2.1 Familiarity and feelings of fluency

The experimental induction of déjà vu converges on one theoretical entity: Familiarity. The implication is that retrieval is incomplete: Participants find a stimulus familiar but at the same time, are unable to retrieve its source (the Brown, Marsh, and Cleary experiments), or know that the familiarity is false (the Urquhart and O'Connor experiments). In terms of memory retrieval, as pointed out by Renoult, Irish, Moscovitch, and Rugg (Reference Renoult, Irish, Moscovitch and Rugg2019), familiarity is often defined by default – it is what occurs in the absence of the rich evocative mental time travel which is experienced in episodic remembering (Renoult et al., Reference Renoult, Irish, Moscovitch and Rugg2019). For a definition of familiarity, we turn to the recognition memory literature, where it is described as a recognition memory decision-making process based on the evaluation of trace strength, a judgement of prior occurrence based simply on a feeling, rather than the retrieval of information (e.g., Yonelinas, Reference Yonelinas2002). A robust literature examining this trace strength account using signal detection models and receiver operating curve analysis exists (for a review, see Yonelinas & Parks, Reference Yonelinas and Parks2007; see also Delay & Wixted, Reference Delay and Wixted2021, for an example application to recognition memory making the case for a continuous process).

The focus here however is on phenomenology rather than recognition memory performance, and perhaps the idea of familiarity is most intuitive when we talk about first-person experiences, such as “the butcher on the bus”:

In Figure 1, we propose that such phenomenological familiarity occurs when there is (temporarily) no content retrieved. It is felt in both involuntary and voluntary retrieval, but as we will explain below, it serves to guide and motivate retrieval processes: A sense of familiarity arises without any intention, but the subsequent search for corresponding information can be intentional. Before examining this metacognitive account in more detail, we quickly review the notion of familiarity in recognition memory.

Recognition of a cue in the environment arguably relies on two retrieval modes or processes, familiarity and recollection (e.g., Yonelinas, Reference Yonelinas2002). The neural and cognitive organisation of these entities is often debated (for an overview of this debate, see Moulin, Souchay, & Morris, Reference Moulin, Souchay and Morris2013, and for a proposal for the integration of these ideas, see Bastin et al., Reference Bastin, Besson, Simon, Delhaye, Geurten, Willems and Salmon2019). Simons, Ritchey, and Fernyhough (Reference Simons, Ritchey and Fernyhough2022) propose that early characterisations (such as Yonelinas's [Reference Yonelinas2002] account of a hippocampal mechanism and a simple threshold that gives rise to recollection) are helpful for explaining binary (laboratory-based) memory decision making. However, with more complex representations of the personal past, they propose that additional to this hippocampal reinstatement, cortical regions such as the angular gyrus are recruited to generate a precise representation of the episode (e.g., Richter, Cooper, Bays, & Simons, Reference Richter, Cooper, Bays and Simons2016). As such, they describe remembering as drawing upon two different dimensions: Retrieval success and the precision of the details retrieved.

We emphasise separable phenomenological experiences of familiarity and recollection, and we are agnostic as to whether these map onto one or two underlying processes, although as will be seen, our ideas converge on the idea of a trace-strength account based on the continuity between familiarity and recollection. A nuanced view would be that complex interactions exist between networks traditionally thought responsible for familiarity (extra-hippocampal regions) and recollection (the hippocampus) such as proposed by Quamme, Yonelinas, and Norman (Reference Quamme, Yonelinas and Norman2007), with also the recruitment of an area around the left angular gyrus which yields the multisensorial information experienced in remembering as outlined by Simons et al. (Reference Simons, Ritchey and Fernyhough2022). As proposed by Montaldi and Mayes (Reference Montaldi and Mayes2010), we suggest that recollection and familiarity are related: “…each is a complex function, likely to depend on several different processes that are probably mediated by different structures that are functionally connected in a system” (p. 1294). Whereas the consideration of recollection and familiarity as processes favours an interaction between shared neural and cognitive processes in a continuum, for the experient, recollection and familiarity are separable signals which differ in their phenomenology and content.

The critical issue, as identified by Bastin et al. (Reference Bastin, Besson, Simon, Delhaye, Geurten, Willems and Salmon2019), is that these signals are interpreted differently, and have consequences for retrieval. They propose an integrative model with a representational hierarchy and an attributional system. The first is a “core” system that stores specific representations based on interactions between a large network of computational operations from the whole brain, split into an entity representation system (guided by feelings of fluency – see below) and a relational representation system (responsible for binding and pattern completion) centring on the hippocampus. A second system acts on the current context and “translates content reactivation into a subjective experience” (p. 5) and includes input from networks in the parietal and prefrontal cortexes. Critically, their model rests on differences in the outputs and representations of their core systems, and an interaction between processes and computations, resonating with the Montaldi and Mayes “kinds” of memory proposal and our view about separable phenomenology. In autobiographical memory, the retrieval of specifics from the personal past is accompanied by a feeling of mental time travel or autonoesis (see Conway, Reference Conway2009), directly comparable with the concept of recollection: A detailed representation of a moment in the personal past. The idea of the “self in past” (Conway, Reference Conway2009; Klein, Reference Klein2013; Marshall, Halligan, & Wade, Reference Marshall, Halligan and Wade1995; Moulin & Souchay, Reference Moulin, Souchay, Perfect and Lindsay2014) is that it enables distinguishing between imagined, false, and actual events and gives episodic memory its particular character (Perrin & Rousset, Reference Perrin and Rousset2014). Recollection gives the feeling of completeness, and the resolution of the feeling of familiarity. In the butcher on the bus, if we are able to identify the person we have seen, we resolve the feeling of familiarity and the feeling of recollection confirms successful retrieval. The phenomenology of remembering is indeed associated with high degrees of confidence (Dunn, Reference Dunn2004).

