R2.1. Episodic construction processes

First, Aronowitz challenged us to account for the fact that recall in episodic memory is highly temporally ordered. Indeed, she is correct that this seems to be a feature that should be taken into account for any mechanistic explanation of episodic memory. She is also correct in pointing out that this phenomenon is not explained by our proposal. Nonetheless, while we agree that a mechanistic account of episodic memory should be able to account for temporally ordered retrieval, we do not perceive said property to stand in contradiction to our functional account. Moreover, it is not clear how competing alternative accounts of the function of episodic memory would make sense of this phenomenon.

Second, De Brigard & Gessell argued that our functional claims are contradicted by the findings that episodic amnesiacs like HM can engage in the appropriate type of reason-giving communication. On our analysis, however, while HM likely lacked the capacity to generate episodic content, the cognitive machinery necessary to represent certain contents under the appropriate attitude (“as experienced”) might have well been intact. This is supported by the observation that HM could make use of information represented in working memory precisely in such a way. As we have argued in the target article, episodic memory does not consist merely of a certain type of content. Rather, episodic content has to be represented in a specific way (i.e., as the outcome of firsthand experience). The cognitive systems responsible for generating episodic content, on one hand, and the embedding of such content under the appropriate attitude, on the other, should therefore be dissociable. While H.M. represents a dissociation in one direction (lack of episodic content construction with intact metarepresentational capacities), Klein & Nichols (Reference Klein and Nichols2012) reported a complementary dissociation case: Patient RB was apparently able to generate episodic content but unable to embed this content under the appropriate attitude.Footnote ¹

Third, one strand of commentaries remarked on issues relevant to the construction process underlying the contents of episodic memory. Keven, in particular, took issue with the claim that the contents of event memory and episodic memory represent the outcome of the same construction process. He pointed out that the contents of these two types of memory should differ significantly because they follow different organizational principles. While event memory, on his view, includes only spatial organization, episodic memory is additionally organized according to temporal, causal, and teleological principles. For this reason, episodic memory acquires narrative character because it becomes possible to relate different events in a causal and teleological structure. We welcome these suggestions. It does seem plausible that causal and teleological relations within and across event representations would require additional inferential processes that are not produced by event construction but rather specify relations between them. Episodic memory constructions that are induced by the need to justify one's assertions in communicative exchanges do indeed call for the creation of rich event representations that go beyond spatial and temporal elements.

In fact, as Kampis, Keszei, & Király pointed out, this is likely true for processes at both the encoding and retrieval stages of memory formation. As they rightly mentioned, information will often have to be encoded and retrieved in relation to a given social partner, taking into account common experience and the ensuing common ground. Similarly, “destination memory,” as mentioned by El Haj & Miller is thus likely the outcome of such additional inferential processes guiding encoding and enriching reconstructed content with causal relations relevant to communicative interaction. These processes likewise serve to maintain and track common ground in communicative interaction. We further agree with Diekelmann, Paulus, & Krach that consolidation processes during sleep have an important role to play in the mechanisms mediating between encoding and retrieval of episodic memories.

R2.2. Episodic memory and the maintenance of common ground

Nagel proposed that autonoesis, rather than serving as a signal of epistemic authority, tracks the informativeness of a given event representation for one's interlocutors. However, if episodic encoding and retrieval processes include elements of tracking and maintaining common ground, information about informativeness would come for free. Informativeness would thus not be a design feature of the metarepresentational component of episodic memory but rather fall out of the way its contents are generated. Moreover, Nagel's view would not provide any explanation for the privileged role that humans assign to firsthand information. As Henry & Craver pointed out, one significant benefit of our account is that it makes intelligible why firsthand information is so often and intuitively treated as epistemically privileged. If autonoesis merely provided information about what would be informative, it could not account for the privileged treatment that autonoetic information receives in communication. In sum, although the co-occurrence of autonoesis and informativeness follows from our characterization of episodic memory, we do not think that one can be derived from the other.

Further, viewing episodic memory as playing a role in the management of common ground might throw light on why the creation of a “shared reality” is crucial in the maintenance of social relations, as emphasized by Coman and Hirst & Echterhoff. The reason why we care about maintaining common ground about the past with those close to us is because of (1) the role that such events play in maintaining cooperative relations in general and (2) the enormous inferential potential that specific episodes might have for learning about others' dispositions. Communicating about past events grounds gossip, determines individual accountabilities, and enables the transmission of inductively derived conclusions. In fact, joint reminiscing and sharing past experiences might do on the group level what private remembering does for the individual: create a collective sense of epistemic authority (Seeman Reference Seeman2016).

