Further advancing theories of retrieval of the personal past

Abstract

In our target article, we presented the idea that involuntary autobiographical memories (IAMs) and déjà vu may both be based on the same retrieval processes. Our core claim was thus straightforward: Both can be described as “involuntary” or spontaneous cognitions, where IAMs deliver content and déjà vu delivers only the feeling of retrieval. Our proposal resulted in 27 commentaries covering a broad range of perspectives and approaches. The majority of them have not only amplified our key arguments but also pushed our ideas further by offering extensions, refinements, discussing possible implications and providing additional empirical, neuroscientific and clinical support. The discussion launched by the commentaries proves to us the importance of bringing IAMs and déjà vu into mainstream discussions of memory retrieval processes.

R1. Introduction

The advancement of knowledge is, among other things, about putting forward ideas for discussion. Our main goal was to launch a discussion about the possibility that two seemingly different phenomena, involuntary autobiographical memories (IAMs) and déjà vu, may have the same neurocognitive origin and point to common retrieval processes. Our main claim was: IAMs and déjà vu are two forms of spontaneous experiences derived from autobiographical memory processing. If nothing else, we wanted to bring déjà vu and involuntary memory phenomena into mainstream discussions of memory retrieval processes.

We received 27 commentaries. Let us thank the commentators for their insightful comments, suggestions and thoughts. They allowed us to reconsider new perspectives that ultimately advance the theoretical proposal. On occasion, as developed below, we agree to disagree, especially with those which argue against studying IAMs and déjà vu together.

At its core, our proposal was about the interaction between the environment and internal representations of the personal past, and two key theoretical concepts underpin our thinking: The roles of cues in retrieval and familiarity. The role of cues was mentioned in many commentaries. For instance, Cleary, Poulos, & Mills (Cleary et al.) propose a mechanism of cue familiarity detection. Depending on the strength of the cue–memory overlap, the resulting familiarity “signal” may pass the critical threshold directing attention inward to initiate a memory search. This notion advances not only our framework but theories of autobiographical retrieval in general. Familiarity and neuroscientific mechanisms of retrieval also make up a large part of the commentaries received.

We also made the claim that to understand déjà vu and IAMs, the processes involved in memory retrieval and in conscious experience appear to be separable, something emphasised in Schwartz & Pournaghdali's commentary. We have recently expanded on the phenomenology of retrieval processes (Moulin, Carreras, & Barzykowski, 2023), including discussion of the tip-of-the-tongue experience (which was largely cut from the target article due to space constraints). Our point is that the brain is constantly trying to make sense of its inputs: In memory, we propose that the phenomenology of retrieval points to there being an attributional system which oversees the function of the temporal lobe. To organise our responses around these key concepts, we discuss conceptual issues, notably the role of cues in retrieval and the definition of familiarity, and debating the very definition and characteristics of the phenomena. Finally, we turn our attention to the neuroscientific aspects of our conceptualisation of déjà vu and IAMs in a second section before attempting to defend our continuum in the conclusions.

R2. How to conceptualise déjà vu and involuntary memories?

Kvavilashvili & Markostamou focus on cues as a means of explaining two issues: Temporal orientation and the frequency of occurrence. They propose that IAMs seem to be triggered by an easily identifiable cue, whereas déjà vu seems to be more dependent on less specific cues (e.g., the situation as a whole, contextual cues, complex constellation of cues, situational gist properties). Non-specific cues may then activate not the fragments of sensory-perceptual experiences (that are part of the autobiographical memory knowledge system) but rather the representations of previously experienced scenes (that are part of long-term perceptual-representation system). We welcome this suggestion and agree that déjà vu may emerge through several not mutually exclusive paths, and this may be one possible route. However, we consider that activating information from any long-term perceptual-representation system would not explain the intense feeling of self. There is a difference between saying “this reminds me of something” and “this feels like I have been here before.” Therefore, there is space to further advance this idea by explaining how such activation leads to having a “personal past” component.

Robins & Afifi argue that our unified account depends upon how easily concepts from the study of voluntary memory transfer to involuntary phenomena. We agree that cues may differently be processed in involuntary and voluntary retrieval, but we do not think that these differences significantly complicate our framework. While the authors say that “the cue can be identified prior to and independently of the memory it activates,” this is only true in successful recall of a sought-for memory. Moreover, it seems that this falls foul of the Norman and Bobrow (1979) conundrum which we reproduced as a footnote in the target article. In an experiment, or with a degree of introspection, it is easy for an expert observer to pinpoint a cue. In a naturalistic setting, the definition of what the cue was is arguably post-hoc and is as subjective as the contents of retrieval, as Robins & Afifi state: “a cue can only ever be identified in retrospect.”