Recognition decisions made based on familiarity are proposed to be faster and more automatic than those based on recollection (e.g., Besson, Ceccaldi, Didic, & Barbeau, Reference Besson, Ceccaldi, Didic and Barbeau2012; Hintzman, Caulton, & Levitin, Reference Hintzman, Caulton and Levitin1998; Hintzman & Curran, Reference Hintzman and Curran1994; Mandler, Reference Mandler2008; see Dewhurst, Holmes, Brandt, & Dean, Reference Dewhurst, Holmes, Brandt and Dean2006, for a discussion of this issue, and Berry, Shanks, Speekenbrink, & Henson, Reference Berry, Shanks, Speekenbrink and Henson2012, for a single-process explanation of this pattern of results). In tasks where recollection and familiarity are compared by asking participants to make decisions based on the information they are able to retrieve, Yonelinas and Jacoby (Reference Yonelinas and Jacoby1994) found that recollection decisions peaked at 800–1100 ms, whereas familiarity decisions peaked earlier at between 600 and 800 ms. In EEG paradigms (e.g., Duzel, Yonelinas, Mangun, Heinze, & Tulving, Reference Duzel, Yonelinas, Mangun, Heinze and Tulving1997; Rugg & Curran, Reference Rugg and Curran2007), two separate neural signatures are found for old/new decisions in episodic memory. The FN400 effect (between 300 and 500 ms, e.g., Curran, Reference Curran1999; a negative potential in the frontal areas) is proposed to reflect familiarity processing, whereas the parietal old/new effect (sometimes called the late parietal effect) is a positive potential in the parietal area arising slightly later, after about 500 ms (e.g., Tsivilis et al., Reference Tsivilis, Allan, Roberts, Williams, Downes and El-Deredy2015).

Familiarity is often described as a low-level process working on perceptual inputsFootnote ⁵ (Bastin et al., Reference Bastin, Besson, Simon, Delhaye, Geurten, Willems and Salmon2019). This part of the memory system tries to match, as quickly and effortlessly as possible, the contents of mental representations stored in memory with the current contents of perception; consistent with the classic neuroanatomical view of the temporal-lobe memory system as being the last point in the ventral visual stream (e.g., Suzuki & Amaral, Reference Suzuki and Amaral1994). Viewed like this, familiarity decision making is the last stage in perception – once we have composed and identified a scene or environment, we can “read off” whether we have encountered it before. More recent neuroanatomical accounts of memory retrieval propose a similar organisation. Moscovitch (Reference Moscovitch2008) describes two phases in episodic retrieval. A first stage involving the hippocampus is non-conscious, based on a rapid interaction between a retrieval cue and a stored representation, whereas a second stage, featuring a more complex interaction between the hippocampus, prefrontal, and parietal cortex, is required for the retrieved information to be re-experienced. These accounts are inspired by the notion of an organisational hierarchy of connectivity (e.g., Margulies et al., Reference Margulies, Ghosh, Goulas, Falkiewicz, Huntenburg, Langs and Smallwood2016). Irish and Vatansever (Reference Irish and Vatansever2020, p. 47) propose a “…concrete to abstract representation spectrum along modality-specific and default mode brain networks which collectively support rich and detailed memories,” such that rather than specifying different stores or types of information as in the standard episodic/semantic distinction (see also Renoult et al., Reference Renoult, Irish, Moscovitch and Rugg2019), the memory system emerges as part of a larger scale network with retrieval of specific events from the personal past as being a type of interrogation of the stored representations based on a common network, rather than something restricted to a specific zone, region or process in the brain. As such, memories might be thought of as reflecting a dynamic pattern of local to regional connectivities, varying along the continuum from abstract to specific and more or less involving the sensorimotor regions (Irish & Vatansever, Reference Irish and Vatansever2020). For an up-to-date review of the neural processes in the subjective experience of remembering taking on board both the large-scale networks and traditional “regions” involved in memory, see Simons et al. (Reference Simons, Ritchey and Fernyhough2022).

The existing cognitive frameworks resonate with the neuroscientific view, but discuss phenomenology in a way which is somewhat lacking in neuroscientific accounts (see Renoult et al., Reference Renoult, Irish, Moscovitch and Rugg2019). In Whittlesea's SCAPE (Selective Construction and Preservation of Experience) model (Whittlesea & Williams, Reference Whittlesea and Williams2001), the memory system is continuously trying to make sense of its inputs, so that it can interpret any signals arising from low-level processing of the environment. Familiarity is thus a subjective feeling arising from the fluent processing of a stimulus. This theory draws upon various illusions of familiarity: Where the fluency of processing leads to the (erroneous) decision that we have seen something before, or to “know” an answer (e.g., Goldinger & Hansen, Reference Goldinger and Hansen2005; Newman, Garry, Bernstein, Kantner, & Lindsay, Reference Newman, Garry, Bernstein, Kantner and Lindsay2012; Whittlesea & Williams, Reference Whittlesea and Williams1998). Phonological and structural regularity can be used to induce feelings of fluency which are misinterpreted as meaning we have seen the information before. For example, in the hension effect, people are proposed to be more likely to make false positives for pseudowords such as hension (which is regular and resembles a real word) than stofwus (for a critique, see Cleary, Morris, & Langley, Reference Cleary, Morris and Langley2007). The idea is that the ease at which they read the word is misinterpreted as it being an item from a study list. Golding and Hansen push the idea one step further: Participants respond to meaningless subliminal vibration cues, interpreting them as a signal they have seen the word before. There is some debate about whether the critical element is “surprise” in false memories, and it seems unlikely that Whittlesea's view could explain all kinds of false memory. For instance, in the DRM paradigm, the critical lures for which participants have false memories are expected as targets, and so the false memory cannot be described as surprising (Karpicke, McCabe, & Roediger, Reference Karpicke, McCabe and Roediger2008). Moreover, Cleary et al. (Reference Cleary, Morris and Langley2007) point out that structural regularity usually leads to benefits in recognition memory, not an increase in false positives, suggesting that empirical support for Whittlesea's ideas on false memories is weak. Cleary et al. (Reference Cleary, Morris and Langley2007) found that the hension effect disappeared completely once controlling for inter-item similarity in the recognition test materials, and so interestingly, this effect which formed the basis of the original theorising receives little empirical support. The role of phenomenological fluency in recognition memory is something which will need continued evaluation, even though the basic idea, that the ease of processing leads to an attribution of prior exposure, is well established (e.g., Jacoby & Dallas, Reference Jacoby and Dallas1981; but see also Berry et al., Reference Berry, Shanks, Speekenbrink and Henson2012).

