Witnessing, Remembering, and Testifying: Why the Past Is Special for Human Beings

Abstract

The past is undeniably special for human beings. To a large extent, both individuals and collectives define themselves through history. Moreover, humans seem to have a special way of cognitively representing the past: episodic memory. As opposed to other ways of representing knowledge, remembering the past in episodic memory brings with it the ability to become a witness. Episodic memory allows us to determine what of our knowledge about the past comes from our own experience and thereby what parts of the past we can give testimony about. In this article, we aim to give an account of the special status of the past by asking why humans have developed the ability to give testimony about it. We argue that the past is special for human beings because it is regularly, and often principally, the only thing that can determine present social realities such as commitments, entitlements, and obligations. Because the social effects of the past often do not leave physical traces behind, remembering the past and the ability to bear testimony it brings is necessary for coordinating social realities with other individuals.

For human beings, the past is special. We think of the past as defining almost all aspects of our lives: where we belong, who our friends are, what our social status is, what kind of person we are. We also love to talk about the past. We share much of our (emotional) experiences with others (Pasupathi, McLean, & Weeks, 2009; Rimé, Finkenauer, Luminet, Zech, & Phillippot, 1998), and, according to one estimate, 40% of our conversational time is spent telling stories about past events (Eggins & Slade, 2005; Hirst & Echterhoff, 2012). In fact, humans seem to have a “retrospective bias” in their conversational behavior: We talk about our personal past two to three times as much as about our personal future (Demiray, Mehl, & Martin, 2018). The special status of the past is also reflected in the fact that humans operate a dedicated “episodic memory system” for cognitively representing specific past events. Whereas other memory systems allow us to simply know what happened in the past, episodic memory also lets us know how we know what happened. When we remember a past event, we not only recall the event, we also know that we experienced it. In other words, episodic memory allows us to become witnesses of the past and thus give testimony about it.

The past is so pervasively important for us that it might seem hard to see that the question of why this should be the case is a genuine puzzle. In this article, we attempt to solve this puzzle. To do so, we have to consider the evolution of the human cognitive architecture for thinking about the past. If it is true that episodic memory is a special way of representing information about past events that lets us know how we know about them, then why did we develop such a system? In other words, what is it about past events that requires a dedicated, metacognitive mechanism to think about them?

Our answer to this question relies on the observation that, for human beings, specific events have not only physical but also social effects that may not leave physical (but only mental) traces behind. For this reason, such events require a dedicated capacity that allows us to negotiate them effectively in communication.

We therefore recapitulate arguments we have made elsewhere in more detail (Mahr & Csibra, 2018) about the nature and human-specific function of episodic memory. The ability to remember allows humans to distinguish knowledge about the past that they acquired on the basis of their own experience from that acquired in other ways. This ability constitutes the basis of testimony about the past: An act of testimony is an account about the past that is claimed to be based on first-hand experience. The term testimony is sometimes (especially in philosophy) used to refer to any transmission of social information. This is emphatically not the way it will be used here. Instead, we use the term to refer to an account about the past on the basis of first-hand experience and the epistemic authority such experience conveys. This use of testimony might evoke associations with the legal domain, in which it is most commonly studied and has been institutionalized as eye-witness testimony under oath. Yet testimony understood as “experience-based” communication about the past is probably the most common form of talk about the past and is not constrained to the courtroom.¹

Remembering (i.e., episodic memory), therefore, is the cognitive basis of testimony. We previously argued on the basis of this assertion that remembering functions to support communication about past events (Mahr & Csibra, 2018). However, the claim that a fundamental function of episodic memory is to facilitate the communication about specific past events by enabling testimony about them implies that communication about such events is important enough to justify the evolution of such a dedicated mechanism. It is, however, not clear what grounds this high importance of specific past events to human beings.

Here we aim to give a more full-fledged answer to this challenge than previously provided.² Thus, we ask: Why is the past special for humans? In other words, what is so special about the past that requires a dedicated mechanism allowing us to facilitate communication about it? In attempting to answer this question, we argue that although the transmission of information about specific past events might—under some circumstances—allow for the teaching of generic information and the dissemination of reputational information to others, its main purpose is to justify claims about present social entities such as commitments, entitlements, and obligations.

The reason for this is that a large part of our social ontology is reliant on representations of history. Therefore, testimony will sometimes be helpful in coordinating the social realities that we take to be in effect. We thus develop an account of how the dependency of many “social facts” on particular past events might make communication about these events necessary. From this perspective, transmitting the events that causally ground a given social fact will sometimes be the only way in which the existence of this fact can ultimately be established. To the extent that there is no other way to independently track the social effects of a given event, testimony about this event will be important in signaling its existence and thereby coordinating the shared representation of social reality with others.

The Nature and Communicative Function of Episodic Memory

Adult humans seem to have two main ways in which they can cognitively represent information about past events. On the one hand, information about specific past events can be represented propositionally in semantic memory, as in “the Berlin Wall fell on the night of November 9th, 1989.” Although this way of representing information about events is common, humans also recall events as rich, quasiperceptual representations of specific past episodes (Clayton & Russell, 2009; Mahr, in press).³

However, episodic memories are more than just mental representations of specific, past events (Mahr & Csibra, 2018). One represents in episodic memory not only that a given event occurred, uniquely, in the past, but further how one came to acquire information about this event: by having had first-hand experience of it (Dokic, 2001; Perner & Ruffman, 1995). This is illustrated by the fact that although even infants demonstrate the capacity for recalling unique events (Bauer & Leventon, 2013; Maguire & Mullally, 2014), only children around 5 years of age seem to be able to represent such unique events as sources of their beliefs (Haigh & Robinson, 2009). It is this additional piece of source information that grounds the autonoetic character of episodic memory (Tulving, 1983, 2002). Episodic memory proper is therefore the outcome of inferential processes making explicit the way in which a given event representation relates to a given belief about a past event: namely, as a source of this belief (Burge, 1993; Teroni, 2014). In other words, when we remember the past, we commonly take ourselves to believe in the occurrence of whatever we remember because we experienced it.