In voluntary retrieval, we have an explicit idea of what we want to recall, but whether a cue will or will not be successful in retrieving it is not a given. We argue that the same is true for involuntary remembering, as we can never be sure whether something in the environment will or will not trigger a given memory, although on occasion a cue is highly efficient in triggering a given memory, for example, intrusive, recurrent memories triggered by easily identified stimuli. We state that retrieval is an iterative process, and what starts as involuntary retrieval may cross over into voluntary retrieval and vice-versa (see also Barzykowski & Mazzoni, 2022).

Evidently, involuntary retrieval is less predictable, as it is spontaneous and automatic. However, we do not agree with Robins & Afifi since we propose that people can be aware of the cues triggering memories prior to a complete retrieval. We take issue with the idea that “there is also no sense to be made of the idea of a cue failing to activate a memory during involuntary retrieval. If nothing is activated, then there is no cue – and also no memory.” When saying that a cue failed to activate a memory during involuntary retrieval, we do not refer to a situation in which nothing is activated but that something is insufficiently activated. Not having memory content does not mean per se that some stages of retrieval processes did not happen at all. We propose that there is a prior cascade of processes resulting in retrieval of content: There are instances where there is a cue, but nothing is retrieved. We do not think this is contentious. Detecting something in the environment as familiar based on an evaluation of fluency does not imply retrieval or the access of content, and yet we are able, we think, to identify something as a cue from this feeling, even if nothing is “retrieved.”

Our characterisation of a continuously active memory system trying to make sense of its inputs necessitates that anything in the environment is potentially a cue. For the experient, there is not some philosophical division of the environment into cues and non-cues, but a series of activations and associations between the internal and external world which may or may not enter consciousness.

We now turn our attention to familiarity. A number of commentators interpret familiarity in a different manner from that conceptualised in the article. At least three conceptualisations of familiarity exist (as noted in a footnote by Whittlesea and Williams [1998, p. 141]):

One is that a person has actually encountered a stimulus (or even one like it) previously. … A second is that the person has knowledge about a stimulus that permits them to perform appropriately toward an object, without necessarily having an accompanying feeling of having experienced that stimulus previously. … A third is the subjective feeling of having encountered a stimulus on some previous occasion, whether one actually has or not.

Pan & Carruthers helpfully differentiate two types of familiarity. The first type is phenomenological and arises in response to a retrieval failure. The second is akin to that in the recognition-memory literature referring to “a decision-making process based on feelings of fluency.” These definitional issues are critical, and we emphasise that our framework focuses on phenomenological familiarity. A number of commentators point to this lack of a solid basis on which to construct our argument. Stendardi, Basu, Treves, & Ciaramelli (Stendardi et al.) claim that the experience of déjà vu is fundamentally different from other forms of familiarity experience, including the butcher on the bus which we use as a reference point, because it is based on a configuration of a complex set of factors, and not a single item. As such, they invoke a framework of compositional memory formation (Ryom, Stendardi, Ciaramelli, & Treves, 2022). They propose that déjà vu is an incomplete memory state (and thus far our accounts do not differ) but that it occurs when “some” familiar items from a compositional memory (a rich representation of complex scene from the personal past) are activated (see their figure) in the neocortex, but where a full memory is not retrieved from the hippocampus.

The critical distinction between their proposal and our original account would be that a multitude of cues aiming to reconstruct a rich configuration of information is critical, whereas we were somewhat neutral as to the number of elements which might make up a cue. We suppose that differentiating between their elaboration and the initial proposal would depend not only on how you define familiarity, but also how you define “cue.” The grain size of the information required to trigger déjà vu or an involuntary memory is open to debate. Even if the experient in the end identifies any one element as critical, we are sure that cues and representations of the personal past are indeed as complex as Stendardi et al. point out. It is of interest that accounts of over-extrapolation of familiarity for specific cues to the whole scene or “single-element familiarity” has long been proposed as a mechanism of how déjà vu is generated (e.g., Leeds, 1944).

Sikorski & Sitek propose a comparison with misidentification syndromes and familiarity. However, the typical conceptualisation of familiarity in misidentification syndromes is as objective and not phenomenological. As such, when they write “While in Capgras syndrome access to memory content is preserved (patient's proxy face is correctly recognised) [the] feeling of familiarity is missing (the proxy is perceived as an imposter)…” it seems that whether or not this is a genuine first-person “feeling” or an attribution imputed by an observer is a critical question. An important issue therefore to address is whether there is a disorder of phenomenological familiarity in such misidentification syndromes. Whilst we take issue with their implied definition of familiarity, they raise an important point: The role of emotion, proposing that “familiarity has its emotional aspect and cannot be reduced to (meta)-cognitive components.” We think that this is probably more the case in delusional misidentification than it is day-to-day attributions of past experience, but it is something which warrants further investigation. There is sometimes an emotional response to déjà vu and involuntary memories, and moreover, familiarity is invoked in various types of emotional responses (as demonstrated in the mere exposure effect, e.g., Kunst-Wilson & Zajonc, 1980).

