrence. By establishing this theoretical baseline, Gardiner set the stage for mathematical psychophysicists to construct formal, predictive models of these dual ecphoric events.

3. Classical Signal Detection Theory and Recognition Memory

3.1 Fundamentals of Equal-Variance Signal Detection (EVSD)

To appreciate the intellectual revolution ignited by dual-process modeling, one must examine the mathematical foundations of classical psychophysical measurement. For decades, Signal Detection Theory (SDT), originally developed in the context of radar technology and sensory psychophysics by Green and Swets (1966), served as the dominant quantitative framework for human recognition memory. Classical SDT was attractive precisely because it solved the long-standing psychometric problem of distinguishing between an individual’s true discriminative memory sensitivity and their subjective decision bias or response criterion.

The Equal-Variance Signal Detection (EVSD) model assumes that memory strength is a continuous random variable distributed normally across the decision space. When a list of items is studied, these items form a “target” (signal-plus-noise) distribution, while unstudied items constitute a “lure” (noise) distribution. In the standard EVSD formulation, both distributions are assumed to be Gaussian and to possess identical variance ($\sigma_{\text{lure}}^2 = \sigma_{\text{target}}^2 = 1.0$). The lure distribution is centered at zero, while the target distribution is centered at a positive distance denoted by the sensitivity index, $d’$ (d-prime):

$$d’ = \frac{\mu_{\text{target}} – \mu_{\text{lure}}}{\sigma}$$

To make an old/new decision, the observer places a decision criterion ($c$) along this unidimensional strength axis. If a test item evokes a memory trace strength exceeding $c$, the observer responds “Old”; otherwise, the observer responds “New.” The probability of a Hit (correctly classifying a target as old) and the probability of a False Alarm (incorrectly classifying a lure as old) are defined by the cumulative normal distribution function ($Phi$):

$$P(\text{Hit}) = \Phi\left(\frac{\mu_{\text{target}} – c}{\sigma_{\text{target}}}\right) = \Phi(d’ – c)$$

$$P(\text{False Alarm}) = \Phi\left(\frac{\mu_{\text{lure}} – c}{\sigma_{\text{lure}}}\right) = \Phi(-c)$$

Under EVSD, manipulating the observer’s conservatism (e.g., through instructions or payoff matrices) shifts the criterion $c$ along the horizontal axis, but leaves $d’$ invariant. While mathematically parsimonious, empirical memory research quickly exposed a fatal flaw in the EVSD formulation: human recognition memory data almost never conform to the assumption of equal variance.

3.2 The Unequal-Variance Signal Detection (UVSD) Challenge

When researchers evaluated empirical memory performance across multiple levels of response confidence, they plotted empirical Receiver Operating Characteristic (ROC) curves—functions mapping Hit rates against False Alarm rates across varied confidence thresholds. Under the EVSD model, the ROC curve must be symmetrically positioned along the minor diagonal, and the corresponding z-transformed ROC (z-ROC), calculated by transforming Hit and False Alarm probabilities into standard normal deviations ($z$-scores), must produce a straight line with a slope exactly equal to 1.0 ($s = \sigma_{\text{lure}} / \sigma_{\text{target}} = 1.0$).

Empirical findings consistently contradicted this prediction. Across hundreds of experiments involving verbal, pictorial, and associative stimuli, the empirical z-ROC slope systematically averaged between 0.75 and 0.85, rarely approaching 1.0. An empirical slope less than 1.0 mathematically dictates that the variance of the studied target distribution is substantially greater than the variance of the unstudied lure distribution ($\sigma_{\text{target}} > \sigma_{\text{lure}}$). This led to the widespread adoption of the Unequal-Variance Signal Detection (UVSD) model, which formally incorporates the standard deviation ratio parameter ($s$):

$$z(\text{Hit}) = s \cdot z(\text{False Alarm}) + \frac{d_e}{\sigma_{\text{target}}}$$

Within the UVSD framework, proponents of single-process models argued that Gardiner’s Remember/Know dissociations were simply an artifact of criterion placement along a single, continuous, unequal-variance memory strength dimension. According to this view, an item possesses a single scalar value of mnemonic strength. If that strength surpasses a highly conservative decision criterion ($c_{\text{Remember}}$), the subject issues a “Remember” response. If the strength exceeds an intermediate, less conservative criterion ($c_{\text{Know}}$) but falls below $c_{\text{Remember}}$, the subject issues a “Know” response. Thus, UVSD theorists asserted that Remember and Know did not reflect fundamentally different qualitative processes or consciousness states, but were merely labels for regions along a singular strength continuum characterized by an unequal-variance distribution.

3.3 Limitations of Univariate Signal Detection Models

Despite the computational elegance of the UVSD model, its fundamental assertion—that recognition memory is fully reducible to a univariate strength dimension—faced mounting theoretical and empirical limitations. The first limitation was phenomenological: the UVSD model treated the introspective reality of conscious recollection as an epiphenomenon. It offered no psychological explanation for why a participant would explicitly report retrieving an idiosyncratic episodic detail, such as the visual image of a former classmate when reading the word “school,” simply because an internal strength variable had crossed an arbitrary mathematical threshold.

Second, the single-process UVSD model struggled to account for clinical and neurobiological dissociations. If memory strength is a unitary scalar quantity, an individual suffering from focal brain lesions should display a generalized reduction in sensitivity ($d’$), shifting performance downwards along the single-process curve. However, neuropsychological patients with selective bilateral damage to the hippocampus frequently exhibited catastrophic losses of the ability to recollect context, while demonstrating entirely normal discriminability when recognition could be supported by item familiarity. UVSD models could only model these patient populations by ad-hoc adjusting distribution variances in mathematically arbitrary ways.

Third, statistical overparameterization concerns emerged. While UVSD curves fit empirical ROC points with high precision, researchers noted that mathematical flexibility does not guarantee psychological validity. The UVSD model explained the characteristic asymmetry of recognition ROCs solely by positing an arbitrary expansion of target variance during encoding, yet failed to provide a mechanistic, neurocomputational rationale for why such variance expansion occurs specifically for studied items. The field found itself in an intellectual deadlock: phenomenological psychology had demonstrated rich qualitative dissociations that lacked mathematical formalization, while psychophysical signal detection theory possessed mathematical rigor that denied qualitative reality. Resolving this impasse required an entirely new quantitative framework.

4. Andrew P. Yonelinas and the Dual-Process Signal Detection (DPSD) Formulation

4.1 Conceptual Architecture of the Yonelinas Dual-Process Framework

In a series of landmark papers beginning in 1994, Andrew P. Yonelinas formulated a comprehensive mathematical and theoretical model that successfully synthesized Gardiner’s phenomenological insights with psychophysical signal detection theory: the Dual-Process Signal Detection (DPSD) model. Yonelinas asserted that recognition memory is supported by two qualitatively and quantitatively distinct retrieval processes operating in parallel: **Recollection** and **Familiarity**.

Yonelinas formally defined these two components:

• Recollection: Characterized as a threshold retrieval process. Recollection involves the retrieval of specific associative, qualitative, and contextual details regarding the study episode (e.g., spatial location, temporal ordering, internal thoughts). Yonelinas conceptualized recollection as operating akin to a high-threshold mechanism: an individual either retrieves the contextual details of an event with high confidence, or the process fails completely, yielding zero contextual information. It does not operate as a continuous, graded signal; rather, it provides an all-or-none qualitative ecphoric recovery.
• Familiarity: In sharp contrast, familiarity is conceptualized as a continuous, graded signal detection process. It reflects an assessment of memory trace strength in the total absence of qualitative contextual retrieval. Familiarity operates precisely according to classical equal-variance Gaussian signal detection principles. Test items generate varying levels of familiarity along a normal distribution, and the participant renders judgments by setting response criteria along this continuum.