As emphasized by Dessalles, the unexpectedness of a given event should be an important factor governing both its tellability and memorability. This is to be expected under the assumption that sharing episodes is crucial for maintaining common ground with relevant interlocutors. However, the negotiation of communicative commitments via epistemic authority provides the backdrop through which episodic memory can fill this role. It matters whether the event one is told about is coming from one's interlocutor's personal experience, is made up, or is being retold in order to decide how much one should rely on it, how to transmit it further, and who is accountable for its truth. Of course, as Fivush pointed out, these practices of sharing memories have to be considered in relation to the given sociocultural context in which the spearker's identity might additionally determine how claims are evaluated. This, however, does not mean that claims to epistemic authority as such are entirely dependent on cultural context.

In spite of the clear applications of (the sharing of) episodic memories in the creation of common ground, we nonetheless view this role of episodic memory as secondary to its function as we analyzed it. Common ground can be tracked, maintained, and generated through semantic information alone. Of course, as described above, past events might carry implications, which make it particularly important to maintain common ground about them. In sharing those events (as experiences), episodic memory serves its primary function.

R2.3. Autonoesis and metarepresentation

The second, and in some sense central, aspect of our characterization of episodic memory has been our analysis of autonoesis. On our understanding, autonoesis is the outcome of a metarepresentational attitude taken toward the outputs of episodic construction processes. Consequently, we represent the fact that information about the event in question was acquired through firsthand experience. Thus, the intensional direction of the representational content of the memory points no longer toward the event itself but rather toward the relationship between one's own mind and the event in question. In other words, by taking the attitude of remembering toward a given event simulation, we do not represent the event but rather how we came to acquire a given belief based on that event. This is why we took episodic memory to be “epistemically generative”: It represents explicitly the justification for endorsing a given belief.

Some commentators have claimed that it is a mistake to think that such a metarepresentational attitude is essential to the concept of episodic memory. Both Carruthers and Werning & Cheng remarked that such a metarepresentational form of memory is incidental and not the obligatory format of episodic memory. On their view, all of what episodic memory seems to do can be explained purely in terms of first-order content. Thus, although we might sometimes metarepresent the contents of episodic memory, just as we do with other mental contents, it might not be essentially metarepresentational. Indeed, we agree that it is not obvious why episodic memory should have a second-order component. The question why it nonetheless seems to include such a component provides the motivation for our functional account in the first place. Thus, even if it turned out that we were wrong in claiming that episodic memory obligatorily includes a metarepresentational component, an explanation would still be required for why we regularly metarepresent it in this way.

There are, however, good reasons to think that episodic memory proper indispensably invokes a dedicated form of metarepresentation. While it is true that the contents of episodic memory will usually possess features that will easily allow one to infer that the associated event was experienced, this inference still has to be carried out. And the outcome of such an inference will necessarily be metarepresentational by virtue of being source information (i.e., information about one's own beliefs). The second-order component, moreover, seems to have a distinct developmental trajectory (Haigh & Robinson Reference Haigh and Robinson2009; Robinson & Whitcombe Reference Robinson and Whitcombe2003; Whitcombe & Robinson Reference Whitcombe and Robinson2000) and seems to produce signature deficits when lost (Klein & Nichols Reference Klein and Nichols2012). Indeed, the circumstance that there seems to be clear separability between the first- and second-order components of episodic memory prompted us to draw a distinction between episodic and event memory.

It is worth providing an example here. Take an instance of alcohol-induced amnesia: You might not remember that you texted your ex-partner when you were drunk last night. Now your friend who was with you at the time of the texting tells you about this episode. As a consequence, you form an accurate event simulation based on your friend's account of the events. The resulting representation will now have a lot of the features of episodic memory without being, strictly speaking, a memory: It will include a first-person perspective, it will be vivid, you will believe that the events it represents occurred, and you will also believe that you were the agent involved. Importantly, however, this representation will not include the crucial piece of source information telling you that you know about the events of last night through your own experience. After all, the source of your knowledge about the events in this case will be your friend's testimony and not your own experience. Thus, it will not be autonoetic in our senseFootnote ² and will not fill the functional role of an episodic memory.Footnote ³ Hence, it seems clear that the bearer of such a representation will not claim to remember.