A further consideration is that of gist, which somewhat overlaps with the idea of a non-specific feeling driven by general information as implied in some conceptualisations of familiarity. Pan & Carruthers describe a system with recollection-based identification of the gist of the current scene or event. In agreement, Moulin et al. (2023) propose that a retrieval model is created at the moment of accessing a memory which includes a general model (i.e., gist) of pertinent information from various stores including a metacognitive template of what can be expected from retrieval. In the target article, we did not explicitly discuss the processing of gist information, but this general sense of a scene or event is something captured in the Gestalt account of déjà vu (e.g., Cleary, 2008). Clearly, when cues are not strong enough to sufficiently activate a given memory, there still may be sufficient activation for a gist identification.

With these comments about gist as a specific mechanism in retrieval, we see similar conceptual issues about retrieval schema and the interaction with semantic structures identified in the commentaries by Irish, Renoult & Debruille and Sikorski & Sitek. Renault refers to jamais vu, which has been described as the opposite of déjà vu, whereby an objectively familiar stimulus feels as being unfamiliar (Brown, 2003). It seems to be even less frequent than déjà vu in daily life, although it proves relatively easy to reproduce in the laboratory (Moulin et al., 2021). It can be provoked by repeatedly writing a word until it loses its meaning, and so at the time of writing, it seems to be more of a semantic phenomenon than one about the personal past, and as such it escapes our interest here. However, as Renoult & Debruille point out, there is a need to better consider semantic networks and schema more generally in our thinking: Either as a repeated context, a gist representation or even the use of personal semantics to act as a plausibility check.

Indeed, Morales-Torres & De Brigard raise the possibility of a relationship between déjà vu and “nonbelieved memories.” If we organise our concepts around two dimensions: (a) access to content (full/absent) and (b) feeling of familiarity (minimal/maximal), we think that non-believed memories differ from déjà vu in the access to content dimension. While déjà vu lacks content, leaving us only with a feeling of familiarity, the non-believed memory has content. Importantly, both are believed to be untrue. In this regard, we agree that non-believed memories – also an unusual and infrequent phenomenon – fall out of our way of thinking about content and feelings (see Moulin et al., 2023, for an explanation of this idea), but it is possible that semantic structures play a greater role in the identification of implausibility in non-believed memories.

To best tackle these issues (definitions of familiarity, involvement of gist and semantic retrieval), we should draw on the neuroscientific literature, and restrict our thinking to known mechanisms of familiarity. Recently, the literature has converged on two types of familiarity, both of which can accommodate our framework, echoing the Pan & Carruthers commentary: One which is an attribution based on fluency and the other which derives from perceptual or conceptual overlap (Montaldi & Kafkas, 2022). We explore in more detail these ideas with reference to the brain (see below).

A number of commentaries extend our work beyond memory systems. Our focus was on memory, but it was not our intention to imply that only memory processes are important for understanding IAMs and déjà vu, since a number of other cognitive operations come into play in our characterisation of a constantly active process surveying the environment. We are thus in agreement with Vannucci & Hańczakowski who propose that attentional factors (e.g., monitoring the stream of consciousness) may affect the awareness threshold. We also find their idea appealing that attention may be a “‘switch’ mechanism responsible for whether common retrieval processes may result in such different states.” Similarly, attentional factors may contribute to the differences in phenomenological characteristics between, for instance, involuntary and voluntary memories (see Barzykowski, Niedźwieńska, & Mazzoni, 2019; Barzykowski & Staugaard, 2018). We are not sure whether attention plays a leading role but we surely agree that attention has an important part to play in both the pre-retrieval stage (i.e., how either peripheral or central/focal potential cues are processed) and the post-retrieval stage (i.e., how it changes the threshold of awareness).