The role of sources in communication

Communication amounts to an attempt by a sender of information to influence a receiver’s mind in a specific way (Dawkins & Krebs, 1978; Krebs & Dawkins, 1984; Sperber & Wilson, 1995). For complex, reciprocal systems of communication to remain evolutionarily stable, mechanisms have to be in place that ensure that communicative behavior remains (on the whole) beneficial for both senders and receivers.

On the one hand, to ensure that the overall influence of the communicated information is beneficial, receivers must be capable of scrutinizing sources for trustworthiness, reliability and competence, and communicated information for believability (Mercier, 2017, 2020; Sperber et al., 2010). If such “epistemic-vigilance” mechanisms were not in place, receivers would not be able to judge which pieces of communicated information they ought to believe. Consequently, receivers would often be misled and exploited by senders and thus, on average, not gain from attending to communicative signals. On the other hand, speakers must be capable of influencing receivers’ minds effectively despite such vigilance. Both the mechanisms of epistemic vigilance and the mechanisms to overcome such vigilance are crucially dependent on the cognitive capacity to represent source information.

One way in which source information is important is because it can serve as a reason (Mercier, 2016; Mercier & Sperber, 2011, 2017). On the side of the sender, reasons are important insofar as one can supply them to convince an interlocutor who would otherwise not accept what one has to say on the basis of trust alone. Thus, if someone can point to whatever caused them to believe something (their own reasons), this might also be good enough for others to believe it. On the side of the receiver, one has to be able to tell good reasons from bad ones when deciding what to believe. This also means that the better senders are at giving reasons the better receivers should be at processing and scrutinizing those reasons, and vice versa.

The communicative function of episodic memory

Source information is also important in communication because it allows speakers to regulate their conversational commitments. Making an assertion commits the speaker to the truth of whatever he or she asserts (Brandom, 1983; Turri, 2011): The speaker accepts that if he or she is found to be wrong, there will be direct or reputational costs. The fact that the speaker is willing to incur such costs can serve as a signal for the audience to accept whatever he or she claims. Thus, the stronger a speaker commits to a claim, the more convincing he or she should be (Mazzarella, Reinecke, Noveck, & Mercier, 2018; Vullioud et al., 2017). Therefore, speakers should be able to regulate such commitment appropriately because overcommitment (or undercommitment) can be costly. One of the basic ways in which such commitments can be regulated is by claiming or deferring epistemic authority about whatever one asserts (McMyler, 2007). If the speaker claims to have acquired the information in question firsthand, he or she at once claims epistemic authority and makes him- or herself directly accountable for the truth of his or her assertion. This should in turn cause the speaker to be more strongly committed and hence more convincing than if he or she were to defer accountability to another, secondhand source. Representing and being able to communicate sources can thereby serve a variety of goals: Sources can be used to convince (“I saw it with my own eyes”), to take credit (“I found the solution on my own”), or to hedge one’s bets (“It’s only something I’ve heard”; see, e.g., Altay & Mercier, 2019; Shaw & Olson, 2015; Silver & Shaw, 2018).

Considering the extent to which source information is therefore useful in communication, it is not surprising that such information is grammaticalized in about a quarter of all recorded languages as evidential markers (Aikhenvald, 2004; Nagel, 2015). Moreover, even languages that do not grammatically encode the evidential basis of a claim have numerous other ways to express a speaker’s source (Aikhenvald, 2004).

Episodic memory thus allows us to (a) represent the grounds on which we formed a given belief in the first place, which we can then transmit as reasons to others or use to decide when to change our mind and (b) regulate the extent of our commitments in discourse by highlighting whether a given event representation originated in our own firsthand experience (see also Jablonka, 2017; Poole, 2008; for a view applying a similar idea to collective memory, see Seeman, 2016). Of course, this is only the first layer in a complex web of potential source information. We can discern on the basis of an episodic representation whether we have seen, heard, or inferred information about a given event (Johnson, Hashtroudi, & Lindsay, 1993). Such more fine-grained source distinctions are important because they allow one to calibrate the communicative effects of one’s statements as well as answer potential challenges to one’s authority (“How do you know?”) more precisely than simple expressions of confidence would (Vullioud et al., 2017) and might further be useful in deciding what is informative for one’s audience (Nagel, 2015).

Why do we care about the past?

If the above account of episodic memory is correct, remembering allows humans to give testimony: It allows us to decide when we can speak about the past as witnesses, that is, on the basis of firsthand experience. But why are representations of specific, past events important enough to require the inclusion of source information in the first place? In principle, one might think that such a system could apply to the representation of any type of information, including semantic, propositional facts. Autonoetic remembering, however, is specific to representations of unique, past events. It allows speakers to adaptively regulate their communicative commitments for claims about such events and allows listeners to decide when to revise their beliefs. In other words, if autonoesis indeed serves as a signal of epistemic authority, why is it specific to past events? What is so special about the past that requires a dedicated mechanism to manage claims of epistemic authority about it? Why are claims about history important enough to require justification?

One way to approach this question might be by thinking about how knowledge about the past can be relevant to human fitness. Changes in fitness can inherently exploit possibilities only in the present and future (Klein, Cosmides, Tooby, & Chance, 2002), but there are two obvious ways in which knowledge about the past can nonetheless be relevant to fitness. On the one hand, knowledge about the past might support the learning of regularities in our environment. If we know what happened, we might be able to use this information to inductively infer regularities in the way our environment works and therefore form appropriate expectations about what will happen. On the other hand, however, knowing about the past can be important because some past causes have effects that manifest only after some time in the future (e.g., infections). In this way, knowing what happened might allow one to predict what will happen or what is the case. Now, can we apply these insights to the question of when the transmission of information about the past might be relevant to fitness? After all, episodic memory is structured to facilitate the transmission of information about the past, and the special status of the past seems to be particularly prominent in human social life.

