Readiness to remember: predicting variability in episodic memory

Abstract

Learning and remembering are fundamental to our lives, so what causes us to forget? Answers often highlight preparatory processes that precede learning, as well as mnemonic processes during the act of encoding or retrieval. Importantly, evidence now indicates that preparatory processes that precede retrieval attempts also have powerful influences on memory success or failure. Here, we review recent work from neuroimaging, electroencephalography, pupillometry, and behavioral science to propose an integrative framework of retrieval-period dynamics that explains variance in remembering in the moment and across individuals as a function of interactions among preparatory attention, goal coding, and mnemonic processes. Extending this approach, we consider how a ‘readiness to remember’ (R2R) framework explains variance in high-level functions of memory and mnemonic disruptions in aging.

Preparing to remember

We use episodic memory to think about the past, live in the present, and plan for the future. Given that learning and remembering are fundamental to our experience as humans, scientists have long studied what causes us to forget. Extensive research from neuroscience and behavioral science has focused on the neural and psychological mechanisms at play before and during learning that have consequences for later remembering [1–11]. For example, the ability to sustain attention and selectively attend to, and code, features of learning events based on task goals are important predictors of subsequent memory. Indeed, one foundational principle in many domains of cognitive science [12–19] is that moment-to-moment fluctuations in engagement of frontoparietal brain networks of attention and cognitive control within individuals and differences across individuals are linked to variability in preparatory processes that impact goal-directed behavior.

More recently, within the context of episodic memory, emerging evidence (e.g., [20]) indicates that preparatory processes are at play not only at the time of learning, but also immediately before an attempt to retrieve a memory, and that variability in preparatory processes explains variability in whether and how we remember. Accumulating data indicate that attentional preparedness and control modulate episodic retrieval attempts, with both moment-to-moment and individual differences before remembering influencing behavioral and neural expressions of memory, a phenomenon termed ‘R2R’ (see Glossary) [21]. This is a relatively underexplored aspect of remembering, because most research on episodic retrieval has focused on mechanisms during the act of retrieval itself that predict memory, including retrieval cue processing, memory weakening and updating, and mnemonic interference [22–29].

Here, we describe an R2R framework (Figure 1, Key figure) hypothesizing that: (i) preparatory attention may partially predict remembering in the moment directly and as mediated by the strength of preparatory goal-state coding; and (ii) individuals may differ in retrieval performance, in part, because of differences in their ability to be attentionally prepared and to represent and use mnemonic goals. Complementing the rich literature on preparatory processes at the time of learning [1–11], the proposed R2R framework advances basic science understanding of how retrieval depends on attentional preparedness, goal states, and their interactions with each other and with mnemonic mechanisms. The R2R framework also connects with translational science explorations of forgetting and ways to improve remembering, and is relevant for understanding variability in remembering across putatively healthy older adults and in other populations. In addition to reviewing multimodal evidence on how variability in episodic remembering in the moment and across individuals may be explained by R2R factors that impact behavioral and neural mnemonic expressions, we also consider how these processes may impact other functions of memory, such as prospection and creative thinking.

Figure 1.

(A) Mechanistic pathways depict preparatory attention, goal coding, and memory interactions in the moment and across individuals. Time-ordered fluctuations in attentional preparedness may have direct effects on remembering, and may also have mediated effects by giving rise to fluctuations in goal-state representation. Trait-level differences in sustained attention, aging, and other factors may explain variability in goal-directed remembering directly and via preparatory attention and goal-state coding indirectly. In terms of in-the-moment remembering, R2R focuses on time-ordered effects from attention to goal coding to memory (note: post-goal attentional influences on memory, and memory influences on subsequent attention and goal coding, as shown with the respective arrows, are also undoubtedly important). (B) An example of this phenomenon that occurs in real life (e.g., being tested on material learned in an earlier school lecture) highlights why examining R2R factors has value. The thought bubbles illustrate what attentional state an individual may be in (e.g., off- and on-task thoughts), the art question illustrates the goal code, and the painting and lightbulbs serve as a retrieval cue and potential memory expressions, respectively. (C) Assays to operationalize R2R constructs in neuroscience and behavioral science are shown in respective colors (note: these mappings are carried across all figures). While we focus on the dorsal attention network (DAN) for preparatory attention, some work has shown that the default mode network (DMN) can also serve as an attentional predictor. Abbreviations: CCN, cognitive control network; ERP, event-related potential; FP, frontoparietal; gradCPT, gradual-onset continuous performance task; HIPP, hippocampus; LC, locus coeruleus; RT, reaction time.

Understanding preparatory attention and preparatory goal coding

To understand how preparatory processes modulate learning and memory, it is important to define what encompasses preparatory attention and preparatory goal coding in the vast landscape of cognitive operations.

Preparatory attention is a broad term that encompasses different mechanisms of the ability to focus before task performance, including dimensions of arousal, vigilance, selective monitoring, and motivation [30–47]. Understanding the temporal frequencies of forms of preparatory attention is informative for predicting variability in goal-directed task performance, particularly as related to slow rhythms of attention and inattention and quick micro-state lapses of attention (e.g., [38,43]). Treating preparatory attention in this general manner allows for understanding how psychological states of preparedness may impact performance within an individual within a trial or across N trials, as well as between individuals. At the same time, it will be critical for future work to specify the mechanistic form or forms of preparatory attention that are causal to behavior (e.g., arousal versus motivation, etc.).