A crucial theoretical axiom of the Yonelinas framework is the **Independence Assumption**. Yonelinas postulated that recollection and familiarity operate as statistically independent cognitive mechanisms. The probability that an item evokes recollection is independent of the probability that it evokes familiarity. Consequently, an item can evoke both recollection and familiarity, recollection without familiarity, familiarity without recollection, or neither process. This independence assumption provided the mathematical leverage required to resolve the measurement confounds that had historically plagued the Remember/Know paradigm.

4.2 Mathematical Formulation of the DPSD Model

The mathematical formulation of the Dual-Process Signal Detection model is elegant yet rigorous. Because recollection is formalized as a high-threshold probability parameter and familiarity as a continuous Gaussian process, the probability of a participant endorsing a studied target item as “Old” across a given response criterion ($c_i$) is the union of two independent probabilities: the probability that the item is recollected, plus the probability that the item fails to be recollected but exceeds the familiarity criterion.

Let $R$ represent the probability of high-threshold recollection, $d’$ represent the discriminability index of the continuous familiarity distribution, and $c_i$ represent the familiarity decision criterion for a specific confidence category $i$. The cumulative probability of a Hit at criterion $c_i$ is expressed as:

$$P(\text{Hit}_i) = R + (1 – R) \cdot \Phi\left(\frac{d’}{2} – c_i\right)$$

For unstudied lure items, because they were never presented during the study phase, the probability of true recollection is assumed to be zero ($R = 0$). Therefore, the cumulative probability of a False Alarm at criterion $c_i$ is driven entirely by the familiarity distribution of lures exceeding the decision criterion:

$$P(\text{False Alarm}_i) = \Phi\left(-\frac{d’}{2} – c_i\right)$$

Here, $Phi$ denotes the standard cumulative normal distribution function, with the midpoint between the lure and target familiarity distributions centered at zero, meaning the lure distribution has a mean of $-d’/2$ and the target familiarity distribution has a mean of $+d’/2$, both possessing unit variance ($\sigma = 1$).

To extract the parameters $R$ and $d’$ from experimental data, non-linear optimization algorithms (such as Levenberg-Marquardt or Nelder-Mead simplex algorithms) or Maximum Likelihood Estimation (MLE) are applied across multi-point confidence ratings. Alternatively, when working with standard Remember/Know data, Yonelinas formulated the **Independence Remember/Know (IRK)** estimation method. Because participants in standard R/K tasks are forced to choose mutually exclusive categories—rendering a “Know” response only when they do not “Remember”—the observed proportion of Know responses, $P(K)$, systematically underestimates the true underlying familiarity rate. Under the independence assumption, familiarity has the opportunity to express itself only when recollection fails. Thus:

$$P(R) = R$$

$$P(K) = (1 – R) \cdot F$$

Algebraically rearranging this equation yields the true, uncontaminated estimate of Familiarity ($F$):

$$F = \frac{P(K)}{1 – P(R)}$$

This simple mathematical correction transformed the quantitative utility of Gardiner’s paradigm, permitting researchers to estimate pure familiarity rates uncontaminated by fluctuations in recollection.

4.3 Bridging Phenomenological Reports and Formal Measurement

The development of the DPSD model successfully bridged the long-standing divide between Gardiner’s phenomenological approach and formal psychophysical measurement. Yonelinas demonstrated that the subjective reports collected in Remember/Know experiments mapped directly onto the underlying mathematical parameters extracted from confidence-based Receiver Operating Characteristics.

In a series of landmark empirical validations, Yonelinas tested participants across both paradigms simultaneously: subjects were asked to provide 6-point confidence ratings while also rendering Remember/Know judgments. When Yonelinas calculated the parameter estimates of $R$ and $d’$ from the confidence ROC curves using non-linear curve fitting, and independently calculated $R$ and $F$ from the Remember/Know reports using the IRK formulas, the resulting estimates exhibited near-perfect alignment. Experimental manipulations that selectively suppressed the $R$ parameter derived from ROC curves (such as shallow encoding or rapid response deadlines) exerted a mathematically identical suppression on the subjective “Remember” reports collected under Gardiner’s instructions.

This convergence resolved the historical criticism that Remember/Know responses were merely subjective artifacts or arbitrary confidence thresholds. If Remember and Know judgments were driven by non-specific shifts along a single strength continuum, they could not have demonstrated consistent parameter congruence with mathematically independent ROC curve properties across diverse experimental paradigms. Yonelinas’s mathematical architecture proved that Gardiner’s standardized instructions had tapped directly into the underlying dual-process psychophysics of human memory.

5. Receiver Operating Characteristic (ROC) Analysis in Dual-Process Research

5.1 Empirical Construction and Properties of Recognition ROCs

The construction and analysis of Receiver Operating Characteristics (ROCs) represents the central empirical tool used to arbitrate between competing recognition models. In a typical recognition ROC paradigm, participants are presented with studied and unstudied items and asked to rate their recognition confidence along an ordinal scale (e.g., 1 = Sure New, 2 = Probably New, 3 = Maybe New, 4 = Maybe Old, 5 = Probably Old, 6 = Sure Old). By cumulating the hit and false alarm probabilities from the most conservative criterion (Category 6: “Sure Old”) down to the most liberal criterion (Category 2: including all ratings except “Sure New”), a sequence of empirical points is generated and plotted on a Cartesian coordinate space, where the y-axis represents the cumulative Hit Rate and the x-axis represents the cumulative False Alarm Rate.

Empirical recognition ROCs exhibit three fundamental geometric properties:

• Curvilinearity: The ROC is non-linear and bows upward toward the upper-left coordinate $(0, 1)$, indicating that memory discriminability is substantially greater than chance performance.
• Asymmetry: Unlike the symmetric ROC curves characteristic of elementary sensory psychophysics, recognition memory ROCs are strikingly asymmetrical. The curve rises sharply along the y-axis at very low false alarm rates, and then flattens as it approaches the upper right quadrant.
• Non-Zero Y-Intercept: When an empirical ROC curve is fitted back to the y-axis (where the false alarm rate equals zero), it does not intersect the origin $(0, 0)$. Instead, it exhibits a distinct, positive, non-zero y-intercept. In the DPSD model, this non-zero intercept provides an empirical signature of high-threshold recollection: even under the most conservative response criteria, a substantial proportion of studied items are recollected with absolute certainty, generating hits in the complete absence of false alarms.

When these empirical ROCs are converted into $z$-space ($z$-ROCs) by plotting $z(\text{Hit})$ against $z(\text{False Alarm})$, empirical recognition data exhibit a characteristic, slight U-shaped curvature. While the single-process UVSD model predicts that $z$-ROCs must be strictly linear, the DPSD model predicts precisely this slight non-linear, upward-bowing curvature in $z$-space, because the curve is formed by the mixture of a linear threshold process and a continuous Gaussian process.

5.2 Extracting Recollection and Familiarity from ROC Parameters

To formally derive the quantitative parameters of the DPSD model from multi-point confidence ROCs, researchers utilize non-linear regression and Maximum Likelihood Estimation (MLE). The fitting procedure iteratively optimizes two parameters: the recollection capacity parameter ($R$, constrained between 0.0 and 1.0) and the familiarity sensitivity parameter ($d’$, constrained to non-negative values), alongside the set of familiarity decision criteria ($c_i$).