Whilst there is not unequivocal empirical support from recognition memory studies, Whittlesea describes an interaction of attributions of memory and processing fluency, something that is critical for thinking about how cognitive “feelings” guide retrieval processing (Bastin et al., Reference Bastin, Besson, Simon, Delhaye, Geurten, Willems and Salmon2019; Moulin & Souchay, Reference Moulin, Souchay, Perfect and Lindsay2014). The idea that we adopt here is that the current context generates a top–down expectation of processing fluency. When an expectation of fluency is violated, it triggers a search for as to why that has arisen. This is the search triggered in the case of the butcher on the bus; the person feels intensely familiar precisely because the source of the familiarity is unknown. In this way, we align our reasoning with Whittlesea's observation that strangers feel familiar, but our friends do not (Whittlesea & Williams, Reference Whittlesea and Williams1998): The phenomenology of familiarity is produced by a mismatch between processing fluency and the expectations about fluency. The spontaneity and phenomenology of the experience are one and the same: It feels familiar because it was unexpected.

Fluency is also a variable of interest in IAM studies. Sanson et al. (Reference Sanson, Cardwell, Rasmussen and Garry2020; see also Barzykowski et al., Reference Barzykowski, Staugaard and Mazzoni2021b) investigated the influence of fluency on classification of intention in a paired-associates design. Participants studied a series of phrases, half of which were accompanied by a phrase-related photo. During retrieval, participants were presented with the phrase only and then instructed to retrieve additional information about it. Crucially (Experiment 3; Sanson et al., Reference Sanson, Cardwell, Rasmussen and Garry2020, p. 12), participants reported if they experienced retrieval as spontaneous (“the information came to mind without intent”) or deliberate (“the information was brought to mind intentionally”). Although they were explicitly instructed to intentionally retrieve memories, fluency overruled intentionality when classifying their memories. When participants intentionally, yet fluently, brought a memory to mind, they were prone to judge the retrieval as spontaneous. Similar results were obtained in study by Barzykowski et al. (Reference Barzykowski, Staugaard and Mazzoni2021b) who investigated the effect of retrieval effort while classifying memories. They created experimental conditions that maximise the probability of one type of retrieval (e.g., involuntary or voluntary) while minimising the probability of the other. Participants recalled autobiographical memories in each condition, retrieval time was measured, and they classified their memories as either voluntary or involuntary and rated them on retrieval effort. There were four categories of memories: Experimentally defined voluntary and involuntary memories with an objective measure of effort (retrieval time) and subjectively classified involuntary and voluntary memories with a subjective measure of effort. Similar to Sanson et al., the majority of memories were classified as involuntary, whether they were retrieved in the experimentally defined voluntary or involuntary condition. Interestingly, the relationship between the rating of subjective fluency and objective retrieval time was only significant in the voluntary conditions, indicating that people are more aware of subjective effort when deliberately trying to retrieve a memory than for an involuntary memory. In contrast with existing theories, these results indicate that the subjective feeling of fluency is more important than intention for the involuntary–voluntary distinction. Such experiments suggest that low-level feelings are used when making explicit decisions about memory.

These findings question the interpretability of existing studies on involuntary memories in which fluency has not been controlled for. Sanson et al. (Reference Sanson, Cardwell, Rasmussen and Garry2020, p. 2) argued “[…] subjects themselves classify their memories as voluntary or involuntary, and so the validity of what scientists know about involuntary memories rests upon the validity of this task. How well do subjects perform this task, then? Here, we show that, at least under certain circumstances, the answer is not very well.” Our view is more moderate: Classifications are internally consistent in terms of retrieval latency and retrieval effort, especially during voluntary retrieval. At the same time, it is much less of an issue during involuntary retrieval, since the vast majority of participants' classifications align with the classification set by the experimental condition. As such, these findings do not undermine previous results on involuntary and voluntary memories but rather extend our understanding of retrieval. Accordingly, the involuntary–voluntary distinction is not as clear-cut as the conceptual foundation of intentional versus non-intentional retrieval (or indeed retrieval with or without cognitive control) suggests. As such, there are no involuntary/voluntary memories but a memory representation that is constituted by certain factors operating during retrieval, an attributional process. We see retrieval as a dynamic process consisting of interactions between different dimensions (e.g., intention, retrieval effort, memory accessibility, and monitoring processes) that all together make up the particular “flavour” of a retrieved memory.