Another modulatory readiness factor is preparatory goal-state coding, which is the ability to instantiate, maintain, and use representations of the structure, rules, cues, or objectives of a task [12,48–60]. Part of being able to use a goal is first to instantiate its representation when required. Bringing a goal online may depend on selective attention mechanisms that resolve interference from competing goals and reinstate the target representation (e.g., task-set reconfiguration) [12,15,17,50,52]. Once a goal is reinstated, another dimension of being able to use a goal is maintaining its representation over time; the representation and persistence of a goal are important for dynamically guiding the engagement of selective and sustained attention mechanisms [12,15,17,50,52]. Indeed, as a goal is processed and maintained, a third dimension is being able to apply the feature-to-categorization/response mappings of a goal, which guide attentional mechanisms to select over external and internal goal-relevant and goal-irrelevant features [12,15,17,50,52]; this dimension is thought to go awry in conditions such as goal neglect [12].

In terms of delineating between preparatory attention and preparatory goal coding, our view is that the representation of a goal can exist without engagement or application of subsequent selective attention; representing a retrieval goal can govern perceptual selection or action selection, but the goal representation, and variance in the strength of said representation, can exist independent of the implementation of these selection processes (see [12], but also [49]). Preparatory lapses may have direct effects on all three dimensions of goal coding, and one consequence of these interactions is that subsequent goal-directed task performance may be reduced. Understanding the relationships between preparatory attention and preparatory goal coding in the broad landscape of goal-directed behavior has implications for identifying the consequences of these processes additively and multiplicatively.

Given these constructs, it is also important to identify how preparatory attention and preparatory goal coding are measured and how they modulate multimodal signals of goal-directed cognitions in contexts outside episodic memory. Streams of evidence indicate that, both in isolation and when combined, neural, ocular, and behavioral measurements or assays of preparatory attention and goal coding produce downstream consequences for cognitive computations, explaining important within-individual variance (Box 1) and between-individual variance (Box 2) in goal-directed task performance.

Box 1. Attention and goal-state variability in the moment.

Insights into how and why preparatory processes contribute to some moments being better than others for goal-directed cognition, including memory, come from multimodal measurement of attentional state and goal coding within the individual.

There are multiple ways to assay preparatory attention signals in terms of arousal, vigilance, and selective monitoring (Figure I). One approach involves posterior alpha oscillatory power with EEG [33,38,44]. Increases in posterior alpha before engaging in a task are thought to reflect a loss of top-down attentional control or release from inhibition [33], and also mark periods of mind wandering in slow and fast rhythms, as assayed via thought probes [38,44]. Such increases in alpha relate to reaction time (RT) slowing and variability, and task errors [33,38,44] (see also work on selective monitoring [122]).

Fluctuations in pupil diameter provide another metric [36,37,40]. Decreases in tonic pupil diameter are thought to be due, in part, to hypoarousal in the locus coeruleus (LC), which projects noradrenaline throughout the brain [40]. Constrictions in pupil diameter pre-trial are often linked to lapses of internal attention in slow and fast rhythms [37], while increases in pupil diameter (hyperarousal in the LC) are often attributed to lapses of external attention [36]. Decreases and variability in pupil diameter relate to RT slowing and errors on multiple tasks [36,37,40]. Blink rate and fixation stability are additional ocular preparatory signals of attention [35].

Two other metrics appear to measure performance-related moment-to-moment attention. First, RT is often used as an index of ‘on’ or ‘off’ preparatory state. Research with the canonical gradual-onset continuous performance task (gradCPT) shows that both increased RT variability and windows of very fast and slow RTs mark being ‘out of the zone’ [14], and these state metrics predict subsequent periods of performance [14,30]. Second, univariate fMRI BOLD amplitude within the DAN fluctuates with lapses and relates to gradCPT behavior [14].

Similarly, there are multiple ways to measure preparatory goal-related operations (Figure II). Signals of goal coding strength with EEG before engaging in various tasks predict RT and accuracy metrics (e.g., [48]); these include left- (F3 and/or F7) and midfrontal (Fz) positive-peaking ERPs from –800 to –100 ms epochs. Multivariate decoding and pattern similarity analyses of fMRI data in the CCN, particularly the strength and content of distributed patterns in inferior frontal and intraparietal sulci [54], indicate an explanatory role for goal operations before many cognitive tasks [50,52,54,55]. The strength of goal coding, assayed via these patterns, can predict goal-directed behavior (e.g., [55]).

Emerging research has also examined variability in goal-directed behavior from preparatory interactions. EEG-fMRI data indicate that alpha fluctuations relate to variations in univariate CCN and DAN signals before various tasks, and these associations predict performance (e.g., [32]). EEG-fMRI data with pupillometry and thought-probe sampling show that periods of off-task cognitions can be predicted by changes in alpha, pupil, and frontoparietal BOLD [45]. While direct evidence for goal coding is lacking, the indirect evidence from neural assays that are thought to support goal coding points to potentially meaningful interactions with attention.

Figure I. Tracking fluctuations in preparatory attention within the individual.

In-the-moment tonic increases in posterior alpha power and decreases in pupil size are linked to hypoarousal and mind wandering, and predict accuracy reductions and reaction time (RT) slowing on goal-directed tasks (e.g., Sustained Attention to Response Task; SART with odd versus even numbers). Based conceptually on [36–38,40,43].