The extraction of these parameters relies on specific geometric properties of the ROC curve:

• Degree of Asymmetry: The parameter $R$ is directly derived from the height of the y-intercept and the degree of asymmetry in the ROC curve. When recollection is high, the curve is pulled strongly upward at the left vertical boundary, creating high asymmetry and an elevated y-intercept. If recollection is eliminated ($R = 0$), the y-intercept drops to the origin, and the ROC curve becomes entirely symmetrical.
• Degree of Curvature: The parameter $d’$ reflects the degree of continuous, curvilinear bowing exhibited by the ROC points as they progress across the interior of the coordinate space. Higher familiarity discriminability causes the curve to bow more deeply toward the upper-left coordinate independent of the intercept height.

Model fitting reliability is established through quantitative model comparison metrics, most prominently the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). By penalizing models for structural overparameterization, these metrics allow researchers to determine whether the inclusion of separate $R$ and $d’$ parameters provides a statistically superior account of empirical confidence data compared to the single-process UVSD model with its standard deviation ratio ($s$). Decades of model fitting have demonstrated that across diverse recognition paradigms, DPSD fits empirical confidence data with exceptional mathematical stability, yielding low residuals and high predictive power.

5.3 Contrasting DPSD Predictions with Single-Process UVSD ROC Predictions

The mathematical debate between DPSD and UVSD models centers on their competing geometric accounts of ROC properties. The Unequal-Variance Signal Detection model accounts for ROC asymmetry not by positing a qualitative retrieval threshold, but by asserting that the standard deviation of the target distribution ($\sigma_{\text{target}}$) is larger than that of the lure distribution ($\sigma_{\text{lure}}$). Under UVSD, as an observer shifts their decision criterion from conservative to liberal, they sweep continuously through two overlapping normal distributions, generating an asymmetrical ROC curve that terminates at the origin $(0, 0)$ without a true discontinuous y-intercept.

This fundamental mathematical divergence leads to radically different empirical predictions at the extreme boundaries of performance:

• Boundary Conditions at Extreme Conservatism: The DPSD model predicts that at the most conservative response criterion, the hit rate will remain elevated at a discrete value equal to $R$, even when the false alarm rate is driven asymptotically toward zero. The UVSD model predicts that as the false alarm rate approaches zero, the hit rate must continuously decline toward zero. High-powered psychophysical studies sampling extreme conservative criteria have routinely revealed that hit rates flatten out at a non-zero plateau, precisely as predicted by the DPSD threshold formulation.
• Line-Fitting in z-Space: The UVSD model mathematically mandates that z-transformed ROCs must be strictly linear, with the slope corresponding to the ratio $\sigma_{\text{lure}} / \sigma_{\text{target}}$. In contrast, the DPSD model predicts that $z$-ROCs must display a subtle, systematic U-shaped upward curvature due to the high-threshold component. Large-scale meta-analyses encompassing hundreds of experimental conditions have repeatedly confirmed the presence of this non-linear U-shaped curvature in empirical $z$-ROCs, providing powerful quantitative support for the dual-process architecture over univariate formulations.

6. The Methodological Debate: Single-Process versus Dual-Process Models

6.1 The 1-df Single-Process Critiques (Wixted, Stretch, Donaldson)

The ascendancy of the Dual-Process Signal Detection model did not go unchallenged. A formidable intellectual counter-offensive was mounted by proponents of single-process models, led prominently by John T. Wixted, alongside researchers such as Wayne Donaldson and Vincent Stretch. Wixted argued that the dual-process framework was theoretically unparsimonious, unnecessary, and built upon fundamentally flawed interpretations of experimental dissociations.

Wixted’s primary contention was that recognition memory is governed by a singular, unidimensional continuous variable—which he termed “memory strength” or “decision variable strength.” According to Wixted’s 1-degree-of-freedom single-process formulation, when a participant is presented with a test probe, the cognitive system generates a unitary scalar value representing the aggregate evidence for past occurrence. The participant does not consult separate, independent cognitive mechanisms (recollection versus familiarity); rather, they simply place multiple decision criteria along this single continuum. An item that evokes strong memory strength crosses the most conservative criterion and is labeled “Remember”; an item evoking intermediate strength crosses a lower criterion and is labeled “Know”; and an item evoking weak strength is rejected as “New.”

The single-process critique reinterpreted Gardiner’s functional dissociations as mathematical artifacts of criterion placement along an unequal-variance strength axis. Donaldson (1996) demonstrated through mathematical simulations that if an experimental manipulation increases the distance between target and lure distributions while participants hold their internal criteria constant, the mathematical changes in the proportions of items falling above the high criterion versus falling between criteria can create the statistical illusion of a double dissociation. Wixted further demonstrated that when participants are asked to provide confidence ratings nested within their Remember judgments, they exhibit a continuous range of confidence, a finding he argued was fundamentally incompatible with Yonelinas’s claim that recollection is an all-or-none, discrete threshold process.

6.2 Rebuttals and Counter-Evidence from Yonelinas and Gardiner

Yonelinas, Gardiner, and their colleagues met the single-process critique with extensive empirical and computational rebuttals. First, they demonstrated that while certain simple dissociations could theoretically be simulated by shifting criteria along a UVSD axis, true double dissociations mathematically violate any model positing monotonic shifts along a single continuum. If an experimental manipulation simultaneously drives the proportion of Remember responses up while driving the proportion of Know responses down—in a manner that cannot be fitted by any single criterion shift without violating distribution properties—the single-process hypothesis collapses.

Yonelinas highlighted critical empirical paradigms demonstrating such unyielding dissociations:

• Perceptual Fluency Manipulations: Manipulations that selectively enhance the speed or ease of perceptual processing (such as masked perceptual priming immediately prior to a test probe) systematically increase Know responses and elevate the continuous familiarity parameter $d’$, while leaving Remember judgments and the recollection parameter $R$ entirely unchanged.
• Response Deadline Paradigms: Imposing ultra-rapid response deadlines (forcing participants to make recognition judgments within 300 to 500 milliseconds) completely eliminates the recollection parameter $R$, collapsing the ROC curve into a fully symmetrical, zero-intercept function, while leaving familiarity discriminability ($d’$) intact. Because recollection requires time-intensive search and feature-binding operations, it drops out under tight temporal constraints, an outcome that cannot be explained by criterion shifts on a singular strength continuum.
• Deconstructing Nested Confidence: In response to Wixted’s observation of continuous confidence within Remember judgments, Yonelinas demonstrated that this variance does not reflect varying levels of trace strength on an old/new decision continuum. Instead, it reflects the *quantity and diversity of contextual features* retrieved (source detail, associative richness, emotional memory). When participants are recollecting an event, they can recall a single isolated detail (yielding lower recollective confidence) or a vivid flood of multisensory context (yielding higher recollective confidence), but the underlying process of retrieval remains qualitatively categorical context reinstatement.

6.3 Alternative Formalizations: Continuous Dual-Process Models

The intense debate between discrete dual-process models and single-process continuous models catalyzed the development of alternative mathematical formulations that sought to refine or synthesize these competing architectures. The most significant among these was the **Continuous Dual-Process Model**, advanced by researchers such as Slotnick and Dodson (2005). These theorists accepted the fundamental assertion that recollection and familiarity are distinct cognitive and neural processes, but challenged Yonelinas’s assumption that recollection operates as a high-threshold, all-or-none event. Instead, they formalized recollection as a continuous variable possessing its own normal distribution, suggesting that both recollection and familiarity operate as continuous signal detection processes with differing distributional properties.