We see familiarity as metacognitive; a feeling that is generated from the fluent processing of a stimulus.Footnote ⁶ It is a signal which guides memory processing – orientating us to information in the environment which we have already encountered. As such, it can be described as an epistemic feeling (Moulin & Souchay, Reference Moulin, Souchay, Perfect and Lindsay2014). Arango-Muñoz (Reference Arango-Muñoz2011) defines epistemic feelings as quick-acting intuitive processes, based on how things feel, and déjà vu is often used by philosophers as an example of an epistemic feeling (Arango-Muñoz, Reference Arango-Muñoz2014; Bortolotti, Reference Bortolotti2010; de Sousa, Reference de Sousa2009; Gerrans, Reference Gerrans2014). Epistemic feelings neatly map onto a set of metacognitive phenomena and paradigms that are better researched and understood than déjà vu, such as the TOT state (Greely, Reference Greely2021) and the feeling of knowing (FOK; Souchay, Moulin, Clarys, Taconnat, & Isingrini, Reference Souchay, Moulin, Clarys, Taconnat and Isingrini2007). These are instances of metacognitive awareness at the moment of retrieval failure (Fig. 1). When familiarity is high and retrieval is incomplete, this will lead to such feelings, even for the personal past. Critically, in the TOT (or the FOK) for the personal past, the experient must evaluate retrieval as plausible, since otherwise the experience would be one of conflict, possibly resulting in déjà vu.

Souchay and Moulin (Reference Souchay and Moulin2009) synthesised the existing proposals into a schematic representation about the relationship between familiarity and recollection (Fig. 2). Various similar proposals exist in the literature (see Hintzman & Curran, Reference Hintzman and Curran1994; Koriat & Levy-Sadot, Reference Koriat and Levy- Sadot2001; Reder, Reference Reder1987; Reder & Ritter, Reference Reder and Ritter1992).Footnote ⁷ It explains the search for information given a cue in the environment, and pertains most clearly to the FOK. In the FOK, a rapid initial feeling of familiarity with a cue triggers the attempted retrieval of episodic detail about the cue, and any associated information. In the absence of any contextual information, people can make accurate predictions of future recognition based on the strength of familiarity for the cue, and what “partial information” about the cue or searched-for-target comes to mind. Such research shows that even when information is not currently retrievable, outputs from the recognition memory system (in particular fluency) can be acted on metacognitively in order to accurately predict the likelihood of subsequent retrieval, suggesting the existence of a sequential arrangement as shown in Figure 2, which is why it is helpful to think of familiarity and recollection as lying along a continuum.

Figure 2.

Representation of the metacognitive evaluation of familiarity and recollection in recognition memory decision making.

In the context of IAMs, it is possible to retrieve information directly without feeling anything until the moment there is a fully formed mental representation. We would argue that this is due to there being very little delay between the first familiarity stage and the complete retrieval. Memory retrieval in this case feels spontaneous because the subject had both the feeling of familiarity and recollected specifics at approximately the same time. Familiarity is most strongly felt when we are not able to retrieve contextual specifics. In the butcher on the bus, familiarity is high but the recollected specifics are (momentarily) absent, whereas in déjà vu, the familiarity is similarly high, but a metacognitive evaluation favours the interpretation that the familiarity is false or impossible. Here we suggest for the first time that this is based on access to information about the personal past.Footnote ⁸

2.2 Autobiographical retrieval mechanisms

Here, we examine in more detail autobiographical memory retrieval processes with a focus on metacognition, and propose a threshold mechanism. According to Conway and Pleydell-Pearce (Reference Conway and Pleydell-Pearce2000; for later modifications, see also Conway, Reference Conway, Dere, Easton, Nadel and Huston2008, Reference Conway2009; Conway & Jobson, Reference Conway, Jobson, Berntsen and Rubin2012), autobiographical memory consists of a hierarchical network of interconnected nodes that differ in terms of their level of specificity. At the bottom are fragments of events with specific sensory details (e.g., details experienced when riding a bike for the first time). Such vivid information is further connected to super-ordinate levels of general events (e.g., riding a bike to work), common themes (riding, commuting to work), and, at the highest level, important periods in one's life (e.g., when I moved to France). The activation of autobiographical information spreads across the network resulting in the construction of a particular memory. Such activation may be elicited by different types of cues, internal or external. While voluntary autobiographical memories are the result of a top–down search process that eventually arrives at an episode, directly retrieved (i.e., involuntary) memories are thought to circumvent the search process and enter consciousness very quickly. Conway and Pleydell-Pearce argued that fragments of memory representations are constantly activated at the bottom level of the hierarchy, but the vast majority of such memories never reach consciousness due to being suppressed by constantly operating executive control system. Moreover, some low level and/or mundane information will be completely in line with expectancies (cf. Whittlesea's SCAPE) and will not be raised to awareness. Only some of these activated memories, especially those that are consistent with current self-goals, may occasionally get through this executive control mechanism and reach awareness.

This implies that IAMs occur as a result of two concurrent processes, namely spreading activation and inhibition that works against such activation (see Ball & Hennessey, Reference Ball and Hennessey2009). The retrieval of IAMs depends on absent or inefficient inhibitory control. For instance, when entering a hotel room for the very first time (i.e., we have never been to the city, in general, nor to this hotel chain), the environment is being continuously processed and automatically matched with any corresponding representation in memory. Aside of deliberative searches in memory cued by conscious thoughts (e.g., being asked by the receptionist if we have ever stayed at the hotel before), we propose that there are several possible outcomes, which lie on a continuum: (a) An involuntary memory may be triggered, (b) one may get only a general feeling of familiarity, or (c) the room does not trigger any retrieval process. The key premises delivered from this model are that (1) autobiographical memory information is constantly and automatically activated by cues and that (2) only some of them may enter awareness.