It might seem plausible to answer this question by pointing to the fact that the communicative transmission of a past occurrence could function as quasiexperience for others to form judgments about. If we can transmit our own experiences to other people, to the extent that our interlocutors believe us, they might vicariously learn from this experience just as if it were their own. This fact alone would be a good reason to sometimes require justification for claims about the past. However, not all judgments are equally well transmitted in this way: Learning from the transmission of social information will usually benefit most from generic statements rather than claims about specific events. After all, one of the greatest benefits of human communication is that we can transmit generic information directly to others without being reliant on individual learning episodes (Csibra & Gergely, 2011).

Moreover, it is not clear what role the “pastness” of our experience would play in allowing others to learn from it. The transmission of information about specific past events is not identical to the transmission of information about specific events in general. That is, instead of asking what we can vicariously learn from the transmission of information about specific events, we must ask what we can learn from the retrospective representation of such events that is important enough for “pastness” to play a role in its transmission.

Therefore, in what follows, we explore what kinds of judgments (a) are particularly sensitive to the kind of evidence provided by claims about specific past events and (b) carry particularly high social consequence so that humans would care about, and consequently regularly require, additional reassurance in their transmission.

The Past Supports Learning: Generics and Reputations

When people are asked what kind of inferences are well supported by reference to past events, it is common to point to induction. Clearly, to the extent that a given judgment is supported or supportable by inductive inference, it will benefit from reference to past experience.⁴ We might thus care about what happened on particular occasions in the past because such events increase the potential sampling base behind our inductive inferences, leading to our generic beliefs. After all, one way to arrive at a generic belief is by generalizing over specific instances. Thus, the claim that “It rains on Thursdays in Los Angeles” could be supported by pointing out that it rained when I was there on Thursday last week (i.e., my testimony).

However, although inductive learning can be supported by evidence from specific past events, neither the specificity nor the pastness of such events is important for such learning (if it is Thursday today and it is raining in Los Angeles now this takes nothing away from the inductive power of the event). What matters instead is that the specific instance of an occurrence follows a regularity that allows for generalization. Thus, generic beliefs are suboptimally justified only by retrospectively pointing to particulars simply because individual cases might not say much about the general pattern under scrutiny. If one inductively generalizes over a number of instances, one disregards exactly what is particular about each one. The effectiveness of pointing to a specific experience in justifying a generic claim will therefore commonly be limited simply because that experience could be an outlier.⁵ The fact that it rained last Thursday in Los Angeles does not necessitate, after all, that it will normally rain there on Thursdays.

Moreover, although pointing to specific past events can be helpful for the justification of inductively derivable generics, generic beliefs are sensitive to all kinds of evidence. Your general meteorological knowledge, for example, might tell you that the weather is unlikely to conform to the days of the week. Thus, generic beliefs are not dependent on reference to specific past events to be justifiable and can also be effectively transmitted by reference to other generic facts one holds true (Prasada, 2000).

Bounding and exemplifying generics

There are, however, two other potentially more effective ways in which the transmission of generics can be supported by claims about specific events. First, as Klein et al. (2002) and Cosmides and Tooby (2000) have pointed out, claims about specific events can set bounds on how far a generalization might extend. To return to the example of “It rains on Thursdays in Los Angeles,” pointing out that it did not rain when I was there last Thursday provides a good counterexample. The universally quantified version of this assertion (“It rains every Thursday in Los Angeles”) can simply not be true if this specific event occurred. Knowledge about specific events can therefore allow listeners to debate the scope of an assertion. The bounding function of specific events seems particularly useful for the purposes of epistemic vigilance: If we are confronted with a universal claim, but we can come up with a specific instance in which it did not hold, we should, at the very least, accept only a more modest version of the claim in question.

Second, instead of being just one more data point for an inductive generalization, communicated information about a specific event might serve as an exemplar (Shafto & Goodman, 2008; Shafto, Goodman, & Griffiths, 2014): A general pattern might be illustrated and thereby supported by pointing to one specific, diagnostic instance in which it occurred (cf. “strong sampling” in Xu & Tenenbaum, 2007). You might, having never encountered a panda bear, wonder whether they are dangerous. You ask a zookeeper, who tells you that he was bitten by one once. On the one hand, as discussed above, this might cause you to inductively increase your belief in the hypothesis that panda bears are indeed dangerous. On the other hand, however, the simple fact that the zookeeper chose this specific episode from his experiences with panda bears to answer your question should cause you to treat this information as being diagnostic of a more general pattern of panda bear behavior. In other words, this episode would not make the hypothesis that panda bears are dangerous more believable because it would provide one more instance from which to inductively generalize. Instead, the listener will assume that the speaker picked that episode to share because it provides the best example from which to learn, and this in turn would make the target claim more convincing. A specific event can therefore serve to justify a general claim in virtue of its exemplary character.

One benefit of pointing to specific events as exemplars is thereby that one does not have to make explicit the target claim one aims to transmit. Simply pointing out that “I was bitten by a panda bear once” will sometimes be enough to make one’s audience infer that panda bears must be dangerous. Providing exemplars, however, will likely be mostly necessary when reasons in support of a prior claim are requested. After all, as mentioned above, one of the main advantages of communication in the first place is that we can transmit ready-made generalizations to others. Only when challenged will pointing to a specific episode (in the form of testimony) become necessary.