Concurrently, cognitive psychologists turned to **State-Trace Analysis**, a non-parametric mathematical methodology designed to determine the dimensionality of the underlying cognitive latent variables without making arbitrary assumptions regarding distribution shapes or linear scales. State-trace analyses conducted across diverse recognition datasets systematically rejected unidimensional models, confirming that recognition memory performance requires at least two distinct latent dimensions to account for the empirical data.

Furthermore, researchers successfully mapped Remember/Know and ROC data into **Multinomial Processing Tree (MPT)** models. MPT architectures formalize memory tasks as branching trees of discrete, probabilistic cognitive states. These models confirmed that architectures incorporating distinct, independent branches for recollection and familiarity provide superior descriptions of categorization behavior across varied experimental contexts. Today, the broad consensus in experimental psychology affirms the dual-process architecture, viewing Yonelinas’s DPSD framework as an exceptionally robust approximation of human memory psychophysics.

7. Neuroanatomical Architectures Underlying Recollection and Familiarity

7.1 Medial Temporal Lobe Functional Specialization

The behavioral and mathematical validity of the Dual-Process Signal Detection model received its most decisive confirmation from cognitive neuroscience. Decades of structural neuroimaging, functional magnetic resonance imaging (fMRI), and clinical lesion studies have converged to demonstrate that recollection and familiarity are supported by anatomically dissociable structures within the Medial Temporal Lobe (MTL).

The Medial Temporal Lobe is not a monolithic memory engine; rather, it is a highly organized hierarchical network exhibiting clear functional specialization:

• The Hippocampus: At the apex of the MTL hierarchy, the hippocampus (comprising the dentate gyrus, CA3, CA1, and subicular subfields) is selectively dedicated to **Recollection**. Computational neural network models demonstrate that the sparse coding properties of the dentate gyrus facilitate *pattern separation*, preventing interference between similar episodes, while the recurrent collateral fibers of the CA3 field support *pattern completion*. These unique biological mechanisms enable the hippocampus to perform arbitrary relational binding—linking disparately processed neocortical representations of sights, sounds, spatial coordinates, and internal thoughts into a single, unified episodic engram. When a partial retrieval cue is presented, the hippocampus reconstructs the complete engram in a threshold-like, qualitative ecphoric event.
• The Perirhinal Cortex: Situated along the parahippocampal gyrus of the ventral temporal lobe, the perirhinal cortex is primarily dedicated to **Familiarity**. The perirhinal cortex receives direct, rich sensory inputs from polymodal neocortical visual and auditory processing streams. It processes item-level representations and decontextualized object features. When an item is repeatedly encountered, perirhinal neurons display a phenomenon known as *repetition suppression*—a reduction in neuronal firing rates that signals item priming and perceived familiarity. The perirhinal cortex computes this continuous familiarity signal rapidly and autonomously, without engaging hippocampal relational binding.
• The Parahippocampal Cortex: Located posterior to the perirhinal cortex, the parahippocampal cortex is functionally dedicated to processing spatial, environmental, and scene context. It projects contextual information directly to the hippocampus, where it is bound together with the item representations transmitted from the perirhinal cortex.

7.2 Prefrontal and Parietal Cortex Contributions

While the medial temporal lobe serves as the computational hub for engram storage and initial ecphory, the conscious deployment of recollection and familiarity depends upon extensive distributed networks spanning the prefrontal and posterior parietal cortices.

Within the **Prefrontal Cortex (PFC)**, functional imaging reveals a division of labor between strategic retrieval control and post-retrieval monitoring:

• Ventrolateral Prefrontal Cortex (VLPFC): Plays an indispensable role during the encoding and retrieval phases by orchestrating the strategic, top-down selection of goal-relevant item features, directly modulating perirhinal and hippocampal representations.
• Dorsolateral Prefrontal Cortex (DLPFC): Heavily engaged during post-retrieval monitoring and evaluation. When threshold recollection occurs, the DLPFC operates to verify the veracity, coherence, and source validity of the retrieved episodic fragments, maintaining recollected context within working memory for active decision-making.

Simultaneously, the **Posterior Parietal Cortex (PPC)** has emerged as a crucial node in the episodic retrieval network, formalized under the Dual-Attentional Framework:

• Dorsal Posterior Parietal Cortex (DPPC): Encompassing the superior parietal lobule and intraparietal sulcus; mediates top-down, goal-directed allocation of attention during effortful memory search, exhibiting heightened engagement when familiarity signals are weak or ambiguous.
• Ventral Posterior Parietal Cortex (VPPC): Encompassing the angular gyrus and supramarginal gyrus; functions as an *episodic buffer* or “attention-to-memory” system. The angular gyrus shows selective, profound activation during threshold recollection. When the hippocampus successfully completes an engram, the reconstructed contextual details are projected to the angular gyrus, which holds the rich, multi-sensory episodic representation within conscious awareness, creating the subjective phenomenological experience of autonoetic mental time travel.

7.3 Neuropsychological Lesion Evidence

The definitive neuroanatomical proof for the DPSD model comes from neuropsychological patients suffering from focal brain lesions. If recollection and familiarity were merely points along a single memory strength continuum supported by a unitary neural system, focal damage to that system should cause a proportional, parallel decline in both parameters.

Strikingly, neuropsychological investigations have revealed profound double dissociations:

• Selective Hippocampal Lesions: Patients suffering from selective bilateral hypoxic brain injury, which damages the CA1 pyramidal cell layers of the hippocampus while leaving surrounding perirhinal and parahippocampal cortices intact, display an isolated collapse of recollection. Extensive evaluations of hypoxic amnesic patients (e.g., patient Jon, and cohorts investigated by Yonelinas, Kroll, Dobbins, et al.) demonstrate that their recognition ROC curves lose their characteristic asymmetry and non-zero y-intercepts, becoming entirely symmetrical. Quantitative parameter estimation reveals that their recollection parameter $R$ drops to zero or near-zero levels, whereas their familiarity discriminability index ($d’$) remains identical to age-matched healthy controls. These patients can accurately identify which items were on a study list based on pure familiarity, but they cannot retrieve any associative context, source information, or temporal order.
• Selective Perirhinal Lesions: Conversely, patients with focal damage encompassing the anterior temporal lobe and perirhinal cortex (often secondary to localized herpes simplex encephalitis or targeted surgical resections) display the inverse deficit: severely impaired familiarity discriminability with preserved threshold recollection for associative, relational engrams bound by intact hippocampi.
• Animal Neurotoxic Models: Targeted neurotoxic lesion studies in rodents and non-human primates confirm these findings. Bilateral excitotoxic lesions confined strictly to the rodent hippocampus selectively disrupt associative context retrieval and non-zero ROC intercepts, whereas neurotoxic perirhinal lesions selectively eliminate the continuous discriminability metric $d’$, providing indisputable cross-species validation of the dual-process architecture.

8. Electrophysiological and Neuroimaging Correlates

8.1 Event-Related Potential (ERP) Old/New Effects

Human electrophysiology offers exceptional millisecond-level temporal resolution, providing a precise temporal window into the cognitive architecture of recognition. In scalp-recorded Event-Related Potentials (ERPs), the presentation of recognized studied items elicits more positive-going waveforms than unstudied items—a phenomenon known as the “ERP Old/New Effect.” Decades of high-density EEG research have demonstrated that this old/new effect is bifurcated into two distinct temporal and topographic components that map directly onto Familiarity and Recollection.