This example illustrates a number of things. First, it is clear here that autonoesis should not be taken to be part of the content of the memory. The content of the event simulation might be exactly the same regardless of whether you formed it on the basis of your friend's testimony or not. Second, we can see that source information is importantly different from agentive information: That is, knowing that “I did X” is not the same as knowing that “I know about X because I experienced it.”

This is relevant for our understanding of the kind of self-reference at issue here. According to Szpunar & Chan, episodic memory is “key to the self in time.” Indeed, the content of episodic memory might include agentive information about which actions you performed yourself and which you did not. As the previous example illustrates, however, episodic memory is not necessary for this because agentive and source information can be dissociated. While a temporally extended conception of the self is certainly part of what makes episodic memory effective, autonoetic memory is not necessary for it to occur.

Thus, Rau & Botterill might well be correct that trace information could be constituted by perceptual and motor models, which might underlie representations of agency at recall and help inform motor learning and action control. However, again, this alone is not sufficient for autonoesis (and hence episodic memory) to occur.

In the same vein, we find Morton's description of research on patients with dissociative identity disorder fascinating. However, we do not perceive these phenomena as posing a problem for our account. On the contrary, Morton's description of the underlying mechanisms seems entirely compatible with the general picture we at least intended to paint of the way autonoesis occurs. That is, we agree that an “agentive” notion of self might be part of the “addressable” content of episodic memory, while the “autonoetic” sense of self is likely the outcome of a metacognitive attribution process at the retrieval stage.

De Brigard & Gessell challenged our characterization of episodic memory from another angle. They argued that certain kinds of simulations of past counterfactual events have all of the features we identified as distinctive of episodic memory without being memories themselves. This claim seems surprising to us, because, if this was indeed the case, we would have difficulties telling these two kinds of mental activities apart in our daily lives. But the difference between past counterfactuals and actual episodic memories seems glaringly obvious: In the latter case, one takes oneself to have indeed experienced the event as it actually occurred, whereas, in the former case, one makes an inference from that experience as to what could have occurred if circumstances had been different. There are two important points to make here: (1) “Firsthand” information in the way we used the term was not intended to mean something akin to “only I could know this.”Footnote ⁴ It rather meant that the bearer of a given representation also represents that its contents were acquired through her own perception – something that is clearly not the case for the “subjunctive replays” of the kind described by De Brigard & Gessell. In other words, we do not consider counterfactual simulations – or any other epistemic metarepresentational attitudes (except remembering) – to be autonoetic. (2) Even though we agree with De Brigard (2013) that the mechanisms underlying content generation for both episodic memories and episodic counterfactuals are closely similar, their outputs crucially differ in intensional direction. Episodic memories present themselves as being directed toward actual experiences, while episodic counterfactuals do not. The fact that in actuality their representations are both outcomes of a best guess as to what is most likely to have occurred given certain assumptions does not make this difference irrelevant. In fact, it is exactly this difference that underlines Michaelian's observation that episodic memory is systematically misleading (see R3.3.).

Importantly, the proposal that episodic memory proper is metarepresentational in nature does not entail that this representational format is due to a domain-general metarepresentational faculty. In fact, there are good reasons to think that metarepresentational capacities might be domain specific (e.g., Mercier & Sperber Reference Mercier and Sperber2017; Sperber & Wilson Reference Sperber and Wilson2002) and that general mind-reading abilities are (while related) importantly distinct from the kind of source representations involved in episodic memory (Rosenbaum et al. Reference Rosenbaum, Stuss, Levine and Tulving2007; Naito Reference Naito2003). Thus, it strikes us as unlikely that, as Redshaw & Suddendorf claimed, episodic memory should be a by-product of the application of domain-general metarepresentational capacities to the outputs of the scenario construction system. Instead, episodic memory is the outcome of a dedicated metarepresentational mechanism, which allows us to take the attitude of remembering toward certain mental contents.

R2.4. Epistemic generativity and the justificatory role of episodic memory

Werning & Cheng further argued that epistemic generativity and the epistemic privilegedness of episodic memory can be viewed as outcomes of purely first-order content. Here, unfortunately, our use of the term epistemic generativity has been misunderstood. Werning & Cheng seemed to have taken us to claim that the beliefs we form on the basis of episodic memory are epistemically justified in a normative sense. To be clear, we did not claim anything about whether episodic memory does in fact generate epistemic justification so understood. Rather, we propose that episodic memory plays the psychological role of supplying a justification for a given belief. It might be true that the processes underlying episodic memory construction are statistically reliable, and thus our beliefs formed on their basis can be taken to be epistemically justified such that we are epistemically entitled to them. But our point was a different one; namely, that when we remember, we know that we are justified in believing that a given event occurred precisely because we remember it.