Figure II. Tracking in-the-moment goal coding within the individual.

Multivariate frontoparietal cognitive control network (CCN) patterns represent the strength and content of different task goals (e.g., shape versus color orienting on perceptual discrimination, as represented by separability in neural state space), with impacts on goal-directed performance (e.g., accuracy) at the trial level. Based conceptually on [54,55]. Abbreviations: IFS, inferior frontal sulcus; IPS, intraparietal sulcus.

Box 2. Attention and goal-state variability from individual to individual.

In addition to insights that can be drawn from examining attention and goal processing in goal-directed contexts within an individual, a focused subject-level approach [41,58] is necessary to understand why some individuals remember better than others. In behavioral science and neuroscience, different types of subject-level approaches are adopted to probe relationships between multimodal signals of cognitive constructs. These include relating subject-level variance in multiple constructs within the same task and relating multiple constructs in independent tasks and/or self-report surveys [34,89]. These individual-difference approaches are fruitful for characterizing trait relationships between preparatory lapsing and various goal-directed constructs. For example, individual differences in preparatory task-based sustained attention (arousal), as assayed by pupil diameter and pupil diameter variability, and behavioral expressions of lapsing are predictive of working memory capacity, attentional control, and mind wandering, both within and across tasks [34].

While individual-difference approaches typically involve identifying relationships among task- and survey-dependent assays, another powerful approach that has recently gained traction is leveraging neuroimaging to identify intrinsic person-specific functional brain connectomes; that is, systems-level circuits that vary from one individual to the next and relate to cognitive abilities [13,92] (Figure I). In terms of preparatory lapsing, individual variability in connectivity strength of the DAN with other brain networks during resting-state functional (BOLD) scanning has been related to person-to-person differences in the number and detail of off-task thoughts during separate working memory tasks [39]. Significant evidence indicates that wide-spread networks not reducible to DAN also predict differences in preparatory attention between [45], as well as within [45,111], individuals.

With respect to goal coding, less is known about how individual differences in the ability to represent and use goal states relate to other core cognitive abilities. Some initial insights stem from understanding how variable use of frontoparietal-dependent proactive (versus reactive) control from person to person relates to goal-directed performance within a task [51]. Proactive control refers to sustained/preparatory activation and use of goal representations that allocate selective attention, whereas reactive control refers to transient/stimulus-driven activation and use of goal representations that also allocate selective attention [51]. Trait differences in these computations affect goal-directed behavior during working memory, switching, and executive functioning tasks [51]. Related work indicates that individual differences in the ability to switch between rules (task relevant versus irrelevant) within a block of a cognitive control task is related to subject-level variability in working memory, response inhibition, and executive functioning [57] (see [56] for related evidence). In addition, the systems-level brain connectome approach has revealed that individual differences in the strength of resting-state and task-dependent functional connectivity between the CCN and the putative mnemonic default mode network (DMN) [123] relates to separate task-based assays of cognitive control [16].

Figure I. Individual differences in attention and goal-state neural variability.

Systems-level resting- and task-based interactions between the frontoparietal cognitive control network (CCN) and dorsal attention network (DAN), and the default mode network (DMN), predict individual differences in neurocognitive abilities, including sustained attention and the cognitive control of on- and off-task thoughts. Based conceptually on [13,39,45,92].

Readiness to learn

Efforts to understand how preparatory processes drive variability in episodic memory are informed by an extensive literature on the influences of preparatory attention and preparatory goal-state coding at the time of episodic encoding, or the act of acquiring knowledge of events that will be expressed later.

Research on attention, goal, and encoding mechanisms has a rich history in the cognitive neuroscience of human memory. One longstanding line of work demonstrates that inducing blocks of divided (versus full) attention at encoding can alter behavioral and neural signals of subsequent memory, particularly (although not exclusively) recollection-based memory and a canonical parietal old/new signal from event-related potentials (ERPs) [2,8]. Divided attention alters subsequent memory, in part, via disruption of preparatory mechanisms of arousal, vigilance, and selective monitoring [2,8] (see [61] for related evidence). Another line of work has characterized how the representation of different task goals before an event is encoded can impact selective attention to different aspects or features of the event, thus impacting memory strength, content, hippocampal mechanisms, and goal-directed behavior at the time of learning, as well as later retrieval accuracy and content [8]. Factors such as emotion, stress, and curiosity (e.g., [62]) can also impact preparatory attention (including motivation), goal coding, and encoding-side mnemonic processing. Indeed, a large number of neural studies have indicated that differences in ocular, oscillatory, ERP, and blood oxygen level-dependent (BOLD) sources of neural activity in the moment preceding an encoding event relate to subsequent memory, phenomena known as prestimulus subsequent memory effects (e.g., [1,3,10,11]). Work on slow and fast temporal rhythms of ‘on’ and ‘off’ attentional states [63], and on trait-level differences in learning across individuals under various task contexts [9,64], has also gained momentum.

Extant data indicate that preparatory attention and goal coding, which depend, in part, on distinct frontoparietal networks [4,65], influence how subsequently experienced sensory signals are transformed into cortical representations of event content. In addition, frontoparietal-dependent preparatory attention and goal coding influence how cortical representations of event features propagate to, and are conjunctively bound in, the hippocampus. During successful memory encoding, it is thought that the cognitive control network (CCN) codes the current goal state and the dorsal attention network (DAN) mediates selective attention to, and computes over, relevant sensory signals (e.g., [4]), which are projected to, and encoded by, the hippocampus via pattern separation and conjunctive encoding [5,6].