These two electrophysiological signatures are:

• The Early Mid-Frontal Old/New Effect (The FN400): Occurring between 300 and 500 milliseconds post-stimulus onset, this effect is characterized by a positive-going voltage deflection over frontal and central scalp electrodes. The FN400 correlates systematically with familiarity: its amplitude varies continuously as a graded function of familiarity strength, confidence, and perceptual fluency. The FN400 is elicited equally by items accompanied by successful source retrieval and items recognized in the total absence of context, confirming its role as an electrophysiological index of context-free noetic familiarity.
• The Late Positive Complex (LPC) / Parietal Old/New Effect: Occurring between 500 and 800 milliseconds post-stimulus onset, this component manifests as a prominent, sustained positive deflection centered over posterior parietal recording sites, maximal over the left parietal scalp. The LPC corresponds strictly to threshold recollection: it is elicited exclusively when an item is accompanied by successful retrieval of episodic context, source details, or a “Remember” judgment. The LPC exhibits a threshold-like, all-or-none activation profile, remaining absent for items recognized solely on the basis of high familiarity.

Although some researchers have debated whether the FN400 might represent conceptual priming rather than declarative familiarity, comprehensive studies controlling for semantic processing have confirmed that familiarity-based recognition reliably modulates the early mid-frontal waveform independently of conceptual priming, providing robust temporal validation for the two-stage DPSD model.

8.2 Functional Magnetic Resonance Imaging (fMRI) Discoveries

Functional neuroimaging has allowed researchers to map the mathematical parameters of the DPSD model directly onto voxel-level changes in Blood-Oxygen-Level-Dependent (BOLD) signals. By scanning participants during the retrieval phase of Remember/Know and confidence ROC paradigms, neuroscientists have uncovered distinct activation kinetics across medial temporal subregions.

These fMRI studies reveal clear computational dissociations:

• Perirhinal BOLD Responses: The perirhinal cortex exhibits a continuous, linear relationship with memory strength. As an item’s perceived familiarity increases (across graded confidence levels from “Sure New” to “Sure Old”), perirhinal BOLD signals display a monotonic, linear decrease in activity (repetition suppression). The perirhinal cortex tracks the familiarity parameter $d’$ continuously, firing vigorously to novel items and progressively scaling down its response as familiarity strength rises.
• Hippocampal BOLD Responses: The hippocampus displays a non-linear, step-function activation profile. Hippocampal BOLD activity remains at basal levels across all levels of familiarity confidence (from low-confidence familiar up to high-confidence familiar), but shows a sudden, dramatic spike in activation exclusively for items that receive a “Remember” judgment or result in successful source recollection. The hippocampus behaves as a threshold detector, igniting only when the ecphoric threshold is crossed and relational context is retrieved.

Furthermore, advances in **Multivoxel Pattern Analysis (MVPA)** have revealed that during threshold recollection, the pattern of BOLD activations across sensory neocortical regions mirrors the neural pattern recorded during the initial perceptual encoding of the event—a neural reinstatement effect that is completely absent during familiarity-based decisions.

8.3 Intracranial Recordings and Neural Oscillations

The most temporally and anatomically precise evidence for dual-process mechanics stems from intracranial electroencephalography (iEEG) and local field potential (LFP) recordings conducted in neurosurgical patients undergoing invasive monitoring for medically intractable epilepsy. Electrodes implanted directly within the human hippocampus and rhinal cortices reveal striking oscillatory dissociations between recollection and familiarity.

Intracranial recordings reveal that Familiarity is marked by rapid, transient **gamma-band synchrony (30–80 Hz)** in the perirhinal cortex, peaking as early as 200–300 milliseconds post-stimulus onset. This rapid gamma burst reflects local, feed-forward processing of item-specific perceptual representations. In contrast, Recollection is characterized by sustained **theta-band oscillations (4–8 Hz)** within the hippocampus, beginning around 400 milliseconds and extending past 800 milliseconds. Crucially, during successful contextual recollection, the human hippocampus orchestrates **theta-gamma phase-amplitude coupling**, in which the phase of slow hippocampal theta oscillations modulates the amplitude of high-frequency cortical gamma activity across neocortical sites, physically coordinating the multi-regional reinstatement of distributed episodic memories.

9. Experimental Paradigms and Manipulations Dissociating R and K

9.1 Encoding Manipulations and Cognitive Load

The empirical viability of the Gardiner-Yonelinas dual-process framework rests heavily upon the discovery of experimental manipulations that reliably dissociate the parameters $R$ and $d’$. The most well-established dissociations emerge from encoding-stage interventions that manipulate cognitive load and depth of processing.

Consider the following critical encoding manipulations:

• Levels-of-Processing (LoP): When participants are presented with words and instructed to perform either a shallow structural task (e.g., “Does the word contain the letter ‘e’?”) or a deep semantic task (e.g., “Is the object concrete or abstract?”), the deep encoding manipulation selectively boosts the recollection parameter $R$ by several hundred percent, while exerting minimal influence on the familiarity parameter $d’$. Elaborative semantic processing provides rich conceptual hooks that facilitate hippocampal binding, whereas familiarity operates on basic perceptual and conceptual trace enhancements that occur even during shallow processing.
• Divided Attention: Requiring participants to perform a concurrent, resource-demanding auditory task (such as continuous random number generation or tone tracking) during the study phase devastates the recollection parameter $R$, driving it toward zero. In dramatic contrast, familiarity discriminability ($d’$) remains virtually unscathed. Conscious episodic binding requires the full deployment of central executive resources, whereas familiarity signals are computed via automatic neocortical plasticity that operates outside focal attention.
• Self-Referential Processing: Encoding items in relation to one’s own self-concept (e.g., “Does this trait adjective describe you?”) induces an exceptional, selective increase in the $R$ parameter. The self-concept serves as a powerful, hyper-elaborated associative hub, maximizing the subsequent autonoetic retrieval of personal encoding context.

9.2 Retrieval Dynamics and Temporal Constraints

The temporal dynamics of memory ecphory provide another rich domain for functional dissociations. The Dual-Process Signal Detection model asserts that because familiarity is computed through local neocortical circuits, it operates substantially faster than recollection, which requires recurrent hippocampal circuitry and distributed neocortical reinstatement.

Experimental manipulations targeting retrieval timing confirm this prediction:

• Response Deadlines: In speeded response paradigms, participants are forced to render their recognition judgments within narrow temporal windows (e.g., 300 ms, 500 ms, 800 ms, 1200 ms). At extremely short deadlines (300–500 ms), the recollection parameter $R$ is mathematically zero, and the empirical ROC is completely symmetrical ($s = 1.0$). Only when the response window is extended beyond 600–800 milliseconds does the ROC curve begin to exhibit asymmetry, with the parameter $R$ rising monotonically as more time is granted for hippocampal pattern completion. Familiarity, conversely, reaches its asymptotic $d’$ discriminability within 400 milliseconds.
• Speed-Accuracy Trade-Off (SAT) Functions: Comprehensive SAT studies demonstrate that the retrieval function for recognition memory is biphasic. The earliest available mnemonic information emerges at an early temporal intercept and rises along an exponential curve reflecting familiarity. At a distinctly later temporal intercept, a second retrieval component initiates, corresponding to the arrival of recollected episodic information.
• Environmental Context Shifts: Altering the physical environment between study and test (e.g., changing room odors, background music, or physical architecture) severely impairs recollection ($R$) by disrupting environmental cue reinstatement, while leaving the item-familiarity parameter ($d’$) invariant.