According to what we have discussed so far then, we should expect people to care about what happened at specific occasions in the past primarily because (a) past events can set bounds on generics, allowing us to evaluate and contradict them, and (b) because past events can serve as exemplars for transmitting generics to others in argumentation and teaching.

Disseminating reputations

Humans can teach, argue for, and evaluate almost anything by pointing to exemplifying events (or chains of events). In these cases, providing reasons in the form of specific events functions according to the same principles as argumentation in general (Mercier & Sperber, 2017). A claim will require justification if the audience does not trust the speaker enough to accept his or her claim on that basis alone. Consequently, justification will be required when the stakes of being misled or the incentives to mislead are high.

Although this is true in many specific contexts, these conditions are met very consistently in claims affecting others’ reputation. According to Dunbar (2004), the most common topic of conversation is social evaluations: As much as 65% of casual conversation concerns social topics (about others’ interactions, behaviors, and traits; Dunbar, Marriott, & Duncan, 1997). People’s interest in others’ behaviors and interactions is enormous even when they themselves are not involved (DeScioli & Kurzban, 2009), and this pattern does not seem to be exclusive to Western societies: Zinacantan people in Mexico similarly have been reported to spend 78% of conversational time talking about such social topics (Haviland, 1977).

This phenomenon is commonly termed “gossip” (Foster, 2004) and has been proposed to be essential in the stabilization of cooperative group living (Dunbar, 1998; Wu, Balliet, & Van Lange, 2015, 2016a, 2016b). In particular, Dunbar (2004) has argued that the transmission of social evaluations plays an essential role in the stabilization of our conditions of communal living. The reason for this is that information about one’s past behaviors is often taken to be diagnostic about one’s future behavior; that is, that such information potentially licenses trait inferences.⁶ Once trait judgments become shared across a group, they develop into reputational information.

By disseminating reputational information through a given social group, the transmission of social evaluations allows us to go beyond our personal experience when assessing the state of our social network and the dispositions of others (Sommerfeld et al., 2007). This in turn is taken to fulfill “policing” functions (Foster, 2004), effectively implementing a form of social control because what one believes about others’ traits will determine who one associates with and how one interacts with them. Therefore, if one can manipulate others’ opinions about a specific person, one can effectively control coalitional associations as well as cooperative opportunities. This explains why people gossip so much about others’ past behaviors and are careful about tracking (their own and others’) sources of this information (Wilson, Wilczynski, Wells, & Weiser, 2000). They have to justify and be vigilant against being misled about claims affecting others’ reputation because of the various ways one could take advantage of changing someone’s reputation (Hess & Hagen, 2006). Indeed, one’s conversational commitments are rarely more important than in the domain of gossip. Where a piece of gossip originates from, and how far the speaker is removed from having personally experienced the episode in question, are crucial both for how believable the gossip is and who is responsible for it (Giardini & Conte, 2011).

The Past Generates Entitlements, Obligations, and Commitments

So far, we have identified generic beliefs as one kind of judgment, the transmission of which can be supported by referring to specific past events. Moreover, we have argued that the transmission of judgments about others’ traits and dispositions is a domain in which incentives to mislead and risks to be misled are regularly high. That is, we should expect source claims and the modification of conversational commitments they allow (i.e., testimony) to be particularly important in the transmission of trait judgments, that is, in gossip. Claims about specific past events are, however, not only relevant in the transmission of generic beliefs, and generic beliefs can be transmitted without ever referring to such events. In fact, as mentioned above, past events play a role in the transmission of generic beliefs not necessarily in virtue of the specificity or pastness of these events but rather as input to inductive-learning machinery or as examples of a more general pattern. To explain why the past for its own sake seems to have such a special status for human beings, we should look for a domain in which retrospective reference to particulars is required.

A further reason for why the transmission of generics, including reputation, may not provide a sufficient evolutionary pressure for the development of testimony is the fact that the main beneficiaries of teaching and of the spread of reputational information are the recipients of such communication: They acquire knowledge to be used in the future, which may have fitness consequences. Although there might be other factors that would make these types of communication fitness enhancing for communicators (e.g., reputational gain), these would only be indirect benefits. It is therefore worth asking whether there is an explanation that relies on direct fitness enhancement for the person who gives testimony. How could the speaker’s, as opposed to others’, fitness benefit from testimony?

Type and token causes

One domain in which this might be the case is causal judgments. The relationship between past and present is commonly conceived of in terms of causal relations. People constantly infer causal relations between events unfolding around them (Gopnik, 2000). Nevertheless, although causes are events, the representation of causal relations intrinsically does not require the representation of specific events: Causes, after all, can be represented in terms of “type causation” (“being shot kills people”). What does require the representation of a specific event, however, is the retrospective inference from a specific, token effect to its token cause (“Mark died because he was shot”). Crucially, such a retrospective inference requires not only the representation of an event as the cause of an effect but also its representation as having occurred temporally before the effect, that is, in the past. Moreover, although in principle an unbounded set of causes underlies any given effect, humans commonly represent causes and their effects as standing in a one-to-one relationship; in other words, specific token effects are often represented as having specific token causes. Token-causal judgments then have all the qualities that would seem to make the representation of past events necessary in order to link causes to effects.

Token-causal judgment is a domain in which particulars (i.e., specific past events and their counterfactual derivatives) are crucial (Campbell, 1996). Epistemic authority about the actual occurrence of specific past events might thus matter particularly in the transmission of causal judgments as causal explanations.⁷ Note that a type-causal explanation requires claims about the past experience of a specific event only insofar as they are relevant to the transmission of generic causal beliefs more generally. Claims to personal experience will thus be particularly powerful in the transmission of token-causal judgments.