Building on this work, the field is now well positioned to examine how tonic signals of attention in the moment before a goal cue and event are encoded interact and relate to behavioral and neural expressions of goal coding, learning, and remembering. For example, representations of goal states in the CCN at learning may themselves be influenced by attentional preparedness. Pre-goal moment-to-moment attention lapses may relate to a concomitant weakening of goal representations in the CCN, which may result in the DAN failing to select goal-relevant sensory representations for mnemonic encoding. Consequently, lapses of attention and goal state may affect prestimulus attention and the precision of cortical representations of event content, as well as pattern separation and conjunctive encoding in the hippocampus. Lapsing at the onset of a learning episode, before the presentation of a goal cue, is predicted to explain significant variance in subsequent goal coding strength and memory, the latter of which should include difference-due-to-memory ERP and fMRI subsequent memory signals of learning. Similarly, lapses in the maintenace of a goal code before a stimulus is encountered should hinder encoding of goal-relevant event features, negatively predicting subsequent memory.

Preparatory attention, goals, and episodic retrieval

At retrieval, the R2R framework posits that preparatory frontoparietal processes that occur at encoding also impact episodic remembering. From this perspective, domain-general preparatory attention and preparatory goal coding mechanisms contribute to why some moments are better for memory than others, and why some individuals remember better than others.

Before successful retrieval, the CCN is thought to code goal states that, in part, guide DAN selective attention to, and processing of, subsequent retrieval cues [24–26]. In this manner, preparatory goal coding may facilitate how retrieval cues elicit sum similarity signals in medial temporal lobe (MTL) cortex that subserve item familiarity and pattern completion in the hippocampus that subserves reinstatement of accompanying event features. Goal states at retrieval also establish the evidence (or decision category)-to-action mappings and, thus, have a role in mnemonic readout/evidence integration and action selection [66]. While goal states are thought to interact with arousal, vigilance, and selective attentional allocation, the strength of a goal state in the CCN may, itself, depend upon attentional preparedness processes that impact goal-state instantiation and maintenance. That is, fast and slow temporal fluctuations in attentional preparedness may affect goal coding, influencing the strength of goal representations that are subsequently used for the act of goal-directed retrieval (or other cognitive acts) and memory-guided decision-making. As a consequence, the efficacy of retrieval cues to trigger medial temporal cortical familiarity signals and hippocampal pattern completion and the readout of mnemonic evidence may be affected both within, and between, individuals.

With advances in the ability to quantify the strength of attentional preparedness and goal states at the trial level through EEG time-frequency analyses, pupillometry, and multivariate analyses over fMRI data (e.g., [67]), the field is now well positioned to examine how in the moment attention and goal-state fluctuations at the time of retrieval affect core mnemonic mechanisms, including MTL familiarity and hippocampal pattern completion computations. Given the posited centrality of preparatory attention and goal coding for goal-directed task behavior, the R2R framework predicts that interactions between these two preparatory processes also contribute to individual differences in goal-directed abilities that propagate to influence MTL mnemonic computations and subsequent mnemonic evidence readout/integration, offering a partial account of across-subject variability in episodic remembering.

Empirical evidence for readiness to remember

Emerging evidence (Figure 2) supports the hypothesized R2R pathways (Figure 1) between pre-goal preparatory attention to goal coding, pre-goal preparatory attention to episodic retrieval, and goal coding to episodic retrieval, within and across individuals.

Figure 2. Readiness-to-remember (R2R) explains variance in episodic memory and its extensions.

(A) Orienting to different types of mnemonic goals (e.g., thematic/conceptual versus spatial/perceptual) modulates goal-directed remembering. How the content and strength of frontoparietal neural patterns of mnemonic goals modulate neural signals of remembering is an open question. (B) Pre-goal attentional lapsing, indexed by posterior alpha power, relates to behavioral forgetting directly, and as partially mediated via midfrontal event-related potential (ERP) goal coding strength. Subplots for pre-goal alpha and goal code indicate that increases in alpha (continuous x-axis) and decreases in goal coding strength (continuous x-axis) predict a higher probability of a miss (categorical y-axis) from trial to trial. (C) Receiving an episodic specificity induction, a goal-directed strategy that biases for detailed retrieval, has downstream state effects on memory-related functions, including imagination and divergent creative thinking. Neural effects of induction on activity and connectivity of default mode and cognitive control networks are supported by behavioral generation of episodic details in imagination and creative uses in divergent thinking. Abbreviations: Assoc, associations; CCN, cognitive control network; Def, definitions; DMN, default mode network; ESI, episodic specificity induction; HIPP, hippocampus; MF, midfrontal; MTL, medial temporal lobe; PCC, posterior cingulate cortex; PreC, precuneus; retrieval_S, spatial retrieval; retrieval_T, thematic retrieval. Based conceptually on [69,83] (A), [20] (B), and [100,105,106] (C).