9.3 Perceptual Fluency and Perceptual Manipulations

While encoding manipulations often selectively alter recollection, perceptual manipulations provide the mirror-image dissociation, selectively altering familiarity while leaving recollection unaffected.

Key experimental demonstrations include:

• Masked Perceptual Priming: In this paradigm, a test probe is preceded immediately by a subliminal presentation of the identical word or an unrelated word, presented for a fraction of a second (e.g., 30–50 ms) and masked from conscious awareness. The subliminal presentation of the matching prime enhances the perceptual fluency of the subsequent test probe. This fluency enhancement causes participants to experience a sudden surge in felt familiarity, selectively inflating the proportion of “Know” responses and shifting the familiarity parameter $d’$. Crucially, masked priming has zero effect on the recollection parameter $R$, proving that familiarity can be causally induced through pure fluency manipulation without engaging episodic context.
• Perceptual Clarity and Duration: Modulating the visual clarity (contrast, font blur) or the presentation duration of test probes selectively alters the speed and magnitude of the familiarity signal, leaving threshold recollection unaffected.
• Modality Crossing: Studying items visually and testing them auditorily (or vice versa) selectively degrades perceptual familiarity, depressing $d’$ while leaving the semantic recollection of episodic details intact. These continuous demonstrations of double dissociations provide overwhelming empirical evidence against single-process models.

10. Life-Span Developmental Changes and Clinical Manifestations

10.1 Normal Aging: Asymmetric Trajectories of Memory Sub-Processes

The application of the Dual-Process Signal Detection model to cognitive aging has resolved long-standing controversies regarding the nature of age-related memory decline. Healthy older adults routinely report that their memory feels less vivid, and standard neuropsychological tests document pronounced impairments in episodic recall. However, older adults often perform surprisingly well on basic recognition tests.

The DPSD framework accounts for this discrepancy through the **Associative Deficit Hypothesis**. When older adults are evaluated using Remember/Know paradigms and confidence-based ROCs, researchers observe a profound, asymmetric trajectory:

• Marked Decline in Recollection: The recollection parameter $R$ exhibits steep, progressive declines across the adult lifespan. Older adults experience severe difficulties binding disparate episodic elements (such as an item and its source, or a face and a name) into cohesive engrams, resulting in a dramatic loss of autonoetic recollection.
• Relative Preservation of Familiarity: In sharp contrast, the familiarity parameter $d’$ remains remarkably intact in healthy aging. Older adults retain the capacity to gauge item familiarity with high sensitivity, allowing them to perform effectively on standard old/new recognition tests where familiarity signals can be leveraged.

This cognitive asymmetry directly mirrors structural and functional brain aging. Normal aging is characterized by progressive volume loss in the CA1 field and dentate gyrus of the hippocampus, alongside structural degeneration of the fornix and the fronto-temporal white matter tracts that mediate strategic episodic retrieval. Conversely, the perirhinal cortex exhibits substantially less structural atrophy in normal aging, maintaining its capacity to compute intact familiarity signals. Functional neuroimaging demonstrates that high-performing older adults frequently exhibit compensatory prefrontal cortex hyper-recruitment to maximize the utility of their spared familiarity-based judgments.

10.2 Pediatric Development and the Emergence of Recollection

The developmental trajectory of recognition memory during childhood provides the chronological mirror image of aging. Longitudinal and cross-sectional developmental studies demonstrate that recollection and familiarity follow distinct ontogenetic timelines.

The developmental milestones unfold as follows:

• Early Emergence of Familiarity: Familiarity-based recognition is present very early in ontogeny, functioning robustly in infants and young children. Toddlers readily discriminate between novel and familiar objects, a capacity supported by early-maturing neocortical and perirhinal structures.
• Protracted Development of Recollection: In contrast, autonoetic recollection exhibits a protracted developmental trajectory, continuing to mature throughout middle childhood and adolescence. Children aged 4 to 6 display low $R$ parameters, frequently recognizing items without the capacity to retrieve source context or episodic associations. The parameter $R$ rises significantly between ages 7 and 12, reaching adult levels only in late adolescence.

This prolonged maturation maps directly onto the delayed neurodevelopment of the human brain: while rhinal structures mature early, the structural connectivity between the hippocampus and the prefrontal cortex, alongside the microstructural refinement of the CA fields and dentate gyrus, undergoes continuous synaptic pruning and myelination well into adolescence. Furthermore, the development of metamemory—the meta-cognitive capacity to monitor, calibrate, and accurately report internal states of conscious recollection—requires mature prefrontal executive networks that continue developing through young adulthood.

10.3 Clinical Pathologies and Neuropsychiatric Conditions

The DPSD framework provides profound clinical utility as a diagnostic and translational tool across neuropsychiatric and neurodegenerative pathologies.

The dual-process parameters serve as sensitive biomarkers across multiple clinical conditions:

• Mild Cognitive Impairment (MCI) and Alzheimer’s Disease (AD): While normal aging selectively impairs recollection, Alzheimer’s disease pathology targets the medial temporal lobe in a distinct pathological sequence. Neurofibrillary tau tangles originate in the transentorhinal and perirhinal cortices before spreading into the hippocampus. Consequently, patients with Amnestic MCI and early AD exhibit an early, catastrophic collapse of *both* recollection ($R$) and familiarity ($d’$). The premature degradation of the familiarity parameter $d’$ serves as a powerful neuropsychological biomarker distinguishing early AD pathology from normal age-related cognitive decline.
• Temporal Lobe Epilepsy (TLE): Patients suffering from pharmacoresistant TLE characterized by unilateral hippocampal sclerosis exhibit asymmetric recollection impairments. Left hippocampal sclerosis produces severe, selective deficits in verbal recollection ($R$), leaving verbal familiarity intact; right hippocampal sclerosis selectively decimates spatial and non-verbal recollection, confirming material-specific functional lateralization within the human episodic system.
• Schizophrenia: Schizophrenic pathology is characterized by pronounced, aberrant familiarity attributions and profound relational binding deficits. Patients with schizophrenia frequently misattribute strong familiarity signals to novel items or unstudied lures, generating elevated false alarm rates and delusional interpretations of past experiences. DPSD analyses reveal that schizophrenia involves a structural failure of hippocampal pattern separation, leading to catastrophic interference and severely reduced $R$ parameters.
• Post-Traumatic Stress Disorder (PTSD): Individuals with PTSD exhibit trauma-related alterations in dual-process mechanics: they display heightened, intrusive, and dysregulated recollection for traumatic cues, alongside generalized, over-inclusive familiarity that causes neutral everyday stimuli to trigger profound feelings of past trauma.

11. Practical Methodological Guide to Implementing DPSD and R/K

11.1 Experimental Design Protocols and Subject Instructions

To successfully implement the Remember/Know paradigm and obtain uncontaminated DPSD parameters, experimental protocols must be engineered with psychometric precision. The primary danger remains participant misunderstanding: without clear, standardized instructions, participants routinely treat “Remember” simply as high confidence and “Know” as low confidence.