Other commentators consider the very definition of the phenomena being discussed, and given their subjective nature, it is unsurprising that commentators develop, extend and contest our description of these phenomena, especially déjà vu. Some of these commentaries draw out differences between déjà vu and involuntary memory using specific characteristics. One such characteristic was the frequency of the experiences. Kvavilashvili & Markostamou invite us to wonder “why déjà vu experiences are so rare compared to the frequent occurrences of IAMs.” In preliminary findings (Zareen, Ahraf, Barzykowski, & Moulin, in preparation) from a large scale-survey of respondents (mean age = 22) in France (n = 178), Pakistan (n = 370) and Poland (n = 312), we found that déjà vu was experienced fewer times per week than IAMs (median values respectively 0 and 1–2 times) and fewer times per year (median values respectively, 3–5 times and 11–20 times).¹

Why would there be these differences in frequency of occurrence? One possibility is that there are different types of IAMs. Curot, Servais, & Barbeau (Curot et al.) remind us of this issue. They explicitly ask whether there are different types of IAMs with eloquent reference to Marcel Proust's madeleine. We find this extremely relevant, and existing research can contribute to this discussion. According to Mace (2004) while 30% of all memories reported were triggered by sensory/perceptual cues, only 3% of them were triggered by Proustian-type sensory cues (i.e., tastes or smells). Curot et al. refer to a study by Ball and Little (2006) showing that IAMs retrieved without easily identified cues were a minority. Curot et al. thus argue that these types of IAMs and déjà vu actually share similar prevalence. First, this clearly shows that, while IAMs may be frequently triggered by cues, corroborating their relative high frequency, we overlook IAMs that were retrieved without any identifiable cue or by a “rare” type of cue. As such, déjà vu and IAMs may be more similar in their frequency (also phenomenological characteristics like surprise, or strength) when we look at IAMs triggered in similar situations. Therefore, we should look at IAMs triggered by certain types of cues to make an adequate comparison with déjà vu. Our focus on the cues and different types of IAMs may increase not only their conceptual but also empirical proximity.

Putting aside these issues, let us explore why déjà vu does not happen more frequently. This is exactly the question that has often been asked in IAM research, which we identified in the target article: Why are we not constantly flooded by IAMs in daily life? It seems to us that the frequency question pertains then to both déjà vu and IAMs. An interesting avenue proposed by Morales-Torres & De Brigard is the possibility to look at the extent to which “novelty is proportional to the mismatch between the stimulus and its context” and the prediction error in which “novelty relates to the divergence between a prior likelihood and a posterior probability.” It seems to us that these possibilities may be relatively easily examined in laboratory settings, and although empirical approaches to déjà vu have considered generating conflict (e.g., Urquhart, Sivakumaran, Macfarlane, & O'Connor, 2021), it would also be of interest to manipulate people's expectations of what they should experience. If you go into an experiment expecting to have déjà vu, it should be less likely to occur if surprise is its key phenomenology, which may be what renders it so difficult to produce anything like genuine déjà vu in an experimental situation!

We do not feel obliged to explain the infrequency of déjà vu as compared to IAMs, so long as we have a sensible, theory-based explanation of how it occurs in the memory system. As an analogue, lightning strikes are relatively infrequent but we understand the conditions by which they are produced. When wanting to explain their infrequency, we fall into a circular (but completely reasonable) argument: Lightning strikes are infrequent because the conditions which lead to their occurrence are infrequent. If déjà vu and IAMs are infrequent, then it is because happily, the cognitive system is such that the conditions by which they are produced are also infrequent.

Andonovski & Michaelian go beyond the frequency issue to claim that our proposal fails to account for the strangeness of déjà vu. We wonder why explaining this one phenomenological property would serve as sine qua non to explain déjà vu properly by the reference to typically occurring processes in autobiographical memory? First, the phenomenological strangeness might simply fall out of the frequency issue. It feels strange because it is rare. Second, per analogiam, we do not have concerns about integrating involuntary memories into autobiographical memory research despite the robust phenomenological differences between involuntary and voluntary memories.

At this stage, it is important to look for mechanistic reasons for similarities and differences. Similar work has been carried out over the years in involuntary and voluntary memory research; an attempt to explain possible mechanisms resulting in phenomenological differences between involuntary and voluntary memories. A recently proposed answer was the threshold idea which formed the core of the target article. Briefly, retrieval favours phenomenologically “juicy” memories, which we can consider along different dimensions: Attentional factors, memory accessibility, retrieval intentionality, attribution and metacognitive concerned processes, etc. There is, at least for us, no reason to argue why this would be different in retrieval of déjà vu and IAMs.

It is possible that the phenomenological strangeness felt in déjà vu is what makes it pass the awareness threshold. It may be, as suggested by Cleary et al. that depending on the strength of the cue–memory overlap, the resulting familiarity signal has to pass the critical threshold directing attention inward to initiate a memory search. In a retrieval failure, it would bring phenomenological familiarity to one's attention, because it was strange, unexpected or inappropriate. We have accounted for such a possibility in our proposal, even if we did not use the word “strange.” Bastin further expanded on this, proposing that an unexpected content retrieval attempt clashes with feelings and sensations resulting from that retrieval failure that cannot be easily understood or explained. This leads to attribution processes becoming “hyperactive” resulting in strong feelings such as strangeness.