Physical and social effects

The benefits of making token-causal judgments may depend on the nature of the effect in question. Finding a token cause for a physical effect is an inference to a specific, past event, but establishing this causal relation affects our future fitness only to the extent that it allows us to inductively generalize it (and use this generalization, for example, in planning future actions). Although causal thinking is a powerful learning engine that allows us to understand, predict, and explain contingencies in our environment, these benefits only partially apply to thinking in terms of token causes: Particular instances of causation serve as learning opportunities mostly in light of our capacity for building causal maps from representations of type-causal relations. Moreover, token instances of causation can usually be interpreted only through the application of type-causal assumptions and serve as opportunities for learning only insofar as they inform these assumptions. In the physical domain, token-causal judgments are therefore most important as inputs for inductive learning mechanisms. As we have argued above, however, inductive inferences are not optimally transmitted through testimony.

This is similarly true for retrospective causal inference: Say you arrive in your office one day to find that your computer screen is lying on the floor and does not work anymore. When you ask one of your coworkers what happened, she informs you that another one of your coworkers threw your screen to the ground in a fit of frustration. Regarding the purely physical cause-and-effect relations at play here, this information will be relevant to you insofar as it informs you that computer screens tend to stop working when thrown to the ground. That is, you will benefit from knowing the physical cause of why your screen stopped working insofar as you can infer a type-causal relation from this specific instance. You might then, for example, affix your screen to your desk so that it does not fall or get thrown to the ground in the future. In this way (i.e., via a type-causal inference), information about token-causal relations can affect your future behavior adaptively.

For humans, however, causes instantiated in specific past events are often more than opportunities for learning about our physical environment via type-causal inferences; they may also have important social implications. Many physical or biological causes produce not only physical but also social effects. If I manually create an artifact, I may earn rights to use it or own it; if my aunt dies, I may inherit some of her property; if your dog kills my lamb, you may have to compensate me; if I father a child, I may have to contribute to his or her upbringing; if a landslide destroys my crop, I may be relieved from the duty of contributing to the common good; and so on.

Note that although some, but not all, of these events are actions of social agents, all of them produce lasting changes in the physical environment. However, crucially, they also produce social facts: Someone becomes or ceases to be the owner of a resource, a father, a debtor, and so forth, thereby producing entitlements and obligations that have fitness consequences for the future. And because these social facts are not perceivable, unlike the physical or biological effects of the same causes, only the causes that produced them in the past can prove that they hold in the present. In fact, these causes are thought to play a constitutive role in them. To a large extent, this explains why we are bound to care so much about specific past events (including their actual details).

Going back to the example of the broken screen above: The crucial inference you will likely draw from the information that your coworker threw your screen to the ground will in fact not be a type-causal relation. Instead, you will form an accountability judgment that serves as the basis for a claim to an entitlement for compensation. That is, in this case, knowing the specific token cause will have an effect on your future fitness not (primarily) by allowing you to learn about type-causal relations but by allowing you to infer and transmit the social effects produced contingently with the physical ones.

When testimony is necessary: communicatively generated commitments

People see certain physical events as generating (and sometimes even constituting) social effects. Thus, occurrences of certain specific past events can inform us about present and future social entitlements and obligations, and because these social effects exist primarily as mental representations, testimony about such events can be an important argument during negotiations of entitlements. However, strictly speaking, testimony is not the only way to prove the occurrence of such events. Even though the social consequences of such events (e.g., the entitlement for compensation) are not perceivable, their contingent physical effects (the broken screen) can still be traceable. This in turn, at least in principle, may allow retrospective inference from effects to their causes without relying on the testimony of others. One can always try to do the detective work backward from the physical effects to infer, and argue for, the cause and thereby for its social effect. The craftwork on an artifact may show who created it; the exhumed corpse of my aunt can prove that she really died; the injuries of my lamb may reveal that your dog was the culprit; fatherhood can be inferred from DNA tests or from facial resemblance; the change of the landscape provides evidence of a landslide; and so on. Thus, although testimony (and the episodic memory it requires) is useful to argue for the validity of a given social fact in all of these cases, it is not mandatory: Contingent physical effects may allow us to infer the past physical (or biological) causes that induced the present social facts in question.

Nonetheless, once the ability to represent the social effects of events emerged in human evolution, it likely made the ability to refer to the past on the basis of remembering it (i.e., testimony) extremely useful. Once in place, however, this ability could then have given rise to new forms of commitments that do not necessarily rely on traceable physical effects: Promises, agreements, bets, and marriages are all examples of social effects that do not necessarily leave physical traces behind. Instead, they are generated by communicative acts. These instances of communication normally have no correlated, lasting physical effects. Therefore, not only their social effect but also the cause itself exists only in the mind of the participants.

If Margaret promises Elena that she will be back home by 7 p.m., the effect of this promise (i.e., that Margaret is now committed to a certain behavior) is not observable; it survives—if at all—only in the minds of Margaret and Elena (and any possible witnesses). Nonetheless, the promise-commitment relation here seems to be of the same kind as the relation between token cause and token effect described above.

The proof that such a cause occurred could come only from testimony—hence the necessity of episodic memory. In a sense then, social effects of this sort have an inherent “dual temporality.” On the one hand, they are about the future: A promise obliges to a behavior, an ownership transfer entitles the beneficiary to privileged use, and so forth. Nonetheless, once established, the existence of the ensuing obligations and entitlements can be justified or proven only by reference to the past event that established them. Without the ability to communicatively refer to the past, such practices could not have developed.