Researchers should adhere to the following best practices:

• Explicit, Structured Instructions: Provide detailed, written and verbal instructional scripts (modeled on Gardiner’s protocols). Define “Remember” as conscious mental reinstatement of specific details from the encoding phase (thoughts, feelings, perceptual features, adjacent words). Define “Know” as a state of certain knowledge that the item was present, accompanied by an absolute absence of contextual detail. Define “Guess” as uncertainty or chance responding.
• Interactive Practice Sessions: Always require participants to complete supervised practice trials prior to the experimental test phase. When a participant renders a “Remember” judgment during practice, the experimenter must ask: “What specific detail do you remember about when this word appeared?” If the participant responds, “I’m just really sure it was on the list,” the experimenter must correct them, clarifying that such an experience must be classified as “Know.”
• Balanced Stimulus Lists: Construct target and lure lists matched rigorously on linguistic variables: word frequency, syllable count, concreteness, imageability, and emotional valence. Avoid using lures that possess idiosyncratic associations with studied items, which can induce aberrant false recollection.
• Mitigating Response Bias: Counterbalance lists across conditions, randomize stimulus presentation orders, and interleave filler trials to stabilize internal decision criteria throughout lengthy testing sessions.

11.2 Computational Estimation and Parameter Fitting Software

Once data are collected, researchers must apply appropriate mathematical solvers to extract the latent parameters. For confidence-based ROC data, non-linear least squares estimation or Maximum Likelihood Estimation (MLE) should be deployed.

The computational workflow involves the following mathematical procedures:

• DPSD Formulation in Scripting Environments: Using statistical computing environments such as R or Python, researchers define the objective function based on the DPSD hit and false alarm equations. In R, non-linear optimization functions like nls, optim, or specialized packages (such as mrecem or the DPSD toolbox) optimize $R$, $d’$, and criteria thresholds ($c_i$).
• The Independence Remember/Know (IRK) Correction: When analyzing tripartite Remember/Know/Guess or binary R/K paradigms, researchers must avoid using raw Know proportions. Apply the Independence correction formulas:
  $$R = P(\text{Remember}_{\text{Target}}) – P(\text{Remember}_{\text{Lure}})$$
  $$F_{\text{Target}} = \frac{P(\text{Know}_{\text{Target}})}{1 – P(\text{Remember}_{\text{Target}})}$$
  $$F_{\text{Lure}} = \frac{P(\text{Know}_{\text{Lure}})}{1 – P(\text{Remember}_{\text{Lure}})}$$
  $$d’ = \Phi^{-1}(F_{\text{Target}}) – \Phi^{-1}(F_{\text{Lure}})$$
  This algebraic transformation guarantees that the derived familiarity metric reflects true signal detection discriminability uncontaminated by the opportunity structure of the task.
• Handling Zero-Cell Frequencies: When participants exhibit extreme hit rates (1.0) or false alarm rates (0.0), standard signal detection z-transforms become undefined ($\pm\infty$). Researchers must apply the standard Snodgrass and Corwin (1988) correction: adding 0.5 to all raw response frequencies and dividing by $N + 1$, where $N$ is the number of target or lure trials.

11.3 Common Methodological Pitfalls and Best Practices

Researchers embarking on dual-process research must navigate several common technical and analytical pitfalls that can compromise experimental validity:

• Collinearity and Parameter Instability: If confidence ROC datasets lack sufficient points in the conservative or liberal regions, non-linear optimization algorithms can suffer from parameter trade-offs, where increases in $R$ are artificially offset by decreases in $d’$. To prevent collinearity, employ at least a 6-point confidence rating scale (generating 5 independent ROC points), ensuring that participants are encouraged to distribute their ratings across the full breadth of the scale.
• Inadequate Trial Counts: Fitting non-linear dual-process models requires adequate statistical power. Using fewer than 40–50 target trials and 40–50 lure trials per experimental condition leads to extreme parameter instability and uninterpretable error variance. High-powered psychophysical designs should target 80 to 120 trials per condition.
• Failure of Post-Experiment Verification: At the conclusion of testing, administer a brief debriefing questionnaire asking participants to describe how they distinguished between Remember and Know judgments. Discard or separately analyze data from participants who explicitly report using “Remember” merely to signal high confidence.
• Reporting Standards: Always report raw Hit and False Alarm rates for Remember, Know, and Guess categories alongside derived parameters ($R$, $d’$, $c$). Report the goodness-of-fit statistics (e.g., chi-square, AIC, BIC) for model fits, providing full transparency regarding the empirical stability of the mathematical solutions.

12. Synthesis, Contemporary Developments, and Future Directions

12.1 Integration with Modern Computational and Neurocomputational Models

As cognitive science advances into the twenty-first century, the Dual-Process Signal Detection model continues to evolve, integrating deeply with sophisticated computational and neurocomputational frameworks.

Prominent contemporary intersections include:

• Complementary Learning Systems (CLS) Theory: The DPSD model provides the empirical cornerstone for the CLS framework developed by McClelland, McNaughton, and O’Reilly. CLS posits that the brain resolves the “stability-plasticity dilemma” by utilizing two specialized learning systems: the neocortex, which acts as a slow, continuous learner extracting statistical regularities (providing the biological substrates for continuous familiarity $d’$), and the hippocampus, which operates as a rapid, sparse learner performing immediate episodic pattern separation and completion (yielding threshold recollection $R$). The mathematical parameters of the DPSD model correspond directly to the functional outputs of these two computational systems.
• Sequential Sampling and Drift-Diffusion Models (DDM): Recent theoretical advances have bridged static signal detection models with dynamic evidence-accumulation architectures. In these hybrid models, familiarity and recollection are modeled as two distinct drift rates operating across time: a fast, early drift rate driven by sensory familiarity, followed by a delayed, high-gain drift rate initiated when hippocampal pattern completion succeeds. This synthesis explains not only choice probabilities and ROC shapes, but also the complete reaction-time distributions of Remember and Know judgments.
• Deep Neural Networks and Artificial Intelligence: Contemporary machine learning architectures designed for episodic memory simulations increasingly incorporate dual-memory buffers: dense, distributed associative networks that compute continuous vector similarity (familiarity), paired with explicit, non-parametric episodic memory buffers that store and retrieve uncompressed experiential vectors (recollection).

12.2 Emerging Frontiers: Beyond Verbal Episodic Memory

While the Remember/Know and DPSD paradigms were originally formulated using verbal stimuli, modern researchers have expanded these models into entirely new empirical domains:

Expanding applications include:

• Immersive Virtual Reality (VR) and Spatial Navigation: Utilizing advanced VR environments, cognitive scientists now test dual-process mechanics within rich, three-dimensional, naturalistic environments. These studies demonstrate that spatial context reinstatement relies strictly upon hippocampal recollection mechanisms, whereas object recognition within dynamic environments can be maintained through neocortical familiarity networks alone.
• Eyewitness Identification and Legal Testimony: The application of DPSD to legal psychology has fundamentally transformed how courts evaluate eyewitness reliability. In eyewitness lineups, high-confidence identifications driven by immediate, threshold recollection exhibit near-perfect accuracy, whereas identifications driven by familiarity in the absence of recollection are prone to false identifications under cross-racial lineups or weapon focus. Measuring eyewitness decisions through signal detection ROCs has overturned decades of flawed legal reliance on uncalibrated confidence statements.
• Cross-Species Episodic Memory: Innovative testing paradigms have demonstrated recollection-like and familiarity-like memory in non-human animals. Using olfactory and spatial ROC paradigms, researchers have shown that rodents display asymmetric, non-zero ROC curves that are selectively flattened by hippocampal lesions, demonstrating that autonoetic-like threshold retrieval is an evolutionarily ancient memory adaptation.
• Pharmacological Neuromodulation: Pharmacological challenges utilizing NMDA receptor antagonists, GABAergic modulators, and targeted deep brain stimulation of the fornix demonstrate that recollection and familiarity can be chemically and electrically titrated, paving the way for targeted neurotherapeutics.