Andonovski & Michaelian also conclude that our proposal does not account for the sudden onset of déjà vu. We did not argue against a sudden onset of déjà vu, and so apologise if we gave that impression. However, over the course of experiencing déjà vu and certainly reflecting metacognitively upon it, it can evolve and change in intensity. In a preliminary analysis (again from Zareen et al. as above), we asked participants to categorise their experience using a series of dichotomies, one of which was to ask participants if the experience came on suddenly or gradually over time. Both déjà vu and IAMs were seen as coming on suddenly by the majority of participants (85% for déjà vu and 62% for IAMs). Interestingly, for tip-of-the-tongue experiences, the majority of participants also reported it as sudden (67%), in contrast with Andonovski & Michaelian's comment. We are not sure therefore that suddenness is a defining feature of déjà vu, and it does not seem in any case to distinguish IAMs and déjà vu.

We can group their criticism together with a number of other points about micro-phenomenology, and state that we welcome further research into factors such as strangeness, surprise and abruptness. The first step is to accept or reject these somewhat subjective descriptions of déjà vu with reference to empirical studies, and consensual, population-based descriptions, as we have done in the Zareen et al. study cited here. Perhaps the strongest contribution to this phenomenology debate comes from Perrin, who proposes that we should further consider déjà vécu as a variety of cognitive experience. Déjà vécu can be thought of as a recollection-based recognition memory error, the direct sibling of déjà vu (see Funkhouser, 1995; Moulin, Conway, Thompson, James, & Jones, 2005; Perrin, Moulin, & Sant'Anna, 2023, for definitions), whereby we feel that the present moment is not only familiar, but could lead to something being remembered. It seems reasonable to posit that a recollection form of déjà vu exists given the mechanisms we have proposed in our framework, but we were hesitant to add it at this stage, since empirical work attesting to the existence of the two types and the reliability of the definitions of the familiarity and recollection forms is severely lacking.

Bastin (see also Kvavilashvili & Markostamou) propose that IAMs and déjà vu differ in temporality: Déjà vu is retrieved “with some glimpse into the future with the feeling of prescience.” Similarly, Addis & Szpunar suggest that our framework may allow us to better understand related phenomena such as simulations of the future. This leads them to an intriguing speculation; namely, that some instances of déjà vu may be a result of not only weakly activated representations of the past, but also weakly activated (by goal-relevant cues) representations of the future, which they call “déjà vu for the future,” something which may be worthy of future examination in surveys: How often does déjà vu involve a glimpse into the future? It seems to us, that whilst interesting, such prescience is not a ubiquitous feature of this experience (for a single case description of prescience in déjà vu, see Curot et al., 2021), and that future research needs to consider the micro-phenomenology of these experiences in line with Addis & Szpunar's hypothesis.

A further conceptual concern is the question of functionality. Mace elaborated further on the idea that “IAMs, like déjà vu and other similar phenomena, may be cognitive failures.” First of all, we would like to clarify that in the target article, we did not want to adjudicate between two opposing views whereby IAMs are or are not functional. These experiences clearly expose some adaptive function as identified by other commentators (e.g., a constantly active system with a bias for making type I errors [see the evolutionary perspective proposed by Veit & Browning]). In addition, we propose that the ability to set an appropriate awareness threshold might have been beneficial from evolutionary perspective for preventing being flooded by spontaneous thoughts (for a case study, see Parker, Cahill, & McGaugh, 2006).

In Table 1, we wrote that IAMs “may be a side effect of typical involuntary processing of contextual/environmental cues, IAMs may serve several functions and roles,” and in retrospect, the word “may” should have been highlighted. We believe that IAMs and déjà vu are most likely the by-products of other cognitive processes, and in this way, any one instance of déjà vu or IAM for the individual is not functional. However, this is not to say that in their apparent randomness, they cannot be functional on some occasions – the individual may choose to interpret their IAMs (or indeed déjà vu) or as meaningful or as serving a function. Spontaneous cognitions help us to solve a problem creatively, or remember a prospective task, for example. In any case, IAMs are not completely random, as we were at pains to demonstrate in the target article, since there are several processes and mechanisms showing that the likelihood of their occurrence can be predicted and even manipulated (e.g., Barzykowski & Niedźwieńska, 2018; Mace, 2005).

IAMs and déjà vu “can be explained in the absence of a functional account” (Mace). Does this mean that we should treat them as a cognitive failure? If we agree that they are a result of a failure, then we somehow treat them as some sort of exception, but they arise because this is how our cognition works: They are neither failures nor triumphs of the cognitive system, since this evaluation can only be made in such terms in a given context. A helpful parallel might be forgetting. If you forget a friend's birthday, you may describe it as a failure of the cognitive system, but if you forget the hurtful comments made by a colleague the last time you presented your work at a seminar, we might describe it as adaptive.

