Mahr & Csibra (M&C) bring together considerable evidence and ideas to advance their primary thesis that an important function of episodic memory is to support communicative interactions. We find merit in this discussion, but focus on a different aspect of M&C's target article.

M&C highlight what they believe is a problem with theories that hold that the constructive nature of episodic memory, and associated vulnerability to errors and distortions, is related to the role of episodic memory in constructing simulations of future events: The fact that episodic memory is often veridical “becomes mysterious in this view” (sect. 2.1.1, para. 2). M&C cite Schacter and Addis (Reference Schacter and Addis2007) as proponents of this view, and others have advanced similar ideas (cf. Dudai & Carruthers Reference Dudai and Carruthers2005; Suddendorf & Corballis Reference Suddendorf and Corballis2007). In what follows, we focus on our own perspective, the constructive episodic simulation hypothesis (Schacter & Addis Reference Schacter and Addis2007). We disagree with M&C's assertion that in our view, it is mysterious that episodic memory can be veridical.

According to the constructive episodic simulation hypothesis, remembering past experiences and simulating future experiences rely on similar kinds of episodic information and processes, including relational processing (e.g., Eichenbaum & Cohen Reference Eichenbaum and Cohen2014): the capacity to bind together distinct elements of experiences (for additional discussion, see Roberts et al. Reference Roberts, Schacter and Addis2017). Our perspective further maintains that simulating future events requires a system that supports flexible recombination of details from past events into novel scenarios, which entails relational processing. Thus, episodic memory possesses key characteristics that make it adaptive for simulating alternative future scenarios and “trying out” novel approaches to those scenarios based on past experiences. However, the flexible recombination processes that are critical for episodic simulation can also make the system prone to errors that arise from mistakenly combining elements of stored episodes. Such memory conjunction errors (Reinitz et al. Reference Reinitz, Lammers and Cochran1992) and related source misattributions (Johnson et al. Reference Johnson, Hashtroudi and Lindsay1993) are common forms of memory distortion (Devitt & Schacter Reference Devitt and Schacter2016; Schacter Reference Schacter2001).

Critically, our emphasis on the perils of flexible recombination need not mean that the veridicality of episodic memory is mysterious. From our perspective, episodic memory is often accurate because constructing useful simulations of future experiences that benefit from what actually happened in relevant past experiences requires some degree of accurate representation; a memory system entirely lacking the capacity for accurate representation would be virtually useless for simulating the future based on the past. We already know from work on affective forecasting errors that limitations on episodic memory can produce inaccurate simulations of future happiness (Gilbert & Wilson Reference Gilbert and Wilson2007), highlighting that some degree of veridicality in episodic memory is critical for constructing accurate simulations of future experiences (see also, Schacter Reference Schacter2012). M&C (sect. 2.1.2) acknowledge that memory veridicality is important for future simulation, but seek to attribute this feature to semantic memory. We agree that semantic memory is critical for future thinking and may contribute to its accuracy, but we have shown in several studies that episodic memory makes distinct contributions to future imagining that are experimentally dissociable from semantic memory (Schacter & Madore Reference Schacter and Madore2016). Moreover, future thinking that relies solely on semantic memory (e.g., established schemas) would lack the flexibility that we think characterizes episodic memory and is crucial for future simulation.

In Schacter and Addis (Reference Schacter and Addis2007), we addressed the issue of memory veridicality, emphasizing that although “a memory system that simply stored rote records of what happened in the past would not be well suited to simulating future events,” the system we envisaged “can draw on elements of the past and retain the general sense or gist of what has happened” (p. 778). Remembering gist information may partly reflect the influence of semantic memory, but the key point is that we rejected the notion of an episodic memory that draws on “rote records” and did not reject the broader idea that episodic memory is to some extent veridical.

It should therefore be clear that from our view, the link between memory errors and constructing novel event simulations based on elements of past experiences depends specifically on the involvement of flexible recombination processes in episodic simulation, not on a wholesale dismissal of veridicality in episodic memory. Although in Schacter and Addis (Reference Schacter and Addis2007) we cited no experimental evidence for a link between flexible recombination and memory errors, we have recently produced such evidence. Devitt et al. (Reference Devitt, Monk-Fromont, Schacter and Addis2016) used an experimental recombination paradigm, where people, places, and objects that occurred together in actual past experiences are recombined into novel arrangements by an experimenter, and participants are then asked to imagine events involving the recombined details. Devitt et al. (Reference Devitt, Monk-Fromont, Schacter and Addis2016) reported that such recombinatory processing sometimes resulted in autobiographical memory conjunction errors, where participants claimed that a novel, recombined event actually happened in the past. Carpenter and Schacter (Reference Carpenter and Schacter2017) adapted an associative inference paradigm (Zeithamova & Preston Reference Zeithamova and Preston2010) that requires combining information from separate episodes involving people, objects, and contextual settings to make inferences about individuals who are linked to one another because each is paired with the same object. Across four experiments, Carpenter and Schacter (Reference Carpenter and Schacter2017) showed that participants are more susceptible to false memories that result from mistakenly combining contextual details from related episodes when they make correct compared with incorrect inferences about the relations between the people in these episodes. Critically, this boost in false memories occurred only when contextual details were probed after (versus before) an associative inference test that engaged flexible recombination processes.

These findings highlight that flexible recombination – not mental time travel into the future per se – likely underpins the link between episodic simulation and subsequent memory errors. Episodic future simulations rely on flexible recombination, but so do simulations of present and past events (Addis et al. Reference Addis, Pan, Vu, Laiser and Schacter2009; Schacter et al. Reference Schacter, Addis, Hassabis, Martin, Spreng and Szpunar2012; Reference Schacter, Benoit, De Brigard and Szpunar2015). Future simulation is perhaps the main adaptive function served by flexible recombination processes, but is not the only one (Schacter et al. Reference Schacter, Benoit, De Brigard and Szpunar2015). A memory system that supports episodic simulation requires some veridicality to benefit from past experiences, but reliance on flexible recombination to construct simulations can undermine memory accuracy when elements of experiences are mistakenly combined.