Attentional influences on goal states and memory in the moment

Most evidence for attention to goal and attention to memory interactions comes from research on processes during goal-directed retrieval itself, rather than during preparatory periods. Top-down and bottom-up attention and control mechanisms, supported by the CCN, DAN, and ventral attention network, interact during the act of making a goal-directed retrieval decision and, in some cases, especially when such decisions are marked by uncertainty (e.g., [24,68–70]). A rich fMRI and, to a lesser extent, lesion, literature has revealed multiple distinct influences of attention and control during remembering, and their interactions with mnemonic signals, with the mechanistic implications remaining a focus of ongoing work and theorizing [22,26–28,68]. Recent EEG/MEG studies also highlight the roles of phasic (stimulus locked rather than tonic) posterior alpha during different kinds of mnemonic decisions [71,72]; larger decreases in posterior alpha are observed during successful associative retrieval than during item recognition [72]. Moreover, posterior alpha during the presentation of a retrieval cue is sensitive to cue-specific mnemonic precision, and appears to recapitulate alpha power patterns from the encoding of the cue [72]. At the same time, extensive behavioral work on the impacts of concurrent divided attention during the act of remembering shows mixed evidence for impacts on retrieval outcomes, with the match/mismatch between selective attentional allocation and memory modalities being predictive [8].

Goal-state and memory interactions in the moment

Extant evidence bears on the pathway from goal coding to episodic retrieval within the individual. For example, scalp EEG studies demonstrate that the strength or maintenance of preparatory retrieval goal signals, as indexed by left and midfrontal ERPs (including F3/F7 for left frontal and Fz for midfrontal), relate to mnemonic accuracy [73,74] (for evidence from single-neuron recordings in medial prefrontal cortex, see [75]). Given that such signals onset before a subsequent cue that probes a specific memory, they may represent a retrieval orientation or mode [76,77], which is a subset of goal coding that is often examined from task block to task block (e.g., [78]), but can also be meaningful on shorter temporal scales, from trial to trial (e.g., [20]); these signals are sometimes referred to as a prestimulus ERP [78]. Goal coding is not solely reducible to retrieval orientation or mode, however, because some evidence indicates that preparatory signals also represent cognitive conflict, flexibility, and gating (e.g., [75]).

Functional MRI evidence (Figure 2A) also indicates that the processing and use of goal states impact whether and how we retrieve, with the adoption of a specific retrieval goal or orientation modulating memory signals [69,79–84]. For example, univariate fMRI data [79,80,82] indicate that left hemisphere nodes of the CCN (including left inferior frontal sulcus, left intraparietal sulcus, and frontopolar cortex) are engaged when orienting to retrieval goals that guide recollection, while right inferior frontal sulcus and right intraparietal sulcus are differentially engaged when representing the goal of novelty detection and, thus, may be part of a network that subserves familiarity-based memory decisions (although left hemisphere regions can be sensitive to perceived oldness [80]). Mnemonic behaviors, including accuracy, confidence, and RT, are also impacted by the retrieval goal [78–84]. In addition, proactive control computations drive hippocampally dependent retrieval [85], as do memory inhibition/control computations [86], such that different task goals that precede remembering impact the likelihood that particular mnemonic representations will be activated/deactivated or read out during a goal-directed retrieval attempt.

Individual differences

Across individuals [87,88], some evidence points to trait-level abilities that relate to memory variability, offering initial support for the R2R framework. In terms of the pathway from attention to memory, trait relationships between the ability to sustain attention (arousal and vigilance) and the ability to remember are evident in studies using self-report and independent task-based behavioral measures of attention and memory [89]. Moreover, individuals with attention-deficit/hyperactivity disorder (ADHD) also typically show worse episodic memory [90]. In terms of the pathway from goal coding to memory, individual differences in the ability to process and engage various goal operations before retrieval (as assayed by midfrontal ERPs) predict memory [91]. Concerning the pathway from attention to goal coding, the person-specific functional connectome, or systems-level, approach (Box 2) reveals that variability in memory may be driven, at least in part, by variability in person-to-person interactions between networks of attention and goal coding. For example, variability in the specificity of different types of memory behavior (e.g., episodic and semantic) from person to person has been related to fMRI resting-state connectivity profiles [92], with MTL, CCN, and DAN coupling being predicted by variance in self-reported memory abilities.

Examining R2R in the moment and between individuals

To integrate and test the posited attention, goal, and memory pathways, EEG and pupillary metrics were recently combined during goal-directed episodic encoding and retrieval to test whether lapses of attention before each trial, in the moment and across individuals, predict neural fluctuations in goal coding and variability in neural and behavioral expressions of retrieval success or failure (i.e., forgetting) [20] (Figure 2B). Pre-goal attention (arousal and vigilance) was indexed by posterior alpha and pupil diameter during the retrieval task, and an individual-differences battery of task-based (gradCPT) and self-report sustained attention constructs were also included to assess whether putative trait-level differences in the propensity to lapse partially explain why some individuals are more likely to forget.

Multiple outcomes from this study support the R2R framework. First, spontaneous moment-to-moment pre-goal lapses before a retrieval attempt predicted a higher likelihood of subsequent retrieval failure and weaker neural signals of memory, as assayed via the canonical left posterior parietal old/new (including P3; recollection/pattern completion) and left midfrontal FN400 (including F3 and F7; familiarity/sum similarity) ERP components of episodic retrieval [78]. Second, a neural marker of goal coding, a midfrontal ERP component previously implicated in task set processing [48], was negatively predicted by pre-goal lapses; in turn, weaker goal coding predicted subsequent retrieval failure. Third, this fluctuation in goal coding strength partially mediated the relationship between pre-goal attention lapsing and memory, indicating that a lapse in preparatory attention leads to fluctuations in goal coding strength before a retrieval attempt, which in turn affects both recollection and familiarity-based memory retrieval and decision processes. Fourth, complementing the within-subject effects, individual differences in the ability to sustain preparatory attention partially explained why some individuals are more likely to forget. As such, multiple independent assays of preparatory attention predicted multiple behavioral and biological goal coding and memory outcomes.