To sum up this section, we note that some commentaries extend our thinking into domains other than memory (e.g., attention or inhibitory function discussed by Burns) and draw our attention to aspects of the two experiences that we did not discuss in detail (e.g., strangeness, temporality) or which beg questions about our notion of a continuum (e.g., differences in frequency between IAMs and déjà vu). We defend our continuum in a final section, but first offer some concrete mechanisms on which to base our thinking, we respond to the commentaries about the brain.

R3. An elaboration of the neuroscientific basis of déjà vu and IAMs

Neuroanatomically, our proposal is built upon the idea that in human memory, there are separable representational and attribution systems. The importance of the attributional system was highlighted by Bastin. We are grateful for her endorsement of our framework since our original neuroscientific view was inspired by Bastin et al.'s (2019) integrative memory framework. Their account explains an interaction between core systems and an attributional system, and it seems to us that any neuroscientific account of memory retrieval must encompass higher-order epi-memorial processes captured in the attributions we are able to make about fluency, etc., as well as more mechanistic accounts of the core systems. In a similar way, Montaldi and Kafkas (2022, p. 5) discuss familiarity mechanisms in light of two systems: A global matching system and a fluency-attribution system:

Although the theoretical development of these two familiarity mechanisms have been somewhat distinct, they are highly compatible. … the global-matching mechanism explains how familiarity signals are computed by brain regions within a potential familiarity memory network … while the fluency-attribution mechanism explains how the feeling of memory is generated and how decision-making contributes to familiarity memory.

When discussing familiarity, we think we should adhere to these definitions. Of note, these global matching and fluency attribution systems map neatly onto the ideas of gist and phenomenological familiarity discussed above.

A number of commentaries specify the medial temporal-lobe mechanisms which may be responsible for the generation of déjà vu. Addis & Szpunar identify pattern completion as a critical mechanism involved in involuntary memory generation. They propose that the hippocampus responds to an overlap between a cue and a stored representation, such that when activation surpasses a certain threshold, retrieval is automatic, the pattern is completed, and then a layer of neurons in the hippocampus, CA3, recruits neocortical areas where different elements of the scene are stored. Their account is compelling, since it is a neuroscientific proposal of a mechanism (i.e., pattern completion) which could account for a threshold-like process in retrieval of the personal past. Yang, Martin, & Köhler (Yang et al.) provide a useful coupling between intentionality and déjà vu. We had overlooked in our original article the fact that the overwhelming bias in neuroscientific accounts of familiarity is of research into intentional, strategic detection of familiarity. However, our argument hinges upon an automatic and spontaneous activation of familiarity circuitry. Happily for us, Yang et al. specify research findings which point to involuntary detection of familiarity in the MTL. They point out that the perirhinal cortex specifically is also activated in response to previously seen stimuli in implicit tasks independently of intentionality. Moreover, if we extend their argument further, it is clear that the perirhinal cortex is also activated during encoding and not simply retrieval (see Montaldi & Kafkas, 2022). As such, we might pinpoint the perirhinal cortex as a critical starting point in a broader network for retrieval.

Although other commentaries specify different structures, such as the anterior thalamic nuclei and the retrosplenial cortex (e.g., Sikorski & Sitek), and Zorns, Sierzputowski, Pardillo, & Keenan (Zorns et al.) highlight that the right hemisphere might be specifically involved, these commentaries lack the specific mechanisms that are involved in the same way as the Yang et al. and Addis & Szpunar contributions. While Yang et al. seem to emphasise a view of familiarity as found in the perirhinal cortex, Addis & Szpunar focus on complete retrieval and pinpoint the hippocampus, and in particular CA3 as being a critical feature. Together, these commentaries echo the proposal of Stendardi et al., which essentially suggests déjà vu arises from a lack of hippocampal involvement in retrieval.

Varma & Yu elucidate the aspects of the default mode network in involuntary memories and déjà vu, and further refine the neurological characteristics of these phenomena. As with the Stendardi et al. and Addis & Szpunar commentaries, they see déjà vu as arising when there is a lack of the usual hippocampal activation in memory retrieval. They invoke the cascaded memory systems framework (Kaefer, Stella, McNaughton, & Battaglia, 2022), and propose spontaneous neural replay as the precise neurocognitive mechanism. Spontaneous neural replay is associated with characteristic bursts of neural activity and occurs in both hippocampal and neocortical areas, and it is thus appealing to think of it as underpinning spontaneous phenomena such as involuntary memories and déjà vu. However, the conscious experience of neural replay is not well understood, and we cannot go too far with this way of thinking since many neuroscientific explorations of spontaneous activity (sharp-wave ripples) are based on rodents, and many analogical processes in the human brain occur during sleep (see Kaefer et al., 2022). That is, whereas in the cascaded memory systems framework, there is constant reactivation of neural patterns which aids in the consolidation and maintenance of memories, it seems unlikely that this spontaneous activity is obligatorily raised to consciousness (especially when we are asleep).

