Mahr & Csibra's (M&C's) account rests in part on the literature on human cognition but stands full weight upon Morgan's canon – namely, that in nonhuman behaviour, it is to the simplest levels of explanation that we must first appeal. Our approach is not to argue that M&C's thesis is wrong. However, Morgan's canon does not permit a rejection outright of higher cognitive processes in other species, because it is not evidence. Therefore, the unfalsifiability of M&C's account correlates perfectly with the provability of the presence of episodic memory in non-linguistic species. The canon is at best a sort of subjective (and likely anthropocentric) Bayesian prior, with a theoretically tenuous default setting of zero. We briefly present the case against zero, and for reasons of space focus on only two species: scrub jays and rhesus monkeys.

M&C's definitions of episodic memory and autonoesis are intrinsically human-specific, so we adopt a broad, “traditional” definition of episodic memory as the integrated knowledge of what, where, and when (WWW) something occurred, coupled with a sense of recollection or re-experience (autonoesis) when accessing this information. M&C redefine autonoesis somewhat more specifically than this, as an outcome of the metarepresentation of a scenario under the “epistemic attitude of remembering” (sect. 1.1.2, para. 5). We thus briefly examine evidence for WWW memory and the metarepresentation of memory, the latter being as close as empirical research on nonhumans can presently get to the authors' own definition, and invite the readers to draw their own conclusion.

Our first point is that WWW memory has been observed in a number of species (for a review, see Cheke & Clayton Reference Cheke and Clayton2010), but it is in scrub jays that it has been most thoroughly examined. Jays selectively recover cached food according to when they cache it (i.e., if too long a time has passed, the worms perish, but the peanuts are fine), what it is (worms are tastier than peanuts, given the choice), and where they bury it (Clayton & Dickinson Reference Clayton and Dickinson1998; Reference Clayton and Dickinson1999b). Moreover, they can flexibly adapt to recover food types they have not recently been fed to satiety on (Clayton & Dickinson Reference Clayton and Dickinson1999a), show flexible awareness of the rate at which different food types decline in freshness (Clayton et al. Reference Clayton, Yu and Dickinson2001) or improve as they “ripen” (de Kort et al. Reference de Kort, Dickinson and Clayton2005), can update their information about cached foods even between caching and recovery (Clayton et al. Reference Clayton, Yu and Dickinson2003), show awareness of which individuals may have been watching them while they cached (Dally et al. Reference Dally, Emery and Clayton2006), and can dissociate the location of specific food from other, less preferred food in the same container (Clayton & Dickinson Reference Clayton and Dickinson1999b). Clayton and colleagues, mindful of the autonoetic component of episodic memory that only language users can realistically convey, termed this “episodic-like” memory (see also Clayton & Russell Reference Clayton and Russell2009). Overall, this pattern of results suggests that the jays form an integrated and cognitively rich WWW memory (de Kort et al. Reference de Kort, Dickinson and Clayton2005). Moreover, this memory is grounded in the jay's own agentive experience. This need not be a trivial point – the ability to judge a signal as unreliable, even from a conspecific (e.g., Cheney & Seyfarth Reference Cheney and Seyfarth1988), suggests there is theoretical utility in retaining some aspect of the self as the source of information.

Second, there is also by now a substantial body of evidence suggesting nonhumans can metarepresent their own memory – essentially “know what they remember.” Rhesus monkeys are the species for which the greatest amount of evidence is available. For example, in delayed matching-to-sample tests, the subject is presented with a stimulus to remember and then given a multiple-choice memory task after an intervening delay. The subject is either forced to take the test or given the opportunity to opt out before doing so. Rhesus monkeys showed evidence of greater accuracy on memory tests they chose to take than on memory tests that were forced (Hampton Reference Hampton2001) and were more likely to “bet” more on tests they were about to answer correctly or felt they already answered correctly (Morgan et al. Reference Morgan, Kornell, Kornblum and Terrace2014), suggesting that they could assess their own memories, even in the absence of the stimuli, and chose to take tests when their confidence was relatively high. In another task, rhesus monkeys were presented with a number of tubes and either witnessed one of them being baited with food or had no knowledge of which tube the food was in. The monkeys were more likely to look into the tubes first when they had no memory of where the food was than when they had seen the baiting, indicating that they “knew when they didn't know” (Basile et al. Reference Basile, Schroeder, Brown, Templer and Hampton2015; Templer & Hampton Reference Templer and Hampton2012; see also Hampton & Hampstead Reference Hampton and Hampstead2006). Moreover, these apparently metacognitive judgments have been shown to transfer across different tasks, lending weight to the argument that they are cognitively independent of associative learning and stimulus – or task-specific factors (Kornell et al. Reference Kornell, Son and Terrace2007; Morgan et al. Reference Morgan, Kornell, Kornblum and Terrace2014).

Crucially, we do not claim that nonhuman animals must have episodic memory or metamemory (Hampton Reference Hampton2009), and we believe that M&C's central argument may be correct, particularly as it pertains to humans, and may furthermore be testable. For example, in English we can make the distinction between the episodic past (e.g., “I remember posting the letter”) versus non-episodic past (e.g., “I remembered to post the letter”). One possible test of M&C's account could involve asking which sentence was more “ill-willed” when the participant knew the speaker was lying. Insofar as we may tie our reputations to the mast of our epistemic authority, we suspect the majority would be particularly irked by the invocation of specifically episodic rather than semantic memory in “I remember posting … .” However, the validity of the human story rests on the invalidity of the nonhuman one. Whether we should begin to look in more detail at the crux of M&C's profoundly human account of episodic memory, we leave to the reader to decide.