Synthesis

Extant data provide initial evidence for R2R pathways within and across individuals that can be extended in future work, including through the use of EEG-fMRI and machine learning analytics. Building on current EEG-fMRI retrieval data [93,94], it will be of value to further quantify and understand interactions between frontoparietal networks of attention and goal coding, and between the mechanisms subserved by these networks and MTL cortical and hippocampal mechanisms of memory. Multivariate BOLD pattern analyses that enable discrimination between goal states and that quantitatively assay the strength of specific states [54,55] may be a particularly fruitful avenue, especially when combined with temporally sensitive EEG and ocular measures of attention, and spatially sensitive fMRI univariate and multivariate measures of memory states and evidence (e.g., familiarity versus recollection; cortical reinstatement of specific content).

Given this landscape, there are several implications (Box 3) of the R2R framework for theory and practice that stem from an increased understanding of how preparatory factors drive memory variability in the moment and between individuals.

Box 3. Implications of readiness for theory and practice.

Scholars and the public are well positioned to leverage readiness to learn and to remember in at least four ways. First, the differentiating and interacting roles of readiness to learn and to remember in memory expressions will propel theoretical progress. Within individuals, mixed-effects modeling enables examination of whether and how memory of an episode in the moment is best predicted by preparedness before encoding, before an attempt to retrieve, or a combination of both. Between individuals, recent work indicates that memory is predicted by retrieval preparedness even after accounting for encoding preparedness [20]. These kinds of advances will also inform the understanding of boundary conditions (as with the modulatory effects of acetylcholine [124]).

Second, the identification of preparatory effects at retrieval will inform theorizing about the mechanistic causes of memory variability. For example, variability attributed to experimental manipulations of core retrieval processes may be confounded with pre-retrieval preparedness fluctuations. Under such conditions, if preparedness fluctuations are not also measured, then the effects of the experimental manipulation might be incorrectly attributed to effects on retrieval processes. In this context, the computation of match/mismatch mnemonic prediction error is considered a core mechanism of an adaptive learning and memory system, driving dopaminergic modulation of plasticity and performance. However, prediction errors also may yield an upregulation of attention and control that influences retrieval and results in trial-history effects. Examination of relationships between targeted constructs and episodic memory should account for preparedness.

Third, understanding the consequences of preparedness at encoding versus retrieval has practical implications. Consider a school lecture: if an individual is unprepared to learn and does not encode the material, then they have missed the opportunity to do so and cannot access this knowledge later. Here, learning-stage interventions will be critical to enable future memory-guided performance. At the same time, if an individual is prepared to learn and effectively encodes the material, they might nonetheless be unprepared to retrieve their knowledge when later prompted. Such moments provide an opportunity to intervene by transitioning the individual into a more ready state. This distinction makes contact with intervention science, including closed-loop feedback [125]. When states of unpreparedness before learning or retrieval are detected within the individual, reorienting signals may influence the probability of memory success or failure. Understanding memory variability as a function of preparedness may also help to uncover why only some individuals respond to closed-loop feedback [126].

Fourth, the R2R framework is likely to be broadly relevant, in the sense that the preparedness factors emphasized here for episodic memory likely also predict variability in expressions of other forms of memory. Indeed, evidence indicates that preparedness signals predict goal-directed behavior in semantic memory tasks (e.g., [59,63]). Given that effects of preparedness are unlikely to be selective to declarative memory, it will be important to explore further whether and how R2R factors penetrate other memory systems. As has been posited and debated, given variable effects of divided attention on implicit forms of memory (e.g., priming), it may be that some forms of memory are differentially impacted by preparedness dynamics [127].

Extensions of readiness to remember

Extrapolating from readiness to learn and to remember in memory tasks, the R2R framework also has relevance for understanding variability in performance on complex tasks that depend, in part, on mnemonic functions. Over the past decade, behavioral, neuropsychological, and neuroimaging data indicate that we draw on episodic memories to prospect about novel events specific in time and place (episodic simulation), as well as to perform other generative tasks, such as divergent creative thinking and means-end problem solving/planning [95,96]. R2R pathways at the state and trait levels help to predict variability in these high-level cognitive acts.

Some variability in complex expressions of mnemonic functions can be attributed to preparatory attention directly and via goal coding interactions. Data from creativity studies show the potential of this approach, because state and trait variability in posterior alpha (resting and task-based), attentional allocation (e.g., defocused/leaky, focused, or flexible), and disengagement (e.g., mind wandering, incubation, or switching) can predict creative thinking [97]. In addition, state and trait variability in control and attention interactions, as supported by resting and task-based CCN, DMN, and DAN connectivity, predict creative thinking, problem-solving, future planning, and navigation [16,97].