To carry this important neuroscientific work further into rich human experiences like déjà vu and involuntary remembering would require more advanced models of rodent memory: As it stands, we are unsure that a rodent would be able to generate the complex metacognitive evaluation of false familiarity that is at the core of the déjà vu experience. Another strategy proposed by Varma & Yu is to use neuroscientific methods, and particularly MEG. This approach has much to offer. For instance, we could use simple associative memory tasks and contrast voluntary and involuntary forms of remembering and contrast neural signatures of the two on otherwise identical tasks. Curot et al. offer concrete proposals about how to examine the brain networks involved in déjà vu and IAMs using intracranial electrical brain stimulation, and this is perhaps the best place to start, drawing upon classifications of patients' experiences of IAMs and déjà vu.

Several commentators identify the particular form of temporal-lobe degeneration in dementia as an interesting extension to our work, in keeping with Kvavilashvili, Niedźwieńska, Gilbert, and Markostamou (2020) who propose that deficits in spontaneous cognition could be used as an early marker of Alzheimer's disease (AD). The commentaries do not form a concordant set of proposals, and before discussing them we should point out that there is a lack of empirical work on déjà vu in dementia, and relatively little on involuntary memories. It was beyond the scope of the target article to consider too fully the impact of the healthy ageing process on déjà vu and IAMs, but whilst déjà vu frequency decreases with age, involuntary memory experiences seem to stay somewhat constant across the lifespan (Moulin et al., 2014; see also Maillet & Schacter, 2016; Yeung & Fernandes, 2021, for discussions of age differences in spontaneous cognitions and autobiographical memories). As such, perhaps we need to know more about déjà vu in healthy ageing before considering it in pathological ageing, since base rates of the experience may be so low in dementia as to render it impossible to study.

Bastin; Gautier, Bulteau, Chapelet, & El Haj (Gautier et al.); Irish and Zorns et al. all discuss dementia or the delusions sometimes seen in dementia patients. Whilst Irish is too generous in her appraisal to say so explicitly, it is very difficult to reconcile our ideas with what is known about dementia, and thus we too would like to see involuntary memories and déjà vu as “testbeds” for theorising in health and disease. Irish focuses on the literature on mindwandering in AD concluding that although some authors report fewer instances of mindwandering in AD, experimental studies show that some self-referential mindwandering is possible. Gautier et al. propose that autobiographical retrieval in AD can be viewed “under the lens of a ‘déjà vu’ perspective, lacking the richness of contextual and phenomenological information,” seemingly proposing higher levels of déjà vu in AD. Irish also considers whether higher levels of déjà vu might be expected to occur in AD, but concludes this seems unlikely. She points to a lack of endogenous generation of mental content in AD as likely to explain the lack of déjà vu and involuntary memories in AD, similar to Kvavilashvili et al. (2020).

We agree with Irish, but add that people with AD are known to have deficient metacognitive access in episodic memory (Souchay, 2007; Souchay, Isingrini, & Gil, 2002). As such, people with AD lack the metacognitive access to episodic memory with which to generate the conflict at the core of the déjà vu experience. The schema-driven sense of familiarity described by Irish would therefore be more likely to lead to ungated false memories, as there would not be sufficient metacognitive and recollective information to correct this familiarity (e.g., a “recall to reject strategy,” Gallo, Sullivan, Daffner, Schacter, & Budson, 2004; see also Souchay & Moulin, 2009, and for a review of false memories in AD, see El Haj, Colombel, Kapogiannis, & Gallouj, 2020). A straightforward extension of this hypothesis would be that in memory-impaired groups more generally, déjà vu would be negatively correlated with false memories.

Zorns et al. compare déjà vu to different forms of delusional misidentification (see also Sikorski & Sitek), and particularly reduplicative paramnesia. These delusions (reviewed in Moulin, 2018) have been explicitly compared with déjà vu. For instance, Feinberg and Shapiro (1989, p. 40) describe a form of reduplication where “the patient maintains that his current experiences are a repeat of past experiences.” Reduplicative paramnesia shows a clear overlap with the concept of recollective confabulation, which has been described as like permanent déjà vu (see Moulin, 2013). The delusion arguably arises because the experient attempts to resolve intact factual knowledge of the world with erroneous feelings or inappropriate affective signals. According to Langdon and Coltheart (2000), this calls for a two-factor account of the delusion: An underlying cognitive deficit (presumably a problem with over-active familiarity) combined with an erroneous interpretation of the familiarity (the inability to reject the familiarity as false). This is different from déjà vu where the experient is aware of the inappropriateness of the familiarity. In sum, whereas we echo Zorns et al.'s call to explore these delusion syndromes with reference to our proposal, and especially to consider the role of the self in IAMs and déjà vu, there are clear differences between delusion and healthy experiences; notably people with reduplicative paramnesia are anosognosic for their overactive familiarity.