Crucially, it is exactly because of the dependence on testimony of these causal events that societies developed ways to ensure their provability by recruiting witnesses for ceremonies, and (only more recently) by creating correlated physical effects of these “nonphysical’ causes in the form of documents such as contracts, certificates, memoranda, bills, records, and so forth (e.g., Basu, Dickhaut, Hecht, Towry, & Waymire, 2009). That is, the ephemeral nature of the cause-effect relationship in social commitments induced the cultural evolution of a host of “commitment devices” (Fessler & Quintelier, 2013) designed to alleviate reliance on individual memory alone by requiring the commitment event to become physically traceable in one form or another. Further, events grounding explicit commitments are often ritually structured so as to be public and easily referable: A promise is accompanied by a handshake, a marriage by a ceremony, and so forth. Making a commitment public, for example, not only increases the cost of possible defection but also coordinates the representation of this social fact in the community. It is worth noting, however, that although unperceivable social facts (ownership, kinship relations, social status, etc.) are frequently signaled publicly to make sure that others are aware of them without having to prove them again and again, these documents do not simply indicate that certain social facts hold but are also designed to prove that the specific cause that brought them about indeed occurred (this is why date and place, which together individuate a specific episodic event, are included in them).

Before we continue, let us recap the argument we have developed in the last two sections. Information about specific past events can be used for various purposes. It can support inductive inferences about projectable properties of objects, agents, situations, and causal relations, which support the acquisition of generic knowledge about kinds, individuals, and type-causal relations. However, this purpose can be achieved in various other ways as well, and so it does not require the preservation, representation, or testimony about past events. A special subset of token-causal events, however, produces not only (or no traceable) physical effects but also social facts that allocate entitlements and obligations to specific individuals or groups. These social facts are generated by their own token causes, and therefore the ultimate proof of their existence is evidence of the occurrence of these token events. Episodic memory and testimony of past events can thus be crucial for the stable maintenance of such social facts in the community (see below). In fact, reliance on communicatively established commitments (such as promises), which may not leave any physical trace behind, could not even emerge without cognitive mechanisms that ground both prospective memory, to ensure fulfillment, and retrospective memory and testimony, to ensure accountability.

The Historicity of Social Facts

What then is the relationship between social facts (obligations, entitlements, commitments, etc.) and episodic memory? It has been proposed that memory capacities are necessary for enabling the tracking of certain forms of social interaction of the “who did what to whom” sort (e.g., Stevens & Hauser, 2004). However, tracking social relations of this kind can also be accomplished by cognitive “bookkeeping” mechanisms that keep and update scores of interacting agents on each encounter. A given interaction would then be interpreted depending on the score of each agent involved (e.g., Nowak & Sigmund, 1998). In this way, nothing about the event in question has to be remembered because its outcome simply updates such a score. Say Isa lends €5 to Rahmeed. For Rahmeed to reciprocate and pay Isa back, all he has to keep in memory is that he now owes Isa €5. Nothing else about the lending event itself has to be remembered.

A number of different authors have proposed that such a tracking mechanism could have been implemented through an “attitudinal” (Brosnan & DeWaal, 2002) or emotional-scoring system in nonhuman animals (Schino & Aureli, 2009; see also Gervais & Fessler, 2017). It is likely that in many situations exchange-related information is tracked in a similar manner in humans (Bell, Koranyi, Buchner, & Rothermund, 2017; for a modeling approach related to this issue, see Kleiman-Weiner, Ho, Austerweil, Littman, & Tenenbaum, 2016). The representation of specific past events is therefore not a requirement for maintaining stable pairwise social relations. Learning from “exchange events” is similar to learning from events that have no social consequences: One can draw inferences from such events without storing much of what happened.

Although commitments and entitlements can only ultimately be proven by reference to specific past events, the representation of their existence does not depend on the capacity to recall specific past events. In order to believe that John owns his car, you do not have to remember anything about the event in which he acquired it (even though you have to assume that there was such an event). Why should the transmission of the privately represented social effects of a given event be important then?

Maintaining and stabilizing social facts

The social effects produced by ordinary physical or biological events have to be maintained by some forms of public representation, such as face-to-face communication. After all, they often exist only in the minds of individuals, and communication is the main means by which these effects become and remain shared. If they are not shared, social facts do not fulfill their function, so it is in everyone’s interest to coordinate them appropriately.

Put differently, to become social facts, the privately represented social effects of events have to be shared, and agreement about them has to be established. Ownership, social structure, and social roles are good examples here: They may also be marked by permanent public signals to ensure common acceptance even in the absence of direct verbal communication. Social facts such as these may be generated by token causes, but their shared maintenance depends on communication, and if their existence is disputed, they can be negotiated by reference to the events that brought them about. Social facts inherently depend on public agreement, and to achieve such agreement, the past events grounding a given fact have to be available. This is important not only in cases of conflict. Rather, it is simply not possible to decide privately whether a given social fact indeed applies. Although one can represent a social fact in itself without entertaining the (historical) reasons why it should hold, in communication such reasons might have to be explicitly invoked as the ultimate argument for its existence.

Crucially, for humans, the social implications that an event establishes can be ambiguous. A given episode is often important not only because of factual occurrences but also for the myriad ways in which these events could have turned out. What a person did not do and what his or her intentions were in acting, for example, are essential in computing the ways in which commitments should be distributed (e.g., Gerstenberg et al., 2018). Although humans have a host of specialized cognitive mechanisms that enables them to carry out such computations online, the transmission of the conclusion will often require justification. Distinguishing between, for example, incompetence and malevolence will sometimes require that one refers to details of the specific action in question. Although malevolence should trigger punishment or ostracism, incompetence does not necessarily call for these reactions (for why the ability to make such a distinction might be important, see Nowak & Sigmund, 2005). The social coordination of the representations of implications of specific events will thus often unavoidably require communication about such events (for a similar point, see Pietraszewski, 2016).