Preparing to retrieve a memory, or becoming ready to remember with a goal-directed strategy, also contributes to variability in performance on cognitive tasks that nominally involve episodic retrieval. One recent experimental method, called episodic specificity induction (ESI), involves biasing cognition toward remembering an event in specific detail, or becoming ready to remember via adoption and maintenance of an extended mnemonic goal state. Evidence of this bias is then examined through transfer state effects on subsequent tasks that nominally involve episodic retrieval [98] (Figure 2C) (for related evidence in clinical populations, see [99]). Biasing for specific episodic detail typically leads to increases in episodic expressions on memory, simulation, divergent thinking, and means-end problem-solving tasks, but does not increase performance on tasks that are thought to have minimal episodic retrieval components, such as describing pictures, defining words, or convergent thinking [100]. ESI manipulations can also modulate subsequent assays of false memory [101], navigation [102], discounting [103], and empathy [104], highlighting how variability in mnemonic goal states modulates complex behavior that is thought to depend, at least in part, on episodic memory.

Functional MRI data associated with ESI manipulations, and concomitant behavioral correlates, further support a pathway between goal states and mnemonic function. Anterior hippocampus and inferior parietal lobule, key nodes of the DMN, show greater univariate activity during simulation following ESI versus nonepisodic/control inductions [105]. Moreover, anterior hippocampus shows greater univariate activity, and multivariate BOLD patterns from nodes of the DMN and CCN show stronger coupling, during divergent thinking after ESI [106] (for related evidence, see [107]). These data suggest that some variance in induction effects are due to the priming of goal-directed proactive control [51], which could bias a memory-guided task set [108]. This work also makes contact with studies suggesting that lingering mnemonic states impact retrieval processes [109], perhaps via acetylcholine-linked effects on control and attention processing [94,110].

Variability in episodic memory-like tasks may be partially explained by preparatory attention and goal coding interactions at state and trait levels. New explanations for understanding variability in high-level cognitive expressions that emerge, in part, from memory retrieval, are likely to come from future studies that link tonic changes in posterior alpha and pupil diameter with concomitant changes in multivariate goal-state, attention, and mnemonic activity patterns in the CCN, DAN, and DMN. Delineating how R2R factors subserve variability in everyday cognitions is also important for research and practice identifying why state and trait memory can differ as a function of aging (Box 4).

Box 4. Attention, goals, and memory in healthy and pathological aging.

Individuals vary in many ways, including along dimensions (e.g., age) that co-vary with or impact neurocognitive functions. The R2R framework makes contact with studies and theories suggesting links between diminished attention- and goal-related function and altered memory ability in groups of individuals (e.g., older versus younger adults) and between individuals where there is marked variability in remembering (e.g., across older adults).

While the cognitive neuroscience of healthy and pathological aging and memory has justifiably focused on identifying changes in the medial temporal lobe [128], it has long been known, and is becoming increasingly more precisely specified, that changes in neurocognitive networks of attention and goal processing critically relate to neurobiological markers of Alzheimer’s disease (AD) pathology, and also account for variance in memory in putatively healthy older adults [129–132] (Figure I). While beta-amyloid (Aβ) accumulation, especially when accompanied by tau burden, is a predictor of memory decline and progression to dementia in older adulthood [128], recent data also suggest that, in older adults, regional Aβ can appear in nodes of frontoparietal networks, as well as in the DMN [133]; Aβ positivity further relates to changes in connectivity between frontoparietal attention and control networks and the DMN [133]. Recent data also suggest that variance in locus coeruleus (LC) integrity, which, as noted, is a core structure that mediates attention, arousal, and goal-directed cognition, and is related to pupil diameter changes [40], predicts subject-level differences in memory performance in older adulthood [134]. Importantly, reduced LC integrity is likewise related to distractibility, and the LC itself is an early site of AD pathology [135]. These data support the attention-to-goal, goal-to-memory, and attention-to-memory R2R pathways. Finally, emerging data suggest that behavioral assays of ‘trait’-level sustained attention (e.g., from RT and error rates) predict early AD, and are also associated with white matter changes in frontoparietal networks and projections from the LC [136], further pointing to impacts due to dysfunction in the attention to memory pathway.

One prediction of R2R is that aging is associated with increases in the propensity and intensity of attention and goal-state lapses before remembering, which partially account for age-related memory decline. This prediction is consistent with known age-related differences in sustained attention [14], reactive (versus proactive) control [132], dedifferentiation of neural responses [128], motivational selectivity [137], and the representation and use of task-irrelevant goals and information [138,139]. These age-related differences may partially explain why older adults often exhibit differential declines in pattern completion-dependent episodic remembering (e.g., tests of associative versus item-based memory [140]), as well as in episodic simulation [97]. A meaningful extension will be to determine and formalize which pathway(s) of R2R (e.g., attention to goal) drive the most variability in memory change from one older adult to the next.

Figure I. Readiness to remember (R2R) explains variance in memory change in aging.

Sustained attention ability, as measured by the Sustained Attention to Response Task (SART) total errors (commission and omission), differentiates those with early Alzheimer’s disease (AD) from putatively healthy older adults (left panel). Conceptually from [107]. Integrity of the locus coeruleus (LC), a core brain structure that mediates arousal, is correlated with memory ability in older adults; this is particularly observed for rostral LC (right panel). Based conceptually on [107] (left) and [134,135] (right).

Concluding remarks

The R2R framework synthesizes multimodal streams of research to suggest that variability in episodic retrieval in the moment and across individuals can be driven by fluctuations in states of preparatory attention, goal coding, and their interactions with each other and with core mnemonic computations. R2R posits that episodic retrieval, and other mnemonic functions, such as prospection and creative thinking, can be driven by several factors that precede the act of remembering itself. The consideration of factors that interact in the service of remembering, such as attention and goal coding, may also advance understanding of age- and disease-related changes in memory and goal-directed behaviors.