Given that IAMs occur automatically in response to incidental external and internal cues, one may ask why are we not constantly flooded by memories? According to the principle of cue overload, it is most likely that a cue will match several past events. Berntsen (Reference Berntsen2009) proposed a mechanism of cue-item discriminability, defined as “how easily a given cue isolates an item” (Rubin, Reference Rubin1995, p. 151 as cited in Berntsen, Reference Berntsen2009, p. 107; see also Norman & Bobrow, Reference Norman and Bobrow1979). That is, the more events that are associated with a particular cue, the less efficient this cue will be in triggering any one of them. Berntsen, Staugaard, and Sørensen (Reference Berntsen, Staugaard and Sørensen2013) examined the relationship between cue and memory and were able to predict the occurrence of an involuntary episodic memory based on a manipulation at encoding. Across four experiments, they found that involuntary episodic memories were retrieved more often in response to unique compared to repeated cues, in keeping with the principle of cue-overload.Footnote ⁹

The déjà vu field has yet to manipulate such variables, although a neuroscientific account relating déjà vu to novelty mechanisms has been advanced by O'Connor, Lever, and Moulin (Reference O'Connor, Lever and Moulin2010). At least in clinical samples, delusional beliefs of prior occurrence are more likely to appear in novel locations and events (i.e., environments with unique cues) than those that are familiar. Our prediction is that déjà vu would arise in situations where the cues are relatively unique, since in situations where there is cue overload, the attribution will be that any sense of familiarity is normal given the current situation.

It can be argued that cues that are unique and distinctive are infrequent in daily life and most cues map onto many different memories and events. Given how frequent IAMs are, it must be the case that other processes are at play. In response, it has been suggested that priming processes (i.e., increasing the activation of a memory information by prior encounter with the contents of memory representation) may not only increase cue-item discriminability allowing for efficient activation of a particular involuntary memory (e.g., Barzykowski & Niedźwieńska, Reference Barzykowski and Niedźwieńska2018b; Mace, Reference Mace2005) but also increase the familiarity of a given cue.Footnote ¹⁰ It is thus possible that on some occasions, a new event/setting may consist of cues that map onto many different past memories, or even that there is a relevant configural, contextual similarity between the current situation and some past events (exactly as proposed in the Gestalt Similarity hypothesis; Cleary et al., Reference Cleary, Ryals and Nomi2009). While these cues/contexts are not efficient in triggering any given memory, it is nonetheless possible that some of them may induce the feeling of familiarity, leading to the experience of déjà vu. This would be the case if the memory is vague, or non-specific: Associated with many previous events/information in the AM organisational hierarchy and unable to enter awareness. The activation is not strong enough to trigger a given memory, but it is strong enough to trigger a feeling of familiarity. This suggests that while on some occasions the memory content (e.g., being in a similar hotel room in Rome a few years ago) does not enter consciousness, leaving us without a straightforward reference to our past, the feeling of familiarity still appears. Conway and Pleydell-Pearce (Reference Conway and Pleydell-Pearce2000) assumed an inhibitory control mechanism suppresses irrelevant memory information, preventing it from reaching consciousness. Findings from several recent studies (e.g., Barzykowski et al., Reference Barzykowski, Radel, Niedźwieńska and Kvavilashvili2019b, Reference Barzykowski, Hajdas, Radel, Niedźwieńska and Kvavilashvili2021a, Reference Barzykowski, Hajdas, Radel and Kvavilashvili2022) do not provide full support for the existence of such a direct control mechanism,Footnote ¹¹ and favour an account based on a retrieval threshold (see Barzykowski & Staugaard, Reference Barzykowski and Staugaard2016, Reference Barzykowski and Staugaard2018; Barzykowski et al., Reference Barzykowski, Niedźwieńska and Mazzoni2019a).

In recognition memory decision making, activation/threshold models have long been the norm, stemming from signal detection theory (e.g., Banks, Reference Banks1970; Wixted, Reference Wixted2007). In autobiographical memory, it has recently been proposed that in order to reach consciousness, a memory must pass an awareness threshold that determines the minimum amount of activation for awareness. Entering awareness can be achieved either when a memory reaches levels of activation that is greater than a given threshold, or when the threshold is lowered by some additional factors. For activation, for instance, it may be easier for a phenomenologically pronounced and intense memories to pass the threshold because they may be especially good at drawing one's attention (see also Barzykowski & Staugaard, Reference Barzykowski and Staugaard2016, Reference Barzykowski and Staugaard2018; Barzykowski et al., Reference Barzykowski, Niedźwieńska and Mazzoni2019a). For the threshold level, it is argued that it may be momentarily modified by different factors (e.g., expecting memory to happen, placing the focus of attention on one's stream of thoughts), which would consequently modify the experience of otherwise less accessible memories. The threshold hypothesis states that, while both highly and poorly accessible memories can be retrieved either voluntarily or involuntarily, the processes operating during memory retrieval can influence the frequency of each type of retrieval by modifying the awareness threshold. Barzykowski and Staugaard (Reference Barzykowski and Staugaard2018; also, Barzykowski et al., Reference Barzykowski, Niedźwieńska and Mazzoni2019a) proposed these processes to be intention (i.e., wanting to retrieve a memory) and selective monitoring (i.e., expecting a memory to appear). This idea has been verified in several empirical studies (e.g., Barzykowski & Mazzoni, Reference Barzykowski and Mazzoni2022; Barzykowski, Skopicz-Radkiewicz, Kabut, Staugaard, & Mazzoni, Reference Barzykowski, Skopicz-Radkiewicz, Kabut, Staugaard and Mazzoni2023; Barzykowski & Staugaard, Reference Barzykowski and Staugaard2016, Reference Barzykowski and Staugaard2018; Barzykowski et al., Reference Barzykowski, Niedźwieńska and Mazzoni2019a, Reference Barzykowski, Staugaard and Mazzoni2021b). For instance, involuntary memories compared to voluntary memories were rated by participants as more phenomenologically accessible; namely, more clear, unusual, personally relevant, important, recent, and more frequently rehearsed in the past (e.g., Barzykowski & Staugaard, Reference Barzykowski and Staugaard2016; Barzykowski et al., Reference Barzykowski, Niedźwieńska and Mazzoni2019a). These findings corroborate the idea that when not being actively engaged in memory retrieval, a memory has to be phenomenologically sound to pass the threshold. Also, Barzykowski and Niedźwieńska (Reference Barzykowski and Niedźwieńska2016) demonstrated that involuntary memories retrieved in a selective monitoring condition (i.e., when one is expecting a memory to appear) were less clear, less detailed, less vivid, and elicited less intense physical reactions compared to memories retrieved without selective monitoring. In another study, selective monitoring decreased the personal importance and emotional intensity of memories (Barzykowski & Staugaard, Reference Barzykowski and Staugaard2018). Finally, these results pertain not only to the phenomenological characteristics of memories but also to their objectively observable descriptions (Barzykowski et al., Reference Barzykowski, Skopicz-Radkiewicz, Kabut, Staugaard and Mazzoni2023). When looking at the descriptions of memories, on average involuntary memories were indeed more accessible (i.e., rated by independent judges as having more indicators of memory accessibility such as being scored higher on the emotional impact scale, etc.) than voluntary memories and intentionality lowered the awareness threshold and increased the number of low-accessibility memories.