Testimony, therefore, helps to maintain and coordinate the validity of entitlements, obligations, and commitments within a social group. Although it might also serve other important functions, testimony can play a “signaling” role in advertising the existence of certain social facts: I claim that this knife is mine and I justify it by the fact that I made it, or that I inherited it, or that it was donated to me, and so forth. It does not necessarily require an open challenge or violation of property rights to make these assertions; in the absence of a permanent symbol system to mark ownership and other entitlements, repeated declarations of social facts may be necessary to maintain their shared nature and to let newcomers know about them.

Note that this conclusion does not require that communicatively coordinating social facts should always, or even necessarily, involve pointing to specific past events. After all, beliefs can be transmitted on the basis of trust alone, without requiring the representation or transmission of reasons. Moreover, social effects affecting a whole community are commonly structured to be purposefully independent of individual testimony. Events establishing important social effects are ritualized or designed to generate public knowledge from the outset by either generating concomitant physical effects (e.g., documents) or many witnesses. In this case, testimony becomes less important.⁸ In fact, the motivation to make such events independent of individual testimony has likely led to the development of technologies allowing for the generation of public knowledge.

Once public knowledge about a given social fact has been established it will rarely be challenged: Marriages or kinship relations rarely, if ever, become a matter of dispute. Therefore, social effects that affect the whole community will often not depend on testimony to be maintained because they generate public knowledge or are otherwise made traceable for everyone. One reason for this, however, is that they are assumed to be ultimately appropriately justified by past events, which can be made available in one form or another in case of doubt. Only the possibility of pointing to the establishing event in any given case ensures that disagreements about the fact in question could be resolved in principle. Many social facts are simply such that only their establishing event can ultimately arbiter whether (or in what way) they hold. If there were no way to refer to or to make available these events, there would be no way to ultimately ensure the appropriateness of claims about such social facts and consequently to justify the ensuing entitlements and to enforce their fulfillment.

Nonetheless, if our analysis here is correct then the capacity for testimony, underpinned by human episodic memory, must have enabled the capacity for coordinating certain social facts in the first place and for generating new types of social facts that could not even exist without testimony. The reason for this is not that such commitments require episodic memory to be cognitively traceable and behaviorally implementable by the individuals involved. Rather, they require testimony to be shared in a community. This is because in some cases there could be no fact of the matter whether a given obligation, entitlement, or commitment applies without the potential of testimony about the specific past event causally grounding the social effect under dispute.

Motivated remembering, memory bias, and narrativity

The above argument predicts that episodic memory is motivated (and hence to some extent biased) by design to justify one’s own present entitlements (see also Lambek, 1996). Mahr and Csibra (2018) argued that one way in which such a bias manifests is through “recollective my-side bias.” Episodic-memory construction is more likely to confirm and support our prior beliefs than to contradict them. However, this might not be the only source of architectural bias in episodic memory. Episodes are neither retrieved nor communicated as atomized particles but as narratives. Testimony is not only given as a series of propositions but also narrativized in a way that makes it more likely for the audience to draw certain inferences over others. According to Keven (2016) and Keven, Kurczek, Rosenbaum, and Craver (2017), episodic memory retrieval includes a mechanism (referred to as “narrative binding”) connecting isolated event representations by inferring not only temporal (X happened before or after Y) but also causal (X occurred because of Y) and teleological (X occurred in order to bring about Y) relations between them. From this perspective, episodic memories allow us to “make sense” of the past in light of causal and teleological relations between different events as well as their connection to the present (Bietti, Tilston, & Bangerter, 2019).

How did I get to work this morning? I went to the bus stop to get the bus at 7:45 but the stop was closed because of a construction site and so I had to take the metro to work instead. Already this minimalistic account of the events of this morning includes a significant amount of selection and interpretation insofar as certain events and their causal or teleological relations are highlighted and others left out. In order to effectively argue for the validity of a given social fact, making the temporal relations between events available is not enough. Instead, we must be able to bind events in a way that highlights the causal connections in question (e.g., having been forced to change the mode of transport this morning added delay to my travel and caused me to be late for work). Narrative-binding processes therefore always include a modicum of interpretation: Relating events causally and teleologically includes a selection process in which certain events are highlighted over others. Narratives are often effective because they display events as being (causally and teleologically) related in a way that suggests certain conclusions over others. The fact that episodic memory is narrativized and often biased in favor of present beliefs, attitudes, and goals (Anderson & Hanslmayr, 2014; Coman, Stone, Castano, & Hirst, 2014; Conway, 2005; Kappes & Crocket, 2016) follows from its crucial communicative role in establishing social facts in the present through reference to history. Thus, on our account, remembering (both individually and collectively) does not necessarily only or even primarily function to produce an accurate representation of the past.⁹

Moreover, our account provides a new perspective on the question of why we care so much about the accuracy of our representations of the past in the first place. Phenomena of memory fallibility and inaccuracy contributed to the birth of experimental psychology (Bartlett, 1932; Ebbinghaus, 1885) and continue to constitute one of its major areas of research (e.g., Kurkela & Dennis, 2016; Schacter, 2001). More generally, humans have devoted entire academic fields to the accurate reconstruction of history. Although humans surely have a general instrumental interest in the accuracy of their representations, to the extent that humans in general (and academics in particular) have a special interest in the accuracy of representations of the past, this interest is likely at least partly explained by the social importance of the past for the present.

Accountability judgments and the role of social norms

One might propose that testimony not only plays a role in signaling and coordinating the validity of social facts, as we argued above, but is moreover of crucial importance in the enforcement of social norms themselves. After all, the context in which testimony seems to matter most in contemporary societies is the legal domain. Testimony has been investigated by cognitive psychologists mostly in the form of eyewitness testimony for crimes (e.g., Neisser, 1981; Wright, Memon, Skagerberg, & Gabbert, 2009), and ethnographic accounts have often particularly focused on the role of witnesses in the legal domain (e.g., Gluckman, 1955). Thinking about the relationship between our capacity for testimony and norm enforcement, one might thus conclude that testimony enables the enforcement of social norms by informing others about the violation of those norms. Without the ability to share information about such violations, people would always be dependent on firsthand experience in judging whether a norm has been violated, which would not make it possible for norms to be widely enforced by third parties or communities in general.