Non-believed memories, as mentioned by Nash and Blank, provide a perfect illustration of this. In this case we do not take the fact that we remember to be good-enough reason to believe that a given event indeed occurred. Remembering the occurrence of an event is not always sufficient, let alone the only reason, to indeed believe in its occurrence. The fact that belief and episodic memory can dissociate in this way thus showcases that they stand in a psychological relation of justification.

Relatedly, our account aims to explain not why we should treat episodic memory as epistemically privileged in a normative sense. Rather, as pointed out by Henry & Craver, we explain why we often do treat contents made available through episodic memory as privileged. And that is because firsthand experience bestows epistemic authority in communication by making us accountable for its truth. And, to reiterate, representing a given episode as experienced firsthand requires a metacognitive inference.

R2.5. Distinguishing different kinds of memory

Another aspect relevant to determining the nature of episodic memory is its distinction from other kinds of memory. On the one hand, Berntsen and Nelson took issue with the fact that we did not discuss autobiographical memory. As mentioned in the target article (footnote 1), we do not take autobiographical memory to be a natural kind (Michaelian Reference Michaelian2011b; Reference Michaelian2016b). Rather, autobiographical memory is a specific type of memory content, which can be represented both in episodic and semantic memory. As such, the question of why we form such autobiographical memory contents is substantially different from the question of the function of episodic memory. Relatedly, Rubin and Berntsen pointed out that we did not reference work investigating the uses of autobiographical memory (e.g., Bluck et al. Reference Bluck, Alea, Habermas and Rubin2005; Harris et al. Reference Harris, Rasmussen and Berntsen2014). This strand of research did not make a clear distinction between episodic and autobiographical memory, and it investigated the function of autobiographical memory by explicitly asking participants about the role these memories play for them. Although the answers of participants in such situations can be revealing, it is not obvious what significance they have for investigating the proper function of episodic memory. As mentioned in the target article (sect. 2), the function of a cognitive capacity needs to be importantly distinguished from its uses, and therefore we did not take this work to be relevant to our question.

On the other hand, several authors have commented on the distinction between episodic and semantic memory. Rubin claimed that the distinction is becoming increasingly hard to see, and Hayes, Ramanan, & Irish (Hayes et al.) argued that semantic memory plays a significant role in episodic memory itself. Indeed, we agree that episodic and semantic memory are intricately intertwined. The scenario construction process underlying episodic memory draws heavily on semantic information, given that encoded traces are not event representations themselves (a point also emphasized by Werning & Cheng). In fact, we cited some of the work mentioned by Hayes et al. ourselves in pointing out that the construction of future-oriented as compared to past-oriented scenarios should rely more heavily on semantic information (Irish et al. Reference Irish, Addis, Hodges and Piguet2012a). Despite this intricate relationship, we nonetheless maintain that it remains useful and meaningful to distinguish between the outputs of scenario construction (i.e., event and episodic memory) and the semantic memories serving as its inputs.

In a slightly different vein, Nelson expressed discontent with the developmental primacy of semantic over episodic memory implied in our account. Nelson seemed to suggest that both semantic and episodic memory emerge from a third kind of memory, which she termed infant memory. However, we do not perceive this view to necessarily stand in contradiction to our own account.

R3.1. Episodic memory veridicality and mental time travel

One significant point of controversy emerging from the commentaries was our claim that the “mental time travel” view of episodic memory function would have trouble accounting for the veridicality that episodic memories frequently display. Schacter, Carpenter, Devitt, Roberts, & Addis (Schacter et al.) emphasized that the constructive episodic simulation hypothesis (Schacter & Addis Reference Schacter and Addis2007) assumes that accurate episodic memory supports the simulation of future scenarios. However, their use of the term episodic memory seems to differ from our own. Schacter & Madore (Reference Schacter and Madore2016), which Schacter et al. cited to illustrate the contribution of veridical episodic memory to future imagination, stated that “remembering past experiences and imagining future experiences both rely heavily on a particular form of memory known as episodic memory” (p. 246, their emphasis). This suggests that episodic memory is viewed here as something that is prior to, and underlies, both future and past-directed simulation processes. According to Schacter (personal communication, June 28 2017), “episodic memory” here is meant to refer to “retrieved episodic information.” To the extent that this implies that one does not first have to construct a simulation of a past event in order to construct a simulation of a future scenario, this view does not contradict our own account. Indeed, in the target article we concurred that future-oriented mental time travel should depend on, and make use of, episodic information in the form of memory traces originating from past experience. However, we questioned the claim that future-oriented mental time travel would benefit substantially from the capacity of veridical past-oriented mental time travel. A combination of established semantic memory plus trace information from past events could do all of the work here.