Note that these threshold processes are metacognitive, about expectations and intentions. Our proposal is that in complete retrieval, there is little need to reflect too much on the process of retrieval: Information comes to mind, and is recollectively experienced, whether or not it was sought for. Naturally, the recall of contextual specifics may even justify to the experient why the information has come to mind, for example, “it was in this post office where I got my foreign currency last year.” However, in cases where retrieval is incomplete or low-level signals violate expectations (i.e., finding something familiar in a novel location), then intense feelings are generated.

Autobiographical retrieval is therefore determined by a complex interplay between factors. Intention and monitoring can be conceived as processes that enable access to otherwise less accessible memories. The most important assumption of the threshold hypothesis is that because of certain phenomenological properties (e.g., emotional intensity, clarity, vividness, personal relevance, unusualness), some mental contents may be especially good at drawing one's memory-related attention and, thus, they may pass the awareness threshold more easily and be more likely to be reported. However, on occasions where we may not be aware of a given memory content, but where activation is sufficient, we can still feel familiarity. If the familiarity is logically impossible and unexpected, it will trigger a search in memory to explain the feeling. Thus, both déjà vu and IAMs start from the same point in shared retrieval processes.

We turn our attention to how these processes can be split down into their component parts. Autobiographical memory retrieval has the following four stages (e.g., Barzykowski & Mazzoni, Reference Barzykowski and Mazzoni2022; Wilckens, Erickson,& Wheeler, Reference Wilckens, Erickson and Wheeler2012): Pre-retrieval, retrieval, post-retrieval, and retrieval outcome report stages.Footnote ¹² The pre-retrieval stage is associated with any cognitive processes that either facilitate or impair retrieval. For example, during this stage, an individual may be in “retrieval mode” in which “the cognitive system is prepared for or expects memory construction and recollection” (Conway, Reference Conway2001, p. 1379). The effect of priming would occur here, enhancing the likelihood that the memory will be triggered and/or will enter a person's awareness. The retrieval stage relates to the forming and developing of an autobiographical memory, but without explicit self-reflection; namely, a given memory might have been formed but one may not be explicitly aware of it yet. During this stage, a memory is triggered by and/or accessed via a given cue. Once the memory is formed, during the post-retrieval phase, people may become aware of having the memory in mind. Thus, this stage relates to the ability to, for example, extract autobiographical content from the stream of consciousness to explicitly become aware of having a memory that is autobiographical. One is fully aware that an autobiographical memory was actually retrieved, and in voluntary retrieval, if the memory content meets the given criteria, then the search may be terminated. In the last stage, the retrieved memory may be shared with others and reported by giving a verbal account of the content.

This four-stage account highlights two main factors, the first being intentionality during the pre-retrieval phase, the second being a metacognitive process (see Koriat & Goldsmith, Reference Koriat and Goldsmith1996; Mazzoni & Kirsch, Reference Mazzoni, Kirsch, Perfect and Schwartz2002). Koriat and Goldsmith, inspired by signal detection models, describe a situation whereby the subject can decide whether to report retrieved information (p. 181): “people monitor the subjective likelihood that an item of information that comes to mind is correct and then apply a control threshold to the monitoring output in order to decide whether to volunteer that item or not.” Their emphasis, necessitated by the use of a forced/free report recall procedure (see Goldsmith & Koriat, Reference Goldsmith, Koriat, Gopher and Koriat1999), is on the participant evaluating the quality of information before choosing to output it, whereas our proposal is that internally such a process also exists to motivate and regulate the search for information for the subject.

When looking at these retrieval stages, we argue that both déjà vu and IAMs are natural forms of involuntary and spontaneous processing of cues that can be found in one's surroundings, and that memory information when retrieved is not only accompanied by the exact memory content but also the feeling of familiarity that, importantly, may arise and/or be experienced separately from each other. In particular, déjà vu can be thought of as arising at the retrieval stage, and before the person is aware of having any content in mind.

