A good test of a framework is the way in which it handles rare cases. The challenging example I wish to introduce for Mahr & Csibra's (M&C's) consideration is that of the episodic memory of individuals with dissociative identity disorder.

The Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM-V) diagnostic category for dissociative identity disorder (DID) has two main criteria:

1. A. Disruption of identity characterized by two or more distinct personality states. …

2. B. Recurrent gaps in the recall of everyday events, important personal information, and for traumatic events that are inconsistent with ordinary forgetting. (American Psychiatric Association 2013, p. 292)

Other criteria include the ruling out of cultural factors and general medical conditions. Any gap in the recall of everyday events is usually filled by the recall of another personality state. Thus, detail of the previous day's activities might be traced by piecing together the (non-overlapping) episodic recall of three or four alters.

With DID patients, then, the phenomenon of interest relates to what one alternative personality state (alter) knows about what happened to another alter. One experimental demonstration of this involves an alter learning 24 nouns. A second alter, who denies all knowledge of the preceding procedure, is taught a different set of nouns. A week later, without warning, the second alter is brought out and asked to follow a recognition memory test with the 48 stimuli together with distractors. Huntjens et al. (Reference Huntjens, Postma, Peters, Woertman and van der Hart2003; Reference Huntjens, Peters, Woertman, van der Hart and Postma2007) found that their DID subjects responded to the words presented to the other alter as though they had previously seen them, in spite of having no recollection of the presentation. These authors conclude that “dissociators … seem to be characterised by the belief of being unable to recall information instead of an actual retrieval inability” (2007, p. 788, their italics). This situation, where there is no phenomenal experience of an event, but where the event is exerting a clear influence on behaviour, matches the phenomenon of post-hypnotic amnesia (e.g., Smith et al. Reference Smith, Oakley and Morton2013). Here, subjects claim no recollection of recent experiences which, nonetheless, affect current behaviour. Smith et al. (Reference Smith, Oakley and Morton2013) have suggested that executive processes are responsible for controlling the initial access to material and then determine whether retrieved information is allowed into consciousness. However, material that has been accessed will exert some influence on processing even though it is not allowed into consciousness. Morton (Reference Morton2017) gives a similar account for the results of Huntjens et al. (Reference Huntjens, Postma, Peters, Woertman and van der Hart2003; Reference Huntjens, Verschuere and McNally2012) described above.

Using the same experimental procedure as Huntjens et al. (Reference Huntjens, Postma, Peters, Woertman and van der Hart2003), Morton (Reference Morton and Sinason2012; Reference Morton2017) found two individuals with DID where one alter responded to the words that had been presented to another alter in exactly the same way as they responded to the control words. In other words, this material could not even be accessed by the second alter despite being a full part of the first alter's phenomenal past.

Similar results have been shown with more complex material. Reinders et al. (Reference Reinders, Nijenhuis, Paans, Korf, Willemsen and Den Boer2003) studied DID patients who were in either a trauma-related identity state or a neutral identity state. The former generated an autobiographical traumatic memory that the latter failed to recognise as relating to themselves. These memories were contrasted to neutral memory scripts, which both states accepted as autobiographical. The two scripts were put into the third person and read in a neutral tone to the patients while they were in a scanner. The scans were similar with the neutral script for the two states, and there were only small differences between the scans of the two scripts for the neutral identity state. The big difference occurred when the trauma-related state listened to the trauma script. Note that not only did the neutral identity states claim they did not recognise the trauma story as relevant to themselves, but also their brain response backed that up; there was an amnesic barrier between the alters.

The possible problem for the M&C position, then, is that an episodic memory, fully self-referential and autonoetic when accessed by one alter, behaves totally differently when accessed by another alter. In some cases, the material is treated as knowledge, and in other cases it is treated as though it does not exist. This seems to create problems if the differences between episodic memory, event memory, and beliefs are simply there in the representations, as M&C seem to claim. The alternative way of thinking about this seeming paradox has two components. First, the emphasis would be on meta-cognitive processes: what other people refer to as executive processes. This is distinct from the metarepresentational format referred to by M&C, which would be seen as the product of current processing. The second component is the retrieval process itself. If one thinks in terms of context-sensitive memory (e.g., Godden & Baddeley Reference Godden and Baddeley1975), it is natural to treat some notion of self as a part of the addressable section of a memory representation. This would have a normal use of distinguishing between representations of events that were first- or secondhand. Its interpretation by the executive (meta-cognitive) processes would give rise to autonoesis.

Within such a framework, the treatment of the DID case is straightforward as outlined above. Executive processes use the self marker specific to the alter that is currently active either to restrict whether material can be made conscious or, in other cases, whether the memory representations can be accessed at all. In this way a particular representation can either have an autonoetic character or not. The apparent paradox is solved here by the use of processing. It is not immediately apparent how a representational view such as that put forward by M&C would deal with it. I leave it in their hands.