12.3 The Gardiner and Yonelinas Legacy in Cognitive Science

The collective intellectual contributions of John M. Gardiner and Andrew P. Yonelinas represent an enduring milestone in the history of cognitive science. Prior to their work, the scientific study of human memory was fractured: one camp operated in the realm of qualitative, introspective philosophy, while the other was confined to rigid, univariate psychophysical metrics that denied the reality of human conscious phenomenology.

John M. Gardiner took the visionary, philosophical insights of Endel Tulving and transformed them into a robust, systematically standardized experimental science. Through hundreds of meticulous behavioral studies, Gardiner proved that the subjective textures of human memory—autonoetic remembering and noetic knowing—could be reliably isolated, experimentally manipulated, and validated through uncompromising scientific rigor.

Andrew P. Yonelinas achieved the grand synthesis: uniting Gardiner’s phenomenological paradigm with the mathematical power of Signal Detection Theory. By formulating the Dual-Process Signal Detection model, Yonelinas gave the field an elegant, highly predictive mathematical architecture that mapped conscious experiences onto quantifiable parameters, resolved decades of psychophysical anomalies, and provided cognitive neuroscience with the precise empirical framework required to map the functional neuroanatomy of the human medial temporal lobe.

Together, Gardiner and Yonelinas redefined recognition memory. Their legacy is an enduring testament to the power of integrating phenomenology, mathematical modeling, and cognitive neuroscience—demonstrating that the deepest mysteries of human consciousness can be understood when empirical rigor is matched by theoretical imagination.

Conclusion

The scientific journey from early univariate trace-strength models to the sophisticated architecture of the Dual-Process Signal Detection (Remember/Know) paradigm reflects the maturation of cognitive psychology into a rigorous, integrative science. By demonstrating that recognition memory is not an undifferentiated, scalar decision variable, but rather the synergistic product of two functionally, mathematically, and neurobiologically dissociable mechanisms—threshold recollection and continuous familiarity—John M. Gardiner and Andrew P. Yonelinas reshaped our understanding of the human mind.

Through Gardiner’s behavioral standardization of autonoetic and noetic consciousness, cognitive science acquired the tools to investigate the qualitative experiences of memory retrieval. Through Yonelinas’s mathematical formalization of the DPSD model and the Independence Remember/Know procedure, psychophysicists obtained the quantitative machinery to isolate these mechanisms cleanly from response biases, confidence artifacts, and criterion shifts. The convergence of behavioral double dissociations, asymmetric ROC geometries, localized BOLD kinetics, distinct electrophysiological old/new effects, and focal neuropsychological lesion profiles provides one of the most thoroughly validated frameworks in modern cognitive neuroscience.

As memory science looks forward, the Gardiner-Yonelinas dual-process framework remains as intellectually vibrant as ever. Its principles continue to guide the development of biologically realistic neurocomputational networks, illuminate the cognitive trajectories of aging and neuropsychiatric disease, refine legal standards for eyewitness identification, and guide philosophical debates on the evolutionary emergence of mental time travel. By successfully bridging the subjective experience of remembering with the quantitative mechanics of signal detection, the Remember/Know paradigm and the Dual-Process Signal Detection model stand as foundational pillars in the enduring scientific quest to decode the architecture of human memory.

References

• Donaldson, W. (1996). The role of decision processes in remembering and knowing. Memory & Cognition, 24(4), 523–533. https://doi.org/10.3758/BF03200940
• Eichenbaum, H., Yonelinas, A. P., & Ranganath, C. (2007). The medial temporal lobe and recognition memory. Annual Review of Neuroscience, 30, 123–152. https://doi.org/10.1146/annurev.neuro.30.051606.094328
• Gardiner, J. M. (1988). Functional aspects of recollective experience. Memory & Cognition, 16(4), 309–313. https://doi.org/10.3758/BF03197041
• Gardiner, J. M., & Java, R. I. (1991). Forgetting in recognition memory with and without recollective experience. Memory & Cognition, 19(6), 617–623. https://doi.org/10.3758/BF03199584
• Gardiner, J. M., & Parkin, A. J. (1990). Attention and recollective experience in recognition memory. Memory & Cognition, 18(6), 579–583. https://doi.org/10.3758/BF03197103
• Gardiner, J. M., Ramponi, C., & Richardson-Klavehn, A. (1998). Experiences of remembering, knowing, and guessing. Consciousness and Cognition, 7(1), 1–26. https://doi.org/10.1006/ccog.1997.0321
• Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. John Wiley & Sons. https://mitpress.mit.edu/9780262691109/signal-detection-theory-and-psychophysics/
• Mandler, G. (1980). Recognizing: The judgment of previous occurrence. Psychological Review, 87(3), 252–271. https://doi.org/10.1037/0033-295X.87.3.252
• McClelland, J. L., McNaughton, B. L., & O’Reilly, R. C. (1995). Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory. Psychological Review, 102(3), 419–457. https://doi.org/10.1037/0033-295X.102.3.419
• Rugg, M. D., & Curran, T. (2007). Event-related potentials and recognition memory. Trends in Cognitive Sciences, 11(6), 251–257. https://doi.org/10.1016/j.tics.2007.04.004
• Slotnick, S. D., & Dodson, C. S. (2005). Support for a continuous dual-process model of recognition memory and source memory. Memory & Cognition, 33(1), 151–170. https://doi.org/10.3758/BF03195305
• Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory: Applications to dementia and amnesia. Journal of Experimental Psychology: General, 117(1), 34–50. https://doi.org/10.1037/0096-3445.117.1.34
• Tulving, E. (1985). Memory and consciousness. Canadian Psychology/Psychologie Canadienne, 26(1), 1–12. https://doi.org/10.1037/h0080017
• Wixted, J. T. (2007). Dual-process theory and signal-detection theory of recognition memory. Psychological Review, 114(1), 152–176. https://doi.org/10.1037/0033-295X.114.1.152
• Wixted, J. T., & Stretch, V. (2004). In defense of the signal detection interpretation of remember/know judgments. Psychonomic Bulletin & Review, 11(4), 616–641. https://doi.org/10.3758/BF03196616
• Yonelinas, A. P. (1994). Receiver operating characteristics in recognition memory: Evidence for a dual-process model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(6), 1341–1354. https://doi.org/10.1037/0278-7393.20.6.1341
• Yonelinas, A. P. (2001). Components of episodic memory: The contribution of recollection and familiarity. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 356(1413), 1363–1374. https://doi.org/10.1098/rstb.2001.0939
• Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of 30 years of research. Journal of Memory and Language, 46(3), 441–517. https://doi.org/10.1006/jmla.2002.2864
• Yonelinas, A. P., Kroll, N. E., Dobbins, I., Lazzara, M., & Knight, R. T. (1998). Recollection and familiarity deficits in amnesia: Convergence of remember-know, process dissociation, and receiver operating characteristic approaches. Neuropsychology, 12(3), 323–339. https://doi.org/10.1037/0894-4105.12.3.323
• Yonelinas, A. P., & Parks, C. M. (2007). Receiver operating characteristics (ROCs) in recognition memory: A review of critical issues. Psychological Bulletin, 133(5), 800–832. https://doi.org/10.1037/0033-2909.133.5.800