Several open and promising avenues related to R2R should be probed (see Outstanding questions). Given that preparatory attention and goal coding are not monoliths and there are multiple sources of variance [15,47,111] that impact the strength and implementation of each of these processes at multiple temporal frequencies, it will be important to pinpoint what sources of variance underlie differences in memory expressions, as well as their relative weightings. While R2R is compatible with a spontaneous mind-wandering or mind-blanking interpretation linked to hypoarousal/underload [19,112,113], future work should expand on this mechanistic question. Extending from current work in contexts outside episodic memory, the use of precise assays of thought content (e.g., [37]) before and during the context of goal-directed remembering will be particularly informative, as will closed-loop intervention paradigms [114] with ocular and EEG assays of temporally sensitive impacts of reorienting attention, goal, and motivation lapses before and during remembering.

Outstanding questions.

What sources of variance (e.g., mind blanking or mind wandering) underlie fluctuations in attentional lapsing and goal coding?

What are the relative weightings of preparatory attention and preparatory goal-state coding in explaining variability in episodic memory?

Do fluctuations in preparatory attention and goal coding have different predictive ability when spontaneous or deliberate, when fast or slow, and when involving different individuals completing different tasks?

Can real-time closed-loop attention, goal, and motivation intervention approaches be developed from oscillatory, ocular, and neural signals, with concomitant impacts on remembering?

Are particular assays of attention lapsing and goal coding predictive biomarkers in populations characterized by memory decline, such as aging and Alzheimer’s disease?

Can preparatory attention lapsing and goal coding explain variance in temporally relevant dimensions of remembering, such as how events are segmented, naturalistic stimuli are processed, and representations are updated via prediction errors?

Are there positive aspects of attention and goal lapsing before acts of memory and other goal-directed cognitions, including those related to exploration, curiosity, and reward?

In terms of bidirectional relationships in R2R, how does memory guide preparatory attention and goal coding in the moment and across individuals?

Scholars often analyze remembering in an all-or-none (success or failure) fashion, while also noting that memory strength varies continuously (see [115] for work with continuous memory precision). Utilizing continuous measures of memory, as well as of preparatory processes, and integrating them with computational modeling to probe R2R pathways may be particularly fruitful, because relationships among attention, goal coding, and memory may not be linear. Related to pathways, it will also be important to examine whether and how post-goal interactions with attention (e.g., from midfrontal theta power [116]) impact memory, as well as how memory guides attention and goal coding [8,117–119]. We anticipate that theories of memory-related phenomena with a sensitive temporal dimension [120], such as event segmentation and prediction errors, will likewise be advanced through probing R2R factors, both with ocular and EEG assays (including slow-wave assays [121]).

One feature of a constructive and adaptive episodic memory system is forgetting. Identifying the positive aspects of attention and goal-state lapsing, potentially related to exploration, curiosity, reward, and motivated forgetting effects, warrants further experimental effort [62]. Investigating these contemporary questions promises to build on, and inform, R2R as a framework of episodic retrieval variability and the broader consequences of such variability for high-level cognition.

Highlights.

Two processes before episodic memory retrieval (our abilities to prepare attention and to represent and use task goals) relate to whether an imminent retrieval attempt will succeed or fail.

We propose an integrative readiness-to-remember (R2R) framework that explains variance in episodic remembering as a function of preparatory attention, preparatory goal coding, and their interactions with each other and with core mnemonic processes both within and between individuals.

This R2R framework also accounts for state- and trait-level variability in higher-level cognitive functions that depend, in part, on episodic retrieval, such as prospection and creativity, and informs understanding of the mnemonic disruptions linked to aging.

Glossary

EEG-fMRI

concurrent acquisition of electroencephalography and functional magnetic resonance imaging measures, which non-invasively capture temporally sensitive electrical and spatially sensitive hemodynamic activity, respectively

Episodic remembering

pattern completion and reinstatement of the details of a past event, which is often accompanied by the subjective experience of recollection

Episodic simulation

neurocognitive set of processes that support prospection of a novel future event, specific to a time and place

Episodic specificity induction (ESI)

goal-directed strategy or orientation that biases for detailed episodic retrieval of an event via a generative set of questions, which often has downstream consequences on subsequent tasks that nominally involve episodic retrieval as a component

ERP components of episodic retrieval

event-related brain potentials that index recollective- and familiarity-based memory signals, including the parietal old/new component over left posterior electrodes (with P3) associated with recollection and the FN400 component over midfrontal electrodes (with F3) associated with familiarity

Pattern completion

one mechanism postulated to underlie memory retrieval, whereby a stored mnemonic representation is reconstructed from a partial cue that is part of the representation

Preparatory attention

ability to engage and sustain vigilance, arousal, selective attention, and motivation before an upcoming task or experience

Preparatory goal-state coding

representation of the structure or rules for performance of a task, which involves processing, maintaining, and using the representation, often in terms of specifying the feature-to-categorization/response mappings required by the task

Readiness-to-remember (R2R)

neurocognitive interactions between factors, including preparatory attention, preparatory task goal processing, and pattern completion, that predict moment-to-moment and across-subject variability in remembering and higher-level cognitions that involve remembering

Retrieval orientation

goal-directed strategy or memory state that biases processing of a subsequent cue to mnemonically relevant information, often episodic