Benoit, Berkers, & Paulus (Benoit et al.) proposed an intriguing extension of the mental time travel view. One way to read their proposal is that episodic memory and episodic simulations in general serve to motivate prospective decisions based on non-occurrent information. This proposal resonates with views by Boyer (Reference Boyer2008) and Hoerl & McCormack (Reference Hoerl, McCormack, Michaelian, Klein and Szpunar2016), who similarly argued that episodic simulation in general, and episodic memory specifically, might function to implement certain emotion-based decisions. Indeed, one version of this proposal seems plausible: Episodic simulation does function to implement emotions in time. Because we can simulate a given non-occurrent event, we can concomitantly experience its emotional consequences, which should in turn influence decision making.Footnote ⁷ Benoit et al., however, seemed to restrict the role of episodic simulation merely to prospective emotions. This strikes us as unlikely: If anything, past-directed episodic simulations should be primarily geared toward guiding the emotional assessments of past events in the form of gratitude, guilt, shame, and so forth.Footnote ⁸

All in all, as long as it is recognized that simulations of specific past events play a distinct, stand-alone role in motivating decisions based on non-occurrent information, we are sympathetic to views such as the one proposed by Benoit et al. Nonetheless, the functional role of veridicality in such an account remains unclear because appropriate (i.e., adaptive) motivation can be achieved independently from veridical representation. In fact, many motivational systems seem to depend on misrepresentations to be effective (Kurzban Reference Kurzban2012).

Moreover, this view is compatible with our own account only to the extent that the motivational role of episodic simulation is not dependent on a metarepresentational format. In fact, it is questionable whether Benoit et al. can account for the metarepresentational component distinctive of episodic memory.Footnote ⁹ While to some extent, this is surely an empirical question, there are some reasons to think that the distinctive metarepresentational component of episodic memory is not responsible for its emotional impact. Counterfactual and future-directed simulations have emotional consequences while differing in metarepresentational content from past-directed simulations. The fact that I have not yet experienced the future does nothing to lessen its potential emotional impact on me. Indeed, the earlier example of alcohol-induced amnesia illustrates this point perfectly: One can feel ashamed about having texted one's ex-partner when drunk without (strictly speaking) remembering anything about the event. Thus, it seems unlikely that the metarepresentational format of episodic memory is required for its role in generating emotions. Rather, its content (i.e., the simulation of a specific past event) seems to be doing the work here.

Note that the relationship between emotions and episodic memory emphasized here is different from the one discussed by Eaton & Anderson. They rightly remark on the role of emotions for the encoding of memory trace information and argue that this role is likely prior to the social functions of episodic memory. Indeed, as we have argued in the target article (sect. 4: Conclusions), the encoding of memory traces is likely influenced by many distinct mechanisms, because these traces themselves inform a plethora of processes. This fact in itself, however, does not preclude the finished product (episodic memory proper) to serve a crucial sociocommunicative function.

R3.2. The relationship between autonoesis, veridicality, and confidence

A range of commentators discussed our account of the relationship between autonoesis, veridicality, and confidence. On our view, autonoesis serves as a signal for the fact that one has epistemic authority about the contents of one's memories. Crucially, this has to be distinguished from both a signal of veridicality (Blank) and of confidence/certainty (Werning & Cheng).

On the one hand, distinguishing autonoesis from assessments of veridicality makes intelligible how “non-believed memories,” as mentioned by Nash and Blank, are possible. Because judgments of the source of one's beliefs have to be distinguished from judgments about the truth of those beliefs, one can remember to have experienced something that one takes to not actually have occurred. This is presumably the reason why Johnson et al. (Reference Johnson, Hashtroudi and Lindsay1993) distinguished two kinds of mechanisms responsible for these judgments: reality and source monitoring, respectively. Thus, pace Blank, the role of autonoesis as a signal of epistemic authority is not dependent on it signaling veridicality but rather accountability. By claiming that it originates in my own experience, I take personal responsibility for a given statement, and I make myself accountable in case of being found wrong, which should make my interlocutor more likely to rely on whatever I am stating. While this view predicts the statistical co-occurrence of autonoesis and veridicality, the relation between the two is mediated by other factors (such as whether or not veridicality can be monitored by others and the costs of being found wrong). This should also make clear why certainty or confidence judgments about the veridicality of one's recollections should be distinguished from autonoesis. First, confidence about the accuracy of the contents of a given event representation and confidence about the way one acquired it can be estimated separately from each other. Second, confidence reports have to be distinguished from the kind of evidential reports that autonoesis allows us to perform. This distinction is analogous to the distinction between evidentiality and modality in linguistics (e.g., De Haan Reference De Haan2001).