A critical issue is to understand why in some cases, retrieval is incomplete and déjà vu arises rather than the retrieval of an IAM. Here, we advance two hypotheses which are not mutually exclusive and may map onto the top–down/bottom–up descriptions of déjà vu in the literature. These are proposals based on cue familiarity/trace strength and the metacognitive evaluation of cues in the environment. It is beyond the scope of this paper to delve too deeply into retrieval failure per se on a mechanistic level, but it is important to note that the dynamic patterns of inhibition and activation seen in episodic memory (e.g., Anderson, Bjork, & Bjork, Reference Anderson, Bjork and Bjork1994; for a review, see Kuhl & Wagner, Reference Kuhl, Wagner, Berntson and Cacioppo2009) are likely to contribute to the failed retrieval referred to here. For instance, it is known that competition between similar cues leads to an inhibition process whereby the retrieval of one memory actively inhibits the retrieval of another. It is not impossible that information is not simply forgotten or lost according to this process, but that it gives rise to phenomenological states such as referred to here, something that would warrant further investigation.

Our first hypothesis is that there is only weak activation of the associated cues and memories from the environment, and thus there is not sufficient activation for any content to enter into consciousness. This is not unlike the experience of information in the TOT (developed in detail below), where knowledge or lexical information is partially activated and frustratingly will not come to mind, even if we are aware of its existence. In this weak version of déjà vu, a parallel is seen in autobiographical memory. A cue reminds us of something stored in autobiographical memory, but the cue is undetected and does not yield sufficient activation, giving a feeling of familiarity without a known source. Accordingly, it would be expected that déjà vu will most likely occur in a context that is rather familiar/known, in contradiction to what we proposed above when discussing cue discrimination. Whilst Brown and Marsh (Reference Brown and Marsh2008) compared mundane and distinctive scenes in their experimental analogue of déjà vu, finding that mundane scenes were more likely to generate feelings of having visited a place before, this rather common-sense interpretation of déjà vu in mundane, familiar locations has yet to be tested (although questionnaire studies suggest that in healthy groups, déjà vu happens more in familiar than novel contexts; see Moulin, Reference Moulin2018). One possibility to test this idea in experimental analogues of déjà vu is to further probe the déja vu once experienced. For instance, one may amplify the weak activation of the associated cues and memories by lowering the awareness threshold and/or drawing the attention to the possible sources of déjà vu. This way, the feeling of familiarity caused by weak activation of memory content could be overcome leading eventually to memory content retrieval; that is, reinstating cues experimentally may shift the experience from déjà vu to a memory of the past. This weak activation account also suggests that the déjà experience may be more likely to arise when one is engaged in attention demanding activities where ongoing task performance elevates the awareness threshold making weakly activated memories difficult to enter consciousness.

Our second hypothesis centres on metacognitive evaluations and posits that déjà vu is felt precisely because it is unexpected, aligning with the top–down view. In certain contexts, the expectation of familiarity should be low. In such cases, any involuntary process of retrieval, especially when it does not result in any retrieved information will be unexpected and will lead to a metacognitive evaluation which will be raised to consciousness. Until this feeling of familiarity is resolved or explained, it will grow and the experiment may well search for a reason “why” they have this feeling. Experimentally, it would be relatively simple to manipulate expectations about the likelihood of finding information familiar. In Cleary-type recognition without identification experiments, the idea that expectations about familiarity lead to the feeling of conflict should mean that déjà vu is more intense and more frequent when the expectation is lower. This could either be manipulated by giving explicit instructions to this effect to participants or manipulating the amount of similarity between the initial scene and its analogue in the recognition phase. It would be intense when the context is surprising, such as in the unrelated experiments conceit (e.g., Lakin & Chartrand, Reference Lakin and Chartrand2003): A first study phase is conducted in the context of one task and then the test phase is presented as a separate experiment. This design could assess the role of the experimental context (either presented as the same experiment or a separate experiment) on the report of déjà vu.

There is little existing research to discriminate between our two hypotheses, but Urquhart and colleagues successfully manipulated both the novelty and the activation (which they described as “familiarity”) of the critical non-presented stimulus. For novelty, they manipulated how specific the letter string was in referring to the critical lure (e.g., monitor for words that begin “SLE” in the list, vs. monitor words that begin with “B”). In the test phase, the string SLE was highly novel because it did not correspond to any items presented in the study phase, whereas the less specific “B” string applied to two words in the studied list, blanket and bed. For activation of the cue, they used results from previous versions of the DRM task: Some lists of associates are more likely to activate the critical lure than others: Thus they had more strongly and more weakly activated critical lures. In their experiment, the recognition memory results replicated the expected patterns. However, for the rates of experiencing déjà vu, there were no effects or interactions involving activation: The strength activation of the critical lure in the presented lists had no bearing on the generation of déjà vu. Thus, in this case, a weakly activated associate was not more likely to generate déjà vu than a more strongly activated one. The effect of novelty was however significant: Participants were more likely to experience déjà vu when the critical lure was unambiguously impossible to have been viewed in the study phase due to its unique string.