R4. Concluding remarks: The continuum between déjà vu and IAMs

A number of commentaries take issue with the notion of a continuum, and we want to avoid any metaphysical discussions about what a continuum is, only so far as to say that déjà vu and involuntary memories must, according to our view, share some common processes and characteristics. Water, ice and steam share an important common characteristic, but they differ according to context, and might be thought of as lying on a continuum. Is that comparable to the type of shared characteristics we have hypothesised to be at the core of déjà vu and IAMs? Probably not. But until more is known about these phenomena, it seems reasonable to us to point to shared processes available “off-the-shelf” in the memory literature to explain how they occur. The extent to which they operate on a continuum rests upon our notion of the “botched” retrieval of an involuntary memory (to use Stendardi et al.'s terminology): On the way to having a fully fledged involuntary memory, we may “only” achieve the strange feeling of familiarity which leads to a déjà vu. Staugaard argues explicitly that “the relationship between déjà vu and autobiographical memories is not continuous, but more akin to a path diagram.” Although they might share a starting point, at some point, their paths diverge depending on the presence or absence of memory content. In principle, we agree and this briefly summarises the key claim of our proposition.

Our boldest claim is that déjà vu and IAMs belong to the same family since they operate on the same autobiographical memory base. Staugaard and Gülgöz & Ergen question whether déjà vu should be considered a part of autobiographical memory. We argue that déjà vu and IAMs may, at least in some circumstances, originate from the same common processes, and are built on autobiographical oriented information (Would we experience déjà vu without having access to our personal past?). But does this make déjà vu an autobiographical memory? We agree with Staugaard that this would cause a serious problem for the conventional definition of autobiographical memory. Thus, as long as the instance of déjà vu is embedded within and occurs in the context of one's personal past, we consider déjà vu as an autobiographical memory-affiliated phenomena: IAMs and déjà vu share cognitive mechanisms but at some point their paths diverge. We agree with Staugaard that this may happen once the memory activation reaches consciousness. Thus, we agree to disagree with Gülgöz & Ergen saying that “clustering these constructs can be counterproductive for research.” We believe just the opposite.

To conclude, we are indebted to our peers for their comments on our work, and they have raised many challenges for empirical and theoretical developments. Some issues, such as the very notion of a continuum, are ideas which are not central to our argument. We can let that idea go as long as we can keep the idea that déjà vu and IAMs are related phenomena which are central to understanding a person's relationship with their personal past. Future research may identify how we should think about the role of gist extraction, pattern matching or whether there is one type of phenomenological familiarity or two, but at this moment, we would like our field to converge on the idea of core and attributional systems which are critical for examining our dimensions of content and feelings which clearly make up our experiences of the personal past. Central to this idea is the notion that there is a constantly active system surveying the external environment and assessing its relation to the personal past. This proposes that anything in the external world could possibly act as a cue. The extent to which this cue arouses feelings or retrieval of content from the personal past can be thought of using a threshold mechanism, and the quality and classification of what comes to mind rests on what level of activation the cue affords, and how this activation is interpreted. We believe that, as suggested by Li, Jones, & Laird (Li et al.), in future we will be able to develop computational models of both IAMs and déjà vu that may allow testing at least some ideas presented in our framework. We welcome such a possibility with interest, especially since Li et al. already successfully modelled IAMs and prospective memory.

As a final remark, we are convinced that discussing déjà vu in the context of autobiographical memory research and its theory, especially given the fact that personal past is a key context in which it occurs, may advance our understanding of déjà vu. Over the years, déjà vu was not even close to being linked with autobiographical memory although it clearly is about the personal past. We believe that it is time to deepen our understanding of not only déjà vu but also IAMs. There are plenty of questions to be answered. Some of which, we argue, can only be solved by taking a new perspective based on thinking about memory retrieval being rich and multidimensional, and above all dynamic. In their commentary, Markowitsch, Kordon, & Staniloiu remind us of the classical conceptualisation of memory systems. In our framework, we quite deliberately and extensively discussed the dynamic aspects of memory retrieval processes which are not system-unique. If nothing else, studying déjà vu and involuntary memories should help us shift focus away from this traditional systems approach onto a more nuanced, ever-active, person-centred view of the cognitive neuroscience of memory retrieval.