In our view, however, the role that testimony plays in norm enforcement is just a special case of the more general role we have outlined above. In essence, sharing event information pertinent to norm violations aims to transmit a judgment—an accountability judgment—to establish a social fact about such accountability. That is, although accountability judgments by themselves are private, they can be justified and thereby transmitted to others by pointing to the event in which a norm was violated.¹⁰ The transmission of such a judgment aims to establish a shared representation of accountability, through which it would become a social fact. The enforcement of the norm in question, however, may follow from the accountability judgment itself, not from its transmission. Only once accountability has been established and is shared within a group may norm enforcement ensue. The transmission of the norm-violation event serves to coordinate only the representation of accountability and is not directly involved in the enforcement of the norm.

Thus, testimony is common in the social negotiation of accountability judgments because additional reassurance in communication about the past is required when the stakes are high. Arguably, as the domain of norm enforcement has become institutionalized, the forms in which testimony is given in this context (e.g., as eyewitness testimony under oath) have also become cultural institutions. Norm enforcement is facilitated through testimony because the fact of a norm being violated is sometimes not physically traceable. In fact, contrary to other domains, accountability transmission can hardly be alleviated from its dependence on testimony through cultural or technological solutions in principle: Norm violations are usually carried out in a way that avoids their publicity or documentation.

Testimony, therefore, will often help in justifying and determining punishment. In essence, however, norm enforcement and our capacity for testimony are not dependent on one another. Norms can be enforced without the involvement of testimony, and testimony is effective and occurs outside the domain of norm violations and their enforcement. It is important to note here, however, that the role of testimony in the transmission and coordination of social facts depends, to some extent, on social norms in the first place. After all, what social consequences follow from a given event is commonly governed by social norms.

For example, in East Timor land ownership is negotiated on the basis of a norm of first possession (Fitzpatrick & Barnes, 2010).¹¹ As a result of such a norm, it becomes relevant who (or whose ancestors) first settled on a given piece of land in deciding disputes about land ownership. This particular past event would, however, entirely lose its importance as a way of determining present land ownership in the absence of a norm of first possession.

It follows from this point that our propensity to represent (and observe) social norms must have existed before the emergence of the role of testimony outlined here.

Nonetheless, although testimony might not play a role in the enforcement of social norms and commitments, it certainly makes them more effective. Our capacity to bear testimony changes the dynamics of social interactions in crucial ways. The possibility of testimony transforms the payoff structure of two-person interactions into one in which third parties are always at least potentially present. In fact, even the possibility of report has been shown to be highly effective in promoting cooperation (e.g., Wu et al., 2015, 2016a).

People intensely care about whether their behavior is being witnessed by others, and this concern emerges relatively early in development: By the age of 5 years, children have developed a robust sense of the consequences of someone else witnessing their norm violations (and the communicative forms of aggression that can ensue) and adjust their behaviors accordingly. Five-year-olds behave more prosocially in the presence of peers (Engelmann, Herrmann, & Tomasello, 2012) when they believe to be watched (Piazza, Bering, & Ingram, 2011) or when their actions are witnessed by others (Leimgruber, Shaw, Santos, & Olson, 2012). On the one hand, these effects are likely to occur because witnessing someone’s behavior might cause third parties to draw inferences about their traits and accountabilities. On the other hand, people are likely aware that witnesses could pass on their evaluations via testimony. This would spread their judgments to the community, which might in turn motivate alliance recruitment against the observed individuals (Boehm, 2012; Pietraszewski, 2016) and potentially influence their reputation. In fact, gossip is likely to play an important role in this process: Although (as discussed above) gossip regulates the spread of reputational information, it also serves as a norm-enforcing device by transmitting others’ accountabilities through reference to norm-violating behaviors.

Conclusion

The main aim of this article is to give an account of why the past has such a special status for humans. The past plays a particularly crucial role in human social life. For humans, events have effects not only in our physical but also our social environments. The representation of such effects affects our future fitness to the extent that we can establish the validity of the ensuing social effects with others. The only way this can commonly be achieved is by retrospectively pointing to the event that produced the social effect in the first place. For this reason, the past becomes highly important to us and is also so frequently contested. It is this circumstance that makes the ability to remember the past in episodic memory particularly beneficial.

Episodic memory allows one to become a witness of the past and give testimony about it. Testimony is a way to facilitate the transmission of information about past events by conferring epistemic authority and increasing speaker commitment. Such facilitation is required in cases in which one’s audience requires additional reassurance about whatever one is asserting. Given the extraordinary importance of representations of the past in deciding what social realities apply in the present, we should expect claims about history to be a context in which listeners regularly require such additional reassurance.

Testimony will thereby be most important in the negotiation and transmission of our own and others’ commitments, entitlements, and accountabilities. Communicatively pointing to the past allows us to justify assertions about the existence of social facts, and a large range of cultural practices has developed exactly to alleviate the reliance of social reality on individual memory and testimony.

This view has consequences for what the evolution of episodic memory (the cognitive basis of testimony) must have looked like. Episodic memory might have developed only once humans were able to represent the social effects of the events in their environment. This ability, however, must have required the prior emergence of social norms determining these social cause-and-effect relationships. If no one represented or followed social norms, the past would lose its importance as a way of coordinating social realities. Once in place, our ability to testify about past events could then also be used to transmit generic beliefs to others and thereby make reputation dissemination more effective.