Our explanation of déjà vu as being “unexpected” would involve processes downstream, corresponding with the post-retrieval stage. During this stage, one becomes aware of having either an involuntary memory in mind or having a feeling of familiarity. More precisely, during both the post-retrieval and final stage, the unexpected feeling of familiarity without having any particular memory in mind may start metacognitive processing (e.g., thinking of plausibility of the event, etc.) leading people to explicitly experience déjà vu, especially in cases when the current situation is judged as novel and, crucially, “encountering it in the past is in fact highly implausible” (Mazzoni & Hanczakowski, Reference Mazzoni, Hanczakowski, Higham and Leboe2011, p. 104). This would be the conflict described by Urquhart and O'Connor (Reference Urquhart and O'Connor2014): Familiarity is generated bottom–up from the environment, but higher-level autobiographical belief or knowledge opposes the feeling. This should be a rare occurrence, since as Conway (Reference Conway2005) proposes, autobiographical remembering is typically coherent with current goals and processing, but this would be a situation where the experience is incoherent. An interesting way to test this idea would be to re-evaluate reported déjà vu experiences in questionnaires according to self-reports of implausibility and novelty, on one hand, and also their stability across time. More precisely, although a given situation may provoke déjà vu, it may do so because in fact, there is not sufficient access to information relating directly to a past situation. However, such access may be restored later, and this additional information may change the evaluation of what was initially experienced as déjà vu, raising the question of the stability of déjà vu over time. In the only work related to this idea, Milton, Butler, and Zeman (Reference Milton, Butler and Zeman2011) report a case of epileptic amnesia where the spontaneous retrieval of previously inaccessible autobiographical memories was “heralded” by a period of déjà vu. Finally, we may also hypothesise that if déjà vu experiences arise in response to the unexpected feeling of familiarity without having any particular memory in mind, then increasing the expectancy of a feeling of familiarity in response to the current situation should lower the frequency of reporting déjà vu experiences.

In our proposal, memory retrieval is not an all-or-nothing process, and numerous forms of partial or incomplete memory retrieval exist.Footnote ¹³ Notably, memory content and a feeling of familiarity are separable (e.g., Ryals & Cleary, Reference Ryals and Cleary2012). Studies on the TOT phenomenon (e.g., Brown, Reference Brown1991; Heine, Ober, & Shenaut, Reference Heine, Ober and Shenaut1999; Schwartz, Reference Schwartz2001, Reference Schwartz2002, Reference Schwartz2011) demonstrate that on some occasions while we cannot access memory content (e.g., trying to recall the last time we were in cinema), we have a strong feeling of familiarity, with the attribution that this memory exists somewhere. Whilst this phenomenon is usually discussed in terms of linguistic or semantic processes, our proposal is that it occurs for the personal past too. Thus, it may occur for a given cue (e.g., a colourful beach ball): There is a strong feeling of that there is something in our past relating strongly to this cue or that this cue reminds us of something that cannot be yet recalled.

In Figure 1, we propose a flow chart summarising the possible memory phenomena on three dimensions: (1) Retrieval intentionality (involuntary vs. voluntary), (2) memory content accessibility (the presence vs. absence of memory content), and (3) phenomenological familiarity (feeling familiar or not). It shows that TOTs/FOKs are the result of unrecalled content despite having strong feelings of familiarity. They can arise both when retrieval is intentional and unintentional, and in this way, they act as a bridge between more controlled and less-controlled processes. That is, an involuntary retrieval may provoke a sense of familiarity, that once detected will lead to an intentional retrieval or at least a wilful memory search. In this way, metacognition plays a functional role in recruiting intentional retrieval processes to something for which retrieval was cued in an involuntary manner.

Figure 1 makes the case that familiarity is something that is experienced (or not) on the way to retrieving content, and that there is something of a sequential relationship between feelings of familiarity and the retrieval of content. We think that this is necessarily the case in failed retrieval, where there is a phenomenological response despite there being a lack of content, but this is not necessarily the case in fluent, successful retrieval where the presentation of content to consciousness is so rapid as to feel direct. The organisation may be sequential, but it is also recursive: Feelings of familiarity and content retrieval may jump back-and-forth in a manner inspired by the large-scale synchronisations of activation in the DMN; more precisely, this phenomenology-content loop is inspired by the interactions between retrieval and control networks stated in various models of memory retrieval described above (e.g., Eichenbaum, Reference Eichenbaum2017; Irish & Vatansever, Reference Irish and Vatansever2020; Sekeres et al., Reference Sekeres, Winocur and Moscovitch2018).

Our view is that one underlying retrieval process yields two different experiences: Phenomenological familiarity is experienced when retrieval is incomplete, but it is not experienced when retrieval is fluent and routine. That is why our friends do not feel familiar, since we are able to rapidly retrieve specific information about them. That is, high trace strength memories can be accessed directly without giving rise to phenomenological fluency. In Figure 1, voluntary memories can be retrieved with familiarity or more directly, but involuntary memories never give rise to familiarity during retrieval. The study of déjà vu and IAMs helps us arrive at this conclusion since we need to understand why there is sometimes unbidden retrieval of the personal past without giving rise to any particular phenomenology, and on the other hand, in déjà vu, there is the opposite: An intense feeling of familiarity with the conclusion that the familiarity is erroneous, and devoid of content.

In sum, both déjà vu and IAMs are products of involuntary retrieval but, as presented in Figure 1, while IAMs are accompanied by memory content, déjà vu lacks the access to memory content and the feeling of familiarity is judged as implausible (if judged plausible, then one would most likely experience FOK/TOT).Footnote ¹⁴ An interesting consequence emerging from Figure 1 is the possibility of differentiating involuntary and voluntary instances of FOK/TOT, which for now in our characterisation are the same experience whether or not they are encountered in intentional or unintentional retrieval. Although they have not been explicitly contrasted with each other in previous studies, such a distinction seems to us to be possible. Finally, all these phenomena (involuntary/voluntary memories, déjà vu, FOK/TOT) can be described along the same retrieval dimensions and so the integration of déjà vu into autobiographical retrieval has opened up new way of thinking about and classifying the range of possible retrieval experiences. We discuss this now, pointing out where research is needed.