R3.3. Epistemic authority and issues of accuracy

Pace Werning & Cheng, we would thus insist that our account has little to say about the relationship between confidence reports and memory accuracy. This said, confidence reports can surely function as commitment devices (Vullioud et al. Reference Vullioud, Clément, Scott-Phillips and Mercier2017). As such, while they should certainly be sensitive to accuracy, they should be influenced by other factors as well. Indeed, the relationship between these constructs is quite nuanced (see, e.g., DeSoto & Roediger Reference DeSoto and Roediger2014; Koriat & Adiv Reference Koriat, Adiv, Dunlosky and Tauber2016; Wixted & Wells Reference Wixted and Wells2017).

One might think that we should instead claim that judgments of autonoesis should accurately track the origins of our event simulations. However, as Michaelian pointed out, scenario construction always makes use of an amalgamation of different sources of information. Episodic memory might thus be said to be misleading in principle about its metarepresentational component. In fact, there is no such thing as a piece of information purely acquired through firsthand experience. Rather, “firsthand experience” is a construct that serves a certain sociocommunicative purpose. It is attributed to certain mental contents in order to justify them rather than to accurately represent their origin. Thus, Michaelian is right: As a rule, autonoesis cannot be wholly accurate. Crucially, however, reason-providing cognitive capacities in general share this feature (Mercier & Sperber Reference Mercier and Sperber2017): Because reasons serve to justify a given conclusion rather than to manage its truth value, they do not have to correspond to the actual causes for why one endorsed the conclusion in question in the first place.

Of course, this seemingly counter-intuitive implication of our proposal does not entail that there should not be certain constraints on when we take ourselves to have experienced something firsthand. Nonetheless, even the case of confabulations, as mentioned by Robins, is nothing extraordinary from this perspective. Rather, confabulations simply represent one extreme of the systematic misleadingness mentioned by Michaelian. In the memory manipulation paradigms a la Loftus & Pickrell (Reference Loftus and Pickrell1995), participants construct an event simulation based on an experimentally induced belief about their own past and try to justify it through autonoetic remembering. The fact that this kind of behavior can be reliably induced experimentally in the laboratory only shows that episodic memory is sensitive to social demand characteristics. The phenomenon of confabulations itself thus in no way challenges the claim that autonoesis in general serves as a signal of epistemic authority.

R3.4. The relation between episodic memory and commitments

A few authors commented on the relationship between autonoetic remembering and diachronic commitments in our account. Specifically, Nagel suggested that we got the role of remembering and social commitments the wrong way around. Instead of merely allowing us to negotiate commitments communicatively, it might be that self-referential remembering plays a role in implementing our capacity for diachronic commitment itself. Similarly, Michael, Székely, & Christensen argued that episodic memory contents can be important in determining whether a commitment has been incurred in the first place. Indeed, it might well be that there has been pressure on the episodic memory system to encode details of events in order to be able to later justify certain interpretations of them. And this mechanism in turn might play a role in implementing commitment insofar as one takes oneself to be committed to whatever is justifiable. This might explain findings such as those described by Fabbro & Crescentini, in which a connection between anti-social behavior (particularly in the communicative domain) and episodic memory deficits emerges.

One way in which this mechanism might operate is through the kind of emotional retrospection mentioned in our discussion of Benoit et al.'s remarks. That is, the commitment situation might carry motivational force through time by way of the emotional consequences of remembering it. Crucially, however, this mechanism would be based exclusively on an “agentive” notion of self without appealing to the “autonoetic” self that operates in episodic memory. Put differently, first-order memory content, referring to the self, could be entirely sufficient for the implementation of commitment itself without representing the source of this content. The constitutive role of episodic memory in social commitments suggested by Nagel might thus be supported by the agentive notion of self involved in episodic simulations. In sum, the role of self-referential memory in social commitments may go both ways by supporting both their motivational force as well as their social enforcement.